Battery state-of-charge prediction method and equipment based on ensemble learning model

Abstract

A battery state-of-charge prediction method, apparatus and device based on an ensemble learning model, and a medium, the method comprising: constructing a sample data set according to charge and discharge characteristic data and a real state-of-charge of a battery, and dividing a part in the sample data set as a training sample set into a plurality of mutually exclusive subsets; taking one subset as a verification set and the remaining subsets as a training set in turn, performing cross verification training on the base learner, and obtaining a first predicted charge state output by the base learner according to the verification set; training a meta-learner by using the first predicted charge state in combination with the training sample set; through the charge state prediction model constructed by integrating the trained base learner and meta learner, the charge state of the to-be-detected battery is predicted, so that the technical problems of insufficient prediction precision of the charge state of the battery, weak model generalization ability and poor adaptability to complex battery behaviors in related technologies are solved; the precision and stability of battery charge state prediction are remarkably improved, and the generalization ability of the model and the adaptability to the complex battery state are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery state prediction, and in particular to a battery state of charge prediction method and device based on an ensemble learning model, equipment and a medium. BACKGROUND

[0002] With the wide application of electric vehicles and the continuous development of mobile devices, lithium ion batteries, as one of the key components of electric vehicles, have many problems such as flammability, explosiveness, volatility, and corrosion due to their liquid organic electrolyte. In order to overcome these problems, solid-state batteries are being accelerated. Solid-state batteries have great application potential due to their high safety and high energy density. However, the research on solid-state batteries is still in its infancy, and the internal solid-state electrolyte replaces the liquid electrolyte, resulting in more complex internal electrochemical processes, which brings many difficulties to the construction of electrochemical modeling and equivalent circuit model.

[0003] In the battery management system, accurately predicting the state of charge (SOC) of the solid-state battery is of great significance to the battery management system, the efficiency of the device operation, and the battery life. At present, the commonly used state of charge estimation methods for lithium batteries can be roughly divided into three categories: experimental test-based methods, model-driven methods, and data-driven methods. Although the experimental test-based method is accurate, it is high in cost, time-consuming, and difficult to apply in real time; the model-driven method relies on accurate physical models, which is difficult to accurately model the complex electrochemical process of solid-state batteries, and the model parameter estimation is complex and has poor adaptability; the data-driven method does not rely on accurate physical models, but learns the relationship between the state of charge and the related features through a large amount of measured data, and has good adaptability to the model parameter changes and complex electrochemical processes of solid-state batteries.

[0004] However, the traditional machine learning model often has problems such as insufficient model generalization ability and inaccurate fitting of complex data relationships when predicting the state of charge of solid-state batteries. Therefore, it is of great practical significance to develop a more efficient, accurate and robust state of charge prediction method for solid-state batteries. SUMMARY

[0005] The present application provides a battery state of charge prediction method, device, equipment and medium based on an ensemble learning model, which can solve the technical problems of insufficient battery state of charge prediction accuracy, weak model generalization ability and poor adaptability to complex battery behavior in the prior art.

[0006] In a first aspect, the embodiments of the present application provide a battery state of charge prediction method based on an ensemble learning model, which comprises: constructing a sample data set according to the charge-discharge characteristic data and the real state of charge of the battery, and dividing the sample data set as part of a training sample set into a plurality of mutually exclusive subsets; alternately taking one of the subsets as a validation set and the remaining subsets as training sets, performing cross-validation training on the base learner, and obtaining a first predicted state of charge output by the base learner according to the validation set; training the meta-learner using the first predicted state of charge in combination with the training sample set; predicting the state of charge of a battery to be tested through a state of charge prediction model constructed by integrating the trained base learner and the meta-learner.

[0007] In combination with the first aspect, in an implementation, constructing a sample data set according to the charge-discharge characteristic data and the real state of charge of the battery includes: constructing an original data set according to the original charge-discharge characteristic data and the real state of charge of the battery during cyclic charge-discharge; performing correlation analysis on each charge-discharge characteristic and real state of charge in the original data set, and eliminating charge-discharge characteristic types with an absolute correlation coefficient less than a preset threshold; evaluating the importance score of each retained charge-discharge characteristic for the real state of charge through a random forest regression model; selecting a specific percentage of charge-discharge characteristics with the highest importance score to construct the sample data set in combination with the real state of charge; wherein the types of original charge-discharge characteristics include voltage, current, internal resistance, temperature, corresponding collection time stamp, and cumulative charge-discharge capacity of the battery.

[0008] In combination with the first aspect, in an implementation, dividing the sample data set as part of a training sample set into a plurality of mutually exclusive subsets includes: dividing the sample data set into independent training sample sets and test sample sets according to a preset ratio; equally dividing the training sample sets into a plurality of mutually exclusive subsets.

[0009] In combination with the first aspect, in an implementation, alternately taking one of the subsets as a validation set and the remaining subsets as training sets, performing cross-validation training on the base learner, and obtaining a first predicted state of charge output by the base learner according to the validation set includes: alternately taking one of the subsets in the training sample set as a validation set and the remaining subsets as training sets; training each sub-model in the base learner using the training set, and obtaining a first predicted state of charge output by the sub-model through the trained sub-model for the corresponding validation set; For each sub-model, the first predicted charge states corresponding to all the validation sets are spliced in the original order to form a complete prediction vector of the sub-model.

[0010] In combination with the first aspect, in an implementation, the meta-learner is trained using the first predicted charge states in combination with the training sample set, including: The complete prediction vectors of all the sub-models in the base learner are transversely spliced with the charge and discharge feature data in the training sample set to obtain an enhanced fusion feature matrix; The meta-learner is trained using the enhanced fusion feature matrix and the real charge states in the training sample set.

[0011] In combination with the first aspect, in an implementation, the method further includes: The charge and discharge feature data in the test sample set is input into the trained base learner to obtain second predicted charge states output by the base learner; The second predicted charge states and the charge and discharge feature data in the test sample set are spliced to generate a test fusion feature matrix; The test fusion feature matrix is input into the trained meta-model to obtain a charge state test prediction result; The confidence of the base learner and the meta-learner is determined according to the absolute error, the mean absolute error and the root mean square error between the real charge states in the test sample set and the charge state test prediction result.

[0012] In combination with the first aspect, in an implementation: The sub-models of the base learner include a random forest model, a gradient boosting decision tree model and a Gaussian process regression model, and the meta-learner includes a gradient boosting decision tree model.

[0013] The second aspect, the application embodiment provides a battery charge state prediction device based on an integrated learning model, the battery charge state prediction device based on the integrated learning model includes: A data processing module is configured to construct a sample data set according to charge and discharge feature data and real charge states of a battery, and divide the sample data set into a plurality of mutually exclusive subsets as part of a training sample set. A first training module is configured to take one of the subsets as a validation set and the remaining subsets as training sets in turn, cross-validate and train a base learner, and obtain first predicted charge states output by the base learner according to the validation set. A second training module is configured to train a meta-learner using the first predicted charge states in combination with the training sample set. a state prediction module configured to predict the state of charge of the battery under test by a state of charge prediction model constructed by integrating the trained base learner and the meta-learner.

[0014] In a third aspect, an embodiment of the present application provides a battery state of charge prediction device based on an integrated learning model, which comprises a processor, a memory, and a battery state of charge prediction program based on an integrated learning model stored in the memory and executable by the processor, wherein the battery state of charge prediction program based on an integrated learning model, when executed by the processor, implements the steps of the battery state of charge prediction method based on an integrated learning model according to any one of the preceding aspects.

[0015] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a battery state of charge prediction program based on an integrated learning model, wherein the battery state of charge prediction program based on an integrated learning model, when executed by a processor, implements the steps of the battery state of charge prediction method based on an integrated learning model according to any one of the preceding aspects.

[0016] The technical scheme provided by the embodiment of the present application has the following beneficial effects: By constructing a sample data set according to the charge-discharge characteristic data and the real state of charge of the battery, and dividing the sample data set as part of the training sample set into a plurality of mutually exclusive subsets; in turn, one of the subsets is used as a validation set, and the remaining subsets are used as training sets, the base learner is cross-validated and trained, and the first predicted state of charge output by the base learner according to the validation set is obtained; the first predicted state of charge is used in combination with the training sample set to train the meta-learner; and the state of charge prediction model constructed by integrating the trained base learner and the meta-learner is used to predict the state of charge of the battery under test, the technical problems of insufficient battery state of charge prediction accuracy, weak model generalization ability, and poor adaptability to complex battery behavior in the related art are solved, the accuracy and stability of battery state of charge prediction are significantly improved, and the generalization ability and adaptability of the model to complex battery states are enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of an embodiment of the battery state of charge prediction method based on an integrated learning model of the present application; Figure 2 A specific flowchart of the battery state of charge prediction method based on an integrated learning model of the present application; Figure 3 A detailed flowchart of model training in the present application; Figure 4A functional module schematic diagram of an embodiment of a battery state of charge prediction device based on an integrated learning model of the present application; Figure 5 A hardware structure schematic diagram of a battery state of charge prediction device based on an integrated learning model involved in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] In order to make the purpose, technical solutions and advantages of the present application clearer, the embodiments of the present application will be described in further detail below with reference to the drawings.

[0020] In a first aspect, the embodiments of the present application provide a battery state of charge prediction method based on an integrated learning model. The method can be used for state of charge prediction of solid-state batteries or other types of batteries.

[0021] In an embodiment, with reference to Figure 1 , Figure 1 A flowchart of a first embodiment of a battery state of charge prediction method based on an integrated learning model of the present application. As Figure 1 shown, the battery state of charge prediction method based on an integrated learning model includes: Step S101, constructing a sample data set according to the charge-discharge characteristic data and the true state of charge of the battery, and dividing the sample data set as part of the training sample set into a plurality of mutually exclusive subsets. Wherein, in the present embodiment, Specifically, constructing a sample data set according to the charge-discharge characteristic data and the true state of charge of the battery, specifically including the following steps: Step S1011, constructing an original data set according to the original charge-discharge characteristic data and the true state of charge of the battery during cyclic charging and discharging.

[0022] As an example, as Figure 2 shown, the cyclic charging and discharging test data of the solid-state battery under different temperature conditions (such as -20℃, -10℃, 0℃, 25℃, 45℃) are obtained as the original charge-discharge characteristic data, and the true state of charge (true SOC) of the battery is recorded as the label to construct the original data set. When performing the cyclic charging and discharging test, the sampling frequency of the data is set to be no less than 1 Hz to ensure the integrity and continuity of the data.

[0023] wherein the acquired original charge-discharge characteristics include voltage (V), current (I), internal resistance (R), temperature (T), corresponding acquisition time stamp (t) and cumulative charge-discharge capacity (Ah) of the battery.

[0024] As a preferred embodiment, after the original charge-discharge characteristic data is acquired, the original charge-discharge characteristic data is preprocessed, including: using sliding average filtering or wavelet denoising technology to eliminate measurement noise of the original charge-discharge characteristic data, using linear interpolation or KNN interpolation method to fill in missing values, and normalizing all features. After preprocessing the original charge-discharge characteristic data, the original data set D ∈ R N×M , wherein N is the total number of samples, and M is the feature dimension (charge-discharge characteristic type).

[0025] Step S1012, correlation analysis is performed on each charge-discharge characteristic in the original data set and the true state of charge, and the charge-discharge characteristic type with an absolute value of correlation coefficient less than a preset threshold is removed.

[0026] Step S1013, the importance score of each charge-discharge characteristic data retained for the true state of charge is evaluated by a random forest regression model, and the charge-discharge characteristic type with the highest importance score in a specific percentage is selected, and the true state of charge is combined to construct the sample data set.

[0027] For example, in this embodiment, a two-stage feature selection strategy is adopted. In the first stage, feature selection is performed by correlation analysis, specifically including: calculating the Pearson correlation coefficient between various charge-discharge characteristics in the original data set and the true state of charge label, and removing weakly correlated features with an absolute value of correlation coefficient less than a preset threshold of 0.3 to reduce the interference of redundant features on model training.

[0028] In the second stage, feature selection is realized by sorting the importance of features by random forest, specifically including: using a random forest regression model to train a preliminary model, outputting the importance score of each charge-discharge characteristic retained in the first stage for the true state of charge, and retaining the charge-discharge characteristics with top 80% importance score, thereby improving the calculation efficiency and prediction performance of the model.

[0029] After two-stage feature selection, the optimized sample data set D ∈ R N×K is obtained as the subsequent model input, wherein K < M.

[0030] It is worth noting that in this embodiment, a two-stage feature selection strategy is adopted, which can effectively reduce the interference of redundant features on model training and improve the calculation efficiency and accuracy of subsequent prediction results.

[0031] In an embodiment, the sample data set is divided into multiple mutually exclusive subsets as part of the training sample set, specifically including: dividing the sample data set into independent training sample set and test sample set according to a preset ratio; dividing the training sample set into multiple mutually exclusive subsets.

[0032] For example, the sample data set is randomly divided into independent training sample set D train and test sample set D test . Among them, the test sample set D test does not participate in any model training, validation or parameter adjustment throughout the training process, and is only used for final performance evaluation.

[0033] In order to support the generation of unbiased meta features in the meta-learner training process, the training sample set D train is divided into 5 mutually exclusive and equal subsets: D1∪D2∪D3∪D4∪D5, D i ∩ D j = ∅(i ≠ j) It is worth noting that the sample data set is divided into independent training sample set and test sample set according to a ratio of 5:1, ensuring that the test sample set does not participate in model training throughout the process to ensure the objectivity of the evaluation results. Further dividing the training sample set into five mutually exclusive subsets lays the foundation for subsequent 5-fold cross-validation training.

[0034] Step S102, in turn, one of the subsets is used as a validation set, and the remaining subsets are used as training sets, and the base learner is cross-validated and trained, and the first predicted charge state output by the base learner according to the validation set is obtained.

[0035] Specifically, in turn, one of the subsets in the training sample set is used as a validation set, and the remaining subsets are used as training sets; for each sub-model in the base learner, the training set is used for training, and the corresponding validation set is predicted through the trained sub-model, and the first predicted charge state output by the sub-model is obtained; for each sub-model, the first predicted charge state corresponding to all validation sets is spliced in the original order to form a complete prediction vector of the sub-model.

[0036] For example, the first predicted charge state of each sub-model is spliced in the original order to form a complete prediction vector of the base learner. Figure 3As shown, the base learner in this embodiment includes three sub-models: Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Gaussian Process Regression (GPR).

[0037] For each sub-model M k For each ∈{RF, GBDT, GPR}, perform the following cross-validation training process: Select one subset i = 1, 2, 3, 4, 5 from the five subsets in turn as the validation set, and use the remaining four subsets. Training sub-model Using the trained sub-model Based on the i-th subset of the verification set Predict the battery's state of charge (SOC) and obtain the result based on... The first predicted charge state .

[0038] For a sub-model in the base learner, the first predicted charge states of the five validation sets are concatenated in their original order to form the complete prediction vector of that sub-model:

[0039] in, For sub-model M k The complete prediction vector, The first predicted charge state when the first subset D1 is used as the validation set. The first predicted charge state when the second subset D2 is used as the validation set. The first predicted charge state when the third subset D3 is used as the validation set. The first predicted charge state when the fourth subset D4 is used as the validation set. This represents the first predicted charge state when the fifth subset D5 is used as the validation set. The complete prediction vector is the generalized prediction output of this sub-model on the training set, also known as the meta-feature.

[0040] Repeat the above process to obtain the complete prediction vectors of the random forest model in the base learners. The complete prediction vector of the gradient boosting decision tree model And the complete prediction vector of Gaussian process regression. .

[0041] It is worth mentioning that the present application adopts 5-fold cross-validation (5-Fold Cross-Validation) to train the base learners and generate meta-features for the training of the meta-learner. The core purpose of this design is to generate a prediction value made by a model that has not seen the sample under the premise of not introducing information leakage for each training sample, thereby ensuring the generalization ability and reliability of the subsequent meta-learner fusion.

[0042] Compared with the traditional method of directly using the fitting results of the training set as input, the present application realizes unbiased construction of meta-features through a structured process, significantly enhances the robustness of the integrated model, and embodies substantial improvement in the technical implementation path. The present application introduces a strict cross-validation mechanism in the field of solid-state battery SOC prediction, aiming to generate unbiased meta-features, effectively avoiding the information leakage problem caused by model self-prediction, and improving the reliability of the overall architecture.

[0043] Step S103, training the meta-learner using the first predicted state of charge in combination with the training sample set.

[0044] In an embodiment, the complete prediction vector of all sub-models in the base learner is horizontally spliced with the charge and discharge feature data in the training sample set to construct an enhanced fusion feature matrix:

[0045] wherein, is the enhanced fusion feature matrix, is the charge and discharge feature data in the training sample set, and the corresponding true state of charge label in the training sample set is .

[0046] In this embodiment, gradient boosting decision tree (GBDT) can be selected as the meta-learner, which is trained using . The task of the meta-learner is to automatically learn the nonlinear combination relationship between the prediction results of each base learner, realizing optimal weighted fusion.

[0047] It is worth mentioning that GBDT is selected as the meta-learner because of its strong non-linear fitting ability, which can dynamically adjust the weights of each base model and avoid subjective bias caused by manually setting the fusion coefficient. In this step, the meta-features of the three sub-models in the base learner are horizontally spliced with the charge and discharge feature data in the training sample set to form an enhanced fusion feature matrix, and gradient boosting decision tree is used as the meta-learner for non-linear fusion training. This design breaks through the limitations of simple average or fixed weight weighting, uses the strong non-linear fitting ability of GBDT to automatically learn the optimal combination relationship between each base model, realizes dynamic weight distribution, makes the fusion process more intelligent and adaptive, and further improves the prediction accuracy.

[0048] In an embodiment, the method further comprises: inputting the charge and discharge feature data in the test sample set into the trained base learner to obtain a second predicted state of charge output by the base learner; splicing the second predicted state of charge and the charge and discharge feature data in the test sample set to generate a test fusion feature matrix; and inputting the test fusion feature matrix into the trained meta-model to obtain a test prediction result of the state of charge.

[0049] For example, the three sub-models in the base learner are used to predict the state of charge SOC according to the test sample set Dtest, and the second predicted state of charge output by the three sub-models is obtained. 、 、 .

[0050] The second predicted state of charge results of the base learner are spliced with the charge and discharge feature data in the test sample set 、 、 to form a test fusion feature matrix:

[0051] wherein, is the test fusion feature matrix.

[0052] Then, the test fusion feature matrix is input into the trained meta-learner to obtain the output test prediction result of the state of charge:

[0053] wherein, is the test prediction result of the state of charge.

[0054] It is worth mentioning that this test process ensures the consistency and integrity of the training and testing stages, and ensures the stable performance of the model in real application scenarios. ​

[0055] Further, according to the absolute error, the mean absolute error and the root mean square error between the true charge state in the test sample set and the charge state test prediction result, the credibility of the base learner and the meta learner is determined.

[0056] Exemplarily, in order to objectively evaluate the performance and prediction accuracy of the model, the present application adopts three error evaluation indexes of absolute error (AE), mean absolute error (MAE) and root mean square error (RMSE) to evaluate the credibility of the model.

[0057] The absolute error (AE) represents the absolute deviation between the predicted value and the true value of a single sample, and the calculation formula is:

[0058] The mean absolute error (MAE) reflects the average absolute level of the prediction error of all samples, and the calculation formula is:

[0059] The root mean square error (RMSE) measures the concentration degree of the deviation between the predicted value and the true value, and is more sensitive to large errors, and the calculation formula is:

[0060] Wherein, n represents the total number of samples, is the true SOC label of the i th sample, is the corresponding charge state test prediction result. The above indexes jointly constitute a multi-dimensional evaluation system for the accuracy of the model, which can comprehensively reflect the performance of the method in different error scales.

[0061] Step S104, predicting the charge state of the battery to be tested by the charge state prediction model integrated by the trained base learner and meta learner.

[0062] Specifically, the charge and discharge feature data of the battery to be tested is input into the charge state prediction model, the base learner therein outputs the predicted charge state of the battery to be tested according to the charge and discharge feature data of the battery to be tested, a fusion feature matrix is constructed according to the predicted charge state and the charge and discharge feature data of the battery to be tested, and the fusion feature matrix is input into the meta learner, and the meta learner outputs the predicted charge state of the battery to be tested.

[0063] The battery state of charge prediction method based on the integrated learning model provided in the application solves the technical problem of insufficient state estimation accuracy under a complex electrochemical system by applying a heterogeneous integrated learning framework based on cross-validation to solid-state battery SOC prediction through systematic step design.

[0064] Specifically, first, the training data is divided into 5 folds, and a "round-robin validation" method is used, that is, each time 4 folds are used to train a model (including random forest, gradient boosting tree and Gaussian process regression), and then the remaining 1 fold is predicted. The advantage of this is that the prediction result of each sample comes from a "model that has not seen it", avoiding overly optimistic or distorted prediction results, thereby generating more objective and reliable "meta-features". Subsequently, these meta-features are combined with the battery charge and discharge feature data (such as voltage, current, temperature, etc.) to form a more comprehensive feature set, which is input to a stronger model (gradient boosting tree GBDT) as the decision center for learning. This center will automatically determine which base model output to trust more under different conditions, achieving intelligent weighted fusion rather than simply averaging. Finally, after training is complete, the trained base learner is used to predict new data, and then the prediction results are combined with the charge and discharge feature data of the new data as input to the trained meta-learner, which performs the final prediction and outputs the SOC result. The entire process not only ensures the unbiasedness of the intermediate features, but also realizes the complementary advantages of multiple models, significantly improving the accuracy and robustness of SOC prediction.

[0065] The advantages of this method include: using a 5-fold cross-validation mechanism to generate meta-features, ensuring that the prediction value of each sample is output by a "model that has not participated in the training of this sample", effectively avoiding the information leakage problem caused by model self-prediction in traditional integrated methods. This approach significantly improves the unbiasedness and reliability of meta-features, allowing the subsequent fusion model to obtain more realistic and more generalizable input signals during training, thereby improving the stability and accuracy of the overall prediction.

[0066] By introducing three sub-models of random forest, gradient boosting decision tree and Gaussian process regression to construct heterogeneous base learners, and then inputting their prediction results and original features to GBDT for nonlinear fusion, the complementary advantages of different models are fully utilized. This fusion structure no longer relies on fixed weights or simple averaging set by humans, but instead uses GBDT to automatically learn the contribution of each model under different input conditions, achieving dynamic weighting and significantly enhancing the fitting ability for complex nonlinear relationships, further improving the accuracy of SOC estimation.

[0067] After the training is completed, the test set features are combined with the independent prediction results of the three sub-models in the base learner to construct a fusion input matrix in the test stage, and the final result is output by the trained meta-learner. This approach ensures the consistency of the training and inference processes and avoids feature space mismatch problems; at the same time, since the meta-learner comprehensively considers the original physical features and the model prediction information from multiple perspectives, the output has stronger robustness and anti-interference ability, especially in actual working conditions such as battery aging and temperature changes, which can still maintain excellent performance.

[0068] In a second aspect, the embodiments of the present application also provide a battery state of charge prediction device based on an ensemble learning model.

[0069] In an embodiment, the battery state of charge prediction device based on the ensemble learning model comprises: Figure 4 Figure 4 FIG. 1 is a schematic diagram of the functional modules of an embodiment of the battery state of charge prediction device based on the ensemble learning model of the present application. As shown in FIG. 1, the battery state of charge prediction device based on the ensemble learning model comprises: Figure 4 a data processing module configured to construct a sample data set according to the charge-discharge feature data and the true state of charge of the battery, and divide the sample data set into multiple mutually exclusive subsets as part of the training sample set; a first training module configured to use one of the subsets as a validation set and the remaining subsets as training sets to perform cross-validation training on the base learner, and obtain a first predicted state of charge output by the base learner according to the validation set; a second training module configured to train the meta-learner using the first predicted state of charge in combination with the training sample set; a state prediction module configured to predict the state of charge of the battery to be tested by a state of charge prediction model integrated by the trained base learner and meta-learner.

[0070] Further, in an embodiment, the data processing module is further configured to: construct an original data set according to the original charge-discharge feature data and the true state of charge of the battery during the cycle charge-discharge; perform correlation analysis on each charge-discharge feature and the true state of charge in the original data set, and eliminate the charge-discharge feature types with an absolute correlation coefficient less than a preset threshold; evaluate the importance score of each retained charge-discharge feature for the true state of charge by a random forest regression model; select a specific percentage of the charge-discharge features with the highest importance score to construct the sample data set in combination with the true state of charge; wherein the types of original charge-discharge features include the voltage, current, internal resistance, temperature, corresponding collection timestamp, and cumulative charge-discharge capacity of the battery.​​

[0071] Further, in an embodiment, the data processing module is further configured to: divide the sample data set into an independent training sample set and a test sample set according to a preset proportion; divide the training sample set into a plurality of mutually exclusive subsets.

[0072] Further, in an embodiment, the first training module is further configured to: take one subset in the training sample set as a validation set and the remaining subsets as training sets in turn; for each sub-model in the base learner, use the training set to train and use the trained sub-model to predict the corresponding validation set to obtain a first predicted state of charge output by the sub-model; for each sub-model, splice the first predicted state of charge corresponding to all validation sets in the original order to form a complete prediction vector of the sub-model.

[0073] Further, in an embodiment, the second training module is further configured to: splice the complete prediction vectors of all sub-models in the base learner and the charge and discharge feature data in the training sample set horizontally to obtain an enhanced fusion feature matrix; use the enhanced fusion feature matrix and the real state of charge in the training sample set to train the meta-learner.

[0074] Further, in an embodiment, the device further comprises a test module configured to: input the charge and discharge feature data in the test sample set into the trained base learner to obtain a second predicted state of charge output by the base learner; splice the second predicted state of charge and the charge and discharge feature data in the test sample set to generate a test fusion feature matrix; input the test fusion feature matrix into the trained meta-model to obtain a state of charge test prediction result; determine the credibility of the base learner and the meta-learner according to the absolute error, the mean absolute error, and the root mean square error between the real state of charge in the test sample set and the state of charge test prediction result.

[0075] Further, in an embodiment, the sub-models of the base learner include a random forest model, a gradient boosting decision tree model, and a Gaussian process regression model, and the meta-learner includes a gradient boosting decision tree model.

[0076] The functions of each module in the battery charge state prediction device based on the integrated learning model correspond to the steps in the battery charge state prediction method based on the integrated learning model, and the functions and implementation processes will not be repeated here.

[0077] In a third aspect, the embodiments of the present application provide a battery charge state prediction device based on an integrated learning model. The battery charge state prediction device based on the integrated learning model can be a personal computer (PC), a notebook computer, a server, or other device with data processing function.

[0078] Reference Figure 5 , Figure 5 FIG. 1 is a schematic diagram of a hardware structure of a battery charge state prediction device based on an integrated learning model according to an embodiment of the present application. In the embodiments of the present application, the battery charge state prediction device based on the integrated learning model can include a processor, a memory, a communication interface, and a communication bus.

[0079] The communication bus can be of any type, used to interconnect the processor, the memory, and the communication interface.

[0080] The communication interface includes an input / output (I / O) interface, a physical interface, and a logical interface, etc., which are used to interconnect the devices inside the battery charge state prediction device based on the integrated learning model, and are used to interconnect the battery charge state prediction device based on the integrated learning model with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, a fiber interface, an ATM interface, etc.; the user device can be a display (Display), a keyboard (Keyboard), etc.

[0081] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0082] The processor can be a general-purpose processor, which can invoke a battery charge state prediction program based on an ensemble learning model stored in the memory and execute the battery charge state prediction method based on the ensemble learning model provided by the embodiments of the present application. For example, the general-purpose processor can be a central processing unit (CPU). The method performed when the battery charge state prediction program based on the ensemble learning model is invoked can refer to various embodiments of the battery charge state prediction method based on the ensemble learning model of the present application, which will not be described here.

[0083] Those skilled in the art can understand that the hardware structure shown in the above-mentioned embodiments is not a limitation of the present application, and can include more or fewer components than those shown, or combine certain components, or different component arrangements. Figure 5

[0084] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium.

[0085] The computer readable storage medium of the present application stores a battery charge state prediction program based on an ensemble learning model, wherein when the battery charge state prediction program based on the ensemble learning model is executed by a processor, the steps of the battery charge state prediction method based on the ensemble learning model as described above are implemented.

[0086] The method implemented when the battery charge state prediction program based on the ensemble learning model is executed can refer to various embodiments of the battery charge state prediction method based on the ensemble learning model of the present application, which will not be described here.

[0087] It should be noted that the above-mentioned sequence numbers of the embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0088] The terms "include" and "have" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device. The terms "first", "second" and "third" and the like descriptions are used to distinguish different objects, and do not represent the order or limit the types of "first", "second" and "third".

[0089] ​In the description of the embodiments of the present application, "exemplary", "for example", "e.g." or "for instance" are used on the basis that a person of ordinary skill in the art will be able to bring to mind many examples of a given implementation as the implementation described in the embodiments of the present application is exemplary. Any embodiment or design scheme described as "exemplary", "for example" or "for instance" in the embodiments of the present application should not be interpreted as being more preferred than other embodiments or design schemes. Rather, the use of "exemplary", "for example", "e.g." or "for instance" is intended to present concepts in a particular manner.

[0090] In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in the text only describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. In addition, in the description of the embodiments of the present application, "multiple" means two or more than two.

[0091] In some of the processes described in the embodiments of the present application, a plurality of operations or steps are included in a specific order, but it should be understood that these operations or steps can be executed or performed in parallel or in an order different from that in which they appear in the embodiments of the present application. The serial number of the operation is only used to distinguish different operations, and the serial number itself does not represent any execution order. In addition, these processes can include more or fewer operations, and these operations or steps can be executed in sequence or in parallel, and these operations or steps can be combined.

[0092] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a plurality of instructions for making a terminal device execute the method described in each embodiment of the present application.

[0093] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation using the content of the specification and drawings of the present application, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A battery state of charge prediction method based on an ensemble learning model, characterized by, The battery charge state prediction method based on the ensemble learning model comprises: constructing a sample data set according to the charge-discharge characteristic data and the real charge state of the battery, and dividing the sample data set as part of the training sample set into a plurality of mutually exclusive subsets; alternately taking one of the subsets as a validation set and the remaining subsets as training sets, cross-validation training the base learner, and obtaining a first predicted charge state output by the base learner according to the validation set; training the meta-learner using the first predicted charge state in combination with the training sample set; predicting the charge state of the battery to be tested through a charge state prediction model constructed by the trained base learner and meta-learner. 2.The battery state of charge prediction method based on an ensemble learning model of claim 1, wherein, The sample data set is constructed according to the charge-discharge characteristic data and the real charge state of the battery, comprising: constructing an original data set according to the original charge-discharge characteristic data and the real charge state of the battery during cyclic charge-discharge; performing correlation analysis on each charge-discharge characteristic and real charge state in the original data set, and eliminating charge-discharge characteristic types with an absolute correlation coefficient less than a preset threshold; evaluating the importance score of each retained charge-discharge characteristic for the real charge state through a random forest regression model; selecting a specific percentage of charge-discharge characteristics with the highest importance score to construct the sample data set in combination with the real charge state; wherein the types of original charge-discharge characteristics include voltage, current, internal resistance, temperature, corresponding collection time stamp, and cumulative charge-discharge capacity of the battery. 3.The battery state of charge prediction method based on an ensemble learning model of claim 1, wherein, The sample data set is divided as part of the training sample set into a plurality of mutually exclusive subsets, comprising: dividing the sample data set into independent training sample sets and test sample sets according to a preset ratio; dividing the training sample set into a plurality of mutually exclusive subsets. 4.The battery state of charge prediction method based on an ensemble learning model of claim 3, wherein, Alternately taking one of the subsets as a validation set and the remaining subsets as training sets, cross-validation training the base learner, and obtaining a first predicted charge state output by the base learner according to the validation set, comprising: alternately taking one of the subsets in the training sample set as a validation set and the remaining subsets as training sets; training each sub-model in the base learner using the training set, and obtaining a first predicted charge state output by the sub-model through the trained sub-model for the corresponding validation set; for each sub-model, concatenating the first predicted charge states corresponding to all validation sets in the original order to form a complete prediction vector for the sub-model. 5.The battery state of charge prediction method based on ensemble learning model of claim 4, wherein, Training the meta-learner using the first predicted charge state in combination with the training sample set, comprising: horizontally concatenating the complete prediction vectors of all sub-models in the base learner with the charge-discharge characteristic data in the training sample set to obtain an enhanced fusion feature matrix; training the meta-learner using the enhanced fusion feature matrix and the real charge state in the training sample set. 6.The battery state of charge prediction method based on ensemble learning model of claim 3, wherein, The method further comprises: inputting the charge-discharge characteristic data in the test sample set into the trained base learner to obtain a second predicted charge state output by the base learner; concatenating the second predicted charge state and the charge-discharge characteristic data in the test sample set to generate a test fusion feature matrix; The test fusion feature matrix is input into the trained meta-model to obtain a charge state test prediction result; According to the absolute error, the mean absolute error, and the root mean square error between the real charge state in the test sample set and the charge state test prediction result, the credibility of the base learner and the meta-learner is determined.

7. The battery charge state prediction method based on an ensemble learning model according to claim 1, wherein: The sub-models of the base learner include a random forest model, a gradient boosting decision tree model, and a Gaussian process regression model, and the meta-learner includes a gradient boosting decision tree model.

8. A battery state of charge prediction apparatus based on an ensemble learning model, characterized by, The battery charge state prediction device based on an ensemble learning model includes: A data processing module configured to construct a sample data set according to the charge and discharge feature data of the battery and the real charge state, and divide the sample data set as part of a training sample set into a plurality of mutually exclusive subsets; A first training module configured to use one of the subsets as a validation set and the remaining subsets as training sets, cross-validate and train the base learner, and obtain a first predicted charge state output by the base learner according to the validation set; A second training module configured to train the meta-learner using the first predicted charge state in combination with the training sample set; A state prediction module configured to predict the charge state of the battery to be tested by a charge state prediction model constructed by the trained base learner and meta-learner.

9. A battery state of charge prediction device based on an ensemble learning model, characterized by, The battery charge state prediction device based on an ensemble learning model includes a processor, a memory, and a battery charge state prediction program based on an ensemble learning model stored on the memory and executable by the processor, wherein when the battery charge state prediction program based on an ensemble learning model is executed by the processor, the steps of the battery charge state prediction method based on an ensemble learning model according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a battery charge state prediction program based on an ensemble learning model, wherein when the battery charge state prediction program based on an ensemble learning model is executed by the processor, the steps of the battery charge state prediction method based on an ensemble learning model according to any one of claims 1 to 7 are implemented.

Images