Photovoltaic power prediction method based on LSTM and EBRB
By combining LSTM and EBRB in a stacking ensemble model, the nonlinearity and interpretability issues in photovoltaic power generation prediction are resolved, achieving higher accuracy and more stable prediction results.
Patent Information
- Application Number
- CN202411974659.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-11-07
- Estimated Expiration
- 2044-12-30
AI Technical Summary
Existing photovoltaic power generation prediction methods perform poorly when dealing with complex nonlinear relationships and large amounts of data, and deep learning models lack interpretability.
A Stacking ensemble model based on LSTM and EBRB is adopted. The extreme gradient boosting algorithm is used to extract the influencing factors of photovoltaic power generation, an LSTM model is constructed to capture nonlinear and long-term dependencies, an EBRB model is used to assess data uncertainty, and a linear regression model is combined for prediction.
It improves the accuracy of photovoltaic power generation prediction and the interpretability of the model, providing higher prediction accuracy and stability.
Smart Images
Figure CN119918728B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the field of photovoltaic power prediction, and particularly relates to a photovoltaic power prediction method based on LSTM and EBRB. BACKGROUND
[0002] With the transformation of global energy structure and the rapid development of renewable energy, photovoltaic power generation as an important form of clean energy has achieved significant growth in installed capacity and power generation. However, photovoltaic power generation has randomness and volatility, mainly affected by factors such as light intensity, weather conditions, seasonal changes, etc., which brings challenges to power grid dispatching and operation, and also affects the economic benefits and market competitiveness of photovoltaic power generation. Therefore, photovoltaic power prediction technology is of great significance for improving the efficiency of photovoltaic conversion and optimizing energy production configuration. Traditional photovoltaic power prediction methods are mainly based on physical models and statistical methods, such as electronic component model prediction method and time series method, etc. These methods may perform worse than new algorithms such as deep learning when dealing with complex nonlinear relationships and large data volumes. In recent years, prediction methods based on artificial intelligence technology have gradually emerged, especially deep learning plays an important role in photovoltaic power prediction, which uses its own automatic learning ability to build complex mapping relationships from multiple inputs to single or multiple outputs, and realizes accurate prediction of photovoltaic power. However, deep learning models are usually regarded as black box models, and their internal mechanism and decision-making process are difficult to explain, lacking the interpretability of the model. SUMMARY
[0003] The purpose of the present application is to solve the problems raised in the prior art and provide a photovoltaic power prediction method based on LSTM and EBRB.
[0004] To achieve the above-mentioned purpose, the technical solution of the present application is: a photovoltaic power prediction method based on LSTM and EBRB, comprising
[0005] Step one, extracting photovoltaic power influence factors based on feature engineering: using the extreme gradient boosting algorithm to extract key factors affecting photovoltaic power generation, and constructing a key feature set with high correlation and representativeness;
[0006] Step two, building a prediction model based on LSTM: using the extracted key features to train the LSTM model to capture the nonlinear and long-term dependence relationship in the photovoltaic power time series;
[0007] Step three, building an inference model based on EBRB: building an EBRB model according to the extracted key features and their distribution, evaluating the uncertainty and complexity of the data, and generating reliable prediction output;
[0008] Step four, constructing a stacking integrated model based on LSTM and EBRB: taking the constructed LSTM model and EBRB model as base learners, taking a linear regression model as a meta-learner, and then combining the outputs of the base learners to train the meta-learner;
[0009] Step five, inputting data into the stacking integrated model to obtain the photovoltaic power prediction result.
[0010] In an embodiment of the present application, in step one, the extreme gradient boosting algorithm is used to extract the photovoltaic power influencing factors, the feature importance score provided by the XGBoost model is used to evaluate the contribution of each feature to the photovoltaic power prediction, and the features are sorted according to the importance, and the top 5 key features are selected as the independent variables of the model.
[0011] In an embodiment of the present application, in step two, the LSTM model is trained to capture the nonlinear and long-term dependence in the photovoltaic power time series, that is, in the construction process of the LSTM model, the input layer matching the shape of the training data is first set, the LSTM layer is added to capture the long-term dependence of the photovoltaic power data, and the Dropout layer is used to reduce overfitting, and finally the full connection layer is used as the output layer to predict the target variable, and the Adam optimizer and the mean square error loss function are configured to compile the model.
[0012] In an embodiment of the present application, in step three, an EBRB model set is constructed, an EBRB evaluation model is constructed, and an ER rule is used as an inference engine, and the modeling process is as follows:
[0013] (1) EBRB type knowledge base description
[0014] The EBRB model is constructed by IF-THEN rules;
[0015] (2) EBRB model inference process
[0016] For the EBRB model, an ER rule is used as an inference engine, and a photovoltaic power prediction value is obtained based on the EBRB model inference.
[0017] In an embodiment of the present application, in the IF-THEN rule, the kth extended confidence rule is defined as follows:
[0018]
[0019] Where Aki represents the candidate value of the ith premise attribute in the kth extended confidence rule; T k represents the number of premise attributes in the kth extended confidence rule; β n,k represents the result attribute D nthe confidence of the nth rule, if n,k <1, it is called that the kth extended belief rule contains complete information, otherwise the kth extended belief rule contains incomplete information; δ k,i (i = 1, …, T k ) represents the attribute weight of the ith premise attribute in the kth extended belief rule; θ k represents the rule weight of the kth extended belief rule.
[0020] In an embodiment of the present application, the EBRB model reasoning process is as follows:
[0021] Step 1, initialize the values of key parameters in the EBRB model: use expert knowledge to give all candidate values in all premise attributes, utility values of all evaluation levels in the result attribute, and attribute weights of all premise attributes;
[0022] Step 2, calculate individual matching degree: assuming that there are T groups of data, denoted as <x i ,y i >(i = 1, …, T), and its candidate value set is denoted as A i ={A i,j ; j = 1, …, J i}, that is, all evaluation levels of each index affecting photovoltaic power generation; wherein, x i represents the input value vector of the ith premise attribute, that is, the data of each index affecting photovoltaic power generation; y i represents the ith output value, that is, photovoltaic power generation; then, according to the rule-based information transformation technology, the individual matching degree of each candidate value is calculated; the corresponding formula is as follows:
[0023]
[0024] When the individual matching degrees of all candidate values are calculated, the individual matching degrees are represented by a distributed framework; the corresponding formula is as follows:
[0025] S(x i ) = {(A ij , α ij ); i = 1, 2, …, T; j = 1, 2, …, J i}
[0026] Wherein, α i,j represents the individual matching degree of the jth candidate value A i,j in the ith premise attribute;
[0027] Step 3, calculate the activation weight: integrate the rule weight, the attribute weight and the individual matching degree to obtain the activation weight of the kth extended belief rule:
[0028]
[0029] wherein, ak i represents the individual matching degree of the i th premise attribute candidate value in the k th extended confidence rule; when ω k >0, it indicates that the k th extended confidence rule is activated;
[0030] Step 4, synthesizing the confidence by applying the ER rule: synthesizing all activated rules once through the analysis formula in the ER rule, and the new synthesized confidence corresponds to the formula as follows:
[0031]
[0032] When the input value is x={x1,…,x M}, the prediction value obtained based on the EBRB model reasoning is as follows:
[0033] f(x)={(D i ,β i (x)),i=1,2,…,N}.
[0034] In an embodiment of the present application, step four is specifically implemented as follows:
[0035] (1) Data division
[0036] The time series data is divided into a training set and a test set, and is performed in time sequence; let T be the total number of time periods, T train be the number of time periods of the training set, T test be the number of time periods of the test set, then the training set and the test set are respectively represented as:
[0037] D train ={X1,X2,...,X T_train}
[0038] D test ={X T_train+1 ,X T_train+2 ,...,X T}
[0039] (2) Base learner training
[0040] Each base learner M i is trained on the training set; let M i be the i th base learner, and θ i be its parameters; the training process is represented as:
[0041]
[0042] wherein L is a loss function, X t is the data of time t in the training set, and Y tis the true label of time t;
[0043] (3) Cross-validation prediction
[0044] For each base learner M i , use time series cross-validation to divide D train into K folds, D train (k) is the k-th fold; for each fold k, divide the training set and the validation set, where the amount of data in the validation set is 10% of the original training set; use the training set of D train (k) to train M i , and make predictions on the validation set of D train (k) to obtain K prediction results, denoted as Y' t (k);
[0045] (4) Combined learner training
[0046] Use the prediction values of the base learners on the validation set as input features to train the combined learner; let Y' t (i) be the prediction value of the i-th base learner at time t, then the new feature vector Z t = [Y' t (1), Y' t (2),..., Y' t (N)], where N is the number of base learners; the parameters θ comb of the combined learner M comb are trained by the following method:
[0047]
[0048] Use the combined learner M comb to make predictions on the test set to obtain Y' test .
[0049] In an embodiment of the present application, in step five, the Stacking integrated model is a photovoltaic power prediction model with higher prediction accuracy and stability constructed by combining an LSTM model and an EBRB model. The photovoltaic power prediction value can be obtained by inputting data into the model.
[0050] The present application also provides a photovoltaic power prediction system based on LSTM and EBRB, which comprises a memory, a processor and computer program instructions stored in the memory and capable of being executed by the processor, and when the processor executes the computer program instructions, the method steps described above can be realized.
[0051] The application further provides a computer readable storage medium, which stores computer program instructions capable of being executed by a processor, and the computer program instructions can realize the method steps as described above when the processor executes the computer program instructions.
[0052] Compared with the prior art, the application has the following beneficial effects: the method combines LSTM and EBRB, constructs a Stacking integrated model based on LSTM and EBRB, effectively improves the interpretability and prediction accuracy of the model, and provides support for further application and development of LSTM and EBRB in the field of photovoltaic power prediction. BRIEF DESCRIPTION OF DRAWINGS
[0053] Figure 1 A flowchart of the application;
[0054] Figure 2 A schematic diagram of the LSTM model of the application;
[0055] Figure 3 An EBRB model construction process of the application;
[0056] Figure 4 A schematic diagram of the Stacking integrated model of the application. DETAILED DESCRIPTION
[0057] The technical solutions of the application will be specifically described below with reference to the drawings.
[0058] The application provides a photovoltaic power prediction method based on LSTM and EBRB, including
[0059] Step one, extracting photovoltaic power influence factors based on feature engineering: using an extreme gradient boosting algorithm to extract key factors affecting photovoltaic power, and constructing a key feature set with high correlation and representativeness;
[0060] Step two, constructing a prediction model based on LSTM: using the extracted key features, training an LSTM model to capture the nonlinear and long-term dependence in the photovoltaic power time series;
[0061] Step three, constructing an inference model based on EBRB: constructing an EBRB model according to the extracted key features and their distribution, evaluating the uncertainty and complexity of the data, and generating reliable prediction output;
[0062] Step four, constructing a Stacking integrated model based on LSTM and EBRB: using the constructed LSTM model and EBRB model as base learners, using a linear regression model as a meta-learner, and combining the outputs of the base learners to train the meta-learner;
[0063] Step five, input data into the Stacking integrated model to obtain the photovoltaic power prediction result.
[0064] The following is the specific implementation process of the present application.
[0065] Please refer to Figures 1 to 4 The present application provides a photovoltaic power prediction model based on LSTM and EBRB, and the steps are as follows:
[0066] Step one, based on feature engineering, extract the influencing factors of photovoltaic power: use the extreme gradient boosting algorithm to extract the key factors affecting photovoltaic power, and construct a feature set with high correlation and representativeness;
[0067] Step two, build a prediction model based on LSTM: use the extracted key features to train the LSTM model to capture the nonlinear and long-term dependence in the photovoltaic power time series;
[0068] Step three, build an inference model based on EBRB: build EBRB based on the extracted features and their distribution, evaluate the uncertainty and complexity of the data, and generate reliable prediction output;
[0069] Step four, build a Stacking integrated model based on LSTM and EBRB: use the constructed LSTM model and EBRB model as base learners, use a linear regression model as a meta-learner, and then combine the output of the base learners to train the meta-learner;
[0070] Step five, input data into the Stacking integrated model to obtain the photovoltaic power prediction result.
[0071] In step one, based on the extreme gradient boosting algorithm, the influencing factors of photovoltaic power are extracted, the feature importance score provided by the XGBoost model is used to evaluate the contribution of each feature to the photovoltaic power prediction, and the features are sorted by importance, and the top 5 features are selected as the independent variables of the model.
[0072] In step two, the LSTM model is trained to capture the nonlinear and long-term dependence in the photovoltaic power time series. In the construction process of the LSTM model, first set the input layer matching the shape of the training data, add the LSTM layer to capture the long-term dependence of the photovoltaic power data, and reduce overfitting through the Dropout layer, finally use the fully connected layer as the output layer to predict the target variable, and configure the Adam optimizer and mean square error loss function to compile the model.
[0073] In step three, an EBRB model set is constructed, and the EBRB is used to build an evaluation model, using the ER algorithm as the inference engine of the model, and the modeling process is as follows:
[0074] (1) EBRB knowledge base description
[0075] The model is constructed by IF-THEN rules, wherein the kth extended belief rule is defined as follows:
[0076]
[0077] wherein Aki represents the candidate value of the ith premise attribute in the kth extended belief rule; T k represents the number of premise attributes in the kth extended belief rule; β n,k represents the confidence degree of the result attribute D n assigned in the kth extended belief rule, wherein if ΣNn=1β n,k <1, it is called that the kth extended belief rule contains complete information, otherwise the kth extended belief rule contains incomplete information; δ k,i (i = 1, …, T k ) represents the attribute weight of the ith premise attribute in the kth extended belief rule; θ k represents the rule weight of the kth extended belief rule.
[0078] (2) EBRB reasoning process
[0079] For the EBRB model, the ER algorithm is usually used as the reasoning engine, and the specific EBRB model reasoning process is as follows:
[0080] Step 1: initialize the value of the key parameter in EBRB. Use expert knowledge to give all candidate values in all premise attributes, utility values of all evaluation levels in the result attribute, and attribute weights of all premise attributes.
[0081] Step 2: calculate the individual matching degree. Suppose there are T groups of data, denoted as <x i , y i >(i = 1, …, T), and its candidate value set is denoted as A i = {A i,j ; j = 1, …, J i}, that is, all evaluation levels of each index affecting photovoltaic power generation; wherein x i represents the input value vector of the ith premise attribute, that is, the data of each index affecting photovoltaic power generation; y i represents the ith output value, that is, photovoltaic power generation; then, according to the rule-based information transformation technology, the individual matching degree of each candidate value is calculated. The corresponding formula is as follows:
[0082]
[0083] When the individual matching degrees of all candidate values are calculated, the individual matching degrees can be represented by a more clear and intuitive distributed framework. The corresponding formula is as follows:
[0084] S(x i )={(A ij ,α ij );i=1,2,…,T;j=1,2,...,J i}
[0085] wherein, α i,j represents the individual matching degree of the jth candidate value A i,j in the ith premise attribute.
[0086] Step 3: Calculate the activation weight. Integrate the rule weight, attribute weight and individual matching degree to obtain the activation weight of the kth extended confidence rule:
[0087]
[0088] wherein, αk i represents the individual matching degree of the ith premise attribute candidate value in the kth extended confidence rule. When ω k > 0, it indicates that the kth extended confidence rule is activated.
[0089] Step 4: Apply the ER algorithm to synthesize the confidence. Synthesize all activated rules at one time by the analytical formula in the ER algorithm. The new confidence synthesized is as follows:
[0090]
[0091] When the input value is x = {x1,..., x M}, the prediction value obtained based on the EBRB reasoning is:
[0092] f(x) = {(D i , β i (x)), i = 1, 2,..., N}.
[0093] In step four, the integrated model combines the characteristics of multiple models for optimization, which is implemented as follows:
[0094] (1) Data division
[0095] The time series data is divided into a training set and a test set, usually in chronological order. Let T be the total number of time periods, T train be the number of time periods of the training set, and T test be the number of time periods of the test set. Then the training set and the test set can be represented as:
[0096] D train ={X1,X2,...,X T_train}
[0097] D test ={X T_train+1 ,X T_train+2 ,...,X T}
[0098] (2) Base learner training
[0099] For each base learner M i , training is performed on the training set. Let M i be the i-th base learner and θ i its parameters. The training process can be represented as:
[0100]
[0101] where L is the loss function and Yt is the true label at time t.
[0102] (3) Cross-validation prediction
[0103] For each base learner M i , time-series cross-validation is used to divide D train into K folds, D train (k) is the k-th fold. For each fold k, the training set and validation set are divided, with the validation set being 10% of the original training set. Using the training set of D train (k), M i is trained, and the prediction is made on the validation set of D train (k), resulting in K prediction results, denoted as Y' t (k).
[0104] (4) Ensemble learner training
[0105] The prediction values of the base learners on the validation set are used as input features to train the ensemble learner. Let Y' t (i) be the prediction value of the i-th base learner at time t, then the new feature vector Z t = [Y' t (1), Y' t (2),..., Y' t (N)], where N is the number of base learners. The parameters θ comb of the ensemble learner M comb are trained by:
[0106]
[0107] The ensemble learner M comb is used to make predictions on the test set, resulting in Y' test .
[0108] In step five, by combining the LSTM model and the EBRB model, a photovoltaic power prediction model with higher prediction accuracy and stability is constructed, and the photovoltaic power prediction value can be obtained by inputting data into the model.
[0109] To verify the effectiveness of the present application, the measured data of the Yulara photovoltaic power station of the Australian Desert Knowledge Solar Centre is taken as an example for experimental verification, mainly including the following processes:
[0110] (1) Extracting photovoltaic power influence factors based on feature engineering
[0111] Firstly, the data is read by using the third-party library Pandas in Python, and the data is cleaned by filling in missing values, replacing abnormal values, and deleting duplicate records; secondly, the extreme gradient boosting algorithm is introduced to iteratively establish a new tree model, and the importance of the obtained characteristics is sorted. Select the top 5 influencing factors as independent variables, namely air temperature, total radiation, air pressure, component temperature and solar radiation intensity, and the corresponding photovoltaic power as dependent variable to build the model.
[0112] (2) Constructing LSTM model for photovoltaic power prediction
[0113] The LSTM model is introduced for modeling, the input layer matching the shape of the training data is set, the LSTM layer, the Dropout layer and the full connection layer are added in turn, and the Adam optimizer and the mean square error loss function are configured to compile the model.
[0114] (3) Constructing EBRB model for photovoltaic power prediction
[0115] The values of the key parameters in the EBRB are initialized, the evaluation levels and utility values of the premise attributes and result attributes and the attribute weights of the premise attributes are obtained as shown in Table 1. Based on the set parameter values, the individual matching degree and the activation weight are calculated, and the confidence is synthesized according to the ER algorithm.
[0116] Table 1 Evaluation levels and utility values of each attribute
[0117]
[0118]
[0119] (4) Constructing Stacking integrated model based on LSTM and EBRB
[0120] Firstly, the training set and the test set are divided according to 7:3, and the first 70% of the data is used as the training set, and the last 30% of the data is used as the test set; secondly, the time series cross-validation method is used to divide the training set 6 times, each division ensures the time continuity of the divided data, and 40% is used as the starting proportion, and 10% is used as the moving step proportion, and 6 different training-validation pairs are obtained, and the data amount of the validation set of each pair of data is 10% of the original training set; then, the LSTM model and the EBRB model constructed are used as the base learner, and the linear regression model is used as the meta-learner, and the Stacking integrated model based on LSTM and EBRB is constructed.
[0121] (5) inputting data into the Stacking integrated model to obtain a prediction result
[0122] The photovoltaic power generation power and the meteorological data are inputted, the Stacking integrated model is trained and tested, and part of the photovoltaic power generation power prediction values are as shown in Table 2.
[0123] Table 2 part of photovoltaic power generation power prediction values on January 23
[0124]
[0125]
[0126] The obtained prediction values are compared and evaluated with the true values, R2 reaches 0.949, and it is indicated that the prediction effect of the model is good, and the model can be applied to photovoltaic power generation power prediction.
[0127] The application provides a photovoltaic power generation power prediction model based on LSTM and EBRB, the model combines LSTM and EBRB, constructs a Stacking integrated model based on LSTM and EBRB, effectively improves the interpretability and prediction accuracy of the model, and provides support for further application and development of LSTM and EBRB in the field of photovoltaic power generation power prediction.
[0128] The application further provides a photovoltaic power generation power prediction system based on LSTM and EBRB, which comprises a memory, a processor and computer program instructions stored in the memory and capable of being executed by the processor, and when the processor executes the computer program instructions, the method steps described above can be realized.
[0129] The application further provides a computer readable storage medium, which stores computer program instructions capable of being executed by a processor, and when the processor executes the computer program instructions, the method steps described above can be realized.
[0130] Those skilled in the art will appreciate that embodiments of the application can be readily used as software, hardware, or a combination of software and hardware. In one
[0131] The present application is described in reference to the flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing system or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart Figure 1 one or more functions specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks.
[0132] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks.
[0133] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks.
[0134] The above description is only preferred embodiments of the application, not intended to limit other forms of the application. Any person skilled in the art can make modifications or alterations to the above-mentioned disclosed technical contents without departing from the technical solution of the application. Any simple modification, equivalent change and alteration of the above embodiments according to the technical essence of the application still belong to the protection scope of the technical solution of the application.
Claims
1. A photovoltaic power generation power prediction method based on LSTM and EBRB, characterized in that, Comprising Step one, based on feature engineering to extract photovoltaic power influencing factors: using extreme gradient boosting algorithm to extract the key factors affecting photovoltaic power generation, and construct a set of key features with high correlation and representativeness; Step two, based on LSTM to build a prediction model: using the extracted key features, train the LSTM model to capture the nonlinear and long-term dependence in the photovoltaic power time series; Step three, based on EBRB to build an inference model: according to the extracted key features and their distribution, build an EBRB model to evaluate the uncertainty and complexity of the data, and generate reliable prediction output; Step four, build a stacking integrated model based on LSTM and EBRB: use the constructed LSTM model and EBRB model as base learners, and use a linear regression model as a meta-learner, then combine the outputs of the base learners to train the meta-learner; Step five, input the data into the stacking integrated model to get the photovoltaic power prediction result; In step two, the LSTM model is trained to capture the nonlinear and long-term dependence in the photovoltaic power time series, that is, in the construction process of the LSTM model, first set the input layer matching the shape of the training data, add the LSTM layer to capture the long-term dependence of the photovoltaic power data, and reduce overfitting through the Dropout layer, finally use the fully connected layer as the output layer to predict the target variable, and configure the Adam optimizer and mean square error loss function to compile the model; Step four is implemented as follows: (1) Data division The time series data is divided into a training set and a test set in chronological order; let T be the total number of time periods, T train be the number of time periods of the training set, T test be the number of time periods of the test set, then the training set and the test set are represented as follows: D train = {X1, X2,..., X T_train} D test = {X T_train+1 , X T_train+2 ,..., X T} (2) Base learner training for each base learner M i Train on the training set; let M i for the i-th base learner, θ i its parameters; The training process is represented as: where L is a loss function, X t is data at time t in the training set, Y t is the true label at time t; (3) Cross-validation prediction For each base learner M i , use time series cross-validation to divide D train into K folds, D train (k) is the k-th fold; for each fold k, divide the training set and the validation set, where the amount of data in the validation set is 10% of the original training set; use the training set of D train (k) to train M i , and make predictions on the validation set of D train (k) to obtain K prediction results, denoted as Y t (k). (4) Combined learner training The prediction values of the base learners on the validation set are used as input features to train the ensemble learner; let Y t (i) be the prediction value of the i-th base learner at time t, then the new feature vector Z t = [Y' t (1), Y' t (2),..., Y' t (N)], where N is the number of base learners; the parameters θ comb of the ensemble learner M comb are trained by Using the combined learner M comb Making predictions on the test set, yielding Y' test .
2. The photovoltaic power forecasting method based on LSTM and EBRB according to claim 1, characterized in that, In step one, based on the extreme gradient boosting algorithm to extract the photovoltaic power influencing factors, use the feature importance score provided by the XGBoost model to evaluate the contribution of each feature to the photovoltaic power prediction, and sort the features by importance, select the top 5 key features as the independent variables of the model. 3.The LSTM and EBRB based photovoltaic power generation power prediction method of claim 1, wherein, In step three, build an EBRB model set, use EBRB to build an evaluation model, and use ER rules as the inference engine, the modeling process is as follows: (1) EBRB knowledge base description EBRB model is constructed by IF-THEN rules; (2) EBRB model inference process For EBRB model, use ER rule as inference engine, and get photovoltaic power prediction value based on EBRB model inference.
4. The photovoltaic power forecasting method based on LSTM and EBRB according to claim 3, characterized in that, In IF-THEN rule, the kth extended confidence rule is defined as follows: wherein Aki represents the candidate value of the ith premise attribute in the kth extended confidence rule; T k represents the number of premise attributes in the kth extended confidence rule; β n,k represents the confidence degree assigned to the result attribute D n in the kth extended confidence rule, if ΣNn=1β n,k <1, the kth extended confidence rule contains complete information, otherwise the kth extended confidence rule contains incomplete information; δ k,i (i=1,…,T k ) represents the attribute weight of the ith premise attribute in the kth extended confidence rule; θ k represents the rule weight of the kth extended confidence rule.
5. The photovoltaic power forecasting method based on LSTM and EBRB according to claim 4, characterized in that, The EBRB model inference process is as follows: Step 1, initialize the value of the key parameters in the EBRB model: use expert knowledge to give all candidate values in all premise attributes, utility values of all evaluation levels in the result attribute, and attribute weights of all premise attributes; Step 2, calculating individual matching degree: assuming there are T groups of data, denoted as <x i ,y i >(i=1,…,T), and the candidate value set thereof is denoted as A i ={A i,j ;j=1,…,J i}, that is, all evaluation grades of each index affecting photovoltaic power generation; wherein x i represents an input value vector of the i-th premise attribute, that is, data of each index affecting photovoltaic power generation; y i represents the i-th output value, that is, photovoltaic power generation; then, according to a rule-based information transformation technology, the individual matching degree of each candidate value is calculated; the corresponding formula is as follows: After calculating the individual matching degree of all candidate values, the individual matching degree is represented by a distributed framework; the corresponding formula is as follows: S(x i ) = {(A ij , a ij ) ; i = 1, 2,..., T; j = 1, 2,..., J i} wherein, α i,j represents the individual matching degree of the jth candidate value A i,j in the ith premise attribute; Step 3, calculate the activation weight: integrate the rule weight, attribute weight and individual matching degree to get the activation weight of the kth extended confidence rule: wherein, ak i represents the individual matching degree of the i th premise attribute candidate value in the k th extended confidence rule; when ω k > 0, it indicates that the k th extended confidence rule is activated; Step 4, application of ER rule synthesis confidence: through the analysis formula in ER rule, all activated rules are synthesized once, and the synthesized new confidence corresponds to the formula as follows: When the input value is x = {x1,..., x M} the predicted value inferred based on the EBRB model is: f(x) = {(D i , β i (x)), i = 1, 2,..., N}.
6. The photovoltaic power forecasting method based on LSTM and EBRB according to claim 1, characterized in that, In step five, the stacking integrated model is a photovoltaic power prediction model with higher prediction accuracy and stability, which is constructed by combining the LSTM model and the EBRB model.
7. A photovoltaic power generation power prediction system based on LSTM and EBRB, characterized in that, The computer program instructions stored in the memory and capable of being executed by the processor can implement the method steps of any one of claims 1-6 when the processor executes the computer program instructions.
8. A computer readable storage medium having stored thereon computer program instructions capable of being executed by a processor, the computer program instructions capable of implementing the method steps of any one of claims 1-6 when the processor executes the computer program instructions.