Lithium battery SOC adaptive estimation method based on multi-mode fusion

Abstract

The invention discloses a lithium battery SOC adaptive estimation method based on multi-mode fusion. A data acquisition and processing module, a multi-mode fusion estimation module and an adaptive parameter optimization module are included. According to the method, multi-source data such as voltage, current, temperature and the like are collected in real time, online parameter identification is carried out by adopting a least square method with a joint adaptive forgetting factor and noise covariance, the online parameter identification is combined with extended Kalman filtering, ampere-hour integration and static calibration, dynamic fusion is carried out according to confidence, and an optimal SOC estimated value is output; according to the method, the problem that the precision of a traditional single estimation method is reduced under a complex working condition is effectively solved, and the robustness and accuracy of SOC estimation of the lithium battery under the conditions of a full life cycle, a wide temperature range and variable loads are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of lithium battery technology, and more specifically to an adaptive estimation method for lithium battery SOC based on multi-mode fusion. Background Technology

[0002] With the rapid popularization of electric vehicles and intelligent energy storage systems, the accuracy and reliability of the management system for power batteries, as the core energy storage unit, are crucial. The state of charge (SOC) of a battery is a key internal state characterizing its remaining usable energy. High-precision estimation of SOC is a prerequisite for achieving efficient energy dispatching, preventing overcharging and over-discharging, ensuring system safety, and extending battery life.

[0003] Currently, the mainstream methods for estimating State of Charge (SOC) include the ampere-hour integration method, the open-circuit voltage method, the neural network method, and the Kalman filter method. However, they all face varying degrees of challenges in practical engineering applications.

[0004] The ampere-hour integration method estimates charge throughput by integrating the operating current over time, thereby calculating the state of charge (SOC) change. It has advantages such as clear physical meaning, simple implementation, and low computational cost. However, its estimation results heavily rely on the accuracy of the initial SOC; errors in the initial value are inherited throughout the process. Furthermore, due to inherent limitations in sensor measurement accuracy and zero-point drift, small deviations in current measurement are amplified over long integration periods, resulting in significant cumulative errors. This causes the estimated SOC to gradually deviate from the true value, making it difficult to use independently in scenarios requiring long-term, high-precision estimation.

[0005] The open-circuit voltage method is calibrated based on the characteristic that there is a definite correspondence between the battery's terminal voltage and its state of charge (SOC) after the battery has been fully rested, and it has extremely high accuracy under resting conditions. However, in actual operation, the battery's terminal voltage is affected by the combined effects of ohmic internal resistance, electrochemical polarization, and concentration polarization, making it difficult to apply to online, real-time dynamic SOC estimation.

[0006] Neural network methods do not rely on the internal mechanisms of the battery. Instead, they train a nonlinear mapping model using a large amount of comprehensive test data, thereby directly inferring the State of Charge (SOC) based on external characteristics. While demonstrating potential in handling highly nonlinear systems, their performance is extremely dependent on the coverage and quality of the training data. Furthermore, complex network models typically involve significant computational overhead and storage requirements, posing a severe challenge to the computing power and memory resources of embedded hardware, thus limiting their large-scale application in cost-sensitive and resource-constrained BMS products.

[0007] The Kalman filter method models the battery as a dynamic system, constructing state-space equations and using voltage observations to optimally estimate the system state. It effectively handles measurement noise and achieves closed-loop correction. However, the core bottleneck of this type of algorithm lies in its high dependence on the accuracy of the battery model. In practical applications, the battery model parameters undergo drastic and complex time-varying changes with SOC, operating temperature, current rate, and state of health, making it impossible for the model to accurately describe its dynamic characteristics.

[0008] In summary, although the ampere-hour integration method, open-circuit voltage method, neural network method, and Kalman filter method are currently the mainstream SOC estimation techniques, each exhibits unique advantages under specific conditions, their inherent methodological defects make it difficult for them to independently meet the daily application requirements for SOC estimation accuracy and robustness under complex dynamic operating conditions.

[0009] Therefore, we make improvements by proposing an adaptive estimation method for lithium battery SOC based on multi-mode fusion. Summary of the Invention

[0010] This invention provides a method for adaptive estimation of lithium battery SOC based on multi-mode fusion, enabling the system to dynamically select the optimal estimation strategy according to the actual operating conditions of the battery, and achieve high-accuracy estimation of SOC across all scenarios and the entire life cycle.

[0011] To achieve the above objectives, embodiments of the present invention provide a method for adaptive estimation of lithium battery SOC based on multi-mode fusion, the method comprising: The target type of lithium battery was subjected to standard charge-discharge tests at multiple typical ambient temperatures, and its performance under different conditions was recorded. The value is the open-circuit voltage value after sufficient rest and stabilization.

[0012] Curve fitting was performed on the collected data to construct a functional relationship with SOC as the independent variable and OCV as the dependent variable, i.e. This relationship will serve as a baseline reference for the initial online estimation algorithm.

[0013] The system acquires the continuous running time since power-on and the duration of the battery in a static state, and initializes the state and model.

[0014] If the system determines that the battery is in an inactive state and the continuous resting time exceeds a preset threshold, it will use the open-circuit voltage method to calculate and obtain the current voltage. The baseline value is used to update the confidence level in static calibration mode.

[0015] When the battery is determined to be in an active operating state (charging / discharging), the system uses the recursive least squares method based on the forgetting factor (hereinafter referred to as FFRLS) to identify the time-varying parameters (including ohmic internal resistance, polarization parameters, etc.) in the battery equivalent circuit model in real time online.

[0016] The system inputs the parameter values ​​identified online by FFRLS into the extended Kalman filter module to obtain the estimated terminal voltage and the first SOC estimate. The confidence level in the dynamic high-precision calibration mode is updated based on the error between the estimated terminal voltage and the voltage value measured by the actual sensor.

[0017] If the confidence level of the dynamic high-precision calibration mode is higher than the preset value, calculate... If the trace of the covariance matrix is ​​lower than the expected threshold, the system will... Data updates to The relation table is updated using a recursive weighted least squares algorithm to improve the high-precision estimation model.

[0018] When the confidence levels of both static calibration and dynamic high-precision calibration modes are lower than the preset levels, the system sets the confidence level for that mode and uses the ampere-hour integration method to obtain the first SOC estimate. However, the result will be corrected or replaced by a higher-precision mode as soon as possible.

[0019] The system decides to use the corresponding model estimation result as the second estimate based on the confidence level of each model. The estimated value is output.

[0020] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section.

[0021] Beneficial effects The lithium battery SOC joint estimation method based on multi-mode fusion and adaptive models provided by this invention achieves high-precision and robust estimation across all operating conditions and the entire life cycle. Through a multi-mode intelligent switching scheme involving static calibration, dynamic high-precision calibration, and ampere-hour integral-based estimation, it overcomes the inherent limitations of single algorithms under specific operating conditions. Furthermore, through online parameter identification and self-learning, it effectively overcomes model mismatch and aging drift problems.

[0022] This paper proposes a recursive least squares method with a forgetting factor for online identification of battery model parameters, deeply coupled with an extended Kalman filter. This fundamentally solves the problem of model mismatch and decreased estimation accuracy caused by the time-varying characteristics of batteries in fixed-parameter models. A unique [method / mechanism] is also introduced. The curve-reverse update mechanism can adapt to battery aging and degradation. It utilizes data from high-confidence operating conditions to update online and gradually. The baseline curve enables the system to automatically track and compensate for aging phenomena such as battery capacity degradation and internal resistance increase. Relationship drift.

[0023] In summary, this invention achieves the co-evolution of battery models and algorithms by constructing a three-layer linkage full life cycle adaptive architecture of "online parameter identification - multimodal fusion - benchmark self-evolution", providing a systematic solution for high-precision estimation of lithium battery SOC. Attached Figure Description

[0024] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 Flowchart for multi-mode fusion estimation of SOC; Figure 2 System power-on initialization flowchart; Figure 3 Flowchart for estimating SOC in dynamic high-precision calibration mode; Figure 4 The SOC estimation results are obtained through multi-mode fusion without initial error; Figure 5 The SOC estimation results of multi-mode fusion are given with an initial error of 20%. Detailed Implementation

[0025] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0026] Figure 1 This is a flowchart of a multi-mode fusion estimation SOC according to an embodiment of the present invention.

[0027] In step S1, an nth-order polynomial is fitted based on the OCV values ​​corresponding to different SOCs at different temperatures obtained during the experimental phase.

[0028] In step S2, when the system is powered on, state and model initialization operations are performed.

[0029] In step S3, the system performs data acquisition and preprocessing in each sampling period.

[0030] The battery's terminal voltage is collected via a built-in sensor. Operating current and temperature It is then preprocessed with a sliding filter to suppress high-frequency noise; In step S4, the system determines whether the static calibration conditions are met.

[0031] The system detected that the battery had been idle for more than 2 hours and the operating current was... If the absolute value is less than 0.02C, the static calibration condition is deemed met; otherwise, it is deemed not met. If the static calibration condition is met, proceed to step S5; otherwise, proceed to step S6.

[0032] In step S5, the system outputs the first SOC estimate using the static calibration mode and updates the confidence level in this mode.

[0033] When the system enters a static state, the confidence level for this mode is set to high. The currently acquired stable terminal voltage is then used. With ambient temperature Substituting this into the nth-order polynomial fitted in S1, we obtain the first polynomial corresponding to the current open-circuit voltage. value.

[0034] In step S6, the system uses a dynamic high-precision calibration mode to estimate the SOC value and update the confidence level in this mode.

[0035] In step S7, the system determines whether the dynamic high-precision calibration conditions are met.

[0036] If the error of the terminal voltage value estimated using the dynamic high-precision calibration mode is less than the preset range, the condition for the dynamic high-precision calibration mode is deemed to be met; otherwise, the condition is not met. If the condition for the dynamic high-precision calibration mode is met, steps S8 and S9 are executed; otherwise, step S10 is executed.

[0037] In step S8, the system outputs the first SOC estimate in dynamic high-precision calibration mode and the confidence level in this mode.

[0038] In step S9, the system triggers a self-learning mechanism to update the parameters.

[0039] The system collects the terminal voltage at the current moment. ,temperature With high confidence To form a high-quality data pair This data is used to analyze the original... The fitted curve is corrected online and incrementally. This "reverse update" process enables... The relationship can adapt to battery aging and degradation, ensuring its long-term accuracy.

[0040] In step S10, the system outputs the first SOC estimate using the ampere-hour integration mode and updates the confidence level in this mode.

[0041] This mode is used during the initial power-on startup, when static calibration is not triggered, when the dynamic high-precision mode has not yet converged, or when the dynamic high-precision calibration mode is temporarily unavailable due to abnormal conditions, to provide continuous SOC estimation. The integral calculation formula is as follows:

[0042] in: For the end of the previous cycle Output value For Coulomb efficiency (set to a fixed value of 1 in this model). Sampling interval The rated capacity of the battery In this mode, the inherent cumulative uncertainty of the estimation results is the highest because no closed-loop correction is performed for current measurement errors and initial errors. This uncertainty increases over time, and its confidence level decreases over time.

[0043] In step S11, the system arbitrates the confidence level of each mode and outputs the second SOC value, and writes the state variable into the ROM.

[0044] The system will finalize the second [phase] in this cycle. Estimates, updated EKF error covariance matrix FFRLS parameter vector The corresponding key state variables, such as the covariance matrix, are written into the ROM. This operation ensures that in the event of an unexpected power outage or the next power-on, the algorithm can seamlessly recover from the most recent high-precision state, rather than starting from the initial default value. This achieves continuity and consistency of SOC estimation throughout its entire lifecycle, effectively avoiding estimation jumps or accuracy loss caused by system restarts.

[0045] Compared with existing technologies, this invention organically integrates multi-source information and multi-model estimation capabilities through an online confidence assessment mechanism, constructing a closed-loop estimation architecture with self-perception, self-evaluation and self-optimization characteristics. This effectively overcomes the accuracy bottleneck of traditional methods under complex working conditions and achieves high accuracy, high reliability and strong robustness in SOC estimation.

[0046] To ensure the continuity and convergence accuracy of the SOC estimation, and to guarantee stable connection and reliable initialization of the algorithm across different operation cycles, it is necessary to persistently store and manage the key intermediate states of the algorithm. Therefore, in this embodiment, the management methods for state saving and loading can be various those known to those skilled in the art.

[0047] like Figure 2 As shown, the following steps may be included: In step S21, the key status data saved in the previous working cycle is read from and loaded into the ROM, including: initial value Covariance matrix of recursive least squares method Parameter vector Data vector and forgetting factor .

[0048] In step S22, the system determines whether the key status data is valid by verifying the hash value of the key status data. If the key status data is found to be valid, step S23 is executed; otherwise, step S24 is executed.

[0049] In step S23, the system loads a set of preset battery model default parameters as initial parameters.

[0050] In step S24, the system initializes the correlation matrix and model using the state data loaded in the ROM.

[0051] To achieve high-precision dynamic estimation of SOC values, this implementation method uses the least squares method with joint adaptive forgetting factor and noise covariance for online parameter identification, which improves the system's ability to track time-varying characteristics of battery model parameters and its robustness.

[0052] like Figure 3 As shown, the following steps may be included: In step S61, a second-order RC equivalent circuit model is adopted and discretized to construct a standard least squares scheme. = .(in For the observed values, For data vectors, (This is the parameter vector to be identified.) In step S62, a forgetting factor is used. The recursive least squares method is used to identify battery parameters.

[0053] Calculate the gain matrix:

[0054] Update parameter estimates:

[0055] Update the covariance matrix:

[0056] From the convergent parameter vector The specific physical parameters of the battery model, including the ohmic internal resistance, are analyzed. Resistance of each RC network and and capacitor and The corresponding time constant and .

[0057] In step S63, the online covariance estimate of the updated information is calculated using a smoothing factor. The formula for calculating the online estimated covariance of the information is as follows:

[0058] in: For new information Smoothing factor In step S64, the model uncertainty is evaluated using the total uncertainty predicted by the current filtering theory and the total uncertainty actually observed, and an adjustment signal is generated, along with the innovation covariance ratio. The calculation formula is as follows:

[0059] in: The total uncertainty predicted by the current filtering theory Total uncertainty of actual observation In step S65, the formula for calculating the joint conditional factor based on the information covariance ratio is as follows:

[0060] in: As a regulatory factor The preset maximum conditional amplitude In step S66, as uncertainty increases, the forgetting factor is reduced to accelerate the tracking speed of battery model parameters, while the process noise covariance in the state equation increases simultaneously. This adaptive adjustment of the forgetting factor... Covariance of process noise The calculation formula is as follows:

[0061]

[0062] in: Linkage gain coefficient In step S67, the optimal estimate of SOC is performed using the identified real-time model parameters and process noise covariance.

[0063] The system predicts the prior state at the current time step based on the state equation and the SOC estimate from the previous time step. 1. Predict the prior error covariance matrix. Calculate the Kalman gain based on the prediction covariance and observation noise. Then, using the currently measured terminal voltage... By correcting the prior state using Kalman gain, the estimated SOC of the posterior state is obtained, which is the first estimated SOC value for the final output of this cycle. Finally, the posterior error covariance matrix is ​​updated to prepare for the calculation at the next time step.

[0064] Using data from lithium battery laboratory tests as the object, the verification results of online SOC estimation of lithium batteries under UDDS dynamic discharge conditions are as follows: Figure 4 , Figure 5 As shown.

[0065] from Figure 4 It can be seen that using the multi-mode fusion algorithm proposed in this invention... The estimated curve closely matches the reference true value, and the estimation error remains within 1% throughout the entire process, proving its high accuracy characteristics under all working conditions.

[0066] from Figure 5 It can be seen that the initial With an error value set to 20%, the multi-mode fusion algorithm proposed in this invention can quickly complete the correction, converging the estimation error to within 1% in a very short time.

[0067] In summary, the advantages of this invention are: This invention overcomes the contradiction between dynamic tracking lag and steady-state fluctuation caused by the fixed forgetting factor and process noise assumption in conventional recursive least squares method by using the least squares method with joint adaptive forgetting factor and noise covariance for online parameter identification, and achieves the optimal balance between parameter identification speed and accuracy.

[0068] This invention uses a collaborative algorithm that integrates multi-dimensional confidence and multi-mode fusion to estimate the state of charge (SOC) of lithium batteries. Compared with traditional single extended Kalman filtering or ampere-hour integration methods, it has stronger adaptability to operating conditions, fault tolerance, and consistency throughout the entire life cycle, ultimately comprehensively improving the accuracy, robustness, and long-term reliability of SOC estimation.

[0069] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive estimation method for lithium battery SOC based on multi-mode fusion, characterized in that, The method includes: Fitting based on lithium battery experimental data Functional relational expressions; Initialize the system's state and model; Real-time data acquisition and preprocessing; Determine whether the static calibration conditions are met; If the static calibration conditions are met, the first SOC estimate is output using the static calibration mode and the confidence level in this mode is updated. If the static calibration conditions are not met, the SOC is estimated using a dynamic high-precision calibration mode and the confidence level under that mode is updated. Determine whether the dynamic high-precision calibration conditions are met; If the conditions for dynamic high-precision calibration are met, the first SOC estimate and the confidence level under this mode are output using the dynamic high-precision calibration mode, and then the parameters are updated automatically. If the dynamic high-precision calibration conditions are not met, the first SOC estimate is output using the ampere-hour integration mode and the confidence level in this mode is updated. The second SOC value is output based on the confidence level arbitration, and the state variable is written to the ROM.

2. The method as described in claim 1, characterized in that: Initialize the system's state and model, including: Read and load the critical status data saved in the previous operation from the ROM; Determine if the key data is valid; If the key data is deemed valid, the key data will be used to initialize the correlation matrix and model; If the key data is determined to be invalid, the relevant matrix and model will be initialized using preset default parameters.

3. The method as described in claim 1, characterized in that: If the static calibration conditions are not met, a first SOC estimate is output using a dynamic high-precision calibration mode, and the confidence level in this mode is updated, including: Build a battery model; Battery parameters are identified by recursive least squares with a forgetting factor. Calculate and update online covariance estimates for new information; Assess model uncertainty and generate adjustment signals; Dynamically generate joint regulatory factors; Adaptive adjustment of forgetting factor and process noise covariance; The first SOC estimate is output based on the extended Kalman filter algorithm.

Images