CN109978284A - Photovoltaic power generation power time-sharing prediction method based on hybrid neural network model - Google Patents
Photovoltaic power generation power time-sharing prediction method based on hybrid neural network model Download PDFInfo
- Publication number
- CN109978284A CN109978284A CN201910337078.5A CN201910337078A CN109978284A CN 109978284 A CN109978284 A CN 109978284A CN 201910337078 A CN201910337078 A CN 201910337078A CN 109978284 A CN109978284 A CN 109978284A
- Authority
- CN
- China
- Prior art keywords
- sample
- value
- neural network
- training
- data
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000010248 power generation Methods 0.000 title claims abstract description 18
- 238000003062 neural network model Methods 0.000 title abstract description 4
- 238000012549 training Methods 0.000 claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 210000002569 neuron Anatomy 0.000 claims description 18
- 241000131894 Lampyris noctiluca Species 0.000 claims description 11
- 230000015572 biosynthetic process Effects 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000003786 synthesis reaction Methods 0.000 claims description 6
- 238000004519 manufacturing process Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 3
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 210000005036 nerve Anatomy 0.000 claims 1
- 241000254158 Lampyridae Species 0.000 abstract 1
- 230000003287 optical effect Effects 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 6
- 238000012360 testing method Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000007796 conventional method Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 230000019771 cognition Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 229910052751 metal Inorganic materials 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Human Resources & Organizations (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Strategic Management (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Water Supply & Treatment (AREA)
- Entrepreneurship & Innovation (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a hybrid neural network model-based time-sharing prediction method for optical method generated power, which comprises the steps of obtaining historical data of a photovoltaic power station; performing comprehensive clustering grouping on historical meteorological data; according to grouping results, constructing the historical data into a plurality of training sets; establishing a neural network prediction model, and training a neural network prediction model in each training set; acquiring real meteorological data and dividing into groups; and predicting the photovoltaic power generation power by using the prediction model trained by the grouping. The method establishes a prediction model aiming at each specific training set, so that the accuracy and precision of prediction are greatly improved, the parameters of the neural network are optimized by utilizing a mixed algorithm of a differential algorithm and a firefly algorithm, and meanwhile, the model has higher robustness, is easier to obtain a global optimum value and has stronger solution optimizing capability by adopting the combination of branch evolution and global evolution.
Description
Technical field
The invention belongs to photovoltaic power generation power prediction fields, and in particular to a kind of photovoltaic based on hybrid production style
Generated output timesharing prediction technique.
Background technique
The principle of photovoltaic power generation is photovoltaic effect, so-called photovoltaic effect refer to illumination make inhomogeneos semiconductor or metal with
A kind of phenomenon of potential difference is generated between the different parts that semiconductor combines.On the one hand, it is electronics, luminous energy that it, which is by converting photons,
The phenomenon that being converted into electric energy；On the other hand, it forms voltage, there is voltage, if be connected between the two, can form electric current
Circuit.It follows that the activity of photovoltaic power generation and the sun is closely bound up, it is protected from environmental very greatly, daytime, generated energy was big,
At night almost without generated energy.Also, the influence of the factors such as generated energy climate and region is more obvious, has Diurnal Variation
And climatic change characteristics；Height above sea level, weather, irradiation intensity are different, and photovoltaic power generation quantity gap is accordingly bigger.Not with forecasting wind speed
With the prediction of photovoltaic generation power has apparent periodicity and timeliness.If not considering that the time of photovoltaic power generation is special
Property, then then granularity is too big to the power prediction that it is carried out, precision and accuracy are difficult to reach current demand, as a result, this hair
The bright physical characteristic for photovoltaic power generation proposes a kind of photovoltaic generation power timesharing prediction based on hybrid production style
The thought of method, timesharing is, original whole learning sample is grouped on time, the sample of grouping is more accurate, prediction
Precision is higher, but in the actual moving process of photovoltaic power generation, the time is only only intended to a reference of sample packet
Standard, not absolute standard, because of such as two days synchronizations, but since one day is the cloudy day, another day is fine day, then this two
A sample, which is obtained a result, to differ greatly, so then to need the grouping that sample is more corresponded to reality to sample data
It is clustered, similar sample is divided into one kind, with such sample training model, to obtain the predicted value of such data,
Higher accuracy could be obtained in this way.Prediction technique proposed by the present invention as a result, on the one hand, learning sample is carried out fine
Change pretreatment, fining training is carried out to prediction model, on the other hand, neural network model is improved, its training is enhanced
The ability of gain of parameter optimized parameter and time, as a result, on the one hand method proposed by the present invention could improve the standard of prediction model
On the other hand exactness also can guarantee and calculate the time in the controllable range of a reality.
Summary of the invention
It is a primary object of the present invention to be grouped the influence for portraying external environment to photovoltaic power generation power prediction, to obtain
More fine-grained prediction model provides prediction accuracy and the higher method of precision, and to achieve the above object, the present invention provides
Following technical solution, comprising the following steps:
Step 1, the historical data of photo-voltaic power generation station is obtained, the historical data includes with history meteorological data and power generation
Power data；
Step 2, comprehensive Clustering is carried out to the history meteorological data；
Step 3, by group result, the historical data is constructed into several training sets in groups；
Step 4, neural network prediction model is established, input layer, hidden layer and output layer number is determined, utilizes modern optimization
Algorithm is trained the parameter in prediction model, each training set trains a neural network prediction model；
Step 5, real meteorological data is obtained, real meteorological data is divided according to the synthesis cluster result and is grouped；
Step 6, according to group result, the prediction model gone out using the station work carries out photovoltaic generation power pre-
It surveys.
Modern optimization algorithm described in step 4 is the hybrid algorithm of difference algorithm and glowworm swarm algorithm, to prediction model
In parameter be trained the following steps are included:
Step 401, the neuron of hidden layer in the BP neural network is sequentially divided into two groups, by each group of hidden layer
The relevant parameter to be optimized of neuron is divided into one；
Step 402, branch's training is carried out, including,
It step 40201, by every parameter initialization to be optimized is respectively random value, remaining parameter to be optimized is set as solid
Definite value thus generates two random populations；
Step 40202, two populations are updated with glowworm swarm algorithm respectively；
Step 40203, difference operator operation, including selection, intersection and variation are carried out to two populations respectively, updated respectively
The population；
Step 40204, two training is allowed to reach preset maximum number of iterations or reach designated precision respectively；
Step 403, global training is carried out, including
Step 40301, half individual optimal in two populations is chosen respectively, by optimal one of two populations
Half individual carries out crossover operation, merges into a population；
Step 40302, a population is updated with glowworm swarm algorithm；
Step 40303, difference operator operation, including selection, intersection and variation are carried out to a population, respectively more
The new population；
Step 40304, global training is allowed to reach preset maximum number of iterations or reach designated precision；
Step 404, optimal individual is selected, the parameter of the optimal value of each parameter and the threshold value of output layer neuron is calculated
Value.
Synthesis Clustering described in step 2 be the distance between comprehensive consideration sample value and angle value and determination it is poly-
Class method, the similarity between sample indicate are as follows:
Wherein, Max (d (X, Y)) indicates the maximum manhatton distance in sample set between two samples, α be it is preset can
Adjust parameter, for adjust distance metric and angle measurement between weighted value, the manhatton distance be expressed as d (x, y)=|
x1-y1|+|x2-y2|+…+|xp-yp|, x and y are two samples, x1,x2,…xpFor each attribute value in sample x, y1,
y2,…,ypFor each attribute value in sample y, there is p attribute in each sample,Wherein,
| | x | | it is sample vector x=(x1,x2,…xp) Euclid norm, | | y | | be sample vector y=(y1,y2,…yp) Europe
Norm is obtained in several.
The step of synthesis Clustering described in step 2 includes:
Step 201: in historical climate data sample, selecting k point at random as cluster centre, C={ c1,c2,…
ck}；
Step 202: the sample in traversal sample set, if x and cluster centre ciSimilarity be greater than it arrive cluster centre cj
Similarity, then x is just divided into ciIn class；
Step 203: obtaining in each classification, two-by-two the similarity matrix simi (x, y) of sample；
Step 204: calculating the sum of the similar value of each sample and other samples in the category in each classification, obtain most
Big similar value and, the sample of phase Sihe value maximum in the category is redefined as cluster centre；
Step 202,203 and 204 are repeated, until cluster centre does not change, or until reaching maximum number of iterations；
Dividing grouping to real meteorological data described in step 5 includes calculating separately real meteorological data sample xtWith
The similarity of the k cluster centres, by xtIt is divided into the maximum class group of similar angle value；
In step 6 using the real meteorological data divide prediction model that station work goes out to photovoltaic generation power into
Row prediction.
The history meteorological data includes irradiation intensity, temperature, humidity and air pressure, the input layer quantity
For 4, hidden layer neuron quantity be 8 and output layer neuron quantity is 1.
0.5 is set by the α in step 2, sets 0.5 for the fixed value in step 40201, the k in cluster is arranged
It is 4.
Prediction technique proposed by the present invention is improved conventional method in terms of two, first, is clustered using comprehensive
Method fully considers the similitude of distance and angle, for more suitably being classified to data, is divided using classification data
Class model training so that model have stronger predictive ability, second, utilize difference algorithm and glowworm swarm algorithm mixing calculate
Method optimizes the parameter of neural network, while being evolved using branch and being combined with global evolution, so that model has more Shandong
Stick, it is easier to obtain global optimum.
Detailed description of the invention
Fig. 1 is overall flow schematic diagram of the invention.
Specific embodiment
The present invention will be further described below with reference to the drawings, but the present invention is limited in any way, base
In present invention teach that it is made it is any transform or replace, all belong to the scope of protection of the present invention.
In practical application, the factor for influencing photovoltaic power station power generation power is very more, there is geographical factor, such as precision, latitude
, also there are factor of equipment, such as setting angle, equipment energy consumption, the transfer efficiency of tabula rasa etc. in degree, height above sea level, position etc., but weather
Factor, such as intensity of illumination, incident angle, temperature, air pressure etc., in many factors, what is had can quantify and measure,
Have and be difficult to quantify really, so in actual prediction, it is not necessary that all factors are considered, only need will wherein influence compared with
Big factor consider it is clear just it is big otherwise just to will appear measurement difficulty during prediction, calculate time explosion and increase etc. and ask
Topic.In fact, influence of the climatic factor to the generated output of photovoltaic plant the most also influences maximum.Photovoltaic generating system
Output power with the number of solar irradiation and variation, due to the variation of the meteorological conditions such as temperature, humidity, air pressure, wind speed, cloud layer
Uncertainty, so that the variation of solar power station power, which is also affected by it, generates randomness variation.For specific photovoltaic plant,
Its geographic factor and apparatus factor are basicly stable constant, thus only consider in the present invention by meteorological data to photovoltaic power generation
The influence of power.
Since photovoltaic generation power has many characteristics, such as very strong time, place, weather, restricted very by external environment
Obviously.If predicting the data in evening with the training pattern in morning, precision of prediction is obviously relatively low；It is similar, if with the cloudy day
Training pattern predicts the data of fine day, and precision of prediction is also very low.Since Generalization Ability of Neural Network is strong, there is very strong return
It receives summary ability, there is good prediction effect to the input variable with same characteristic features.Thus the present invention is proposed with suitable
Method input variable is grouped, then predicted by establishing grouping model, then available better precision of prediction.
There are many kinds of the methods of grouping, can be based on time point based on seasonality, can also be with the closely located degree of sample, and traditional presses
When the method granularity that is grouped it is excessively thick, so the present invention proposes a kind of grouping strategy of comprehensive consideration sample attribute, this grouping
Method can allow the grouping of sample with more practical directive significance, then establish its proprietary model respectively for each group, comprehensive
Each group of prediction result obtains the prediction result of final total data.Sample Similarity compared with conventional model, in group
It is higher, therefore each model is stronger to the inducing ability of this group of data, fitting precision is higher, and model is more accurate.
As shown in Figure 1, a kind of photovoltaic generation power timesharing based on hybrid production style of the embodiment of the present invention is pre-
Survey method, includes the following steps:
Step 1, the historical data of photo-voltaic power generation station is obtained, the historical data includes with history meteorological data and power generation
Power data；
Step 2, comprehensive Clustering is carried out to the history meteorological data；
Step 3, by group result, the historical data is constructed into several training sets in groups；
Step 4, neural network prediction model is established, input layer, hidden layer and output layer number is determined, utilizes modern optimization
Algorithm is trained the parameter in prediction model, each training set trains a neural network prediction model；
Step 5, real meteorological data is obtained, real meteorological data is divided according to the synthesis cluster result and is grouped；
Step 6, according to group result, the prediction model gone out using the station work carries out photovoltaic generation power pre-
It surveys.
Modern optimization algorithm described in step 4 is the hybrid algorithm of difference algorithm and glowworm swarm algorithm, to prediction model
In parameter be trained the following steps are included:
Step 401, the neuron of hidden layer in the BP neural network is sequentially divided into two groups, by each group of hidden layer
The relevant parameter to be optimized of neuron is divided into one；
Step 402, branch's training is carried out, including,
It step 40201, by every parameter initialization to be optimized is respectively random value, remaining parameter to be optimized is set as solid
Definite value thus generates two random populations；
Step 40202, two populations are updated with glowworm swarm algorithm respectively；
Step 40203, difference operator operation, including selection, intersection and variation are carried out to two populations respectively, updated respectively
The population；
Step 40204, two training is allowed to reach preset maximum number of iterations or reach designated precision respectively；
Step 403, global training is carried out, including
Step 40301, half individual optimal in two populations is chosen respectively, by optimal one of two populations
Half individual carries out crossover operation, merges into a population；
Step 40302, a population is updated with glowworm swarm algorithm；
Step 40303, difference operator operation, including selection, intersection and variation are carried out to a population, respectively more
The new population；
Step 40304, global training is allowed to reach preset maximum number of iterations or reach designated precision；
Step 404, optimal individual is selected, the parameter of the optimal value of each parameter and the threshold value of output layer neuron is calculated
Value.
Branch's training be in order to obtain in each group preferably parameter particle faster, on the basis of branch's training into
The global training of row, can converge in global optimum faster.
The relevant parameter to be optimized of each group of neuron in 401 described in step includes this group of hidden layer mind
The weight connecting through member with all input layers, the threshold value of this group of hidden layer neuron and described group are implicit
The weight that layer neuron is connect with all output layer neurons.
Synthesis Clustering described in step 2, which refers to, is grouped sample using comprehensive a variety of Clusterings.It is common
Cluster mode have levels clustering method, partition clustering method, Grid Clustering method, Model tying method and Density Clustering side
Method.During cluster, the measurement to similitude is core, and common method for measuring similarity includes distance metric, correlation
Coefficient measurement and cosine similarity etc..For problem to be solved of the present invention, it is most intuitive using distance metric, but compares
It is more realistic, distance metric is only used, and will cause the deviation in some cognitions, so think of of the present invention using comprehensive cluster
Want to be grouped, using manhatton distance and cosine similarity collectively as clustering target, actually refer to so that the present invention has more
Lead meaning.
The manhatton distance is expressed as
D (x, y)=| x1-y1|+|x2-y2|+…+|xp-yp|
X and y is two samples, x1,x2,…xpFor each attribute value in sample x, y1,y2,…,ypIt is each in sample y
A attribute value has p attribute in each sample.The manhatton distance of two samples is bigger, and similitude is bigger, and manhatton distance is got over
Small, similitude is smaller.
The cosine similarity is expressed as
Wherein, | | x | | it is sample vector x=(x1,x2,…xp) Euclid norm, | | y | | be sample vector y=
(y1,y2,…yp) Euclid norm, under this measure of criterions, the similarity degree of two samples only has with their angle
It closes, and apart from unrelated.When angle is 0, similarity is up to 1, since the numerical value in sample is positive number, without negative, institute
It is up to pi/2 with angle, similarity minimum is equal to 0.
The thought for the comprehensive cluster that the present invention uses as a result, is the cluster side of comprehensive consideration distance value and angle value and determination
Method.Its similarity indicates are as follows:
Wherein, Max (d (X, Y)) indicates the maximum manhatton distance in sample set between two samples, α be it is preset can
Parameter is adjusted, for adjusting the weighted value between distance metric and angle measurement.
The clustering method that the present invention uses is by improved K Mean Method.The basic principle is that: assuming that be clustered
Data object has n, first in testing data object, selects k point at random as initial cluster centre；Successively calculate sample
This concentrates remaining point and the similitude of k cluster centre, according to the size of similitude, i.e., with which center is most like is just divided into
Where the center in class, all samples are divided into k class；The center that each class is recalculated according to division result is carrying out
It divides, passes through successive ignition, it is known that central sample is not changing or reaching maximum number of iterations, the cluster process knot
Beam.
Step 201: in historical climate data sample, selecting k point at random as cluster centre, C={ c1,c2,…
ck}；
Step 202: the sample in traversal sample set, if x and cluster centre ciSimilarity be greater than it arrive cluster centre cj
Similarity, then x is just divided into ciIn class；
Step 203: obtaining in each classification, two-by-two the similarity score matrix simi (x, y) of sample；
Step 204: calculating the sum of the similar value of each sample and other samples in the category in each classification, obtain most
Big similar value and, the sample of phase Sihe value maximum in the category is redefined as cluster centre；
Step 202,203 and 204 are repeated, until cluster centre does not change, or until reaching maximum number of iterations.
Dividing grouping to real meteorological data described in step 5 includes calculating separately real meteorological data sample xtWith
The similarity of the k cluster centres, by xtIt is divided into the maximum class group of similar angle value；
In step 6 using the real meteorological data divide prediction model that station work goes out to photovoltaic generation power into
Row prediction.
For the factor of influence photovoltaic generation power, there are many documents to be discussed both at home and abroad, it is most common several
A is irradiation intensity, temperature, humidity, air pressure, height above sea level etc..In view of being all same value for a certain its height above sea level of specific generating field,
Therefore cannot form variable influences the predicted value of its generated output, so in the embodiment of the present invention only consider irradiation intensity, temperature,
Four humidity, air pressure influence factors.
Data of the present invention come from the measured data of certain external photovoltaic power generation company.Each hour adopts
Collect a data, acquires the data of ten hours daily.The experimental data of the present embodiment includes 300 days 3000 groups of data,
In 2900 groups of data carry out classification based trainings, 100 groups of data are used as test data, and all neural network input layer quantity is 4, defeated
Layer number is 1 out.
In order to acquire best hidden layer neuron number, when the neuron number of hidden layer is 8, the neural network
There is best performance solving this experiment problem, the parameter total number to be optimized of the neural network is 49, in the present embodiment
0.5 is set by the α in step 2, sets 0.5 for the fixed value in step 40201, sets 4 for the k in cluster.Pass through
It is as shown in the table that experiment can obtain partial results.
| Classification number | Actual value | It does not cluster | The method of the present invention |
| 1 | 0.021 | 0.033 | 0.023 |
| 1 | 0.034 | 0.041 | 0.032 |
| 1 | 0.027 | 0.034 | 0.026 |
| 2 | 0.068 | 0.075 | 0.065 |
| 2 | 0.085 | 0.072 | 0.080 |
| 2 | 0.043 | 0.062 | 0.051 |
| 2 | 0.049 | 0.064 | 0.054 |
| 3 | 0.065 | 0.051 | 0.059 |
| 3 | 0.077 | 0.083 | 0.076 |
| 3 | 0.081 | 0.065 | 0.074 |
| 4 | 0.054 | 0.064 | 0.061 |
| 4 | 0.046 | 0.041 | 0.043 |
As can be seen from the table, data train the model come to each by carrying out model training respectively after cluster
The prediction effect of class is substantially better than the prediction effect not clustered.Therefore, it can be seen that training sample and test sample into
Row cluster, then to training and test neural network model, there is higher prediction accuracy.
By above-mentioned experiment it is found that prediction technique proposed by the present invention, carries out neural network prediction model in terms of two
It improves, first, using comprehensive clustering method, fully considers the similitude of distance and angle, it is more suitable for being carried out to data
Classification, disaggregated model training is carried out using classification data so that model is with stronger predictive ability, second, utilize difference
Algorithm and the hybrid algorithm of glowworm swarm algorithm optimize the parameter of neural network, at the same evolved using branch and it is global into
Change combines, so that model has more robustness, it is easier to obtain global optimum.As a result, therefore the present invention is compared with conventional method meter
It is relatively lower to calculate complexity, and prediction accuracy is higher.
Claims (6)
1. a kind of photovoltaic generation power timesharing prediction technique based on hybrid production style, which is characterized in that including following
Step:
Step 1, the historical data of photo-voltaic power generation station is obtained, the historical data includes with history meteorological data and generated output
Data；
Step 2, comprehensive Clustering is carried out to the history meteorological data；
Step 3, by group result, the historical data is constructed into several training sets in groups；
Step 4, neural network prediction model is established, input layer, hidden layer and output layer number is determined, utilizes modern optimization algorithm
Parameter in prediction model is trained, each training set trains a neural network prediction model；
Step 5, real meteorological data is obtained, real meteorological data is divided according to the synthesis cluster result and is grouped；
Step 6, according to group result, photovoltaic generation power is predicted using the prediction model that the station work goes out.
2. photovoltaic generation power timesharing prediction technique according to claim 1, which is characterized in that show described in step 4
It is the hybrid algorithm of difference algorithm and glowworm swarm algorithm for optimization algorithm, the parameter in prediction model is trained including following
Step:
Step 401, the neuron of hidden layer in the BP neural network is sequentially divided into two groups, by each group of hidden layer nerve
The relevant parameter to be optimized of member is divided into one；
Step 402, branch's training is carried out, including,
It step 40201, by every parameter initialization to be optimized is respectively random value, remaining parameter to be optimized is set as fixed
Value, thus generates two random populations；
Step 40202, two populations are updated with glowworm swarm algorithm respectively；
Step 40203, difference operator operation, including selection, intersection and variation are carried out to two populations respectively, respectively described in update
Population；
Step 40204, two training is allowed to reach preset maximum number of iterations or reach designated precision respectively；
Step 403, global training is carried out, including
Step 40301, half individual optimal in two populations is chosen respectively, by half of optimal 1 of two populations
Body carries out crossover operation, merges into a population；
Step 40302, a population is updated with glowworm swarm algorithm；
Step 40303, difference operator operation, including selection, intersection and variation are carried out to a population, updates institute respectively
The population stated；
Step 40304, global training is allowed to reach preset maximum number of iterations or reach designated precision；
Step 404, optimal individual is selected, the parameter value of the optimal value of each parameter and the threshold value of output layer neuron is calculated.
3. photovoltaic generation power timesharing prediction technique according to claim 2, which is characterized in that comprehensive described in step 2
Close the clustering method that Clustering is the distance between comprehensive consideration history meteorological data sample value and angle value and determination, sample
Between similarity indicate are as follows:
Wherein, Max (d (X, Y)) indicates that the maximum manhatton distance in sample set between two samples, α are preset adjustable ginseng
Number, for adjust distance metric and angle measurement between weighted value, the manhatton distance be expressed as d (x, y)=| x1-y1
|+|x2-y2|+…+|xp-yp|, x and y are two samples, x1,x2,…xpFor each attribute value in sample x, y1,y2,…,yp
For each attribute value in sample y, there is p attribute in each sample,Wherein, | | x | | it is
Sample vector x=(x1,x2,…xp) Euclid norm, | | y | | be sample vector y=(y1,y2,…yp) Euclid
Norm.
4. photovoltaic generation power timesharing prediction technique according to claim 3, which is characterized in that comprehensive described in step 2
Close Clustering the step of include:
Step 201: in historical climate data sample, selecting k point at random as cluster centre, C={ c1,c2,…ck}；
Step 202: the sample in traversal sample set, if x and cluster centre ciSimilarity be greater than it arrive cluster centre cjPhase
Like degree, then x is just divided into ciIn class；
Step 203: obtaining in each classification, two-by-two the similarity matrix simi (x, y) of sample；
Step 204: calculating the sum of the similar value of each sample and other samples in the category in each classification, obtain maximum
Similar value and, the sample of phase Sihe value maximum in the category is redefined as cluster centre；
Step 202,203 and 204 are repeated, until cluster centre does not change, or until reaching maximum number of iterations；
Dividing grouping to real meteorological data described in step 5 includes calculating separately real meteorological data sample xtWith it is described
The similarity of k cluster centres, by xtIt is divided into the maximum class group of similar angle value；
The prediction model that station work goes out is divided using the real meteorological data in step 6 to carry out in advance photovoltaic generation power
It surveys.
5. photovoltaic generation power timesharing prediction technique according to claim 4, which is characterized in that the history meteorology number
According to including irradiation intensity, temperature, humidity and air pressure, the input layer quantity is 4, hidden layer neuron quantity is 8
It is 1 with output layer neuron quantity.
6. the photovoltaic generation power timesharing prediction technique according to claim 5 or 4, which is characterized in that by the α in step 2
It is set as 0.5, sets 0.5 for the fixed value in step 40201, sets 4 for the k in cluster.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910337078.5A CN109978284B (en) | 2019-04-25 | 2019-04-25 | Photovoltaic power generation power time-sharing prediction method based on hybrid neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910337078.5A CN109978284B (en) | 2019-04-25 | 2019-04-25 | Photovoltaic power generation power time-sharing prediction method based on hybrid neural network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109978284A true CN109978284A (en) | 2019-07-05 |
| CN109978284B CN109978284B (en) | 2021-05-14 |
Family
ID=67086214
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910337078.5A Active CN109978284B (en) | 2019-04-25 | 2019-04-25 | Photovoltaic power generation power time-sharing prediction method based on hybrid neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109978284B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110348648A (en) * | 2019-08-02 | 2019-10-18 | 国网电子商务有限公司 | A kind of predicting power of photovoltaic plant method and device |
| CN110717623A (en) * | 2019-09-20 | 2020-01-21 | 中国科学院自动化研究所 | Photovoltaic power generation power prediction method, device and equipment integrating multiple weather conditions |
| CN111369070A (en) * | 2020-03-13 | 2020-07-03 | 西安理工大学 | Envelope clustering-based multimode fusion photovoltaic power prediction method |
| CN111767982A (en) * | 2020-05-20 | 2020-10-13 | 北京大米科技有限公司 | Training method and device for user conversion prediction model, storage medium and electronic equipment |
| CN113033898A (en) * | 2021-03-26 | 2021-06-25 | 国核电力规划设计研究院有限公司 | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network |
| CN113919606A (en) * | 2021-12-14 | 2022-01-11 | 山东建筑大学 | Intelligent site selection method and system for distributed photovoltaic power station |
| CN116706907A (en) * | 2023-08-09 | 2023-09-05 | 深圳航天科创泛在电气有限公司 | Photovoltaic power generation prediction method based on fuzzy reasoning and related equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101404071A (en) * | 2008-11-07 | 2009-04-08 | 湖南大学 | Electronic circuit fault diagnosis neural network method based on grouping particle swarm algorithm |
| CN103034904A (en) * | 2013-01-11 | 2013-04-10 | 上海电机学院 | Firefly grouping method, as well as power dispatching system and power dispatching method based on same |
| CN104463349A (en) * | 2014-11-11 | 2015-03-25 | 河海大学 | Photovoltaic generated power prediction method based on multi-period comprehensive similar days |
| CN106529701A (en) * | 2016-10-21 | 2017-03-22 | 国网吉林省电力有限公司信息通信公司 | Optical fiber state prediction method for optimizing neural network based on improved firefly algorithm |
-
2019
- 2019-04-25 CN CN201910337078.5A patent/CN109978284B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101404071A (en) * | 2008-11-07 | 2009-04-08 | 湖南大学 | Electronic circuit fault diagnosis neural network method based on grouping particle swarm algorithm |
| CN103034904A (en) * | 2013-01-11 | 2013-04-10 | 上海电机学院 | Firefly grouping method, as well as power dispatching system and power dispatching method based on same |
| CN104463349A (en) * | 2014-11-11 | 2015-03-25 | 河海大学 | Photovoltaic generated power prediction method based on multi-period comprehensive similar days |
| CN106529701A (en) * | 2016-10-21 | 2017-03-22 | 国网吉林省电力有限公司信息通信公司 | Optical fiber state prediction method for optimizing neural network based on improved firefly algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 代倩等: "基于天气类型聚类识别的光伏***短期", 《中国电机工程学报》 * |
| 侯越等: "基于萤火虫优化的BP神经网络算法研究", 《兰州交通大学学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110348648A (en) * | 2019-08-02 | 2019-10-18 | 国网电子商务有限公司 | A kind of predicting power of photovoltaic plant method and device |
| CN110717623A (en) * | 2019-09-20 | 2020-01-21 | 中国科学院自动化研究所 | Photovoltaic power generation power prediction method, device and equipment integrating multiple weather conditions |
| CN110717623B (en) * | 2019-09-20 | 2022-11-08 | 中国科学院自动化研究所 | Photovoltaic power generation power prediction method, device and equipment integrating multiple weather conditions |
| CN111369070A (en) * | 2020-03-13 | 2020-07-03 | 西安理工大学 | Envelope clustering-based multimode fusion photovoltaic power prediction method |
| CN111369070B (en) * | 2020-03-13 | 2023-06-27 | 西安理工大学 | Multimode fusion photovoltaic power prediction method based on envelope clustering |
| CN111767982A (en) * | 2020-05-20 | 2020-10-13 | 北京大米科技有限公司 | Training method and device for user conversion prediction model, storage medium and electronic equipment |
| CN113033898A (en) * | 2021-03-26 | 2021-06-25 | 国核电力规划设计研究院有限公司 | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network |
| CN113919606A (en) * | 2021-12-14 | 2022-01-11 | 山东建筑大学 | Intelligent site selection method and system for distributed photovoltaic power station |
| CN116706907A (en) * | 2023-08-09 | 2023-09-05 | 深圳航天科创泛在电气有限公司 | Photovoltaic power generation prediction method based on fuzzy reasoning and related equipment |
| CN116706907B (en) * | 2023-08-09 | 2024-01-23 | 深圳航天科创泛在电气有限公司 | Photovoltaic power generation prediction method based on fuzzy reasoning and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109978284B (en) | 2021-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109978284A (en) | Photovoltaic power generation power time-sharing prediction method based on hybrid neural network model | |
| CN109754113A (en) | Load forecasting method based on dynamic time warping Yu length time memory | |
| CN102298707B (en) | Wind power prediction method based on continuous time slice clustering and support vector machine (SVM) modeling | |
| CN105512745A (en) | Wind power section prediction method based on particle swarm-BP neural network | |
| CN109426889A (en) | Short-term load forecasting method based on KPCA in conjunction with improvement neural network | |
| CN109002915A (en) | Photovoltaic plant short term power prediction technique based on Kmeans-GRA-Elman model | |
| CN104951834A (en) | LSSVM (least squares support vector machine) wind speed forecasting method based on integration of GA (genetic algorithm) and PSO (particle swarm optimization) | |
| CN104123595B (en) | A kind of distribution network load prediction technique and system | |
| CN105701572B (en) | Photovoltaic short-term output prediction method based on improved Gaussian process regression | |
| CN106485093B (en) | Based on the solar irradiance time series synthetic method for improving Markov chain | |
| CN109086928A (en) | Photovoltaic plant realtime power prediction technique based on SAGA-FCM-LSSVM model | |
| CN105389634A (en) | Combined short-term wind power prediction system and method | |
| CN105447572A (en) | Wind power prediction system and method based on neural network optimized by genetic algorithm | |
| CN102184337A (en) | Dynamic combination analysis method of new energy generating capacity influenced by meteorological information | |
| CN110766134A (en) | Photovoltaic power station short-term power prediction method based on cyclic neural network | |
| CN108062720A (en) | A kind of load forecasting method based on similar day selection and random forests algorithm | |
| CN104820877A (en) | Photovoltaic system generation power prediction method based on cloud adaptive PSO-SNN | |
| CN109886567A (en) | A kind of short-term load forecasting method considering sendible temperature and radiation intensity | |
| CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
| CN110070228A (en) | BP neural network wind speed prediction method for neuron branch evolution | |
| CN109978275A (en) | A kind of extreme wind wind speed forecasting method and system mixing CFD and deep learning | |
| CN110070227A (en) | Migration neural network power prediction method suitable for grid-connected photovoltaic power generation | |
| CN109961173A (en) | A kind of intelligent Forecasting for distributed generation resource generated output | |
| CN109444840A (en) | A kind of radar clutter suppression method based on machine learning | |
| CN106786499B (en) | Based on the short-term wind power forecast method for improving AFSA optimization ELM |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |