Parallel battery system multi-mechanism degradation modeling and health state estimation method

Abstract

The invention discloses a parallel battery system multi-mechanism degradation modeling and health state estimation method, and belongs to the technical field of lithium ion battery health management. The method comprises the steps that a teacher model is constructed, the teacher model generates health state pseudo labels and soft constraints based on a multi-mechanism physical model, and the multi-mechanism physical model comprises a thermal degradation model, a cyclic degradation model, a calendar aging model, a parallel system effect model and mechanism coupling terms; a student model is constructed, the student model integrates various machine learning algorithms, and training is carried out through a teacher-student training process under the guidance of the teacher model so as to estimate the health state of the parallel battery system; the machine learning algorithm comprises a random forest algorithm, a gradient boosting algorithm, a support vector regression algorithm and a lightweight multilayer perceptron; and performing real-time or offline estimation on the health state of the parallel battery system based on the trained student model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lithium ion battery health management, and particularly relates to a parallel battery system multi-mechanism degradation modeling and health state estimation method. BACKGROUND

[0002] Lithium ion batteries face complex working conditions such as high-frequency power cycling, wide temperature range changes and system imbalance in electric aviation, and traditional single mechanism models or pure data-driven methods are difficult to accurately estimate the health state thereof. Existing researches mainly focus on single battery degradation mechanisms (such as Arrhenius thermal aging, rainflow counting cycle aging, calendar aging diffusion model) or simple equivalent circuit models, and lack of systematic modeling capability for multi-mechanism coupling such as current imbalance, capacity mismatch and thermal coupling in parallel systems.

[0003] In the prior art, some researches use physical models (such as Arrhenius equation, rainflow counting method) or data-driven methods (such as support vector regression, random forest) to estimate the health of the battery. For example, some methods use double Kalman filtering for parameter identification, or use long short-term memory network for time series prediction. However, these methods either rely on accurate physical parameters (which are difficult to obtain) or require a large amount of labeled data (which is not available in practice), and fail to effectively integrate multi-mechanism coupling and system-level parallel effects.

[0004] To solve the above problems, the present application provides a parallel battery system multi-mechanism degradation modeling and health state estimation method. SUMMARY

[0005] The present application aims to provide a parallel battery system multi-mechanism degradation modeling and health state estimation method to solve the problem that traditional battery health estimation methods are difficult to accurately model multi-mechanism coupling degradation (such as thermal stress, cycle aging, calendar aging, parallel system imbalance, etc.) in parallel systems, and lack direct battery health state capacity measurement labels in actual operation, resulting in low estimation accuracy and poor generalization ability, and failing to provide reliable and interpretable optimization strategies for the battery management system.

[0006] To solve the above technical problems, the present application adopts the following technical solutions: A parallel battery system multi-mechanism degradation modeling and health state estimation method, comprising the following steps: A teacher model is constructed, which takes a multi-mechanism physical model as the core and is used to generate health state pseudo-labels and soft constraints. The multi-mechanism physical model covers five key degradation mechanisms: 1) Thermal degradation model: based on the acceleration effect of temperature on the internal electrochemical side reactions of the battery (such as solid electrolyte interface growth, active material dissolution), an Arrhenius equation is constructed to describe the relationship between temperature and degradation rate; 2) Cycle degradation model: Combined with the rainflow counting method and the Palmgren-Miner cumulative damage rule, the different effects of different charge and discharge depths (shallow cycle, medium cycle, deep cycle) on battery degradation are distinguished; 3) Calendar aging model: Based on the material diffusion theory, two key acceleration factors of state of charge and temperature are introduced to describe the aging law of the battery in the static storage process; 4) Parallel system effect model: For the characteristics of parallel battery system, the interaction between single batteries is described, including uneven current distribution, capacity mismatch, thermal coupling and cycle out-of-sync; 5) Mechanism coupling term: Quantify the nonlinear interaction between different degradation mechanisms, including the coupling relationship between thermal degradation and cycle degradation, cycle degradation and calendar aging, and thermal degradation and calendar aging.

[0007] A student model is constructed, which is an integrated machine learning model integrating random forest algorithm, gradient boosting algorithm, support vector regression algorithm and lightweight multilayer perception. Through the teacher-student training process, the training is completed, and finally used for parallel battery system health state estimation. The teacher-student training process includes: 1) Data windowing: Time series features are extracted from real-time running data of battery system using sliding window to ensure data time series continuity and representativeness; 2) Pseudo-label generation: The total degradation of the battery is calculated by the teacher model, and the total degradation is converted into health state pseudo-label to provide training target for the student model; 3) Feature engineering: A multi-dimensional feature set is constructed, including statistical features (such as state of charge distribution, temperature statistical data), physical features (such as quantitative indicators of each degradation mechanism) and time series features (such as degradation trend coefficient, rolling fluctuation of operating parameters); 4) Soft constraint design: Two types of physical constraints are introduced to guide the training, one is monotonicity constraint (to ensure that the degradation degree increases when the temperature rises), and the other is directionality constraint (to ensure that the state of charge and calendar aging are positively correlated), to ensure that the model prediction conforms to the basic rules of electrochemistry; 5) Ensemble learning: The prediction results of multiple basic algorithm models are weighted and fused, and the weights are adaptively adjusted according to the performance of each model on the validation set to improve the overall prediction accuracy and stability.

[0008] Based on the trained student model, the health state of the parallel battery system can be estimated in real time or offline data analysis, and the health state quantitative results are output.

[0009] To verify the effectiveness of the battery management system intervention strategy, it also includes the counterfactual intervention analysis step: 1) Simulated intervention strategies: Three typical intervention programs were designed, including peak temperature reduction (inhibiting high-temperature accelerated degradation), upper limit of state of charge (slowing down calendar aging), and power control (as a negative control to verify causality). 2) Quantitative causal effects: Calculate the impact of intervention strategies on degradation, including individual treatment effects (degradation changes in a single sample), average treatment effects (average degradation changes in the overall sample), and quantile treatment effects (differences in effects among samples with different degrees of degradation). 3) Statistical confidence assessment: The Bootstrap resampling method was used to construct confidence intervals. The 95% confidence interval of the average treatment effect was calculated by repeated sampling (specifically 200 times) to verify the statistical significance of the intervention strategy.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention constructs a two-layer architecture with a multi-mechanism physical model as the "teacher" and an integrated machine learning model as the "student". The teacher model generates pseudo-labels based on the Arrhenius equation, rain flow counting method and parallel system dynamics, and applies monotonicity constraints to guide the student model to train in the absence of real labels, thus realizing the deep integration of physical knowledge and data-driven approaches. (2) This invention establishes a complete model that includes system-level effects such as thermal degradation, cycle degradation, calendar aging, current unevenness, and capacity mismatch. For the first time, it introduces coupling coefficients to quantify the three coupling effects of thermal-cycle, cycle-calendar, and thermal-calendar, thus solving the problem of multi-mechanism coupling modeling in parallel battery systems. (3) This invention innovatively introduces counterfactual analysis and Bootstrap resampling method. By calculating the 95% confidence interval of the average treatment effect through 200 repeated sampling, it provides a statistically reliable quantitative basis for BMS control thresholds such as temperature > 42°C and current deviation > 1.5A, realizing a technological leap from experience-based decision-making to data-driven decision-making. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings involved in the embodiments are now briefly described. Obviously, the drawings in the following description are merely illustrative of some embodiments of the present invention. For those skilled in the art, other forms of drawings can be constructed based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the system implementation process of a parallel battery system multi-mechanism degradation modeling and health state estimation method proposed in this invention. Detailed Implementation

[0013] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0014] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0015] This invention proposes a method for multi-mechanism degradation modeling and health state estimation of parallel battery systems. It aims to construct a physically constrained teacher-student learning framework that can accurately model and estimate the multi-mechanism degradation of parallel lithium-ion battery systems without direct capacity labels, and provide quantitative and verifiable control strategy suggestions for battery management systems.

[0016] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0017] Example 1: Please see Figure 1 This example proposes a method for multi-mechanism degradation modeling and health state estimation of parallel battery systems, including: A teacher model is constructed, which generates health status pseudo-labels and soft constraints based on a multi-mechanism physical model, wherein the multi-mechanism physical model includes a thermal degradation model, a cyclic degradation model, a calendar aging model, a parallel system effect model, and a mechanism coupling term; A student model is constructed, which integrates multiple machine learning algorithms and is trained through a teacher-student training process under the guidance of the teacher model to estimate the health status of the parallel battery system; the machine learning algorithms include random forest algorithm, gradient boosting algorithm, support vector regression algorithm, and lightweight multilayer perceptron. Based on the trained student model, the health status of the parallel battery system is estimated in real time or offline.

[0018] Based on the above methods, this example proposes a corresponding parallel battery system for multi-mechanism degradation modeling and health state estimation, combining... Figure 1 The content shown is explained in detail below.

[0019] 1. System Architecture Overview This example proposes a two-stage "teacher-student" framework: 1) Teacher Model: Based on a multi-mechanism physical model (thermal degradation, cyclic degradation, calendar aging, parallel system effects and their coupling), pseudo-labels and soft constraints are generated; 2) Student Model: Integrates multiple machine learning algorithms (random forest, gradient boosting, support vector regression, lightweight multilayer perceptron) to estimate health status under the guidance of the teacher model.

[0020] 2. Physical degradation modeling (teacher model) 1) Thermal degradation model: Based on the Arrhenius equation, considering the accelerating effect of temperature on the growth of the solid electrolyte interface film (SEI film) and the dissolution of active substances; the algorithm formula is as follows:

[0021] in, , It is the pre-exponential factor (normalized to 1). J / mol (SEI layer growth kinetics of NCM811 battery cell). J / (mol·K), Absolute temperature (Thermal degradation rate factor).

[0022] 2) Cyclic Degradation Model: Rainflow counting and Palmgren-Miner cumulative damage rules are used to distinguish between shallow, medium, and deep cycles; the algorithm formula is as follows:

[0023]

[0024] Among them, shallow loops (DOD<0.2) Next, in the middle loop (0.2≤DOD<0.5) Next, deep loop (DOD≥0.5): times; when At that time, the minimum loop depth threshold was adjusted from 0.1 to 0.15.

[0025] 3) Calendar Aging Model: Based on diffusion theory, it introduces the State of Charge (SOC) and temperature acceleration factor; the algorithm formula is as follows:

[0026] in, day -0.5 (Baseline calendar aging rate). (SOC stress factor) It is a temperature acceleration factor based on the Arrhenius model (isothermal degradation model).

[0027] 4) Parallel system effects: including current unevenness, capacity mismatch, thermal coupling, and cycle out-of-step; the algorithm formula is as follows: Current imbalance: ,

[0028] Capacity mismatch: ,

[0029] Thermal coupling: ,

[0030] Cyclic step loss: ,

[0031] 5) Mechanism Coupling Terms: Modeling the nonlinear interactions between heat-cycle, cycle-calendar, and heat-calendar. The algorithm formula is as follows:

[0032]

[0033]

[0034]

[0035] in: , .

[0036] 3. Teacher and Student Training Process 1) Data windowing: A 5000-point sliding window (approximately 15 minutes) is used to extract time-series features with a window overlap rate of 50% to avoid information leakage.

[0037] 2) Pseudo-label generation: The teacher model calculates the total degradation. Convert to a State of Health (SOH) pseudo-label using the following formula: .

[0038] 3) Feature engineering: Construct a 168-dimensional feature set, including statistical features (56 dimensions, such as SOC mean / variance, temperature distribution), physical features (84 dimensions, such as degradation components of various mechanisms), and time series features (28 dimensions, such as trend coefficient, rolling standard deviation).

[0039] 4) Soft constraint design: Introducing monotonicity constraints ( , , ), directional constraints ( , ).

[0040] 5) Ensemble learning: Multiple student models are weighted and fused together, with the weights adaptively adjusted based on the performance of the validation set. The weight formula is as follows:

[0041] Final ensemble prediction:

[0042] 4. Counterfactual Intervention Analysis 1) Intervention strategy: Simulating peak temperature reduction ( SOC upper limit ( ), power control ( )wait; 2) Causal effect quantification: Calculating individual treatment effects ( ), average treatment effect ( ), quantile treatment effect ( ); 3) Statistical confidence assessment: Bootstrap resampling was used to construct 95% confidence intervals to verify the significance of the strategy.

[0043] Example 2: Based on Example 1, but with a difference, an experiment was designed to illustrate the performance of the multi-mechanism degradation modeling and health state estimation method for parallel battery systems protected by this invention, as detailed below.

[0044] 1. Experimental vehicle Using the RX4E electric aircraft as the experimental platform, its battery system adopts a dual-module parallel configuration at the front and rear, with high-energy-density NCM811 ternary lithium batteries (nominal capacity 20 Ah, nominal voltage 3.6 V) as the cells. The total system voltage is 350 V, and the rated output power is 80 kW, which is used to meet the power requirements of multiple stages during flight, such as ground start-up, climb, cruise, and maneuvering.

[0045] 2. Data Acquisition System A high-frequency data acquisition module was used with a sampling interval of 0.18 seconds to simultaneously collect the operating parameters of the battery system at the individual cell level, module level, and system level. The acquisition time covered 12 complete flight missions, with a total data volume of 257,500 data points, corresponding to an actual operating time of 12.9 hours.

[0046] 3. Monitoring Parameter Details

[0047] The collected parameters cover battery status, system operation, and environmental context, as detailed below: 4. Data Coverage Scenarios The data covers the entire operational lifecycle of electric aircraft, and is distributed as follows: Ground operation: 3.2 hours (24.8%), temperature range 22-28℃, SOC fluctuation ±3%; Climbing phase: 2.1 hours (16.3%), temperature range 35-42℃, SOC decrease rate 0.8% / min; Cruise phase: 5.8 hours (45.0%), temperature range 38-45℃, SOC stabilized at 60%-85%; Mobile / emergency phase: 1.8 hours (13.9%), temperature range 48-52℃, SOC fluctuation ±8%.

[0048] In summary, this invention achieves high-precision health status estimation without direct capacity measurement. It employs a teacher-student learning framework that generates pseudo-labels using a physical model. The teacher model, constructed through the Arrhenius equation, rainflow counting method, and parallel system dynamics model, provides reliable supervision signals, enabling the student model to learn effectively even in the absence of real labels, overcoming the dependence of traditional data-driven methods on large amounts of labeled data. Secondly, this invention possesses more comprehensive system-level degradation modeling capabilities, establishing a complete model encompassing the unique effects of parallel systems such as thermal, cyclic, calendar aging, current unevenness, and capacity mismatch. It also quantifies the coupling effects between multiple mechanisms for the first time, whereas existing technologies typically consider only a single or limited number of degradation mechanisms, failing to accurately reflect the complex degradation behavior of parallel systems. Furthermore, this invention provides statistically significant engineering decision support, achieved through the introduction of counterfactual intervention analysis and Bootstrap confidence interval assessment. This provides statistically reliable quantitative evidence for BMS parameters such as temperature thresholds and current balance criteria, whereas traditional methods often rely on empirical recommendations and lack rigorous statistical validation. Finally, this invention achieves good generalization performance while maintaining physical interpretability. It designs a hybrid feature space that includes physical features, statistical features, and temporal features, and embeds physical laws such as monotonicity and directionality into the learning process through a soft constraint mechanism. This maintains the flexibility of data-driven approaches while ensuring the physical rationality of the prediction results.

[0049] The above specific embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Those skilled in the art should understand that the above embodiments do not limit the present invention in any way, and all similar technical solutions obtained by equivalent substitution or equivalent transformation are within the protection scope of the present invention.

Claims

1. A parallel battery system multi-mechanism degradation modeling and state of health estimation method, characterized in that, The method comprises the following steps: constructing a teacher model that generates health state pseudo-labels and soft constraints based on multi-mechanism physical models, wherein the multi-mechanism physical models include a thermal degradation model, a cycle degradation model, a calendar aging model, a parallel system effect model, and a mechanism coupling term; constructing a student model that integrates multiple machine learning algorithms and is trained under the guidance of the teacher model through a teacher-student training process to estimate the health state of a parallel battery system; the machine learning algorithms include a random forest algorithm, a gradient boosting algorithm, a support vector regression algorithm, and a lightweight multi-layer perceptron; based on the trained student model, the health state of the parallel battery system is estimated in real time or offline.

2. The method of claim 1, wherein, The thermal degradation model models the acceleration of solid electrolyte interface growth and active material dissolution by temperature based on the Arrhenius equation; The cycle degradation model uses the rainflow counting method and the Palmgren-Miner cumulative damage rule to distinguish the effects of shallow cycling, medium cycling, and deep cycling on battery degradation; The calendar aging model is based on diffusion theory and introduces state of charge and temperature acceleration factors to model the aging process under static storage; The parallel system effect model includes current imbalance, capacity mismatch, thermal coupling, and cycle out-of-sync effects to describe the interaction between individual batteries in a parallel battery system; The mechanism coupling term models the nonlinear interaction between thermal degradation and cycle degradation, cycle degradation and calendar aging, and thermal degradation and calendar aging, and quantifies it through coupling coefficients.

3. The method of claim 1, wherein, The soft constraints include: monotonicity constraints to ensure that temperature increases lead to increased degradation; directionality constraints to ensure that state of charge is positively correlated with calendar aging.

4. The method of claim 2, wherein, The teacher-student training process includes: data windowing step: time series features are extracted from battery system operation data using a sliding window; pseudo-label generation step: the total degradation is calculated by the teacher model and converted into health state pseudo-labels; feature engineering step: a multi-dimensional feature set is constructed, including statistical features, physical features, and time series features; soft constraint design step: monotonicity constraints and directionality constraints are introduced to guide student model training; ensemble learning step: multiple student models are weighted and fused, and the weights are adaptively adjusted based on the performance of the validation set.

5. The method of claim 1, wherein, It also includes a counterfactual intervention analysis step: simulate intervention strategies, including temperature peak reduction, state of charge upper limit restriction, and power control; quantify causal effects, calculate individual treatment effects, average treatment effects, and quantile treatment effects; statistical confidence evaluation: Bootstrap resampling method is used to construct confidence intervals to verify the significance of the intervention strategy.

6. The method of claim 4, wherein, The statistical confidence evaluation includes calculating the 95% confidence interval of the average treatment effect by repeated sampling, wherein the number of repeated sampling is 200 times.

7. A parallel battery system multi-mechanism degradation modeling and health state estimation system applying the method of any one of claims 1-5, comprising: The teacher model module generates health state pseudo-labels and soft constraints based on multi-mechanism physical models, wherein the multi-mechanism physical models include a thermal degradation model, a cycle degradation model, a calendar aging model, a parallel system effect model and a mechanism coupling term; The student model module integrates multiple machine learning algorithms and is trained by a teacher-student training process under the guidance of the teacher model to estimate the health state of the parallel battery system; the machine learning algorithms include a random forest algorithm, a gradient boosting algorithm, a support vector regression algorithm and a lightweight multi-layer perception; The training module: based on the trained student model, the health state of the parallel battery system is estimated in real time or offline; The counterfactual intervention analysis module specifically includes: The intervention strategy simulation unit includes temperature peak reduction, state of charge upper limit restriction and power control; The causal effect quantification unit is used to calculate individual treatment effects, average treatment effects and quantile treatment effects; The confidence evaluation statistics unit adopts a Bootstrap resampling method to construct a confidence interval to verify the significance of the intervention strategy.

8. A computer device, comprising: The computer device includes a processor and a memory, and the memory stores at least one instruction, at least one program, a code set or an instruction set, which is loaded and executed by the processor to realize the parallel battery system multi-mechanism degradation modeling and health state estimation method according to any one of claims 1-5.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one instruction, at least one program, a code set or an instruction set, which is loaded and executed by the processor to realize the parallel battery system multi-mechanism degradation modeling and health state estimation method according to any one of claims 1-5.

Images