Power supply cell state diagnosis method and device, storage medium and program product

Abstract

The invention discloses a power supply cell state diagnosis method and device, a storage medium and a program product, and belongs to the technical field of battery safety management. The method comprises the following steps: acquiring internal stress, temperature, terminal voltage and charge and discharge current values of a battery unit; based on the parameters, constructing a thermal runaway symptom index through a statistical correlation algorithm; inputting the historical state parameters into a long-short term memory (LSTM) network model, and predicting a future terminal voltage value; calculating the residual error between the predicted terminal voltage and the actual terminal voltage, and carrying out anomaly detection on the residual error; and performing comprehensive state diagnosis by combining the thermal runaway symptom index and the anomaly detection result. Internal stress parameters are introduced, LSTM prediction and OC-SVM detection are combined, early degradation symptoms can be captured from mechanical and electrochemical double angles, and the accuracy and advance of thermal runaway early warning are improved.

Description

Technical Field

[0001] This application relates to the field of battery safety management technology, and in particular to a method, device, storage medium and program product for diagnosing the state of a power cell. Background Technology

[0002] As the core energy unit of various electronic devices and power supply systems, the safety, reliability and service life of power cells are key indicators for measuring product performance.

[0003] Current technologies for diagnosing battery cell condition primarily rely on the ampere-hour integration method. Its basic principle involves using a high-precision current sensor to monitor the current flowing into or out of the cell in real time, integrating the current over time to calculate the cell's current state of charge (SOC). However, the ampere-hour integration method is susceptible to factors such as current sensor sampling errors, cell self-discharge, and capacity decay due to aging, resulting in persistent cumulative errors that lead to increasingly inaccurate capacity estimations over time. Its limitation lies in tracking only the inflow and outflow of charge, neglecting the complex physicochemical processes that cause cell aging. The decline in the cell's state of health (SOH) is the result of cumulative stress on its internal materials under long-term, complex operating conditions, including different temperatures, charge / discharge rates, and depths of discharge.

[0004] Therefore, this method of state assessment can lead to misjudgments of the long-term performance evolution trend of the battery cell, ultimately resulting in low accuracy of the predicted state of health (SOH) of the battery cell. Summary of the Invention

[0005] This application provides a method, device, storage medium, and program product for diagnosing the condition of power cell cells, which can improve the accuracy of power cell condition assessment, the foresight of safety warnings, and the refinement of life cycle management, thereby comprehensively enhancing the safety, reliability, and service life of battery systems.

[0006] In a first aspect, this application provides a power cell state diagnosis method, applied to a power cell state diagnosis system. The method includes: collecting real-time operating data from the battery cell within a preset sampling period, the real-time operating data including terminal voltage, charging / discharging current, and internal stress values ​​of the cell; after the battery cell completes one preset charge / discharge cycle, calculating the difference between the internal stress value of the cell at the end of the current cycle and the internal stress value of the cell at the beginning of the current cycle, obtaining the irreversible stress increment; constructing an input sequence sample with a preset sequence length from the real-time operating data and the irreversible stress increment based on a preset time step and sliding window; and transmitting the input sequence sample to the battery cell. The input sequence samples are fed into a pre-built Long Short-Term Memory (LSTM) network model. The LSM network model includes at least one LSM network layer configured to output a single hidden state vector only at the end of the sequence, and an output layer connected to the LSM network layer, containing three neurons and a linear activation function. The LSM network layer is controlled to process the input sequence samples and generate a final hidden state vector, which encodes the temporal dependence of the input sequence samples. The final hidden state vector is sent to the output layer, which calculates and outputs the corrected health state value, the corrected state of charge value, and the predicted terminal voltage value for the next sampling period.

[0007] This embodiment introduces the "internal stress value" and its "irreversible stress increment," which directly characterize the cumulative damage of the battery cell, as key input features, and utilizes a Long Short-Term Memory (LSTM) network model to process the input sequence containing these features. Through its unique gating structure, this model can capture the nonlinear dependencies of multidimensional data such as voltage, current, and stress over long time scales, and encode these relationships in the final hidden state vector. This scheme expands the diagnostic dimension from simple charge measurement to the quantitative tracking of internal aging mechanisms, and a single output layer collaboratively outputs corrected health and state of charge values. This not only provides a health status assessment that more closely reflects the actual physical degradation of the battery cell, but also uses health status information to reverse-calibrate the estimated state of charge, thereby improving the accuracy of the entire battery management system's state perception.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the method further includes: calculating the core temperature rise rate based on multiple core temperature values ​​of the battery cells collected within a preset time window; using a statistical correlation algorithm to calculate the correlation coefficient between the core temperature rise rate and the charge / discharge current value within the preset time window to obtain a thermal runaway symptom index; inputting the thermal runaway symptom index into a pre-trained single-class support vector machine model, the single-class support vector machine model including a hyperplane boundary for defining the normal operating state of the battery cell; determining whether the data point position of the thermal runaway symptom index in the single-class support vector machine model crosses the hyperplane boundary, and if the determination result is yes, then marking the state of the battery cell as a fault warning state and generating a first alarm command.

[0009] This embodiment constructs a "thermal runaway symptom index" by calculating the core temperature rise rate and analyzing its statistical correlation with charge / discharge current values. This index reflects the relationship between heat generation and electrical load within the battery cell. Subsequently, a pre-trained single-class support vector machine model is used to classify this index. This model defines a complex normal behavior boundary based on a large amount of normal operating data. This approach elevates the judgment criterion from a static threshold to a dynamic behavioral model. By identifying whether there is an abnormal mismatch between the heat generation rate and electrical load, it can capture the initial signs of thermal runaway earlier, before the battery cell temperature reaches a dangerous threshold. This improves the reliability and foresight of fault warnings while avoiding false alarms caused by normal operating conditions.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, after sending the final hidden state vector to the output layer, and having the output layer calculate and output the corrected health state value and the corrected state of charge value, the method further includes: recording the operating mode as charging mode when the charging / discharging current value is positive, recording it as discharging mode when the charging / discharging current value is negative, and recording it as standby mode when the charging / discharging current value is zero; retrieving and retrieving a set of initial PID control parameters from a preset control parameter library based on the operating mode, the corrected health state value, the corrected state of charge value, and the fault warning state; and setting the battery cell... The core temperature value is compared with a preset upper limit temperature threshold. When the core temperature value of the battery cell exceeds the upper limit temperature threshold, the target charging or discharging current limit is calculated using the initial PID control parameters, and the battery management system is controlled to reduce the charging or discharging current to the target current limit by adjusting the pulse width modulation signal. The corrected state of charge value is compared with a preset constant current charging cutoff state of charge threshold. When the corrected state of charge value is less than the threshold, the battery management system is controlled to execute a constant current constant voltage charging strategy. When the corrected state of charge value is greater than or equal to the threshold, the battery management system is controlled to execute a trickle charging strategy.

[0011] This embodiment constructs a state-aware closed-loop control mechanism. This mechanism uses the "corrected health status value," "corrected state of charge value," and "fault warning status" obtained in previous steps as decision-making basis. It dynamically retrieves and calls up a set of initial PID control parameters from a preset control parameter library that match the current overall state of the battery cell, and adjusts the charging strategy and the current limit for over-temperature protection accordingly. This makes charge and discharge control no longer a rigid execution based on a preset script, but rather able to match the actual health and safety risks of the battery cell in real time. In this way, the system can perform more refined management under different aging stages and operating conditions, thereby ensuring operational safety while helping to delay the performance degradation of the battery cell.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the method further includes: setting a maximum allowable charge / discharge power value based on a modified health status value, and limiting the output power and input power of the battery cell.

[0013] This embodiment links the corrected health status value directly to the power management strategy, dynamically setting the maximum allowable charge and discharge power value. This allows the power boundary to adaptively adjust according to the actual health status of the battery cell. In the early stages of use, the battery cell is allowed to realize its performance potential, while in the later stages of aging, the power output is automatically reduced to avoid the risk of overheating or abnormal voltage caused by increased internal resistance. This improves the safety of the battery throughout its entire life cycle and achieves a more rational use of battery performance.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the method further includes: continuously recording the actual change trajectory of the core temperature value of the battery cell and the corrected state of charge value; after the preset adjustment period ends, calculating the difference between the actual change trajectory and the preset ideal change trajectory to obtain a performance deviation value; inputting the performance deviation value into a machine learning optimization algorithm, which outputs a set of PID parameter correction values, and using the PID parameter correction values ​​to update the initial PID control parameters in the control parameter library.

[0015] This embodiment introduces a closed-loop optimization mechanism to compare the "actual change trajectory" after control execution with the "ideal change trajectory" to quantify the "performance deviation value." The deviation value is input into a machine learning optimization algorithm to back-calculate the correction amount for the PID parameters, and this is used to continuously update the control parameter library. This solution endows the control system with online self-learning and self-optimization capabilities, enabling the control parameters to start from a "relatively good initial value" and gradually converge to an optimal solution that better reflects the current true characteristics of the battery cell through continuous learning and fine-tuning, thereby improving the system's control accuracy and robustness throughout its entire lifecycle.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after inputting the thermal runaway symptom index into a pre-trained single-class support vector machine model, the method further includes: controlling a long short-term memory network model to predict the terminal voltage value of the next sampling period based on historical input sequence samples, and generating a predicted terminal voltage value; extracting the terminal voltage value of the same sampling period as the predicted terminal voltage value from real-time operating data as the actual terminal voltage value; subtracting the actual terminal voltage value from the predicted terminal voltage value to calculate the voltage prediction residual; collecting the voltage prediction residuals of multiple consecutive sampling periods within a preset residual analysis window to construct a voltage prediction residual sequence; calculating the mean of the voltage prediction residual sequence to generate a residual mean; inputting the thermal runaway symptom index and the residual mean together into a preset logic judgment module, the logic judgment module having built-in judgment rules for defining fault warning conditions; the logic judgment module, according to the judgment rules, determining whether the combination of the thermal runaway symptom index and the residual mean satisfies the fault warning condition, and if the determination result is yes, marking the state of the battery cell as a fault warning state and generating a second alarm command.

[0017] This embodiment introduces a parallel predictive analysis dimension, utilizes a long short-term memory network model to predict the terminal voltage in a time series, and quantifies the deviation between the predicted and actual values ​​(i.e., the voltage prediction residual), thereby constructing a new index characterizing the degree of abnormality in the electrical behavior of the battery cell. This approach is not a simple replacement, but rather a joint judgment using the "thermal runaway symptom index," representing physical degradation, and the "mean value of voltage prediction residual," representing abnormal electrical behavior. This judgment mechanism based on multi-source information fusion, through cross-validation of two indices with different mechanisms, can detect subtle deviations in electrical characteristics caused by factors such as internal short circuits before physical symptoms appear, more early. Simultaneously, the confirmation by dual indices reduces the possibility of false alarms due to fluctuations in a single data source, improving the overall reliability and timeliness of the fault early warning system.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the battery cell includes multiple individual cells connected in series. The method further includes: for each individual cell, obtaining its corresponding corrected state of charge (SOC) value; when it is detected that the difference between the corrected SOC value of any individual cell and the corrected SOC value of other individual cells exceeds a preset imbalance threshold, activating an active balancing circuit; controlling the active balancing circuit to transfer energy from the individual cell with a higher SOC value to the individual cell with a lower SOC value in the same battery cell through energy transfer, until the difference between the corrected SOC values ​​of multiple individual cells is less than a preset balancing target threshold.

[0019] This embodiment utilizes a corrected state of charge (SOC) value, incorporating aging characteristics such as internal stress, as the basis for balancing judgment. This enhances the reliability of balancing triggering and target setting, improving the problem of balancing failure or exacerbated cell inconsistency caused by inaccurate SOC estimation. Simultaneously, by activating the active balancing circuit for energy transfer, energy from cells with higher SOC values ​​is transferred to cells with lower SOC values, promoting a more consistent SOC among the individual cells within the battery cell. This consistency prevents some cells from remaining on the verge of overcharging or over-discharging due to SOC differences during charging and discharging, mitigating performance degradation caused by the bottleneck effect in the entire battery cell and increasing the usable capacity of the battery cell throughout its lifespan.

[0020] In a second aspect, embodiments of this application provide a power cell status diagnostic device, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the power cell status diagnostic device to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a power cell status diagnostic device, cause the power cell status diagnostic device to execute the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a power cell status diagnostic device, cause the power cell status diagnostic device to perform the method described in the first aspect and any possible implementation thereof.

[0023] It is understood that the power cell status diagnostic device provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. This application uses "internal stress value" and "irreversible stress increment," which directly characterize the cumulative damage of the battery cell, as key input features. It also utilizes a long short-term memory network model to deeply mine the nonlinear dependencies of battery cell operating data over long time scales, achieving coordinated assessment and correction of the state of health (SOH) and state of charge (SOC). This expands the diagnostic dimension from simple charge measurement to the quantitative tracking of internal aging physical mechanisms, thereby obtaining state assessment results that more closely reflect the actual physical degradation of the battery cell and improving the accuracy of overall state perception.

[0026] 2. This application introduces a parallel predictive analysis dimension, jointly judging the "thermal runaway symptom index" representing the physical degradation trend of the battery cell and the "mean value of voltage prediction residuals" representing abnormal electrical behavior, thus constructing a fault early warning mechanism based on multi-source information fusion. Through cross-validation of two indicators with different mechanisms, it can capture weak fault symptoms caused by factors such as internal short circuits earlier, and reduce the possibility of false alarms due to fluctuations in a single data source, thereby improving the reliability and timeliness of fault early warning.

[0027] 3. This application constructs a state-aware closed-loop control and self-optimization mechanism. On the one hand, it utilizes the accurate state assessment results obtained in the preceding steps to dynamically match control parameters from the parameter library that are adapted to the current cell health and safety risks. On the other hand, it uses machine learning optimization algorithms to continuously correct the parameter library based on the deviation between the actual operating trajectory and the ideal trajectory. This scheme enables the charge and discharge control strategy to start from a "relatively good initial value" and gradually converge to the optimal solution that better reflects the current real characteristics of the cell through online learning and fine-tuning. Thus, while ensuring operational safety, it optimizes the performance of the battery throughout its entire life cycle. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating a power cell status diagnosis method in an embodiment of this application;

[0029] Figure 2 This is a flowchart illustrating a power cell fault early warning method in an embodiment of this application;

[0030] Figure 3 This is a flowchart illustrating a power cell optimization control method in an embodiment of this application.

[0031] Figure 4 This is a flowchart illustrating a power cell fault diagnosis method in an embodiment of this application.

[0032] Figure 5 This is a schematic diagram of the physical device structure of a power cell status diagnostic device in the embodiments of this application. Detailed Implementation

[0033] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0034] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0035] To facilitate understanding, the application scenarios of the embodiments of this application are described below.

[0036] As the core component of modern electronic devices and energy storage systems, the accurate diagnosis of the state of power cells, especially the accurate assessment of their state of charge (SOC) and state of health (SOH), is directly related to the safety, reliability and service life of the system.

[0037] In related technologies, the ampere-hour integration method is commonly used to estimate the state of charge (SOC) of a battery cell. This method tracks the change in charge by monitoring and integrating the current flowing through the cell in real time. However, sampling errors of the current sensor in the ampere-hour integration method, cell self-discharge, and capacity decay during aging all introduce cumulative errors that are difficult to eliminate, causing the SOC estimate to become inaccurate over time. Furthermore, this method only stays at the macroscopic measurement level of charge inflow and outflow, ignoring the complex physicochemical processes that lead to cell aging, namely the cumulative stress on the internal materials of the cell under long-term, variable operating conditions. Therefore, relying solely on the ampere-hour integration method for state assessment cannot accurately capture the true degradation trajectory of the cell's state of health (SOH), ultimately leading to misjudgments of the long-term performance evolution trend of the cell and low accuracy in SOH prediction.

[0038] This application provides a method for diagnosing the condition of power supply cells.

[0039] The power cell status diagnosis method described in this application is applied to a power cell status diagnosis system. The hardware architecture of this system mainly includes a power cell assembly, a multi-dimensional sensing module, and a main controller.

[0040] The power cell assembly, as the monitored object, can be composed of one or more battery cells (e.g., 18650 or pouch lithium-ion batteries) connected in series and parallel. The multi-dimensional sensing module is the data acquisition foundation for realizing the technical solution of this application. Its design goal is to simultaneously acquire key parameters characterizing the physical and electrochemical states of the cell. To obtain the "internal stress value of the cell," which directly reflects the mechanical fatigue and cumulative damage within the cell, this application embeds or attaches a high-sensitivity thin-film composite sensor at key locations in the cell assembly (e.g., the core area of ​​the cell winding or the surface of the casing). This sensor can be up to the micrometer level in thickness to ensure that its impact on the original structure and thermal field distribution of the cell is minimized. It integrates a high-sensitivity strain gauge and a thermistor, enabling simultaneous and point-to-point measurement of the minute strain of the electrode material caused by lithium-ion insertion / extraction (used to convert into stress values) and the core temperature of the cell. This point-to-point measurement eliminates errors caused by spatial differences, providing a data foundation for subsequent analysis of the "thermal-mechanical coupling" effect.

[0041] Simultaneously, the system is equipped with a high-precision differential voltage acquisition unit. This unit can independently and at high resolution (e.g., 16-bit or higher) sample the "terminal voltage value" between the positive and negative terminals of each battery cell in the series-connected cell group. High-resolution measurement is crucial for capturing millivolt-level voltage anomalies caused by internal micro-short circuits or early aging, and is a prerequisite for accurate prediction and voltage residual analysis by the subsequent LSTM model (as described in claim 6). It also includes a wide dynamic range, high-precision bidirectional current monitoring unit (e.g., based on a high-precision shunt and a dedicated fuel gauge chip such as the INA226). This unit is responsible for accurately measuring the "charge / discharge current value" flowing into (charging) or out of (discharging) the battery cell. Its characteristics include the ability to cover leakage currents ranging from microamperes in standby mode to tens of amperes during fast charging and discharging, while maintaining high linearity and low drift across the entire range. This is not only a key input for constructing the "thermal runaway symptom index," but also provides a reliable charge measurement benchmark for the collaborative correction of the state of charge (SOC) in this scheme.

[0042] All the aforementioned physical and electrical sensing data streams ultimately converge to a high-performance main controller (such as an STM32 series microcontroller) that serves as the system's "brain." This main controller possesses sufficient computing power and storage space to embed and run the core software modules described in this application, such as the Long Short-Term Memory (LSTM) network model, the One-Class Support Vector Machine (OC-SVM) model, and the PID control algorithm, in real time, and issue control commands to the charge and discharge management system based on the calculation results.

[0043] Through the coordinated operation of the aforementioned hardware systems, this application is able to obtain a high-fidelity real-time data stream containing multiple dimensions such as stress, temperature, voltage, and current, providing a solid data foundation for subsequent precise condition diagnosis, proactive fault warning, and adaptive closed-loop control.

[0044] To facilitate understanding, the method provided in this implementation will be described in detail below, using the above scenario as an example. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating a power cell status diagnosis method in an embodiment of this application.

[0045] S101. Within a preset sampling period, collect real-time operating data from the battery cell. The real-time operating data includes terminal voltage value, charging and discharging current value, and internal stress value of the cell.

[0046] Real-time operational data refers to a set of parameters synchronously collected at each sampling moment, which can fully characterize the current physical and electrical state of the battery cell, and is used to provide basic input for subsequent time-series analysis models. Terminal voltage value represents the actual potential difference between the positive and negative output terminals of the battery cell under specific load or charging conditions, measured in volts (V). It is a comprehensive electrical indicator reflecting the electrochemical potential, polarization, and ohmic voltage drop within the cell. Charging and discharging current value refers to the amount of current flowing through the main circuit of the battery cell, measured in amperes (A), used to represent the rate and direction of energy exchange within the cell. In this application, charging current is defined as positive and discharging current as negative. The internal stress value of the battery cell is a key physical quantity used to represent the mechanical stress generated inside the battery cell (such as the electrode, separator, and current collector interface) due to the lattice expansion / contraction caused by the insertion and extraction of lithium ions in the electrode active material, as well as the thermal expansion and contraction effect caused by temperature changes. It can be measured by special sensors (such as fiber Bragg grating sensors or high-sensitivity strain gauges) that are pre-integrated inside the battery cell or attached to the surface of the battery cell. Its value is directly related to the mechanical fatigue and cumulative damage of the battery cell.

[0047] S102. After the battery cell completes one preset charge-discharge cycle, calculate the difference between the internal stress value of the cell at the end of the current cycle and the internal stress value of the cell at the beginning of the current cycle to obtain the irreversible stress increment.

[0048] The trigger condition for this step is that the Battery Management System (BMS) detects that the battery cell has completed a preset charge-discharge cycle. Specifically, the BMS first needs to identify the start and end points of the cycle. For example, the system continuously records the state of charge (SOC) of the cell. When it detects that the SOC has dropped from a preset upper limit (e.g., 98%) and then recharged back to that upper limit, it determines that a cycle is complete. Subsequently, the system accurately retrieves the internal stress values ​​(σ_start and σ_end) of the cell corresponding to the start time (T_start) and end time (T_end) of this cycle from historical data records. It is important to note that, to ensure the validity of the comparison, these two stress values ​​should be measured under similar temperature and SOC conditions, or corrected using a preset compensation model to eliminate reversible stress components caused by temperature and SOC changes. Then, a subtraction operation is performed: Δσ_irreversible = σ_end - σ_start, to obtain the irreversible stress increment generated during this cycle.

[0049] S103. Based on the preset time step and sliding window, construct an input sequence sample with a preset sequence length from real-time running data and irreversible stress increments.

[0050] The trigger condition for this step is that the system has collected a sufficient amount of real-time operational data and has calculated at least one irreversible stress increment. Specifically, the data processing module first reads the most recent real-time operational data cache from the memory. Then, based on the preset time step and sequence length, a sliding window mechanism is used to truncate the data. For example, assuming the sequence length is L and the sliding step is S, at time t, the window will truncate the data points from t-L+1 to t. For each truncated data point, the system extracts three dynamic features: its terminal voltage value, charging / discharging current value, and internal cell stress value. Simultaneously, the system obtains the most recently calculated "irreversible stress increment" value and uses it as a static feature, copying and appending it to the data of all L time points within the window. Thus, each input sequence sample becomes an Lx4 matrix (L time points, each containing 4 features). A new input sequence sample is generated each time the window slides forward by S steps. The expected outcome of this step is to transform the original, continuous data stream into a series of discrete, fixed-length samples containing multi-dimensional dynamic and static features. These samples have a uniform format and can be directly fed into the Long Short-Term Memory network model for processing.

[0051] S104. Input the input sequence sample into a pre-built long short-term memory network model. The long short-term memory network model includes at least one long short-term memory network layer configured to output a single hidden state vector only at the end of the sequence, and an output layer connected to the long short-term memory network layer, which contains three neurons and a linear activation function.

[0052] Specifically, the BMS main control chip or dedicated AI accelerator loads the input sequence sample matrix generated by the S103 into memory and initiates the forward propagation computation of the LSTM model. The data is first fed into the LSTM layer. The LSTM layer processes each data point in the sequence (a vector containing four features) sequentially over time steps. At each time step, the LSTM unit updates the cell state (long-term memory) and hidden state (short-term memory) based on the current input, the hidden state from the previous time step, and the cell state, using its internal gating mechanism. This process proceeds sequentially along the sequence length (e.g., 60 time steps). Because the LSTM layer is configured to output only at the end of the sequence, the calculated hidden state is only propagated internally within the layer during the first 59 time steps and is not output externally. The LSTM layer only outputs its final calculated hidden state vector after processing the data from the 60th time step. This step utilizes the temporal information processing capabilities of the LSTM layer to compress and encode an input sequence containing complex time dependencies into a fixed-dimensional, highly condensed feature vector, laying the foundation for subsequent accurate multi-task prediction.

[0053] This approach employs a specific choice regarding the LSTM layer output mode: instead of outputting the hidden states of the entire sequence, it utilizes only the final hidden state at the end of the sequence. The logic behind this design is that the goal is not to label every point in the sequence, but rather to provide a comprehensive evaluation of the "current state" represented by the entire historical window and to predict the "next moment." Therefore, using only the final hidden state as a "summary" of the entire sequence is a more efficient and purposeful approach, forcing the model to learn how to integrate all historical information into a single, highly efficient representation. Simultaneously, there is a precise correspondence between the output layer neurons and the prediction task. The output layer is precisely designed to contain three neurons, directly corresponding to the three highly coupled yet distinct core tasks: the "corrected health state value," the "corrected state of charge value," and the "predicted terminal voltage value for the next sampling period." This is not a simple multi-objective output, but rather based on a profound insight: SOH, SOC, and future voltage are an inseparable "trinity" describing the battery state, influencing and constraining each other.

[0054] S105. Control the Long Short-Term Memory network layer to process the input sequence samples and generate the final hidden state vector. The final hidden state vector encodes the time dependency of the input sequence samples.

[0055] Specifically, when the input sequence sample (an Lx4 matrix) is fed into the LSTM layer, the computation process begins at the first time step (t=1) and ends at the last time step (t=L). At each time step t, the LSTM unit executes the following internal logic: 1) The forget gate examines the cell state c_(t-1) from the previous time step and decides which long-term memories should be forgotten; 2) The input gate analyzes the current input x_t (including voltage, current, stress, and irreversible stress increments) and decides which new information is important and should be added to memory; 3) Cell state update: combining forgetting and input decisions, the old cell state c_(t-1) is updated to the new cell state c_t, completing the long-term memory update; 4) The output gate, based on the new cell state c_t, decides what to output as the current hidden state h_t. This h_t will be used as one of the inputs for the next time step t+1. This process repeats continuously, with information constantly being transferred and evolving along the time axis. When the sequence is processed to the last time step t=L, the generated hidden state h_L is the "final hidden state vector", which encapsulates the key temporal information from t=1 to t=L and is then output to the next layer.

[0056] S106. The final hidden state vector is sent to the output layer, which calculates and outputs the corrected health state value, the corrected state of charge value, and the predicted terminal voltage value for the next sampling period.

[0057] Specifically, when the final hidden state vector (a D_h-dimensional vector, where D_h is the number of hidden units in the LSTM layer) is input to the output layer, a matrix multiplication and bias addition operation are performed internally within the output layer. This output layer contains a weight matrix W (dimension D_h x 3) and a bias vector b (dimension 1 x 3). The calculation process is [output 1, output 2, output 3] = [final hidden state vector] * W + b. Since this layer contains three neurons and uses a linear activation function, the first neuron calculates and outputs the corrected SOH value, the second neuron calculates and outputs the corrected SOC value, and the third neuron calculates and outputs the predicted terminal voltage value for the next sampling period. The linear activation function (i.e., f(x) = x) ensures that the output can be directly used as a continuous regression value without nonlinear compression, preserving the original scale and physical meaning of the values. This step decodes an abstract feature vector containing rich temporal information into three core state parameters that have direct application value to BMS.

[0058] Having achieved accurate diagnosis of cell state of health (SOH) and state of charge (SOC) through the above embodiments, this application further considers the issue of early warning of sudden safety risks to battery cells, especially for the critical failure mode of thermal runaway. In another embodiment, this application proposes a forward-looking fault warning method by constructing a new index that reflects the "electro-thermal coupling" characteristics inside the cell and using a machine learning model for anomaly detection.

[0059] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is a flowchart illustrating a power cell optimization control method in an embodiment of this application.

[0060] S201. Calculate the core temperature rise rate based on the core temperature values ​​of multiple battery cells collected within a preset time window.

[0061] This step is executed periodically, for example, once per second, to update the assessment of battery thermal dynamics in real time. Specifically, the BMS processor first retrieves all core temperature sampling points within the most recent "preset time window" from memory. Assuming the time window is Δt and the sampling frequency is f, n = Δt * f temperature data points (t_1, T_1), (t_2, T_2), ..., (t_n, T_n) will be obtained. To calculate a stable "core temperature rise rate" that is resistant to single-point measurement noise, this scheme preferably uses a linear regression method. The system fits these n data points into a straight line T(t) = a * t + b, where a is the slope of the line. This slope a, calculated using the least squares method, is defined as the core temperature rise rate within the current time window. This method can effectively smooth out individual noise points caused by sensor fluctuations or electromagnetic interference, extract the main temperature change trend throughout the time window, and thus obtain a rate value that is more robust and reliable than the simple two-point difference method.

[0062] S202. Using a statistical correlation algorithm, calculate the correlation coefficient between the core temperature rise rate and the charge / discharge current value within a preset time window to obtain the thermal runaway symptom index.

[0063] This step is executed synchronously with S201, performing periodic calculations to continuously monitor the evolution of the battery's "electro-thermal coupling" state. Specifically, the system first needs to acquire two synchronized data sequences: one is a single "core temperature rise rate" value representing the trend of the entire time window (calculated by S201 and can be considered a constant R_T); the other is a sequence of n discrete "charge and discharge current values" {I_1, I_2, ..., I_n} collected within the same time window. To calculate the correlation, a better approach is to decompose the temperature rise rate in S201 into instantaneous rates. That is, the system first calculates the instantaneous temperature rise rate r_i = (T_i - T_{i-1}) / Δt_s (Δt_s is the sampling period) for each sampling point t_i within the time window. This results in two synchronized sequences of length n-1: the instantaneous temperature rise rate sequence {r_2, ..., r_n} and the instantaneous current absolute value sequence {|I_2|, ..., |I_n|}. Then, the classic Pearson correlation coefficient algorithm is used to calculate the correlation coefficient between the two sequences. The calculation formula is: ρ = Cov(r, |I|) / (σ_r * σ_{|I|}), where Cov(r, |I|) is the covariance of the two sequences, and σ_r and σ_{|I|} are their respective standard deviations. The calculated Pearson correlation coefficient ρ is directly assigned as the "thermal runaway symptom index". The expected effect is to accurately quantify the battery's response sensitivity to current load through a single, normalized value.

[0064] The aforementioned statistical correlation algorithms encompass various mathematical statistics methods capable of achieving the above mapping modeling. Those skilled in the art can select the appropriate specific algorithm based on the accuracy requirements, computational resource constraints, and model interpretability needs in practical applications.

[0065] For example, in a simpler implementation, the Multiple Linear Regression algorithm can be used. This method aims to find a set of optimal weight coefficients to construct a linear equation that directly maps the linear combination of input parameters such as voltage, current, temperature, and stress to the thermal runaway symptom index. The advantages of this method are its simplicity, speed, and strong model interpretability. In another implementation, to capture the potentially complex nonlinear relationships between the parameters and the thermal runaway symptom index, the Support Vector Regression (SVR) algorithm can be used. SVR maps the data to a high-dimensional feature space through a kernel function and searches for a "hyperplane" in this space that best fits the data, demonstrating good performance in handling high-dimensional and highly nonlinear problems.

[0066] It is understood that other methods may also be used, and this application does not impose specific restrictions on them.

[0067] S203. Input the thermal runaway symptom index into a pre-trained single-class support vector machine model, which includes a hyperplane boundary used to define the normal operating state of the battery cell.

[0068] Specifically, the processor in the BMS takes the "thermal runaway symptom index" output by S202 as a new data point and inputs it into the decision function of the OCSVM model, which is already embedded in the BMS firmware. This decision function essentially calculates the directed distance between the new data point and the pre-learned "hyperplane boundary." Since the input feature in this embodiment is one-dimensional (only the thermal runaway symptom index), the "hyperplane boundary" is mathematically represented as the endpoints of one or more "normal intervals." For example, the model may learn through training that all index values ​​in normal states fall within the interval [0.75, 0.98]. The task of the decision function is then to determine whether the newly input index value is within this "safe interval." If the index value is 0.85, the decision function will return a specific value indicating "inside" (e.g., +1); if the index value is 0.5, exceeding the lower limit of the interval, the decision function will return a specific value indicating "outside" (e.g., -1).

[0069] S204. Determine whether the data point position of the thermal runaway symptom index in the single-class support vector machine model crosses the hyperplane boundary. If the determination result is yes, mark the state of the battery cell as a fault warning state and generate the first alarm command.

[0070] Specifically, the BMS processor receives the output value returned by the OCSVM decision function in S203. The system has a pre-defined judgment logic: if the return value equals a specific value representing "external / abnormal" (e.g., -1), the judgment condition "yes" is met. Once the judgment is met, the BMS immediately executes two parallel core operations: First, it updates the internal state machine, changing the health status flag (StatusFlag) of the specific battery cell from "Normal" to "Pre-fault_Warning". This status flag is the basis for all subsequent control strategies. Second, it generates and broadcasts an alarm command. This command is sent to other controllers via the CAN (Controller Area Network) bus. The command content can be designed as a message containing multiple layers of information, such as: alarm level (e.g., Level 1 / Severe Warning), the unique identifier (ID) of the faulty cell, the current "thermal runaway symptom index" value, and related snapshot data such as temperature and current. For example, the dashboard displays "Power battery system fault, please stop immediately for inspection," while the vehicle information system pushes a warning message to the user's mobile app. The expected outcome is to achieve a closed loop from "internal symptom detection" to "effective external response," ensuring that once a definite early risk is detected, multi-level and multi-dimensional safety measures can be immediately initiated, buying valuable time for personnel evacuation and risk control.

[0071] The above embodiments achieve accurate diagnosis of cell status and reliable early warning of potential faults. How to transform this information into effective control actions to achieve optimized management of the battery's entire lifecycle is a further problem to be solved. In another embodiment, this application proposes a state-aware closed-loop control mechanism that uses the state information obtained from previous steps as decision input to dynamically adjust the charging and discharging strategy.

[0072] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 3 This is a flowchart illustrating a power cell optimization control method in an embodiment of this application.

[0073] S301. When the charging / discharging current value is positive, the operating mode is recorded as charging mode; when the charging / discharging current value is negative, it is recorded as discharging mode; when the charging / discharging current value is zero, it is recorded as standby mode.

[0074] Specifically, the BMS's microcontroller (MCU) reads the current sensor's (such as a shunt or Hall sensor) value after analog-to-digital conversion (ADC). To improve the stability of state judgment, instead of simply comparing it to "zero," a small current dead-band is set, for example, [-50mA, +50mA]. When the current value is greater than +50mA, the system sets a predefined flag or register value to "charging mode"; when the current value is less than -50mA, it is set to "discharging mode"; and when the current value falls within the [-50mA, +50mA] dead-band, it is set to "standby mode." This threshold-based judgment logic effectively filters out the influence of sensor noise and the system's static, small power consumption, preventing frequent and invalid transitions in operating mode near critical points. The expected effect of this step is to provide the system with a clear, stable, and unambiguous label of the current operating mode, serving as the basis for subsequent multi-dimensional and complex control decisions.

[0075] S302. Based on the operating mode, the corrected health status value, the corrected state of charge value, and the fault warning status, retrieve and call up a set of initial PID control parameters from the preset control parameter library.

[0076] This step is executed before the system determines that power adjustment is needed (e.g., current limiting due to overheating) and activates the PID controller. Specifically, the BMS processor uses the four most recent state variables—"operating mode," "corrected SOH," "corrected SOC," and "fault warning status"—as query keywords. The control parameter library can logically be viewed as a four-dimensional matrix. To implement this within limited storage space, a discretization approach is typically used. For example, SOH can be divided into multiple intervals (e.g., 100%-95%, 95%-90%, etc.), and the same applies to SOC. The BMS first maps continuous SOH and SOC values ​​to their respective discrete intervals, then combines these four discretized indices to form a unique address, directly reading the pre-stored P, I, and D values ​​from the parameter library. For example, the system might retrieve a specific set of {P1, I1, D1} parameters corresponding to [charging mode, SOH interval (90-95%), SOC interval (70-80%), no warning]. This process is rapid and deterministic, achieving a direct mapping from the battery's "comprehensive health profile" to "optimal control strategy parameters." The expected result is to provide the PID controller with an "optimal starting point" that closely matches the current operating conditions, avoiding the use of fixed, "one-size-fits-all" PID parameters.

[0077] S303. Compare the core temperature value of the battery cell with the preset upper limit temperature threshold. When the core temperature value of the battery cell exceeds the upper limit temperature threshold, use the initial PID control parameters to calculate the target charging or discharging current limit and control the battery management system to reduce the charging or discharging current to the target current limit by adjusting the pulse width modulation signal.

[0078] This step is applied to online thermal management and active safety protection of batteries during operation (especially during high-power charging and discharging). It is executed immediately upon the occurrence of an over-temperature event. Specifically, once triggered, the PID control algorithm within the BMS is activated. This algorithm takes the temperature deviation (Error = core cell temperature - upper limit temperature threshold) as input. Then, using the "initial PID control parameters" {Kp, Ki, Kd} provided by S302, it calculates a control output based on the PID control law (control output = Kp * Error + Ki * ∫Error * dt + Kd * dError / dt). This output is converted and defined as the value that needs to be reduced from the current current limit, thus obtaining the "target current limit" (target current limit = current maximum allowable current - control output). Subsequently, the BMS main controller sends this target current limit to the corresponding power execution unit (such as the charger) via the CAN bus. Upon receiving the instruction, the controller of the power execution unit immediately adjusts the duty cycle of the PWM drive signal of its internal power switching transistor (such as a MOSFET). For example, it reduces the duty cycle to decrease the output current until the actual charging and discharging current smoothly decreases and stabilizes below the new "target current limit." The purpose of this step is to achieve closed-loop negative feedback regulation of the battery current, precisely controlling the core temperature near a safe threshold. This avoids the risk of overheating and prevents functional interruption caused by a simple "one-size-fits-all" current cutoff.

[0079] S304. Compare the corrected state of charge value with the preset constant current charging cutoff state of charge threshold. When the corrected state of charge value is less than the threshold, control the battery management system to execute the constant current and constant voltage charging strategy. When the corrected state of charge value is greater than or equal to the threshold, control the battery management system to execute the trickle charging strategy.

[0080] Specifically, when the battery is in charging mode, the BMS's microcontroller (MCU) executes this comparison logic in each control cycle. When it determines that the "corrected SOC value" has not yet reached the "constant current charging cutoff state of charge threshold" (e.g., current SOC is 85%, threshold is 95%), it sends a command to the charger to execute the standard CC-CV charging procedure. The charger operates according to the maximum charging current and charging cutoff voltage set by the BMS; this is the main charging phase. Once the MCU detects that the "corrected SOC value" has reached or exceeded the threshold (e.g., current SOC is 95.1%), the BMS immediately sends a new control command to the charger, requiring it to switch to "trickle charging mode". In this mode, the charger no longer maintains a constant voltage but instead charges the battery intermittently or continuously with a small current (e.g., C / 20 or less) preset by the BMS and much smaller than the CC phase current, until 100% SOC is reached or the charging is terminated by other conditions (such as charging time). The purpose of this step is to ensure that the battery is fully charged safely and gently while maintaining charging efficiency, and to suppress aging phenomena such as lithium deposition that may be caused by continuous high voltage at the end of the charging process.

[0081] In some other embodiments of this application, the method further includes setting a maximum allowable charge / discharge power value based on the modified health status value, and limiting the output power and input power of the battery cell.

[0082] Specifically, the BMS internally stores one or more pre-calibrated "Power-SOH" two-dimensional or multi-dimensional lookup tables, which are preset at the battery manufacturing stage. When the BMS obtains the latest "corrected SOH value" (e.g., 85%), it uses this value as an index to look up the table and obtain the corresponding "maximum allowable charging power" (e.g., reduced from the rated 120kW to 90kW) and "maximum allowable discharging power" (e.g., reduced from the rated 150kW to 110kW). These two newly calculated power limits override the original settings in the BMS, becoming the new global power ceiling. During subsequent driving or charging, any power request will first be "approved" by the BMS. If the requested power exceeds the maximum allowable value corresponding to the current SOH, the BMS will not approve the request, but will instead issue a reduced power command to the execution end (inverter or charger) that does not exceed the upper limit.

[0083] In some preferred embodiments, the method further includes continuously recording the actual change trajectory of the core temperature value and the corrected state of charge value of the battery cell; after the preset adjustment period ends, calculating the difference between the actual change trajectory and the preset ideal change trajectory to obtain the performance deviation value; inputting the performance deviation value into the machine learning optimization algorithm, which outputs a set of PID parameter correction values, and using the PID parameter correction values ​​to update the initial PID control parameters in the control parameter library.

[0084] This step is applied in the background self-learning and self-optimization process of the battery management system (BMS), which is a non-real-time, long-cycle strategic adjustment. Specifically, during battery operation (e.g., a fast charge), the BMS's data recording unit collects and stores core temperature and SOC values ​​at a fixed frequency (e.g., 1Hz), forming an actual operating trajectory. When the charge ends, the BMS's optimization processor is activated. It first retrieves the corresponding "ideal change trajectory" from memory based on the boundary conditions of this charge (initial SOC, charging current, ambient temperature). Then, through numerical calculations (e.g., calculating the root mean square error (RMSE) between two curves), it obtains one or a set of "performance deviation values." This deviation value is used as an input feature and sent to a pre-trained machine learning model. The output layer of this model (e.g., a small neural network) directly provides three correction values ​​{ΔKp, ΔKi, ΔKd} for the PID parameters under this operating condition. Finally, the BMS's parameter management module locates the set of PID parameters in the parameter library corresponding to the current battery state (SOH, temperature range, etc.) and updates them with the correction values.

[0085] The above embodiments mainly address how to perform responsive and refined closed-loop control based on the instantaneous state of the battery under specific operating conditions. However, ensuring the safety of the battery throughout its entire life cycle from a more macro-strategic perspective is equally crucial. To this end, this application proposes a forward-looking power limiting strategy based on health status in another embodiment. This strategy limits the maximum load that the battery can withstand at different stages of its life by setting a global power ceiling that adaptively shrinks as the battery ages.

[0086] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 4 This is a flowchart illustrating a power cell fault diagnosis method in an embodiment of this application.

[0087] S401. The control long short-term memory network model predicts the terminal voltage value of the next sampling period based on historical input sequence samples and generates the predicted terminal voltage value.

[0088] This step is applied to real-time online fault diagnosis during vehicle operation or charging, particularly for monitoring early signs of thermal runaway. Specifically, a pre-trained LSTM model is deployed on the BMS's microprocessor (MCU). At any given time t, the MCU maintains a fixed-length (e.g., the past 100 sampling points) sliding window of "historical input sequence samples." When new data [current t, temperature t, SOCt] is collected at time t, this data point is added to the latest end of the sequence, while the oldest data point is removed. This updated complete sequence is used as input and fed into the LSTM model for a forward propagation calculation. After the model's output layer undergoes an activation function, it directly generates the predicted value V_predicted(t+1) for the terminal voltage at time t+1. The logic behind this process is that the LSTM model has learned a profound mapping relationship between the "input sequence" and the "next time voltage" through massive amounts of healthy battery data; therefore, its predicted value is essentially a "digital twin" simulation of the healthy battery's behavior based on the current state and historical load.

[0089] S402. Extract the terminal voltage value with the same sampling period as the predicted terminal voltage value from the real-time operation data, and use it as the actual terminal voltage value.

[0090] This step is executed immediately after S401, running almost in parallel with S401 in each control loop of the BMS. Specifically, the BMS software architecture contains a data acquisition task and a fault diagnosis task. At time t, the fault diagnosis task calls the LSTM model and generates a predicted voltage V_predicted(t+1) for time t+1 based on historical data from time t and earlier. Almost simultaneously, the BMS hardware timer triggers the ADC (analog-to-digital converter) to sample the voltage of all cells a new time. After this sampling is completed and preliminarily processed, the actual data measured at time t+1 is broadcast to the BMS's internal data bus. The fault diagnosis task then subscribes to (or directly reads) this data stream and precisely extracts the voltage measurement value V_measured(t+1) corresponding to the cell being analyzed, with a timestamp of t+1. This extracted V_measured(t+1) is the "actual terminal voltage value" defined in this step.

[0091] S403. Subtract the actual voltage value from the predicted voltage value to calculate the voltage prediction residual.

[0092] S404. Within the preset residual analysis window, collect the voltage prediction residuals from multiple consecutive sampling periods to construct a voltage prediction residual sequence.

[0093] The triggering condition for both steps is the generation and processing of a new round of data in the BMS control loop. Specifically, after S402 obtains the actual voltage V_measured(t+1) at time t+1, S403 immediately performs a subtraction operation: Residual(t+1) = V_predicted(t+1) - V_measured(t+1), obtaining an instantaneous residual value. Then, S404 stores this newly generated Residual(t+1) into a predefined, fixed-length First-In-First-Out (FIFO) queue, which serves as the carrier of the "voltage prediction residual sequence." Simultaneously, the oldest residual value in the queue is dequeued, thus maintaining the queue length (i.e., the "window" size) unchanged. This process is repeated continuously, ensuring that the BMS always maintains a dynamic sequence reflecting the abnormal battery behavior over a recent period.

[0094] S405. Calculate the mean of the voltage prediction residual sequence and generate the residual mean.

[0095] S406. Input the thermal runaway symptom index and the residual mean together into the preset logic judgment module. The logic judgment module has built-in judgment rules for defining fault warning conditions.

[0096] Specifically, after the residual sequence is updated in S404, S405 immediately sums all the values ​​in the sequence and divides them by the sequence length to obtain a smooth and stable "residual mean". Simultaneously, other monitoring tasks of the BMS are also calculating the current "thermal runaway symptom index" in parallel. Subsequently, S406 passes these two key feature values ​​representing different dimensions of risk information—the "residual mean" and the "thermal runaway symptom index"—as two input parameters to the logic judgment module. This module may contain a multi-dimensional judgment matrix or a set of nested logical statements, such as the rule: "IF (residual mean > threshold A) AND (thermal runaway symptom index > threshold B) THEN...", or more complex non-linear relationships.

[0097] S407 The logic judgment module determines whether the combination of thermal runaway symptom index and residual mean value meets the fault warning conditions according to the judgment rules. If the judgment result is yes, the battery cell status is marked as fault warning status and a second alarm command is generated.

[0098] Specifically, once the "residual mean" and "thermal runaway symptom index" are sent to the logic judgment module, the module immediately begins calculation. For example, a rule might be IF (residual mean > 0.005V) AND (thermal runaway symptom index > 1.2), and the module will determine whether the two input values ​​simultaneously satisfy these two inequalities. If the determination result is "true" (i.e., the condition is met), the BMS processor will immediately execute two parallel operations: First, it will locate the data structure representing the battery cell in memory and modify the status field from 0x00 (normal) to 0x01 (level 1 warning); second, according to the preset communication protocol, it will package a CAN message containing the faulty battery cell ID and warning code and push it into the CAN bus transmission queue.

[0099] In some other embodiments of this application, after the alarm command is generated, the method further includes: disconnecting the main relay connected in series with the charging and discharging circuit of the battery cell; driving the audible and visual alarm device to emit preset sound and light alarms; packaging the thermal runaway symptom index and the residual mean to generate a fault information message; and sending the fault information message to a remote cloud server so that the remote cloud server pushes a notification containing the fault information message to the application on the user terminal bound to the battery management system.

[0100] The trigger condition for this series of additional actions is that the S407 generates a specific level (usually the highest level) alarm command. These actions are initiated in parallel or sequentially according to priority, almost simultaneously with the generation of the alarm command. Specifically, taking a vehicle as an example, when the logic judgment module determines that the fault has reached the preset highest emergency level, the BMS processor immediately sends a "disconnect" signal to the main relay driver chip, physically isolating the high-voltage battery and preventing further energy output or input—this is the highest priority safety action. Simultaneously, the BMS broadcasts a command to the instrument cluster controller and body controller via the CAN bus, instructing them to illuminate a specific red fault light and trigger an emergency buzzer. At the software level, the BMS packages the two core criteria for triggering this alarm—the "residual mean" and the "thermal runaway symptom index"—into a predefined format message. Finally, the BMS delivers this message to the onboard communication module (TCU), which then uploads it to the cloud server via a 4G / 5G network. While receiving and recording the data, the cloud server calls the push service to send an emergency notification containing the core fault information to the bound mobile app.

[0101] In another embodiment of this application, the battery cell includes multiple individual cells connected in series. The method further includes: for each individual cell, obtaining its corresponding corrected state of charge (SOC) value; when the difference between the corrected SOC value of any individual cell and the corrected SOC value of other individual cells exceeds a preset imbalance threshold, activating an active balancing circuit; controlling the active balancing circuit to transfer energy from the individual cell with a higher SOC value to the individual cell with a lower SOC value in the same battery cell through energy transfer, until the difference between the corrected SOC values ​​of the multiple individual cells is less than a preset balancing target threshold.

[0102] In this context, a single cell refers to a single lithium-ion battery cell that independently constitutes the smallest energy storage unit within the battery pack. Each cell has independent positive and negative terminals, and its voltage, current, temperature, and stress can be measured independently. A preset imbalance threshold is used to indicate the boundary for triggering the active balancing mechanism. This threshold is calibrated based on cell type, number of cells connected in series, and application scenario experience, and is typically set to a voltage difference of 50mV to 100mV or the corresponding SOC difference. Once the difference in the corrected state of charge (SOC) value between any two cells exceeds this threshold, the battery pack is considered to have an unacceptable state inconsistency. An active balancing circuit is a hardware topology capable of bidirectional energy transfer between cells. Unlike passive balancing, which only dissipates energy through heat, this circuit actively transfers excess energy from cells with high SOC to cells with low SOC using technologies such as DC-DC converters, inductive energy storage units, or capacitive coupling, thereby achieving zero or low energy loss in energy redistribution. The preset equalization target threshold is used to represent the termination condition of the active equalization process. This threshold is strictly less than the imbalance threshold (e.g., set to 20mV or less) to ensure that the energy transfer stops only after the equalization action achieves the expected fine adjustment effect, thus avoiding over-equalization or oscillation.

[0103] This step is applied to the state monitoring and control scenarios of battery cells containing multiple series-connected individual cells during actual operation. Specifically, for each individual cell in the battery cell, the system calls the output of the preceding long short-term memory network model to obtain its corresponding corrected state of charge value, and temporarily stores these values ​​in a one-dimensional array to form a complete state snapshot of the battery pack at the current moment. Then, pairwise comparison operations are performed on this array to calculate the absolute difference |SOC_i-SOC_j| between any individual cell SOC_i and all other cells SOC_j (j≠i), and these differences are compared one by one with a preset imbalance threshold (e.g., 5% or the corresponding 50mV voltage difference). Once the system detects that the difference between any pair of cells satisfies |SOC_i - SOC_j| > the imbalance threshold, it immediately sets the global balancing enable flag to an active balancing circuit. Using its internal topology identification algorithm, the system precisely locates the cell with the highest SOC value (cell_H) and the cell with the lowest SOC value (cell_L) in the current state snapshot, and establishes an energy transfer path from cell_H to cell_L. In practical implementation, this circuit uses a topology based on a flyback or forward DC-DC converter. It drives the power switching transistors with a high-frequency PWM signal, temporarily stores energy through magnetic components (inductors or transformers), and then injects it into the target cell. The energy transfer rate is controlled by an embedded PID closed-loop controller, which monitors the SOC difference between cell_H and cell_L in real time and dynamically adjusts the PWM duty cycle to ensure a smooth and controllable energy flow. When the difference between each pair of corrected state of charge values ​​of all individual cells converges to less than the preset equalization target threshold (e.g., 1% or 20mV), the system automatically clears the equalization enable flag, the active equalization circuit stops energy transfer and enters standby mode.

[0104] In some preferred embodiments, the preset imbalance threshold is not a fixed constant, but dynamically adjusted according to the real-time state of the battery pack. Specifically, the system maintains a two-dimensional lookup table, with the two index dimensions being the corrected state of health (SOH) value and the ambient temperature range. For example, when SOH > 90% and the ambient temperature is between 20-30°C, the imbalance threshold can be set to a more lenient 80mV or 3% SOC difference to reduce unnecessary balancing actions; while when SOH < 70% or the temperature is below 0°C, the threshold is tightened to 50mV or 2% SOC difference to more strictly maintain consistency in aged cells or under low-temperature conditions. The values ​​in this lookup table are obtained through statistical analysis of a large amount of cyclic experimental data from cells at different aging stages and temperatures. Specifically, in a laboratory environment, cells from the same batch are subjected to more than 500 charge-discharge cycles. After every 50 cycles, the impact of different imbalance degrees on capacity retention is recorded, and a three-dimensional mapping relationship between SOH, temperature, and the optimal threshold is fitted using linear regression or a neural network.

[0105] In other embodiments, the active balancing circuit can be implemented using various hardware topologies. For example, a flyback converter topology can be used, employing a single high-frequency transformer in conjunction with a multi-channel switching matrix to achieve point-to-point energy transfer between any cells. Advantages include good isolation performance, transfer efficiency of 85%-92%, and moderate cost, making it suitable for medium-sized battery packs of 12-24 cells. In specific implementations, the transformer turns ratio is preferably set to 1:1, and the operating frequency is selected from 100kHz to 200kHz to balance efficiency and EMI. Alternatively, a capacitive coupling topology can be used, utilizing flying capacitors to repeatedly switch between cells to achieve charge transfer. Advantages include no magnetic components, small size, and low EMI, but the transfer rate is limited by the switching frequency and capacitor value. This is suitable for scenarios where balancing time requirements are not high but size is sensitive (such as wearable devices).

[0106] In some extreme cases (such as power device failure in the equalization circuit or abnormal increase in the internal resistance of a battery cell), energy transfer may fail to converge or even exacerbate the imbalance. To address this, the system incorporates multiple protection logics: 1. Set a maximum equalization duration threshold (e.g., continuous equalization exceeding 2 hours), after which it will forcibly stop and record a fault code; 2. Monitor the equalization current direction in real time to ensure it matches expectations. If reverse current is detected for more than 5 seconds, the equalization path will be immediately cut off; 3. Set an imbalance deterioration detection mechanism. If the SOC standard deviation increases instead of decreasing during equalization and shows an upward trend for three consecutive sampling periods, equalization is deemed a failure, and the system should switch to safety mode and issue an alarm.

[0107] The power cell status diagnostic device in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference]. Figure 5 This is a schematic diagram of the physical device structure of a power cell status diagnostic device in an embodiment of this application.

[0108] It should be noted that, Figure 5 The structure of the power cell status diagnostic device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0109] like Figure 5 As shown, the power cell status diagnostic device includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes based on a program stored in a Read-Only Memory (ROM) 502 or a program loaded from a storage section 508 into a Random Access Memory (RAM) 503, such as performing the methods described in the above embodiments. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An Input / Output (I / O) interface 505 is also connected to the bus 504.

[0110] The following components are connected to I / O interface 505: input section 506 including audio input devices, push-button switches, etc.; output section 507 including liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 508 including hard disks, etc.; and communication section 509 including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 509 performs communication processing via a network such as the Internet. Drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0111] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs the various functions defined in the present invention.

[0112] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0113] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0114] Specifically, the power cell status diagnostic device of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the power cell status diagnostic method provided in the above embodiment.

[0115] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the power cell status diagnostic device described in the above embodiments; or it may exist independently and not assembled into the power cell status diagnostic device. The storage medium carries one or more computer programs, which, when executed by a processor of the power cell status diagnostic device, cause the power cell status diagnostic device to implement the power cell status diagnostic method provided in the above embodiments.

[0116] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0117] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0118] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for diagnosing the condition of a power supply cell, characterized in that, The method, applied to a power cell condition diagnostic system, includes: Within a preset sampling period, real-time operating data is collected from the battery cells, including terminal voltage values, charging and discharging current values, and internal stress values ​​of the cells. After the battery cell completes one preset charge-discharge cycle, the difference between the internal stress value of the cell at the end of the current cycle and the internal stress value of the cell at the beginning of the current cycle is calculated to obtain the irreversible stress increment. Based on the preset time step and sliding window, an input sequence sample with a preset sequence length is constructed from the real-time running data and the irreversible stress increment; The input sequence sample is input into a pre-built long short-term memory network model, which includes at least one long short-term memory network layer configured to output a single hidden state vector only at the end of the sequence, and an output layer connected to the long short-term memory network layer, which contains three neurons and a linear activation function. The Long Short-Term Memory (LSTM) network layer is controlled to process the input sequence samples and generate a final hidden state vector, which encodes the temporal dependencies of the input sequence samples. The final hidden state vector is sent to the output layer, which calculates and outputs the corrected health state value, the corrected state of charge value, and the predicted terminal voltage value for the next sampling period.

2. The method according to claim 1, characterized in that, The method further includes: The core temperature rise rate is calculated based on multiple core temperature values ​​of the battery cells collected within a preset time window. A statistical correlation algorithm is used to calculate the correlation coefficient between the core temperature rise rate and the charge / discharge current value within the preset time window, thereby obtaining the thermal runaway symptom index. The thermal runaway symptom index is input into a pre-trained single-class support vector machine model, which includes a hyperplane boundary for defining the normal operating state of the battery cell. Determine whether the data point position of the thermal runaway symptom index in the single-class support vector machine model crosses the hyperplane boundary. If the determination result is yes, mark the state of the battery cell as a fault warning state and generate a first alarm command.

3. The method according to claim 2, characterized in that, After the step of sending the final hidden state vector to the output layer, whereby the output layer calculates and outputs the corrected health state value and the corrected state of charge value, the method further includes: When the charging and discharging current value is positive, the operating mode is recorded as charging mode; when the charging and discharging current value is negative, it is recorded as discharging mode; when the charging and discharging current value is zero, it is recorded as standby mode. Based on the operating mode, the corrected health status value, the corrected state of charge value, and the fault warning status, a set of initial PID control parameters is retrieved from the preset control parameter library. The core temperature of the battery cell is compared with a preset upper temperature threshold. When the core temperature of the battery cell exceeds the upper temperature threshold, the target charging or discharging current limit is calculated using the initial PID control parameters, and the battery management system is controlled to reduce the charging or discharging current to the target current limit by adjusting the pulse width modulation signal. The corrected state of charge (SOC) value is compared with a preset constant current charging cutoff SOC threshold. When the corrected SOC value is less than the threshold, the battery management system is controlled to execute a constant current and constant voltage charging strategy. When the corrected SOC value is greater than or equal to the threshold, the battery management system is controlled to execute a trickle charging strategy.

4. The method according to claim 3, characterized in that, The method further includes: Based on the corrected health status value, a maximum allowable charge / discharge power value is set to limit the output power and input power of the battery cell.

5. The method according to claim 4, characterized in that, The method further includes: The actual change trajectory of the core temperature value of the battery cell and the corrected state of charge value is continuously recorded. After the preset adjustment period ends, the difference between the actual change trajectory and the preset ideal change trajectory is calculated to obtain the performance deviation value; The performance deviation value is input into a machine learning optimization algorithm, which outputs a set of PID parameter correction values. The initial PID control parameters in the control parameter library are then updated using these PID parameter correction values.

6. The method according to claim 2, characterized in that, After inputting the thermal runaway symptom index into a pre-trained single-class support vector machine model, the method further includes: The Long Short-Term Memory network model is controlled to predict the terminal voltage value of the next sampling period based on the historical input sequence samples, and generate the predicted terminal voltage value. Extract the terminal voltage value with the same sampling period as the predicted terminal voltage value from the real-time operating data, and use it as the actual terminal voltage value; The voltage prediction residual is calculated by subtracting the actual voltage value from the predicted voltage value. Within a preset residual analysis window, the voltage prediction residuals from multiple consecutive sampling periods are collected to construct a voltage prediction residual sequence. Calculate the mean of the voltage prediction residual sequence to generate the residual mean; The thermal runaway symptom index and the residual mean are input together into a preset logic judgment module, which has built-in judgment rules for defining fault warning conditions. The logic judgment module determines, according to the judgment rules, whether the combination of the thermal runaway symptom index and the residual mean satisfies the fault warning condition. If the judgment result is yes, the state of the battery cell is marked as the fault warning state, and a second alarm command is generated.

7. The method according to claim 4, characterized in that, The battery cell comprises multiple individual cells connected in series, and the method further includes: For each individual battery cell, obtain its corresponding corrected state of charge value; When the difference between the corrected state of charge value of any single cell and the corrected state of charge value of other single cells exceeds a preset imbalance threshold, the active balancing circuit is activated. The active balancing circuit is controlled to transfer energy from individual cells with higher state of charge (SOC) values ​​to individual cells with lower SOC values ​​within the same battery cell, until the corrected SOC value difference between the multiple individual cells is less than a preset balancing target threshold.

8. A power cell status diagnostic device, characterized in that, The power cell status diagnostic device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the power cell status diagnostic device to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the power cell status diagnostic device, the power cell status diagnostic device performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the power cell condition diagnostic device, the power cell condition diagnostic device performs the method as described in any one of claims 1-7.

Images