A charging strategy optimization method based on battery health and charging behavior analysis

Abstract

The present application relates to a kind of based on battery health and charging behavior analysis charging strategy optimization method, comprising: by extracting the slope of the internal resistance variation of battery in complete charge-discharge cycle, voltage platform offset and temperature rise response hysteresis, normalization is unified feature vector, mapping to healthy state interval, realize the accurate determination of battery current health state.Joint coding is generated simultaneously in combination with user 24 hours charging behavior, and the decision index is retrieved, and the dynamic anchor point sequence suitable for adaptation is searched, and the charging constraint parameter is structuredly managed.Further, according to the actual temperature rise and capacity supplement efficiency in the charging process, the charging strategy is dynamically fine-tuned, and parameter adaptive correction is realized.The method considers battery health state and user charging habit, can effectively improve charging safety, prolong battery life and enhance the intelligence and robustness of charging process.

Description

Technical Field

[0001] This invention relates to the field of battery management system and charging strategy optimization technology, and in particular to a method for optimizing charging strategies based on battery health and charging behavior analysis. Background Technology

[0002] Currently, battery health management and charging strategy optimization are the core directions of battery management system (BMS) R&D. Existing technologies generally adopt constant current-constant voltage charging modes, adaptive charging curve generation algorithms, and cloud-based big data training-based charging strategy self-learning mechanisms. Some solutions introduce equivalent circuit modeling, impedance identification, or data-driven machine learning algorithms. Due to limitations in device-side computing power and storage resources, embedded BMS systems mainly rely on lightweight solutions such as lookup table methods, parameter rule mapping, and simplified first-order model optimization. The adaptability to personalized user charging behavior usually depends on coarse threshold switching, lacking fine-grained dynamic adaptation capabilities.

[0003] Existing technologies suffer from the following shortcomings: The charging strategy generation process mostly uses battery health as an evaluation reference rather than a core driving variable; health status is merely an output rather than actively involved in charging decisions, leading to a disconnect between the strategy and the actual electrochemical state. Rule templates or simple lookup table methods cannot achieve dynamic offsetting and fine-grained adjustment of staged charging anchor points, resulting in slow response to special scenarios such as abnormal thermal runaway, extreme aging characteristics, and extreme short-charging by users. Cloud-based collaborative optimization methods face bandwidth, latency, and data privacy barriers, making efficient deployment in offline environments difficult. The industry urgently needs a charging curve generation method that can operate efficiently in embedded BMS, strongly couple charging strategies with health status, and possess segmented fine-tuning and adaptive behavior capabilities. Summary of the Invention

[0004] This application provides a charging strategy optimization method based on battery health and charging behavior analysis, aiming to solve one of the problems or issues of the prior art mentioned in the background.

[0005] This application provides a charging strategy optimization method based on battery health and charging behavior analysis, specifically including: S1: Obtain the slope of internal resistance change, voltage plateau offset and temperature rise response hysteresis data of the battery in the most recent complete charge and discharge cycle, and perform normalization processing to generate a normalized feature vector. S2: Project the normalized feature vector onto a preset low-dimensional health state mapping space, and determine the health state interval label of the current battery based on the matching relationship between the projection coordinates and the electrochemical behavior directional boundary. S3: Analyze the user plugging and unplugging time series data within a preset time before this charging, identify the user charging behavior pattern label through a finite state machine, and jointly encode the health state interval label and the user charging behavior pattern label to generate a context-aware decision index. S4: Use the context-aware decision index to retrieve the segmented anchor template library and extract the dynamic anchor sequence that is bound to the health state interval label and contains several dynamic anchors, wherein the dynamic anchor contains constraint operation instructions and trigger logic within a specific SOC interval. S5: Calculate the actual temperature rise deviation and capacity replenishment efficiency deviation of the most recent preset number of charging cycles of the same type, generate an anchor point offset coefficient based on the actual temperature rise deviation and the capacity replenishment efficiency deviation, and apply the anchor point offset coefficient to the constraint operation instructions in the dynamic anchor point sequence to generate a corrected dynamic anchor point sequence.

[0006] S6: Real-time adjustment of charging current and voltage is performed according to the modified dynamic anchor point sequence. During the charging process, the execution response data of each dynamic anchor point is continuously monitored. If the execution response data meets the preset abnormal conditions, the anchor point reliability decay mechanism is triggered and the sequence is switched to the conservative anchor point sequence of the adjacent healthy state interval.

[0007] This application provides a charging strategy optimization method based on battery health and charging behavior analysis, which has the following beneficial effects: (1) This application constructs a low-dimensional health state mapping space, transforming the traditional health assessment that relies on continuous SOH values ​​into a discrete health state interval determination based on electrochemical behavior. It does not require high-precision sensors and complex model training, and can quickly assign the aging stage by extracting features from key fragment data within a single charge-discharge cycle, thus significantly reducing computational overhead and storage requirements.

[0008] (2) A segmented anchor point template library is introduced as the hub for health status and control actions. Each dynamic anchor point is defined as a constraint-type operation instruction within a specific SOC range, and context-aware decision-making is completed by integrating user plug-in behavior patterns. The system continuously tracks the deviation between actual temperature rise and capacity replenishment efficiency, and only incrementally updates the anchor point offset coefficient to avoid reconstructing the entire charging curve. When a critical anchor point is detected to deviate continuously from the expected response, it automatically switches to a conservative fallback strategy to improve the safety boundary under abnormal operating conditions. Attached Figure Description

[0009] Figure 1 This is the main flowchart of a charging strategy optimization method based on battery health and charging behavior analysis; Figure 2 This is a sub-flowchart of a charging strategy optimization method based on battery health and charging behavior analysis. Figure 3 This is another sub-flowchart of a charging strategy optimization method based on battery health and charging behavior analysis. Detailed Implementation

[0010] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0011] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0012] like Figure 1 As shown, this application provides a charging strategy optimization method based on battery health and charging behavior analysis, specifically including: S1: Obtain the slope of internal resistance change, voltage plateau offset and temperature rise response hysteresis data of the battery in the most recent complete charge and discharge cycle, and perform normalization processing to generate a normalized feature vector. S2: Project the normalized feature vector onto a preset low-dimensional health state mapping space, and determine the health state interval label of the current battery based on the matching relationship between the projection coordinates and the electrochemical behavior directional boundary. S3: Analyze the user plugging and unplugging time series data within a preset time before this charging, identify the user charging behavior pattern label through a finite state machine, and jointly encode the health state interval label and the user charging behavior pattern label to generate a context-aware decision index. S4: Use the context-aware decision index to retrieve the segmented anchor template library and extract the dynamic anchor sequence that is bound to the health state interval label and contains several dynamic anchors, wherein the dynamic anchor contains constraint operation instructions and trigger logic within a specific SOC interval. S5: Calculate the actual temperature rise deviation and capacity replenishment efficiency deviation of the most recent preset number of charging cycles of the same type, generate an anchor point offset coefficient based on the actual temperature rise deviation and the capacity replenishment efficiency deviation, and apply the anchor point offset coefficient to the constraint operation instructions in the dynamic anchor point sequence to generate a corrected dynamic anchor point sequence.

[0013] S6: Real-time adjustment of charging current and voltage is performed according to the modified dynamic anchor point sequence. During the charging process, the execution response data of each dynamic anchor point is continuously monitored. If the execution response data meets the preset abnormal conditions, the anchor point reliability decay mechanism is triggered and the sequence is switched to the conservative anchor point sequence of the adjacent healthy state interval.

[0014] Step S1: Acquire the battery's internal resistance change slope, voltage plateau offset, and temperature rise response hysteresis data during the most recent complete charge-discharge cycle, and perform normalization processing to generate a normalized feature vector. A complete charge-discharge cycle refers to the complete process of the battery discharging from the upper threshold of its state of charge (SOC) to the lower threshold of its SOC, and then recharging back to the upper threshold of its SOC; or the complete process of the battery discharging from a fully charged state to a preset SOC threshold (e.g., 20%) and then recharging back to a fully charged state. In this embodiment, a complete charge-discharge cycle is defined as the complete cycle from when the battery voltage first reaches the charging cutoff voltage until the discharge cutoff voltage. The system automatically triggers the data acquisition process in step S1 after detecting that the battery has completed a complete charging process (i.e., the current drops to the cutoff current and the voltage reaches the charging upper limit). Specifically, this includes: S1.1: Obtain the original time-series monitoring data of the battery during the most recent complete charge-discharge cycle. The original time-series monitoring data includes instantaneous current value, terminal voltage value and cell surface temperature value. Perform differential processing on the original time-series monitoring data using sliding window differential processing to generate a primary feature dataset containing the internal resistance dynamic change rate sequence, voltage plateau drift sequence and temperature rise rate response sequence.

[0015] Sub-step S1.1 is located in the data preprocessing stage of main step S1. Its core function is to transform the original physical quantities sampled at high frequency into primary feature sequences that reflect the dynamic characteristics of battery electrochemical processes, providing standardized input for subsequent feature extraction and normalization.

[0016] The system reads the raw timing monitoring data of the most recent complete charge and discharge cycle from the ring buffer of the embedded storage unit. This data stream is acquired by a high-precision ADC at a fixed sampling frequency and includes the timestamp-aligned instantaneous current sequence I(t), terminal voltage sequence V(t), and cell surface temperature sequence T(t).

[0017] For the instantaneous current sequence and the terminal voltage sequence, a sliding time window of length N is constructed. A first-order forward differential operation is performed within the window to calculate the ratio of voltage difference to current difference between adjacent sampling points, so as to estimate the instantaneous value of dynamic internal resistance R_dyn(t) in real time.

[0018] The instantaneous value of dynamic internal resistance is calculated using the following formula: Where V(t) is the terminal voltage at time t, I(t) is the instantaneous current at time t, and V(t-1) and I(t-1) are the corresponding values ​​at the previous sampling time, respectively.

[0019] The generated dynamic internal resistance instantaneous value sequence is then subjected to sliding window differential processing to calculate the change in internal resistance per unit time, generating the internal resistance dynamic change rate sequence dR / dt. This sequence characterizes the impedance evolution rate caused by the growth of the SEI film or lithium deposition inside the battery.

[0020] For the terminal voltage sequence, the slope of voltage change over time is extracted during the constant current charging stage. Noise interference is removed by local linear fitting to generate a voltage plateau drift sequence, which reflects the nonlinear shift characteristics of battery polarization voltage with aging degree.

[0021] For the cell surface temperature sequence, the temperature difference between adjacent sampling points is calculated and divided by the time interval to generate the temperature rise rate response sequence dT / dt. This sequence captures the thermal accumulation effect and thermal conduction hysteresis characteristics of the battery under high-rate charge and discharge.

[0022] Outlier removal is performed on the internal resistance dynamic change rate sequence, voltage plateau drift sequence, and temperature rise rate response sequence. Median filtering is used to eliminate spike noise caused by poor contact or sensor jitter, ensuring the smoothness and continuity of the primary feature dataset.

[0023] S1.2: Based on the aforementioned primary feature dataset, the least squares fitting technique is used to perform linear regression analysis on the internal resistance dynamic change rate sequence to calculate the internal resistance change slope value. At the same time, the mean steady-state deviation in the voltage plateau drift sequence is extracted as the voltage plateau offset value, and the phase delay time relative to the theoretical thermal model in the temperature rise rate response sequence is statistically analyzed as the temperature rise response hysteresis value. Thus, the original three-dimensional feature combination consisting of the internal resistance change slope value, the voltage plateau offset value, and the temperature rise response hysteresis value is output.

[0024] Receive the primary feature dataset generated in S1.1, which includes the internal resistance dynamic change rate sequence, voltage plateau drift sequence, and temperature rise rate response sequence, as the data input source for this step.

[0025] For the dynamic change rate sequence of internal resistance, the least squares method is used for linear regression fitting to construct a linear equation model with timestamp as the independent variable and internal resistance change rate as the dependent variable. By solving the problem of minimizing the sum of squared residuals, the slope coefficient of the sequence is extracted as the internal resistance change slope value to characterize the evolution trend of battery internal impedance with aging process.

[0026] For the voltage plateau drift sequence, identify the steady-state interval in the charge-discharge curve where the current is constant and the voltage fluctuation is small, calculate the set of deviations between the measured voltage value and the standard open-circuit voltage reference value in this interval, perform an arithmetic mean operation on the set of deviations, and the resulting mean value is the voltage plateau offset value, which reflects the degree of potential reference drift caused by the loss of electrode material activity.

[0027] When processing the temperature rise rate response sequence, a lumped parameter method theoretical thermal model based on thermal capacity and thermal resistance parameters is introduced to generate an ideal temperature rise response curve under the corresponding load current. The measured temperature rise rate sequence and the ideal curve are cross-correlation analysis or peak alignment detection is performed to calculate the time lag between the measured response peak and the theoretical response peak. This lag is defined as the temperature rise response lag value, which reflects the thermodynamic characteristics of obstructed internal heat conduction paths or intensified interfacial side reactions.

[0028] The calculated internal resistance change slope, voltage plateau offset, and temperature rise response hysteresis values ​​are assembled in a predetermined order to form the original three-dimensional feature combination.

[0029] S1.3: Call the historical calibration extreme value parameters preset in the embedded storage unit. The historical calibration extreme value parameters include the maximum threshold of the slope of internal resistance change, the maximum threshold of the voltage plateau offset, and the maximum threshold of the temperature rise response hysteresis. Use range normalization processing to map the values ​​of the slope of internal resistance change, the voltage plateau offset, and the temperature rise response hysteresis in the original three-dimensional feature combination to the dimensionless interval of zero to one, so as to generate a normalized feature vector with uniform dimension characteristics.

[0030] Receive the original three-dimensional feature combination output from step S1.2, which includes the internal resistance change slope value, voltage plateau offset value, and temperature rise response hysteresis value, as the input data source for normalization processing.

[0031] The historical calibration extreme value parameter set preset in the embedded read-only memory is invoked. The parameter set includes the maximum threshold of the slope of internal resistance change, the maximum threshold of voltage plateau offset, and the maximum threshold of temperature rise response hysteresis calibrated for a specific cell chemical system, to establish the range boundaries of each dimension.

[0032] A range normalization mapping model is constructed, and linear scaling is performed using the following formula for the slope value of the internal resistance change: in, This represents the normalized internal resistance slope component. This represents the slope value of the currently measured internal resistance change. and These are the preset maximum and minimum calibration thresholds, respectively.

[0033] For the voltage plateau offset value, the same range transformation logic is performed to map the measured offset to a dimensionless range of zero to one, eliminating the interference of the absolute voltage magnitude on the feature weights.

[0034] For the temperature rise response hysteresis value, based on the preset upper and lower limits of phase delay time, the deviation ratio relative to the theoretical thermal model is calculated to generate a third-dimensional normalized component.

[0035] The three independently calculated normalized components are assembled in a fixed order to construct a normalized feature vector with uniform dimensional properties.

[0036] like Figure 2 As shown, step S2: Project the normalized feature vector onto a preset low-dimensional health state mapping space, and determine the health state interval label of the current battery based on the matching relationship between the projection coordinates and the electrochemical behavior directional boundary.

[0037] The low-dimensional health state mapping space refers to a pre-constructed three-dimensional feature space coordinate system with normalized internal resistance change slope, normalized voltage plateau offset, and normalized temperature rise response hysteresis as three orthogonal coordinate axes. This system is used to project the battery's real-time state feature vector onto this space and determine the health state interval. It is obtained as follows: Based on the aging mechanism of lithium-ion batteries, the entire battery lifecycle is divided into three typical health state intervals: SEI stable state, lithium inventory slow decline state, and interface impedance sensitive state. For each health state interval, the coordinates of the feature cluster center points obtained during the laboratory calibration stage are extracted as the geometric center reference points of that interval. Hyperplane segmentation technology is used to calculate the perpendicular bisector of the line connecting the geometric centers of adjacent intervals, generating decision boundary surfaces that distinguish different health states. The coefficients of the decision boundary surface equations and the geometric center coordinates are encapsulated as structured health state interval boundary parameters and stored in an embedded non-volatile storage unit, forming the low-dimensional health state mapping space.

[0038] The electrochemical behavior directional boundary refers to the decision boundary surface in the low-dimensional health state mapping space used to distinguish different health state intervals. This boundary surface is formed by the perpendicular bisector of the line connecting the geometric centers of two adjacent health state intervals, and has a clear electrochemical mechanism orientation—that is, one side of the boundary corresponds to the electrochemical characteristics of the SEI stable state, and the other side corresponds to the electrochemical characteristics of the lithium inventory slow-decline state or the interface impedance sensitive state. It is obtained as follows: During the laboratory calibration stage, feature data of a large number of battery samples are collected for each health state interval, and the coordinates of the cluster center points of the samples in each interval are calculated as geometric center reference points; then, hyperplane segmentation technology is used to calculate the perpendicular bisector of the line connecting the geometric centers of two adjacent health state intervals to generate the decision boundary surface that distinguishes different health states; the equation coefficients and geometric center coordinates of the decision boundary surface are encapsulated as structured health state interval boundary parameters and stored in an embedded non-volatile storage unit to form the electrochemical behavior directional boundary.

[0039] S2 specifically includes: S2.1: Obtaining multiple health state interval boundary parameters based on electrochemical mechanism calibration contained in the low-dimensional health state mapping space, and using the health state interval boundary parameters to construct a multi-dimensional feature space coordinate system to generate a health state interval partitioning model with electrochemical behavior orientation.

[0040] S2.1 Obtain the low-dimensional health state mapping space pre-set in the embedded non-volatile memory unit. This low-dimensional health state mapping space contains multiple health state interval boundary parameters calibrated based on electrochemical mechanisms. A multi-dimensional space construction process is invoked, constructing a three-dimensional feature space coordinate system using the normalized internal resistance change slope, normalized voltage plateau offset, and normalized temperature rise response hysteresis as three orthogonal coordinate axes. Based on the lithium-ion battery aging mechanism, the entire battery lifecycle is divided into three typical health state intervals: SEI stable state, lithium inventory slow decline state, and interface impedance sensitive state. For each health state interval, the coordinates of the feature cluster center points obtained during the laboratory calibration stage are extracted as the geometric center reference points for that interval. Using hyperplane segmentation technology, the perpendicular bisector of the line connecting the geometric centers of two adjacent health state intervals is calculated, generating decision boundary surfaces that distinguish different health states. The coefficients of the decision boundary surface equations and the geometric center coordinates are encapsulated as structured health state interval boundary parameters, forming a health state interval partitioning model with electrochemical behavior orientation.

[0041] S2.2: Receive the normalized feature vector generated from the previous step, perform vector projection operation based on the health state interval partitioning model, and map the normalized feature vector to the multi-dimensional feature space coordinate system to generate real-time projection coordinate points representing the current state of the battery.

[0042] The system receives the normalized feature vector output from step S1.3. This vector contains three components: the normalized internal resistance change slope, the voltage plateau offset, and the temperature rise response hysteresis. It then calls the health state interval partitioning model parameters preset in the embedded read-only memory. These parameters define the orthogonal basis vector matrix of the low-dimensional health state mapping space. A vector projection operation is performed, multiplying the input three-dimensional feature vector with the orthogonal basis vector matrix to eliminate coupling correlations between features. The real-time projection coordinates are calculated using the following formula: Where P is the generated real-time projected coordinate point vector, M is the orthogonal basis vector matrix, V is the input normalized eigenvector, and T represents the matrix transpose operation. Boundary clamping is applied to each component of the calculated projected coordinate points to ensure that the coordinate values ​​are strictly limited to the effective range of the predefined multidimensional feature space coordinate system. The clamped coordinate data is encapsulated into a structured data object containing spatial location identifiers, serving as the real-time projected coordinate points characterizing the current electrochemical state of the battery.

[0043] S2.3: Calculate the distance metric between the real-time projected coordinate point and the boundary parameters of each health state interval, and determine the target area range in which the real-time projected coordinate point falls based on the minimum distance principle, so as to generate a preliminary health state attribution result.

[0044] Receive the real-time projected coordinate points generated from the previous steps. These coordinate points consist of normalized values ​​of three dimensions: the slope of the internal resistance change, the voltage plateau offset, and the temperature rise response hysteresis. These coordinate points serve as the reference input for distance measurement.

[0045] The healthy state interval boundary parameters preset in the embedded storage unit are invoked. The healthy state interval boundary parameters define the centroid coordinate vectors in the three-dimensional feature space of typical electrochemical states such as SEI stable state, lithium inventory slow-decline state and interface impedance sensitive state.

[0046] The geometric distance between the real-time projected coordinates and the centroid coordinates of each healthy interval is calculated using the Euclidean distance formula, and then a dimension-wise difference square summation operation is performed using the following formula: Where D is the Euclidean distance between the real-time projection point and the centroid of the k-th healthy interval, and P i Let C be the eigenvalue of the real-time projection point in the i-th dimension. ki Let be the calibration value of the centroid of the k-th healthy interval in the i-th dimension.

[0047] Iterate through all predefined healthy state intervals, calculate the Euclidean distance between the real-time projection point and the centroid of each interval, and generate a candidate distance set containing multiple distance values.

[0048] The candidate distance set is sorted and filtered by applying the minimum distance principle, and the healthy state interval index corresponding to the distance value with the smallest value is identified.

[0049] The health state interval index corresponding to the minimum distance value is mapped to the preliminary health state attribution result to determine the most likely electrochemical aging stage of the battery.

[0050] In other embodiments, the distance metric is used to quantify the spatial proximity between the real-time projected coordinate points and the boundary parameters of each healthy interval, and can also be calculated using metrics such as Manhattan distance or Mahalanobis distance.

[0051] S2.4: Based on the preliminary health status attribution result, retrieve the preset health status interval label mapping table, convert the target area range into a health status interval label with clear electrochemical semantics, and output the final determined health status interval label of the current battery.

[0052] The system receives the preliminary health status attribution result output from step S2.3. This result includes the target area range identifier where the real-time projected coordinates fall and the corresponding minimum Euclidean distance index. It then calls the health status interval label mapping table pre-installed in the embedded Flash storage area. This table uses a key-value pair structure, where the key is the spatial geometric code of the target area range, and the key is a health status interval label string with explicit electrochemical semantics. A hash lookup is performed, using the target area range identifier as the retrieval key, to locate the corresponding storage unit address in the mapping table and read the original label data bound to that spatial area. The read original label data undergoes integrity verification, checking whether the label fields contain illegal characters or null values. If the verification passes, the label data is directly loaded into a temporary register; if the verification fails, exception handling logic is triggered, forcibly resetting the label to the default safe conservative state label. The loaded health status interval label is associated and bound with the current battery's physical ID, generating a structured state object containing a timestamp, battery ID, and health status label.

[0053] Step S3: Analyze the user plug-in / plug-out time series data within a preset time period before this charging, identify user charging behavior pattern labels through a finite state machine, and jointly encode the health state interval label and the user charging behavior pattern label to generate a context-aware decision index. The preset time can be set to 24 hours, 36 hours, or 48 hours depending on the actual scenario. In this embodiment, the preset time is preferably 24 hours.

[0054] The finite state machine (FSM) is a mathematical model constructed based on preset states and state transition rules, used for pattern recognition of user charging behavior event streams. This FSM defines three typical states: high-frequency short charging, nighttime long charging, and commuting fast charging, as well as the transition probability matrix between each state. It is obtained by: calculating the ratio of the time interval between two adjacent charging sessions to the SOC increment using statistical methods based on the charging start time, duration, and charging frequency density distribution in the standardized user behavior event stream; mapping the behavior sequence to the corresponding state according to preset behavior pattern judgment rules; and statistically analyzing the transition frequency between states to generate the state transition probability matrix, thereby constructing a complete finite state machine.

[0055] S3 specifically includes: S3.1: Obtaining user plug-in / plug-out time series data within the 24 hours prior to this charging, performing timestamp alignment and outlier removal processing on the user plug-in / plug-out time series data to generate a standardized user behavior event stream.

[0056] Read the raw logs of all charging and plugging events recorded in the embedded non-volatile memory during the current cycle, extract the absolute timestamps of each connection establishment and disconnection time and the corresponding initial state of charge value, and construct a raw behavior dataset containing time and energy dimensions.

[0057] The timestamps in the original behavior dataset are aligned to the UTC standard time zone to eliminate time base deviations caused by device sleep clock drift or cross-time zone movement, ensuring that all event sequences are sorted based on a unified global timeline.

[0058] The time interval difference between two adjacent insertion / removal events is calculated. Abnormal breakpoints with intervals less than a preset minimum settling threshold or greater than the maximum effective observation window are identified and marked. These abnormal breakpoints are typically caused by sensor malfunctions or system restarts. In this embodiment, the preset minimum settling threshold is set to 5 minutes to filter false insertion / removal events caused by sensor malfunctions, and the maximum effective observation window is set to 24 hours to eliminate invalid intervals caused by settling across days.

[0059] Linear interpolation is used to complete the missing intermediate state data, while a sliding median filter is used to remove false insertion and removal records caused by instantaneous current spikes due to poor contact, generating a cleaned list of continuous behavioral events.

[0060] The cleaned list of continuous behavioral events is converted into a standardized sequence of binary tuples, each containing a normalized start time and duration, forming a standardized flow of user behavioral events with uniform dimensional characteristics.

[0061] S3.2: Based on the standardized user behavior event stream, construct a state transition matrix including high-frequency short charging state, nighttime long charging state, and commuting fast charging state. Use a finite state machine to perform trajectory matching analysis on the state transition matrix to output user charging behavior pattern labels.

[0062] The system receives a standardized user behavior event stream after timestamp alignment and outlier removal, and extracts the start time, end time, and duration of each charging session as basic temporal features. It calculates the time interval between two adjacent charging sessions, constructs a histogram of charging frequency density distribution, and identifies the peak charging trigger frequency within a unit time window. It then calculates the ratio of the average duration of a single charging session to the SOC increment, generating a charging depth feature vector to distinguish between shallow charging and deep saturation charging behaviors.

[0063] Based on preset behavior pattern determination rules, high-frequency, short-interval, low SOC increment behavior sequences are mapped to high-frequency short-charging states; nighttime, long-duration, high SOC increment behavior sequences are mapped to nighttime long-charging states; and commuting-time, medium-to-high-power behavior sequences are mapped to commuting fast-charging states. A state transition matrix containing these three typical states and their transition probabilities is constructed, and a finite state machine is used to match and analyze the behavior trajectory within the current 24 hours. Using the standardized behavior event stream within the current 24 hours as the input sequence, starting from the current state, the state is updated event by event according to the state transition matrix, and the state dwell time and transition frequency are recorded. Finally, the current user charging behavior pattern label is determined using the maximum a posteriori probability criterion. This current user charging behavior pattern label is used for subsequent joint encoding with the healthy state interval label to achieve adaptive adjustment of the charging strategy to the user's personalized behavior.

[0064] S3.3: Obtain the health status interval label determined in the previous steps and the user charging behavior pattern label, and perform bit concatenation and hash mapping operations on the health status interval label and the user charging behavior pattern label to generate a context-aware decision index.

[0065] The system receives the health status interval label and the user charging behavior pattern label output from the previous steps. The labels respectively represent the current electrochemical aging stage of the battery and the user's recent electricity consumption habits.

[0066] The health state interval labels are binary encoded and mapped, and a fixed bit width is allocated according to the preset number of intervals. Semantic labels such as 'SEI stable state' and 'lithium inventory slow decline state' are converted into corresponding low-bit sequences to ensure that different health states have unique identifiers at the bit level.

[0067] The same binary encoding process is performed on the user charging behavior pattern label. High-bit sequences are assigned according to behavior categories such as 'high-frequency short charging' and 'nighttime long charging' to form a digital fingerprint that corresponds one-to-one with the behavior pattern, ensuring the distinguishability of behavior features.

[0068] The encoded health state low-bit sequence and the behavior pattern high-bit sequence are concatenated bitwise to construct the original joint codeword containing dual-dimensional information. This joint codeword fully preserves the orthogonal feature information of the health state and behavior pattern.

[0069] A hash map is introduced to process the original joint encoded word. The hash value is calculated using the division remainder method or the folding method to compress the data length and eliminate coding redundancy, thereby generating a fixed-length context-aware decision index.

[0070] The generated context-aware decision index is subjected to conflict detection and verification. If a hash collision is found, the index value is adjusted by linear probing to ensure that each unique 'health-behavior' combination corresponds to a unique memory addressing key value.

[0071] By using bit concatenation and hash mapping, the health status interval labels and user charging behavior pattern labels from the previous step are transformed into context-aware decision indexes with unique addressing capabilities. This enables a rapid mapping from a multi-dimensional semantic space to a one-dimensional storage address space, providing a low-overhead and unambiguous index basis for efficient retrieval of the subsequent segmented anchor template library, and significantly improving the response speed of the embedded system in real-time charging strategy switching.

[0072] Step S4: Utilize the context-aware decision index to retrieve the segmented anchor template library, and extract a dynamic anchor sequence containing several dynamic anchors bound to the healthy state interval label. The dynamic anchors contain constraint-type operation instructions and triggering logic within a specific SOC interval. Specifically, this includes: S4.1: Initialize and build the segmented anchor template library to generate an initialized segmented anchor template library with multi-dimensional indexing capabilities. Specifically, this includes defining a combination mapping relationship between multiple health state interval labels and multiple user charging behavior pattern labels based on laboratory calibration data, associating each combination mapping relationship with a set of preset dynamic anchor sequence, where each dynamic anchor sequence consists of several dynamic anchors arranged in order of charge state interval, and each dynamic anchor encapsulates constraint-type operation instructions and trigger logic within a specific charge state interval, ultimately forming an initialized segmented anchor template library with a joint encoded label as the unique retrieval key.

[0073] Receive the context-aware decision index generated in the previous steps. This index contains health status interval identifier bits and user charging behavior pattern identifier bits, which serve as unique keys for retrieval of the segmented anchor point template library.

[0074] Based on laboratory-calibrated electrochemical aging data and typical user charging logs, a multidimensional mapping matrix is ​​constructed to define all legal combinations of healthy state interval labels and user charging behavior pattern labels.

[0075] For each combination method, a set of dynamic anchor points is pre-set according to the order of the State of Charge (SOC) intervals. Each dynamic anchor point encapsulates the constraint-type operation instructions and trigger logic within a specific SOC interval.

[0076] Each set of combined mapping relationships is linked to its corresponding dynamic anchor sequence by establishing a bidirectional pointer link, forming an initialized segmented anchor template library with the joint encoded label as the unique retrieval key.

[0077] Memory address alignment optimization is performed on the initialized segmented anchor template library to ensure that the dynamic anchor sequence is stored continuously in the embedded Flash memory, thereby reducing read latency and improving cache hit rate.

[0078] Through the above initialization process, the discrete health assessment results and behavior recognition results are transformed into standardized charging strategy templates with direct execution capabilities. This achieves static decoupling and dynamic correlation between charging strategies, battery status, and user habits, providing a benchmark reference framework for subsequent real-time fine-tuning.

[0079] For example, a 12KB space is allocated in the Flash memory area of ​​the embedded MCU to build a segmented anchor point template library. Three health state interval labels (SEI stable state, lithium inventory slow-down state, interface impedance sensitive state) and three user behavior mode labels (high frequency short charge, nighttime long charge, commuting fast charge) are defined to form nine combined mapping relationships. Taking the combination of "interface impedance sensitive state" and "commuter fast charging state" as an example, the corresponding dynamic anchor point sequence includes 5 anchor points: The SOC 0-20% range anchor point sets the maximum current to 1.5C, triggered by a temperature <40℃; the SOC 20-50% range anchor point sets the current step-down size to 0.1C / 5min, triggered by a temperature rise rate >0.5℃ / min; the SOC 50-80% range anchor point sets the constant voltage switching threshold to 4.35V, triggered by a voltage plateau duration >10s; the SOC 80-95% range anchor point sets the trickle charging current to 0.05C, triggered by an SOC change rate <0.1% / min; and the SOC 95-100% range anchor point sets the cutoff current to 0.01C. All anchor point parameters have been calibrated and verified in the laboratory to ensure that the battery surface temperature rise is controlled within 45℃ and the capacity replenishment efficiency is higher than 95% under the corresponding operating conditions. The template library has a fixed structure and the index lookup time is always less than 2μs, which meets the real-time requirements of embedded systems.

[0080] S4.2: Perform hash addressing retrieval processing on the segmented anchor template library based on the context-aware decision index to locate the storage address of the target dynamic anchor sequence; specifically, this includes parsing the health state interval identifier and the user charging behavior mode identifier in the context-aware decision index, calculating the hash value after their joint encoding, and using the hash value to match the corresponding storage pointer in the index table of the segmented anchor template library, thereby locking the base address of the target dynamic anchor sequence that is completely bound to the health state interval label and the user charging behavior mode label.

[0081] Receive the context-aware decision index generated in the previous steps. This index is composed of health state interval identifier bits and user charging behavior pattern identifier bits concatenated by bits, and serves as the unique key value for retrieving the segmented anchor point template library.

[0082] The binary data structure of the context-aware decision index is parsed, and the high N bits are extracted as the health status interval encoding value, and the low M bits are extracted as the user behavior pattern encoding value to ensure the accuracy of the identifier bit separation.

[0083] The BKDR hash algorithm is used to jointly process the extracted health status interval encoding value and the user behavior pattern encoding value to generate an unsigned integer hash value, thereby eliminating the risk of address collisions between different tag combinations. The specific formula is as follows: Where H is the final hash value, S[i] is the value of the i-th identifier bit, n is the total length of the identifier bits, and 131 is the hash seed coefficient. This formula maps the multidimensional label to a one-dimensional linear space index.

[0084] Perform a modulo operation on the generated hash value, with the modulus being the maximum capacity of the segmented anchor template library index table, to constrain the hash value within the range of legal memory addresses and prevent array out-of-bounds access.

[0085] The calculated index value is used to directly address the corresponding storage pointer data in the index table pre-installed in the embedded Flash. This pointer points to the starting physical address of the dynamic anchor sequence in the SRAM.

[0086] The validity of the read storage pointer is verified. If the pointer value is null or points to an illegal memory area, it is determined that the index is missing and the pointer is automatically reset to the address pointer of the default conservative strategy to ensure system robustness.

[0087] S4.3: Read the target dynamic anchor sequence data from the segmented anchor template library according to the storage address to generate a dynamic anchor sequence object to be corrected; specifically, this includes continuously reading the structured data block stored in memory through the storage pointer, parsing the multiple dynamic anchor data units contained in the data block, loading the constraint operation instructions and trigger logic in each dynamic anchor data unit into a temporary buffer, and reassembling them into a complete dynamic anchor sequence object to be corrected, ensuring that the dynamic anchor sequence object to be corrected completely retains the original calibrated electrochemical behavior directional characteristics.

[0088] The system receives a storage pointer generated in the preceding steps, which points to the starting memory address of the target dynamic anchor sequence in the segmented anchor template library. Based on the embedded system's bus access protocol, and using 32-bit word length as the basic unit, it performs continuous read operations starting from the starting memory address to obtain a raw binary data block containing multiple dynamic anchor data units. The read raw binary data block is parsed byte-wise to identify the data block header identifier, confirming data structure version compatibility and preventing data format misalignment due to firmware upgrades. According to the predefined structure offset definition, the header information of each dynamic anchor data unit is extracted byte-by-byte. This header information includes the SOC interval start and end values, instruction type encoding, and trigger condition parameter length. Based on the instruction type encoding, memory space is dynamically allocated in a temporary buffer to temporarily store the constraint-type operation instructions and trigger logic to be loaded. The parsed constraint-type operation instruction data, such as current threshold, voltage cutoff value, and time window parameters, are converted to floating-point or integer values ​​in big-endian mode and stored in the instruction field of the temporary buffer. The comparison operator code and reference threshold in the trigger logic are parsed synchronously to construct a structured object containing logical branch information, ensuring the executability of the trigger logic. The parsed individual dynamic anchor data units are appended to the end of the linked list of the dynamic anchor sequence object to be corrected, maintaining the anchors in ascending SOC order. This parsing and loading process is repeated until all dynamic anchor data units in the current data block have been traversed, or a sequence end marker is encountered. The number of anchors in the dynamic anchor sequence object to be corrected is checked against the expected length recorded in the index table; if they do not match, a data integrity warning is issued. Through memory copying and structural reorganization, the scattered binary data is transformed into dynamic anchor sequence objects with electrochemical behavior orientation, enabling rapid instantiation and context association of the charging strategy template, providing a standardized input basis for subsequent fine-tuning.

[0089] For example, the storage pointer points to Flash address 0x08004000, and the system reads a 128-byte data block. The first anchor header information is parsed, identifying the SOC range as 30%-50% and the instruction type as constant current regulation. The current threshold parameter 0x41200000 is extracted and converted to a floating-point number of 10.0A; the temperature rise trigger threshold 0x40400000 is extracted and converted to 3.0℃ / min. A trigger logic object is constructed: IF (dTemp / dt>T threshold THEN (I set = I act * k), where dTemp / dt is the current real-time temperature rise rate of the battery cell, T threshold The preset temperature rise trigger threshold is preferably 3.0℃ / min; I set The new target current command value set by the system after the trigger logic is satisfied; I act To trigger the logical judgment, the system currently executes the action current value, where k is the current reduction coefficient (dimensionless). In this embodiment, the current reduction coefficient is set to 0.85. This anchor point is added to the sequence object. Subsequent anchor points are parsed until the SOC reaches 95%. Finally, a sequence object containing 5 dynamic anchor points is generated, with a total processing time of less than 2ms and a memory footprint of 400 bytes, ensuring the strategy is immediately available.

[0090] S4.4: Perform structural integrity verification on each dynamic anchor point in the dynamic anchor point sequence object to be corrected, and output a dynamic anchor point sequence; specifically, this includes traversing each dynamic anchor point in the dynamic anchor point sequence object to be corrected, checking whether the range of its internal constraint-type operation instruction parameters exceeds the limit and whether the syntax tree of the triggering logic is closed. If any dynamic anchor point is found to have a data structure abnormality, it is automatically replaced with a conservative dynamic anchor point in the adjacent healthy state interval. Finally, a dynamic anchor point sequence in which all dynamic anchor points conform to the preset safety specifications is output for subsequent fine-tuning unit to call.

[0091] The system receives a sequence of dynamic anchor points to be corrected generated in step S4.3. This object contains multiple dynamic anchor point data units sorted by SOC interval. It iterates through each dynamic anchor point in the sequence, extracting its encapsulated constraint-type operation command parameters, including current threshold, voltage cutoff value, and duration upper limit. The extracted parameters are compared with a preset safety boundary table, which defines the maximum allowable charging current and highest cutoff voltage for each health state interval. If any parameter exceeds the allowable range of the corresponding health state interval, the anchor point is marked as an out-of-bounds abnormal state. The trigger logic in each dynamic anchor point is parsed, and a logic syntax tree structure is constructed. The leaf nodes of the syntax tree are checked to ensure complete connection to the root node, verifying the closure of the logical expression. If an unclosed logic branch or illegal operator combination is found, the anchor point is marked as a structural defect state. The number of dynamic anchor points marked as out-of-bounds abnormal or structural defect is counted. If abnormal anchor points exist, the adjacent health state interval mapping table is retrieved to determine the conservative alternative interval corresponding to the current health state. The default dynamic anchor point sequence bound to the conservative alternative interval is read from the segmented anchor point template library. Replace the abnormal anchor points in the original sequence with the corresponding anchor points from the default sequence, ensuring that the replaced anchor points have stricter current limits and more lenient temperature rise tolerance. Perform a final integrity scan on the sequences that have been replaced or passed verification to confirm that all anchor points comply with preset safety specifications. Through structural integrity verification and automatic anomaly replacement processing, the raw data read in the previous step is transformed into a dynamic anchor point sequence, providing a safety fallback before the charging strategy is executed and preventing the risk of battery overcharging due to data corruption or calibration deviations.

[0092] like Figure 3 As shown, step S5: Calculate the actual temperature rise deviation and capacity replenishment efficiency deviation of the most recent preset number of charging cycles of the same type; generate an anchor point offset coefficient based on the actual temperature rise deviation and the capacity replenishment efficiency deviation; and apply the anchor point offset coefficient to the constraint-type operation instructions in the dynamic anchor point sequence to generate a corrected dynamic anchor point sequence. Specifically, this includes: S5.1: Obtain the measured temperature curve data and nominal temperature model data from the most recent preset number of charging processes of the same type. Perform point-by-point difference calculation on the measured temperature curve data and the nominal temperature model data to generate the actual temperature rise deviation, and extract the maximum positive deviation value in the actual temperature rise deviation as the key temperature rise feature. The preset number of times can be set to two, three, or more times according to the actual scenario. In this embodiment, the preset number of times is preferably three times.

[0093] Historical charging cycle data that most recently matched the current charging behavior pattern label with a preset number of times is retrieved from embedded non-volatile memory. The data includes a battery surface temperature sampling sequence aligned to timestamps and the corresponding state of charge (SOC) sequence.

[0094] Based on a pre-set electrochemical thermal model, the theoretical reference temperature value at each moment is obtained by looking up a table according to the real-time SOC value of the current battery, the ambient temperature and the charging current, and a nominal temperature model curve is constructed.

[0095] The measured temperature curve and the nominal temperature model curve are synchronized on the time axis. The difference between the measured temperature and the theoretical temperature at the same moment is calculated by the method of successive differences to generate the actual temperature rise deviation.

[0096] By iterating through the actual temperature rise deviations, all positive deviation terms greater than zero are selected to characterize the abnormal temperature rise range where the actual heat generation of the battery exceeds the theoretical expectation.

[0097] The maximum value is extracted from the selected positive deviation terms and defined as the key temperature rise characteristic quantity. This value directly reflects the margin of thermal runaway risk of the battery in its current healthy state.

[0098] S5.2: Obtain the measured charging capacity data and theoretical target capacity data from the most recent preset number of charging processes of the same type, perform ratio calculation based on the measured charging capacity data and the theoretical target capacity data to generate capacity replenishment efficiency deviation, and combine the key temperature rise characteristic quantity and the capacity replenishment efficiency deviation to construct a multi-dimensional feedback error vector.

[0099] Read the complete charging cycle record that matches the current charging behavior pattern label the most recent preset number of times from the embedded non-volatile memory, and extract the cumulative charging capacity value at the end of each charging cycle as the measured charging capacity data sequence.

[0100] The nominal capacity parameters in the underlying library of the battery management system are called, and combined with the theoretical available capacity coefficient corresponding to the current health state interval, the theoretical target capacity value sequence for each charging cycle is calculated.

[0101] An element-wise ratio operation is performed on the measured charging capacity data sequence and the theoretical target capacity value sequence to generate an intermediate vector containing charging efficiency indicators for three historical cycles.

[0102] The intermediate vector is processed using a weighted average method, with the weight coefficient of the most recent charging cycle set to 0.5, and the weight coefficients of the previous two charging cycles set to 0.3 and 0.2 respectively, to highlight the impact of recent changes in battery state.

[0103] The capacity replenishment efficiency deviation is calculated using the following formula: Where δ represents the capacity replenishment efficiency deviation. The weighting coefficient for the i-th charge. The measured charging capacity for the i-th charge. The theoretical target capacity for the i-th charge.

[0104] Obtain the key temperature rise characteristic quantity output from the preceding step S5.1. This characteristic quantity represents the maximum positive temperature rise deviation of the battery during the most recent preset number of charging processes of the same type.

[0105] The calculated capacity replenishment efficiency deviation and key temperature rise characteristics are concatenated into a vector to construct a two-dimensional feedback error vector containing thermodynamic deviation and electrochemical efficiency deviation.

[0106] S5.3: Call the linear mapping lookup table, input the multidimensional feedback error vector as an index into the linear mapping lookup table, output the corresponding initial adjustment gain parameter through table lookup interpolation, and normalize the initial adjustment gain parameter to generate a standardized anchor offset coefficient.

[0107] The linear mapping lookup table is a two-dimensional lookup table. Its row index corresponds to the quantization range of key temperature rise characteristics, and its column index corresponds to the quantization range of capacity supplementation efficiency deviation. Each grid node stores the optimal adjustment gain parameter verified by laboratory calibration under the corresponding operating condition. The feedback error space is uniformly divided into N×N grids according to the historical distribution range of temperature rise characteristics and efficiency deviation, with N ranging from 8 to 16. During table lookup, the actual error vector is mapped to its corresponding grid cell, and bilinear interpolation is used to calculate the gain value at any point within the grid. If the error vector exceeds the maximum boundary defined by the lookup table, the gain value of the boundary node is used as the extrapolation result, and an out-of-bounds flag is triggered for upper-level monitoring.

[0108] Receive the multidimensional feedback error vector generated by the preceding steps, which contains key temperature rise characteristics and capacity replenishment efficiency deviation.

[0109] The linear mapping lookup table preset in the embedded read-only memory is invoked. This lookup table establishes a discrete mapping relationship from the feedback error space to the initial adjustment gain space.

[0110] Determine the coordinates of the adjacent nodes of each component in the multidimensional feedback error vector in the lookup table grid, and identify the set of hypercube vertices surrounding the current error point.

[0111] Obtain the pre-stored initial adjustment gain parameter values ​​corresponding to each vertex. These parameter values ​​are pre-calculated and stored based on laboratory full-condition calibration data.

[0112] Bilinear or trilinear interpolation is used to perform a weighted summation of the gain parameters of each vertex based on the relative position weight of the current error point within the hypercube.

[0113] The initial adjustment gain parameter is calculated using the following formula: Where G is the calculated initial adjustment gain parameter, w1 to w4 are the interpolation weight coefficients calculated based on the error point position, and G1 to G4 are the pre-stored gain parameter values ​​of adjacent grid vertices.

[0114] The calculated initial adjustment gain parameter is subjected to a limiting process to ensure that its value is within the preset safe adjustment range, thus preventing over-adjustment due to extreme errors.

[0115] The initial adjustment gain parameter after limiting is normalized and mapped to a uniform numerical range to generate standardized anchor offset coefficients. This can be achieved using methods such as Min-Max linear normalization, normalization based on historical extreme values, or exponential decay mapping. In this embodiment, Min-Max linear normalization is preferred, linearly mapping the initial adjustment gain parameter after limiting to the [0, 1] interval to maintain the original proportional relationship of the parameter values ​​while also considering computational efficiency.

[0116] By using table lookup interpolation and normalization, the multidimensional feedback error vector from the previous step is transformed into a standardized anchor offset coefficient, enabling fine-tuning of the expected technical effect based on individual battery differences.

[0117] S5.4: Read the original dynamic anchor sequence bound to the context-aware decision index in the segmented anchor template library, parse the constraint operation instructions contained in each dynamic anchor in the original dynamic anchor sequence, and use the standardized anchor offset coefficient to perform incremental correction calculations on the current threshold parameter and voltage duration parameter in the constraint operation instructions.

[0118] Read the original dynamic anchor sequence bound to the context-aware decision index from the segmented anchor template library, and parse the constraint-type operation instructions contained in each dynamic anchor. Extract the reference current threshold parameter and reference voltage duration parameter from the instructions as the objects to be corrected.

[0119] Obtain the standardized anchor point offset coefficient generated in the previous steps. This coefficient represents the global fine-tuning weight calculated based on historical temperature rise deviation and capacity efficiency deviation. Map the standardized anchor point offset coefficient to the SOC interval weight factor where the current dynamic anchor point is located to eliminate the difference in parameter sensitivity between different state of charge intervals.

[0120] A linear incremental correction operation is performed on the reference current threshold parameter. The corrected current threshold is calculated using the following formula: in, The corrected current threshold for the i-th anchor point. The original reference current threshold, This is the mapping value of the standardized anchor point offset coefficient within the SOC range.

[0121] A nonlinear proportional correction calculation is performed on the reference voltage duration parameter. Considering the hysteresis effect of voltage plateau duration on heat accumulation, the corrected duration is calculated using the following formula: in, The corrected voltage duration for the i-th anchor point. For the original baseline duration, The time compensation coefficient is adjusted based on temperature rise hysteresis. Its value is obtained by low-pass filtering the mapping value of the standardized anchor offset coefficient in this SOC range.

[0122] The calculated corrected current threshold and corrected voltage duration are written back to the dynamic anchor point data structure, replacing the original baseline parameters. The anchor point status bit is updated to indicate that it has been fine-tuned, ensuring that subsequent control modules use strategy parameters that have been corrected through real-time feedback.

[0123] S5.5: The constraint operation instructions after the incremental correction operation are re-encapsulated into the corresponding dynamic anchor data structure, the trigger logic state bit of the dynamic anchor is updated to mark it as corrected, and finally the corrected dynamic anchor sequence containing the updated constraint operation instructions is output for the subsequent charging control module to execute.

[0124] Receives a set of constraint-type operation command parameters after incremental correction calculation by S5.4, including the corrected current threshold, voltage duration, and trigger condition logic tree.

[0125] Traverse each dynamic anchor data unit in the sequence of dynamic anchors to be corrected, and locate the memory address space of its internal storage constraint operation instructions.

[0126] Write the corrected current threshold parameter into the maximum charging current field in the dynamic anchor structure, overwriting the original calibration value to adapt to the current temperature rise deviation.

[0127] The corrected voltage duration parameter is written into the constant voltage maintenance time field in the dynamic anchor structure, and the charging cutoff strategy is adjusted to compensate for the capacity replenishment efficiency deviation.

[0128] Update the status flag field in the dynamic anchor structure, resetting the "uncorrected" status bit to the "corrected" status bit, indicating that the anchor has applied the local fine-tuning coefficient.

[0129] Perform CRC cyclic redundancy check on the updated dynamic anchor data structure to ensure data integrity and prevent bit flipping errors during instruction encapsulation.

[0130] The validated dynamic anchors are reassembled in SOC interval order to construct a corrected dynamic anchor sequence object containing complete trigger logic and correction parameters.

[0131] The final execution weight of the corrected anchor point instruction is calculated using the following formula to ensure control stability under the combined effect of multiple parameters: Where W is the execution weight coefficient, k is the sensitivity adjustment factor, and ΔT is the actual temperature rise deviation value.

[0132] The corrected dynamic anchor point sequence object is stored in the real-time control buffer and used as the direct input source for the PWM signal generation in step S6.

[0133] Step S6: Real-time adjustment of charging current and voltage is performed according to the corrected dynamic anchor point sequence. During charging, the execution response data of each dynamic anchor point is continuously monitored. If the execution response data meets a preset abnormal condition, an anchor point reliability decay mechanism is triggered, and the sequence is switched to a conservative anchor point sequence in an adjacent healthy state interval. The preset abnormal condition is: the execution response data deviates from the expected response threshold after a preset number of consecutive detections. In this embodiment, the preset number of consecutive detections is two. In other embodiments, the preset number of consecutive detections can be adaptively adjusted according to actual needs. Specifically, this includes: S6.1: Obtain the constraint-type operation commands in the corrected dynamic anchor point sequence and the currently collected real-time battery terminal voltage, charging current and surface temperature data, and perform charging power output adjustment based on pulse width modulation to generate a real-time charging control signal that meets the requirements of the dynamic anchor point sequence.

[0134] The system reads constraint-type operation commands from the corrected dynamic anchor point sequence and parses the target current threshold and voltage cutoff condition corresponding to the current SOC range. Real-time telemetry data from the battery management system is simultaneously acquired, including instantaneous terminal voltage, charging circuit current, and cell surface temperature sensor values. Digital filtering is performed on the acquired raw analog signals, and a moving average is used to remove high-frequency noise interference, generating a smoothed state monitoring dataset. The smoothed terminal voltage and current data are input to the PWM control module, and the duty cycle adjustment is calculated based on preset switching frequency parameters. The PWM duty cycle is determined using the following formula based on the volt-second balance principle: Where D is the PWM duty cycle. The target voltage or equivalent voltage command specified for the dynamic anchor point. This is the real-time battery terminal voltage. The voltage is the charger bus voltage. The calculated duty cycle D is limited to a safe range to prevent overshoot or undervoltage. The power switch is driven to perform on / off operations, adjusting the inductor's energy release rhythm, thereby precisely controlling the charging current injected into the battery. The actual output current is compared with the current threshold in the anchor command in real time. If the deviation exceeds the allowable tolerance, closed-loop feedback correction is initiated, and the duty cycle of the next cycle is fine-tuned. The charging power output is adjusted through the above pulse width modulation, converting the corrected dynamic anchor sequence generated in the previous step into a real-time charging control signal that conforms to electrochemical constraints, achieving precise hardware mapping of the charging strategy.

[0135] S6.2: Obtain the actual battery temperature rise rate data and voltage ramp-up delay time data under the action of real-time charging control signal, perform statistical processing based on sliding window and time-series feature extraction processing to generate real-time execution response data characterizing the execution effect of the current dynamic anchor point.

[0136] The battery terminal voltage sampling sequence, charging current sampling sequence, and cell surface temperature sampling sequence generated by S6.1 under the action of the real-time charging control signal are received as the raw input data for time-series feature extraction.

[0137] The sampling sequence of the cell surface temperature is processed by a sliding window statistical method. The window length is set to N sampling points and the step size is 1 sampling point. The linear regression slope of the temperature data within the window is calculated to generate a real-time temperature rise rate data sequence.

[0138] The rate of temperature rise at time t is calculated using the following formula: in, Let be the rate of temperature rise at time t. Let t be the temperature sample value at time ti, N be the sliding window length, and i be the index variable within the window.

[0139] The battery terminal voltage sampling sequence is subjected to first-order differential processing to calculate the voltage change between adjacent sampling points, and the starting point of the voltage ramp-up stage is determined in combination with the charging current state.

[0140] The system calculates the time required for the voltage to reach a preset threshold after switching from the constant current stage to the constant voltage stage or after triggering a specific SOC anchor point, in order to generate voltage ramp-up delay time data.

[0141] The real-time temperature rise rate data and voltage ramp-up delay time data are timestamped to construct a binary data structure that includes thermal response characteristics and electrochemical response characteristics.

[0142] Outlier filtering is performed on the binary data structure to remove instantaneous jump data caused by sensor noise, and retain smooth feature data that conforms to the laws of physical change.

[0143] By using the above sliding window statistics and time-series alignment processing method, the original sampling data from the previous step is transformed into real-time execution response data that characterizes the execution effect of the current dynamic anchor point. This enables a quantitative assessment of the actual execution deviation of the charging strategy and provides accurate data support for the subsequent reliability decay mechanism.

[0144] S6.3: Obtain the real-time execution response data and the preset expected response threshold range, and perform continuous deviation judgment processing based on the deviation counting logic unit to generate an abnormal trigger flag indicating that the execution response data exceeds the allowable range twice consecutively.

[0145] The system receives real-time execution response data generated in step S6.2, which includes the actual temperature rise rate and voltage ramp-up delay time under the current dynamic anchor point. It calls a preset expected response threshold range table in the embedded storage unit, which defines the upper and lower limits of the allowable temperature change rate and the voltage response time tolerance range for each dynamic anchor point. The system compares the actual temperature rise rate in the real-time execution response data with the upper and lower limits of the expected temperature rise rate to determine if it exceeds the allowable temperature fluctuation range. Simultaneously, it compares the real-time voltage ramp-up delay time with the preset maximum allowable delay threshold to determine if the voltage response lags behind the theoretical model. A deviation counting logic unit is constructed, which maintains a continuous deviation counter for the current dynamic anchor point, initially set to zero. When either the actual temperature rise rate or the voltage ramp-up delay time exceeds the corresponding expected response threshold range, it is determined as a single execution deviation event. The single execution deviation event is marked, and the value of the continuous deviation counter is incremented. If the execution response data for the current detection period is all within the expected response threshold range, the continuous deviation counter is reset to zero, indicating that the current anchor point is executing normally. The value of the continuous deviation counter is continuously monitored to determine whether it has reached a preset continuous deviation trigger threshold, which is set to 2. When the value of the continuous deviation counter equals 2, it indicates that the current dynamic anchor point has failed to achieve the expected charging control effect in two consecutive control cycles. An abnormal trigger flag is generated to indicate that the response data exceeds the allowable range for two consecutive executions, and this flag is set to an active state. Through continuous comparison processing between the real-time response data and the expected threshold by the deviation counting logic unit, the real-time execution response data of the previous step is transformed into an abnormal trigger flag characterizing the risk of strategy failure, realizing real-time quantitative evaluation of the reliability of the charging strategy execution and rapid identification of anomalies.

[0146] S6.4: Obtain the anomaly trigger flag and the credibility weight coefficient of the current dynamic anchor sequence, and perform weight update calculation based on the exponential decay function to generate an updated credibility weight coefficient that reflects the decrease in the reliability of the current dynamic anchor sequence.

[0147] Receive the anomaly trigger flag generated from the previous step. This flag indicates that the response data of the current dynamic anchor sequence deviates from the expected threshold within two consecutive monitoring periods. Read the confidence weight coefficient of the current dynamic anchor sequence preset in the memory. The coefficient has an initial value of 1.0, representing the fully trustworthy state of the strategy.

[0148] Determine whether the abnormal trigger flag is active. If the flag is true, start the exponential decay calculation process; if the flag is false, keep the current confidence weight coefficient unchanged and skip the subsequent calculation.

[0149] The floating-point arithmetic unit within the embedded microcontroller is invoked to load a preset decay factor parameter. This parameter is set to a value between 0.85 and 0.95 to control the rate at which reliability decreases.

[0150] The weight update operation is performed based on the exponential decay model, and the updated confidence weight coefficient is calculated using the following formula: in, This is the updated credibility weight coefficient. λ represents the current credibility weight coefficient, and λ is a preset decay factor.

[0151] The calculated updated credibility weight coefficient is compared with a preset minimum credibility threshold, which is set to 0.3, to define the policy failure boundary.

[0152] If the updated credibility weight coefficient is lower than the minimum credibility threshold, the credibility weight coefficient will be forcibly clamped to the minimum credibility threshold to prevent the weight from being too low, which could lead to numerical overflow or control logic disorder.

[0153] The finalized updated credibility weight coefficients are written to the designated address space of the non-volatile memory, thus completing the persistent state storage.

[0154] By using an exponential decay function, the abnormal triggering results of the previous step are transformed into a quantified credibility weight index, which enables rapid weight reduction of abnormal charging strategies. This ensures that the system can promptly trigger a conservative switching mechanism when the risk of strategy failure increases, thus protecting battery safety.

[0155] S6.5: Obtain the updated credibility weight coefficient and the preset mapping relationship table of adjacent healthy state intervals, and perform conservative strategy retrieval and replacement processing based on the fault fallback switching logic to generate a conservative anchor point sequence of adjacent healthy state intervals for taking over charging control.

[0156] The updated credibility weight coefficients and the preset adjacent health state interval mapping table are obtained as the input data source for the fault fallback switching logic. The adjacent health state interval mapping table is parsed to extract the conservative strategy candidate set corresponding to the current health state interval label. This candidate set includes anchor sequence identifiers bound to adjacent intervals with deeper or shallower aging. The final failure probability of the current dynamic anchor sequence is calculated based on the exponential decay function. If this probability exceeds a preset safety threshold, the conservative strategy retrieval instruction is activated. Anchor sequences with the lowest current slope and the widest voltage tolerance interval are selected from the conservative strategy candidate set as target takeover objects. The storage base address of the target takeover object in the segmented anchor template library is read, and its contained constraint operation instructions are loaded into the execution buffer. The loaded conservative anchor sequence undergoes integrity verification to ensure that all triggering logic conforms to the embedded system syntax specification. The pointer of the current charging control is redirected to the verified conservative anchor sequence, overwriting the original corrected dynamic anchor sequence configuration. By using fault fallback switching logic, the dynamic strategy that failed in the previous step due to response deviation is replaced with a conservative anchor point sequence of adjacent healthy state intervals, thereby achieving smooth takeover of charging control and battery safety protection.

[0157] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0158] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0159] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing charging strategies based on battery health and charging behavior analysis, characterized in that, Specifically, it includes: S1: Obtain the slope of internal resistance change, voltage plateau offset and temperature rise response hysteresis data of the battery in the most recent complete charge and discharge cycle, and perform normalization processing to generate a normalized feature vector. S2: Project the normalized feature vector onto a preset low-dimensional health state mapping space, and determine the health state interval label of the current battery based on the matching relationship between the projection coordinates and the electrochemical behavior directional boundary. S3: Analyze the user plugging and unplugging time series data within a preset time before this charging, identify the user charging behavior pattern label through a finite state machine, and jointly encode the health state interval label and the user charging behavior pattern label to generate a context-aware decision index. S4: Use the context-aware decision index to retrieve the segmented anchor template library and extract the dynamic anchor sequence that is bound to the health state interval label and contains several dynamic anchors, wherein the dynamic anchor contains constraint operation instructions and trigger logic within a specific SOC interval. S5: Calculate the actual temperature rise deviation and capacity replenishment efficiency deviation of the most recent preset number of charging cycles of the same type, generate an anchor point offset coefficient based on the actual temperature rise deviation and the capacity replenishment efficiency deviation, and apply the anchor point offset coefficient to the constraint operation instructions in the dynamic anchor point sequence to generate a corrected dynamic anchor point sequence.

2. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 1, characterized in that, Also includes: S6: Real-time adjustment of charging current and voltage is performed according to the modified dynamic anchor point sequence. During the charging process, the execution response data of each dynamic anchor point is continuously monitored. If the execution response data meets the preset abnormal conditions, the anchor point reliability decay mechanism is triggered and the sequence is switched to the conservative anchor point sequence of the adjacent healthy state interval.

3. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 2, characterized in that, The preset abnormal condition is: the execution response data deviates from the expected response threshold after a preset number of consecutive detections.

4. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 1, characterized in that, Step S2 specifically includes: Obtain the boundary parameters of multiple healthy state intervals calibrated based on electrochemical mechanisms contained in the low-dimensional healthy state mapping space, and use the boundary parameters of the healthy state intervals to construct a multi-dimensional feature space coordinate system to generate a healthy state interval partitioning model; The normalized feature vector is received, and a vector projection operation is performed based on the health state interval partitioning model to map the normalized feature vector to the multi-dimensional feature space coordinate system to generate real-time projection coordinate points representing the current state of the battery. Calculate the distance metric between the real-time projected coordinate point and the boundary parameters of each health state interval, and determine the target area range in which the real-time projected coordinate point falls based on the minimum distance principle, so as to generate a preliminary health state attribution result; Based on the preliminary health status attribution result, a preset health status interval label mapping table is retrieved, and the target area range is converted into the health status interval label.

5. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 4, characterized in that, Calculating the distance metric between the real-time projected coordinate point and the boundary parameters of each healthy interval includes: using the Euclidean distance formula to calculate the geometric distance between the real-time projected coordinate point and the centroid coordinates of each healthy interval, and obtaining the distance metric.

6. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 1, characterized in that, Step S4 specifically includes: The segmented anchor point template library is initialized and constructed to generate the initialized segmented anchor point template library; Based on the context-aware decision index, perform hash addressing retrieval processing on the segmented anchor template library to locate the storage address of the target dynamic anchor sequence; The target dynamic anchor sequence data is read from the segmented anchor template library according to the storage address to generate a dynamic anchor sequence object to be corrected; The structural integrity of each dynamic anchor point in the dynamic anchor point sequence object to be corrected is verified to output the dynamic anchor point sequence.

7. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 6, characterized in that, The segmented anchor point template library is initialized and constructed to generate an initialized segmented anchor point template library, including: Based on laboratory calibration data, a combination mapping relationship is defined between multiple health state interval labels and multiple user charging behavior pattern labels. Each combination mapping relationship is associated with a set of preset dynamic anchor point sequences, wherein each dynamic anchor point sequence consists of several dynamic anchor points arranged in order of charge state intervals, resulting in the initialized segmented anchor point template library with the joint encoding label as the unique retrieval key.

8. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 1, characterized in that, Step S5 specifically includes: Acquire measured temperature curve data and nominal temperature model data from the most recent preset number of charging processes of the same type, perform point-by-point difference calculation on the measured temperature curve data and the nominal temperature model data to generate actual temperature rise deviation, and extract the maximum positive deviation value in the actual temperature rise deviation as the key temperature rise feature quantity. Obtain the measured charging capacity data and theoretical target capacity data from the most recent preset number of similar charging processes. Calculate the ratio between the measured charging capacity data and the theoretical target capacity data to generate the capacity replenishment efficiency deviation. Combine the key temperature rise characteristic quantity with the capacity replenishment efficiency deviation to construct a multi-dimensional feedback error vector. The linear mapping lookup table is invoked, and the multidimensional feedback error vector is used as an index to input into the linear mapping lookup table. The corresponding initial adjustment gain parameter is output through table lookup interpolation, and the initial adjustment gain parameter is normalized to generate a standardized anchor offset coefficient. Read the original dynamic anchor sequence bound to the context-aware decision index in the segmented anchor template library, parse the constraint operation instructions contained in each dynamic anchor in the original dynamic anchor sequence, and use the standardized anchor offset coefficient to perform incremental correction calculation on the current threshold parameter and voltage duration parameter in the constraint operation instructions. The constraint operation instructions after the incremental correction operation are re-encapsulated into the corresponding dynamic anchor data structure, the trigger logic state bit of the dynamic anchor is updated to mark it as corrected, and finally the corrected dynamic anchor sequence containing the updated constraint operation instructions is output.

9. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 8, characterized in that, Normalizing the initial adjustment gain parameter to generate standardized anchor offset coefficients includes: linearly mapping the initial adjustment gain parameter to the [0,1] interval using Min-Max linear normalization.

10. The charging strategy optimization method based on battery health and charging behavior analysis according to claim 1, characterized in that, The preset time is 24 hours.

Images