A method, device and equipment for joint prediction of hydropower, wind and solar power generation

Abstract

The present invention discloses a method, device and equipment for joint prediction of water, wind and photovoltaic power generation, which includes: data set acquisition, the data set includes historical meteorological data and historical hydropower, wind power and photovoltaic power data; using the data set, respectively, for hydropower, wind power and photovoltaic power, according to the divided seasons and weather, training and verifying multiple deep learning models, and ranking the deep learning models with the highest prediction accuracy as candidate models; inputting the current meteorological data into the candidate models under the corresponding seasons and weather conditions, obtaining the hydropower, wind power and photovoltaic predicted power, and forming a water, wind and photovoltaic joint predicted power set; under the condition of minimum residual load, extracting the optimal hydropower, wind power and photovoltaic power from the water, wind and photovoltaic joint predicted power set. The technical solution provided by the present invention improves the accuracy of water, wind and photovoltaic joint power prediction.

Description

Technical Field

[0001] The present invention relates to the field of electric energy, and in particular to a method, device and equipment for jointly predicting hydropower, wind power and solar power generation. Background Art

[0002] Today, as we vigorously develop a low-carbon economy, hydropower, wind power, and photovoltaics, as environmentally friendly and clean renewable energy sources, have been increasingly valued by countries around the world. However, the inherent volatility and randomness of hydropower, wind power, and photovoltaics have brought serious challenges to the safety, stability, and power quality of the power system. To this end, by predicting the power of hydropower, wind, and photovoltaic power generation, the water, wind, and light energy sources are aggregated to form a multi-energy complementary power generation system, which is of great significance to reducing the impact of clean energy grid connection and improving the utilization rate of river basin resources. Since meteorological factors are the main factors causing fluctuations in hydropower, wind, and photovoltaic power generation, the existing technology uses meteorological data to predict the power of hydropower, wind power, and photovoltaic power generation, thereby improving the accuracy of power prediction. However, the current power prediction methods all predict power for hydropower, wind power, and photovoltaic power generation separately, without considering the correlation between water, wind, and light energy sources. Therefore, how to comprehensively consider the mutual influence between water, wind, and light, and jointly predict the power of hydropower, wind, and photovoltaic power generation, is a problem that needs to be considered to further improve the accuracy and practicality of hydropower, wind power, and photovoltaic power prediction. Summary of the invention

[0003] In view of this, the embodiments of the present invention provide a method, device and equipment for joint prediction of water, wind and solar power generation, thereby improving the accuracy and practicality of the joint power prediction of water, wind and solar.

[0004] According to the first aspect, an embodiment of the present invention provides a method for jointly predicting hydropower, wind power and photovoltaic power generation, the method comprising: obtaining a data set, the data set comprising historical meteorological data as training samples and historical hydropower power data, historical wind power data and historical photovoltaic power data as labels; using the data set to train multiple hydropower power prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models; inputting current meteorological data into each power prediction model to predict the corresponding hydropower, wind power and photovoltaic power sets; selecting from the hydropower, wind power and photovoltaic power sets the optimal hydropower predicted power, the optimal wind power predicted power and the optimal photovoltaic predicted power under the condition of minimum residual load when hydropower, wind power and photovoltaic power are jointly output.

[0005] Optionally, acquiring the data set includes: dividing the historical meteorological data and the corresponding historical hydropower power data, historical wind power data and historical photovoltaic power data according to seasons and weather; according to the divided seasons and weather, identifying the main meteorological factors affecting hydropower, wind power and photovoltaic power from the historical meteorological data, and using the main meteorological factors as training samples.

[0006] Optionally, the use of the data set to train multiple hydropower power prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models includes: using the historical meteorological data and the historical hydropower data to train and verify multiple deep learning models to obtain the multiple hydropower power prediction models; using the historical meteorological data and the historical wind power data to train and verify multiple deep learning models to obtain the multiple wind power prediction models; using the historical meteorological data and the historical photovoltaic power data to train and verify multiple deep learning models to obtain the multiple photovoltaic power prediction models; using the historical meteorological data and the historical photovoltaic power data to train and verify multiple deep learning models to obtain the multiple photovoltaic power prediction models; using the mean absolute percentage error and the determination coefficient as evaluation indicators of prediction accuracy, selecting a preset number of power prediction models ranked in the front with a preset accuracy from the multiple hydropower power prediction models, the multiple wind power prediction models and the multiple photovoltaic power prediction models, as candidate models for power prediction.

[0007] Optionally, the current meteorological data is input into each power prediction model to predict the corresponding power sets of hydropower, wind power and photovoltaic power, including: inputting the current meteorological data into the candidate model under the corresponding season and weather conditions; using the hydropower prediction power sequence, wind power prediction power sequence and photovoltaic prediction power sequence output by each candidate model to respectively form the corresponding power sets of hydropower, wind power and photovoltaic power.

[0008] Optionally, the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power under the condition of minimum residual load when the combined output of water, wind and photovoltaic is selected from the power sets of hydropower, wind power and photovoltaic power, including: respectively combining the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence obtained from different models, performing sum operations to obtain multiple total power sequences; using the total load sequence and each total power sequence to calculate separately to obtain multiple residual load standard deviations; selecting the optimal total power sequence corresponding to the minimum residual load standard deviation, and using the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence used when summing and calculating the optimal total power sequence as the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power.

[0009] Optionally, the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power under the condition of minimum residual load when the combined output of water, wind and photovoltaic is selected from the power set of hydropower, wind power and photovoltaic, includes: respectively combining the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence obtained from different models, performing summation operation to obtain multiple total power sequences; using the total load sequence and each total power sequence to calculate respectively to obtain multiple residual load standard deviations; using each residual load standard deviation to create corresponding evaluation matrices from the perspective of hydropower, wind power and photovoltaic respectively; performing fuzzy optimization decision based on the created hydropower evaluation matrix, wind power evaluation matrix and photovoltaic evaluation matrix to determine the hydropower power prediction model, wind power prediction model and photovoltaic power prediction model with the largest relative superiority from the hydropower, wind power and photovoltaic power prediction models used; and using the predicted power output by the hydropower power prediction model, wind power prediction model and photovoltaic power prediction model with the largest relative superiority as the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power.

[0010] Optionally, before dividing the historical meteorological data and the corresponding historical hydropower power data, historical wind power data and historical photovoltaic power data according to season and weather, the method also includes: checking the abnormal values ​​and missing values ​​in the historical meteorological data, historical hydropower power data, historical wind power data and historical photovoltaic power data, deleting the historical meteorological data with missing power data, and correcting the abnormal power data using the K nearest neighbor complementary method.

[0011] According to the second aspect, an embodiment of the present invention provides a device for jointly predicting water, wind and photovoltaic power generation, the device comprising: a data acquisition module, used to obtain a data set, the data set comprising historical meteorological data as training samples and historical hydropower power data, historical wind power data and historical photovoltaic power data as labels; a model training module, used to use the data set to train multiple hydropower power prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models; an initial prediction module, used to input current meteorological data into each power prediction model to predict the corresponding hydropower, wind power and photovoltaic power sets; a prediction result optimization module, used to select the optimal hydropower prediction power, optimal wind power prediction power and optimal photovoltaic prediction power under the condition of minimum residual load when the water, wind and photovoltaic power sets are jointly output.

[0012] According to the third aspect, an embodiment of the present invention provides a joint prediction device for water, wind and solar power generation, including: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method described in the first aspect or any optional implementation manner of the first aspect by executing the computer instructions.

[0013] According to a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method described in the first aspect or any optional implementation manner of the first aspect.

[0014] The technical solution provided by this application has the following advantages:

[0015] The technical solution provided by the present application first obtains historical meteorological data within the historical time range and the corresponding historical hydropower power data, historical wind power data and historical photovoltaic power data. And use the historical meteorological data as the training sample of the machine learning model, use various types of historical power data as labels, and train multiple deep learning models for predicting power generation for hydropower, wind power and photovoltaic respectively. Then obtain the current meteorological data (for example, obtain the meteorological data of the day or the previous day according to the weather forecast), and then input the current meteorological data into all the power prediction models trained above in sequence, so as to obtain multiple hydropower prediction power sequences, multiple wind power prediction power sequences and multiple photovoltaic prediction power sequences. Then the output predicted power is stored in the corresponding power set according to the types of hydropower, wind power and photovoltaic respectively. After that, the hydropower prediction power sequence, wind power prediction power sequence and photovoltaic prediction power sequence are taken from each set for free combination and matching. The combined predicted power is applied to load consumption (for example, the power value required by various load equipment in the industrial park), and the best power combination is selected according to the minimum residual load, so that the prediction models of hydropower, wind power and photovoltaic are optimal and the power generation efficiency of multi-energy complementarity is the highest. On the one hand, it improves the accuracy of the combined power forecast of water, wind and solar power, and on the other hand, it enhances the application value of the combined power forecast results of water, wind and solar power in the preparation of short-term scheduling plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0017] Figure 1 A schematic diagram showing the steps of a method for joint prediction of hydropower, wind-solar power generation in one embodiment of the present invention is shown;

[0018] Figure 2A schematic diagram of the final scores of candidate models for power prediction on April 30, 2017 in one embodiment of the present invention is shown;

[0019] Figure 3 A schematic diagram of the final scores of candidate models for power prediction on July 30, 2017 in one embodiment of the present invention is shown;

[0020] Figure 4 A schematic diagram of the final scores of candidate models for power prediction on October 30, 2017 in one embodiment of the present invention is shown;

[0021] Figure 5 A schematic diagram of the final scores of candidate models for power prediction on December 31, 2017 in one embodiment of the present invention is shown;

[0022] Figure 6 A schematic diagram of the structure of a combined prediction device for hydropower, wind power and solar power generation in one embodiment of the present invention is shown;

[0023] Figure 7 A schematic structural diagram of a hydro-wind-solar power generation combined prediction device in one embodiment of the present invention is shown. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0025] See also Figure 1 In one embodiment, a method for jointly predicting hydropower, wind power and solar power generation comprises the following steps:

[0026] Step S101: Acquire a data set, where the data set includes historical meteorological data as training samples and historical hydropower data, historical wind power data, and historical photovoltaic power data as labels.

[0027] Step S102: using the data set to train multiple hydropower prediction models, multiple wind power prediction models, and multiple photovoltaic power prediction models.

[0028] Step S103: input the current meteorological data into each power prediction model to predict the corresponding power set of hydropower, wind power and photovoltaic power.

[0029] Step S104: Select the optimal hydropower predicted power, the optimal wind power predicted power and the optimal photovoltaic predicted power under the condition of minimum residual load when the water, wind and photovoltaic power are jointly output from the power sets of hydropower, wind power and photovoltaic power.

[0030] Specifically, considering that the prior art mainly focuses on power prediction of single energy sources such as water, wind and light, the mutual influence between water, wind and light has not been comprehensively considered to jointly predict the power of water, wind and light. Based on this, an embodiment of the present invention proposes a method for jointly predicting water, wind and light power. First, a data set is obtained, which includes historical meteorological data within a preset time range, such as meteorological data in the past year, and also includes historical hydropower data, historical wind power data and historical photovoltaic power data actually collected within the time range. Among them, the historical meteorological data is a training sample, and the historical hydropower data, historical wind power data and historical photovoltaic power data are all used as labels. Based on the above data, multiple deep learning models are trained and verified for hydropower, wind power and photovoltaic power, respectively, to obtain corresponding multiple hydropower prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models. The model algorithms used in this embodiment include but are not limited to BP neural network, LSTM neural network, bidirectional LSTM neural network (BI-LSTM), GRU neural network, bidirectional GRU neural network (BI-GRU) and extreme learning machine (ELM). After the training of each power prediction model is completed, the current meteorological data is collected, for example, the meteorological data of the day or the previous day is collected according to the weather forecast, and the inflow forecast data provided by the hydrological department is collected. Then the current meteorological data is input into each power prediction model to obtain the hydropower, wind power, and photovoltaic prediction power sequences, and the power data in the hydropower, wind power, and photovoltaic power sequences are stored in the power sets of hydropower, wind power, and photovoltaic according to the power generation type. After that, any predicted power sequence is taken out from the power sets of hydropower, wind power, and photovoltaic for summation operation, and different combinations of calculations correspond to multiple different total power sequences. Then, the residual load corresponding to each total power sequence after the total load consumption is supplied is calculated, and then the optimal hydropower prediction power, the optimal wind power prediction power, and the optimal photovoltaic prediction power are obtained according to the total power under the minimum residual load. Thereby, the prediction models of hydropower, wind power, and photovoltaic are optimized, the power generation accuracy of multi-energy complementarity is the highest, and the accuracy of the joint prediction of water, wind, and photovoltaic power generation is improved.

[0031] Specifically, in one embodiment, the above step S101 specifically includes the following steps:

[0032] Step 1: Divide the historical meteorological data and the corresponding historical hydropower data, historical wind power data, and historical photovoltaic power data according to season and weather.

[0033] Step 2: According to the divided seasons and weather, identify the main meteorological factors that affect hydropower, wind power, and photovoltaic power from historical meteorological data.

[0034] Step 3: Use the main meteorological factors as training samples.

[0035] Specifically, in this embodiment, in order to further improve the accuracy of model training, the training samples are first preprocessed, and the historical meteorological data obtained from the water-wind-solar multi-energy complementary power generation system includes a variety of meteorological factors, such as wind speed, wind direction, air pressure, precipitation, runoff, direct radiation, scattered radiation and temperature. In addition, for hydropower, the training samples can also include the water level, outflow flow, output data of the hydropower station, and the dispatching procedures of the hydropower station. Afterwards, the historical meteorological data of water, wind, and light and the corresponding power data are divided into spring, summer, autumn, and winter four seasons data, and the photovoltaic data is mainly divided into three types according to the data of each season: sunny days, cloudy days, and rainy and snowy days; the season division adopts the Gregorian calendar: March to May is spring, June to August is summer, September to November is autumn, and December to February is winter. Then, the main meteorological factors of water, wind and light participating in the training are selected according to the season and weather. Specifically, the Pearson correlation coefficient between the meteorological factors of wind power and hydropower data in different seasons and their power generation is calculated, and the Pearson correlation coefficient between the meteorological factors of photovoltaic power generation data in different seasons and weather and their power generation is calculated. The Pearson correlation coefficient calculation formula is as follows:

[0036]

[0037] Where: x i is the value of a meteorological factor in the i-th period, y i is the power generation value in the i-th period, is the average value of a certain meteorological factor, is the average value of the generated power, and n is the total number of time periods in the entire data set.

[0038] After that, the correlation coefficient is greater than the preset threshold (ρ xy The factors with a value of ≥0.6) are used as the basis for dividing the main meteorological factors and the input factors of the prediction model.

[0039] In addition, in one embodiment, the pre-processed data can be further clustered according to the divided seasons and weather to further improve the accuracy of the data season and weather division. Taking photovoltaic data as an example, the specific method is: cluster analysis of photovoltaic power generation data based on the K-means++ algorithm, using the mean and variance of the screened meteorological factors as input factors of the K-means++ algorithm for cluster analysis, and using the silhouette coefficient to obtain the best clustering results of photovoltaic power generation data in different seasons and weather conditions {K spr , K sum , Kaut , K win}(where K spr , K sum , K aut , K win Represent the best clustering results of different weather type data in spring, summer, autumn and winter respectively).

[0040] Therefore, the power prediction model can be trained in a targeted manner using data from different seasons and weather conditions to obtain model sets under different seasons and weather conditions. This will facilitate the selection of a more matching model set for prediction in the subsequent prediction process based on the season and weather of the current meteorological data, thereby further improving the accuracy of power prediction.

[0041] Specifically, in one embodiment, before the above step one, the outliers and missing values ​​in the historical meteorological data, historical hydropower data, historical wind power data, and historical photovoltaic power data are also checked, and the historical meteorological data with missing power data is deleted, and the abnormal power data is corrected using the K nearest neighbor complementary method. Specifically, the outliers and missing values ​​of meteorological and power data are preprocessed. Data that is missing for more than 16 consecutive time periods is regarded as missing data and deleted, and the outliers that appear are identified using a box plot, and processed using the K nearest neighbor complementary method, so as to further improve the accuracy of the data. The calculation formula is as follows:

[0042]

[0043] In the formula, represents the value after K-nearest neighbor complementary method, x i-k Indicates an outlier x i The sample value of the kth period after, x i+k Indicates an outlier x i The sample value of the kth period after .

[0044] Specifically, in this embodiment, the above step S102 specifically includes the following steps:

[0045] Step 4: Use historical meteorological data and historical hydropower data to train and verify multiple deep learning models to obtain multiple hydropower prediction models.

[0046] Step 5: Use historical meteorological data and historical wind power data to train and verify multiple deep learning models to obtain multiple wind power prediction models.

[0047] Step 6: Use historical meteorological data and historical photovoltaic power data to train and verify multiple deep learning models to obtain multiple photovoltaic power prediction models.

[0048] Step 7: Using the mean absolute percentage error and the coefficient of determination as evaluation indicators of prediction accuracy, select a preset number of power prediction models ranked in front of a preset accuracy from multiple hydropower power prediction models, multiple wind power prediction models, and multiple photovoltaic power prediction models as candidate models for power prediction.

[0049] Specifically, in this embodiment, several deep learning models are pre-initialized, such as BP neural network, LSTM neural network, bidirectional LSTM neural network (BI-LSTM), GRU neural network, bidirectional GRU neural network (BI-GRU) and extreme learning machine (ELM), and 4 to 10 models can be selected for training. The preprocessed data set is divided into three parts: training set (60%), verification set (20%) and test set (20%). Each model is trained with the training set, verified and tested on the verification set and test set, and the model parameters are adjusted reasonably.

[0050] Taking photovoltaic data as an example, the same is true for wind power data and hydropower data. The test set data divided by season and weather is input into each model trained with the training set, and then the mean absolute percentage error MAPE and determination coefficient R of the photovoltaic power generation prediction output by each model are calculated. 2 , used to evaluate the accuracy of each power prediction model, among which MAPE and R 2 The calculation formula is:

[0051]

[0052]

[0053] In the formula, MAPE k and They represent the mean absolute percentage error and determination coefficient of the k-th deep learning model in predicting photovoltaic power, represents the photovoltaic power prediction value of the i-th period obtained by using the k-th deep learning model, y i represents the true value of the power in the i-th period of the corresponding test set, represents the average value of the true power value in the i-th period of the corresponding test set, and m represents the number of samples of photovoltaic data divided by season and weather in the test set.

[0054] Afterwards, a preset number of power prediction models with the highest accuracy are selected based on the calculation results (for example, 3 to 5 power prediction models with the highest accuracy ranking), thereby further improving the accuracy of model training and improving the accuracy of subsequent power predictions.

[0055] Specifically, in this embodiment, the above step S103 specifically includes the following steps:

[0056] Step 8: Input the current meteorological data into the corresponding candidate model under the corresponding season and weather conditions.

[0057] Step nine: Use the hydropower prediction power sequence, wind power prediction power sequence and photovoltaic prediction power sequence output by each candidate model to form corresponding hydropower, wind power and photovoltaic power sets respectively.

[0058] Specifically, to predict the power generation according to the current meteorological forecast information, first obtain the meteorological factors selected in the above step 2 from the current meteorological data, then calculate the mean and variance of the forecast data for the day to be predicted according to the meteorological factors, and use the calculation results to divide the day into the season and weather categories corresponding to the water, wind, and light data, and select the prediction model set corresponding to the current season and weather according to the corresponding season and weather category, and then select the candidate models with the highest accuracy from the prediction model set according to the average absolute percentage error and the determination coefficient to perform power prediction. Thus, through the steps of this embodiment, a more accurate prediction power sequence of different models of water, wind, and light power generation on that day is calculated, for example, the prediction power sequences output by the three most accurate hydropower prediction models, the three most accurate wind power prediction models, and the three most accurate photovoltaic power prediction models are:

[0059]

[0060]

[0061]

[0062] In the set, the superscripts h, w, and p represent hydropower, wind power, and photovoltaic power, respectively; the subscripts 1, 2, and 3 represent the top three prediction models selected from the candidate models; and pow represents the predicted power sequence (a power sequence can be understood as the power prediction value of the model for each period of time within a period of time). For example, this parameter represents the power forecast value of the first model in the i-th period in the wind power forecast set. The meanings of other parameters in the power set are similar.

[0063] Specifically, in this embodiment, the above step S104 specifically includes the following steps:

[0064] Step 10: Combine the hydropower prediction power series, wind power prediction power series and photovoltaic prediction power series obtained from different models respectively, perform sum operations, and obtain multiple total power series.

[0065] Step 11: Calculate multiple residual load standard deviations using the total load sequence and the total power sequence.

[0066] Step 12: Select the optimal total power sequence corresponding to the minimum standard deviation of the residual load, and use the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence used in summing up and calculating the optimal total power sequence as the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power.

[0067] Specifically, a predicted power sequence is randomly selected from the hydropower power set, the wind power set and the photovoltaic power set, and the three predicted power sequences are summed to obtain a total power sequence. Multiple total power sequences can be obtained through multiple groups of different combinations. For each total power sequence, the difference between it and the total load sequence is calculated to obtain the residual load sequence, and then the standard deviation of the residual load sequence is calculated. The residual load standard deviation corresponding to each total power sequence can be obtained, and the minimum residual load situation can be determined based on the minimum residual standard deviation. When the residual standard deviation of a certain combination is the smallest, the corresponding predicted power sequence taken out from the power set is used as the optimal hydropower predicted power, the optimal wind power predicted power and the optimal photovoltaic predicted power.

[0068] The residual load standard deviation is calculated as follows:

[0069]

[0070] δ i,c =L i -SP i,c

[0071]

[0072] In the formula, sl c represents the standard deviation of the residual load of the cth model combination, m represents the total number of test sets, L i represents the total load in the ith period, SP i,c Represents the sum of the predicted values ​​of water, wind and solar power generation in the i-th period under the c-th model combination. δ i，c represents the residual load in the i-th period under the c-th model combination, represents the average residual load in all periods under the cth model combination, where c represents different candidate model combinations of the three energy sources and takes values ​​of 1 to 27.

[0073] By finding the smallest residual standard deviation among all combinations, the optimal hydropower forecast power, the optimal wind power forecast power and the optimal photovoltaic forecast power, as well as the power forecast models with the highest accuracy for each, can be determined.

[0074] Specifically, in another embodiment, the above step S104 specifically includes the following steps:

[0075] Step 13: Combine the hydropower prediction power series, wind power prediction power series and photovoltaic prediction power series obtained from different models respectively, perform sum operations, and obtain multiple total power series.

[0076] Step 14: Calculate multiple residual load standard deviations using the total load sequence and the total power sequence.

[0077] Step 15: Create corresponding evaluation matrices from the perspectives of hydropower, wind power and photovoltaic power using the standard deviation of each residual load.

[0078] Step 16: Perform fuzzy optimization decision based on the created hydropower evaluation matrix, wind power evaluation matrix and photovoltaic evaluation matrix, and determine the hydropower power prediction model, wind power prediction model and photovoltaic power prediction model with the greatest relative superiority from the hydropower, wind power and photovoltaic power prediction models used.

[0079] Step 17: The predicted power output by the hydropower prediction model, wind power prediction model and photovoltaic power prediction model with the largest relative superiority is taken as the optimal hydropower prediction power, the optimal wind power prediction power and the optimal photovoltaic prediction power.

[0080] Specifically, in this embodiment, after calculating the residual standard deviations of various combinations, the minimum residual standard deviation is not directly used for judgment, but the residual standard deviations of each combination are processed by the fuzzy optimal decision method to select the optimal predicted power under the minimum residual load condition through analysis and calculation, thereby further improving the accuracy of the optimal predicted power. Specifically, the corresponding evaluation matrix is ​​first created from the perspectives of hydropower, wind power and photovoltaic power using the residual load standard deviation.

[0081] The standard deviation of the residual load at the angle of interest is calculated as follows:

[0082]

[0083] δ i | nn,b =L i -SP i | nn,b

[0084]

[0085] In the formula, sl nn,b represents the standard deviation of the residual load of this model combination, m represents the total number of test sets, L i represents the total load in the ith period, SP i | nn,b Represents the sum of the predicted values ​​of water, wind and solar power generation in the ith period under this model combination. i |nn,b represents the residual load in the i-th period under this model combination, Represents the average residual load of all periods under this model combination. Tag | nn,b Represents the model combination method. The subscript nn represents the candidate model number of the current focus angle (hydropower, wind power or photovoltaic), which can be 1, 2, or 3. The subscript b represents different candidate model combinations of the other two energy sources, which can be 1 to 9.

[0086] Taking three candidate models for hydropower, wind power and photovoltaic power prediction as examples, the evaluation matrix created according to the definition of fuzzy optimal decision-making is as follows:

[0087] Hydropower Evaluation Matrix:

[0088] Wind power evaluation matrix:

[0089] Photovoltaic evaluation matrix:

[0090] The above evaluation matrix can be understood as creating evaluation matrices respectively by focusing more on the three angles of hydropower, wind power and photovoltaic power, so as to find a more accurate predicted power value. In the hydropower evaluation matrix, each row represents the standard deviation of the remaining load formed by the combination of the same hydropower power prediction model with different wind power and different photovoltaic power prediction models; each column represents the standard deviation of the remaining load formed by the combination of three different candidate hydropower power prediction models with the same wind power prediction model and the same photovoltaic power prediction model, thus forming an evaluation matrix of three rows and nine columns. The wind power evaluation matrix is ​​the same as the photovoltaic evaluation matrix. For ease of distinction, the letters k, f, and g are used to represent the row numbers of the hydropower evaluation matrix, the wind power evaluation matrix, and the photovoltaic evaluation matrix, respectively.

[0091] Afterwards, the entropy weight method is used to calculate the weight of each indicator in the above matrix:

[0092] For the hydropower evaluation matrix:

[0093] For wind power evaluation matrix:

[0094] For the photovoltaic evaluation matrix:

[0095] In the formula, They represent the weights corresponding to the indicators in column b in the hydropower evaluation matrix, wind power evaluation matrix, and photovoltaic evaluation matrix, respectively. They represent the information entropy corresponding to the indicators in column b in the hydropower evaluation matrix, wind power evaluation matrix, and photovoltaic evaluation matrix, respectively. They represent the probability values ​​of the indicators in the kth row and bth column of the hydropower evaluation matrix, the fth row and bth column of the wind power evaluation matrix, and the gth row and bth column of the photovoltaic evaluation matrix, respectively. The values ​​of k, f, or g can all be 1 to 3.

[0096] Then, the relative superiority of the water, wind, and light prediction models is calculated using the above evaluation matrix and index weights, so as to select the model with the lowest power prediction effect that can achieve the lowest residual load. The relative superiority vectors of the three candidate model sets are: The parameter represents the relative superiority of the corresponding candidate model, and the specific calculation formula is as follows:

[0097] where k = 1, 2, 3

[0098] Where f=1,2,3

[0099] Where g = 1, 2, 3

[0100] In the formula, They represent the normalized results of the indicators in the kth row and bth column of the hydropower evaluation matrix, the fth row and bth column of the wind power evaluation matrix, and the gth row and bth column of the photovoltaic evaluation matrix. The normalization principle is as follows:

[0101] where k = 1, 2, 3

[0102] Where f=1,2,3

[0103] Where g = 1, 2, 3

[0104] After obtaining the three relative superiority vectors of hydropower, wind power and photovoltaic power, the optimal prediction model is selected according to the maximum membership principle, which is calculated according to the following formula:

[0105]

[0106]

[0107]

[0108] For example: The relative membership vector of the hydropower candidate model set is According to the formula Calculate BM h=2, the power prediction model with the highest accuracy in the hydropower candidate model set is the second candidate model, and its prediction results will eventually be used for short-term scheduling and other businesses. The calculation process of wind power and photovoltaic power is the same. Through the above steps, the optimal hydropower prediction power, optimal wind power prediction power and optimal photovoltaic prediction power with higher accuracy under the minimum residual load are obtained from the perspectives of focusing on hydropower, focusing on wind power and focusing on photovoltaic power.

[0109] Specifically, the combined prediction method of hydropower, wind power and solar power generation is explained below using an actual scenario embodiment:

[0110] 1. Obtain meteorological data and power data. Obtain total radiation, 70m wind speed, temperature, precipitation, runoff, potential evaporation, wind power, photovoltaic power and hydropower station output; the time resolution of all data is 1 hour, and the data series length is 2014-2016.

[0111] 2. Data preprocessing. No missing data need to be processed, and abnormal data are identified using box plots. Since meteorological data has strong randomness, the embodiment of the present invention uses the K nearest neighbor complementary method to process single abnormal values, and the K value is 2.

[0112] 3. Weather and season classification. The water, wind and solar power generation data are divided into four categories: spring, summer, autumn and winter; and the photovoltaic power generation data are further divided into sunny days, rainy and snowy days, and cloudy days according to weather conditions.

[0113] 4. Select the main meteorological factors involved in the training of water, wind and light data. For hydropower prediction, the main meteorological factor is the historical runoff data; for wind power prediction, the main meteorological factor is the historical 70m wind speed data; for photovoltaic prediction, the main meteorological factor is the historical total irradiation and temperature data.

[0114] 5. Cluster analysis of water, wind and light data based on K-means++ algorithm. For spring data, the light measurement data is divided into 100 sunny days, 14 rainy and snowy days, and 65 cloudy days; the wind measurement data is divided into 2 categories, 57 days in the first category and 122 days in the second category; the hydraulic data is divided into 3 categories, 43 days in the first category, 48 days in the second category, and 88 days in the third category. For summer data, the light measurement data is divided into 108 sunny days, 10 rainy and snowy days, and 65 cloudy days; the wind measurement data is divided into 3 categories, 94 days in the first category, 71 days in the second category, and 18 days in the third category; the hydraulic data is divided into 3 categories, 75 days in the first category, 104 days in the second category, and 4 days in the third category. For autumn data, the photometric data are divided into 66 sunny days, 48 ​​rainy and snowy days, and 104 cloudy days; the wind measurement data are divided into 2 categories, 54 days in the first category, and 129 days in the second category; the hydraulic data are divided into 2 categories, 152 days in the first category, and 31 days in the second category; for winter data, the photometric data are divided into 67 sunny days, 6 rainy and snowy days, and 110 cloudy days; the wind measurement data are divided into 2 categories, 159 days in the first category, and 25 days in the second category; the hydraulic data are divided into three categories, 85 days in the first category, 25 days in the second category, and 74 days in the third category.

[0115] 6. Build a power generation prediction model based on multiple deep learning algorithms. Input the training and validation data of different categories of water, wind, and light into the LSTM, BI-LSTM, GRU, BI-GRU, BP, and ELM models to adjust the model parameters, and input the test set into the model. According to the average absolute error and determination coefficient of different categories of models, determine several candidate models for power prediction with the highest accuracy. The hydropower results of each season are shown in Tables 1 to 4, the wind power results of each season are shown in Tables 5 to 8, and the photovoltaic results of each season are shown in Tables 9 to 12.

[0116] Table 1 Effects of various hydropower generation prediction models in spring and candidate models

[0117]

[0118] Table 2 Effects of various hydropower generation prediction models in summer and candidate models

[0119]

[0120]

[0121] Table 3 Effects of various hydropower generation prediction models in autumn and candidate models

[0122]

[0123] Table 4 Effects of various hydropower generation power prediction models in winter and candidate models

[0124]

[0125]

[0126] Table 5 Effects of various wind power prediction models and candidate models in spring

[0127]

[0128] Table 6 Effects of various wind power prediction models and candidate models in summer

[0129]

[0130] Table 7 Effects of various wind power prediction models and candidate models in autumn

[0131]

[0132]

[0133] Table 8 Effects of various wind power prediction models and candidate models in winter

[0134]

[0135] Table 9 Effects of various power prediction models and candidate models for photovoltaic power generation in spring

[0136]

[0137]

[0138] Table 10 Effects of various power prediction models and candidate models for photovoltaic power generation in summer

[0139]

[0140] Table 11 Effects of various power prediction models and candidate models for photovoltaic power generation in autumn

[0141]

[0142]

[0143] Table 12 Effects of various power prediction models and candidate models for photovoltaic power generation in winter

[0144]

[0145] 7. Combine the day-ahead forecast information to realize the adaptive joint prediction of water, wind and solar power generation. April 30, July 30, October 30 and December 31, 2017 are selected as the days to be predicted. The seasons and water, wind and solar categories of the four days to be predicted are shown in Table 13.

[0146] Table 13 Season and weather classification of the test days

[0147]

[0148] Table 13 is used to query the candidate models for each test day from Tables 1 to 12 above, as shown in Table 14.

[0149] Table 14. Candidate models for the test day

[0150]

[0151] Based on Table 14, the remaining load standard deviation is calculated according to the output results of the candidate models of each power generation type, and the remaining load standard deviation is evaluated by fuzzy optimal decision making. The results of the optimal prediction model are obtained according to the relative superiority as follows:

[0152] Depend on Figure 2 It can be seen that on April 30, 2017: for hydropower data, the final score of LSTM was 0.66, the final score of GRU was 0.71, and the final score of BI-GRU was 0.34, so the best prediction model was GRU; for wind power prediction data, the final score of GRU was 1, the final score of BI-GRU was 0.01, and the final score of ELM was 0, so the best prediction model was GRU; for photovoltaic prediction data, the final score of BI-GRU was 1, the final score of BP was 0.00094, and the final score of ELM was 0, so the best prediction model was BI-GRU.

[0153] Depend on Figure 3 It can be seen that on July 30, 2017: for hydropower data, the final score of LSTM was 1, the final score of GRU was 0, and the final score of BI-GRU was 0.80, so the best prediction model was LSTM; for wind power prediction data, the final score of GRU was 0.58, the final score of BI-GRU was 1, and the final score of ELM was 0, so the best prediction model was BI-GRU; for photovoltaic prediction data, the final score of BI-LSTM was 0.01, the final score of GRU was 1, and the final score of BI-GRU was 0, so the best prediction model was GRU.

[0154] Depend on Figure 4It can be seen that on October 30, 2017: for hydropower data, the final score of LSTM was 0.001, the final score of GRU was 1, and the final score of BI-GRU was 0, so the best prediction model was GRU; for wind power prediction data, the final score of EGRU was 0.41, the final score of BP was 0.02, and the final score of LM was 0.99, so the best prediction model was ELM; for photovoltaic prediction data, the final score of BI-LSTM was 1, the final score of GRU was 0.20, and the final score of BI-GRU was 0, so the best prediction model was BI-LSTM.

[0155] Depend on Figure 5 It can be seen that on December 31, 2017: for hydropower data, the final score of BI-LSTM was 0.003, the final score of GRU was 0.028, and the final score of BI-GRU was 1, so the best prediction model was BI-GRU; for wind power prediction data, the final score of LSTM was 0.46, the final score of BP was 0.15, and the final score of ELM was 0.77, so the best prediction model was ELM; for photovoltaic prediction data, the final score of BI-GRU was 1, the final score of BP was 0.15, and the final score of ELM was 0, so the best prediction model was BI-GRU.

[0156] Finally, the predicted power output by the best prediction model for each of the above-mentioned test days is obtained as the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power for each test day. Based on this, a day-ahead power generation scheduling plan is compiled to guide the short-term scheduling and operation of the hydro-wind-solar multi-energy complementary system.

[0157] Through the above steps, the technical solution provided by the present application first obtains historical meteorological data within the historical time range and the corresponding historical hydropower power data, historical wind power data and historical photovoltaic power data. And use the historical meteorological data as a training sample for the prediction model based on the deep learning algorithm, and use various types of historical power data as labels, and train multiple models for predicting power generation for hydropower, wind power and photovoltaics, respectively, so as to obtain multiple hydropower power prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models. Then obtain the current meteorological data (for example, obtain the meteorological data of the day or the previous day according to the weather forecast), and then input the current meteorological data into all the power prediction models trained above in turn, so as to obtain multiple hydropower prediction power sequences, multiple wind power prediction power sequences and multiple photovoltaic prediction power sequences. Then the output prediction power sequences are stored in the corresponding power sets according to the types of hydropower, wind power and photovoltaics. After that, take a hydropower prediction power sequence, a wind power prediction power sequence and a photovoltaic prediction power sequence from each set for free combination. The combined total predicted power sequence is applied to load consumption (for example, the power values ​​required by various load equipment in an industrial park), and then the best predicted power combination is selected according to the minimum residual load. This makes the prediction models of hydropower, wind power and photovoltaics optimal and the power generation efficiency of multi-energy complementarity highest, thereby improving the accuracy of the joint prediction of hydropower, wind power and photovoltaic power generation.

[0158] like Figure 6 As shown, this embodiment also provides a combined prediction device for hydropower, wind power and solar power generation, the device comprising:

[0159] The data acquisition module 101 is used to obtain a data set, which includes historical meteorological data as training samples and historical hydropower data, historical wind power data and historical photovoltaic power data as labels. For details, please refer to the relevant description of step S101 in the above method embodiment, which will not be repeated here.

[0160] The model training module 102 is used to train multiple hydropower prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models using the data set. For details, please refer to the relevant description of step S102 in the above method embodiment, which will not be repeated here.

[0161] The initial prediction module 103 is used to input the current meteorological data into each power prediction model to predict the corresponding power set of hydropower, wind power and photovoltaic power. For details, please refer to the relevant description of step S103 in the above method embodiment, which will not be repeated here.

[0162] The prediction result optimization module 104 is used to select the optimal hydropower prediction power, the optimal wind power prediction power and the optimal photovoltaic prediction power under the condition of minimum residual load when the hydropower, wind power and photovoltaic power are combined from the power sets of hydropower, wind power and photovoltaic power. For details, please refer to the relevant description of step S104 in the above method embodiment, which will not be repeated here.

[0163] The combined prediction device for water, wind and solar power generation provided in an embodiment of the present invention is used to execute the combined prediction method for water, wind and solar power generation provided in the above embodiment. Its implementation method and principle are the same. For details, please refer to the relevant description of the above method embodiment and will not be repeated here.

[0164] Through the collaborative cooperation of the above-mentioned components, the technical solution provided by this application first obtains historical meteorological data within the historical time range and the corresponding historical hydropower power data, historical wind power data and historical photovoltaic power data. And use the historical meteorological data as the training sample of the machine learning model, and use various types of historical power data as labels, and train multiple models for predicting power generation for hydropower, wind power and photovoltaics, respectively, so as to obtain multiple hydropower power prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models. Then obtain the current meteorological data (for example, obtain the meteorological data of the day or the previous day according to the weather forecast), and then input the current meteorological data into all the power prediction models trained above in turn, so as to obtain multiple hydropower prediction power sequences, multiple wind power prediction power sequences and multiple photovoltaic prediction power sequences. Then the output prediction power sequences are stored in the corresponding power sets according to the types of hydropower, wind power and photovoltaics. After that, take a hydropower prediction power sequence, a wind power prediction power sequence and a photovoltaic prediction power sequence from each set for free combination and matching. The combined total predicted power sequence is applied to load consumption (for example, the power values ​​required by various load equipment in an industrial park), and then the optimal power combination is selected according to the minimum residual load, so that the prediction models of hydropower, wind power and photovoltaics are optimal and the power generation efficiency of multi-energy complementarity is highest, thereby improving the accuracy of the joint prediction of hydropower, wind power and photovoltaic power generation.

[0165] Figure 7 A hydro-wind-solar power generation combined prediction device according to an embodiment of the present invention is shown. The device includes a processor 901 and a memory 902, which can be connected via a bus or other means. Figure 7 The example of connecting through bus is taken in the following.

[0166] The processor 901 may be a central processing unit (CPU). The processor 901 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0167] The memory 902 is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the methods in the above method embodiments. The processor 901 executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory 902, that is, implementing the methods in the above method embodiments.

[0168] The memory 902 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required by at least one function; the data storage area may store data created by the processor 901, etc. In addition, the memory 902 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 902 may optionally include a memory remotely arranged relative to the processor 901, and these remote memories may be connected to the processor 901 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0169] One or more modules are stored in the memory 902 , and when executed by the processor 901 , the method in the above method embodiment is executed.

[0170] The specific details of the above-mentioned water-wind-solar power generation joint prediction device can be understood by referring to the corresponding relevant descriptions and effects in the above-mentioned method embodiment, and will not be repeated here.

[0171] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the implemented program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0172] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for joint prediction of hydropower, wind and solar power generation, characterized in that: The method comprises: Acquire a data set, the data set including historical meteorological data as training samples and historical hydropower data, historical wind power data, and historical photovoltaic power data as labels; Using the data set to train multiple hydropower prediction models, multiple wind power prediction models, and multiple photovoltaic power prediction models; Input the current meteorological data into each power prediction model to predict the corresponding power combination of hydropower, wind power and photovoltaic power; Selecting the optimal hydropower forecast power, the optimal wind power forecast power and the optimal photovoltaic forecast power under the condition of minimum residual load when the hydropower, wind power and photovoltaic power are jointly output from the power sets of hydropower, wind power and photovoltaic power; The method of using the data set to train multiple hydropower prediction models, multiple wind power prediction models, and multiple photovoltaic power prediction models includes: using the historical meteorological data and the historical hydropower data to train and verify multiple deep learning models to obtain the multiple hydropower prediction models; using the historical meteorological data and the historical wind power data to train and verify multiple deep learning models to obtain the multiple wind power prediction models; using the historical meteorological data and the historical photovoltaic power data to train and verify multiple deep learning models to obtain the multiple photovoltaic power prediction models; using the mean absolute percentage error and the determination coefficient as evaluation indicators of prediction accuracy, respectively selecting a preset number of power prediction models ranked in front of a preset accuracy from the multiple hydropower prediction models, the multiple wind power prediction models, and the multiple photovoltaic power prediction models as candidate models for power prediction; The method of inputting the current meteorological data into each power prediction model to predict the corresponding power sets of hydropower, wind power and photovoltaic power includes: inputting the current meteorological data into the candidate model under the corresponding season and weather conditions; using the hydropower prediction power sequence, wind power prediction power sequence and photovoltaic prediction power sequence output by each candidate model to form the corresponding power sets of hydropower, wind power and photovoltaic power respectively; The optimal hydropower predicted power, the optimal wind power predicted power and the optimal photovoltaic predicted power under the condition of minimum residual load when the water, wind and photovoltaic power are jointly output from the power sets of hydropower, wind power and photovoltaic power, including: respectively combining the hydropower predicted power sequence, the wind power predicted power sequence and the photovoltaic predicted power sequence obtained by different models, performing summation operations to obtain multiple total power sequences; using the total load sequence and each total power sequence to calculate respectively to obtain multiple residual load standard deviations; selecting the optimal total power sequence corresponding to the minimum residual load standard deviation, and using the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence used when summing and calculating the optimal total power sequence as the optimal hydropower predicted power, the optimal wind power predicted power and the optimal photovoltaic predicted power; or, The optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power under the condition of minimum residual load when the combined output of water, wind and photovoltaic is selected from the power set of hydropower, wind power and photovoltaic, includes: respectively combining the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence obtained from different models, performing summation operation to obtain multiple total power sequences; using the total load sequence and each total power sequence to calculate respectively to obtain multiple residual load standard deviations; using each residual load standard deviation to create corresponding evaluation matrices from the perspectives of hydropower, wind power and photovoltaic respectively; performing fuzzy optimization decision-making based on the created hydropower evaluation matrix, wind power evaluation matrix and photovoltaic evaluation matrix to determine the hydropower power prediction model, wind power prediction model and photovoltaic power prediction model with the largest relative superiority from the hydropower, wind power and photovoltaic power prediction models used; and using the predicted power output by the hydropower power prediction model, wind power prediction model and photovoltaic power prediction model with the largest relative superiority as the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power.

2. The method according to claim 1, characterized in that The acquiring of the data set comprises: Dividing the historical meteorological data and the corresponding historical hydropower data, historical wind power data and historical photovoltaic power data according to season and weather; According to the divided seasons and weather, main meteorological factors affecting hydropower, wind power and photovoltaic power are identified from the historical meteorological data, and the main meteorological factors are used as training samples.

3. The method according to claim 2, characterized in that Before dividing the historical meteorological data and the corresponding historical hydropower data, historical wind power data, and historical photovoltaic power data according to seasons and weather, the method further includes: Check the abnormal values ​​and missing values ​​in the historical meteorological data, historical hydropower power data, historical wind power data and historical photovoltaic power data, delete the historical meteorological data with missing power data, and correct the abnormal power data using the K nearest neighbor complementary method.

4. A combined prediction device for hydropower, wind power and solar power generation, characterized in that: The device comprises: A data acquisition module, used to acquire a data set, wherein the data set includes historical meteorological data as training samples and historical hydropower data, historical wind power data, and historical photovoltaic power data as labels; A model training module, used to train multiple hydropower prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models using the data set; the training of multiple hydropower prediction models, multiple wind power prediction models and multiple photovoltaic power prediction models using the data set includes: using the historical meteorological data and the historical hydropower data to train and verify multiple deep learning models to obtain the multiple hydropower prediction models; using the historical meteorological data and the historical wind power data to train and verify multiple deep learning models to obtain the multiple wind power prediction models; using the historical meteorological data and the historical photovoltaic power data to train and verify multiple deep learning models to obtain the multiple photovoltaic power prediction models; using the mean absolute percentage error and the determination coefficient as evaluation indicators of prediction accuracy, respectively selecting a preset number of power prediction models ranked in front of a preset accuracy from the multiple hydropower prediction models, the multiple wind power prediction models and the multiple photovoltaic power prediction models as candidate models for power prediction; The initial prediction module is used to input the current meteorological data into each power prediction model to predict the corresponding power set of hydropower, wind power and photovoltaic power; the inputting the current meteorological data into each power prediction model to predict the corresponding power set of hydropower, wind power and photovoltaic power includes: inputting the current meteorological data into the candidate model under the corresponding season and weather conditions; using the hydropower prediction power sequence, wind power prediction power sequence and photovoltaic prediction power sequence output by each candidate model to form the corresponding power set of hydropower, wind power and photovoltaic power respectively; A prediction result optimization module is used to select the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power under the condition of minimum residual load when the water, wind and photovoltaic power are jointly output from the power sets of hydropower, wind power and photovoltaic power; the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power under the condition of minimum residual load when the water, wind and photovoltaic power are jointly output from the power sets of hydropower, wind power and photovoltaic power, including: respectively combining the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence obtained by different models, performing summation operations to obtain multiple total power sequences; using the total load sequence and each total power sequence to calculate separately to obtain multiple residual load standard deviations; selecting the optimal total power sequence corresponding to the minimum residual load standard deviation, and using the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence used when summing and calculating the optimal total power sequence as the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power; or, The optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power under the condition of minimum residual load when the combined output of water, wind and photovoltaic is selected from the power set of hydropower, wind power and photovoltaic, includes: respectively combining the hydropower predicted power sequence, wind power predicted power sequence and photovoltaic predicted power sequence obtained from different models, performing summation operation to obtain multiple total power sequences; using the total load sequence and each total power sequence to calculate respectively to obtain multiple residual load standard deviations; using each residual load standard deviation to create corresponding evaluation matrices from the perspectives of hydropower, wind power and photovoltaic respectively; performing fuzzy optimization decision-making based on the created hydropower evaluation matrix, wind power evaluation matrix and photovoltaic evaluation matrix to determine the hydropower power prediction model, wind power prediction model and photovoltaic power prediction model with the largest relative superiority from the hydropower, wind power and photovoltaic power prediction models used; and using the predicted power output by the hydropower power prediction model, wind power prediction model and photovoltaic power prediction model with the largest relative superiority as the optimal hydropower predicted power, optimal wind power predicted power and optimal photovoltaic predicted power.

5. A combined prediction device for hydropower, wind and solar power generation, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 3 by executing the computer instructions.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method according to any one of claims 1 to 3.

Images