CN114386718A - Wind power plant output power short-time prediction algorithm combined with particle swarm neural network - Google Patents
Wind power plant output power short-time prediction algorithm combined with particle swarm neural network Download PDFInfo
- Publication number
- CN114386718A CN114386718A CN202210256315.7A CN202210256315A CN114386718A CN 114386718 A CN114386718 A CN 114386718A CN 202210256315 A CN202210256315 A CN 202210256315A CN 114386718 A CN114386718 A CN 114386718A
- Authority
- CN
- China
- Prior art keywords
- wind
- wind power
- prediction
- power
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 71
- 239000002245 particle Substances 0.000 title claims abstract description 54
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 46
- 238000000034 method Methods 0.000 claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 43
- 230000008569 process Effects 0.000 claims abstract description 41
- 238000010248 power generation Methods 0.000 claims abstract description 21
- 238000003062 neural network model Methods 0.000 claims abstract description 20
- 238000004458 analytical method Methods 0.000 claims abstract description 11
- 238000006243 chemical reaction Methods 0.000 claims abstract description 4
- 238000000556 factor analysis Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 26
- 238000005457 optimization Methods 0.000 claims description 17
- 230000000875 corresponding effect Effects 0.000 claims description 16
- 238000011156 evaluation Methods 0.000 claims description 14
- 239000000126 substance Substances 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000010408 sweeping Methods 0.000 claims description 9
- 230000007547 defect Effects 0.000 claims description 7
- 230000002596 correlated effect Effects 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 6
- 230000002441 reversible effect Effects 0.000 claims description 5
- 238000004088 simulation Methods 0.000 claims description 5
- 238000012544 monitoring process Methods 0.000 claims description 4
- 150000001875 compounds Chemical class 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 230000002829 reductive effect Effects 0.000 description 8
- 230000008859 change Effects 0.000 description 6
- 238000003066 decision tree Methods 0.000 description 6
- 230000003068 static effect Effects 0.000 description 6
- 238000007637 random forest analysis Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 230000002159 abnormal effect Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000000670 limiting effect Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000010355 oscillation Effects 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 241001155961 Baris Species 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000002431 foraging effect Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000002945 steepest descent method Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
- H02J3/381—Dispersed generators
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
- H02J2300/20—The dispersed energy generation being of renewable origin
- H02J2300/28—The renewable source being wind energy
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/70—Wind energy
- Y02E10/76—Power conversion electric or electronic aspects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Power Engineering (AREA)
- Marketing (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Water Supply & Treatment (AREA)
- Quality & Reliability (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
Abstract
The invention discloses a short-time prediction algorithm for wind power plant output power combined with a particle swarm neural network, which comprises the following steps of: the method comprises the following steps: wind power prediction characteristic analysis is carried out, and analyzed data comprise uncertainty, conditionality and regionality of wind power; step two: the wind power generation influence factor analysis method comprises the following specific analysis processes: in the whole wind power generation system, a fan converts the kinetic energy of air into mechanical energy on a shaft system of the fan, transmits the mechanical energy to a rotor shaft of a wind driven generator, and finally converts the mechanical energy into electric energy, which is the first link step for realizing energy conversion; step three: and combining a short-time dynamic prediction model of the output power of the wind power plant of the particle swarm neural network to obtain a BP-PSO hybrid neural network model, wherein a set of input values and output values are given as training data. The dynamic prediction method can adopt a BP neural network algorithm optimized by particle swarm, dynamically adjust model parameters through an error discrimination function, and realize the dynamic prediction of the wind power.
Description
Technical Field
The invention relates to the field of wind power generation, in particular to a short-time prediction algorithm for output power of a wind power plant combined with a particle swarm neural network.
Background
Under the influence of policies for constructing a novel power system, wind energy is developed and utilized on a large scale, the total capacity of installed wind power equipment is obviously improved, and the difficult problem brought by wind power integration is also brought. The output power of the wind power plant is directly related to wind energy, and the wind power has the characteristics of randomness, volatility, uncertainty and the like, so that the wind power generation has stronger randomness and volatility. Although wind power has obvious advantages as a non-water clean energy source, the exhaustion and the environmental pollution of the traditional fossil energy sources can be effectively relieved, and the carbon emission is reduced, the difficulty of wind power grid connection is greatly increased due to the characteristics of inverse peak regulation, volatility, intermittence and the like. With the continuous improvement of the attention degree of the whole society on the electric energy quality, the stable operation of the electric power system is improved to a new height, and therefore, the prediction of the wind power generation power is a necessary premise for guaranteeing the safe and stable operation of the high-proportion new energy electric power system. From the perspective of the wind power plant, the accurate power prediction result can provide the operation and maintenance level of the wind power plant, and the wind abandon rate is reduced; from the perspective of an electric power system, the reliable power prediction result can effectively reduce the uncertainty of the wind power generation power and the adverse effect caused by the uncertainty, is beneficial to a dispatching department to make and timely adjust a dispatching plan, reduces the rotating reserve capacity of the system, and reduces the operation cost. The uncertainty of a current power grid is enhanced by the wind power grid-connected proportion with the source end rising year by year, higher requirements are put on the prediction precision of wind power generation power for ensuring the electric energy quality and the safe and economic operation of a power system, a wind power generation power prediction method needs to be explored urgently, and the high-precision prediction of the wind power generation power is realized. So far, research on dynamic adjustment of a wind power plant output power prediction model is few, and the model is used for establishing a corresponding model based on numerical weather forecast, such as a wind power prediction system based on a BP artificial neural network, a short-term wind power prediction method based on particle swarm optimization, a short-term wind power prediction model based on chaos DNA genetic algorithm and PSO combined optimization, and the like.
These wind farm output power prediction models share a common drawback, and once the model is determined, the model will continue to be used in the future prediction process regardless of the situation. It is not considered that the existing training model is not suitable for the prediction work at the present stage due to the weakened correlation between the training data and the real-time measurement data, so that the prediction error is increased. Therefore, the invention provides a short-time prediction algorithm for the output power of a wind power plant combined with a particle swarm neural network, which adopts a BP neural network algorithm optimized by the particle swarm and dynamically adjusts model parameters through an error discrimination function to realize the dynamic prediction of the wind power.
Disclosure of Invention
The invention provides a short-time prediction algorithm for wind power plant output power combined with a particle swarm neural network.
The invention solves the technical problems through the following technical scheme, and the invention comprises the following steps:
the method comprises the following steps: wind power prediction characteristic analysis is carried out, and analyzed data comprise uncertainty, conditionality and regionality of wind power;
step two: the wind power generation influence factor analysis method comprises the following specific analysis processes: in the whole wind power generation system, a fan converts the kinetic energy of air into mechanical energy on a shaft system of the fan, transmits the mechanical energy to a rotor shaft of a wind driven generator, and finally converts the mechanical energy into electric energy, which is the first link for realizing energy conversion;
wherein the content of the first and second substances,as a matter of time, the time is,in order to be the density of the air,to achieve the wind speed before entering the swept surface of the fan blades,the area of the wind is swept, and the wind speed is controlled,in order to be the quality of the air,is the air flow energy.
The fan paddle can have energy loss in the rotating process, namely not all wind energy can be captured by the wind turbine, and the wind energy captured by the wind turbine can be pushed out by combining the formula (1):
wherein the content of the first and second substances,to achieve the wind speed before entering the swept surface of the fan blades,is the wind speed after flowing out of the wind sweeping surface of the fan blade,is the utilization coefficient of the wind energy,indicating that the wind turbine is capable of deriving wind powerThe energy captured.
it can be known that whenThe wind energy utilization coefficient reaches the maximum valueMaximum power captured by wind turbineI.e. the power captured by the wind turbine is presentThus, therefore, it isFinal output power of the whole wind power generation systemMaximum power captured by wind turbineAre positively correlated:
wherein the content of the first and second substances,for the wind energy captured by the wind turbine,for the final output power of the entire wind power system,for the maximum power captured by the wind turbine,the wind speed before entering the swept surface of the fan blade;
the existing wind power plants all adopt a maximum power tracking strategy, namely, the power captured by wind power generation as a wind sweeping area A existsThus, the final output power of the entire wind power systemMaximum power captured by wind turbineIs positively correlated, and in a certain wind field, the fan parameters and the wind sweeping area thereofThe wind power and the air density are not changed, the main influence factors of the wind power are the wind speed and the air density, and the air density is mainly influenced by the air temperature, so that the main influence factors of the output power of the wind power plant are the wind speed and the air temperature;
step three: combining a short-time dynamic prediction model of the output power of the wind power plant of the particle swarm neural network, namely obtaining a BP-PSO hybrid neural network model, giving a group of input values and output values as training data, then respectively carrying out forward propagation and error reverse calculation, updating the weight and the deviation of the network along the gradient direction by calculating the error between a predicted value and an actual value, and stopping training until the maximum iteration number is met or the error meets the requirement;
step four: in order to overcome the defect that the BP neural network learning algorithm is easy to fall into local minimum, the performance of the BP neural network is optimized by adopting the particle swarm optimization. Firstly, a particle swarm algorithm is adopted to obtain an optimal initial weight and a threshold, then the values are assigned to an unoptimized BP neural network to obtain an optimized neural network, and finally, simulation data is used for evaluating the performance of the optimized neural network, wherein the specific process is as follows: wherein, the particle velocity update formula:
is the velocity of the particle;is a weighting coefficient, and the value is between 0.1 and 0.9;is an individual extremum;is the current position of the particle;is a global extremum;andknown as learning factors, in general2;Is a random number between (0, 1);the current iteration number is;is the total number of iterations that have been performed,andrespectively the maximum and minimum of the weighting coefficients.
Wherein, the particle position update formula:
step five: and dynamically predicting the wind power through a BP neural network algorithm optimized by particle swarm, an error discrimination function and dynamically adjusting model parameters.
Further, the uncertainty in the first step includes wind speed, wind direction, air pressure, temperature and humidity, and the conditional requirement for reasonably and effectively predicting the future wind power results from the uncertainty of the prediction process meeting specific conditions.
Further characterized in that: the specific calculation process of the error discriminant function in the step five is as follows: generally, byWhich is indicative of the measured data sequence,representing a predicted data sequence, the absolute error of the prediction is defined as follows:
however, the absolute error does not reflect the actual situation well in the prediction of the wind power plant, and is not representative in the judgment of the output power prediction precision of the wind power plant, and in order to more accurately judge the precision and the practical value of the prediction model, the feasibility and the effectiveness of each model are analyzed by adopting the following two error evaluation criteria. Namely the root mean square error and the average absolute error.
Further, the specific process of the root mean square error and the average absolute error is as follows:
root Mean Square Error (RMSE) refers to: square of deviation between predicted value and actual value and observation frequencyThe formula is as follows:
wherein the content of the first and second substances,is a rated power predicted value of the wind power plant,and the actual value of rated power of the wind power plant. When the predicted value is close to the actual value, the RMSE value reaches the minimum value of 0, and when the predicted value is far away from the actual value, the RMSE value reaches infinity, so that the smaller the RMSE value is, the better the RMSE value is in wind power prediction;
mean Absolute Error (MAE), which is the average of the absolute values of the deviations of all individual observations from the arithmetic mean, the evaluation of the mean magnitude of the prediction error by MAE, is defined as follows:
in the formula (I), the compound is shown in the specification,in order to predict the number of verification data,is a rated power predicted value of the wind power plant,and the actual value of rated power of the wind power plant. When the predicted value and the actual value are almost equal, the MAE value tends to 0, and tends to infinity along with the increase of the error, namely, the MAE value can linearly increase along with the increase of the error between the actual values of the predicted value, so that the smaller the MAE value is, the better the wind power prediction is.
Further, the specific process of the dynamic adjustment is as follows: and adaptively adjusting the model parameters according to the prediction error, and improving the prediction error caused by weakened correlation between the training data and the real-time measurement data.
The concrete process of the step five is as follows: the establishment of a dynamic prediction model of the output power of a wind power plant requires historical wind power data and wind power influence factor data corresponding to time points, the main influence factors of the wind power are wind speed and air temperature, the historical wind power can be provided by an energy management system in a microgrid, meteorological data such as the wind speed and the temperature come from a monitoring and data acquisition system arranged on a fan, and the meteorological data and the historical wind power are input into the dynamic prediction model to complete the dynamic prediction of the wind power, and the specific process is as follows:
firstly, wind speed is influenced by wind powerTemperature ofThe time sequence is input data, and historical wind power corresponding to the time point is used as the input dataIs output data, the model input-output relationship can be represented by equation (13),the corresponding functional relationship of the input and the output of the model is represented as follows:
the particle swarm optimization BP neural network wind power plant output power short-term dynamic prediction model specifically comprises the following steps:
1) inputting training sample data, and finishing the first training of model parameters by adopting a traditional BP-PSO mixed neural network;
2) wind speed of input prediction sample dataAnd temperaturePredicting wind power P(s) by using the trained neural network;
3) judging whether the output of the model reaches a set precision threshold value or not through an error discrimination function MAE;
4) if the average absolute error of the prediction output exceeds the set precision threshold, repeating the step 2), keeping the model unchanged, and continuing the prediction work of the next group of data; otherwise, when the prediction error exceeds the precision threshold, the step 5) is carried out;
5) and (3) dynamically adjusting model parameters, using the near data as a training set, and repeating the step 1) to train the model parameters again.
Compared with the prior art, the invention has the following advantages: the short-time dynamic prediction algorithm for the output power of the wind power plant combined with the particle swarm neural network provides a comprehensive monitoring and diagnosis means for the ultra-high voltage direct current transmission technology, the flexible transmission technology and the large-scale new energy grid connection, supports the positioning and fault removal of the power grid oscillation source, provides guarantee for the safe and stable operation of a large power grid and the large-scale application of the new energy, adopts the Prony algorithm to quickly extract the low-frequency component in a digital signal, has good real-time performance, high speed and accurate data, realizes the intelligent work of signal acquisition, filtering, low-frequency component extraction and the like by utilizing the edge computing technology, improves the efficiency of data processing at a station end, effectively reduces the computing pressure of a cloud end, provides a new modeling receiving end for the wind power prediction, can realize the high-precision prediction of the wind power prediction, and overcomes the defect that the BP neural network is easy to fall into a local minimum value, The method has the advantages that the method has the defects of low convergence speed, over-learning and the like, the performance of the model is greatly improved, the prediction precision and the practical value of the algorithm are more accurately judged by adopting the evaluation index of the average absolute error, the characteristics of the original data are screened by adopting the random forest model, the input characteristics of the screened prediction model contain characteristic information closely related to the wind power prediction, the dimensionality of the input characteristics is reduced, and the accuracy of the prediction model is effectively improved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a diagram of a flow chart dynamic prediction model of the present invention;
FIG. 3 is a wind power dynamic prediction model diagram of the present invention;
FIG. 4 is a graph of the average absolute error of the output power of the static BP-PSO hybrid neural network of the present invention;
FIG. 5 is a graph of the average absolute error of the output power of the static BP neural network of the present invention;
FIG. 6 is a graph of dynamic BP and BP-PSO model predicted power of the present invention;
FIG. 7 is a diagram of the predicted results of the A-group data wind power of the present invention;
FIG. 8 shows the prediction result of the B-group data wind power.
Detailed Description
The following examples are given for the detailed implementation and specific operation of the present invention, but the scope of the present invention is not limited to the following examples.
As shown in fig. 1 to 8, the present embodiment provides a technical solution: a short-time prediction algorithm for wind power plant output power combined with a particle swarm neural network comprises the following steps:
the method comprises the following steps: wind power prediction characteristic analysis is carried out, and analyzed data comprise uncertainty, conditionality and regionality of wind power;
wind is the most important factor influencing wind power, and the wind has the characteristics of randomness and the like, so that the wind power is strong in randomness. In addition, wind power is affected by temperature, humidity, air pressure and other factors. Due to the uncertainty and randomness of the factors, the uncertainty of the wind power is further increased. Therefore, the factors are analyzed, the characteristics of wind power prediction are summarized, and the method has important significance for selecting a wind power prediction model. The wind power prediction characteristics are as follows:
(1) uncertainty
The wind power output power of the wind power plant is influenced by factors such as wind speed, wind direction, air pressure, temperature and humidity, and the factors are in dynamic change at any moment. Due to the comprehensive influence of the complex factors, the wind power is unstable and variable. Therefore, one of the most important characteristics of wind power prediction is uncertainty.
(2) Conditional property
The conditionality of wind power prediction means that a prediction process is required to meet specific conditions in order to reasonably and effectively predict future wind power. Specifically, when the changes of the wind speed, the wind direction and other wind power influence factors in the historical time period and the time period to be predicted are not obvious and the evolution process is stable, the prediction accuracy of the wind power is ideal. When the interference is obvious and each influence factor has strong fluctuation, the prediction effect of the wind power is poor. In fact, however, although the wind farm is built in a region where the wind speed is relatively stable and various influencing factors are not changed strongly, various emergencies still occur, the uncertain events are difficult to predict accurately, and the output power of the wind farm cannot be predicted accurately at the moment. Therefore, these uncertainty issues should also be considered when predicting the wind farm output power.
(3) Regional property
In different regions, the wind characteristics can be greatly different, each influence factor can show great difference, and the output power of the wind power plant can also show strong regional characteristics;
step two: the wind power generation influence factor analysis method comprises the following specific analysis processes: in the whole wind power generation system, a fan converts the kinetic energy of air into mechanical energy on a shaft system of the fan, transmits the mechanical energy to a rotor shaft of a wind driven generator, and finally converts the mechanical energy into electric energy, which is the first link for realizing energy conversion;
wherein the content of the first and second substances,as a matter of time, the time is,in order to be the density of the air,to achieve the wind speed before entering the swept surface of the fan blades,the area of the wind is swept, and the wind speed is controlled,in order to be the quality of the air,is the air flow energy.
The fan paddle can have energy loss in the rotating process, namely not all wind energy can be captured by the wind turbine, and the wind energy captured by the wind turbine can be pushed out by combining the formula (1):
wherein the content of the first and second substances,to achieve the wind speed before entering the swept surface of the fan blades,is the wind speed after flowing out of the wind sweeping surface of the fan blade,is the utilization coefficient of the wind energy,indicating that the wind turbine is capable of deriving wind powerThe energy captured.
it can be known that whenThe wind energy utilization coefficient reaches the maximum valueMaximum power captured by wind turbineI.e. the power captured by the wind turbine is presentThus, the final output power of the entire wind power systemMaximum power captured by wind turbineAre positively correlated:
wherein the content of the first and second substances,for the wind energy captured by the wind turbine,for the final output power of the entire wind power system,for the maximum power captured by the wind turbine,the wind speed before entering the swept surface of the fan blade;
the existing wind power plants all adopt a maximum power tracking strategy, namely, the power captured by wind power generation as a wind sweeping area A existsThus, the final output power of the entire wind power systemMaximum power captured by wind turbineIs positively correlated, and in a certain wind field, the fan parameters and the wind sweeping area thereofThe wind power is not changed, the main influencing factors of the wind power are wind speed and air density, the air density is mainly influenced by air temperature,therefore, the main influencing factors of the wind farm output power are wind speed and air temperature;
step three: combining a short-time dynamic prediction model of the output power of the wind power plant of the particle swarm neural network, namely obtaining a BP-PSO hybrid neural network model, giving a group of input values and output values as training data, then respectively carrying out forward propagation and error reverse calculation, updating the weight and the deviation of the network along the gradient direction by calculating the error between a predicted value and an actual value, and stopping training until the maximum iteration number is met or the error meets the requirement;
bp (back propagation) neural networks are also known as error back propagation algorithms. The principle of the BP network is that a group of input values and output values are given as training data, forward propagation and error reverse calculation are respectively carried out, the error between a predicted value and an actual value is calculated, the weight and the deviation of the network are updated along the gradient direction, and the training is stopped until the maximum iteration times or the error meets the requirement. The BP neural network does not form an explicit mathematical model, but obtains an empirical model among data through self-learning and summarizing the rules among the data. The BP network is the most common model of the artificial neural network in practical application. The BP network has good performance in prediction, so that the wind power is predicted by using the BP network and optimized on the basis.
The structure of the BP neural network generally consists of multiple layers, an input layer, an output layer and a plurality of hidden layers. Each network layer is composed of a plurality of nodes, and the specific number is determined by the research problem. The nodes of two adjacent layers of networks are connected together, and the nodes of each layer do not have any relation. The increase of the number of the network hidden layers can improve the network operation precision and reduce the error of a network output result, but the increase of the number of the layers can greatly improve the operation time of an algorithm, increase the complexity of the network, select a proper excitation function, and under the condition that the excitation functions are continuous, the three-layer neural network can approximate any function. The information input by the BP network is transmitted in one way through an input layer, a hidden layer and an output layer, the nodes of two adjacent layers are connected together through the weight, and the weight is updated through the reverse propagation of the error of the output layer.
The establishment of the BP network is a process for converting a complex practical problem into a nonlinear optimization mathematical problem, and theoretically, the BP network can solve any nonlinear function and can achieve the required precision. The BP network can set any number of nodes of an input layer, an output layer and a hidden layer, can also set a plurality of hidden layers, uses a steepest descent method to achieve an expected result, can adjust the learning rate and can find an optimal value more quickly. In short, the BP network can be manually set with parameters and applied to different problems, and has strong flexibility. The advantages of the BP network are summarized specifically in the following aspects:
(1) non-linear mapping. A large number of researches prove that the BP network can approximate any nonlinear function by setting different network layer numbers and node numbers, and establish the nonlinear relation between the input layer and the output value, so that the relation between wind power related parameters and wind power can be established by utilizing the nonlinear mapping capability of the BP network, and the wind power can be predicted.
(2) The network has generalization. The BP network has the ability of predicting the future through the training of sample historical data, namely the BP network can be popularized to corresponding problems. For the wind power prediction problem, the wind power can be predicted by training historical operation data.
(3) The network can learn and adapt itself. The BP network can learn by itself, extract the main information of training sample data, calculate the error between the predicted value and the true value through forward and backward propagation, and modify the weight and the deviation between the network connection layers until the end condition is met.
(4) The network accuracy and stability are better. If the weight between the BP network connection nodes is small, the weight can be ignored, and the result is not greatly influenced. The wind power generation related parameters have no influence on the wind power prediction accuracy, and the wind power can be well predicted after being ignored.
The BP network is widely applied to many fields, and can well solve some practical problems, but has many disadvantages, for example, adjustment of network parameters does not have a corresponding theoretical basis, and most of the network parameters are summarized by some historical experiences or multiple experiments to obtain optimal parameters. The deficiencies of BP networks are mainly listed in the following aspects:
(1) training tends to fall into local optima. The BP algorithm is a nonlinear optimization process, the mean square error of a true value and a predicted value is used as a fitness function, a gradient descent method is adopted to search for a minimum value, the minimum value is searched along the gradient direction of a curved surface of an error function, and due to the fact that a plurality of curved surfaces exist, the found local minimum value is not necessarily a global optimum value, and the algorithm is trapped in a local optimum state. Therefore, the invention adopts the particle swarm optimization to enhance the global search capability and obtain the optimal weight and bias.
(2) The determination of the network structure is not guided by a particular theory. For a method that the number of layers and the number of nodes of a network are not determined, the nodes of an input layer and an output layer can be set according to specific research problems, the number of layers and the number of nodes of an implicit layer are usually determined by experience and experiments, and a complete theoretical basis is not provided. If the number of layers and the number of nodes are set to be too small, a proper nonlinear relation can not be found, and the error becomes large; if the setting is too large, the complexity of the network is increased, the training time is increased, the network is over-fitted, and the robustness is reduced.
(3) The convergence rate of the algorithm is slow. The learning rate of the classic BP network is a constant, which causes the training time to be longer, and under the condition that the nonlinear relation of the network is not clear and is more complex, the training time becomes longer.
In conclusion, the BP neural network is an effective self-learning neural network, has the capabilities of self-organization, self-learning, distributed storage and parallel processing of information and the like, and is widely applied to the aspects of intelligent control, knowledge engineering, pattern recognition and the like. Although BP networks have achieved some success in many areas, they still suffer from problems such as being prone to local minima; the convergence rate of the algorithm is low; "over-learning", etc.
The particle swarm optimization algorithm is an emerging random global optimization technology, is a simulation of migration and aggregation in the foraging process of a bird swarm, is proposed by Jim Kennedy in 1995 and is successfully used for function optimization, a global optimum is found by following a currently searched optimum value, each particle uses a current position, a current speed, a distance between the current position and the best position of the particle swarm, and a distance between the current position and the best position of the bird swarm, and the four information change the current position;
in order to overcome the defect that the BP neural network learning algorithm is easy to fall into local minimum, the performance of the BP neural network is optimized by adopting the particle swarm optimization. Firstly, a particle swarm algorithm is adopted to obtain an optimal initial weight and a threshold, then the values are assigned to an unoptimized BP neural network to obtain an optimized neural network, and finally, simulation data is used for evaluating the performance of the optimized neural network. Compared with the original BP neural network algorithm, the BP neural network algorithm improved by the particle swarm algorithm can be quickly converged to the target of neural network learning, and has higher calculation precision and faster convergence speed;
step four: in order to overcome the defect that the BP neural network learning algorithm is easy to fall into local minimum, the performance of the BP neural network is optimized by adopting the particle swarm optimization. Firstly, a particle swarm algorithm is adopted to obtain an optimal initial weight and a threshold, then the values are assigned to an unoptimized BP neural network to obtain an optimized neural network, and finally, simulation data is used for evaluating the performance of the optimized neural network, wherein the specific process is as follows: wherein, the particle velocity update formula:
is the velocity of the particle;is a weighting coefficient, and the value is between 0.1 and 0.9;is an individual extremum;is the current position of the particle;is a global extremum;andknown as learning factors, in general2;Is a random number between (0, 1);the current iteration number is;is the total number of iterations that have been performed,andrespectively the maximum and minimum of the weighting coefficients.
Wherein, the particle position update formula:
step five: and dynamically predicting the wind power through a BP neural network algorithm optimized by particle swarm, an error discrimination function and dynamically adjusting model parameters.
The uncertainty in the step one comprises wind speed, wind direction, air pressure, temperature and humidity, and the condition is that the uncertainty is caused when the prediction process is required to meet a specific condition in order to reasonably and effectively predict future wind power.
The method is characterized in that: the specific calculation process of the error discriminant function in the step five is as follows: generally, byWhich is indicative of the measured data sequence,representing a predicted data sequence, the absolute error of the prediction is defined as follows:
however, the absolute error does not reflect the actual situation well in the prediction of the wind power plant, and is not representative in the judgment of the output power prediction precision of the wind power plant, and in order to more accurately judge the precision and the practical value of the prediction model, the feasibility and the effectiveness of each model are analyzed by adopting the following two error evaluation criteria. Namely the root mean square error and the average absolute error.
The specific process of the root mean square error and the average absolute error is as follows:
root Mean Square Error (RMSE) refers to: square of deviation between predicted value and actual value and observation frequencyThe formula is as follows:
wherein the content of the first and second substances,is a rated power predicted value of the wind power plant,and the actual value of rated power of the wind power plant. When the predicted value is close to the actual value, the RMSE value reaches the minimum value of 0, and when the predicted value is far away from the actual value, the RMSE value reaches infinity, so that the smaller the RMSE value is, the better the RMSE value is in wind power prediction.
Mean Absolute Error (MAE), which is the average of the absolute values of the deviations of all individual observations from the arithmetic mean, the evaluation of the mean magnitude of the prediction error by MAE, is defined as follows:
in the formula (I), the compound is shown in the specification,in order to predict the number of verification data,for the rated power of the wind power plant, namely the wind energy power, when the predicted value and the actual value are almost equal, the MAE value tends to 0, and the MAE value tends to infinity along with the increase of the error, namely the MAE value can linearly increase along with the increase of the error between the actual values of the predicted value; therefore, in wind power prediction, the smaller the MAE value, the better.
The index MAE has no problem of positive and negative counteraction, is easy to calculate, and can better reflect the actual situation of predicted value errors. The method can be used for monitoring and predicting the long-term running state of the system and carrying out 'macroscopic' evaluation on the error characteristics of the system. The method adopts the MAE as an error discrimination function to realize the dynamic adjustment of the model training set.
The specific process of the dynamic adjustment is as follows: model parameters are adaptively adjusted according to the prediction error, and the prediction error caused by weakened relevance between training data and real-time measurement data is improved, so that higher prediction precision is achieved;
the wind speed has strong randomness and the temperature change is irregular under the influence of various factors, so that the wind power also shows strong randomness, the wind power is difficult to predict, and the prediction precision is difficult to satisfy. According to past experience, when the trained prediction model is used for prediction work, the prediction error gradually increases with the passage of time. The analysis shows that the relevance of the prediction data at the current stage and the previous training set is weakened, the data tracking cannot be well completed by the existing training model, and the prediction work at the current stage is not adapted any more. Therefore, the invention provides a dynamic adjustment strategy, which adaptively adjusts the model parameters according to the prediction error, improves the prediction error caused by weakened relevance between the training data and the real-time measurement data, and has higher prediction precision, as shown in fig. 2.
When the average prediction error of the model is smaller than the threshold set by the discriminant function, continuing to use the prediction model for prediction; and when the prediction error is larger than the error threshold value, dynamically adjusting the training set, using the nearby data as a training sample, retraining the network, and continuously predicting the wind power by using the trained network parameters.
The concrete process of the step five is as follows: the establishment of the dynamic prediction model of the output power of the wind power plant requires historical wind power data and wind power influence factor data of corresponding time points, and the main influence factors of the wind power are wind speed and air temperature. Historical wind power can be provided by an energy management system in the microgrid, and meteorological data such as wind speed and temperature come from a supervisory control and data acquisition (SCADA) system installed on a fan. The meteorological data and the historical wind power are input into the dynamic prediction model to complete the dynamic prediction of the wind power, and the dynamic prediction model of the wind power is shown in fig. 3:
the specific process is as follows:
firstly, wind speed is influenced by wind powerTemperature ofThe time sequence is input data, and historical wind power corresponding to the time point is used as the input dataIs output data, the model input-output relationship can be represented by equation (13),the corresponding functional relationship of the input and the output of the model is represented as follows:
the particle swarm optimization BP neural network wind power plant output power short-term dynamic prediction model specifically comprises the following steps:
1) inputting training sample data, and finishing the first training of model parameters by adopting a traditional BP-PSO mixed neural network;
2) wind speed of input prediction sample dataAnd temperaturePredicting wind power P(s) by using the trained neural network;
3) judging whether the output of the model reaches a set precision threshold value or not through an error discrimination function MAE;
4) if the average absolute error of the prediction output exceeds the set precision threshold, repeating the step 2), keeping the model unchanged, and continuing the prediction work of the next group of data; otherwise, when the prediction error exceeds the precision threshold, the step 5) is carried out;
5) and (3) dynamically adjusting model parameters, using the near data as a training set, and repeating the step 1) to train the model parameters again.
The method takes the 1-month wind speed, air temperature forecast data, actual fan output measurement data and other data of a certain wind power plant as sample data, and adopts the cycle of 15 minutes and the rated capacity of a wind turbine generator set of 850 kW. The invention explains the implementation process of the short-time dynamic prediction model of the wind power plant output power by combining the data, and the implementation process is concretely as follows.
(1) Wind power data preprocessing
1) Exception data handling
The data of the actual operation of the wind turbine generator set usually contains abnormal or missing data, the abnormal data mainly comes from abnormal conditions such as sensor acquisition errors, NWP system prediction errors, wind random fluctuation, wind abandonment and electricity limiting measures, equipment failure, shutdown and maintenance and the like, when an output power short-term prediction model is established, outliers are mixed in input data, the performance of the prediction model is greatly interfered, and therefore the original wind power data needs to be cleaned. Therefore, in order to keep the prediction accuracy of the wind power consistent with the power grid dispatching level, the original historical data is firstly derived from the SCADA system and then preprocessed. The abnormal data in the actual data sample of the wind power plant mainly comprise the following types:
the first type is a sample point with the wind power output power less than 0, and the processing method comprises the following steps: modifying the power value less than 0 to 0;
the second type is a sample point with the wind speed less than the starting wind speed power value and not zero, and the processing method comprises the following steps: modifying the power value to 0;
sample points with the third power value exceeding the rated power of the fan; the treatment method comprises the following steps: modifying the power value of the power converter into rated power;
the fourth type is that the power value of a sample point which causes discontinuous jump of power due to sudden change of wind speed is replaced by the average value of the output power of adjacent points.
2) Feature selection
The invention adopts a random forest method to select the parameter with the highest correlation with the wind power output power, and the algorithm is as follows:
the CART decision tree is applied to the random forest, the kini coefficient is an evaluation index applied to the work of selecting characteristics, in the decision tree model, the evaluation criterion of evaluating and selecting the kini coefficient is whether each sub-node can realize a purity peak value, and in popular terms: the observed values distributed in the decision tree model and having one sub-node are of the same type, and if the sub-node is at the purity peak value, the Gini coefficient is at the lowest point instead. For a normal decision tree, if there is a totalClass, sample belongs toThe probability of a class isThen the kini coefficient of the probability distribution is:
when the Gini coefficient is larger, the uncertainty is higher; when the kini coefficient is small, the uncertainty becomes lower and the data segmentation is more thorough. Since the CART tree is a binary tree, it can be expressed by the mathematical formula:
if we have seen the corresponding segmentation point of each feature, apply the featureHandle barIs divided into two parts, one isCan makeSet of samples that are true, others that are notSet of established samples. Is characterized inUnder the conditions ofThe Gini index of (A) is:
in the formulaRepresentation collectionThe uncertainty of (a) is determined,representation collectionIs indicative ofThroughUncertainty of the segmented set. The method comprises the following steps that each CART decision tree is formed in a random forest model through the following processes, in the first step, Gini coefficients are used for calculating all splitting points corresponding to all feature subsets in the CART decision tree, if the Gini coefficient of the splitting point is the minimum, the feature subset where the splitting point is located is the target of people, and the feature subset is further divided into two subsets until stopping requirements are met.
The historical data has 12 characteristics of wind speed, air pressure, wind direction, humidity, temperature, blade pitch angle, generator rotating speed, wind wheel rotating speed difference, cabin temperature and cabin direction, in order to further analyze the influence of the wind direction on the power, two characteristic parameters of wind direction sine and wind direction cosine are added, the total 14 characteristics are selected through the characteristics of random forests, the parameter with the highest correlation with the wind power generation output power is the wind speed, and then meteorological factors such as air temperature, air pressure, wind direction, humidity and the like exist. The wind power is less influenced by characteristic parameters such as the angle of the blade to be changed, the blade changing angle, the rotating speed of the generator, the rotating speed of the wind wheel, the rotating speed difference of the wind wheel, the temperature of the engine room, the direction of the engine room and the like. Therefore, the invention adopts 5 parameters of wind speed, air temperature, air pressure, wind direction and humidity as input characteristics of the prediction algorithm.
3) Data normalization
After the characteristic selection, characteristic parameters needing to be input in the process of constructing the wind power prediction algorithm comprise wind speed, wind direction, air temperature and air pressure, the input data have large value difference due to dimension, normalization processing is carried out on the data, the convergence rate of the algorithm is accelerated in the prediction process, and the modeling efficiency and the prediction precision are improved. The invention adopts a maximum and minimum normalization method to carry out normalization processing on input data:
(2) Error evaluation function parameter determination
The error evaluation function MAE herein is defined as follows:
in the formula:andrespectively the measured wind speed and the predicted wind speed;rated power of the wind power plant;verifying the number of data for prediction;in order to number the sequence of predicted points,can be used for measuring the dispersion degree of the error;the method is used for evaluating the error accuracy of the prediction model and reflects the error control capability of the prediction model on individuals.
(3) Setting a threshold value of an error evaluation function
The method takes the MAE as a criterion for setting the threshold of the error evaluation function to realize the dynamic adjustment of the model training set. When the threshold value of the discriminant function is set to be too large, effective dynamic adjustment cannot be carried out, and the expected precision requirement cannot be met; if the setting is too small, unnecessary adjustment of the model is caused, resulting in waste. In addition, the magnitude of the discrimination function accuracy threshold also has a certain influence on the prediction accuracy of the final result.
The MAE values of the prediction results obtained by using 100 sets of data as training samples and 300 sets of data as test samples and simulating by using a BP-PSO hybrid neural network model are shown in fig. 4. As can be seen from fig. 4, the MAE of most of the predicted power of the conventional static BP-PSO model is less than 0.02, the MAE value corresponding to a few time points is too large, the maximum value can reach 0.1, and the average value of the MAE calculated is 0.0156. On the basis of ensuring the prediction accuracy of the model, the accuracy threshold is selected with the aim of minimizing the dynamic adjustment times.
In an actual test, the threshold value is temporarily set to 0.010, the step length is 0.001, the threshold value is gradually increased to 0.021, 300 groups of data are predicted by adopting a BP-PSO hybrid neural network model after dynamic optimization, and the optimal precision threshold value is searched in the process. As is clear from table 1, the most desirable prediction effect is obtained when the accuracy threshold is 0.015.
TABLE 1 MAE and number of dynamic adjustments of samples with different precision thresholds
Threshold of accuracy | 0.01 | 0.011 | 0.012 | 0.013 | 0.014 | 0.015 |
MAE% | 0.8457 | 0.8732 | 0.9120 | 0.9226 | 0.9254 | 0.9465 |
Number of times of adjustment | 63 | 55 | 49 | 40 | 32 | 28 |
Threshold of accuracy | 0.016 | 0.017 | 0.018 | 0.019 | 0.020 | 0.0121 |
MAE% | 0.9831 | 1.174 | 1.194 | 1.198 | 1.200 | 1.219 |
Number of times of adjustment | 26 | 23 | 21 | 21 | 19 | 18 |
(4) Dynamic BP and dynamic BP-PSO
The BP neural network has better nonlinearity and self-learning capability, but the defects of slow convergence speed, easy oscillation, easy falling into local minimum values and the like are not negligible. The average absolute error of the prediction result obtained by using 100 sets of data as training samples and 300 sets of data as test samples and simulating by using a static BP neural network model is shown in fig. 5, and the average value of the MAE obtained by calculation is 0.0647, and the maximum value is 0.1355.
And selecting the average value 0.065 of the MAE as the precision threshold value of the introduced dynamic BP neural network, and taking 0.015 as the dynamic threshold value of the BP-PSO hybrid neural network according to the analysis result of the 3.2 section. And (4) carrying out comparative analysis on the prediction performance of the BP neural network model and the BP-PSO hybrid neural network model after introducing a dynamic adjustment strategy. The predicted power data and corresponding measured power data for the two dynamic prediction models are shown in fig. 6, and the average absolute error is shown in table 2.
TABLE 2 average absolute error of predicted power for dynamic BP and BP-PSO models
Mean absolute error MAE | Dynamic BP neural network model | Dynamic BP-PSO hybrid neural network model | BP-PSO reduction amplitude compared with BP |
Mean value of | 0.0367 | 0.0096 | 73.84% |
Maximum value | 0.7204 | 0.2163 | 69.98% |
As can be seen from fig. 5 and table 2, the dynamic BP-PSO hybrid neural network model has more excellent performance, the predicted wind power is closer to the actual output power of the wind farm, and the average value and the maximum value of the average absolute error MAE are reduced by 73.84% and 69.98% respectively compared with the dynamic BP neural network model.
(5) Model dynamic adjustment
The wind speed is influenced by various factors and has strong randomness, so that the wind power also shows strong randomness. Therefore, the data A with small variation amplitude and the data B with large variation amplitude which are close to the rated electric power of the wind power plant are respectively adopted for analysis and research. And comparing the predicted performance of the BP-PSO hybrid neural network model before and after introducing the dynamic adjustment strategy. The output power of the wind power plant is difficult to keep near the rated power for a long time under the influence of wind speed, and in the sample data adopted in the method, the maximum duration of the group A data meeting the condition is 240 groups of data in 40 hours. The wind power prediction results of the data of the group A and the data of the group B under the dynamic and static BP-PSO hybrid neural network models are respectively shown in fig. 7 and fig. 8.
The prediction errors of the two sets of data were analyzed, and the results are shown in tables 3 and 4. The average value and the maximum value of the relative errors of the group A with small wind power change in the PSO optimized BP neural network model after the dynamic adjustment of the training set are reduced by 52.00 percent and 49.08 percent respectively. The average value and the maximum value of the relative error of the B group with larger variation amplitude are respectively reduced by 64.97 percent and 54.06 percent after the dynamic adjustment training set is adopted. In addition, after the BP-PSO hybrid neural network model introduces a dynamic adjustment strategy, the root mean square error of A, B groups of predicted power is reduced by 50.59% and 63.20%, respectively.
At A, B, the dynamic simulation process of two data sets includes recording whether the sample data in each set has adjusted training set, retraining model parameters, and predicting time. The average prediction time of each group of data is 0.0465s when the dynamic adjustment condition does not occur, and the average prediction time is 0.547s when the dynamic adjustment condition does occur; the group A is dynamically adjusted for 13 times, and the parameters of the prediction model are dynamically adjusted once every 185min on average, so that the total time is 17.68 s; group B was dynamically adjusted 35 times, averaging the predictive model parameters once every 87min, which took 27.58s total.
As can be seen by combining fig. 7 and 8, and tables 3 and 4, after the dynamic adjustment strategy is introduced, the prediction accuracy of the wind power plant output power of the BP-PSO hybrid neural network model is obviously improved; the prediction results of the two groups of data are compared, and the optimization effect of the prediction precision of the B group of data with large wind power fluctuation amplitude after dynamic prediction is adopted is more obvious. The short-term dynamic prediction model of the BP neural network wind power plant output power, which adopts particle swarm optimization, can be continuously used for a period of time after dynamic adjustment is completed each time, has sustainability, the dynamic adjustment time is negligible compared with the prediction sample time, the time requirement is met, and certain availability is achieved.
TABLE 3A group data wind power prediction error
TABLE 4B group data wind power prediction error
The invention provides a short-time prediction algorithm for the output power of a wind power plant combined with a particle swarm neural network, and model parameters are adaptively adjusted according to an average absolute error. Firstly, the prediction performances of the BP neural network model and the BP-PSO hybrid neural network model after introducing a dynamic adjustment strategy are compared, and the superiority of the dynamic BP-PSO model is verified. The dynamic prediction model has higher prediction precision, and the relative error and the root mean square error are obviously improved; in addition, the larger the fluctuation range of the wind power is, the weaker the relevance between the prediction at the present stage and a previous training set is, and the conventional static prediction model cannot well track the change of the actual wind power.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (6)
1. A short-time prediction algorithm for wind power plant output power combined with a particle swarm neural network is characterized by comprising the following steps:
the method comprises the following steps: wind power prediction characteristic analysis is carried out, and analyzed data comprise uncertainty, conditionality and regionality of wind power;
step two: the wind power generation influence factor analysis method comprises the following specific analysis processes: in the whole wind power generation system, a fan converts the kinetic energy of air into mechanical energy on a shaft system of the fan, transmits the mechanical energy to a rotor shaft of a wind driven generator, and finally converts the mechanical energy into electric energy, which is the first link for realizing energy conversion;
wherein the content of the first and second substances,as a matter of time, the time is,in order to be the density of the air,to achieve the wind speed before entering the swept surface of the fan blades,the area of the wind is swept, and the wind speed is controlled,in order to be the quality of the air,is air flow energy;
the fan paddle can have energy loss in the rotating process, namely not all wind energy can be captured by the wind turbine, and the wind energy captured by the wind turbine can be pushed out by combining the formula (1):
wherein the content of the first and second substances,to achieve the wind speed before entering the swept surface of the fan blades,is the wind speed after flowing out of the wind sweeping surface of the fan blade,is the utilization coefficient of the wind energy,indicating that the wind turbine is capable of deriving wind powerThe energy obtained;
it can be known that whenThe wind energy utilization coefficient reaches the maximum valueMaximum power captured by wind turbineI.e. the power captured by the wind turbine is presentThus, the final output power of the entire wind power systemMaximum power captured by wind turbineAre positively correlated:
wherein the content of the first and second substances,for the wind energy captured by the wind turbine,for the final output power of the entire wind power system,for the maximum power captured by the wind turbine,the wind speed before entering the swept surface of the fan blade;
the existing wind power plants all adopt a maximum power tracking strategy, namely, the power captured by wind power generation as a wind sweeping area A existsThus, the final output power of the entire wind power systemMaximum power captured by wind turbineIs positively correlated, and in a certain wind field, the fan parameters and the wind sweeping area thereofThe wind power and the air density are not changed, the main influence factors of the wind power are the wind speed and the air density, and the air density is mainly influenced by the air temperature, so that the main influence factors of the output power of the wind power plant are the wind speed and the air temperature;
step three: combining a short-time dynamic prediction model of the output power of the wind power plant of the particle swarm neural network, namely obtaining a BP-PSO hybrid neural network model, giving a group of input values and output values as training data, then respectively carrying out forward propagation and error reverse calculation, updating the weight and the deviation of the network along the gradient direction by calculating the error between a predicted value and an actual value, and stopping training until the maximum iteration number is met or the error meets the requirement;
step four: in order to improve the defect that a BP neural network learning algorithm is easy to fall into local minimum, the performance of the BP neural network is optimized by adopting a particle swarm algorithm, firstly, the particle swarm algorithm is adopted to obtain the optimal initial weight and threshold, then, the values are assigned to the non-optimized BP neural network to obtain the optimized neural network, and finally, the performance of the optimized neural network is evaluated by using simulation data, wherein the specific process is as follows: wherein, the particle velocity update formula:
is the velocity of the particle;is a weighting coefficient, and the value is between 0.1 and 0.9;is an individual extremum;is the current position of the particle;is a global extremum;andknown as learning factors, in general2;Is a random number between (0, 1);the current iteration number is;is the total number of iterations that have been performed,andrespectively the maximum and minimum values of the weighting coefficients,
wherein, the particle position update formula:
step five: and dynamically predicting the wind power through a BP neural network algorithm optimized by particle swarm, an error discrimination function and dynamically adjusting model parameters.
2. The wind farm output power short-time prediction algorithm combining the particle swarm neural network according to claim 1, characterized in that: the uncertainty in the first step comprises wind speed, wind direction, air pressure, temperature and humidity, and the conditionality is that the future wind power is reasonably and effectively predicted, and the prediction process is required to meet specific conditional uncertainty.
3. The wind farm output power short-time prediction algorithm combining the particle swarm neural network according to claim 1, characterized in that: the specific calculation process of the error discriminant function in the step five is as follows: generally, byWhich is indicative of the measured data sequence,representing a predicted data sequence, the absolute error of the prediction is defined as follows:
however, the absolute error does not well reflect the actual situation in the prediction of the wind power plant, and is not representative in the judgment of the output power prediction precision of the wind power plant, and in order to more accurately judge the precision and the practical value of the prediction model, the feasibility and the effectiveness of each model, namely the root mean square error and the average absolute error, are analyzed by adopting the following two error evaluation standards.
4. The wind farm output power short-time prediction algorithm combining the particle swarm neural network according to claim 3, characterized in that: the specific process of the root mean square error and the average absolute error is as follows:
root Mean Square Error (RMSE) refers to: square of deviation between predicted value and actual value and observation frequencyThe formula is as follows:
wherein the content of the first and second substances,is a rated power predicted value of the wind power plant,the method is characterized in that the method is a rated power actual value of the wind power plant, when a predicted value is close to the actual value, the RMSE value can reach the minimum value of 0, and when the predicted value is far away from the actual value, the RMSE value reaches infinity, so that the smaller the RMSE value is, the better the RMSE value is in wind power prediction;
mean Absolute Error (MAE), which is the average of the absolute values of the deviations of all individual observations from the arithmetic mean, the evaluation of the mean magnitude of the prediction error by MAE, is defined as follows:
in the formula (I), the compound is shown in the specification,in order to predict the number of verification data,is a rated power predicted value of the wind power plant,for a rated power actual value of a wind power plant, when a predicted value and an actual value are almost equal, the MAE value tends to 0, and with the increase of errors, the MAE value tends to infinity, namely with the increase of errors between the actual values of the predicted values, the MAE value can be linearly increased, so that in wind power prediction, the smaller the MAE value is, the better the MAE value is.
5. The wind farm output power short-time prediction algorithm combining the particle swarm neural network according to claim 1, characterized in that: the specific process of the dynamic adjustment is as follows: and adaptively adjusting the model parameters according to the prediction error, and improving the prediction error caused by weakened correlation between the training data and the real-time measurement data.
6. The wind farm output power short-time prediction algorithm combining the particle swarm neural network according to claim 1, characterized in that: the concrete process of the step five is as follows: the establishment of a dynamic prediction model of the output power of a wind power plant requires historical wind power data and wind power influence factor data corresponding to time points, the main influence factors of the wind power are wind speed and air temperature, the historical wind power can be provided by an energy management system in a microgrid, meteorological data such as the wind speed and the temperature come from a monitoring and data acquisition System (SCADA) installed on a fan, and the meteorological data and the historical wind power are input into the dynamic prediction model to complete the dynamic prediction of the wind power, and the specific process is as follows:
firstly, wind speed is influenced by wind powerTemperature ofThe time sequence is input data, and historical wind power corresponding to the time point is used as the input dataIs output data, the model input-output relationship can be represented by equation (13),the corresponding functional relationship of the input and the output of the model is represented as follows:
the particle swarm optimization BP neural network wind power plant output power short-term dynamic prediction model specifically comprises the following steps:
1) inputting training sample data, and finishing the first training of model parameters by adopting a traditional BP-PSO mixed neural network;
2) wind speed of input prediction sample dataAnd temperaturePredicting wind power P(s) by using the trained neural network;
3) judging whether the output of the model reaches a set precision threshold value or not through an error discrimination function MAE;
4) if the average absolute error of the prediction output exceeds the set precision threshold, repeating the step 2), keeping the model unchanged, and continuing the prediction work of the next group of data; otherwise, when the prediction error exceeds the precision threshold, the step 5) is carried out;
5) and (3) dynamically adjusting model parameters, using the near data as a training set, and repeating the step 1) to train the model parameters again.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210256315.7A CN114386718A (en) | 2022-03-16 | 2022-03-16 | Wind power plant output power short-time prediction algorithm combined with particle swarm neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210256315.7A CN114386718A (en) | 2022-03-16 | 2022-03-16 | Wind power plant output power short-time prediction algorithm combined with particle swarm neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114386718A true CN114386718A (en) | 2022-04-22 |
Family
ID=81206118
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210256315.7A Pending CN114386718A (en) | 2022-03-16 | 2022-03-16 | Wind power plant output power short-time prediction algorithm combined with particle swarm neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114386718A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116470496A (en) * | 2023-04-24 | 2023-07-21 | 广东永光新能源设计咨询有限公司 | Renewable energy power generation system |
CN116526478A (en) * | 2023-07-03 | 2023-08-01 | 南昌工程学院 | Short-term wind power prediction method and system based on improved snake group optimization algorithm |
CN116523349A (en) * | 2023-05-19 | 2023-08-01 | 北京协合运维风电技术有限公司 | Wind power station reliability analysis method and system |
CN118040680A (en) * | 2024-04-11 | 2024-05-14 | 国家能源集团陕西电力有限公司吉山梁电厂 | Wind power medium-short term prediction method and system |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104806450A (en) * | 2015-03-25 | 2015-07-29 | 华北电力大学(保定) | Universal gravitation neural network based wind power system MPPT control method |
CN105512745A (en) * | 2015-11-05 | 2016-04-20 | 天津大学 | Wind power section prediction method based on particle swarm-BP neural network |
CN107169612A (en) * | 2017-06-21 | 2017-09-15 | 南京信息工程大学 | The prediction of wind turbine active power and error revising method based on neutral net |
CN109118384A (en) * | 2018-07-16 | 2019-01-01 | 湖南优利泰克自动化***有限公司 | A kind of Wind turbines healthy early warning method |
CN109856969A (en) * | 2018-11-06 | 2019-06-07 | 皖西学院 | A kind of failure prediction method and forecasting system based on BP neural network model |
CN110991743A (en) * | 2019-12-05 | 2020-04-10 | 河海大学 | Wind power short-term combination prediction method based on cluster analysis and neural network optimization |
CN113570120A (en) * | 2021-07-07 | 2021-10-29 | 上海电机学院 | Wind power prediction method and device based on improved particle swarm optimization |
-
2022
- 2022-03-16 CN CN202210256315.7A patent/CN114386718A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104806450A (en) * | 2015-03-25 | 2015-07-29 | 华北电力大学(保定) | Universal gravitation neural network based wind power system MPPT control method |
CN105512745A (en) * | 2015-11-05 | 2016-04-20 | 天津大学 | Wind power section prediction method based on particle swarm-BP neural network |
CN107169612A (en) * | 2017-06-21 | 2017-09-15 | 南京信息工程大学 | The prediction of wind turbine active power and error revising method based on neutral net |
CN109118384A (en) * | 2018-07-16 | 2019-01-01 | 湖南优利泰克自动化***有限公司 | A kind of Wind turbines healthy early warning method |
CN109856969A (en) * | 2018-11-06 | 2019-06-07 | 皖西学院 | A kind of failure prediction method and forecasting system based on BP neural network model |
CN110991743A (en) * | 2019-12-05 | 2020-04-10 | 河海大学 | Wind power short-term combination prediction method based on cluster analysis and neural network optimization |
CN113570120A (en) * | 2021-07-07 | 2021-10-29 | 上海电机学院 | Wind power prediction method and device based on improved particle swarm optimization |
Non-Patent Citations (1)
Title |
---|
柯亭婧: "风电场输出功率短期动态预测及其应用研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》, no. 6, 15 June 2016 (2016-06-15), pages 042 - 341 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116470496A (en) * | 2023-04-24 | 2023-07-21 | 广东永光新能源设计咨询有限公司 | Renewable energy power generation system |
CN116523349A (en) * | 2023-05-19 | 2023-08-01 | 北京协合运维风电技术有限公司 | Wind power station reliability analysis method and system |
CN116523349B (en) * | 2023-05-19 | 2024-01-23 | 北京协合运维风电技术有限公司 | Wind power station reliability analysis method and system |
CN116526478A (en) * | 2023-07-03 | 2023-08-01 | 南昌工程学院 | Short-term wind power prediction method and system based on improved snake group optimization algorithm |
CN116526478B (en) * | 2023-07-03 | 2023-09-19 | 南昌工程学院 | Short-term wind power prediction method and system based on improved snake group optimization algorithm |
CN118040680A (en) * | 2024-04-11 | 2024-05-14 | 国家能源集团陕西电力有限公司吉山梁电厂 | Wind power medium-short term prediction method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114386718A (en) | Wind power plant output power short-time prediction algorithm combined with particle swarm neural network | |
Wang et al. | Wind power interval prediction based on improved PSO and BP neural network | |
CN106529719B (en) | Wind power prediction method based on particle swarm optimization algorithm wind speed fusion | |
CN110717610B (en) | Wind power prediction method based on data mining | |
CN111414717A (en) | XGboost-L ightGBM-based unit power prediction method | |
Zhu et al. | A prediction model for wind farm power generation based on fuzzy modeling | |
CN104299044A (en) | Clustering-analysis-based wind power short-term prediction system and prediction method | |
CN105512766A (en) | Wind power plant power predication method | |
CN105574615A (en) | Spatial correlation and genetic algorithm (GA) based wind power forecast method for wavelet-BP neural network | |
CN113555881A (en) | Wind power plant frequency modulation fan sequencing method | |
CN110991701A (en) | Wind power plant fan wind speed prediction method and system based on data fusion | |
CN104732300A (en) | Neural network wind power short-term forecasting method based on fuzzy partition theory | |
CN116885691B (en) | Wind power climbing event indirect prediction method | |
Eseye et al. | Short-term wind power forecasting using a double-stage hierarchical hybrid GA-ANFIS approach | |
Wang et al. | Ultra-short-term wind power forecasting based on deep belief network | |
Chen et al. | Research on wind power prediction method based on convolutional neural network and genetic algorithm | |
CN113505909B (en) | Error compensation method for short-term wind power trend prediction | |
Jiaran et al. | Intelligent combined prediction of wind power based on numerical weather prediction and fuzzy clustering | |
CN113988421A (en) | Wind turbine generator power intelligent prediction method based on deep learning | |
Asghar et al. | Online estimation of wind turbine tip speed ratio by adaptive neuro-fuzzy algorithm | |
CN115965134A (en) | Regional power grid wind power generation power prediction optimization method | |
Yuan et al. | Short-term wind speed forecasting using STLSSVM hybrid model | |
CN116154875A (en) | Photovoltaic power station active power optimization distribution method based on TCN and error function | |
CN111178601B (en) | Wind turbine generator power prediction method based on meteorological data post-processing | |
CN112668764B (en) | Offshore wind farm energy storage system optimization configuration method based on cloud model and FCM algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |