Hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system

Abstract

The invention discloses a hybrid drive photovoltaic power prediction method based on adaptive feature selection and a self-learning system, and relates to the field of prediction. According to the technical scheme, the method is characterized by comprising the steps of S1, collecting photovoltaic data and performing prediction processing; s2, adaptively selecting photovoltaic power prediction features; s3, constructing a photovoltaic power prediction-oriented physical model and a data driving model; s4, fusing and optimizing the physical model and the data driving model to obtain a dual-driving prediction model; and S5, constructing a self-learning system of the prediction model. According to the invention, through real-time data stream processing and an incremental learning algorithm, the model can rapidly adapt to data changes, the accuracy of photovoltaic power prediction is significantly improved, and errors are reduced. A regular updating and error monitoring mechanism ensures that the model stably operates in different environments and conditions, and ensures that a system prediction result better conforms to an actual application scene.

Description

Technical Field

[0001] The invention relates to the field of prediction technology, and in particular to a hybrid drive photovoltaic power prediction method based on adaptive feature selection and a self-learning system. Background Art

[0002] Photovoltaic power prediction is an important technical means for optimizing the management and scheduling of photovoltaic power plants. At present, there are two main types of photovoltaic power prediction systems on the market: systems based on physical models and systems based on data-driven models. The physical model-based system calculates the power output by establishing a current-voltage characteristic model of the photovoltaic array and combining it with meteorological data such as solar irradiance and ambient temperature collected in real time. The main structure of this type of system includes a photovoltaic array model, a meteorological data input module, and a power calculation module, but it relies on accurate physical parameters and environmental data, is difficult to cope with rapidly changing environmental conditions, has poor flexibility, and is not effective in dealing with complex nonlinear relationships in photovoltaic arrays.

[0003] The system based on data-driven model uses historical power data and meteorological data to train and predict through machine learning algorithms such as long short-term memory network (LSTM) or support vector machine (SVM). Its main structure includes data acquisition module, data preprocessing module, machine learning model and prediction result output module. This type of system performs well in processing nonlinear and complex relationships and can provide high prediction accuracy under changing environmental conditions. However, this method relies on a large amount of historical data, and the prediction ability is limited when the data quality and quantity are insufficient. In addition, the existing data-driven models usually lack adaptive capabilities and cannot dynamically adjust feature selection and model parameters, resulting in unstable prediction results under different environments and conditions. Therefore, the existing methods have limitations in feature selection, model fusion and self-learning, and it is difficult to fully cope with changing environmental conditions and data characteristics. Summary of the invention

[0004] The present invention provides a hybrid-driven photovoltaic power prediction method based on adaptive feature selection and self-learning system. The present invention proposes an innovative combination of adaptive feature selection, hybrid drive of physical model and data-driven model, and self-learning system to improve the accuracy and robustness of prediction. This method can not only dynamically adjust feature selection and model parameters, but also maintain stable prediction results under changing environmental conditions, providing strong technical support for the optimal management and scheduling of photovoltaic power stations.

[0005] To achieve the above object, the present invention provides a hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system, and the specific steps are as follows:

[0006] S1. Collect PV data and forecast processing;

[0007] S2. Adaptive selection of photovoltaic power prediction features;

[0008] S3. Build physical models and data-driven models for photovoltaic power prediction;

[0009] S4. Integrate and optimize the physical model and the data-driven model to obtain a dual-driven prediction model;

[0010] S5. Build a self-learning system for prediction models;

[0011] In particular, the photovoltaic data in step S1 includes historical power data, meteorological data, environmental data, and grid status data.

[0012] In particular, the prediction process in step S1 includes data cleaning and data standardization.

[0013] In particular, the photovoltaic power prediction feature adaptive selection in step S2 specifically includes the following sub-steps:

[0014] Step 1: Based on the physical model and empirical knowledge, a preliminary feature set is extracted, wherein the feature set includes meteorological features, time features, and derived features, wherein the derived features include the moving average of irradiance and the rate of change of temperature;

[0015] Step 2: Random forest model training, divide the preprocessed data into training set and test set; use the training set data to train the random forest model, and make predictions by building multiple decision trees and combining their results;

[0016] Step 3: Feature importance measurement: During the model training process, the contribution of each feature to the model prediction performance is evaluated and the importance of the feature is measured by calculating the feature importance score;

[0017] Step 4: Feature selection, sort the features according to their importance, select the top k most important features, and use the recursive feature elimination method to gradually select the optimal feature set.

[0018] In particular, in step S3, the physical model calculates the photovoltaic power output based on the current-voltage characteristic curve of the photovoltaic array using irradiance and temperature parameters.

[0019] In particular, the data-driven model in step S3 specifically adopts a deep learning model composed of a long short-term memory neural network and a convolutional neural network.

[0020] In particular, the dual-driven prediction model in step S4 is a prediction result that combines the physical model and the data-driven model.

[0021] In particular, the self-learning system in step S5 specifically includes the following sub-steps:

[0022] Step 1: Update model data and parameters using online learning methods;

[0023] Step 2: Establish a user feedback mechanism;

[0024] Step 3: Design model updating and continuous optimization strategies.

[0025] Compared with the prior art, the present invention has the following beneficial effects: through real-time data stream processing and incremental learning algorithms, the model can quickly adapt to data changes, significantly improve the accuracy of photovoltaic power prediction, and reduce errors. Regular updates and error monitoring mechanisms ensure that the model operates stably under different environments and conditions, ensuring that the system prediction results are more in line with actual application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for describing the embodiments are briefly introduced below.

[0027] Figure 1 The present invention is a flow chart of a hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system. DETAILED DESCRIPTION

[0028] The technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limitations on the present invention.

[0030] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0031] The present invention provides a hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system. The model process is as follows: Figure 1 The specific steps are as follows:

[0032] S1. Collect PV data and forecast processing, including:

[0033] (1) Data collection: Collect historical power data of photovoltaic power plants, meteorological data (such as solar irradiance, temperature, wind speed, etc.), environmental data (such as humidity, rainfall, etc.) and other relevant data (such as grid status data).

[0034] (2) Data cleaning: processing missing values ​​and outliers, using interpolation or other methods to fill in missing data and remove outliers.

[0035] (3) Data standardization: The data is standardized so that its mean is 0 and its standard deviation is 1 to improve the stability of model training.

[0036] S2. Adaptive selection of photovoltaic power prediction features, including:

[0037] (1) Feature importance assessment: Use the random forest algorithm to assess the importance of each feature. The steps are as follows:

[0038] 1) Based on the physical model and empirical knowledge, a preliminary feature set is extracted, including meteorological features (such as solar irradiance, ambient temperature, etc.) and temporal features (such as timestamp, season). Further, derived features are constructed, such as the moving average of irradiance, the rate of change of temperature, etc.

[0039] 2) Random forest model training: The preprocessed data is divided into a training set and a test set. The random forest model is trained using the training set data to make predictions by constructing multiple decision trees and combining their results. During the model training process, the contribution of each feature to the model's prediction performance is evaluated, and the importance of each feature is measured by calculating the feature's importance score.

[0040] (2) Feature selection: According to the feature importance ranking, the top k most important features are selected, and the optimal feature set is gradually selected using the recursive feature elimination method.

[0041] S3. Build physical models and data-driven models for photovoltaic power prediction, including:

[0042] (1) Based on the optimized feature set, a physical model is constructed. The physical model is based on the current-voltage characteristic curve of the photovoltaic array and uses input parameters such as irradiance and temperature to calculate the photovoltaic power output.

[0043] (2) Build a data-driven model, specifically a deep learning model composed of a long short-term memory neural network (LSTM) and a convolutional neural network (CNN). Use the optimized feature set for training. The data-driven model can capture complex nonlinear relationships in the data and improve prediction accuracy. The specific steps are as follows:

[0044] 1) Design the LSTM model architecture, including the input layer, LSTM layer, fully connected layer, and output layer, and set hyperparameters such as learning rate, batch size, and number of training rounds.

[0045] 2) Design the CNN model architecture, including convolutional layers, pooling layers, fully connected layers, and output layers, and set hyperparameters such as convolution kernel size, pooling window size, learning rate, etc.

[0046] 3) Use the training set data to train the deep learning model, and use the validation set data for model evaluation and hyperparameter tuning.

[0047] S4. Integrate and optimize the physical model and the data-driven model to obtain a dual-driven prediction model. The specific steps are:

[0048] (1) Combine the prediction results of the physical model and the data-driven model to build a hybrid model. Use the stacking method to fuse the outputs of different models.

[0049] (2) Use a meta-learner (such as linear regression, random forest, etc.) to relearn the outputs of the physical model and the data-driven model to obtain the final prediction result.

[0050] (3) Use the cross-validation method to evaluate the performance of the hybrid model and adjust the fusion strategy and model parameters to improve the prediction accuracy.

[0051] (4) Evaluate the performance of the final model on the validation set and the test set, and calculate the prediction error indicators (such as mean square error, mean absolute error, etc.) to ensure the generalization ability of the model on different data sets.

[0052] S5. Build a self-learning system for the prediction model, so that the PV power prediction model can automatically adapt to new data and changes, and improve the long-term performance and stability of the model, including:

[0053] (1) Use online learning methods to update model data and parameters, as follows:

[0054] 1) Obtain real-time data streams from sensors and weather stations to ensure data continuity and real-time performance.

[0055] 2) Use incremental learning algorithms such as online gradient descent and incremental decision trees to process data streams.

[0056] 3) Incrementally update model parameters based on the arrival of new data instead of retraining the entire model.

[0057] 4) Evaluate the model performance after incremental updates in real time to ensure the accuracy and stability of the model. Dynamically adjust the model parameters and learning rate based on the real-time evaluation results.

[0058] (2) Establish a user feedback mechanism, as follows:

[0059] 1) Design a user feedback interface to collect user feedback on prediction results. Store user feedback data in the database for subsequent analysis.

[0060] 2) Regularly analyze user feedback data to identify common problems and areas for improvement. Adjust model parameters and feature selection strategies based on user feedback.

[0061] 3) Based on feedback information and analysis results, continuously optimize each module of the self-learning system. Regularly evaluate user satisfaction to ensure the effectiveness and practicality of the self-learning system.

[0062] (3) Design model update and continuous optimization strategies, as follows:

[0063] 1) When the prediction error exceeds the set threshold, the model update is triggered. The model update is triggered regularly according to the monthly setting.

[0064] 2) Integrate the updated model into the management system of the PV power plant and use the updated model to perform real-time PV power forecasting.

[0065] 3) Continuously monitor the model’s predictive performance to ensure the model’s accuracy and stability. Integrate user feedback to continuously optimize the model and system.

[0066] The results of the present invention are evaluated using the MAPE evaluation index. The MAPE results are shown in Table 1.

[0067] MAPE results table

[0068] Model-driven approach 7.55％ Data-driven approach 8.29％ The present invention 6.14％

[0069] It shows that the present invention has higher prediction accuracy.

[0070] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A hybrid-driven photovoltaic power prediction method based on adaptive feature selection and self-learning system, the specific steps are as follows: S1. Collect PV data and forecast processing; S2. Adaptive selection of photovoltaic power prediction features; S3. Build physical models and data-driven models for photovoltaic power prediction; S4. Integrate and optimize the physical model and the data-driven model to obtain a dual-driven prediction model; S5. Build a self-learning system for predictive models.

2. According to the hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system according to claim 1, the photovoltaic data in step S1 includes historical power data, meteorological data, environmental data, and grid status data.

3. According to the hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system according to claim 1, the prediction processing in step S1 includes data cleaning and data standardization processing.

4. According to the hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system in claim 1, the photovoltaic power prediction feature adaptive selection in step S2 specifically includes the following sub-steps: Step 1: Extract a preliminary feature set based on the physical model and empirical knowledge, the feature set including meteorological features, time features, and derivative features; Derived characteristics include the moving average of irradiance and the rate of change of temperature; Step 2: Random forest model training, divide the preprocessed data into training set and test set; use the training set data to train the random forest model, and make predictions by building multiple decision trees and combining their results; Step 3: Feature importance measurement: During the model training process, the contribution of each feature to the model prediction performance is evaluated and the importance of the feature is measured by calculating the feature importance score; Step 4: Feature selection, sort the features according to their importance, select the top k most important features, and use the recursive feature elimination method to gradually select the optimal feature set.

5. According to the hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system as described in claim 1, in step S3, the physical model is based on the current-voltage characteristic curve of the photovoltaic array and uses irradiance and temperature parameters to calculate the photovoltaic power output.

6. According to the hybrid-driven photovoltaic power prediction method based on adaptive feature selection and self-learning system according to claim 1, the data-driven model in step S3 specifically adopts a deep learning model composed of a long short-term memory neural network and a convolutional neural network.

7. According to the hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system as described in claim 1, the dual drive prediction model in step S4 is a prediction result combining the physical model and the data drive model.

8. The hybrid drive photovoltaic power prediction method based on adaptive feature selection and self-learning system according to claim 1, wherein the self-learning system in step S5 specifically comprises the following sub-steps: Step 1: Update model data and parameters using online learning methods; Step 2: Establish a user feedback mechanism; Step 3: Design model updating and continuous optimization strategies.

Images