Power battery pack health state monitoring system

Abstract

The invention discloses a power battery pack health state monitoring system, and relates to the field of battery pack health state monitoring, and the system comprises a vehicle information input module which is used for receiving vehicle unique identification information inputted by a user; the standard data acquisition module is used for acquiring historical operation and charging and discharging data corresponding to the unique identification information of the vehicle from a cloud data source; the data fusion and feature construction module is used for constructing a battery state evaluation feature set; the battery health assessment module is used for constructing a battery pack health state assessment model and outputting a battery health state assessment result; the safety risk grading module is used for carrying out grading evaluation on the potential safety risk of the vehicle battery; and the report generation and display module is used for generating a structured detection report. The method has the advantages that the full life cycle of the power battery pack from real-time health state accurate evaluation to future safety risk grading early warning is realized, and intelligent monitoring can be explained.

Description

Technical Field

[0001] This invention relates to the field of battery pack health status monitoring, and more specifically to a power battery pack health status monitoring system. Background Technology

[0002] With the continuous growth in the number of new energy vehicles, accurate verification of the health status of power battery packs has become a key factor affecting vehicle pricing and transaction security in the process of trading used electric vehicles. Currently, the detection of the health status of power batteries mainly relies on the following technical solutions: The first approach is a local detection solution based on an OBD device. This involves using an external OBD device to read vehicle fault codes and some real-time operating parameters, such as SOH, cell voltage, and temperature. However, this solution has the following drawbacks: it requires manual connection and operation, making it unfriendly to non-professionals; the OBD protocols of different vehicle models vary significantly, resulting in poor compatibility of low-cost OBD devices and frequent issues such as inability to read or missing data; and it can only obtain current status data, failing to reflect the long-term degradation trend and historical usage characteristics of the battery. The second method is to rely on the evaluation data displayed by the vehicle's BMS. The battery health information displayed by the vehicle's instrument panel or central control system is mostly the result of the car manufacturer's internal algorithm, lacking the support of original data. It is difficult for third parties to verify its authenticity, especially in the context of used car transactions, where its credibility is insufficient. The third method is offline professional testing or disassembly testing. This method requires professional personnel and specialized equipment, and the testing cost is high and the cycle is long, making it unsuitable for business scenarios that require large-scale and rapid verification. Existing technical solutions generally suffer from problems such as high testing thresholds, insufficient adaptability, inability to obtain historical degradation characteristics, and difficulty in achieving remote and rapid evaluation. Therefore, there is an urgent need for a solution that can meet the needs of used car transactions and individual car owners for low-cost, high-efficiency, and standardized evaluation of the health status of power batteries. Summary of the Invention

[0003] To address the aforementioned technical problems, a power battery pack health status monitoring system is provided. This technical solution solves the problems commonly found in existing technical solutions mentioned in the background section, such as high detection threshold, insufficient adaptability, inability to obtain historical degradation characteristics, and difficulty in achieving remote and rapid evaluation.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A power battery pack health status monitoring system, comprising: The vehicle information input module is used to receive unique vehicle identification information input by the user, including: vehicle VIN code and / or vehicle registration certificate information; A standard data acquisition module is used to acquire historical operation and charging / discharging data corresponding to the vehicle's unique identification information from a cloud data source based on the national standard GBT 32960 protocol. The data fusion and feature construction module is used to clean, verify, and reconstruct the time series of the collected multi-source historical data to construct a battery status assessment feature set. The battery health assessment module is used to construct a battery pack health status assessment model and output battery health status assessment results based on the battery status assessment feature set. A safety risk classification module, which is connected to the battery health assessment module, is used to classify and assess the potential safety risks of vehicle batteries based on safety risk coefficients. The report generation and display module integrates the battery health status assessment results and safety risk assessment results to generate a structured test report, which is then visualized through a mini-program or mobile terminal interface.

[0005] Preferably, the step of cleaning, verifying, and reconstructing the time series of the collected multi-source historical data to construct a battery state assessment feature set specifically includes: For data fields in the GBT 32960 standard, automatic cleaning is performed by setting reasonable physical thresholds based on historical data statistics. Based on the missing data segments from multiple sources, each historical data segment is processed separately. Convert the timestamps of all data sources to UTC time and resample and align data with different sampling frequencies to the same time grid. Based on the upload timestamp, identify and delete identical duplicate data records. For conflicting records with the same timestamp but different values, retain the latest uploaded data version. Check if the timestamps of the uploaded data are strictly incrementing, and mark and reorder the data records that are reverted in time. Based on the logical relationship between historical operation and charge / discharge data parameters, multi-source historical data is verified. Numerical data in multi-source historical data is filtered and normalized, while non-numerical data in multi-source historical data is quantized using one-hot encoding. The cleaned, verified, and reconstructed time series data are organized into a unified time series data structure to construct a battery state assessment dataset. Based on the battery state assessment dataset, battery capacity change features, charge / discharge behavior features, usage intensity features, battery consistency features, and abnormal mode features are extracted to construct a battery state assessment feature set.

[0006] Preferably, the construction of the battery pack health status assessment model, based on the battery status assessment feature set, and the output of the battery health status assessment result specifically includes: A battery pack health status assessment model is constructed, with the battery status assessment feature set as input and battery health, battery degradation value, reference range, cycle usage characteristics and safety risk coefficient as output. The cyclic use characteristics include at least: equivalent number of cycles, cycle depth distribution, and cycle decay coefficient; Based on the historical operating data of new energy vehicle power battery packs, corresponding historical operating and charge / discharge data are collected through the GBT 32960 standard protocol; Based on historical operation and charge / discharge data, combined with real health status labels obtained from laboratory standard tests, a battery status assessment feature set is extracted, a training sample set is constructed, and divided into a training set and a test set. The sample set includes battery status assessment feature sets and corresponding real health status labels for different vehicle models, different usage scenarios, and different degradation cycles. For mixed tasks involving regression and probability estimation with multiple outputs, a composite loss function is used to calculate the difference between the model's predicted values ​​and the true labels. The composite loss function consists of mean squared error loss, binary cross-entropy loss, and physical constraint penalty term, and is a weighted sum of the losses. The physical constraint penalty term introduces rule constraints based on the principles of battery electrochemistry into the loss function, and its mathematical expression includes a quantitative calculation of the degree of violation of key physical laws. With the goal of minimizing the composite loss function, an adaptive optimization algorithm is used to train the model based on the training set, and the trained model is defined as the battery pack health status assessment model. The stability and accuracy of the battery pack health status assessment model are evaluated using performance indicators such as mean absolute error, root mean square error, coefficient of determination, AUC value of risk prediction, false alarm rate, and false alarm rate.

[0007] Preferably, the security risk classification module specifically includes: The historical sequence unit is used to infer the battery status assessment feature set arranged in chronological order using a trained battery pack health status assessment model, and output the safety risk coefficient corresponding to each time point, thereby generating a complete historical safety risk coefficient sequence. The sample set unit is used to construct a training sample set for the LSTM encoder based on the historical security risk coefficient sequence. Each training sample takes the data of the first N time points of the historical security risk coefficient sequence as input and the actual risk coefficients of the next M time points as labels. The prediction unit is used to obtain the latest safety risk coefficient sequence of the vehicle to be queried, set a sliding window, input it into the trained LSTM encoder, and predict the safety risk coefficient prediction values ​​for the next M time points. The grading unit is used to calculate the slope of change of the predicted risk coefficient, the predicted peak value, and the expected time exceeding the threshold based on the predicted safety risk coefficient value, and to classify the long-term safety risk of the battery into four levels: Grade A: Forecasted risk remains low and the trend is stable; Grade B: The predicted risk is rising slowly, with no short-term risk of exceeding the limit; Level C: The predicted risk is rising rapidly, with a risk of exceeding the limit in the medium term; Level D: Current risk is high and is predicted to continue to worsen; The assessment report unit is used to integrate the long-term risk level with the real-time safety risk coefficient to form a complete battery safety status assessment report.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a power battery pack health status monitoring system, which achieves remote and rapid assessment of battery health status based on historical operating data without the need for external hardware by constructing a standard data acquisition module, a data fusion and feature construction module, a battery health assessment module, and a safety risk assessment module. Specifically, on the one hand, the system acquires historical vehicle operation and charging / discharging data from the cloud based on the GBT 32960 national standard protocol, effectively solving the problems of large differences in protocols and poor data compatibility among different vehicle models, and achieving broad compatibility with mainstream and older new energy vehicle models. On the other hand, by constructing a multi-dimensional feature set covering battery capacity changes, charging / discharging behavior, usage intensity, consistency, and abnormal modes, and adopting a deep neural network model that integrates physical constraints, it can not only output real-time status indicators such as current battery health (SOH), degradation value, and reference range, but also conduct long-term hierarchical assessment of future safety risks through a time-series prediction model, realizing full-cycle health management from "current status assessment" to "future risk prediction". Compared with traditional solutions that rely on local detection by OBD devices or display data by the vehicle's BMS, this solution requires no manual intervention and does not rely on specific hardware, significantly reducing the detection threshold and cost. Compared with offline professional testing or disassembly testing methods, this solution can achieve second-level response and standardized report output, making it particularly suitable for high-frequency, rapid verification scenarios such as used car transactions. By employing objective data models and risk grading mechanisms, the transparency, credibility, and safety of battery health assessments have been significantly enhanced, providing reliable technical support for applications such as used car transaction pricing, battery residual value assessment, and safety early warning. Attached Figure Description

[0009] Figure 1 This is a structural block diagram of a power battery pack health status monitoring system according to the present invention; Figure 2 The present invention provides a flowchart for cleaning, verifying, and reconstructing time series data from multiple sources to construct a battery state assessment feature set. Figure 3 To construct the battery pack health status assessment model of this invention, a flowchart is provided to output the battery health status assessment results based on the battery status assessment feature set. Figure 4 The present invention provides a four-level structural block diagram for classifying long-term battery safety risks. Detailed Implementation

[0010] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0011] Reference Figure 1 As shown, a power battery pack health status monitoring system includes: The vehicle information input module is used to receive unique vehicle identification information input by the user, including: vehicle VIN code and / or vehicle registration certificate information; A standard data acquisition module is used to acquire historical operation and charging / discharging data corresponding to the vehicle's unique identification information from a cloud data source based on the national standard GBT 32960 protocol. The data fusion and feature construction module is used to clean, verify, and reconstruct the time series of the collected multi-source historical data to construct a battery status assessment feature set. The battery health assessment module is used to construct a battery pack health status assessment model and output battery health status assessment results based on the battery status assessment feature set. A safety risk classification module, which is connected to the battery health assessment module, is used to classify and assess the potential safety risks of vehicle batteries based on safety risk coefficients. The report generation and display module integrates the battery health status assessment results and safety risk assessment results to generate a structured test report, which is then visualized through a mini-program or mobile terminal interface.

[0012] Reference Figure 1 As shown, the security risk classification module specifically includes: The historical sequence unit is used to infer the battery status assessment feature set arranged in chronological order using a trained battery pack health status assessment model, and output the safety risk coefficient corresponding to each time point, thereby generating a complete historical safety risk coefficient sequence. The sample set unit is used to construct a training sample set for the LSTM encoder based on the historical security risk coefficient sequence. Each training sample takes the data of the first N time points of the historical security risk coefficient sequence as input and the actual risk coefficients of the next M time points as labels. The prediction unit is used to obtain the latest safety risk coefficient sequence of the vehicle to be queried, set a sliding window, input it into the trained LSTM encoder, and predict the safety risk coefficient prediction values ​​for the next M time points. The grading unit is used to calculate the slope of change of the predicted risk coefficient, the predicted peak value, and the expected time exceeding the threshold based on the predicted safety risk coefficient value, and to classify the long-term safety risk of the battery into four levels: Grade A: Forecasted risk remains low and the trend is stable; Grade B: The predicted risk is rising slowly, with no short-term risk of exceeding the limit; Level C: The predicted risk is rising rapidly, with a risk of exceeding the limit in the medium term; Level D: Current risk is high and is predicted to continue to worsen; The assessment report unit is used to integrate the long-term risk level with the real-time safety risk coefficient to form a complete battery safety status assessment report.

[0013] Explained by this, the safety risk grading module extends the assessment from current risk status to future risk trend prediction by combining time-series prediction with multi-threshold decision-making. Based on a trained battery pack health status assessment model, the module generates a sequence of historical vehicle safety risk coefficients, which is then used to construct training samples for the LSTM encoder, enabling the model to learn the evolution of risk over time. It then samples the latest risk sequence of the current vehicle using a sliding window and inputs it into the trained LSTM encoder to predict safety risk coefficients at multiple future time points. Finally, by calculating the slope, peak value, and over-threshold time of the predicted sequence, and combining this with multiple preset quantization thresholds, the long-term safety risk is dynamically divided into four levels: A, B, C, and D, corresponding to continuous states from stable low risk to emergency high risk. This grading result, along with the real-time safety risk coefficients, is output to form a battery safety assessment report that reflects both the current state and warns of future trends, significantly improving the systematic and forward-looking nature of the assessment.

[0014] The specific details of receiving the vehicle's unique identifier information input by the user include: Using WeChat mini-programs, apps, or web pages, users can obtain the unique identification information of the vehicle to be queried from the cloud by entering the vehicle's VIN code and / or uploading a photo of the vehicle registration certificate.

[0015] This can be explained by the fact that users can submit the vehicle's identification information by entering the vehicle's VIN code and / or uploading a photo of the vehicle registration certificate through WeChat mini-program, APP, or web interface. The system automatically parses the submitted information, extracts and verifies the vehicle's unique identifier, which serves as a key index for subsequently obtaining the vehicle's historical operating data from the cloud. This method does not require special equipment, lowers the barrier to entry, and is suitable for individual car owners, used car dealers, and other users to quickly initiate battery health status assessments.

[0016] The specific steps of obtaining historical operation and charging / discharging data corresponding to the vehicle's unique identification information from a cloud data source include: It interfaces with the GBT 32960 standard protocol and / or the vehicle manufacturer's cloud data platform, and sends a request to the data platform based on the input VIN code and / or vehicle registration information to obtain the vehicle's historical operation and charging / discharging data. The data includes, but is not limited to: vehicle static information, basic battery parameters, charging behavior data, and driving-related data.

[0017] This can be explained by the fact that, based on the GBT 32960 national standard protocol, the system connects to a cloud-based data platform for new energy vehicles that conforms to this standard, and collects historical operation and charging / discharging data from multiple sources for the target vehicles in batches. Compared to traditional solutions, the GBT 32960 standard has a unified communication interface and data format, and is compatible with mainstream brands and some older new energy vehicle models, significantly improving system adaptability. The collected data includes vehicle static information (such as vehicle model, battery type, and rated capacity), basic battery parameters (such as cell voltage, temperature, and SOC changes), charging behavior data (such as charging frequency, charging duration, charging voltage range, and fast charging ratio), and driving-related data (such as mileage, driving conditions, and energy consumption level), providing comprehensive data support for subsequent evaluation.

[0018] Reference Figure 2 As shown, the process of cleaning, verifying, and reconstructing the time series of the collected multi-source historical data to construct a battery state assessment feature set specifically includes: For data fields in the GBT 32960 standard, automatic cleaning is performed by setting reasonable physical thresholds based on historical data statistics. Based on the missing data segments from multiple sources, each historical data segment is processed separately. Convert the timestamps of all data sources to UTC time and resample and align data with different sampling frequencies to the same time grid. Based on the upload timestamp, identify and delete identical duplicate data records. For conflicting records with the same timestamp but different values, retain the latest uploaded data version. Check if the timestamps of the uploaded data are strictly incrementing, and mark and reorder the data records that are reverted in time. Based on the logical relationship between historical operation and charge / discharge data parameters, multi-source historical data is verified. Numerical data in multi-source historical data is filtered and normalized, while non-numerical data in multi-source historical data is quantized using one-hot encoding. The cleaned, verified, and reconstructed time series data are organized into a unified time series data structure to construct a battery state assessment dataset. Based on the battery state assessment dataset, battery capacity change features, charge / discharge behavior features, usage intensity features, battery consistency features, and abnormal mode features are extracted to construct a battery state assessment feature set.

[0019] It can be explained that historical data acquisition and preprocessing are important components of power battery pack health status monitoring. They are the primary conditions for accurately assessing the health status of the battery pack and evaluating potential safety risks of vehicle batteries. Therefore, after acquiring historical operation and charging / discharging multi-source data of the vehicle to be queried through the cloud, the source data needs to be preprocessed to ensure the integrity, accuracy, unified timestamps, and unified quantification units of the data, thereby providing data support for subsequent assessment of potential safety risks of vehicle batteries. The aforementioned automatic cleaning of GBT 32960 standard data fields, based on historical data statistics and setting physically reasonable thresholds, includes, but is not limited to: Voltage data cleaning: Remove data records with total voltage below a1 or above a2, and remove data segments with individual voltage ranges exceeding a3 and durations exceeding a4 minutes; Temperature data cleaning: Remove abnormal data where the battery temperature is lower than a5 or higher than a6, and remove abrupt data points with a temperature change rate exceeding a7℃ / minute. SOC data cleaning: Remove data records with SOC values ​​less than a8 or greater than a9, and remove discontinuous data whose SOC jumps more than a11 within a10 seconds; Current data cleaning: Remove instantaneous data with an absolute current value exceeding a12, and remove data points where the sign of the charging and discharging current contradicts the direction of SOC change; The specific steps of processing each historical data entry based on the missing data segments from multiple sources include: Short-term missing data imputation: For data segments with consecutive missing durations within a13 seconds, linear interpolation of the preceding and following valid data is used for imputation. Long-term missing data marking: Data segments with continuous missing data for more than a14 minutes are marked as invalid data intervals and are not included in feature calculation; Handling missing critical fields: If any of the critical fields such as SOC, total voltage, or total current is missing, the entire record will be marked as invalid. It should be noted that the values ​​a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11, a12, a13, and a14 are physically reasonable thresholds set based on historical data statistics and / or combined with expert experience. The purpose is to remove and supplement incomplete or abnormal data from a large amount of multi-source historical operation and charging / discharging data of the vehicle to be queried, so as to avoid interference from this data. The verification of multi-source historical data based on the logical relationship between historical operation and charge / discharge data parameters, in specific implementation examples, includes, but is not limited to: SOC and current direction verification: During charging, the current should be positive and the SOC should increase; during discharging, the current should be negative and the SOC should decrease. Data points that violate this logic are marked. Voltage and current product verification: Calculate instantaneous power = voltage × current, and eliminate physically unreasonable data whose power value exceeds the maximum design power of the vehicle; Mileage accumulation verification: Verify whether the accumulated mileage data is monotonically increasing, and mark cases where mileage rollback is not due to maintenance reset; The unified time-series data structure, each data point contains: Standard timestamp (UTC format); Vehicle identification information (VIN code and / or vehicle registration information); Vehicle static information (such as vehicle model, battery type, rated capacity); Battery basic parameters (such as cell voltage, temperature, and SOC changes); Charging behavior data (such as charging frequency, charging duration, charging voltage range, and fast charging percentage); Driving-related data (such as mileage, driving conditions, and energy consumption levels); Time series type identifier (e.g., driving, charging, stationary); The extracted battery capacity change characteristics, charge / discharge behavior characteristics, usage intensity characteristics, battery consistency characteristics, and abnormal mode characteristics specifically include: I. Battery capacity change feature extraction: Based on complete charging events identified from the cleaned and reconstructed time series data, capacity features are extracted through the following steps: Charging capacity estimation: The charging energy is calculated by integrating the charging current over time, and the actual usable capacity of the battery is estimated by combining the SOC values ​​at the start and end of charging and the charging efficiency coefficient. Capacity degradation analysis: Statistically analyze the estimated capacity values ​​from multiple historical charging events, fit the capacity degradation curve over time or usage mileage, and calculate the capacity retention rate and degradation rate. Incremental capacity analysis: Extract the constant current charging stage from the charging event data, plot the voltage-to-charging capacity curve, smooth and filter the curve, and then take its derivative to obtain the incremental capacity curve. Identify and record the peak position and peak height of the curve. II. Charging and discharging behavior feature extraction: Based on the data sequences of all charging events and driving processes, the following features are extracted: Depth of charge / discharge characteristics: Statistical distribution of SOC change during charging / discharging, including average depth of charge / discharge, ratio of deep to shallow charge / discharge, and initial SOC distribution during charging / discharging; Charge and discharge rate characteristics: Calculate the ratio of fast and slow charging usage, average charging and discharging power, maximum charging and discharging rate, and the percentage of usage time in different power ranges; Charging and discharging habits: Analyze the charging time distribution (such as the proportion of charging at night), charging completion rate, resting time after charging, and charging frequency per unit time. Energy efficiency characteristics: Calculate charge / discharge energy efficiency and regenerative braking efficiency; III. Using intensity feature extraction, based on the time series data of vehicle operating status, the following features are extracted: Intensity of time dimension usage: Calculate total mileage, total running time, equivalent number of cycles, average daily mileage, and average daily running time; Environmental stress characteristics: statistical temperature cycle count and average operating temperature; IV. Battery Consistency Feature Extraction: Based on the voltage of each cell and the temperature data of each measurement point, the following features are extracted: Voltage consistency characteristics: Calculate the range, standard deviation, and coefficient of variation of the voltage of each cell; Temperature consistency characteristics: Calculate the maximum temperature difference and temperature standard deviation between each temperature measurement point; V. Abnormal Pattern Feature Extraction: Based on vehicle alarm records and time-series data of operating parameters, the following features are extracted: Alarm event characteristics: Statistics on the frequency and repetition of various alarms; Parameter out-of-bounds characteristics: Analyze the frequency of parameters such as voltage, current, and temperature approaching or exceeding the threshold. The extracted features are combined in a predetermined order to form a multidimensional feature vector, which is then standardized to construct a battery state assessment feature set, which serves as the input to the battery pack health state assessment model.

[0020] Reference Figure 3 As shown, the specific steps for outputting battery health status assessment results based on the battery status assessment feature set include: A battery pack health status assessment model is constructed, with the battery status assessment feature set as input and battery health, battery degradation value, reference range, cycle usage characteristics and safety risk coefficient as output. The cyclic use characteristics include at least: equivalent number of cycles, cycle depth distribution, and cycle decay coefficient; Based on the historical operating data of new energy vehicle power battery packs, corresponding historical operating and charge / discharge data are collected through the GBT 32960 standard protocol; Based on historical operation and charge / discharge data, combined with real health status labels obtained from laboratory standard tests, a battery status assessment feature set is extracted, a training sample set is constructed, and divided into a training set and a test set. The sample set includes battery status assessment feature sets and corresponding real health status labels for different vehicle models, different usage scenarios, and different degradation cycles. For mixed tasks involving regression and probability estimation with multiple outputs, a composite loss function is used to calculate the difference between the model's predicted values ​​and the true labels. The composite loss function consists of mean squared error loss, binary cross-entropy loss, and physical constraint penalty term, and is a weighted sum of the losses. The physical constraint penalty term introduces rule constraints based on the principles of battery electrochemistry into the loss function, and its mathematical expression includes a quantitative calculation of the degree of violation of key physical laws. With the goal of minimizing the composite loss function, an adaptive optimization algorithm is used to train the model based on the training set, and the trained model is defined as the battery pack health status assessment model. The stability and accuracy of the battery pack health status assessment model are evaluated using performance indicators such as mean absolute error, root mean square error, coefficient of determination, AUC value of risk prediction, false alarm rate, and false alarm rate.

[0021] It can be explained that by using a data acquisition and feature construction process based on the GBT 32960 standard, multi-dimensional historical operating status data of the power battery pack can be effectively obtained, providing a data foundation for health assessment. However, in the actual assessment process, the battery health status is affected by multiple complex mechanisms coupled together, such as capacity decay, internal resistance growth, electrochemical aging, usage habits, and environmental factors. A single indicator or simple formula is difficult to comprehensively and accurately reflect its true health status. Therefore, the solution constructs a battery pack health status assessment model, using a set of battery status assessment features as input. Among them, battery capacity change features are the core quantitative basis for the model to assess the current health status, directly reflecting the main manifestations of battery electrochemical aging; charge and discharge behavior features and usage intensity features quantify the usage conditions that lead to aging from the perspective of external stress; battery consistency features and abnormal mode features provide supplementary criteria from a safety perspective. Thus, by utilizing the physical correlation and statistical regularity between various features, the comprehensive health status of the battery pack can be accurately assessed, and its degree of degradation and remaining range can be quantified, providing an objective and quantitative decision-making basis for scenarios such as used car transactions. The battery pack health status assessment model is based on a deep neural network framework. The input layer of the model receives the battery status assessment feature vector and contains several fully connected hidden layers. It uses the ReLU activation function and outputs battery health (SOH, linear activation), battery degradation value (linear activation), reference range (linear activation), cyclic usage characteristics (linear activation), and safety risk coefficient (Sigmoid activation) through multiple parallel output layers. The safety risk coefficient is calculated by an independent output branch, whose input shares the features of the aforementioned hidden layers. The safety risk coefficient is directly generated by the model's Sigmoid output layer, and its output value itself is the estimated probability of the risk event occurring, which is used as the confidence probability. The logic for obtaining the true health status label is as follows: when standard laboratory test data is available, the actual capacity and internal resistance obtained from standard charge-discharge tests are used as the benchmark labels for SOH and degradation values; when laboratory data is lacking, rigorously verified actual capacity test data at the end of vehicle use (such as when scrapped or when the battery is replaced) or indirect calibration can be used through statistical degradation curves of a large batch of batteries of the same model; for the range label, it is calibrated based on the mileage traveled by the vehicle from full charge to cutoff voltage under standard environmental and driving conditions in actual road tests; for the cyclic use characteristic, it is obtained by accumulating energy and equivalent conversion of historical charge-discharge segments; the physical constraint penalty term refers to adding rule constraints based on battery electrochemical principles to the loss function. This penalty term integrates prior physical knowledge into the data-driven model, improving the rationality and interpretability of the model's prediction results; in a specific embodiment, the physical constraint penalty term can be implemented in the following way: for all consecutive time sample pairs in the training batch, calculate the non-incremental penalty of the SOH prediction value. ,in, For the time sample pair The predicted battery health value for each sample. For the time sample pair The battery health value predicted for each sample is calculated, and for all individual cell voltage data points involved in the training batch, a voltage out-of-bounds penalty is calculated. ,in, This refers to the voltage value of a single battery cell. , These represent the upper and lower limits of the maximum safe voltage for a single battery cell, respectively, with a total physical constraint penalty of... ,in , These are the weighting coefficients.

[0022] Reference Figure 4 As shown, the classification of long-term battery safety risks into four levels specifically includes: Using the trained battery pack health status assessment model, reasoning is performed on the battery status assessment feature set arranged in chronological order, and the safety risk coefficient corresponding to each time point is output, thereby generating a complete historical safety risk coefficient sequence. Based on the historical security risk coefficient sequence, a training sample set for the LSTM encoder is constructed. Each training sample takes the data of the first N time points of the historical security risk coefficient sequence as input and the actual risk coefficients of the next M time points as labels. For the vehicle to be queried, obtain its latest safety risk coefficient sequence, set a sliding window, input it into the trained LSTM encoder, and predict the safety risk coefficient prediction values ​​for the next M time points. Based on the predicted safety risk coefficient, the slope of change of the predicted risk coefficient, the predicted peak value, and the expected time exceeding the threshold are calculated, and the long-term safety risk of the battery is divided into four levels: Grade A: Forecasted risk remains low and the trend is stable; Grade B: The predicted risk is rising slowly, with no short-term risk of exceeding the limit; Level C: The predicted risk is rising rapidly, with a risk of exceeding the limit in the medium term; Level D: Current risk is high and is predicted to continue to worsen; The assessment report unit is used to integrate the long-term risk level with the real-time safety risk coefficient to form a complete battery safety status assessment report.

[0023] This can be explained by the fact that the battery pack health status assessment model outputs safety risk coefficients based on current data and historical data, representing an assessment of the current or near-term state. However, scenarios such as used car transactions, insurance pricing, and battery recycling require predictions of future risks, necessitating the development of long-term risk assessment capabilities. This solution achieves long-term risk prediction through the following technical path: First, the battery status assessment feature set is organized into a sliding window sequence in chronological order. The battery pack health status assessment model then predicts safety risk coefficient sequences under different sliding window sequences. Next, an LSTM encoder is trained based on historical data, learning the mapping relationship from historical risk sequences to future risk values. Finally, the real-time obtained safety risk coefficient sequence is input into the trained LSTM encoder, outputting predicted future safety risk coefficients with temporal relationships, thereby supporting long-term risk grading. In practice, the safety risk coefficient is a parallel output of the battery pack health status assessment model (generated by the Sigmoid branch). During historical data backtracking or real-time monitoring, the model infers the data once for each time point, simultaneously outputting health status, degradation value, range, and safety risk coefficient. Therefore, the "safety risk coefficient sequence" is a naturally generated byproduct of the model's processing of time-series data, eliminating the need to rerun the entire model for each sliding window. Among them, the battery pack health status assessment model is a multi-task model. Its Sigmoid output branch can directly generate safety risk coefficients. Therefore, it is only necessary to perform sequential reasoning on the historical or real-time uploaded feature data stream once to efficiently obtain the complete safety risk coefficient sequence without having to run the model repeatedly for different time segments. The long-term risk classification is based on preset quantitative thresholds, including a slope threshold, a peak threshold, a time threshold, and a current value threshold. By comparing the calculated risk change slope, predicted peak value, expected time exceeding the threshold, and current risk coefficient with the corresponding thresholds, long-term safety risks are classified into four levels: A, B, C, and D, according to a preset decision logic. Specifically, the classification decision logic is implemented through multi-threshold comparison: when the risk change slope is less than the first threshold, the predicted peak value is less than the second threshold, and the current risk coefficient is less than the third threshold, it is determined to be level A; when the risk change slope is greater than or equal to the first threshold and less than the fourth threshold, and the predicted peak value is less than the third threshold, it is classified as level D. When the risk level is at the fifth threshold and the predicted time exceeding the threshold is greater than the sixth threshold, and the current risk coefficient is less than the seventh threshold, it is classified as Grade B. When the risk change slope is greater than or equal to the fifth threshold, or the current risk coefficient is greater than or equal to the seventh threshold and the risk change slope is positive, or the predicted time exceeding the threshold is less than or equal to the sixth threshold, or the predicted peak value is greater than or equal to the ninth threshold, it is classified as Grade D. When the above Grade A, B, and D criteria are not met, but any of the following conditions are met, it is classified as Grade C: the risk change slope is greater than or equal to the fourth threshold; the predicted peak value is greater than or equal to the eighth threshold; the current risk coefficient is greater than or equal to the third threshold and the risk change slope is positive; the predicted time exceeding the threshold is less than or equal to the tenth threshold. The first to tenth thresholds satisfy the following numerical relationships: first threshold < fourth threshold < fifth threshold; second threshold < fifth threshold; third threshold < seventh threshold; sixth threshold < tenth threshold; eighth threshold < ninth threshold; the thresholds can be determined based on historical validation sets using receiver operating characteristic (ROC) curve analysis, and by balancing the false alarm rate and the false negative rate, the thresholds can be adapted to the differentiated requirements of safety warning accuracy in different assessment scenarios.

[0024] In summary, the advantages of this invention are: it enables interpretable intelligent monitoring of the entire lifecycle of a power battery pack, from accurate real-time health status assessment to tiered early warning of future safety risks.

[0025] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A power battery pack health status monitoring system, characterized in that, include: The vehicle information input module is used to receive unique vehicle identification information input by the user, including: vehicle VIN code and / or vehicle registration certificate information; A standard data acquisition module is used to acquire historical operation and charging / discharging data corresponding to the vehicle's unique identification information from a cloud data source based on the national standard GBT 32960 protocol. The data fusion and feature construction module is used to clean, verify, and reconstruct the time series of the collected multi-source historical data to construct a battery status assessment feature set. The battery health assessment module is used to construct a battery pack health status assessment model and output battery health status assessment results based on the battery status assessment feature set. A safety risk classification module, which is connected to the battery health assessment module, is used to classify and assess the potential safety risks of vehicle batteries based on safety risk coefficients. The report generation and display module integrates the battery health status assessment results and safety risk assessment results to generate a structured test report, which is then visualized through a mini-program or mobile terminal interface.

2. The power battery pack health status monitoring system according to claim 1, characterized in that, The specific details of receiving the vehicle's unique identifier information input by the user include: Using WeChat mini-programs, apps, or web pages, users can obtain the unique identification information of the vehicle to be queried from the cloud by entering the vehicle's VIN code and / or uploading a photo of the vehicle registration certificate.

3. The power battery pack health status monitoring system according to claim 2, characterized in that, The specific steps of obtaining historical operation and charging / discharging data corresponding to the vehicle's unique identification information from a cloud data source include: It interfaces with the GBT 32960 standard protocol and / or the vehicle manufacturer's cloud data platform, and sends a request to the data platform based on the input VIN code and / or vehicle registration information to obtain the vehicle's historical operation and charging / discharging data. The data includes, but is not limited to: vehicle static information, basic battery parameters, charging behavior data, and driving-related data.

4. The power battery pack health status monitoring system according to claim 3, characterized in that, The process of cleaning, verifying, and reconstructing the time series of the collected multi-source historical data to construct a battery state assessment feature set specifically includes: For data fields in the GBT 32960 standard, automatic cleaning is performed by setting reasonable physical thresholds based on historical data statistics. Based on the missing data segments from multiple sources, each historical data segment is processed separately. Convert the timestamps of all data sources to UTC time and resample and align data with different sampling frequencies to the same time grid. Based on the upload timestamp, identify and delete identical duplicate data records. For conflicting records with the same timestamp but different values, retain the latest uploaded data version. Check if the timestamps of the uploaded data are strictly incrementing, and mark and reorder the data records that are reverted in time. Based on the logical relationship between historical operation and charge / discharge data parameters, multi-source historical data is verified. Numerical data in multi-source historical data is filtered and normalized, while non-numerical data in multi-source historical data is quantized using one-hot encoding. The cleaned, verified, and reconstructed time series data are organized into a unified time series data structure to construct a battery state assessment dataset. Based on the battery state assessment dataset, battery capacity change features, charge / discharge behavior features, usage intensity features, battery consistency features, and abnormal mode features are extracted to construct a battery state assessment feature set.

5. The power battery pack health status monitoring system according to claim 4, characterized in that, The construction of the battery pack health status assessment model, based on the battery status assessment feature set, specifically outputs the battery health status assessment results including: A battery pack health status assessment model is constructed, with the battery status assessment feature set as input and battery health, battery degradation value, reference range, cycle usage characteristics and safety risk coefficient as output. The cyclic use characteristics include at least: equivalent number of cycles, cycle depth distribution, and cycle decay coefficient; Based on the historical operating data of new energy vehicle power battery packs, corresponding historical operating and charge / discharge data are collected through the GBT 32960 standard protocol; Based on historical operation and charge / discharge data, combined with real health status labels obtained from laboratory standard tests, a battery status assessment feature set is extracted, a training sample set is constructed, and divided into a training set and a test set. The sample set includes battery status assessment feature sets and corresponding real health status labels for different vehicle models, different usage scenarios, and different degradation cycles. For mixed tasks involving regression and probability estimation with multiple outputs, a composite loss function is used to calculate the difference between the model's predicted values ​​and the true labels. The composite loss function consists of mean squared error loss, binary cross-entropy loss, and physical constraint penalty term, and is a weighted sum of the losses. The physical constraint penalty term introduces rule constraints based on the principles of battery electrochemistry into the loss function, and its mathematical expression includes a quantitative calculation of the degree of violation of key physical laws. With the goal of minimizing the composite loss function, an adaptive optimization algorithm is used to train the model based on the training set, and the trained model is defined as the battery pack health status assessment model. The stability and accuracy of the battery pack health status assessment model are evaluated using performance indicators such as mean absolute error, root mean square error, coefficient of determination, AUC value of risk prediction, false alarm rate, and false alarm rate.

6. The power battery pack health status monitoring system according to claim 5, characterized in that, The security risk classification module specifically includes: The historical sequence unit is used to infer the battery status assessment feature set arranged in chronological order using a trained battery pack health status assessment model, and output the safety risk coefficient corresponding to each time point, thereby generating a complete historical safety risk coefficient sequence. The sample set unit is used to construct a training sample set for the LSTM encoder based on the historical security risk coefficient sequence. Each training sample takes the data of the first N time points of the historical security risk coefficient sequence as input and the actual risk coefficients of the next M time points as labels. The prediction unit is used to obtain the latest safety risk coefficient sequence of the vehicle to be queried, set a sliding window, input it into the trained LSTM encoder, and predict the safety risk coefficient prediction values ​​for the next M time points. The grading unit is used to calculate the slope of change of the predicted risk coefficient, the predicted peak value, and the expected time exceeding the threshold based on the predicted safety risk coefficient value, and to classify the long-term safety risk of the battery into four levels: Grade A: Forecasted risk remains low and the trend is stable; Grade B: The predicted risk is rising slowly, with no short-term risk of exceeding the limit; Level C: The predicted risk is rising rapidly, with a risk of exceeding the limit in the medium term; Level D: Current risk is high and is predicted to continue to worsen; The assessment report unit is used to integrate the long-term risk level with the real-time safety risk coefficient to form a complete battery safety status assessment report.

Images