A method and system for analyzing state trajectories of energy storage units

Abstract

The application discloses a kind of energy storage unit state trajectory analysis method and system, it is related to energy storage management field, to realize the advanced high-precision analysis early warning of energy storage unit nonlinear aging attenuation state.Multiple-dimensional time series sensing data in current operating cycle of energy storage unit is collected, and multiple-dimensional operating characteristics are extracted;Based on the operating characteristics, the health status benchmark point of the current operating cycle of the energy storage unit is evaluated, the attenuation trajectory of the future operating cycle is speculated, and the constraint condition of the attenuation trajectory of the energy storage unit is determined;The attenuation trajectory is corrected based on the constraint condition;Based on the corrected attenuation trajectory and the historical health state evolution trajectory, the state trajectory representing the attenuation state of the whole life cycle of the energy storage unit is constructed, which includes the state trajectory of the accelerated aging inflection point.The application can accurately capture the nonlinear accelerated aging inflection point while reflecting the high-fidelity mapping / reasoning of the SOH, RUL and other energy storage units in real time.

Description

Technical Field

[0001] This invention relates to the field of energy storage management technology, and in particular to a method and system for analyzing the state trajectory of an energy storage unit. Background Technology

[0002] With the widespread deployment of high-energy-density energy storage systems in new energy vehicles and megawatt-level energy storage power stations, accurate prediction of the system's health status throughout its entire lifecycle has become fundamental to ensuring the safe operation of the underlying system. Existing technologies typically employ purely data-driven time-series models, such as Recurrent Neural Networks (RNNs) or Long Short-Term Memory (LSTMs), for multi-step autoregressive calculations when extrapolating long-cycle degradation trajectories (e.g., Remaining Useful Life). However, these purely mathematical extrapolation mechanisms are fundamentally at odds with the complex physical laws governing electrochemical fatigue evolution. When pure data models perform dozens or even hundreds of iterations of prediction, small deviations in a single prediction step are rapidly amplified by the autoregressive mechanism, leading to severe error accumulation and prediction trajectory divergence.

[0003] On the other hand, the aging of energy storage units is a nonlinear evolution process driven by the coupling of multiple physical fields such as heat, electricity, and force. Although traditional physical mechanism models have clear evolution boundaries, the model parameters will drift irreversibly under alternating fatigue, making it difficult to achieve online long-term extrapolation.

[0004] There has long been a technological bias in this field, namely that pure mathematical time-series extrapolation and static physical equations cannot achieve deep dynamic integration in the absence of high-frequency calibration data. This directly results in existing methods being unable to detect early, latent decays in energy storage cells, such as microscopic lithium deposition and abnormal thickening of the solid electrolyte interphase (SEI) film, and even more difficult to capture the capacity degradation inflection point (i.e., the lifespan drop point) that occurs in the middle and later stages. This exposes the battery management system (BMS) to a significant risk of blind zone failure. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for analyzing the state trajectory of energy storage units, addressing all or part of the problems mentioned above, and to achieve advanced, high-precision analysis and early warning of the nonlinear aging and decay state of energy storage units.

[0006] The technical solution adopted in this invention is as follows: In a first aspect, this application provides a method for analyzing the state trajectory of an energy storage unit, comprising: S1. Collect multi-dimensional time-series sensing data in the current operating cycle of the energy storage unit, and extract the operating characteristics of the multi-dimensional time-series sensing data, the operating characteristics including macroscopic evolution characteristics and microscopic physical characteristics; S2. Based on the operating characteristics, assess the health status benchmark of the energy storage unit in the current operating cycle; S3. Based on the health status benchmark, predict the decay trajectory of the future operating cycle and determine the constraints on the decay trajectory of the energy storage unit. S4. Correct the predicted decay trajectory based on the constraints; based on the corrected decay trajectory and the historical health state evolution trajectory, construct a state trajectory characterizing the decay state of the energy storage unit throughout its entire life cycle, wherein the decay state covers the inflection point of nonlinear accelerated aging.

[0007] Optionally, multi-dimensional time-series sensing data can be collected during the current operating cycle of the energy storage unit, including: Based on the boundary constraints set for the state of charge of the energy storage unit, within the boundary constraints, multi-dimensional time-series sensing data in the current operating cycle is extracted using a sliding window; the multi-dimensional time-series sensing data is then preprocessed.

[0008] Optionally, methods for extracting the macroscopic evolution features of the multidimensional time-series sensing data include: The multidimensional time-series sensing data is subjected to differential transformation to calculate the incremental capacity parameter curve and the differential voltage parameter curve, respectively. The characteristic peaks and / or valleys of the incremental capacity parameter curve and the differential voltage parameter curve are located using a root-finding algorithm. The characteristic peaks and / or valleys, peak intensity and / or valley intensity, peak and / or valley drift in continuous cycles, and the envelope area of ​​characteristic peaks and / or valleys are extracted from the incremental capacity parameter curve and the differential voltage parameter curve to form the macroscopic evolution characteristics.

[0009] Optionally, methods for extracting the microscopic physical features of the multidimensional time-series sensing data include: The multidimensional time-series sensing data is input into a pre-constructed reduced-order electrochemical equivalent circuit model to identify the electrochemical parameters of the current operating cycle. The reduced-order electrochemical equivalent circuit model covers charge transfer and solid-phase diffusion mechanisms. The electrochemical parameters are mapped to microscopic physical features that characterize the source of degradation within the energy storage unit.

[0010] Optionally, assessing the health status benchmark of the energy storage unit during its current operating cycle includes: The operational characteristics are input into a pre-constructed data-physical information fusion network to obtain the health status benchmark. The data-physical information fusion network consists of a temporal feature extraction module connected to a nonlinear physical partial differential residual constraint branch; wherein: The temporal feature extraction module includes a cascaded bidirectional long short-term memory network, a multi-head self-attention layer, and a fully connected layer; The nonlinear physical partial differential residual constraint branch is the physical constraint on the capacity decay of the energy storage unit introduced by the time-series feature extraction module. The total loss of the data-physical information fusion network is composed of the weighted combination of the data loss of the time-series feature extraction module and the physical loss of the nonlinear physical partial differential residual constraint branch.

[0011] Optionally, the constraints on the decay trajectory of the energy storage unit are determined, including: Based on an empirical assessment method characterizing the capacity decay of energy storage units, the theoretical average capacity for future operating cycles is calculated using the health status benchmark as the initial capacity. Based on error propagation and Wiener process evolution characteristics, the empirical evaluation method is optimized; the theoretical capacity standard deviation for future operating cycles is calculated based on the optimized empirical evaluation method. Based on the theoretical capacity mean and the theoretical capacity standard deviation, the attenuation confidence envelope boundary of the energy storage unit attenuation trajectory is determined using the Laida criterion.

[0012] Optionally, the empirical evaluation method can be optimized based on error propagation and Wiener process evolution characteristics, including: The pre-exponential decay rate factor in the empirical evaluation method is modeled as a random variable following a Gaussian distribution; The system noise variable is introduced into the empirical evaluation method.

[0013] Optionally, the predicted decay trajectory is corrected based on the constraints, including: Determine whether the decay state of each future cycle in the predicted decay trajectory exceeds the decay confidence envelope boundary. If so, use the projection operator to calculate the compensation residual based on the decay confidence envelope boundary, and use the compensation residual to correct the decay state.

[0014] Optionally, based on the modified decay trajectory and historical health state evolution trajectory, a state trajectory characterizing the decay state of the energy storage unit throughout its entire life cycle is constructed, including: By connecting the decay states of each future operating cycle, a decay trajectory from the current operating cycle to the end of the lifespan of the energy storage unit can be obtained. The state trajectory is obtained by splicing the decay trajectory with the historical health state evolution trajectory of the energy storage unit before the current operating cycle.

[0015] In a second aspect, this application provides an energy storage unit state trajectory analysis system, which includes a processor and a storage medium; the storage medium stores a computer program, and the processor runs the computer program to execute the above-described energy storage unit state trajectory analysis method.

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: The energy storage unit state trajectory analysis scheme provided in this application collects multi-dimensional time-series sensing data of the energy storage unit and extracts multi-dimensional features. By combining pure mathematical time-series extrapolation with static physical information constraints, it achieves multi-step recursive prediction of the decay state of future operating cycles (the decay state of multiple operating cycles constitutes the decay trajectory). While performing multi-step recursion, it combines the constraints determined by the energy storage unit decay trajectory to correct the decay state of each predicted future operating cycle in real time, forcibly embedding it into the correct evolution range, blocking the accumulation error of time-series inference, and improving the accuracy of decay trajectory inference. By constructing a state trajectory that reflects the decay state throughout the entire life cycle, it can not only reflect the high-fidelity mapping / inference of the energy storage unit's SOH, RUL, etc. in real time, but also accurately capture the inflection point of nonlinear accelerated aging (i.e., the life drop point), which facilitates early response to the failure risk of the energy storage unit and can improve the long-term security of BMS for energy storage unit management. Attached Figure Description

[0017] The present invention will be described by way of example and with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart of the implementation phase of the energy storage unit state trajectory analysis method.

[0018] Figure 2 This is a flowchart illustrating the implementation of the macroscopic evolution feature extraction method.

[0019] Figure 3 This is a flowchart illustrating the implementation of the microscopic physical feature extraction method.

[0020] Figure 4 This is the network structure diagram of the temporal feature extraction module.

[0021] Figure 5 This is a flowchart of the method for determining the attenuation trajectory constraints of an energy storage unit. Detailed Implementation

[0022] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0023] Any feature disclosed in this specification (including any appended claims and abstract) may be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0024] To address the issues of severe cumulative errors and divergent decay trajectories in known energy storage unit (ESU) degradation state inference methods, as well as the inability to effectively capture the inflection point of nonlinear accelerated aging of ESU units, this application proposes an ESU state trajectory analysis method and system. The aim is to improve the accuracy of nonlinear aging degradation state analysis of ESU units and to provide early warning of aging inflection points in ESU capacity. The ESU unit referred to in this application refers to electrochemical energy storage facilities such as batteries capable of cyclic charging and discharging, such as lithium batteries.

[0025] like Figure 1 As shown, the energy storage unit state trajectory analysis method proposed in this application includes the following stages: S1. Collect multi-dimensional time-series sensing data during the current operating cycle of the energy storage unit and extract the operating characteristics of the multi-dimensional time-series sensing data.

[0026] The multidimensional time-series sensing data collected from the energy storage unit can include voltage V, current I, temperature T, etc. Each operating (charge and discharge) cycle has the same duration. By using a sliding window of fixed width from the continuously collected multidimensional time-series sensing data, a local multidimensional time-series sensing data segment corresponding to the current operating cycle can be extracted.

[0027] In addition, this application also considers avoiding the problems of dependence on the entire life cycle of energy storage units and polarization noise interference at both ends. Therefore, in an optional implementation, in the multi-dimensional time-series sensing data acquisition step of the S1 stage, a boundary constraint condition for the state of charge (SOC) of the energy storage unit is set. For example, the absolute range of 25% to 80% of the state of charge of the energy storage unit is set as the boundary constraint condition, and the multi-dimensional time-series sensing data within this boundary constraint condition is collected for subsequent analysis.

[0028] S11. Within the set boundary constraints, a sliding window is used to extract local multidimensional time-series sensing data segments to obtain the multidimensional time-series sensing data for the current operating cycle. The multidimensional time-series sensing data is collected according to a predetermined time step (corresponding to the acquisition frequency of the corresponding sensor), and therefore belongs to a discrete sequence, which has time-series characteristics.

[0029] S12. Preprocess the extracted multidimensional time-series sensing data.

[0030] The preprocessing includes denoising the multidimensional time-series sensor data, such as performing polynomial smoothing (e.g., using five-point cubic smoothing, least squares method, etc.). Furthermore, to facilitate subsequent processing, the multidimensional time-series sensor data can also be dimensionless.

[0031] The multidimensional time-series sensing data obtained above is used to extract multidimensional operational features, which include macroscopic evolution features and microscopic physical features. Macroscopic evolution features characterize the external parameters of the macroscopic nonlinear decay of the energy storage unit; microscopic physical features reflect the intrinsic parameters of degradation within the energy storage unit.

[0032] In one alternative implementation, such as Figure 2 As shown, methods for extracting macroscopic evolutionary features from multidimensional time-series sensing data include: S13. Perform differential transformation on the multidimensional time-series sensing data, and calculate the incremental capacity parameter curve and the differential voltage parameter curve respectively.

[0033] The incremental capacity parameter is represented by IC, and its calculation method is as follows: ; In the formula, k represents the cycle index; L represents the time distance between adjacent cycles, i.e., the sliding window step size; These represent the energy storage unit capacity in the k-th operating cycle and the energy storage unit capacity in the kL-th operating cycle, respectively. These represent the energy storage unit voltage in the kth operating cycle and the energy storage unit voltage in the kLth operating cycle, respectively.

[0034] The differential voltage parameter is denoted as DV, and its calculation method is as follows: .

[0035] By continuously calculating the incremental capacity parameter and differential voltage parameter for each operating cycle, the incremental capacity parameter curve and differential voltage parameter curve are obtained.

[0036] S14. Locate the characteristic peaks and / or characteristic valleys of the incremental capacity parameter curve and the differential voltage parameter curve using the root-finding algorithm.

[0037] By setting the differential derivative to zero, the coordinates of the characteristic peak and / or characteristic valley can be located in the incremental capacity parameter curve and the differential voltage parameter curve, respectively.

[0038] S15. Extract the characteristic peaks and / or valleys, peak intensity and / or valley intensity, peak and / or valley drift in continuous cycles, and the envelope area of ​​characteristic peaks and / or valleys from the incremental capacity parameter curve and differential voltage parameter curve to form macroscopic evolution characteristics.

[0039] After locating the coordinates of the characteristic peak (and similarly for the characteristic valley; for brevity, only the characteristic peak is used as an example here), the characteristic peak position, peak intensity, peak position shift during continuous cycles, and characteristic peak envelope area are extracted from the incremental capacity parameter curve and the differential voltage parameter curve, respectively. Since the same method is used for feature extraction in each operating cycle k, multiple operating cycles constitute a continuous loop.

[0040] In one alternative implementation, such as Figure 3 As shown, methods for extracting microscopic physical features from multidimensional time-series sensing data include: S16. Input the multi-dimensional time-series sensing data into the pre-built reduced-order electrochemical equivalent circuit model to identify the electrochemical parameters of the current operating cycle.

[0041] The constructed reduced-order electrochemical equivalent circuit model covers charge transfer and solid-phase diffusion mechanisms, which reflects the voltage change of the energy storage unit under the influence of polarization when current flows through it.

[0042] Specifically, the physical equation expression fitted by the least squares estimation method for the reduced-order electrochemical equivalent circuit model is as follows: ; In the formula, V and I represent the voltage and current at time step t, respectively; OCV represents the open circuit voltage of the energy storage unit, which is the voltage of the energy storage unit when there is no load current, and it reflects the voltage level corresponding to the chemical energy inside the energy storage unit. This refers to the polarization internal resistance parameter of the energy storage unit; This refers to the polarization capacitor parameter of the energy storage unit.

[0043] By substituting the voltage V and current I from multiple time steps in the multidimensional time-series sensing data into the physical equations of the aforementioned reduced-order electrochemical equivalent circuit model, real-time parameter optimization is performed to identify the electrochemical parameters of the current operating cycle, including: polarization internal resistance parameters. Polarization capacitance parameters and ohmic impedance parameters ( The aforementioned , which is the equivalent impedance that characterizes the ohmic polarization characteristics of the current collector, electrolyte, and solid electrolyte interface membrane inside the energy storage unit; this parameter is obtained by substituting multidimensional time-series sensing data within a local sliding window into the above model and identifying it through overall fitting.

[0044] S17. Map electrochemical parameters to microscopic physical characteristics that characterize the source of degradation within the energy storage unit.

[0045] Based on the identification of electrochemical parameters, they are mapped to microscopic physical characteristics that reflect the internal degradation source of the energy storage unit.

[0046] Specifically: the identified ohmic impedance parameters ( The increase relative to the initial cycling baseline value is mapped to the thickening characteristics of the solid electrolyte interface film; the polarization internal resistance parameter is then used. and polarization capacitance parameters The change relative to the initial cycle baseline (i.e., parameter drift) is mapped to the characteristics of available lithium-ion loss and active material loss.

[0047] Quantitative mapping of ohmic impedance to solid electrolyte interfacial film thickening characteristics: ohmic impedance of the battery ( The increase in ohmic impedance is mainly due to the continuous thickening of the SEI film. The increase in ohmic impedance parameter Δ relative to the initial cycle reference value during the current operating cycle is extracted. R SEI By combining the intrinsic conductivity parameters of the SEI film, it is directly mapped and calculated as the thickness increase of the SEI film. The specific mapping conversion formula is as follows: Δ δ= Δ R SEI × SEI ; in, SEI The intrinsic conductivity of the solid electrolyte interface membrane of the energy storage unit is pre-calibrated.

[0048] Mapping of polarization parameters to available lithium-ion loss (LLI) and active material loss (LAM): Polarization internal resistance parameter Characterized the charge transfer resistance ( R ct Evolution of solid-state diffusion impedance and polarization capacitance parameters The double-layer effect and diffusion capacitance were characterized. Restricted charge transfer hinders redox reactions, causing some electrodes to lose electrochemical activity, leading to the loss of active material. Increased polarization and SEI thickening jointly trap reversible lithium ions, resulting in the loss of usable lithium ions. The system extracts the ratio of the changes in polarization resistance and polarization capacitance parameters relative to the initial cycling baseline (i.e., parameter drift), using this as a dimensionless characteristic factor to characterize the severity of usable lithium ion loss and active material loss. Finally, the parameter drift is compared with the aforementioned SEI film thickening. Vector concatenation is performed to form multidimensional microscopic physical features (vectors) that are input into the subsequent data-physical information fusion network.

[0049] In this way, the dual-drive feature reconstruction, encompassing both external data appearance and internal physical mechanism, is completed.

[0050] S2. Based on the extracted operating characteristics, assess the State of Health (SOH) baseline of the energy storage unit in the current operating cycle.

[0051] As an optional implementation, a data-physical information fusion network is used to assess the health status benchmark.

[0052] Specifically, operational characteristics (including spliced ​​macroscopic evolutionary characteristics and microscopic physical characteristics) are input into a pre-constructed data-physical information fusion network to obtain a health status benchmark.

[0053] In one alternative implementation, the data-physical information fusion network consists of a temporal feature extraction module (as the backbone network) connected to a nonlinear physical partial differential residual constraint branch.

[0054] (1) Temporal feature extraction module.

[0055] like Figure 4 As shown, the temporal feature extraction module includes a cascaded bidirectional long short-term memory network (Bi-LSTM), a multi-head self-attention layer (Mul-Attention), and a fully connected layer (FC).

[0056] Bidirectional Long Short-Term Memory (LSTM) networks consist of two LSTM layers: a forward LSTM and a backward LSTM. The LSTM utilizes an internal gate control mechanism (input gate). Forgotten Gate and output gate The updated state characteristics of the energy storage unit are represented as follows: ; In the formula, These represent the health status at time step t and its previous time step t-1, respectively. These represent the weight matrix and bias term of the output gate, respectively.

[0057] The running features are input into the Bi-LSTM. At each time step, the state features output by the forward LSTM and the backward LSTM are concatenated or summed to form the final state feature of that time step, which is then used as the input feature for the next time step.

[0058] The final state features from the last time step are input into the multi-head self-attention layer. The multi-head self-attention layer employs a multi-head self-attention mechanism for feature enhancement. Specifically, the multi-head self-attention layer... Capture the global long-range dependencies of the final state features of the Bi-LSTM output, where These are the query matrix, key matrix, and value matrix calculated based on the final state features of the input, respectively. is the feature dimension, and softmax is the activation function.

[0059] After feature enhancement by the multi-head self-attention layer, the final predicted health status of the energy storage unit in the current operating cycle k is driven by the output data of the fully connected layer. If the previous running cycle is k-1, its predicted health status value is similarly expressed as... .

[0060] (2) Nonlinear physical partial differential residual constraint branch.

[0061] The nonlinear physical partial differential residual constraint branch introduces physical constraints on the capacity decay of energy storage units into the time-series feature extraction module.

[0062] A nonlinear empirical formula for capacity decay based on Arrhenius's law is introduced to characterize the physical mechanism of (cumulative) capacity decay in energy storage units. Its theoretical decay rate is expressed as: ; In the formula, This represents the capacity decay rate over operating cycle k. All are empirical attenuation coefficients. Activation energy for energy storage units; The temperature of the energy storage unit during operating cycle k; It is the ideal gas constant, usually taken as 8.314 J / mol·K.

[0063] By extracting and inferring features from multiple operating cycles, the time-difference of the data-physical information fusion network output is obtained, and the physical partial differential residual (i.e., the network's physical loss) is calculated. ; In the formula, This represents physical loss, where N is the total number of operating cycles from the current operating cycle k.

[0064] For the temporal feature extraction module of the data-physical information fusion network, its training data loss Represented as: ; In the formula, M represents the number of samples, and i is the sample index; This represents the predicted health status value of the i-th sample in running period k; This represents the true health status of the i-th sample within the same running cycle.

[0065] Total loss of data-physical information fusion network Data loss due to the time-series feature extraction module Physical loss of nonlinear physical partial differential residual constraint branch Weighted combination.

[0066] Specifically: ; In the formula, These are noise weight parameters that are dynamically updated during training.

[0067] By training and updating the network parameters on the training samples, when on the validation set... Training stops when the change over a predetermined number of consecutive iterations falls below the set convergence threshold or the maximum number of iterations is reached. The final output, constrained by nonlinear physical partial differential residuals, represents a baseline value for the current operating cycle's health status, exhibiting extremely high physical consistency. This provides an absolute initial anchor point for subsequent long-term decay trajectory prediction.

[0068] S3. Based on the above health status benchmark, predict the decay trajectory of future operating cycles and determine the constraints on the decay trajectory of the energy storage unit.

[0069] S31. Predict the decay trajectory of future operating cycles.

[0070] The decay trajectory is a time series consisting of the decay state (also known as the health state) of multiple consecutive operating cycles. Each time, the decay state of a single operating cycle is predicted. The decay trajectory of the future operating cycles is formed by the continuously predicted decay states of multiple future operating cycles, such as from the next operating cycle of the current operating cycle until the end of the life of the energy storage unit (EOL, End of Life, such as 25% capacity).

[0071] In one alternative implementation, the decay trajectory of future operating cycles is inferred through a time-series autoregressive module.

[0072] The temporal autoregressive module backbone uses a Long Short-Term Memory (LSTM) network. The gating mechanism of LSTM has been described in detail previously and will not be repeated here. The number of hidden layer nodes is set to... A Dropout layer is introduced to prevent overfitting. The health status benchmark is the current running period k output in stage S2. To determine the starting point, a rolling "time window" mechanism is used to perform multi-step forward prediction.

[0073] For the time-series autoregressive module, its predicted health status value for the m-th future running cycle is expressed as: The health status prediction value of the previous operating cycle and the hidden state of LSTM As a predicted value for the health status in the next operating cycle The input is used to recursively solve for the predicted health status value for the next running cycle. This enables purely data-driven autoregressive prediction of future nonlinear decay trajectories.

[0074] The health status prediction value inferred by the time-series autoregressive module has a cumulative error characteristic. In this embodiment, the cumulative error is blocked by setting constraints on the energy storage unit's decay trajectory. This constraint can be performed synchronously with the health status prediction value inference process.

[0075] In one alternative implementation, such as Figure 5 As shown, the methods for determining the constraints on the decay trajectory of an energy storage unit include: S32. Based on an empirical assessment method that characterizes the capacity decay of energy storage units, the theoretical average capacity for future operating cycles is calculated using the health status benchmark as the initial capacity.

[0076] The so-called empirical assessment method for characterizing the capacity degradation of energy storage units can be evaluated using the Arrhenius power-law degradation model, which characterizes the cumulative fatigue of energy storage units. This model calculates the theoretical average capacity of the energy storage unit over its future operating cycles k+m. The evolution formula is: ; In the formula, This represents the initial capacity of the energy storage unit, that is, the capacity of the energy storage unit in the current operating cycle k; Indicates the pre-exponential degradation rate factor; This represents the absolute temperature of the energy storage unit during the estimated operating cycle k+m; An aging time index representing the diffusion limitation within an energy storage unit.

[0077] S33. Based on error propagation and Wiener process evolution characteristics, optimize the empirical evaluation method; calculate the theoretical capacity standard deviation for future operating cycles based on the optimized empirical evaluation method.

[0078] Since purely empirical evaluation methods can only output a theoretical deterministic mean, without considering the impact of dynamic environments on capacity, this application also simulates and introduces a parameter uncertainty quantification mechanism to help define the hard boundary of the decay confidence envelope.

[0079] In one optional implementation, the method for optimizing the empirical evaluation method includes: The pre-finite decay rate factor in empirical evaluation methods The model is a random variable that follows a Gaussian distribution, and its variance is expressed as... .

[0080] In empirical evaluation methods, a system noise variable is introduced, and its variance is expressed as: .

[0081] Based on this, the theoretical capacity standard deviation for the (k+m)th operating cycle is calculated. The method is as follows: .

[0082] S34, Based on theoretical capacity average and theoretical capacity standard deviation The Raida criterion (also known as the Raida criterion) is adopted. (Accurate measurement) to determine the attenuation confidence envelope boundary of the attenuation trajectory of the energy storage unit.

[0083] The upper boundary of the attenuation confidence envelope is represented as The calculation method is as follows: ; The lower boundary of the attenuation confidence envelope is represented as The calculation method is as follows: .

[0084] At this point, the constraints for the decay trajectory have been determined.

[0085] S4. Correct the predicted decay trajectory based on the constraints; based on the corrected decay trajectory and the historical health state evolution trajectory, construct a state trajectory characterizing the decay state of the energy storage unit throughout its entire life cycle.

[0086] As an optional implementation, correcting the predicted decay trajectory based on constraints includes: S41. Determine whether the attenuation state of each future running cycle in the predicted attenuation trajectory exceeds the attenuation confidence envelope boundary (including the upper and lower boundaries). If so, use the projection operator to calculate the compensation residual based on the attenuation confidence envelope boundary, and use the compensation residual to correct the attenuation state.

[0087] For the original health status prediction value of the time series autoregressive module with a running period of k+m Real-time judgment of whether it is higher than Or is it lower than If so, the network residual compensation mechanism will be triggered. Force projection back into the attenuation confidence envelope boundary.

[0088] The method for calculating the compensation residual using the projection operator is as follows: ; In the formula, This represents the calculated compensation residual.

[0089] Using compensated residuals Correction : .

[0090] In the formula, This represents the revised predicted health status value.

[0091] The predicted health status values ​​are inferred and corrected in each running cycle using a recursive approach. ,Will Hidden state corresponding to the running cycle As input to the time-series autoregressive module, it is used to infer the predicted health status value for the next running cycle. The system then judges and corrects the attenuation based on the attenuation confidence envelope boundary. This process is repeated in a loop, executing a closed-loop iterative process of single-cycle prediction, confidence envelope boundary judgment, residual compensation correction, and attenuation state feedback, until the end-of-life point of the energy storage unit is predicted (e.g., setting 80% of the nominal capacity as the end-of-life point), triggering a stop command. This completes the prediction and correction of the attenuation state (i.e., the attenuation trajectory of future operating cycles) for all future operating cycles starting from the current operating cycle k.

[0092] S42. Based on the modified decay trajectory and historical health state evolution trajectory, construct a state trajectory that characterizes the decay state of the energy storage unit throughout its entire life cycle.

[0093] By connecting the decay states of each future operating cycle using the above method, the decay trajectory from the current operating cycle k to the end of the energy storage unit's lifespan can be obtained.

[0094] By concatenating this decay trajectory with the historical health state evolution trajectory of the energy storage unit prior to the current operating cycle k, a state trajectory (in curve form) characterizing the decay state of the energy storage unit throughout its entire life cycle is obtained. The decay states in the state trajectory encompass the inflection point of nonlinear accelerated aging.

[0095] For the inflection point of nonlinear accelerated aging, in one optional implementation, a discrete difference and curvature analysis algorithm is applied to the state trajectory to calculate the curvature value of each sequence node (i.e., the running cycle) on the curve. In the formula: : Represents the first derivative of the state trajectory curve at the current operating cycle (obtained in the discrete sequence through first-order backward difference or central difference), and its physical meaning is: the rate of change (or degradation slope) of the energy storage unit's state of health (SOH) with the operating cycle. : Represents the second derivative of the state trajectory curve at the current operating cycle (obtained through second-order difference in the discrete sequence), and its physical meaning is: the rate of change of the health state decay rate of the energy storage unit, that is, the accelerated evolution trend of nonlinear degradation.

[0096] The point of maximum curvature response is located at the inflection point of nonlinear accelerated aging caused by internal lithium plating or a surge in polarization impedance.

[0097] Ultimately, we can obtain the complete remaining useful life cycles, the full life cycle state trajectory, and the advanced prediction of the nonlinear accelerated aging inflection point.

[0098] The algorithm models for data acquisition, feature extraction, and attenuation trajectory prediction in stages S1 to S4 are deployed to application systems such as vehicle BMS and edge computing units. The system supports caching historical data in edge devices and performing local attenuation trajectory (covering SOH and RUL) offline prediction to ensure the reliability of the evaluation under network latency conditions.

[0099] Compared with traditional energy storage unit degradation state inference methods, this application has the following significant advantages: 1. A quantitative closed loop for error convergence boundary is achieved, resulting in an order-of-magnitude improvement in long-range prediction fidelity: Traditional pure data-driven models often exhibit exponential divergence from small initial errors when performing long-range autoregressive extrapolation. This application significantly compresses the divergence space of the predicted trajectory by using electrochemical partial differential equations and empirical evaluation methods as underlying hard constraints. Experimental verification shows that this application only requires a very small amount of early cycle data (specifically, voltage, current, and temperature time-series signals acquired online during the first 20-60 charge-discharge cycles (i.e., operating cycles) of the energy storage unit before commissioning; the model input samples are extracted within a specific state of charge range (such as the aforementioned 25%-80% nominal capacity), and reconstructed through a reduced-order physical model to obtain macroscopic IC / DV peak drift characteristics and microscopic ohmic / polarization impedance multi-dimensional fused feature vectors; during the offline training phase of the model, the early SOH benchmark true value and the full lifecycle actual capacity decay sequence until the end of life (EOL) are used as supervision labels), which can strictly control the mean absolute percentage error (MAPE) of long-duration remaining lifetime extrapolation within the range of 0.2% to 1.5%. Even in the later stages of prediction, the peak value of the mean absolute percentage error (MAPE) can be stabilized at around 1.53%. Compared to traditional time series network models, the root mean square error (RMSE) of this application is reduced by 2 to 6 times, completely cutting off the error divergence path and giving the model extremely high physical consistency and long-term prediction fidelity.

[0100] 2. Overcoming technical biases, achieving advanced and precise positioning and high-sensitivity early warning of capacity aging inflection points: Addressing the critical limitation of pure data models in failing to capture the late-stage nonlinear accelerated aging (lifespan drop-off point) of energy storage units in advance, this application deeply decouples prior physical information such as IC / DV characteristics and impedance evolution at the input end. The model can accurately predict the "lifespan drop-off point" in the later stages of lifespan even with only the first 50 early, minute-scale cycles of data. For conventional power batteries, this can trigger an early warning mechanism hundreds of cycles in advance; simultaneously, the root mean square error of the model's prediction of the drop-off point occurrence cycle can be strictly controlled within 50 cycles, with a prediction error rate of less than 1%, providing an extremely ample decision-making time window and high-confidence safety guarantee for system-level preventative maintenance.

[0101] Based on the inventive concept of this application, this application also provides an energy storage unit state trajectory analysis system, which includes a processor and a storage medium. The storage medium stores a computer program, and the processor runs the computer program to execute the above-described energy storage unit state trajectory analysis method.

[0102] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. A method for analyzing the state trajectory of an energy storage unit, characterized in that, include: S1. Collect multi-dimensional time-series sensing data in the current operating cycle of the energy storage unit, and extract the operating characteristics of the multi-dimensional time-series sensing data, the operating characteristics including macroscopic evolution characteristics and microscopic physical characteristics; S2. Based on the operating characteristics, assess the health status benchmark of the energy storage unit in the current operating cycle; S3. Based on the health status benchmark, predict the decay trajectory of the future operating cycle and determine the constraints on the decay trajectory of the energy storage unit. S4. Correct the predicted decay trajectory based on the constraints; based on the corrected decay trajectory and the historical health state evolution trajectory, construct a state trajectory characterizing the decay state of the energy storage unit throughout its entire life cycle, wherein the decay state covers the inflection point of nonlinear accelerated aging.

2. The energy storage unit state trajectory analysis method as described in claim 1, characterized in that, Collect multi-dimensional time-series sensing data during the current operating cycle of the energy storage unit, including: Based on the boundary constraints set for the state of charge of the energy storage unit, within the boundary constraints, multi-dimensional time-series sensing data in the current operating cycle is extracted using a sliding window; the multi-dimensional time-series sensing data is then preprocessed.

3. The energy storage unit state trajectory analysis method as described in claim 1, characterized in that, The method for extracting the macroscopic evolution features of the multidimensional time-series sensing data includes: The multidimensional time-series sensing data is subjected to differential transformation to calculate the incremental capacity parameter curve and the differential voltage parameter curve, respectively. The characteristic peaks and / or valleys of the incremental capacity parameter curve and the differential voltage parameter curve are located using a root-finding algorithm. The characteristic peaks and / or valleys, peak intensity and / or valley intensity, peak and / or valley drift in continuous cycles, and the envelope area of ​​characteristic peaks and / or valleys are extracted from the incremental capacity parameter curve and the differential voltage parameter curve to form the macroscopic evolution characteristics.

4. The energy storage unit state trajectory analysis method as described in claim 1, characterized in that, Methods for extracting the microscopic physical features of the multidimensional time-series sensing data include: The multidimensional time-series sensing data is input into a pre-constructed reduced-order electrochemical equivalent circuit model to identify the electrochemical parameters of the current operating cycle. The reduced-order electrochemical equivalent circuit model covers charge transfer and solid-phase diffusion mechanisms. The electrochemical parameters are mapped to microscopic physical features that characterize the source of degradation within the energy storage unit.

5. The energy storage unit state trajectory analysis method as described in claim 1, characterized in that, The baseline for assessing the health status of the energy storage unit during its current operating cycle includes: The operational characteristics are input into a pre-constructed data-physical information fusion network to obtain the health status benchmark. The data-physical information fusion network consists of a temporal feature extraction module connected to a nonlinear physical partial differential residual constraint branch; wherein: The temporal feature extraction module includes a cascaded bidirectional long short-term memory network, a multi-head self-attention layer, and a fully connected layer; The nonlinear physical partial differential residual constraint branch is the physical constraint on the capacity decay of the energy storage unit introduced by the time-series feature extraction module. The total loss of the data-physical information fusion network is composed of the weighted combination of the data loss of the time-series feature extraction module and the physical loss of the nonlinear physical partial differential residual constraint branch.

6. The energy storage unit state trajectory analysis method as described in claim 1, characterized in that, Determine the constraints on the attenuation trajectory of the energy storage unit, including: Based on an empirical assessment method characterizing the capacity decay of energy storage units, the theoretical average capacity for future operating cycles is calculated using the health status benchmark as the initial capacity. Based on error propagation and Wiener process evolution characteristics, the empirical evaluation method is optimized; the theoretical capacity standard deviation for future operating cycles is calculated based on the optimized empirical evaluation method. Based on the theoretical capacity mean and the theoretical capacity standard deviation, the attenuation confidence envelope boundary of the energy storage unit attenuation trajectory is determined using the Laida criterion.

7. The energy storage unit state trajectory analysis method as described in claim 6, characterized in that, Based on error propagation and Wiener process evolution characteristics, the empirical evaluation method is optimized, including: The pre-exponential decay rate factor in the empirical evaluation method is modeled as a random variable following a Gaussian distribution; The system noise variable is introduced into the empirical evaluation method.

8. The energy storage unit state trajectory analysis method as described in claim 6, characterized in that, The predicted attenuation trajectory is corrected based on the aforementioned constraints, including: Determine whether the decay state of each future cycle in the predicted decay trajectory exceeds the decay confidence envelope boundary. If so, use the projection operator to calculate the compensation residual based on the decay confidence envelope boundary, and use the compensation residual to correct the decay state.

9. The energy storage unit state trajectory analysis method as described in claim 8, characterized in that, Based on the corrected decay trajectory and historical health state evolution trajectory, a state trajectory characterizing the decay state of the energy storage unit throughout its entire life cycle is constructed, including: By connecting the decay states of each future operating cycle, a decay trajectory from the current operating cycle to the end of the lifespan of the energy storage unit can be obtained. The state trajectory is obtained by splicing the decay trajectory with the historical health state evolution trajectory of the energy storage unit before the current operating cycle.

10. A state trajectory analysis system for an energy storage unit, characterized in that, It includes a processor and a storage medium; the storage medium stores a computer program, and the processor runs the computer program to perform the energy storage unit state trajectory analysis method as described in any one of claims 1-9.

Images