Electrochemical energy storage station value evaluation method based on SVR-XGBoost-Ridge ensemble learning

Abstract

The invention provides an electrochemical energy storage station value evaluation method based on SVR-XGBoost-Ridge ensemble learning, and the method comprises the steps: firstly, training a base learner: carrying out the training of XGBoost and SVM, and obtaining a base model; secondly, generating a base model prediction result: predicting the training set and the test set by using XGBoost and SVM to obtain a prediction result of a base learner; training a meta learner: training a ridge regression model by using a prediction result of the base learner to generate a prediction model; integrated model prediction: integrating prediction results of the base learner by using a ridge regression model to obtain a comprehensive prediction result; evaluating and selecting an optimal model: evaluating the performance of the model through cross validation evaluation, and selecting the model with the optimal performance; model tuning: according to an evaluation result, carrying out necessary tuning on the base learner and the meta learner; and finally, deploying and applying the model: applying the finally optimized stack model to actual data, and predicting and analyzing a result.

Description

Technical Field

[0001] The present invention belongs to the field of value assessment of novel energy storage stations, and in particular to a value assessment method of an electrochemical energy storage station. Background Art

[0002] Since the proposal of the carbon peak and carbon neutrality goals (i.e. the "dual carbon" goals), the importance of new energy storage technologies has become increasingly prominent. The development of new energy storage technologies is not only the key to achieving the "dual carbon" goals, but also an important means to promote energy transformation and sustainable development. Among the new energy storage projects currently in operation, lithium-ion battery energy storage accounts for 97%, lead-carbon (acid) accounts for 0.8%, and flow battery energy storage accounts for 0.4%. Electrochemical energy storage accounts for more than 98% overall, occupying an absolute leading position in the energy storage technologies that have been put into production. However, although electrochemical energy storage technology has become the mainstream technology for new energy storage, a scientific and reasonable evaluation index system for new energy storage power stations has not been established in the existing technology, especially the lack of value assessment methods for electrochemical energy storage stations. Summary of the invention

[0003] The purpose of the present invention is to provide an electrochemical energy storage station value assessment method based on SVR-XGBoost-Ridge ensemble learning, which can quickly realize the value assessment and comparison of electrochemical energy storage stations without manual assignment and scoring.

[0004] The technical solution to achieve the purpose of the present invention is:

[0005] A value assessment method for an electrochemical energy storage station based on SVR-XGBoost-Ridge ensemble learning includes the following steps:

[0006] Step 1: Train the base learner: Use the training data to train XGBoost and support vector machine (SVM) respectively to obtain two base learner models;

[0007] Step 2: Generate the prediction results of the base learner: Use the trained XGBoost and SVM to predict the training set and test set to obtain the prediction results of the base learner;

[0008] Step 3: Train the meta-learner: Use the prediction results of the base learner as feature input, train the ridge regression model, integrate the prediction results of the base learner, and generate the final prediction model;

[0009] Step 4: Integrated model prediction: Use the trained ridge regression model to integrate the prediction results of the base learner to obtain a comprehensive prediction result;

[0010] Step 5: Evaluate and select the best model: Evaluate the performance of the entire stacked model through cross-validation and select the model with the best performance;

[0011] Step 6: Model tuning: Based on the evaluation results, make necessary adjustments to the base learner and meta learner to optimize model performance.

[0012] Step 7: Deploy and apply the model: Apply the final optimized stacked model to the actual data, make predictions, and analyze the results.

[0013] Compared with the prior art, the present invention has the following significant advantages:

[0014] (1) The present invention is based on machine learning technology, does not rely on expert judgment, has strong objectivity, and the evaluation results are also universal and generalizable.

[0015] (2) Traditional methods such as hierarchical analysis models are difficult to handle large-scale data sets and complex systems. With the diversification of application scenarios of energy storage technologies, a single expert evaluation model is difficult to fully and accurately capture the value characteristics of different energy storage technologies. The computational complexity of the hierarchical analysis model is relatively high, especially when it involves complex evaluation systems with multiple levels and multiple indicators. It is easy to cause the calculation time to be too long and it is difficult to meet the needs of rapid decision-making. Electrochemical energy storage systems involve multi-dimensional data, including technical parameters, economic indicators, environmental benefits, etc. When processing these complex data, traditional hierarchical analysis models are often unable to fully tap their potential information, resulting in insufficient depth and accuracy of the evaluation results. The present invention is based on a data-driven approach and shows stronger adaptability and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Flowchart of the value assessment method of electrochemical energy storage station based on multi-model ensemble learning based on SVR-XGBoost-Ridge. DETAILED DESCRIPTION

[0017] Step 1: Train base learners: Select two base learners, XGBoost and support vector regression (SVR), and a meta learner, ridge regression. First, train XGBoost and SVR to get two base learner models; then, use these models to predict the data and generate the prediction results of the base learners.

[0018] Step 2: Generate the prediction results of the base learner: Use the trained XGBoost and SVR to predict the training set and test set to obtain the prediction results of the base learner.

[0019] Step 3: Train the meta-learner: Use the prediction results of the base learner as feature input and train the ridge regression model to integrate the prediction results of the base learner to generate the final prediction model.

[0020] Step 4: Integrated model prediction: Use the trained ridge regression model to integrate the prediction results of the base learner to obtain a comprehensive prediction result.

[0021] Step 5: Evaluate and select the best model: Evaluate the performance of the entire stacked model through cross-validation and select the model with the best performance.

[0022] Step 6: Model tuning: Based on the evaluation results, make necessary adjustments to the base learner and meta learner to optimize model performance.

[0023] Step 7: Deploy and apply the model: Apply the final optimized stacked model to the actual data, make predictions, and analyze the results.

[0024] At this point, the application of the electrochemical energy storage station value assessment method based on SVR-XGBoost-Ridge ensemble learning has been completed. This method improves the accuracy and reliability of value assessment by integrating the prediction results of multiple models. In addition, the method of the present invention can adapt to different data characteristics through model integration and tuning, thereby improving the practicality of value assessment technology.

[0025] Case

[0026] The case uses simulated data of 1,300 electrochemical energy storage power stations, which include nine different dimensions: power station type, power level (MW), energy level (MWh), average annual operating hours, conversion efficiency (%), average comprehensive utilization rate (%), response time (ms), expected cycle life (years), and average annual utilization hours. The simulated data is scored using the hierarchical analysis method, and a weighted score with a full score of 100 is finally obtained. The algorithm is implemented on the Python platform.

[0027] In order to evaluate the accuracy of SVR-XGBoost-Ridge evaluation, the root mean square error (RMSE) and the coefficient of determination (R 2 , Coefficient of Determination) as the evaluation index. The calculation formulas of the root mean square error and the coefficient of determination are as follows:

[0028]

[0029] Among them, y i is the actual value is the predicted value is the mean of the actual values, n is the number of samples. RMSE is the square root of the mean square error (MSE).

[0030]

[0031] R 2 Measures the explanatory power of the model, that is, how well the predicted values ​​fit the actual values. 2 The value range of is [0,1]. The closer it is to 1, the better the model fit is.

[0032] In this case, a 5-fold cross-validation strategy was used during model training. The base model was trained on the entire training set, and the prediction results in each fold were provided as input to the Ridge meta-model. This cross-validation method ensures the generalization performance of the meta-model on unseen data. The random seed was set to 42 in the experiment to ensure the repeatability and stability of the results. The case selected decision tree, random forest and XGBoost as comparison algorithms.

[0033] In the realistic data value scoring of electrochemical energy storage power stations, the coefficients of determination of decision tree, random forest, XGBoost and SVR-XGBoost-Ridge methods are 0.82, 0.93, 0.95 and 0.97, respectively, and the root mean square errors are 0.117, 0.069, 0.058 and 0.041. The results show that the SVR-XGBoost-Ridge method is superior to other comparison methods in both fitting accuracy and prediction accuracy. This method significantly improves the performance of the model by combining the powerful generalization ability of SVR, the efficient boosting algorithm of XGBoost and the regularization processing of Ridge regression. Compared with traditional methods, the SVR-XGBoost-Ridge integrated model shows significant performance improvement when dealing with complex data sets by combining the advantages of support vector regression (SVR), XGBoost and Ridge regression. SVR can effectively handle nonlinear relationships, XGBoost performs well in capturing feature interactions and processing high-dimensional data, and Ridge regression further enhances the generalization ability of the model by introducing L2 regularization. Compared with a single model, SVR-XGBoost-Ridge has obvious advantages in reducing overfitting, improving prediction accuracy, and enhancing model robustness. By comprehensively utilizing the different characteristics of each base model, the model achieves significant optimization in indicators such as root mean square error and determination coefficient, fully demonstrating its superiority in various regression tasks. Therefore, the SVR-XGBoost-Ridge method has a high application value in electrochemical energy storage data analysis, and provides more accurate support for the optimization and development of energy storage systems.

Claims

1. A value assessment method for electrochemical energy storage station based on SVR-XGBoost-Ridge ensemble learning, characterized in that: The following steps are involved: Step 1: Train the base learner: Use the training data to train XGBoost and SVM respectively to obtain two base learner models; Step 2: Generate the prediction results of the base learner: Use the trained XGBoost and SVM to predict the training set and test set to obtain the prediction results of the base learner; Step 3: Train the meta-learner: Use the prediction results of the base learner as feature input, train the ridge regression model, integrate the prediction results of the base learner, and generate the final prediction model; Step 4: Integrated model prediction: Use the trained ridge regression model to integrate the prediction results of the base learner to obtain a comprehensive prediction result; Step 5: Evaluate and select the best model: Evaluate the performance of the entire stacked model through cross-validation and select the model with the best performance; Step 6: Model tuning: Based on the evaluation results, make necessary adjustments to the base learner and meta learner to optimize model performance. Step 7: Deploy and apply the model: Apply the final optimized stacking model to the actual data, make predictions, and analyze the results.

Images