Adaptive fusion prediction method and system for battery state of new energy vehicle

Abstract

The invention provides a battery state adaptive fusion prediction method and system for a new energy vehicle, and belongs to the technical field of industrial equipment life prediction. Firstly, multi-dimensional real-time operation data including voltage, current, temperature, charge and discharge curve characteristics and the like are collected, and preprocessing of data cleaning, time alignment and characteristic extraction is carried out. And according to a prediction task instruction, calling a corresponding grouping algorithm group from the algorithm library to execute prediction, obtaining respective prediction confidence, performing weighted scoring and sorting based on the prediction confidence output by the algorithm and the historical reliability factors read from the storage module, and outputting a prediction result with the highest score. According to the method, the problem that a traditional single algorithm model is insufficient in precision and robustness under non-stationary degradation and complex working conditions of the battery is solved, the prediction precision and effectiveness are remarkably improved, and the system is endowed with self-diagnosis and high robustness in a complex environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial equipment life prediction, in particular to a new energy vehicle battery state adaptive fusion prediction method and system. BACKGROUND

[0002] With the continuous growth of global demand for new energy vehicles, the performance and safety management of lithium-ion batteries as the core power source are crucial. Battery state prediction technology, especially the accurate estimation of battery health state, remaining life, and current power remaining time, is one of the core functions of the battery management system. Accurate prediction is not only the basis for ensuring vehicle driving safety, preventing sudden failures, and prolonging battery life, but also the key technical support for realizing precise range estimation, optimizing charging strategies, guiding user maintenance, and ultimately improving vehicle operation reliability. Therefore, developing a high-precision, robust battery state prediction technology that can reliably operate under actual complex conditions has great engineering significance and economic value.

[0003] Currently, the mainstream battery state prediction systems in the industry mostly use single models or static models for prediction, such as the Kalman filter (EKF) method based on equivalent circuit models, or support vector machines (SVM), recurrent neural networks (RNN), etc. However, these existing technologies have the following obvious defects:

[0004] Insufficient robustness and precision limitations: Single models usually perform well only under specific conditions. Once encountering temperature fluctuations, driving mode differences, and other non-stationary conditions in actual driving, the prediction accuracy will drop sharply, and the complex nonlinearity of the battery degradation process cannot be effectively addressed.

[0005] Lack of closed-loop adaptive ability: Existing systems generally lack an effective and automated prediction result verification and model feedback mechanism. They cannot continuously and dynamically adjust the model or the weight of the prediction according to the aging characteristics of the battery and the actual deviation of historical predictions, leading to model drift as the running time increases, and poor long-term prediction reliability.

[0006] Weak ability to handle abnormal interference conditions: Most systems lack targeted design and cannot accurately identify and effectively intervene according to the historical robustness of the algorithm when facing abnormal interference events (such as long-term storage, collision, or severe overheating), which easily leads to prediction failure or output of false warning information.

[0007] Therefore, there is an urgent need for a dynamic prediction system that can integrate the advantages of multiple algorithms, have a closed-loop adaptive ability, and have an abnormal condition handling mechanism. SUMMARY

[0008] The application aims to provide a new energy vehicle battery state adaptive fusion prediction method and system to solve the problem of insufficient precision and robustness of traditional single algorithm model under non-stationary degradation and complex working conditions of the battery.

[0009] In order to achieve the above-mentioned purpose, the application provides a new energy vehicle battery state adaptive fusion prediction method, which comprises the following steps:

[0010] Step 1: Collecting real-time operation data and historical operation data of the battery, and pre-processing the data;

[0011] Step 2: According to the received prediction task instruction, selecting a corresponding prediction algorithm group from an algorithm library storing a plurality of pre-trained prediction algorithms;

[0012] Step 3: Using a plurality of prediction algorithms in the algorithm group, based on the real-time operation data and the historical operation data, respectively performing prediction to obtain a plurality of prediction results and respective prediction confidence levels;

[0013] Step 4: Reading the respective historical reliability factors of the plurality of prediction algorithms from a storage module, based on the algorithm prediction confidence levels and the historical reliability factors, performing weighted scoring and sorting on the plurality of prediction results, and outputting the prediction result with the highest score as the final prediction result;

[0014] Step 5: After a preset charging and discharging cycle ends, obtaining real state data of the battery as verification data; comparing the verification data with the plurality of prediction results to determine the deviation of each prediction result;

[0015] Step 6: According to the deviation, updating the historical reliability factors corresponding to each algorithm in the storage module.

[0016] A new energy vehicle battery state adaptive fusion prediction system comprises:

[0017] A storage module is used for:

[0018] (a) Storing at least one training data set related to battery data;

[0019] (b) Storing a plurality of pre-trained battery state prediction algorithms, the algorithms including, according to prediction functions: battery state of health (SOH) evaluation, current power prediction remaining time, and battery remaining useful life (RUL) prediction, and are stored in groups;

[0020] (c) Storing historical reliability factors corresponding to each algorithm;

[0021] (d) Storing real-time and historical operation data of the battery, including voltage, current, SOC, and temperature;

[0022] (e) store historical prediction results and periodic verification results;

[0023] (f) store an abnormal interference item database;

[0024] An algorithm management module is connected to the storage module, and is configured to:

[0025] (a) count the number of available files, display system usage history data, and display system usage notes;

[0026] (b) supplement the test data set;

[0027] (c) configure the prediction algorithm;

[0028] A processing module is connected to the storage module and the algorithm management module, and is configured to:

[0029] (a) receive a prediction task instruction and select a corresponding algorithm group according to the instruction;

[0030] (b) call multiple algorithms in the algorithm group to perform prediction tasks respectively;

[0031] (c) according to the preset confidence level, and in combination with the historical reliable factor, weighted scoring and sorting are performed on the prediction results of each algorithm;

[0032] (d) output the prediction result with the highest score;

[0033] (e) after the charge and discharge cycle ends, the real data of the battery is obtained, the prediction result is verified, and the historical reliable factor of the corresponding algorithm in the storage module is updated according to the accuracy of the verification.

[0034] The present application has the following beneficial effects:

[0035] Through the above technical solution, compared with the existing pilot type electromagnetic valve life prediction algorithm test method which is only suitable for a specific algorithm, the present application builds a test system, imports a large number of freely combinable data sets and algorithms, develops online configuration and evaluation functions, and greatly improves the universality and efficiency of the test method. BRIEF DESCRIPTION OF DRAWINGS

[0036] The accompanying drawings are included to provide a further understanding of embodiments of the application, and constitute a part of the specification, and are used together with the following detailed description to explain the application, but do not constitute a limitation on the application. In the drawings:

[0037] Figure 1 : adaptive fusion prediction method flowchart of battery state;

[0038] Figure 2 : more specific flowchart of adaptive fusion prediction method of battery state;

[0039] Figure 3 Figure 1 is a structural diagram of a life prediction algorithm test system. DETAILED DESCRIPTION

[0040] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. In addition, the technical features involved in the various embodiments of the present application described below can be combined with each other as long as they do not conflict with each other. In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions.

[0041] As shown in Figure 1 The present application provides a self-adaptive fusion prediction method for battery state of new energy vehicles, which comprises the following steps:

[0042] The present application provides a self-adaptive fusion prediction method for battery state of new energy vehicles, which comprises the following steps:

[0043] Step 1: Collecting real-time running data and historical running data of the battery, and pre-processing the data;

[0044] Step 2: According to the received prediction task instruction, selecting a corresponding prediction algorithm group from an algorithm library storing a plurality of pre-trained prediction algorithms;

[0045] Step 3: Using a plurality of prediction algorithms in the algorithm group, based on the real-time running data and the historical running data, respectively performing prediction to obtain a plurality of prediction results and respective prediction confidence levels;

[0046] Step 4: Reading the respective historical reliability factors of the plurality of prediction algorithms from the storage module, based on the algorithm prediction confidence level and the historical reliability factors, weighting and ranking the plurality of prediction results, and outputting the prediction result with the highest score as the final prediction result;

[0047] Step 5: After a preset charge and discharge cycle ends, obtaining real state data of the battery as verification data; comparing the verification data with the plurality of prediction results to determine the deviation of each prediction result;

[0048] Step 6: According to the deviation, updating the historical reliability factors corresponding to each algorithm in the storage module.

[0049] Further, in step 1, the real-time operation data is collected from the battery management system (BMS) or cloud data platform, including at least voltage, current, state of charge (SOC), cell temperature, and the number of charge and discharge cycles, as well as historical charging curve data required to support battery health state evaluation, including voltage slope data in the constant current charging phase; historical power output and duration relationship data at different environmental temperatures required to support current power prediction of remaining time to use; and battery operation data in its life cycle and its corresponding real capacity values measured periodically required to support battery remaining life prediction.

[0050] Further, in step 1, the collected real-time operation data and historical operation data are preprocessed, including but not limited to:

[0051] Data cleaning is used to eliminate abnormal data points caused by sensor drift or communication errors;

[0052] Time alignment is used to ensure the synchronization of data collected by different sensors on the time axis; and

[0053] Feature extraction is used to calculate the maximum capacity decay rate of the battery, the internal resistance change rate, etc. as input features for the algorithm.

[0054] Further, in step 2, the prediction task instructions are divided into three types: battery health state evaluation, current power prediction of remaining time to use, and battery remaining life prediction. The algorithms in the algorithm library are divided into three algorithm groups according to the task type, including but not limited to:

[0055] For the algorithm group for battery health state evaluation, the algorithm group includes at least the following two algorithms:

[0056] Diagnostic algorithms based on feature extraction, such as incremental capacity analysis (ICA) or differential voltage analysis (DVA), are used to extract charging curve features to directly estimate the current maximum available capacity; model-based state estimation algorithms, such as extended Kalman filter (EKF) or unscented Kalman filter (UKF), are used to estimate the current battery health state (SOH) based on the real-time changes of equivalent circuit model parameters, specifically using the ratio of the current capacity to the initial state as the SOH value.

[0057] For the algorithm group for current power prediction of remaining time to use, the algorithm group includes at least the following two algorithms:

[0058] Data-driven power prediction algorithms, such as Long Short-Term Memory (LSTM) networks or Gated Recurrent Units (GRUs), are used to predict short-term power demand based on historical driving behavior and temperature data. Lookup table-based energy consumption estimation algorithms are used to combine cell temperature, current SOH value, and historical energy consumption data to output the remaining usable time through a nonlinear mapping relationship.

[0059] The algorithm set for predicting battery remaining life includes at least two of the following algorithms:

[0060] Deep learning-based time-series prediction models, such as LSTM or GRU, are used to learn the long-term degradation sequence of battery capacity; kernel function-based regression models, such as Gaussian process regression (GPR) or support vector regression (SVR), are used to perform nonlinear regression prediction of future capacity.

[0061] Furthermore, obtaining the prediction confidence level in step 3 specifically includes:

[0062] The prediction confidence level is dynamically determined based on the type of prediction algorithm, and the confidence level includes, but is not limited to:

[0063] When the prediction algorithm is a deep learning or regression-based model, the confidence level is determined based on the prediction variance, probability distribution, or error rate of the model on the historical validation set of the model output.

[0064] When the prediction algorithm is a feature extraction-based diagnostic algorithm, the confidence level is determined based on the signal-to-noise ratio of the input data, the clarity of the extracted features, or the goodness of fit between the feature curve and the standard decay curve.

[0065] Furthermore, the historical reliability factor mentioned in step 4 ( The specific definition of ) is:

[0066] The historical reliability factor It is a dynamic index used to quantify the first The prediction algorithm measures the historical prediction accuracy and robustness relative to the actual state of the battery throughout its entire lifespan; the specific calculation formula is as follows:

[0067] In each cycle After completion, based on the actual state value ( ) and predicted state values ​​( ), computational algorithm Instantaneous accuracy within this period :

[0068] ;

[0069] algorithm Historical reliability factors will be updated according to the last cycle factor and the instantaneous accuracy of the current cycle Iterative update:

[0070] ;

[0071] wherein, : algorithm The dynamic forgetting factor (F) of the cycle , the calculation method is:

[0072] ;

[0073] ;

[0074] wherein, is the upper and lower bound of the forgetting factor, is a preset proportional coefficient, used to control the sensitivity of the forgetting factor to error changes.

[0075] Further, the specific method of weighting and ranking the prediction results in step 4 is:

[0076] For the first prediction algorithm in the algorithm group, calculate its final score ; The score is obtained by multiplying the prediction confidence of the algorithm and the historical reliability factor , and its mathematical relationship can be expressed as:

[0077] ;

[0078] The weighting and ranking specifically includes:

[0079] Calculate the score set of all N algorithms in the algorithm group ;

[0080] Sort the score set in descending order;

[0081] Output the prediction result corresponding to the algorithm with the highest score as the final prediction result.

[0082] Further, the method of comparing the deviation of the prediction results in step 5 is specifically divided into the following ways according to the prediction task type:

[0083] For battery state of health evaluation tasks: the deviation ​is the relative error between the predicted current maximum available capacity or SOH value and the verification data;

[0084] For the remaining time prediction task of the current power: the bias is the absolute percentage error between the predicted remaining time or remaining range and the actual observed remaining time or range in the period;

[0085] For the battery remaining life prediction task: the bias is the normalized absolute error between the predicted battery capacity or SOH value at a preset future time point (for example, N future charge and discharge cycles) and the actual measured battery capacity or SOH value at the future time point, and the normalization is based on the nominal capacity of the battery.

[0086] Further, the historical reliability factor updating method in step 6 is divided into two kinds.

[0087] Generally, the factor updating method is as described in step 4;

[0088] When abnormal use conditions of the battery occur, such as long-term vacancy, long-term overheating, impact, etc., the real-time operation data is monitored and compared with the abnormal working conditions stored in the abnormal interference item database;

[0089] When it is judged that the real-time operation data triggers the abnormal working condition, the historical reliability factor (H) of each prediction algorithm in the similar abnormal working condition in history is queried from the database to judge its robustness in the working condition;

[0090] According to the historical robustness performance, the of the algorithm in the current period is adjusted, specifically including increasing the historical reliability factor corresponding to the algorithm that has shown high robustness to such abnormal working conditions in history; and decreasing the historical reliability factor corresponding to the algorithm that has shown sensitivity or large prediction bias to such abnormal working conditions in history.

[0091] As shown in Figure 2 , the embodiment of the present application provides a self-adaptive fusion prediction method for battery state of new energy vehicles, which comprises:

[0092] Step 101, collecting real-time operation data and historical operation data of the battery, collecting real-time operation data of voltage, current, state of charge, cell temperature and number of charge and discharge cycles from the battery management system (BMS), and retrieving the data and historical charging curve data required for supporting battery health state evaluation from the storage module, including voltage slope data in the constant current charging stage.

[0093] ​Step 102, pre-processing the collected real-time operation data and historical operation data:

[0094] Data cleaning, for removing abnormal data points caused by sensor drift or communication errors;

[0095] Time alignment, for ensuring the synchronization of data collected by different sensors on the time axis;

[0096] Feature extraction, for calculating the maximum capacity decay rate of the battery, the internal resistance change rate, etc. as the input features of the algorithm.

[0097] Step 103, selecting the prediction function to be executed, such as battery remaining life prediction. Select the corresponding algorithm group, such as long short-term memory network (LSTM), Gaussian process regression (GPR), and support vector regression (SVR). Based on the real-time operation data and historical operation data, the battery capacity after 20, 100, and 200 charge and discharge cycles is predicted respectively, and multiple prediction results and their respective prediction confidence are obtained.

[0098] Step 104, determining whether there is an abnormal working condition at present, if there is an abnormal working condition, such as cell overheating, if there is no new abnormal working condition, executing step 105.

[0099] Step 105, if there is an abnormal working condition, monitoring the real-time operation data and comparing it with the abnormal working conditions stored in the abnormal interference item database; when it is judged that the real-time operation data triggers the abnormal working condition, querying the historical reliable factor of each prediction algorithm under similar abnormal working conditions in history from the database to judge its robustness under this working condition; according to the historical robustness performance, adjusting the algorithm in the current period, which specifically includes increasing the of the algorithm that has shown high robustness to this type of abnormal working condition in history by 5% by experience; and reducing the of the algorithm that has shown sensitivity or large prediction deviation to this type of abnormal working condition in history by 8% by experience.

[0100] Step 106, reading the historical reliable factor corresponding to each of the multiple prediction algorithms from the storage module, weighting and ranking the multiple prediction results based on the algorithm prediction confidence and the historical reliable factor, and outputting the prediction result with the highest score as the final prediction result.

[0101] Step 107, after a preset charge and discharge cycle ends, obtaining the real state data of the battery as verification data; comparing the verification data with multiple prediction results to determine the deviation of each prediction result.

[0102] ​Step 108: Update the historical reliability factor corresponding to each algorithm in the storage module based on the deviation. Historical reliability factor ( The calculation method for ) is as follows:

[0103] The historical reliability factor It is a dynamic index used to quantify the first The prediction algorithm measures the historical prediction accuracy and robustness relative to the actual state of the battery throughout its entire lifespan; the specific calculation formula is as follows:

[0104] In each cycle After completion, based on the actual state value ( ) and predicted state values ​​( ), computational algorithm Instantaneous accuracy within this period .

[0105] ;

[0106] algorithm Historical reliability factors Based on the previous cycle factor Instantaneous accuracy for this period Perform iterative updates:

[0107] ;

[0108] in, :algorithm In the cycle The dynamic forgetting factor ( The calculation method is as follows:

[0109] ;

[0110] ;

[0111] in, These are the upper and lower bounds of the forgetting factor. It is a preset scaling factor used to control the sensitivity of the forgetting factor to changes in error.

[0112] like Figure 3 As shown, this invention provides an adaptive fusion prediction system for battery state forecasting in new energy vehicles, the system comprising:

[0113] Storage module 201 is used for:

[0114] store at least one training data set related to battery data; store a plurality of pre-trained battery state prediction algorithms, the algorithms are stored in groups according to prediction functions (including: battery state of health (SOH) evaluation, current power prediction remaining time, battery remaining useful life (RUL) prediction); store the historical reliability factor corresponding to each algorithm; store the real-time and historical operation data of the battery (voltage, current, SOC, temperature, etc.); store the historical prediction results and periodic verification results; store the abnormal interference item database;

[0115] The algorithm management module 202 is connected with the storage module, and is used for:

[0116] counting the number of available files, displaying system usage history data, and displaying system usage notes; supplementing the test data set; and configuring the prediction algorithm;

[0117] The processing module 203 is connected with the storage module and the algorithm management module, and is used for:

[0118] receiving a prediction task instruction, selecting a corresponding algorithm group according to the instruction, calling a plurality of algorithms in the algorithm group to respectively execute the prediction task, weighting and ranking the prediction results of each algorithm according to a preset confidence level and in combination with the historical reliability factor, outputting the prediction result with the highest score, obtaining real data of the battery after the charging and discharging cycle ends, verifying the prediction result, and updating the historical reliability factor of the corresponding algorithm in the storage module according to the accuracy of the verification.

[0119] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it 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. An adaptive fusion prediction method for the state of battery in new energy vehicles, characterized in that, The method includes: Step 1: Collect real-time and historical operating data of the battery, and preprocess the data; Step 2: Based on the received prediction task instruction, select the corresponding prediction algorithm group from an algorithm library that stores multiple pre-trained prediction algorithms; Step 3: Using multiple prediction algorithms in the algorithm group, based on the real-time running data and historical running data, perform predictions respectively to obtain multiple prediction results and their respective prediction confidence levels; Step 4: Read the historical reliability factors corresponding to each of the multiple prediction algorithms from the storage module, and perform weighted scoring and sorting on the multiple prediction results based on the prediction confidence and the historical reliability factors, and output the prediction result with the highest score as the final prediction result; Step 5: After a preset charge-discharge cycle, obtain the actual state data of the battery as verification data; compare the verification data with multiple prediction results to determine the deviation of each prediction result; Step 6: Update the historical reliability factor corresponding to each algorithm in the storage module according to the deviation.

2. The adaptive fusion prediction method for the state of a battery in a new energy vehicle according to claim 1, characterized in that, In step 1, the real-time operating data is collected from the battery management system or cloud data platform. The real-time operating data includes at least voltage, current, state of charge, cell temperature, and charge / discharge cycle count, as well as historical charging curve data required to support the state of health (SOH) assessment of the battery, including voltage slope data during the constant current charging phase; historical power output and duration relationship data under different ambient temperatures required to support the prediction of the remaining usage time of the current charge; and battery operating data during its life cycle and its corresponding periodically measured actual capacity values ​​required to support the prediction of the remaining battery life.

3. The adaptive fusion prediction method for battery state in new energy vehicles according to claim 1, characterized in that, Step 1 involves preprocessing the collected real-time and historical operating data. This preprocessing includes, but is not limited to: Data cleaning is used to remove abnormal data points caused by sensor drift or communication errors; Time alignment is used to ensure the synchronization of data acquired by different sensors on the timeline; and Feature extraction is used to calculate the battery's maximum capacity decay rate and internal resistance change rate as input features for the algorithm.

4. The adaptive fusion prediction method for battery status in new energy vehicles according to claim 1, characterized in that, In step 2, the prediction task instructions are divided into three types: battery health status assessment, prediction of estimated remaining usage time based on current battery level, and prediction of remaining battery life. The algorithms in the algorithm library are divided into three groups based on the task type, and each group includes, but is not limited to: The algorithm set for battery health status assessment includes at least two of the following algorithms: The feature extraction-based diagnostic algorithm is used to extract charging curve features to directly estimate the current maximum available capacity; the model-based state estimation algorithm is used to estimate the current battery health state (SOH) based on the real-time changes of the equivalent circuit model parameters, using the ratio of the current battery capacity to the initial state as the SOH value. For the algorithm group that predicts the remaining time of current battery usage, the algorithm group includes at least two of the following algorithms: A data-driven power prediction algorithm is used to predict short-term power demand based on historical driving behavior and temperature data; a lookup table-based energy consumption estimation algorithm is used to combine cell temperature, current battery health status (SOH), and historical energy consumption data to output the remaining usable time through a nonlinear mapping relationship. The algorithm set for predicting battery remaining life includes at least two of the following algorithms: A deep learning-based time-series prediction model is used to learn the long-term degradation sequence of battery capacity; a kernel function-based regression model is used to perform nonlinear regression prediction of future capacity.

5. The adaptive fusion prediction method for the state of a battery in a new energy vehicle according to claim 1, characterized in that, Step 3, obtaining the prediction confidence specifically includes: The prediction confidence level is dynamically determined based on the type of prediction algorithm, and the confidence level includes, but is not limited to: When the prediction algorithm is a deep learning or regression-based model, the confidence level is determined based on the prediction variance, probability distribution, or error rate of the model on the historical validation set of the model output. When the prediction algorithm is a feature extraction-based diagnostic algorithm, the confidence level is determined based on the signal-to-noise ratio of the input data, the clarity of the extracted features, or the goodness of fit between the feature curve and the standard decay curve.

6. The adaptive fusion prediction method for the state of a battery in a new energy vehicle according to claim 1, characterized in that, The historical reliability factor mentioned in step 4 The specific definition is: In each cycle After completion, based on the actual state value and predicted state values Calculation algorithm Instantaneous accuracy within this period ; ； algorithm Historical reliability factors Based on the previous cycle factor Instantaneous accuracy for this period Perform iterative updates: ； in, For the algorithm In the cycle The dynamic forgetting factor is calculated as follows: ； ； in, These are the upper and lower bounds of the forgetting factor. It is a preset proportional coefficient.

7. The adaptive fusion prediction method for the state of a battery in a new energy vehicle according to claim 1, characterized in that, The specific method for weighted scoring and ranking of the prediction results described in step 4 is as follows: For the first in the algorithm group A prediction algorithm is used to calculate the final score. The rating It is determined by the prediction confidence level of the algorithm. With the historical reliability factor Obtained through multiplication and fusion, its mathematical relationship is expressed as: ； The weighted scoring and ranking specifically include: Compute the score set of all N algorithms in the algorithm group. ; For the rating set Sort in descending order; The prediction result corresponding to the algorithm with the highest score is output as the final prediction result.

8. The adaptive fusion prediction method for the state of a battery in a new energy vehicle according to claim 4, characterized in that, In step 5, the verification data is compared with multiple prediction results to determine the deviation of each prediction result. Specifically, based on the prediction task type, the deviation is categorized as follows: For battery health status assessment tasks: Battery health status assessment bias It is the relative error between the predicted current maximum available capacity or SOH value and the validation data; For the task of predicting the remaining time of current battery usage: Remaining time prediction bias It is the absolute percentage error between the predicted remaining time or remaining range and the actual remaining time or remaining range observed during the period. For the task of predicting remaining battery life: Prediction bias of remaining battery life It is the normalized absolute error between the predicted battery capacity or SOH value at a preset future time point and the actual measured battery capacity or SOH value at that future time point, wherein the normalization is based on the nominal capacity of the battery.

9. The adaptive fusion prediction method for the state of a battery in a new energy vehicle according to claim 1, characterized in that, The historical reliability factor update method in step 6 further includes: when abnormal battery usage occurs, monitoring the real-time operating data and comparing it with abnormal operating conditions stored in the abnormal interference item database; when it is determined that the real-time operating data triggers an abnormal operating condition, querying the database for historical reliability factors of each prediction algorithm under similar abnormal operating conditions. The algorithm's robustness under the specified operating conditions is determined by analyzing the changing trends; based on historical robustness performance, the algorithm is adjusted for the current cycle. Specifically, this includes increasing the historical reliability factor for algorithms that have historically demonstrated high robustness to such abnormal operating conditions; and decreasing the historical reliability factor for algorithms that have historically been sensitive to such abnormal operating conditions or had large prediction biases.

10. An adaptive fusion prediction system for the state of a battery in a new energy vehicle, characterized in that, A method for implementing any one of claims 1-9, comprising: Storage module, used for: (a) Store at least one training dataset related to battery data; (b) Store multiple pre-trained battery state prediction algorithms, which include, in order of prediction: battery state of health (SOH) assessment, estimated remaining time of use at current charge level, and battery remaining life (RUL) prediction; (c) Store the historical reliability factor for each algorithm; (d) Store real-time and historical operating data of the battery, including voltage, current, SOC, and temperature; (e) Store historical prediction results and periodic verification results; (f) Store an abnormal interference item database; The algorithm management module is connected to the storage module and is used for: (a) Count the number of currently available files, display system usage history data, and display system usage precautions; (b) Supplement the test dataset; (c) Configure the prediction algorithm; The processing module, connected to the storage module and the algorithm management module, is used for: (a) Receive the prediction task instruction and select the corresponding algorithm group according to the instruction; (b) Invoke multiple algorithms from the algorithm group to perform prediction tasks respectively; (c) Based on the preset confidence level and in combination with the historical reliability factor, the prediction results of each algorithm are weighted, scored and ranked. (d) Output the prediction result with the highest score; (e) After the charge-discharge cycle ends, the actual data of the battery is obtained, the prediction results are verified, and the historical reliability factor of the corresponding algorithm in the storage module is updated according to the accuracy of the verification.

Images