Battery soh prediction method and system based on ensemble learning

Abstract

The application provides a battery SOH prediction method and system based on ensemble learning, and relates to the field of battery state analysis. The method comprises the following steps: obtaining first power consumption data information and first SOH information; processing the first power consumption data information to obtain second power consumption data information and determining a target data set; constructing an initial prediction model and training and testing the initial prediction model to obtain an ensemble learning model; obtaining third power consumption data information of a battery to be predicted; and predicting the battery SOH information corresponding to the battery to be predicted based on the ensemble learning model. The multiple base learners can exert their respective advantages, improve the accuracy of data analysis, and the meta learner intelligently integrates these results, avoiding the bias and limitations that may exist in a single model. Through the fusion of multiple base learners and meta learners, the model can have high generalization ability and can flexibly adapt to various application scenarios and data sets.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery state analysis, in particular to a battery SOH prediction method and system based on ensemble learning. BACKGROUND

[0002] With the continuous expansion of the application field of energy storage batteries, the management requirements of energy storage equipment on the state of health (SOH) of the batteries are also increasingly high. The SOH of the batteries is difficult to directly measure like the voltage, and the existing direct measurement method needs to be in a laboratory environment and has a high precision requirement on the equipment, and the result obtained is only an estimated value, which is not completely accurate. At present, the method for determining the SOH of the batteries is based on data, and decisions and predictions are made by analyzing and processing the data. The accurate prediction of the SOH of the batteries is of great significance for estimating the service life of the batteries, optimizing the use of the batteries, and preventing battery failures. The SOH of the batteries refers to the ability of the batteries to maintain stable performance parameters and safely and reliably operate under a specific working environment, and the prediction of the SOH is of great significance for predicting the service life of the batteries, optimizing the use of the batteries, and preventing battery failures.

[0003] In the related art, the data-driven model mainly utilizes historical data generated during the operation of the energy storage batteries, such as voltage and current, and utilizes machine learning algorithms, such as support vector machine (SVM), long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), and the like, to predict the state of health of the batteries, which is a current research hotspot.

[0004] However, these methods are all single machine learning methods, different models have advantages and disadvantages, and the models are highly dependent on the integrity and representativeness of the training data, but in actual applications, data often has noise, incompleteness, or bias; limited by the prediction ability of the single model, it is difficult to comprehensively capture complex factors and their mutual relationships in the change of the state of health of the batteries, the generalization ability of the model is insufficient, and it is difficult to adapt to complex and variable actual application scenarios. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a battery SOH prediction method and system based on ensemble learning, which solves the problems of low accuracy, insufficient generalization ability, and difficulty in adapting to complex and variable actual application scenarios in battery SOH prediction.

[0006] To achieve the above purpose, the present application realizes the following technical solutions:

[0007] In a first aspect, an embodiment of the present application provides a battery SOH prediction method based on ensemble learning, which comprises: obtaining first power consumption data information and first SOH information of each target battery in a target battery pack; performing outlier screening on the first power consumption data information, and performing data smoothing processing based on a preset signal processor to obtain second power consumption data information; wherein the second power consumption data information comprises voltage, current and charge / discharge time of the target battery; determining a corresponding first IC curve based on the second power consumption data information, and calculating a peak feature of the first IC curve; determining the peak feature of the first IC curve, the first SOH information and the second power consumption data information as a target data set; selecting a base learner and a meta learner from a plurality of machine learning algorithms and a plurality of statistical models to construct an initial prediction model; training and testing the initial prediction model based on a plurality of target data sets corresponding to a plurality of target batteries in the target battery pack to obtain an ensemble learning model; obtaining third power consumption data information of a battery to be predicted, the third power consumption data information comprising voltage, current and charge / discharge time of the battery to be predicted; and analyzing the third power consumption data information based on the ensemble learning model to predict battery SOH information corresponding to the battery to be predicted.

[0008] According to the first aspect of the embodiment of the present application, the preset signal processor is a Savitzky-Golay filter.

[0009] According to the first aspect of the embodiment of the present application, the peak feature of the first IC curve is determined by a preset differential calculation method, and the differential calculation method satisfies the expression:

[0010]

[0011] In the expression, t0 is the time when discharging starts, t i is the time point of the i-th cycle, Q i is the cumulative discharge capacity of the i-th cycle; ic i represents the incremental capacity value at the i-th time point, Q i+1 is the capacity corresponding to the i+1-th data point, V i+1 is the voltage corresponding to the i+1-th data point, V i is the voltage corresponding to the i-th data point, ΔQ i is the capacity difference between adjacent data points, ΔV i is the voltage difference between adjacent data points.

[0012] According to the first aspect of the embodiment of the present application, the aforementioned initial prediction model comprises two layers, the first layer is a base learner combination, and the second layer is a meta learner; the plurality of machine learning algorithms comprise LSTM, RNN, BiLSTM, BP neural network and random forest algorithm, and the meta learner is one of random forest algorithm and Bayesian algorithm.

[0013] According to a first aspect of the embodiments of the present application, the foregoing training and testing of the initial prediction model based on the plurality of target data sets corresponding to the plurality of target batteries in the target battery pack to obtain the ensemble learning model comprises: dividing the plurality of target data sets corresponding to the plurality of target batteries in the target battery pack into a training set and a test set; inputting the training set into the initial prediction model for training to obtain a target prediction model; inputting the test set into the target test model to obtain SOH prediction information; comparing the SOH prediction information with the first SOH information of the corresponding target battery to determine the target deviation of the two; and in the case where the target deviation is less than a preset threshold, determining that the target prediction model is the ensemble learning model.

[0014] According to the first aspect of the embodiments of the present application, the foregoing inputting the training set into the initial prediction model for training to obtain a target prediction model comprises: inputting the training set into each base learner in the base learner combination; in each base learner, analyzing the third power consumption data information based on the N-fold cross-validation method to generate N test results; stacking the N prediction results of each base learner in a row vertically to form a prediction set; splicing the prediction sets obtained by each base learner in the base learner combination in a column to obtain an input data set of the next level; and inputting the input data set as new features into the meta-learner of the second layer of the initial prediction model for data fusion training and obtaining the target prediction model.

[0015] According to the first aspect of the embodiments of the present application, the foregoing analyzing the third power consumption data information based on the ensemble learning model to predict the battery SOH information corresponding to the battery to be predicted comprises: performing outlier screening on the third power consumption data information, and performing data smoothing processing based on a Savitzky-Golay filter to obtain fourth power consumption data information; determining the corresponding second IC curve based on the fourth power consumption data information, and calculating the peak feature of the second IC curve; determining the peak feature of the second IC curve and the fourth power consumption data information as target power consumption information; inputting the target power consumption information into the ensemble learning model for analysis to obtain the battery SOH information corresponding to the battery to be predicted.

[0016] According to the first aspect of the embodiments of the present application, the battery SOH prediction method further comprises: evaluating the performance of the ensemble learning model by mean absolute error MAE, mean square error MSE and R 2 2 The closer to 1, the higher the accuracy of the ensemble learning model; when the ensemble learning model prediction has a large error, the mean square error MSE will rapidly increase.

[0017] According to the first aspect of the embodiments of the present application, the foregoing mean absolute error MAE satisfies the expression:

[0018] ​

[0019] The aforementioned mean square error MSE satisfies the expression:

[0020]

[0021] R 2 satisfies the expression:

[0022]

[0023] In the formula, y represents the true value of the battery SOH, y i represents the estimated value of the battery SOH, N represents the size of the data, represents the average value of the true value of the battery SOH.

[0024] In a second aspect, the embodiments of the present application provide a battery SOH prediction system based on ensemble learning, which comprises an acquisition module, a processing module, a calculation module, a determination module, a construction module, a training and testing module and a prediction module. The acquisition module is configured to acquire first power consumption data information and first SOH information of each target battery in a target battery pack. The processing module is configured to perform outlier screening on the first power consumption data information, and perform data smoothing processing based on a preset signal processor to obtain second power consumption data information. The second power consumption data information includes voltage, current and charge / discharge time of the target battery. The calculation module is configured to determine a corresponding first IC curve based on the second power consumption data information, and calculate a peak feature of the first IC curve. The determination module is configured to determine the peak feature of the first IC curve, the first SOH information and the second power consumption data information as a target data set. The construction module is configured to select a base learner and a meta learner from a plurality of machine learning algorithms and a plurality of statistical models, and construct an initial prediction model. The training and testing module is configured to train and test the initial prediction model based on a plurality of target data sets corresponding to a plurality of target batteries in the target battery pack, to obtain an ensemble learning model. The acquisition module is further configured to acquire third power consumption data information of a battery to be predicted, the third power consumption data information including voltage, current and charge / discharge time of the battery to be predicted. The prediction module is configured to analyze the third power consumption data information based on the ensemble learning model, to predict battery SOH information corresponding to the battery to be predicted.

[0025] In a third aspect, the embodiments of the present application provide an electronic device, which comprises a processor, a memory and a program stored in the memory and executable on the processor. The program is executed by the processor to implement the battery SOH prediction method based on ensemble learning in the first aspect.

[0026] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium, which stores programs or instructions, and the programs or instructions are executed by a processor to implement the battery SOH prediction method based on ensemble learning in the first aspect.

[0027] The present application provides a battery SOH prediction method and system based on ensemble learning. Compared with the prior art, the following beneficial effects are achieved:

[0028] The present application obtains the first power consumption data information of the target battery pack and the first SOH information to which the target battery pack belongs. After the first power consumption data information is subjected to abnormal screening and smoothing processing to obtain the second power consumption data information, the present application extracts the first IC curve corresponding to the target battery pack on the basis of the second power consumption data information, and determines the peak feature of the first IC curve as part of the target data set, so as to improve the learning efficiency and prediction performance of the model. The present application selects a plurality of base learners from a plurality of machine learning algorithms, and selects a meta-learner from a plurality of statistical models, and then constructs a double-layer initial prediction model. The plurality of base learners can exert their respective advantages to improve the accuracy of data analysis, and the meta-learner intelligently integrates these results to avoid the bias and limitations that may exist in a single model. By fusing the plurality of base learners and the meta-learner, the model can have high generalization ability and can flexibly adapt to various application scenarios and data sets. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0030] Figure 1 is a flowchart of a battery SOH prediction method based on ensemble learning provided by the embodiments of the present application;

[0031] Figure 2 is Figure 1 is an exemplary flowchart of S160 in

[0032] Figure 3 is a structural diagram of a battery SOH prediction system based on ensemble learning provided by the embodiments of the present application;

[0033] Figure 4 is a structural diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0034] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application are described clearly and completely. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the present application.

[0035] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between such entities or operations. Also, the terms "comprises", "comprising", or any other variations thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or other elements inherent to such a process, method, article, or apparatus. Without more limitations, the element defined by the statement "comprises a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus including the element.

[0036] The embodiments of the present application provide a battery SOH prediction method and system based on ensemble learning, which solves the problems of low accuracy, insufficient generalization ability and difficulty in adapting to complex and variable actual application scenarios in battery SOH prediction.

[0037] The technical solutions in the embodiments of the present application are as follows to solve the above technical problems:

[0038] With the continuous expansion of the application field of energy storage batteries, the management requirements of energy storage equipment on the state of health (SOH) of batteries are also getting higher and higher. The SOH of the battery is difficult to be directly measured like voltage. The existing direct measurement method needs to be in a laboratory environment and requires high accuracy of the equipment, and this value is not completely accurate, but an estimated value. At present, the mainstream method to determine the SOH of the battery is still a data-driven method, that is, based on data, decisions and predictions are made by analyzing and processing data. Accurate prediction of the SOH of the battery is of great significance for predicting the life of the battery, optimizing the use of the battery, and preventing battery failure. The SOH of the battery refers to the ability of the battery to maintain stable performance parameters and safely and reliably operate under a specific working environment. The evaluation of SOH is of great significance for predicting the life of the battery, optimizing the use of the battery, and preventing battery failure.

[0039] As the battery is used and aged, its maximum available capacity will gradually decrease, resulting in a decrease in SOH value. Since the energy storage battery is a nonlinear system with a complex electrochemical structure, its performance is greatly affected by internal and external factors. As the running time increases, the side reactions inside the battery will cause the internal resistance of the battery to rise and the maximum available capacity to continuously decrease, which is the battery aging. In addition, the complex operation mechanism and extreme environmental conditions will also exacerbate the battery aging process and even cause the battery to fail. Therefore, high-precision state of health estimation and prediction is a core technical problem that needs to be solved urgently.

[0040] In related technologies, the research methods of energy storage battery SOH prediction mainly focus on two types of methods, namely model-based methods and data-driven methods. The model-based prediction method mainly establishes a physical or mathematical model, conducts in-depth research on the internal structure of the lithium battery, analyzes the relevant physical and chemical processes inside the lithium battery during the charging and discharging process, seeks relevant rules, and selects a relevant prediction model that can characterize the performance degradation for life prediction. For example, Kalman filter method, particle filter method, electrochemical impedance spectroscopy model, etc. The data-driven model mainly uses historical data such as voltage and current generated by the operation of the energy storage battery, and uses machine learning models such as Support Vector Machine (SVM), Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM) to predict the state of health of the battery. It is the current research hotspot.

[0041] However, these methods are all single machine learning methods, and different models have their own advantages and disadvantages. Limited by the prediction ability of a single model, it is difficult to fully capture the complex factors and their mutual relationships in the change of the battery state of health. The method of modeling the physical or chemical process inside the battery is complex and difficult to accurately describe the change of battery performance, and cannot adapt to the complex scene of simultaneous use of multiple types of batteries. Moreover, the model is highly dependent on the integrity and representativeness of the training data. In actual application, the noise, incompleteness or deviation of the data makes it difficult to guarantee the prediction accuracy of the original technology.

[0042] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the drawings in the specification and specific embodiments.

[0043] First, a battery SOH prediction method based on ensemble learning provided by an embodiment of the present application will be introduced.

[0044] The flowchart of the battery SOH prediction method based on ensemble learning provided by an embodiment of the present application is shown in Figure 1As shown, the battery SOH prediction method can include steps S110-S180.

[0045] S110, acquire first power consumption data information and first SOH information of each target battery in a target battery pack.

[0046] S120, perform outlier screening on the first power consumption data information, and perform data smoothing processing based on a preset signal processor to obtain second power consumption data information; wherein the second power consumption data information includes voltage, current and charge / discharge time of the target battery.

[0047] S130, determine a corresponding first IC curve based on the second power consumption data information, and calculate a peak feature of the first IC curve.

[0048] S140, determine the peak feature of the first IC curve, the first SOH information and the second power consumption data information as a target data set.

[0049] S150, select a base learner and a meta learner from a plurality of machine learning algorithms and a plurality of statistical models to construct an initial prediction model.

[0050] S160, train and test the initial prediction model based on a plurality of target data sets corresponding to a plurality of target batteries in the target battery pack to obtain an ensemble learning model.

[0051] S170, acquire third power consumption data information of a battery to be predicted, the third power consumption data information including voltage, current and charge / discharge time of the battery to be predicted.

[0052] S180, analyze the third power consumption data information based on the ensemble learning model to predict battery SOH information corresponding to the battery to be predicted.

[0053] In the embodiments of the present application, the first power consumption data information of the target battery pack and the first SOH information to which the target battery pack belongs are acquired, and the first power consumption data information and the first SOH information corresponding to the target battery pack are used to train and test the model. It can be understood that the first SOH information of the target battery pack can be measured in a laboratory environment. After the first power consumption data information is screened and smoothed to obtain the second power consumption data information, the first IC curve is extracted based on the second power consumption data information, the first IC curve is the incremental capacity curve corresponding to the target battery pack, and the peak feature of the first IC curve is determined as part of the target data set to improve the learning efficiency and prediction performance of the model.

[0054] It should be noted that the base learner is selected from a plurality of machine learning algorithms, and the meta learner is selected from a plurality of statistical models, and then a double-layer initial prediction model is constructed; it can be understood that the plurality of base learners can exert their respective advantages to improve the accuracy of data analysis, and the meta learner intelligently integrates these results to avoid the bias and limitations of a single model; by fusing the plurality of base learners and the meta learner, the model can have high generalization ability and can flexibly adapt to various application scenarios and data sets.

[0055] In one example, the preset signal processor is a Savitzky-Golay filter.

[0056] It can be understood that the Savitzky-Golay filter is a special low-pass filter that can be used to smooth noise data; the Savitzky-Golay filter can directly process the method from the data smoothing problem in the time domain, rather than defining the characteristics in the frequency domain first and then converting to the time domain as the usual filter. Compared with other similar average methods, the Savitzky-Golay filter can better preserve the distribution characteristics such as relative maximum value, minimum value and width.

[0057] In one example, the peak characteristics of the first IC curve are determined by a preset difference calculation method, and the difference calculation method satisfies the expression:

[0058]

[0059] In the formula, t0 is the time when the discharge starts, t i is the time point of the ith cycle, Q i is the cumulative discharge capacity of the ith cycle; ic i represents the incremental capacity value of the ith time point, Q i+1 is the capacity corresponding to the i+1th data point, V i+1 is the voltage corresponding to the i+1th data point, V i is the voltage corresponding to the ith data point, ΔQ i is the capacity difference of adjacent data points, ΔV i is the voltage difference of adjacent data points.

[0060] In some embodiments, the initial prediction model comprises two layers, a first layer being a base learner combination and a second layer being a meta-learner; the plurality of machine learning algorithms comprises LSTM, RNN, BiLSTM, BP neural network and random forest algorithm, and the meta-learner is one of random forest algorithm and Bayesian algorithm. It can be understood that the base learner is a single machine learning method or model; the base learner can be selected from the aforementioned plurality of machine learning algorithms as needed, and the aforementioned plurality of machine learning algorithms can also comprise other suitable algorithms or models, for example: the aforementioned plurality of machine learning algorithms can also comprise an LSTM model with an attention mechanism.

[0061] In some embodiments, the initial prediction model is trained and tested based on the plurality of target data sets corresponding to the plurality of target batteries in the target battery pack, to obtain an ensemble learning model, i.e., S160 can specifically comprise the following steps:

[0062] S210, dividing the plurality of target data sets corresponding to the plurality of target batteries in the target battery pack into a training set and a test set.

[0063] S220, inputting the training set into the initial prediction model for training to obtain a target prediction model.

[0064] S230, inputting the test set into the target prediction model to obtain SOH prediction information.

[0065] S240, comparing the SOH prediction information with the first SOH information of the corresponding target battery to determine the target deviation therebetween.

[0066] S250, determining that the target prediction model is an ensemble learning model when the target deviation is less than a preset threshold.

[0067] In the embodiments of the present application, it can be understood that the test set corresponds to part of the target batteries, and the first SOH information of the part of the target batteries has been obtained by the present application. In order to train the model and detect the accuracy of the model, the present application determines the first SOH information corresponding to the target battery in addition to the basic data contained in the first power consumption data information, and the first SOH information is the known true value; the present application detects the target prediction model obtained by training through the test set data corresponding to part of the target batteries, and determines that the prediction accuracy of the target prediction model is higher when the SOH prediction information obtained by the target prediction model is close to the corresponding first SOH information, which can meet the actual use demand, and then determines that the target prediction model is an ensemble learning model, and the SOH of the battery to be predicted is analyzed and predicted through the ensemble learning model.

[0068] It should be noted that the first SOH information of the target battery can be measured in an ideal scenario in a laboratory, but it is difficult to directly determine the SOH of each battery under other working conditions. Based on this, the integrated learning model obtained by training and testing of the present application can be used to predict the SOH of the battery under different environmental conditions.

[0069] In one example, the target battery pack includes three target batteries, which can correspond to different models. The peak feature of the first IC curve, the first SOH information and the second power consumption data information of one target battery are taken as a training set; and the peak feature of the first IC curve, the first SOH information and the second power consumption data information of another target battery or another two target batteries are taken as a test set.

[0070] In some embodiments, the foregoing training set is input into the initial prediction model for training to obtain the target prediction model, that is, the foregoing S220 can specifically include the following steps:

[0071] S310, input the training set into each base learner in the base learner combination.

[0072] S320, in each base learner, the third power consumption data information is analyzed based on the N-fold cross-validation method to generate N test results.

[0073] S330, the N prediction results of each base learner are stacked in a row vertically to form a prediction set.

[0074] S340, the prediction sets obtained by each base learner in the base learner combination are spliced in columns to obtain an input data set of the next level.

[0075] S350, the input data set is input into the meta-learner of the second layer of the initial prediction model as new features for data fusion training and obtaining the target prediction model.

[0076] In the embodiments of the present application, it can be understood that the present application forms a complementary model by combining multiple base learners and meta-learners, which ensures that the model can maintain high prediction accuracy and strong robustness when facing complex and variable prediction scenarios.

[0077] In one example, the present application selects m base learners to train the training set respectively. In order to avoid overfitting, the N-fold cross-validation method is used to train the base learners, and N test results y1, y2,..., yNwill be generated after training. N The N prediction results are stacked in a row vertically to form a prediction set y i,j(i = 1, 2, ..., N; j = 1, 2, ..., m), this prediction set contains the predicted values ​​of all samples at each validation. All base learners are trained in the same way, and then the prediction sets obtained by each base learner are concatenated column-wise to obtain the input dataset y for the next level, which can be represented as: y = [y1, y2, ..., ym]. m The meta-learner is used to combine the prediction results of the base learner in the first layer for further learning. The predicted values ​​in the base learner are used as new features and input into the second meta-learner to finally train the target prediction model.

[0078] In some embodiments, please refer to Figure 2 The aforementioned analysis of third-party electricity consumption data based on an ensemble learning model to predict the SOH information of the battery to be predicted, specifically S180 may include the following steps:

[0079] S410. Filter out outliers from the third electricity consumption data and smooth the data based on the Savitzky-Golay filter to obtain the fourth electricity consumption data.

[0080] S420. Determine the corresponding second IC curve based on the fourth power consumption data information, and calculate the peak characteristics of the second IC curve;

[0081] S430. Determine the peak characteristics of the second IC curve and the fourth power consumption data information as the target power consumption information;

[0082] S440. Input the target power consumption information into the ensemble learning model for analysis to obtain the battery SOH information corresponding to the battery to be predicted.

[0083] In the embodiments of this application, it is understood that, in the process of analyzing the third power consumption data information of the battery to be predicted, in order to delete obviously erroneous data and reduce noise and fluctuations in the data, this application performs outlier screening and smoothing processing on the data, and extracts the peak features of the second IC curve of the battery to be predicted based on the voltage, current and charge / discharge time data included in the fourth power consumption data information, so as to improve the accuracy of the prediction of the ensemble learning model.

[0084] In one example, this ensemble learning-based battery SOH prediction method also includes: using the mean absolute error (MAE), mean squared error (MSE), and R... 2 To evaluate the performance of ensemble learning models. It is understandable that R... 2 The closer the value is to 1, the higher the accuracy of the ensemble learning model; when the ensemble learning model has a large error in its prediction, the mean squared error (MSE) will increase rapidly.

[0085] Exemplarily, the mean absolute error MAE satisfies the expression:

[0086]

[0087] The mean square error MSE satisfies the expression:

[0088]

[0089] R 2 satisfies the expression:

[0090]

[0091] In the expression, y represents the true value of the battery SOH, y i represents the estimated value of the battery SOH, N represents the size of the data, represents the average value of the true value of the battery SOH.

[0092] In some embodiments, to verify the effectiveness of the foregoing battery SOH prediction method based on ensemble learning, taking SOH prediction of a lithium electronic battery as an example, the application collects state data of B0005, B0006 and B0007 lithium ion batteries in the NASA lithium battery dataset, that is, obtains first power consumption data information and first SOH information of the B0005, B0006 and B0007 lithium ion batteries, and the B0005, B0006 and B0007 lithium ion batteries are all target batteries; further, four machine learning methods such as LSTM, LSTM with attention mechanism, BP neural network and random forest algorithm are selected as base learners, and a random forest algorithm is selected as a meta-learner to construct an initial prediction model; the target data set corresponding to the B0005 lithium battery is taken as a training set, and the target data set corresponding to the B0006 and B0007 lithium batteries is taken as a test set for experiment.

[0093] In the model training process, each base learner trains the original data training set respectively. Taking LSTM as an example, to avoid overfitting, a five-fold cross-validation method can be used to train the base learner, and after training is completed, five test results y1, y2, y3, y4 and y5 will be generated, and the five prediction results are stacked vertically to form a prediction set y i,LSTM(i = 1, 2, …, 5), and this prediction set contains the prediction value of each sample at each time of validation. The remaining base learners are trained in the same way, and then the prediction set obtained by each base learner is spliced by column to obtain the input data set of the next level; further, the prediction value in the base learner is input into the second layer meta-learner as a new feature; after the target prediction model is trained, the target data set corresponding to the B0006 and B0007 lithium batteries is input into the target prediction model to obtain the SOH prediction information corresponding to the B0006 and B0007 lithium batteries, and the SOH prediction information is compared with the first SOH information actually measured by the B0006 and B0007 lithium batteries, and the deviation is determined by comparison. The target prediction model can be used as a mature integrated learning model for SOH prediction.

[0094] To further verify the effectiveness of the integrated learning model proposed in this paper, under the condition of ensuring that other experimental parameters are the same, the mean square error MSE, the mean absolute error MAE and R 2 are used as evaluation indexes of estimation results. As shown in Table 1, the lithium battery state of health estimation using the integrated learning model proposed in this application; R 2 , MSE and MAE are all less than those of a single learning model, which proves the superiority of the model in the state of health estimation of energy storage batteries. The quantitative results obtained are shown in Table 1. The experimental results show that the MSE result of the integrated learning model for predicting SOH is 0.007, the MAE result is 0.055, and the R 2 result is 99.99%. The prediction result shows that the integrated learning model proposed in this application has the highest prediction accuracy and has a significant advantage over other models, which proves that the model also has certain superiority in the state of health prediction of cross-type energy storage batteries.

[0095] Table 1 SOH prediction error of multi-class energy storage batteries

[0096]

[0097]

[0098] In some embodiments, referring to Figure 3 , the application also provides a battery SOH prediction system 500 based on integrated learning, which can specifically include:

[0099] The acquisition module 510 is configured to acquire the first power consumption data information and the first SOH information of each target battery in the target battery pack.

[0100] The processing module 520 is configured to perform outlier screening on the first power consumption data information, perform data smoothing processing based on a preset signal processor, and obtain second power consumption data information; the second power consumption data information includes voltage, current, and charging and discharging time of the target battery.

[0101] The calculation module 530 is configured to determine a corresponding first IC curve based on the second power consumption data information, and calculate a peak feature of the first IC curve.

[0102] The determination module 540 is configured to determine that the peak feature of the first IC curve, the first SOH information, and the second power consumption data information are the target data set.

[0103] The construction module 550 is configured to select a base learner and a meta learner from a plurality of machine learning algorithms and a plurality of statistical models, and construct an initial prediction model.

[0104] The training and testing module 560 is configured to train and test the initial prediction model based on a plurality of target data sets corresponding to a plurality of target batteries in the target battery group, and obtain an ensemble learning model.

[0105] The acquisition module 510 is further configured to acquire third power consumption data information of a battery to be predicted, the third power consumption data information including voltage, current, and charging and discharging time of the battery to be predicted.

[0106] The prediction module 570 is configured to analyze the third power consumption data information based on the ensemble learning model, to predict battery SOH information corresponding to the battery to be predicted.

[0107] According to embodiments of the present application, any number of modules among the acquisition module 510, the processing module 520, the calculation module 530, the determination module 540, the construction module 550, the training and testing module 560, and the prediction module 570 can be combined in one module, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of the modules can be combined with at least part of the functions of other modules, and implemented in one module.

[0108] In some embodiments, the training and testing module 560 can specifically include:

[0109] The data division unit 561 is configured to divide a plurality of target data sets corresponding to a plurality of target batteries in the target battery group into a training set and a test set.

[0110] The model training unit 562 is configured to input the training set into the initial prediction model for training, to obtain a target prediction model.

[0111] The data input unit 563 is configured to input the test set into the target test model, to obtain SOH prediction information.

[0112] The information comparison unit 564 is configured to compare the SOH prediction information with the first SOH information of the corresponding target battery, and determine a target deviation therebetween.

[0113] The model determination unit 565 is configured to determine that the target prediction model is the ensemble learning model when the target deviation is less than a preset threshold.

[0114] In some embodiments, the model training unit 562 can be specifically configured to:

[0115] input the training set into each base learner in the base learner combination;

[0116] In each base learner, the third power consumption data information is analyzed based on an N-fold cross-validation method to generate N test results;

[0117] The N prediction results of each base learner are stacked in a row vertically to form a prediction set;

[0118] The prediction sets obtained by each base learner in the base learner combination are spliced in a column to obtain an input data set of the next level;

[0119] The input data set is input as new features into the meta-learner of the second layer of the initial prediction model for data fusion training and obtaining a target prediction model.

[0120] In some embodiments, the prediction module 570 can be specifically configured to:

[0121] perform outlier screening on the third power consumption data information, and perform data smoothing processing based on a Savitzky-Golay filter to obtain fourth power consumption data information;

[0122] determine a corresponding second IC curve based on the fourth power consumption data information, and calculate a peak feature of the second IC curve;

[0123] determine the peak feature of the second IC curve and the fourth power consumption data information as target power consumption information;

[0124] input the target power consumption information into the ensemble learning model for analysis to obtain battery SOH information corresponding to the battery to be predicted.

[0125] Figure 3 Each module in the system has the function of implementing each step in the foregoing battery SOH prediction method based on ensemble learning, and can achieve the corresponding technical effects. For brevity, no further description is given here.

[0126] In some embodiments, the present application provides an electronic device, and a structural schematic diagram of the electronic device is shown in Figure 4 .

[0127] The electronic device can include a processor 610 and a memory 620 having stored computer program instructions.

[0128] In particular, the processor 610 described above can include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or can be configured as one or more integrated circuits that implement embodiments of the present application.

[0129] The memory 620 can include a mass storage that is used for data or instructions. By way of example, and not limitation, the memory 620 can include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive or a combination of two or more of these. The memory 620 can include removable or non-removable (or fixed) media, where appropriate. The memory 620 can be internal or external to the integrated gateway disaster recovery device, where appropriate. In particular embodiments, the memory 620 is non-volatile, solid-state memory.

[0130] The memory 620 can include read-only memory (ROM), random-access memory (RAM), a magnetic disk storage medium, an optical storage medium, a flash memory device, an electrical, optical, or other physical / tangible memory storage device. Thus, in general, the memory 620 includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software that, when executed (by one or more processors), is operable to perform the operations described above as being performed by any of the embodiments of the battery SOH prediction method based on ensemble learning.

[0131] The processor 610 implements the battery SOH prediction method based on ensemble learning in any of the embodiments described above by reading and executing computer program instructions stored in the memory 620.

[0132] In one example, the electronic device can further include a communication interface 630 and a bus 600. As shown, the processor 610, the memory 620, and the communication interface 630 are connected through the bus 600 and complete communication with each other. Figure 4

[0133] The communication interface 630 is mainly used to realize the communication between the modules, systems, units and / or devices in the embodiments of the present application.

[0134] ​Bus 600 includes a hardware, software, or both that couples components of the online data traffic metering device to each other. As an example and not by way of limitation, the bus can include an accelerated graphics port (AGP) or other graphics bus, an enhanced industry standard architecture (EISA) bus, a front-side bus (FSB), a HyperTransport (HT) interconnect, an industry standard architecture (ISA) bus, an infiniband interconnect, a low pin count (LPC) bus, a memory bus, a microchannel architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standards board (VLB) bus, or another suitable bus or a combination of two or more of these. Where suitable, bus 600 can include one or more buses. Although this application describes and illustrates a particular bus, this application contemplates any suitable bus or interconnect.

[0135] In addition, in combination with the battery SOH prediction method based on ensemble learning in the above embodiments, an application embodiment can provide a computer storage medium to implement. The computer storage medium has computer program instructions stored thereon; the computer program instructions are executed by a processor to implement any one of the battery SOH prediction methods based on ensemble learning in the above embodiments.

[0136] It needs to be clear that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of well-known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0137] The functional blocks shown in the above structural block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, etc. When implemented in software, the elements of the present application are program or code segments used to perform the required tasks. The program or code segments can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. The "machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segments can be downloaded via a computer network such as the Internet, an intranet, etc.

[0138] It should also be noted that the example embodiments mentioned in the present application are based on a series of steps or systems. However, the present application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or different from the order in the embodiments, or several steps are performed simultaneously.

[0139] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable data processing apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0140] In summary, compared with the prior art, the present application has the following beneficial effects:

[0141] 1、The present application selects a plurality of base learners from a plurality of machine learning algorithms, and selects a meta-learner from a plurality of statistical models, and then constructs a double-layer initial prediction model. The plurality of base learners can exert their respective advantages to improve the accuracy of data analysis, and the meta-learner intelligently integrates these results to avoid the bias and limitations that may exist in a single model. By fusing a plurality of base learners and meta-learners, the model can have high generalization ability and can flexibly adapt to various application scenarios and data sets. This complementary integrated approach makes the overall prediction result closer to the true situation, providing reliable data support for the decision-making of the battery management system.

[0142] 2、The present application extracts the first IC curve corresponding to the target battery pack on the basis of the second power consumption data information, and determines the peak feature of the first IC curve as part of the target data set. That is, the present application extracts the incremental capacity curve peak value of the lithium battery as a key feature on the basis of some measurable data, thereby improving the learning efficiency and prediction performance of the model.

[0143] 3、By training multiple base learners and introducing multiple types of data samples during the training process, the generalization ability of the model is improved, enabling it to learn more extensive and comprehensive battery state of health change patterns. When faced with new and unknown scenarios, the ensemble learning model can rely on its strong generalization ability to quickly adapt and make accurate predictions. In addition, the dynamic adjustment and optimization mechanism of the meta-learner further enhances the adaptability and stability of the model in complex and variable environments.

[0144] 4、The application can improve the accuracy and generalization of battery state of health prediction, provide more scientific decision basis for the maintenance and management of energy systems, timely detect and prevent potential safety hazards, reduce system failure risk, and ensure long-term stable operation of the system.

[0145] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A battery SOH prediction method based on ensemble learning, characterized by, The method comprises the following steps: acquiring first power consumption data information and first SOH information of each target battery in a target battery pack; performing outlier screening on the first power consumption data information, and performing data smoothing processing based on a preset signal processor to obtain second power consumption data information; wherein the second power consumption data information comprises voltage, current and charging / discharging time of the target battery; determining a corresponding first IC curve based on the second power consumption data information, and calculating a peak feature of the first IC curve; determining the peak feature of the first IC curve, the first SOH information and the second power consumption data information as a target data set; selecting a base learner and a meta learner from a plurality of machine learning algorithms and a plurality of statistical models to construct an initial prediction model; training and testing the initial prediction model based on a plurality of target data sets corresponding to a plurality of target batteries in the target battery pack to obtain an ensemble learning model; acquiring third power consumption data information of a battery to be predicted, the third power consumption data information comprising voltage, current and charging / discharging time of the battery to be predicted; analyzing the third power consumption data information based on the ensemble learning model to predict battery SOH information corresponding to the battery to be predicted. 2.The battery SOH prediction method based on ensemble learning of claim 1, wherein, The preset signal processor is a Savitzky-Golay filter; the peak feature of the first IC curve is determined by a preset difference calculation method, and the difference calculation method satisfies the expression: In the formula, t0 is the time at which discharge begins, t i is the time point of the i-th cycle, Q i is the cumulative discharge capacity of the i-th cycle; ic i represents the incremental capacity value at the i-th time point, Q i+1 is the capacity corresponding to the i+1-th data point, V i+1 is the voltage corresponding to the i+1-th data point, V i is the voltage corresponding to the i-th data point, AQ i is the capacity difference of adjacent data points, AV i is the voltage difference of adjacent data points. 3.The battery SOH prediction method based on ensemble learning according to claim 1 or 2, wherein, The initial prediction model comprises two layers, the first layer is the base learner combination, and the second layer is the meta learner; the plurality of machine learning algorithms comprise LSTM, RNN, BiLSTM, BP neural network and random forest algorithm, and the meta learner is one of random forest algorithm and Bayesian algorithm. 4.The battery SOH prediction method based on ensemble learning of claim 3, wherein, The training and testing of the initial prediction model based on a plurality of target data sets corresponding to a plurality of target batteries in the target battery pack to obtain an ensemble learning model comprises: dividing the plurality of target data sets corresponding to the plurality of target batteries in the target battery pack into a training set and a test set; inputting the training set into the initial prediction model for training to obtain a target prediction model; inputting the test set into the target prediction model to obtain SOH prediction information; comparing the SOH prediction information with the first SOH information of the corresponding target battery to determine the target deviation therebetween; in the case where the target deviation is less than a preset threshold, determining that the target prediction model is an ensemble learning model. 5.The battery SOH prediction method based on ensemble learning of claim 4, wherein, The training of the initial prediction model by inputting the training set into the initial prediction model to obtain a target prediction model comprises: inputting the training set into each base learner in the base learner combination; in each base learner, analyzing the third power consumption data information based on an N-fold cross-validation method to generate N test results; stacking the N prediction results of each base learner in a row vertically to form a prediction set; splicing the prediction sets obtained by each base learner in the base learner combination in a column to obtain an input data set of the next level; Input the input data set as a new feature into a meta-learner of a second layer of the initial prediction model for data fusion training and to obtain a target prediction model. 6.The battery SOH prediction method based on ensemble learning according to claim 1 or 2, wherein, The third power consumption data information is analyzed based on the ensemble learning model to predict the battery SOH information corresponding to the to-be-predicted battery. The third power consumption data information is subjected to outlier screening and data smoothing processing based on a Savitzky-Golay filter to obtain fourth power consumption data information. The second IC curve corresponding to the fourth power consumption data information is determined, and the peak feature of the second IC curve is calculated. The peak feature of the second IC curve and the fourth power consumption data information are determined as target power consumption information. The target power consumption information is input into the ensemble learning model for analysis to obtain the battery SOH information corresponding to the to-be-predicted battery. 7.The battery SOH prediction method based on ensemble learning according to claim 1 or 2, wherein, Further comprising: The performance of the ensemble learning model is evaluated by mean absolute error MAE, mean square error MSE and R 2 2 The closer to 1, the higher the accuracy of the ensemble learning model; when the ensemble learning model predicts a large error, the mean square error MSE will increase rapidly.​ The mean absolute error MAE satisfies the expression: The mean square error MSE satisfies the expression: R 2 satisfies the expression: In the formula, y represents a true value of the battery SOH, y i represents an estimated value of the battery SOH, N represents a size of data, represents an average value of the true value of the battery SOH.

8. A battery SOH prediction system based on ensemble learning, characterized by, Including: An acquisition module is configured to acquire first power consumption data information and first SOH information of each target battery in a target battery pack. A processing module is configured to perform outlier screening on the first power consumption data information and perform data smoothing processing based on a preset signal processor to obtain second power consumption data information, wherein the second power consumption data information includes voltage, current, and charge / discharge time of the target battery. A calculation module is configured to determine a first IC curve corresponding to the second power consumption data information and calculate a peak feature of the first IC curve. A determination module is configured to determine the peak feature of the first IC curve, the first SOH information, and the second power consumption data information as a target data set. A construction module is configured to select a base learner and a meta-learner from a plurality of machine learning algorithms and a plurality of statistical models to construct an initial prediction model. A training and testing module is configured to train and test the initial prediction model based on a plurality of target data sets corresponding to a plurality of target batteries in the target battery pack to obtain an ensemble learning model. The acquisition module is further configured to acquire third power consumption data information of a to-be-predicted battery, wherein the third power consumption data information includes voltage, current, and charge / discharge time of the to-be-predicted battery. A prediction module is configured to analyze the third power consumption data information based on the ensemble learning model to predict battery SOH information corresponding to the to-be-predicted battery.

9. An electronic device, comprising: Including: A processor, a memory, and a program stored on the memory and executable on the processor, wherein the program is executed by the processor to implement the battery SOH prediction method based on ensemble learning according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores programs or instructions, which are executed by the processor to implement the battery SOH prediction method based on ensemble learning according to any one of claims 1 to 7.

Images