SVR intelligent vehicle power battery remaining life prediction method based on multi-source information

Abstract

The invention relates to the technical field of energy storage, discloses an SVR intelligent vehicle power battery remaining life prediction method based on multi-source information, and aims to solve the problem of low prediction precision caused by single input of an existing data driving method. The method comprises the following steps: collecting multi-source information of the running state of the power battery, wherein the multi-source information comprises performance degradation characteristics and running historical characteristics; features with high correlation with the battery capacity retention rate are screened through Pearson correlation analysis, and a target input feature vector is constructed; training a support vector regression model based on the screening features, and processing a nonlinear degradation relationship by adopting a radial basis function kernel; and predicting and calculating the remaining service life by using the training model. According to the method, state information and history information are fused, intelligent feature screening is assisted to reduce noise interference, more comprehensive data dimensions are provided for a prediction model, prediction accuracy, robustness and generalization ability are remarkably improved, and the method has outstanding substantive features and industrial application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage, in particular to a SVR intelligent vehicle power battery residual life prediction method based on multi-source information. BACKGROUND

[0002] The power battery is the energy storage core of modern intelligent vehicles such as electric vehicles and hybrid electric vehicles, and its performance, health state and residual useful life (RUL) are directly related to the vehicle's cruising range, running safety, reliability and overall cost of ownership. In the whole life cycle of the battery, due to the influence of complex internal electrochemical reactions (such as electrolyte decomposition and active material loss) and external working conditions (such as temperature and charge-discharge rate), its capacity will gradually decay. When the battery capacity decays to a certain percentage (usually 80%) of its rated capacity, it is considered that its life ends. Therefore, being able to accurately predict the RUL is of great economic and social value for achieving proactive and predictive maintenance of the battery, avoiding safety accidents caused by sudden failure of the battery during vehicle operation, and optimizing the step utilization and replacement strategy of the battery.

[0003] Currently, the prediction methods for the RUL of the power battery can be mainly divided into two major technical routes: model-based methods and data-driven methods.

[0004] The model-based method is committed to describing the internal physical and chemical processes of the battery or its external electrical characteristics through mathematical equations. This kind of method can be further divided into empirical models, electrochemical models and equivalent circuit models, etc. For example, the empirical model uses a single exponential or double exponential function to fit the capacity decay curve; the electrochemical model tries to establish a complex partial differential equation set to simulate the migration and embedding process of lithium ions in the electrode. In addition, Kalman filter and its variants (such as extended Kalman filter) are also used to dynamically update the model state according to the measured data, so as to predict the life. The advantage of this kind of method is that the model has a clear physical or chemical meaning, which helps to understand the degradation mechanism of the battery. However, its disadvantages are also very significant: the model construction process is extremely complex, which requires a deep understanding of the internal mechanism of the battery; the identification and calibration of model parameters are difficult, and the universality for different models and different batches of batteries is poor; especially it is difficult to accurately capture the complex nonlinear degradation behavior caused by the coupling of multiple factors under varying actual working conditions.

[0005] In contrast, data-driven approaches have gained wide attention in recent years. Such approaches treat the battery system as a “black box” without delving into its internal complex reaction mechanisms, but instead learn the evolution law and degradation pattern from the healthy state to the failure state directly from a large amount of historical operation data. Common algorithms include various neural networks, such as back propagation networks, recurrent neural networks and their improved long short-term memory networks, and the application of support vector machines in their regression problems - support vector regression, etc. These methods eliminate complex mechanism modeling and have strong fitting ability for nonlinear relationships.

[0006] However, the existing data-driven methods, especially the traditional methods as described in the original document, have a fundamental technical defect: the singularity of the input data source. Existing technologies usually only rely on single-dimensional feature data, i.e., mainly using “symptom” information reflecting the current performance state of the battery, such as measurable performance degradation features such as charging time, end voltage, current, temperature curve during charging and discharging, etc. This approach ignores key historical information that has a decisive cumulative impact on battery life, especially the number of discharge cycles.

[0007] This singularity of data source leads to two levels of problems. First, the model fails to capture the complete causal chain of degradation. Performance degradation features (such as accelerated voltage drop) are the result or manifestation of battery aging, while cycle count is an important cause or process leading to aging. A model that only sees the “result” and ignores the “cause” will inevitably be limited in its ability to extrapolate the future. Second, the model lacks context awareness of the life stage of the battery. The degradation process of the battery is not linear, and its sensitivity to the same working conditions is different at different life stages. For example, a traditional model based only on temperature and voltage may have difficulty distinguishing between “a healthy battery experiencing a high-load discharge” and “an aging battery undergoing a normal discharge” because they may exhibit similar instantaneous temperature and voltage characteristics. If the model could also know that the former has a very low cycle count while the latter has a very high cycle count, it could more accurately assess the current state and predict future trends. Therefore, the lack of historical cumulative information such as cycle count makes the model's interpretation of performance degradation features ambiguous, limiting the accuracy and robustness of the prediction.

[0008] Based on the above problems, the present application provides a SVR intelligent vehicle power battery residual life prediction method based on multi-source information, which fuses and filters performance degradation data (discharge cut-off voltage time, discharge cut-off voltage temperature) and discharge cycle count data, constructs an SVR model based on multi-source information, and through experimental simulation, compares and analyzes the SVR method based on multi-source information and the traditional SVR method, verifies the effectiveness of the proposed method, and realizes the accurate prediction of the residual life of the power battery. SUMMARY

[0009] The present application aims to at least solve one of the technical problems existing in the related art. To this end, the present application provides a SVR intelligent vehicle power battery residual life prediction method based on multi-source information.

[0010] A SVR intelligent vehicle power battery residual life prediction method based on multi-source information, the steps comprising: S1 information acquisition: acquiring multi-source information recording the running state of the power battery, the multi-source information being divided into two categories: performance degradation characteristic data reflecting the instantaneous performance state of the battery, and running history characteristic data reflecting the cumulative use history of the battery; S2 feature selection: using a statistical correlation analysis method, calculating the correlation of each feature in the performance degradation characteristic data and the running history characteristic data with the target output as the battery health state indicator, and selecting features with a correlation intensity higher than a first threshold value according to the calculation results to obtain target input features; S3 model training: based on the target input features, constructing and training an SVR model of support vector regression; S4 life prediction step: using the SVR model, inputting the target input features acquired in real time, and predicting the residual service life of the power battery.

[0011] Further, the performance degradation characteristic data includes at least one of the following: discharge cutoff voltage time, discharge cutoff voltage temperature, constant current charging time, and internal resistance increment.

[0012] Further, the running history characteristic data is the number of discharge cycles.

[0013] Further, the preset target output as the battery health state indicator is the battery capacity retention rate, which is the ratio of the current actual capacity of the battery to its rated capacity.

[0014] Further, the statistical correlation analysis method is a Pearson correlation analysis method. The first threshold value is a value greater than the absolute value |r| of the Pearson correlation coefficient.

[0015] Further, the target input features are discharge cutoff voltage time and discharge cycle number.

[0016] Further, the SVR model uses a radial basis function as a kernel function.

[0017] An apparatus for performing a SVR intelligent vehicle power battery residual life prediction method based on multi-source information, comprising: one or more processors; and A memory having instructions stored thereon that are executable by the one or more processors, the instructions, when executed, causing the one or more processors to perform the method for predicting the remaining useful life of the SVR intelligent vehicle power battery based on multi-source information according to any one of claims 1 to 7.

[0018] A battery management system for performing the method for predicting the remaining useful life of the SVR intelligent vehicle power battery based on multi-source information. An apparatus comprising performing the method for predicting the remaining useful life of the SVR intelligent vehicle power battery based on multi-source information.

[0019] A computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a processor of a computer, implements the method for predicting the remaining useful life of the SVR intelligent vehicle power battery based on multi-source information.

[0020] The one or more technical solutions in the embodiments of the present application have at least one of the following technical effects: The prediction accuracy is improved, by fusing the performance degradation data reflecting the instantaneous state (such as the discharge cut-off voltage time) and the operation history data reflecting the cumulative damage (such as the discharge cycle number), the present application provides more comprehensive and rich input information for the prediction model. This enables the model to more deeply understand the nonlinear degradation behavior of the battery from the two dimensions of "state" and "history", thereby significantly improving the prediction accuracy of RUL. Experimental results show that, compared with the traditional SVR method using only a single data source, the prediction absolute error of the present application method is greatly reduced from 108.5160% to 89.0407%, and the performance is improved by 17.95%.

[0021] The robustness and generalization ability of the model are enhanced, by introducing statistical correlation analysis for intelligent feature screening, the present application can automatically and objectively eliminate the features with weak correlation or redundancy with RUL from the multi-source data, reduce the noise interference, and avoid the misleading of irrelevant features to the model. This not only simplifies the model structure and reduces the risk of overfitting, but also enables the model to better adapt to different battery types and working conditions, enhancing the robustness and generalization ability of the model.

[0022] The prediction method proposed by the present application has strong industrial application value and implementability.

[0023] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort based on these drawings.

[0025] Figure 1 is a comparison chart of the predicted value and the actual value of the SVR method based on multi-source information in the present application; Figure 2 is a prediction error curve of the SVR method based on multi-source information in the present application; Figure 3 is a comparison chart of the predicted value and the actual value of the traditional SVR method in the present application; Figure 4 is a prediction error curve of the traditional SVR method in the present application. DETAILED DESCRIPTION

[0026] In order to make the objects, technical solutions and advantages of the present application clearer, the following will combine the drawings in the present application to clearly and completely describe the technical solutions in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort belong to the protection scope of the present application. The following embodiments are used to illustrate the present application, but cannot be used to limit the scope of the present application.

[0027] The following will combine Figures 1 to 4 to describe the technical solutions of the present application.

[0028] The present application aims to solve the technical defects of the existing power battery RUL prediction method. Specifically, the core technical problem to be solved by the present application is that the input data source of the existing prediction model (especially the data-driven model) is single, and the features reflecting the current performance state of the battery and the historical operation features reflecting the cumulative damage degree of the battery cannot be effectively fused, so that the model cannot fully understand the complex degradation mechanism of the battery, and finally the problem of low RUL prediction accuracy and insufficient model generalization ability is caused.

[0029] The present application provides a SVR intelligent vehicle power battery remaining useful life prediction method based on multi-source information, which comprises the following steps: Multi-source information acquisition: acquiring multi-source information recording the running state of the power battery, wherein the multi-source information is logically divided into at least two categories: one category is performance degradation feature data reflecting the instantaneous performance state of the battery, and the other category is operation history feature data reflecting the cumulative use history of the battery.

[0030] Feature screening: using a preset statistical correlation analysis method, the correlation between each feature in the multi-source information and the battery capacity retention rate as the battery health state indicator is calculated, and the target input feature with a correlation intensity higher than a preset threshold is selected according to the calculation result.

[0031] Model training: based on the target input feature, a support vector regression (SVR) model is constructed and trained.

[0032] Life prediction: using the trained SVR model, the target input feature obtained in real time is input to predict the remaining useful life of the power battery.

[0033] The content of the acquisition of multi-source information includes: Feature engineering is the cornerstone of data-driven modeling, and its purpose is to extract the most valuable information for the prediction target from the original measured data. One of the core innovations of the present application is to construct a multi-source, multi-dimensional feature set. In the specific implementation of the present application, the acquired multi-source information is organized into the following types of feature vectors: Performance degradation feature vector ): This type of feature directly reflects the health state and electrochemical performance of the battery at the current time. They are the external "symptoms" of battery aging. In a preferred embodiment, this vector can include but is not limited to the following features: Discharge cut-off voltage time ): refers to the time experienced by the battery from the full charge state (such as the charge cut-off voltage 3.65 V) to the discharge cut-off voltage specified by the manufacturer (such as 2.5 V) under standard discharge conditions (for example, 1C rate discharge). As the battery ages, its actual available capacity decreases, so will gradually shorten.

[0034] Discharge cut-off voltage temperature ): refers to the surface or internal temperature of the battery at the moment of discharging to the cut-off voltage. Battery aging is usually accompanied by an increase in internal resistance, according to Joule's law , under the same discharge current , the increase in internal resistance will lead to an increase in heat generation, so may show some trend with aging.

[0035] Other scalable features: in order to further enhance the expression ability of the model, other performance indicators such as the time of the constant current charging phase, the temperature rise rate during charging, the internal resistance increment measured by a specific frequency alternating excitation, etc. can also be included.

[0036] Running history feature vector ): This feature records the cumulative workload or "age" of the battery since it was put into use, which is one of the fundamental reasons for performance degradation. In the core embodiment of the present invention, this vector mainly contains: Number of discharge cycles ): This refers to the number of equivalent full charge-discharge cycles that the battery has experienced. Each charge-discharge cycle causes irreversible damage to the internal materials of the battery, so is the most intuitive and important indicator of measuring the cumulative damage of the battery.

[0037] Target output vector ): This is the target that the prediction model needs to learn and fit. In this invention, the target output is defined as a direct quantitative indicator of the State of Health (SOH) of the battery: Battery capacity retention rate ): It is defined as the ratio of the current actual capacity of the battery to the rated capacity at the factory, that is . is the total charge that the battery can discharge under the current cycle, while is a fixed value (for example, 20 Ah for the battery in this embodiment). The value of starts from 100% (brand new battery) and decreases with use. When drops to a pre-set failure threshold (for example, the industry standard of 80%), it is declared that the battery has reached the end of its life. Therefore, by accurately predicting the future trend of , we can infer the remaining cycle number when it reaches the 80% threshold, i.e. RUL.

[0038] Feature selection is based on Pearson correlation analysis: After collecting rich raw data and building a multi-dimensional feature set, a key problem arises - not all features contribute equally to predicting RUL. Some features may have weak correlation with RUL and are noise; while others may have high correlation, i.e. information redundancy. Directly inputting these ineffective or redundant features into the machine learning model not only unnecessarily increases the computational complexity of the model and prolongs the training time, but also may introduce noise and mislead the learning process of the model, leading to overfitting and reducing its generalization ability on unseen data.

[0039] To solve this problem, the present invention introduces an intelligent feature selection step, which aims to automatically select the candidate features from the pool in an objective and quantitative way, and to predict the target (i.e. battery capacity retention rate the most relevant feature subset. In the preferred embodiment of the present invention, Pearson correlation analysis method is adopted to achieve this goal.

[0040] Pearson Correlation Coefficient, usually denoted by the symbol , is a statistical measure used to quantify the degree of linear association between two continuous variables. Its calculated result is strictly limited in the range of [-1, 1], which makes the correlation strength between different features directly comparable.

[0041] When approaches 1, it indicates that there is a strong positive linear relationship between the two variables (i.e., as one variable increases, the other also increases).

[0042] When approaches -1, it indicates that there is a strong negative linear relationship between the two variables (i.e., as one variable increases, the other decreases).

[0043] When approaches 0, it indicates that there is almost no linear relationship between the two variables.

[0044] The strength of correlation is determined by the absolute value of r, the larger r is, the stronger the linear relationship is.

[0045] The specific implementation steps of feature selection in the present invention are as follows: Calculate the correlation coefficient: for each candidate feature in the obtained multi-source information feature set, calculate its Pearson correlation coefficient with the target output vector using the above formula respectively .

[0046] Set the selection threshold: set a correlation strength threshold . The selection of this threshold can be based on experience or general statistical guidelines, for example, according to some statistical guidelines, can be considered to have strong correlation. Therefore, in this embodiment, .

[0047] Perform feature selection: traverse all calculated correlation coefficients and select all features that meet the condition . These selected features collectively constitute the final "target input feature" vector used for SVR model training.

[0048] The Pearson correlation coefficients between the "discharge cut-off voltage time", "discharge cut-off voltage temperature" and "discharge cycle number" and the "battery capacity retention rate" are calculated respectively. The calculation results show that the |r| value of the "discharge cut-off voltage time" and the "discharge cycle number" is significantly higher than the threshold, while the |r| value of the "discharge cut-off voltage temperature" is lower. Therefore, the final selected target input features are "discharge cut-off voltage time" and "discharge cycle number", and the noise feature "discharge cut-off voltage temperature" is successfully excluded. This process fully proves the necessity and effectiveness of this screening step.

[0049] It is worth noting that the Pearson correlation coefficient mainly measures the linear relationship. However, the degradation process of the battery is highly nonlinear in nature. Although the Pearson analysis achieved good results in the specific experiments of the present application, this may be due to the fact that the selected features and capacity decay exhibit a strong monotonic trend macroscopically. In order to make the protection scope of the present application more forward-looking and robust, it must be recognized that the core point of creation of the present application lies in "using statistical correlation analysis to guide the selection of multi-source fusion features", and not just "using the Pearson correlation coefficient". Therefore, the Pearson analysis should be regarded as a preferred, specific implementation. In other possible cases, if there is a clear nonlinear but monotonic relationship between the features and the target, other statistical correlation analysis methods such as Spearman's rank correlation coefficient can be used.

[0050] After obtaining high-quality target input features through intelligent feature screening, a machine learning model is selected and constructed to learn the complex mapping relationship between these features and the battery RUL. The present application selects Support Vector Regression (SVR) as the core prediction algorithm. SVR is an extended application of the famous Support Vector Machine (SVM) algorithm in regression problems, which inherits many advantages of SVM in handling small samples, nonlinearity and high dimensionality.

[0051] (I) SVR model construction based on multi-source information Assume that the sample set is wherein, is the performance degradation feature vector, is the discharge cycle number feature vector, is the corresponding output vector. The following nonlinear mapping is performed to map the sample set from a low-dimensional space to a high-dimensional space: (1) wherein, is the input feature data, is the weight, is the intercept. A slack variable is added and (n is the number of samples), as shown in equation (2): (2) where, is a penalty factor, is the maximum allowable error value. By performing Lagrange transformation on equation (2), we obtain, (3) where, and denote the Lagrange multipliers, and by maximizing the Lagrange function, we obtain the nonlinear mapping SVR expression as: (4) where, and are two feature vectors and corresponding kernel functions, respectively.

[0052] The key parameters in the above equation are further described as follows: and are the weight vector and bias term of the regression hyperplane, respectively, which together define the prediction function .

[0053] is a nonlinear mapping function that is responsible for mapping the original input feature vector from a low-dimensional input space to a higher-dimensional or even infinite-dimensional feature space (Hilbert space). The purpose is to hope that in the new high-dimensional space, the complex nonlinear relationship of the original data can be transformed into a linear relationship.

[0054] The first term of the optimization objective is to minimize the L2 norm of the weight vector to achieve the "flatness" of the model.

[0055] is a pre-set hyperparameter that defines the radius of the insensitive band. The existence ofmakes the SVR insensitive to part of the noise in the data, enhancing the robustness of the model.

[0056] and are slack variables used to handle those outlier points that cannot satisfy the- pipeline insensitive band constraint. They respectively measure the excess amount of the sample point's true value above and below the - pipeline.

[0057] is a very important hyperparameter, called the penalty factor or regularization constant. It is used to trade-off between "minimizing model complexity" (i.e. ) and "minimizing training error" (i.e. ). A very large value implies a huge penalty for errors that are outside of - the tube, and the model will try to fit every data point, which can result in overfitting. Conversely, a very small value allows the model to tolerate larger training errors, resulting in a smoother, simpler regression function, but this can result in underfitting.

[0058] The introduction of the Radial Basis Function (RBF) kernel makes it very difficult, if not impossible, to compute directly in the high-dimensional space. This problem is solved in SVR by introducing the "kernel trick". The kernel trick allows us to compute the inner product of samples in that high-dimensional space directly without explicitly defining the mapping function . This computation is done by a function called the kernel function .

[0059] Considering the strong nonlinearity of the battery degradation process, the present application preferably adopts a Radial Basis Function (RBF) as the kernel function. The RBF kernel, also known as the Gaussian kernel, is one of the most powerful and commonly used kernel functions for dealing with nonlinear problems. The power of the RBF kernel lies in its ability to implicitly map the original input space to an infinite-dimensional feature space. In this infinite-dimensional space, even the most complex nonlinear relationships in the original space have a high probability of becoming linearly separable (or linearly regressible), thus enabling the linear SVR algorithm to effectively find a suitable regression hyperplane.

[0060] Before feeding the data into the model, normalization or standardization processing is usually required to eliminate the influence of the dimension of different features. Subsequently, the entire data set is randomly divided into a training set and a test set according to a certain proportion. The training set is used to train the model, while the test set is used to finally evaluate the generalization performance of the model. The model training and prediction process can achieve the purpose of the present application using the usual model training process and the prediction method after the end of model training, and thus will not be described here.

[0061] In order to verify the effectiveness and advancement of the technical scheme of the present application, simulation analysis is carried out.

[0062] The experimental platform mainly consists of a power battery charging and discharging device, a power battery to be tested, a thermostat for controlling the environmental temperature, and a computer for data recording and model operation. The experimental object is a lithium iron phosphate (LFP) single battery of PC20F-L type, and its detailed technical specifications are crucial for ensuring the reproducibility of the experiment. The specific parameters are shown in the following table.

[0063] Table 1: Technical specifications of power battery

[0064] Through continuous charging and discharging cycle experiments on the battery, a total of 3115 data samples were collected. Each sample contains the discharge cutoff voltage time, discharge cutoff voltage temperature, discharge cycle number, and corresponding battery capacity retention rate. According to the aforementioned method, the data set is divided into the first 2000 groups as training samples and the last 1115 groups as test samples. The failure threshold of battery capacity retention rate is set to 80%.

[0065] In order to highlight the technical effect of the present application, the following two comparative experiments are designed: Example 1 (method of the present application): First, Pearson correlation analysis is performed on the multi-source features (discharge cutoff voltage time, discharge cutoff voltage temperature, discharge cycle number, constant current charging time, and internal resistance increment) to select the "discharge cutoff voltage time" and "discharge cycle number" with the strongest correlation with the capacity retention rate as the model input.

[0066] Among them, the SVR model based on multi-source information is constructed to comprehensively consider the performance degradation data (discharge cutoff voltage time, discharge cutoff voltage temperature) and discharge cycle number data, and based on Pearson correlation analysis, the Pearson correlation coefficient between the discharge cutoff voltage time data and the battery capacity retention rate is , and the Pearson correlation coefficient between the discharge cutoff voltage temperature data and the battery capacity retention rate is It can be seen that , Therefore, the discharge cutoff voltage time and the discharge cycle number are selected as the input of the prediction model to predict the remaining life of the power battery.

[0067] Model training: using the 2000 training samples (containing two input features) to train an SVR model with RBF kernel. The input features include two input features of discharge cutoff voltage time and discharge cycle number.

[0068] Performance evaluation: using the 1115 test samples to verify the trained model. The prediction results are as follows Figure 1As shown, the prediction curve is highly consistent with the actual value curve. The prediction absolute error sum is 89.0407%, and the prediction error curve is as shown in Figure 2

[0069] Comparative Example 1 (traditional SVR method): The prior art is simulated by using only a single data source, i.e., only the "discharge cut-off voltage time" is extracted as the model input feature.

[0070] Model training: Similarly, an SVR model with an RBF kernel is trained using 2000 training samples. Only one input feature, 2000 sets of discharge cut-off voltage time data and battery capacity retention rate data, are used to train the SVR model.

[0071] Performance evaluation: The same 1115 test samples are used for verification. The prediction curve is as shown in Figure 3 The prediction absolute error sum is 108.5160%, and the prediction error curve is as shown in Figure 4 There is a significant deviation between the prediction curve and the actual value.

[0072] In order to more intuitively show the superiority of the present application, the core performance indicators of the two experiments are compared.

[0073] Table 2: Comparison of prediction model performance

[0074] From the quantitative data in Table 2, it can be clearly seen that the prediction method based on multi-source information fusion proposed in the present application has achieved a significant and non-obvious improvement in prediction accuracy compared to the traditional method which only relies on single performance degradation data. The reduction in prediction absolute error sum is as high as 17.95%, which is a very considerable progress in the field of RUL prediction. This result strongly proves the core technical concept of the present application, i.e., the fusion of performance degradation features reflecting instantaneous state and running features reflecting cumulative history, supplemented by intelligent feature selection, can effectively overcome the defects of the prior art and more accurately capture the complex degradation law of the battery, thereby successfully solving the technical problem and having outstanding substantial features and significant progress.

[0075] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not limiting; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing examples, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solution to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.​

Claims

1. A method for predicting the remaining life of a power battery in an SVR intelligent vehicle based on multi-source information, characterized by the following steps: include: S1 Information Acquisition: Acquire multi-source information that records the operating status of the power battery. The multi-source information is divided into two categories: performance degradation characteristic data reflecting the instantaneous performance status of the battery, and operating history characteristic data reflecting the cumulative usage history of the battery. S2 Feature Filtering: Using statistical correlation analysis, the correlation between each feature in the performance degradation feature data and the historical operation feature data and the target output as a battery health status indicator is calculated, and features with a correlation strength higher than the first threshold with the target output are filtered out based on the calculation results to obtain the target input features; S3 Model Training: Based on the target input features, construct and train the Support Vector Regression (SVR) model; S4 Lifetime Prediction Step: Using the SVR model, input the target input features acquired in real time to predict the remaining lifetime of the power battery.

2. The method for predicting the remaining lifespan of an SVR intelligent vehicle power battery based on multi-source information according to claim 1, characterized in that, The performance degradation characteristic data includes at least one of the following: discharge cutoff voltage time, discharge cutoff voltage temperature, constant current charging time, and internal resistance increment.

3. The method for predicting the remaining lifespan of an SVR intelligent vehicle power battery based on multi-source information according to claim 1, characterized in that, The historical operational data is the number of discharge cycles.

4. The method for predicting the remaining lifespan of an SVR intelligent vehicle power battery based on multi-source information according to claim 1, characterized in that, The preset target output, which serves as an indicator of battery health status, is the battery capacity retention rate, which is the ratio of the battery's current actual capacity to its rated capacity.

5. The method for predicting the remaining lifespan of an SVR intelligent vehicle power battery based on multi-source information according to claim 1, characterized in that, The statistical correlation analysis method is the Pearson correlation analysis method. The first threshold is the absolute value of the Pearson correlation coefficient |r| greater than a certain value.

6. The method for predicting the remaining lifespan of an SVR intelligent vehicle power battery based on multi-source information according to claim 1, characterized in that, The target input features are the discharge cutoff voltage time and the number of discharge cycles.

7. The method for predicting the remaining lifespan of an SVR intelligent vehicle power battery based on multi-source information according to claim 1, characterized in that, The SVR model uses radial basis functions as kernel functions.

8. An apparatus for performing a method for predicting the remaining life of a power battery in an SVR intelligent vehicle based on multi-source information, characterized in that, include: One or more processors; as well as A memory storing instructions executable by the one or more processors, the instructions, when executed, causing the one or more processors to perform the SVR intelligent vehicle power battery remaining life prediction method based on multi-source information as described in any one of claims 1 to 7.

9. A battery management system for implementing a multi-source information-based SVR intelligent vehicle power battery remaining life prediction method, characterized in that, The apparatus includes the device for performing the SVR intelligent vehicle power battery remaining life prediction method based on multi-source information as described in claim 8.

10. A computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a computer processor, implementing the method for predicting the remaining life of an SVR intelligent vehicle power battery based on multi-source information as described in any one of claims 1 to 7.

Images