CN111242371B - Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model - Google Patents

Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model Download PDF

Info

Publication number
CN111242371B
CN111242371B CN202010025659.8A CN202010025659A CN111242371B CN 111242371 B CN111242371 B CN 111242371B CN 202010025659 A CN202010025659 A CN 202010025659A CN 111242371 B CN111242371 B CN 111242371B
Authority
CN
China
Prior art keywords
data
model
power
correction
prediction
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.)
Active
Application number
CN202010025659.8A
Other languages
Chinese (zh)
Other versions
CN111242371A (en
Inventor
朱红路
尹万思
韩雨彤
史淯城
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North China Electric Power University
Original Assignee
North China Electric Power University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by North China Electric Power University filed Critical North China Electric Power University
Priority to CN202010025659.8A priority Critical patent/CN111242371B/en
Publication of CN111242371A publication Critical patent/CN111242371A/en
Application granted granted Critical
Publication of CN111242371B publication Critical patent/CN111242371B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q10/00Administration; Management
    • G06Q10/04Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/084Backpropagation, e.g. using gradient descent
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q10/00Administration; Management
    • G06Q10/06Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067Enterprise or organisation modelling
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q10/00Administration; Management
    • G06Q10/20Administration of product repair or maintenance
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q50/00Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06Energy or water supply
    • YGENERAL 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
    • Y04INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04SSYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Game Theory and Decision Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Development Economics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Educational Administration (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a photovoltaic power generation short-term prediction correction method based on a non-iterative multi-model, belonging to the technical field of new energy. And then establishing a plurality of correction model correction prediction results through a non-iterative method. The method avoids the influence of accumulated errors, can reduce the prediction error by about 2-6%, effectively improves the prediction precision, and is favorable for the dispatching of a power system and the safe and stable operation of a power grid.

Description

Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model
Technical Field
The invention belongs to the technical field of new energy. In particular to a photovoltaic power generation short-term prediction correction method based on a non-iterative multi-model.
Background
In recent decades, due to the advantages of abundant solar energy resources, cleanness, no pollution and inexhaustibility, the photovoltaic power generation technology is rapidly developed, and the number of photovoltaic power stations is rapidly increased in the global scope. With access of large-scale photovoltaic power stations to a power system, characteristics of intermittency, random fluctuation and the like of photovoltaic power generation bring many difficulties and challenges to power dispatching work. Therefore, an accurate photovoltaic power prediction result has important significance for safe and stable operation of the power system under the condition of high-proportion photovoltaic access.
The photovoltaic power prediction can be divided into ultra-short term prediction, short term prediction and medium and long term prediction according to the predicted time scale. The short-term prediction is generally used for correcting a day-ahead plan curve, can be used for smooth control of photovoltaic output and maintenance and overhaul of components of a photovoltaic power station, is beneficial to processing of emergency states, and has important significance in operation and real-time scheduling of a power system. According to the adopted mathematical physics theory and the prediction output quantity thereof, the photovoltaic power generation short-term prediction method can be divided into two categories: (1) a direct prediction method (statistical method) for directly predicting the output power of the photovoltaic system; (2) firstly, the solar radiation is predicted, and then an indirect prediction method (physical method) of photoelectric output power is obtained according to photoelectric conversion efficiency.
Both the direct prediction method and the indirect prediction method are applied in different occasions, but the two methods cannot avoid the problem of large prediction error. Due to the random fluctuation of the photovoltaic output, when the photovoltaic output fluctuates greatly, an accurate result cannot be obtained only by depending on weather forecast values and historical data. Therefore, in order to reduce the negative impact of the error on the grid system, the power prediction error needs to be corrected. The correction method can improve the prediction precision and reduce the influence of uncertainty, and is an effective means for improving the engineering practicability of the photovoltaic power prediction system.
Disclosure of Invention
The invention aims to provide a photovoltaic power generation short-term prediction correction method based on a non-iterative multi-model, which is characterized by comprising the following steps of:
s1, processing original data, namely dividing historical environment data and numerical weather forecast data according to seasons;
s2, dividing meteorological data and photovoltaic output data according to seasons in the step S1, and establishing a prediction model of each season through training of input and output variables;
s3, forecasting and obtaining a result, namely selecting a proper model from the forecasting models of each season established in the step S2 to forecast, and obtaining the result;
s4, selecting correction data; selecting the required predicted power and the actually measured power from the result obtained in the step S3;
s5, establishing a correction model, and establishing correction models at different moments in a non-iterative mode;
and S6, correcting results, and synthesizing the outputs of different models to obtain correction power.
The collected original data in the step S1 comprise historical power data and historical meteorological data of the photovoltaic power station; the historical power data of the photovoltaic power station comprises historical irradiance, ambient temperature, humidity and wind speed data corresponding to the photovoltaic power station, and the numerical weather forecast data comprises numerical weather forecast irradiance, ambient temperature, humidity and wind speed data of the location of the photovoltaic power station; and classifying the collected historical power data and historical meteorological data of the photovoltaic power station according to seasons.
In step S2, a seasonal prediction model is established according to the classified data, and the method specifically includes the following steps:
s21, training input variables and output variables in different seasons by using a support vector machine, a BP neural network and an extreme learning machine respectively, and establishing a mapping relation between the input variables and the output variables in each season;
s22, obtaining prediction models of the support vector machine, the BP neural network and the extreme learning machine in four seasons of spring, summer, autumn and winter respectively.
In the step S3, the photovoltaic output power is predicted through the seasonal model, and the method specifically includes the following steps:
s31, judging the season of the prediction data, and selecting a proper seasonal model for prediction by using three algorithms;
and S32, obtaining a photovoltaic power prediction result.
In step S4, the training and prediction data of the correction part is selected, which specifically includes the following steps:
s41, selecting the predicted power and the actually measured power quantity during training of a correction part of models;
s42, selecting the predicted power quantity according to the needs to correct; during correction, the predicted power x (t + 1), x (t + 2), … x (t + L) and the actual power x ' (t-k + 1), … x ' (t-1) and x ' (t) are selected to be input into a model, and the corrected power at the corresponding moment is obtained; wherein, (x (t + 1), x (t + 2), … x (t + L) are the predicted power at time t +1, t +2, … t + L, respectively, and x ' (t-k + 1), … x ' (t-1), x ' (t) is the actual power at time t-k +1, … t-1,t, respectively.
In the step S5, the extreme learning machine is used to train the input and output variables, which specifically includes the following steps:
s51, training model parameters by using the selected training data through an extreme learning machine, and establishing a mapping relation between an input variable and an output variable;
is provided with M arbitrary data (x) i ,t i ) Wherein x is i =[x i1 ,x i2 ……x im ] T ∈R m ,t i =[t i1 ,t i2 ……t in ] T ∈R n Then, the output of the feedforward neural network with N hidden node excitation functions as G can be expressed as:
Figure BDA0002362350580000031
in the formula: a is i =[a 1i ,a 2i ,……a mi ] T The weight vectors of the ith hidden node and the input node are obtained; beta is a i =[β i1i2 ,……β in ] T The weight vectors of the ith hidden node and the output node are obtained; b i Is the bias of the ith hidden node; n is the number of hidden nodes; equation (1) can be abbreviated as:
Hβ=T (2)
Figure BDA0002362350580000041
in the formula: h is a hidden layer output matrix, and the ith column of H corresponds to an input x 1 ,x 2 ,……x n An ith hidden layer output vector; the output weight may be obtained by solving a least squares solution of the system of linear equations (3);
Figure BDA0002362350580000042
the least squares solution is:
β=H + T (5)
middle H + Is the Moore-Penrose generalized inverse of the hidden layer output matrix H.
S52, establishing L models by using a non-iterative method through an extreme learning machine.
In step S6, correcting the short-term prediction result specifically includes the following steps:
s61, inputting numerical weather forecast data into the seasonal model of S32 to obtain primary predicted power;
and S62, substituting the predicted power into the correction model in the step S52 to obtain the final photovoltaic correction power.
The photovoltaic power short-term prediction correction method has the advantages that: and the short-term prediction result of the power is corrected through the continuity between the output sequences, so that the prediction precision is improved. And effective information is provided for smooth control of photovoltaic output and maintenance and overhaul of components of the photovoltaic power station, including operation and real-time scheduling of the power system. The invention has the following characteristics:
1. the prediction method based on the seasonal model provides possibility for improving the adaptability of the model to the output under different seasonal conditions, and can obtain higher prediction accuracy.
2. And the continuity between photovoltaic output sequences is considered, and the actual value is introduced into the correction model, so that the prediction precision is improved to a great extent. Provides guarantee for dispatching and safe and stable operation of the power system
3. Compared with the conventional iterative method, the non-iterative method avoids the influence caused by accumulated errors and has higher prediction precision.
Drawings
FIG. 1 is a flow chart of short-term predictive correction of photovoltaic power generation;
FIG. 2 is a diagram of an ELM structure;
FIG. 3 is a non-iterative correction model;
FIG. 4 is a predicted power corrected using 6 actual values; wherein a is the relation between the actual power and the predicted power after the first 6 values are corrected; b is the error distribution condition after the first 6 values are corrected;
FIG. 5 shows predicted output power for different seasons; wherein, a spring, b summer, c autumn and d winter;
FIG. 6 shows an error distribution of different months; wherein, a, RMSE and b, MAE;
Detailed Description
The invention provides a photovoltaic power generation short-term prediction correction method based on a non-iterative multi-model. The photovoltaic power multi-model interval prediction method of the present invention is further described below with reference to the accompanying drawings and specific embodiments.
FIG. 1 is a flow chart illustrating short term photovoltaic power generation prediction correction; the photovoltaic power generation short-term prediction correction method based on the non-iterative multi-model comprises the following specific steps:
s1, collecting historical operation data, historical environment data and numerical weather forecast data of the photovoltaic power station.
The historical operation data comprises historical power data of the photovoltaic power station, the historical environment data comprises historical irradiance, environment temperature, humidity and wind speed data corresponding to the photovoltaic power station, and the numerical weather forecast data comprises numerical weather forecast irradiance, environment temperature, humidity and wind speed data of the location of the photovoltaic power station. Selecting the data of 2018 years of a new energy power system national key laboratory photovoltaic empirical test power station of North China Power university, wherein the power station consists of a photovoltaic power generation system with the capacity of 250Kw, a solar energy meteorological station, a photovoltaic power station monitoring system and a numerical weather forecasting system, and the data sampling time is 15 minutes. According to the method, 10KW photovoltaic arrays are selected for analysis, and photovoltaic historical data are classified according to seasons.
S2, establishing a seasonal prediction model according to the classified data, and specifically comprising the following steps:
and S21, training input variables and output variables in different seasons by using a support vector machine, a BP neural network and an extreme learning machine respectively, and establishing a mapping relation between the input variables and the output variables in each season.
S22, obtaining prediction models of the support vector machine, the BP neural network and the extreme learning machine in four seasons of spring, summer, autumn and winter respectively.
S3, predicting the photovoltaic output power through a seasonal model, and specifically comprising the following steps:
and S31, judging the season of the prediction data, and selecting a proper model for prediction.
And S32, obtaining a power prediction result.
The analysis of the prediction result shows that the frequency of the positive error is higher than that of the negative error, which indicates that the probability of the occurrence of the case that the predicted value is larger than the measured value is higher.
S4, selecting training and prediction data of the correction part, and specifically comprising the following steps:
and S41, selecting the predicted power and the actually measured power quantity during training of the correction part of the model.
And S42, selecting the predicted power quantity according to the requirement to correct. During correction, the predicted power x (t + 1), x (t + 2), … x (t + L) and the actual power x ' (t-k + 1), … x ' (t-1), x ' (t) are selected to be input into a model, and the corrected power at the corresponding moment is obtained.
S5, training input and output variables by using an extreme learning machine, and specifically comprising the following steps:
and S51, training model parameters by using the selected training data through an extreme learning machine, and establishing a mapping relation between the input variable and the output variable.
The principle of the extreme learning machine is shown in FIG. 2
Is provided with M arbitrary data (x) i ,t i ) Wherein x is i =[x i1 ,x i2 ……x im ] T ∈R m ,t i =[t i1 ,t i2 ……t in ] T ∈R n Then, the output of the feedforward neural network with N hidden node excitation functions as G can be expressed as:
Figure BDA0002362350580000071
in the formula: a is i =[a 1i ,a 2i ,……a mi ] T The weight vectors of the ith hidden node and the input node are obtained; beta is a i =[β i1i2 ,……β in ] T The weight vectors of the ith hidden node and the output node are obtained; b is a mixture of i Is the offset of the ith hidden node; n is the number of hidden nodes. Equation (3) can be abbreviated as:
Hβ=T (2)
Figure BDA0002362350580000072
in the formula: h is a hidden layer output matrix, and the ith column of H corresponds to an input x 1 ,x 2 ,……x n The ith hidden layer outputs a vector. The output weight may be obtained by solving a least squares solution of the system of linear equations (3);
Figure BDA0002362350580000073
least squares solution of β = H + T (5)
In the formula H + Is the Moore-Penrose generalized inverse of the hidden layer output matrix H.
S52, establishing L models by using a non-iterative method through an extreme learning machine.
When short-term prediction is performed, a rolling iteration method is often adopted, namely, a value obtained by each prediction is used as an actual value of the current moment and is input into a model. However, due to accumulation of errors, the prediction error increases with the increase of the prediction time, resulting in an increase in deviation of the prediction result. In consideration of the limitation of the iterative method, the iterative method is abandoned, and a non-iterative mode is selected.
Assuming that the photovoltaic output power needs to be corrected at time t for L future times, the input-output variables are shown in fig. 3.
Figure BDA0002362350580000081
Actual power, model input value; x (t + 1), x (t + 2), x (t + 3), …, wherein x (t + L) is the predicted power, and the bold parts are input values of different models respectively; the corrected power portions x '(t + 1), x' (t + 2), x '(t + 3), …, x' (t + L) are the output values of the different models. The input data of training is a predicted power value at a certain moment and k real values before the predicted value. In correction, the input data are predicted values and/or based on x (t + 1) to x (t + L)>
Figure BDA0002362350580000082
To>
Figure BDA0002362350580000083
The output data is the correction value.
S6, correcting the short-term prediction result, which specifically comprises the following steps:
s61, inputting numerical weather forecast data into the seasonal model of S32 to obtain primary predicted power;
and S62, substituting the predicted power into the correction model in the step S52 to obtain the final photovoltaic correction power.
RMSE and MAE are selected as error evaluation indexes, and are as follows:
Figure BDA0002362350580000084
Figure BDA0002362350580000085
in the formula: p is installed capacity; n is the number of samples; e.g. of the type t Representing an error.
As shown in Table 1, RMSE and MAE are 4.4774 and 1.9372, respectively, when corrected with the past value. As the number increases, both the RMSE and MAE values begin to decrease, with RMSE and MAE reaching a minimum at the sixth value. After reaching this minimum point, the RMSE and MAE remain substantially stable and increase even slightly. That is, the best effect is obtained when the number of past values is 6. Here we select the top 6 value corrections.
TABLE 1 error table corrected by different values in the past
Figure BDA0002362350580000091
/>
Fig. 4 is a graph of actual power versus predicted power and corrected power after correction using the first 6 values. Wherein a is the relation between the actual power and the predicted power after the first 6 values are corrected; it can be seen that the deviation between the predicted power and the actual power is large and the distribution is more obvious above the left of the dotted line y = x, i.e. the probability that the predicted power is greater than the actual power is higher than the probability that the actual power is greater than the predicted power. However, the points of correcting power are positioned at two sides of the dotted line, the distribution is concentrated, the integral deviation is not very large, only the individual value of the pole deviates far from the dotted line, and the error is large. b is the corrected error distribution for the first 6 values. As can be seen from the figure, the value of the prediction error is large, the distribution range is wide, and the prediction error has distribution in the range of [ -3000,4000 ]. The corrected error value has obvious large amplitude convergence, the error is greatly reduced, most of the error is in the range of [ -1000,1000], and a small part of the error exceeds the range. This shows that after correction, the error is reduced and the prediction accuracy is improved. Although non-iterative methods can reduce the accumulation of errors, the prediction accuracy also decreases with increasing time scale. Moreover, the predicted value is corrected by using the adjacent actual value, the close connection between the two values is required to be ensured, the continuity on a certain time sequence is ensured, and the higher the continuity is, the higher the precision is. Once the time scale that needs to be corrected is too long, the error becomes larger and larger. In table 2, the error of the first correction value is small, and as the time distance between the correction value and the actual value becomes larger, the RMSE increases almost 1% each time, and the MAE increases about 0.5% each time. In particular, the error of the seventh and eighth correction values is already greater than 10%, increasing the error by more than a factor of two compared to the first 4.3764%. Therefore, a suitable correction timescale needs to be selected. The time resolution of the photovoltaic data used this time was 15 minutes, and the four points were 1 hour. From the above table it can be found that it is relatively reasonable to correct 4 points, i.e. the time is one hour.
TABLE 2 error cases when correction values are different
Figure BDA0002362350580000101
And the table 3 shows the statistics of the prediction errors of the photovoltaic power station in 2017. RMSE of BP, SVM and ELM is about 10%, and MAE is about 5%. And the two errors of the correction model are 5.6795% and 2.5028%, which are almost reduced by one time compared with the errors of other models, and the accuracy of the correction model can be effectively improved.
TABLE 3 different model errors
Figure BDA0002362350580000111
As shown in fig. 5 a spring, b summer, c autumn, and d winter, relationships between predicted values, actual values, and corrected values for different seasons are presented. As can be seen from fig. 5, the predicted curves obtained by the three algorithms have larger deviation and are smoother than the actual output curve, and the fluctuation characteristics of the photovoltaic output cannot be well fitted. Compared with the prediction curve, the correction curve not only reduces the error, but also has higher fitting degree with the actual curve. Since the predicted value is corrected one hour at a time by a non-iterative method, although accumulation of errors due to iteration can be prevented, the accuracy also decreases with time. It can also be seen that the first value of each correction has a reduced error compared to the last value of the previous correction, and that the subsequent correction value has an increased error over time.
As shown in fig. 6 a, RMSE and b, MAE, the overall trend of the three verification algorithms is consistent, with only slight differences in the next few months. The correction algorithm is trending with other algorithms, but there are some differences in some places (e.g., months 2, months 7, etc.). The correction algorithm can reduce most errors by more than one time. In addition, when observing fig. a and b, both RMSE and MAE show a tendency of rising first in descending, and then rising and descending again. Furthermore, the maximum value of the error for the three validation algorithms occurs in July, while the maximum value of the correction error occurs in August. The minimum of the errors for the three validation algorithms occurs in december, while the minimum of the correction errors occurs in january. This means that the prediction error is not the smallest, and the correction error is necessarily the smallest, and the two are not in a definite relationship. Among the three algorithms, BP is high in prediction speed, small in fluctuation at the later stage and relatively stable. The SVM has slow prediction speed and difficult parameter selection, but the error in the first two months is obviously lower than that in the other two methods. The ELM converges quickly, but the error is sometimes large.
The invention considers the seasonal distribution characteristic of errors, establishes a seasonal prediction model, introduces an actual value into a correction model, and corrects the prediction result. In addition, compared with an iterative method, the non-iterative method used by the invention avoids the influence caused by accumulated errors and has higher prediction precision.

Claims (6)

1. A photovoltaic power generation short-term prediction correction method based on a non-iterative multi-model is characterized by comprising the following steps:
s1, processing original data, and dividing historical environment data and numerical weather forecast data according to seasons;
s2, dividing meteorological data and photovoltaic output data according to seasons in the step S1, and establishing a prediction model of each season through training of input and output variables;
s3, forecasting and obtaining a result, namely selecting a proper model from the forecasting models of each season established in the step S2 to forecast, and obtaining the result;
s4, selecting correction data; selecting the required predicted power and the actually measured power from the result obtained in the step S3;
s5, establishing a correction model, and establishing correction models at different moments in a non-iterative mode;
in the step S5, the extreme learning machine is used to train the input and output variables, which specifically includes the following steps:
s51, training model parameters by using the selected training data through an extreme learning machine, and establishing a mapping relation between an input variable and an output variable;
with M arbitrary data (x) i ,t i ) Wherein x is i =[x i1 ,x i2 ……x im ]T∈R m ,t i =[t i1 ,t i2 ……t in ] T ∈R n Then, the output of the feedforward neural network with N hidden node excitation functions as G is expressed as:
Figure FDA0004066884740000011
in the formula: a is i =[a 1i ,a 2i ,……a mi ] T The weight vectors of the ith hidden node and the input node are obtained; beta is a i =[β i1i2 ,……β in ] T The weight vectors of the ith hidden node and the output node are obtained; b i Is the bias of the ith hidden node; n is the number of hidden nodes; equation (1) is abbreviated as:
Hβ=T (2)
Figure FDA0004066884740000021
in the formula: h is a hidden layer output matrix, and the ith column of H corresponds to an input x 1 ,x 2 ,……x n An ith hidden layer output vector; the output weight may be obtained by solving a least squares solution of the system of linear equations (3);
Figure FDA0004066884740000022
the least squares solution is:
β=H + T (5)
in the formula H + The Moore-Penrose generalized inverse of the hidden layer output matrix H;
s52, establishing L models by using a non-iterative method through an extreme learning machine;
and S6, correcting results, and synthesizing the outputs of different models to obtain correction power.
2. The non-iterative multi-model-based photovoltaic power generation short-term prediction correction method according to claim 1, wherein the raw data collected in step S1 comprises historical power data and historical numerical weather forecast data of a photovoltaic power station; the historical power data of the photovoltaic power station comprises historical irradiance, ambient temperature, humidity and wind speed data corresponding to the photovoltaic power station; the numerical weather forecast data comprises numerical weather forecast irradiance, ambient temperature, humidity and wind speed data of the place where the photovoltaic power station is located; and classifying the collected historical power data and historical meteorological data of the photovoltaic power station according to seasons.
3. The non-iterative multi-model-based photovoltaic power generation short-term prediction correction method according to claim 1, wherein in the step S2, a seasonal prediction model is established according to the classified data, and the method specifically comprises the following steps:
s21, training input variables and output variables in different seasons by using a support vector machine, a BP neural network and an extreme learning machine respectively, and establishing a mapping relation between the input variables and the output variables in each season;
s22, obtaining prediction models of the support vector machine, the BP neural network and the extreme learning machine in four seasons of spring, summer, autumn and winter respectively.
4. The non-iterative multi-model-based photovoltaic power generation short-term prediction correction method according to claim 1, wherein in the step S3, the photovoltaic output power is predicted through a seasonal model, and the method specifically comprises the following steps:
s31, judging the season of the prediction data, and selecting a proper seasonal model for prediction by using three algorithms;
and S32, obtaining a photovoltaic power prediction result.
5. The non-iterative multi-model-based photovoltaic power generation short-term prediction correction method according to claim 1, wherein in the step S4, training and prediction data of a correction part are selected, and the method specifically includes the following steps:
s41, selecting the predicted power and the actually measured power quantity during training of a correction part of models;
s42, selecting the predicted power quantity according to the requirement to correct; during correction, the predicted power x (t + 1), x (t + 2), … x (t + L) and the actual power x ' (t-k + 1), … x ' (t-1) and x ' (t) are selected to be input into a model, and the corrected power at the corresponding moment is obtained; wherein, (x (t + 1), x (t + 2), … x (t + L) are the predicted power at time t +1, t +2, … t + L, respectively, and x ' (t-k + 1), … x ' (t-1), x ' (t) is the actual power at time t-k +1, … t-1,t, respectively.
6. The non-iterative multi-model-based photovoltaic power generation short-term prediction correction method according to claim 1, wherein in the step S6, the short-term prediction result is corrected, specifically including the steps of:
s61, inputting numerical weather forecast data into the seasonal model of S32 to obtain primary predicted power;
and S62, substituting the predicted power into the correction model in the step S52 to obtain the final photovoltaic correction power.
CN202010025659.8A 2020-01-10 2020-01-10 Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model Active CN111242371B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010025659.8A CN111242371B (en) 2020-01-10 2020-01-10 Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010025659.8A CN111242371B (en) 2020-01-10 2020-01-10 Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model

Publications (2)

Publication Number Publication Date
CN111242371A CN111242371A (en) 2020-06-05
CN111242371B true CN111242371B (en) 2023-04-18

Family

ID=70876088

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010025659.8A Active CN111242371B (en) 2020-01-10 2020-01-10 Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model

Country Status (1)

Country Link
CN (1) CN111242371B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111967652B (en) * 2020-07-22 2023-10-24 国网浙江省电力有限公司电力科学研究院 Double-layer collaborative real-time correction photovoltaic prediction method
CN112926653B (en) * 2021-02-25 2022-08-26 山东大学 Photovoltaic power short-term prediction method and system based on small-scale weather type correction

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103208037B (en) * 2013-04-26 2016-03-30 国电南瑞南京控制***有限公司 A kind of power forecasting method based on on-line correction being applied to new energy power station
CN103996073B (en) * 2014-05-06 2017-11-24 国家电网公司 Light-metering network real time correction self study arma modeling photovoltaic power Forecasting Methodology
CN105184404B (en) * 2015-08-31 2018-12-18 中国科学院广州能源研究所 Output power classification forecasting system suitable for photovoltaic system Life cycle
CN106446440A (en) * 2016-10-11 2017-02-22 天津大学 Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine
CN109816095B (en) * 2019-01-14 2023-04-07 湖南大学 Network flow prediction method based on improved gated cyclic neural network
CN109961190B (en) * 2019-03-29 2021-09-10 北京鑫泰绿能科技有限公司 Photovoltaic power station day-ahead power prediction method based on learning day fitting degree clustering
CN110598896A (en) * 2019-07-26 2019-12-20 陕西省水利电力勘测设计研究院 Photovoltaic power prediction method based on prediction error correction

Also Published As

Publication number Publication date
CN111242371A (en) 2020-06-05

Similar Documents

Publication Publication Date Title
CN111008728B (en) Prediction method for short-term output of distributed photovoltaic power generation system
CN110474339B (en) Power grid reactive power control method based on deep power generation load prediction
CN109086928A (en) Photovoltaic plant realtime power prediction technique based on SAGA-FCM-LSSVM model
CN110334870B (en) Photovoltaic power station short-term power prediction method based on gated cyclic unit network
Yadav et al. A novel GA-ANFIS hybrid model for short-term solar PV power forecasting in Indian electricity market
Lu et al. A hybrid approach for day-ahead forecast of PV power generation
CN110601250B (en) Method for evaluating stability of static voltage of access point based on new energy power station
CN111626473A (en) Two-stage photovoltaic power prediction method considering error correction
CN111242371B (en) Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model
CN112070311A (en) Day-ahead light power prediction method based on similar day clustering and meteorological factor weighting
CN112508246B (en) Photovoltaic power generation power prediction method based on similar days
Fentis et al. Short-term pv power forecasting using support vector regression and local monitoring data
CN110866633A (en) Micro-grid ultra-short term load prediction method based on SVR support vector regression
CN115796004A (en) Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models
CN115965132A (en) Distributed photovoltaic digital twin system power prediction method based on GA-BP neural network
CN112508279A (en) Regional distributed photovoltaic prediction method and system based on spatial correlation
CN109961190B (en) Photovoltaic power station day-ahead power prediction method based on learning day fitting degree clustering
Karabiber et al. Short term PV power estimation by means of extreme learning machine and support vector machine
CN105741192B (en) Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant
CN113991711A (en) Capacity configuration method for energy storage system of photovoltaic power station
Sun et al. Short-term photovoltaic power prediction modeling based on AdaBoost algorithm and Elman
Shi et al. Short-term photovoltaic power forecast based on long short-term memory network
CN112036672B (en) New energy power generation ultra-short term power prediction method and system based on iterative correction
Kaiju et al. Short-term photovoltaic power prediction based on TS fuzzy neural network
CN113780643A (en) Photovoltaic power station short-term output prediction method based on case reasoning

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