Lithium battery low-temperature SOC cooperative preheating and estimating method, device and equipment

Abstract

The invention provides a lithium battery low-temperature SOC cooperative preheating and estimating method, device and equipment. The method is executed in the battery preheating process and comprises the steps that a deep reinforcement learning TD3 algorithm is adopted, the real-time temperature of a battery is used as the state, alternating current pulse parameters are used as actions, and a heating strategy is optimized through a multi-target reward function; on the basis of preheating real-time data, parameters of the temperature coupling PNGV model are identified online by using a recursive least square method with a forgetting factor; and based on the updated model, cooperatively estimating the SOC of the battery in real time by adopting an unscented Kalman filtering algorithm introducing an attenuation memory factor. According to the method, the problem of whole-course error accumulation caused by separation of preheating and SOC estimation in the low-temperature environment is solved, and rapid, safe and high-precision battery state estimation in the low-temperature environment is realized.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery state estimation and thermal management technology, specifically providing a method, apparatus, and equipment for low-temperature SOC-coordinated preheating and estimation of lithium batteries. Background Technology

[0002] Lithium-ion batteries are a core power source in automobiles, energy storage systems, and other fields. Accurate estimation of their state of charge (SOC) is a key technology for battery management systems (BMS), directly affecting vehicle range prediction, energy distribution strategies, charge / discharge control, and lifespan. However, in low-temperature environments (such as below 0°C), the viscosity of the electrolyte inside the battery increases and the ionic conductivity decreases, leading to a significant exacerbation of electrochemical and concentration polarization phenomena. This results in a dynamic characteristic of rapidly increasing internal resistance and a sharp decrease in usable capacity. This complex time-varying nonlinearity makes traditional SOC estimation methods based on fixed-parameter battery models (such as the ampere-hour integral method) susceptible to the cumulative error of current sensors, while model-dependent algorithms such as the extended Kalman filter (EKF) suffer severe accuracy degradation when parameters are mismatched.

[0003] Lithium-ion batteries experience severe performance degradation at low temperatures (such as below 0°C), with a sharp increase in internal resistance and a decrease in usable capacity. This leads to a deterioration in the accuracy of traditional SOC estimation methods based on fixed-parameter models (such as the ampere-hour integral method and the extended Kalman filter method). Existing technologies typically employ a separate "preheating then estimation" strategy, where the battery is heated to a suitable temperature via external heating or a fixed-mode pulse current before the SOC estimation module is activated. This strategy has significant drawbacks: the preheating process fails to be optimized in conjunction with the real-time battery status, resulting in low heating efficiency and the risk of localized overheating; while internal battery parameters (such as internal resistance and open-circuit voltage) change dynamically during preheating, the SOC estimation module is not activated or the model is not updated, making it impossible to obtain an accurate SOC value throughout the heating process. This severely impacts battery safety management and driving range prediction. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention aims to provide a method for coordinated preheating and estimation of SOC in lithium batteries at low temperatures. This method achieves deep coupling and synchronous optimization of the preheating process and the SOC estimation process, ensuring heating safety and efficiency while enabling high-precision estimation of battery SOC in low-temperature environments.

[0005] In a first aspect, the present invention provides a method for coordinated preheating and estimation of low-temperature SOC of lithium batteries, comprising the following steps: S1, using the deep reinforcement learning TD3 algorithm, taking the real-time battery temperature as the state input, taking the parameters of the AC pulse current as the action output, and taking maximizing heat generation efficiency as the optimization objective, to generate an adaptive heating strategy to preheat the battery; S2, based on the real-time data during the preheating process, identifying the parameters of the temperature-coupled PNGV equivalent circuit model online, wherein the model parameters are functions of battery SOC and temperature; S3, based on the temperature-coupled PNGV model with updated parameters, using an improved unscented Kalman filter algorithm to coordinately estimate the battery SOC in real time throughout the preheating process.

[0006] In one technical solution of the above-mentioned lithium battery low-temperature SOC synergistic preheating and estimation method, in step S1, the action space constituted by the parameters of the AC pulse current is defined as follows: ,in, () represents the pulse amplitude. The pulse frequency, This represents the pulse duty cycle.

[0007] In one technical solution of the above-mentioned lithium battery low-temperature SOC collaborative preheating and estimation method, in step S1, the optimization objective is achieved through the following reward function r(t): ,in, Instantaneous heat generation power, where T is the battery temperature. To reward high instantaneous heat production, This is the effective value of the current. To preset the optimal current value, , , , These are the weighting coefficients. High temperature rise rate is rewarded. To penalize currents that deviate from the optimal value.

[0008] In one technical solution of the above-mentioned lithium battery low-temperature SOC synergistic preheating and estimation method, in step S2, the recursive least squares method with forgetting factor is used for online parameter identification.

[0009] In one technical solution of the above-mentioned lithium battery low-temperature SOC coordinated preheating and estimation method, the improvement in step S3 is: introducing an attenuation memory factor into the covariance prediction step of the unscented Kalman filter algorithm. , Update the predicted covariance as follows: : ,in, This represents the prior prediction covariance matrix obtained based on the state transition equation and the unscented transformation. Let be the process noise covariance matrix, and k be the discrete time step.

[0010] In one technical solution of the above-mentioned lithium battery low-temperature SOC collaborative preheating and estimation method, steps S1, S2 and S3 are executed collaboratively based on the same real-time data stream, and the current, voltage and temperature data generated during the preheating process are directly used for model parameter updates and SOC estimation.

[0011] Secondly, the present invention provides a lithium battery low-temperature preheating and state-of-the-art (SOC) co-estimation device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement a lithium battery low-temperature SOC co-estimation method.

[0012] Thirdly, the present invention provides an automobile, including the aforementioned lithium battery low-temperature preheating and state co-estimation device.

[0013] The beneficial effects of this invention are as follows: 1. This invention simultaneously executes the adaptive preheating process based on deep reinforcement learning, real-time online battery parameter identification, and improved state estimation algorithm, with real-time data sharing. The dynamic data generated during preheating immediately drives model updates, and the updated model is then fed back in real time to optimize the preheating strategy and improve the accuracy of SOC estimation. This closed loop of "perception-decision-execution-correction" ensures that the system can continuously output highly reliable SOC values ​​throughout the entire heating process, from the start of battery low-temperature wake-up to reaching the target temperature, achieving seamless connection and end-to-end reliability of state estimation.

[0014] 2. This invention trains a deep reinforcement learning agent, enabling it to autonomously explore and learn the optimal AC pulse heating waveform under the guidance of a multi-objective reward function that integrates heat generation efficiency, temperature rise rate, current safety limits, and temperature tracking accuracy. This waveform can adaptively adjust according to the real-time state of the battery (such as temperature and internal resistance), maximizing heat generation power and temperature rise uniformity while strictly avoiding harmful operating conditions (such as overcurrent and local overheating), thus providing a rapid, safe, and consistent temperature rise foundation for subsequent processes.

[0015] 3. A temperature-coupled PNGV model with parameters as bivariate functions of SOC and temperature was established. Utilizing the rich real-time data stream generated during the preheating process, a recursive least squares method with a forgetting factor was employed for online parameter identification. The forgetting mechanism ensures that the model parameters can quickly "forget" outdated data and "remember" the latest dynamics, thus closely tracking the transient evolution of key parameters such as internal resistance and polarization time constant during battery heating, providing a highly accurate real-time battery model for state estimation.

[0016] 4. This invention introduces a decaying memory factor to exponentially amplify the predicted covariance. At the algorithmic level, this improvement is equivalent to giving historical state estimates a smaller trust weight and placing greater reliance on the latest observation data. Its direct effect is a significant improvement in the filter's response speed and tracking accuracy to sudden changes in battery voltage and internal resistance caused by heating. This allows it to perfectly complement the aforementioned real-time digital twin model, ensuring rapid convergence and stable reliability of the SOC estimation results even under highly nonlinear and rapidly time-varying preheating conditions. Attached Figure Description

[0017] The disclosure of this invention will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Furthermore, similar numbers in the drawings are used to denote similar components, wherein: Figure 1 This is an overall system flowchart of the low-temperature lithium battery SOC estimation method based on TD3-UKF according to an embodiment of the present invention.

[0018] Figure 2 The temperature-coupled PNGV model used in the embodiments of the present invention.

[0019] Figure 3 The OCV and SOC identified by the HPPC experiment of the lithium battery used in the embodiments of the present invention; Figure 4 The internal resistance and polarization capacitance identified by FFRLS in this embodiment of the invention; Figure 5 The error in FFRLS identification provided by the embodiments of the present invention; Figure 6 Error diagram of voltage prediction and actual voltage of the trained action network provided in an embodiment of the present invention.

[0020] Figure 7 The image shows the predicted SOC and actual SOC of the trained action network at a temperature of 25 degrees Celsius, as provided in this embodiment of the invention.

[0021] Figure 8 The image shows the predicted SOC and actual SOC of the trained action network at a temperature of 10 degrees Celsius, as provided in an embodiment of the present invention.

[0022] Figure 9 The image shows the predicted SOC and actual SOC of the trained action network at 0 degrees Celsius, as provided in this embodiment of the invention.

[0023] Figure 10 The image shows the predicted SOC and actual SOC of the trained action network at a temperature of -10 degrees Celsius, as provided in an embodiment of the present invention.

[0024] Figure 11 The average error diagram for each temperature provided in the embodiments of the present invention. Detailed Implementation

[0025] Some embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0026] Example 1 like Figure 1-11 As shown, this invention discloses a method for synergistic preheating and estimation of low-temperature SOC in lithium batteries, comprising the following steps: S1. The deep reinforcement learning TD3 algorithm is adopted, with the real-time battery temperature as the state input, the AC pulse current parameters as the action output, and the goal of maximizing heat generation efficiency as the optimization objective, to generate an adaptive heating strategy to preheat the battery.

[0027] S2. Based on real-time data during the preheating process, FFRLS is used to identify the parameters of the temperature-coupled PNGV equivalent circuit model online, where the model parameters are functions of battery SOC and temperature.

[0028] S3. Based on the temperature-coupled PNGV model with updated parameters, an improved unscented Kalman filter algorithm is used to estimate the battery SOC in real time throughout the preheating process.

[0029] In one embodiment, step S1 specifically includes: Step S11: Establish a state-space model considering temperature variables. Based on the state-space expression of the dual-state lumped parameter model, this invention treats the ambient temperature as a variable and calculates the internal temperature at time k+N iteratively. After N iterations, the following can be obtained: ; in : is the weight matrix, representing Input at time t to final state The extent of the impact.

[0030] Assuming the model matrix Having two distinct eigenvalues .

[0031] Among them, eigenvalues The modulus is typically less than 1 (for stable systems), and they represent the two characteristic time constants of the system. The corresponding temperature dynamics are composed of the superposition of two independent exponentially decaying modes.

[0032] The diagonalized form is: ; in .

[0033] Furthermore, ; Substituting into the original expression, we get: ; The total heat generated by the battery at time K is: ; Step S12: Construct a reinforcement learning framework based on the TD3 algorithm. Reinforcement learning regulates the interaction between an agent and its environment. The ultimate goal of the algorithm is to learn a policy network that maximizes the cumulative reward. At each discrete time step, given a state s, according to the policy network... Generate action Receive rewards and new status The benefit is defined as the sum of the discount rewards: ; in, The optimal strategy is to determine the discount factor for prioritizing short-term returns. The parameters are The goal is to maximize returns through continuous updates. Expected return is defined as: ; Perform the action in state s. And in subsequent strategies The expected reward obtained when the state remains unchanged is defined as the value function of the state-action pair at that time: ; Temporal Differential Learning (TD Learning) uses the update rule of the Belman equation to illustrate the relationship between the value function of the current state-action pair and the value function of subsequent state-action pairs, in order to learn the evaluation function: ; Step S13: Core Improvements and Training Process of the TD3 Algorithm To prevent getting trapped in local optima, TD3 employs a dual-Q network. By selecting the minimum of the two estimates, it limits the overestimation of the Q-value. Furthermore, TD3 innovatively introduces a technique of delayed policy updates, reducing the frequency of policy updates and thus improving the algorithm's stability. Noise is added to the target policy to increase its exploratory nature and prevent getting trapped in local optima. Based on these improvements, the TD-target generated by the dual-objective evaluation network at each time step is: ; Parameters of two real-world evaluation networks It updates by minimizing TD-error: ; Real-world action network parameters The update rules are as follows: ; The update mechanisms for the target evaluation network and the action network are as follows: ; in, For soft update rate, ≪1.

[0034] The update process of the above parameters constitutes the core idea of ​​the TD3 algorithm.

[0035] Step S14: The optimization strategy using the TD3 algorithm aims to maximize heat generation efficiency. This is achieved by optimizing the pulse frequency and amplitude at each discrete time point to obtain the optimal pulse waveform throughout the heating process. The proposed TD3-based preheating strategy is described in the following section. , as well as The definition is as follows.

[0036] During the high-frequency pulse preheating process of the battery, since the battery's SOC remains constant, the internal parameters of the battery are mainly affected by the internal temperature. Therefore, the state space is defined as follows: ; The specified battery value range is [-20° to 20°].

[0037] For the battery pulse preheating process, the agent's goal is to learn how to optimally adjust the current waveform to achieve the best preheating effect while ensuring good consistency. In this process, the waveform amplitude, frequency, and duty cycle are key variables; therefore, the action space is defined as the adjustment range of these three parameters. ; The pulse amplitude and frequency range are [-25A, 25A], [1000Hz, 10000Hz], and [0, 1], respectively. The reward function is set as follows: ; in, Instantaneous heat generation power, Battery temperature, RMS current value coefficient, To reward high instantaneous heat production, High temperature rise rate is rewarded. To penalize current deviations from optimal values ​​and prevent battery damage, It is designed to provide precise control when approaching the target temperature.

[0038] Step S15: At time t, based on the temperature state, the real action network outputs an action vector containing pulse amplitude and frequency. This action vector, along with the state vector, serves as input to the real evaluation network to calculate the Q-value of the current state-action pair. After the state transition is complete, the state vector for the next time step is obtained and input into the target action network and the target evaluation network, and the same operation is performed. During this process, the smaller Q-value output by the two target evaluation networks is selected and, together with the Q-value of the current state-action pair, is used to update the parameters of the real evaluation network. Simultaneously, the parameters of the real action network are also updated based on the Q-value of the current state-action pair. The action vector, state vector, reward value, and state vector for the next time step are combined into a quadruple and stored in the experience replay pool. The network is trained and updated by iterating through the above process. The action network trained through the above process exhibits the following performance: Figure 6 As shown, the predicted voltage is in high agreement with the actual voltage, with minimal error, proving that the trained policy network can accurately control the heating process.

[0039] After training, the converged Actor network is deployed into the actual battery management system. During online operation, the system senses the battery status in real time, and the TD3 network generates optimal pulse parameters to drive the power circuit to produce AC excitation to heat the battery, forming a closed-loop control of "sensing-decision-execution" until the battery temperature reaches the target value. This method can adapt to environmental changes and achieve rapid and efficient low-temperature self-heating while ensuring safety.

[0040] In one embodiment, to obtain the impedance characteristics of the experimental battery, this invention uses a temperature-coupled PNGV model to model the experimental battery. The open-circuit potential characteristics of the experimental battery are obtained and temperature-corrected, and the open-circuit potential curve is fitted using a function. Furthermore, the impedance characteristics of the experimental battery at different temperatures and SOCs are obtained through parameter identification. Figure 2 The diagram shows the temperature-coupled PNGV equivalent circuit model used in this embodiment of the invention. Based on the traditional PNGV model, this model expresses all key parameters as functions of SOC and temperature T, thereby achieving deep coupling between electrical and thermal characteristics. Step S2 specifically includes: Step S21: Obtain the open-circuit potential characteristics of the experimental battery, perform temperature correction, and fit the open-circuit potential curve using a function. All functions here are considered... and temperature The function.

[0041] According to Kirchhoff's laws, the time-domain equation of the temperature-coupled PNGV equivalent circuit model can be expressed as: Terminal voltage equation: ; Dynamic equation: RC network voltage Updates: ; Capacitive effect voltage Updates: ; SOC definition equation (using the ampere-hour integration method): ; in, It is the battery's rated capacity, and it is also a function of temperature.

[0042] Step S22: The recursive least squares method can effectively overcome the uncertainty of model parameters through parameter correction and updating, thereby accurately capturing the implementation characteristics of the system. However, for systems that change slowly and continuously, the traditional recursive least squares method struggles to obtain reliable and stable estimation results. To address this issue, a recursive least squares method with a forgetting factor can be used to reliably identify system parameters. This method reduces the amount of information in old data by incorporating a forgetting factor into the measurement data, creating conditions for supplementing information from new data.

[0043] Consider the following system: ; in, Zero-mean white noise, For the system's output variables, These are system data variables. System parameter variables.

[0044] The system data variables are defined as follows: ; The system parameter variables are shown in equation (3-10): ; For ease of distinction, the subscript indicates that the data value is at the k-th sampling time. The algorithm flow is shown in the following formula: ; in, This is the forgetting factor; when this value is 1, the method degenerates into the traditional recursive least squares method. This represents the gain of the algorithm. Let be the error covariance matrix of the state estimate.

[0045] To complete the parameter identification of the model, it is first necessary to obtain the basic characteristic curves of the battery. For example... Figure 3 As shown, by conducting HPPC tests on the experimental battery, the relationship curves between open-circuit voltage (OCV) and state of charge (SOC) at different temperatures were obtained. Figure 3 a) and the corresponding SOC change curve. Specifically, a least squares algorithm with a forgetting factor is used to dynamically identify the model parameters. The application of this algorithm in the parameter identification of the PNGV equivalent circuit model of a lithium-ion battery is described in detail below. The Laplace equation can be obtained from the temperature-coupled PNGV model as follows: ; Take impedance function for: ; The impedance function is discretized using a bilinear transform, letting ; Where T is the sampling interval, the discretized transfer function can be obtained as: ; in, For the corresponding constant coefficients, the Laplace equation is transformed into a discrete difference equation: ; To simplify the calculation, we assume that the amount of electricity consumed or absorbed by the power source within a unit sampling interval T has approximately zero impact on the State of Charge (SOC), meaning that the open-circuit potential is approximately the same at each sampling time. Therefore, the above equation can be simplified to: ; Recursive least squares is an algorithm with infinite memory length. For battery systems, as the recursive calculations accumulate older data, the results may fail to accurately reflect the characteristics of new data. To avoid this, a forgetting factor is introduced, effectively overcoming the "data saturation" phenomenon. The sampling time is set to 1 second in the program, consistent with the sampling time of the battery charge / discharge tester. To accelerate convergence, an initial covariance matrix is ​​used, where... in It is an identity matrix.

[0046] The results of parameter identification are as follows Figure 4 As shown. The error of this identification method is as follows. Figure 5 As shown, the error remains at a low level, verifying the accuracy and reliability of the FFRLS algorithm for online identification of temperature-coupled PNGV model parameters.

[0047] After iteratively calculating the system parameter variable k, the model parameters can be derived. , as well as The value can be represented as: ; The coefficients on the right side of the equation can be calculated using a recursive algorithm, while the left side represents the unknown parameters of the model. This concludes the parameter identification process.

[0048] To complete the parameter identification of the temperature-coupled PNGV model, it is necessary to select a suitable battery operating condition as the battery test condition. This paper uses the HPPC dynamic operating condition as the operating condition for parameter identification.

[0049] The pulse current was set to 3C. First, a series of pulse experiments were performed at a SOC of 1, followed by discharge with a 1C current to a SOC of 0.975, and then allowed to stand for 1 hour. This pattern was repeated at 13 SOC points: 1, 0.975, 0.95, 0.925, 0.9, 0.8, down to 0.1, recording the voltage and current data. The HPPC experiment with a pulse amplitude of 3C was selected and conducted at 0°C.

[0050] In one embodiment, after completing the above work, the state iteration equation of the dual-state lumped parameter model is used as the state transition equation, and the SOC is cyclically updated and estimated using a jointly improved unscented Kalman filter (UKF) combined with the temperature model identification results. Considering the inconsistency of real-time temperature changes in actual environments, the unscented Kalman filter algorithm (UKF) adopts covariance estimation based on the decay memory factor (exponential weighting). This improvement is equivalent to assigning smaller weights to the old estimates, making the filter more inclined to trust the latest measurement data, thereby accelerating the response to sudden changes. This effectively achieves real-time and accurate estimation of battery state of charge (SOC) under low-temperature environments. Step S3 specifically includes: Step S31, Initialization: ; ; Step S32: Sigma point sampling (same as standard UKF) For k=1,2,…, based on the posterior estimate from the previous time step… Covariance Calculate the Sigma point set .

[0051] ; ; ; Step S33: State prediction (same as standard UKF) The Sigma point is passed through the state transition function. spread: ; Calculate the weighted sum of the predicted state and the predicted covariance: ; ; in These are the weighting coefficients used to calculate the state. These are the weighting coefficients used to calculate covariance.

[0052] Step S34: Introduce decay memory factor This is the core modification step. Incorporating process noise. Previously, the predicted covariance calculated in the previous step... Multiply by the attenuation factor (in ).

[0053] ; Step S35, Observation Update Resampling Sigma points: using attenuated prior covariance Regenerate a set of Sigma points Or use directly Conduct observation and prediction.

[0054] Observational prediction: Sigma points are predicted using the observation function. spread: ; Calculate the predicted observations and the observed covariance: ; ; ; Status Update: ; ; ; When new measurement data Upon arrival, Kalman gain It will become bigger because and All because The Kalman gain increases due to amplification. A larger Kalman gain implies a larger state update. It will "pull" more significantly toward the new measurement value, thus enabling a rapid response to sudden state changes.

[0055] To verify the overall effectiveness of the invention, SOC estimation tests were conducted at multiple temperature points. For example... Figure 7 , Figure 8 , Figure 9 , Figure 10 The figures show comparison curves of the SOC estimates (predicted) and actual SOC values ​​(actual) obtained using the method of this invention at temperatures of 25°C, 10°C, 0°C, and -10°C. As can be seen from the figures, under different low-temperature conditions, the SOC estimation curves of the method of this invention closely track the actual SOC curves, demonstrating extremely high estimation accuracy.

[0056] Estimation error at various temperatures, for example Figure 11 As shown in the figure, compared with the traditional method, the method of the present invention maintains the average error of SOC estimation at an extremely low level over a wide temperature range of -10°C to 25°C, with the advantage being particularly significant at low temperatures, fully demonstrating the effectiveness and robustness of the proposed solution.

[0057] Example 2 This invention discloses a lithium battery low-temperature preheating and state-of-the-art (SOC) co-estimation device, comprising a memory and a processor. The memory stores a computer program and can be configured to store a program for executing the lithium battery low-temperature SOC co-estimation method described in the above-described method embodiments. The processor can be configured to execute the program stored in the memory, including but not limited to the program for executing the lithium battery low-temperature SOC co-estimation method described in the above-described method embodiments. For ease of explanation, only the parts related to the embodiments of this invention are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of this invention. The control device can be a control device device comprising various electronic devices.

[0058] Example 3 This invention discloses an electrical device that includes a lithium battery low-temperature preheating and state co-estimation device, which has the ability to quickly wake up the battery in extremely cold environments and achieve high-precision range prediction, greatly improving the user experience and safety in low-temperature environments.

[0059] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the original technical features, and the technical solutions resulting from these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for low temperature SOC co-estimation and preheating of a lithium battery, characterized in that, The method comprises the following steps: S1, using a deep reinforcement learning TD3 algorithm, taking the real-time temperature of the battery as the state input, taking the parameters of the alternating pulse current as the action output, and maximizing the heat generation efficiency as the optimization target, to generate an adaptive heating strategy to preheat the battery; S2, based on the real-time data in the preheating process, online identification of the parameters of the temperature coupled PNGV equivalent circuit model, wherein the model parameters are functions of the battery SOC and temperature; S3, based on the temperature coupled PNGV model with updated parameters, using an improved unscented Kalman filter algorithm to estimate the battery SOC in real time during the whole preheating process.

2. The method of claim 1, wherein, In step S1, the action space composed of the parameters of the alternating pulse current is defined as: ； wherein ) is the pulse amplitude, is the pulse frequency, is the pulse duty cycle.

3. The method of claim 1, wherein, In step S1, the optimization target is achieved through the following reward function r(t): ； wherein, instantaneous heat generation power, T is the battery temperature, to reward high instantaneous heat generation power, is the current effective value, is a preset optimal current value, , , , is a weight coefficient, to reward high temperature rise rate, to punish the current deviating from the optimal value.

4. The method of claim 1, wherein, In step S2, a recursive least squares method with a forgetting factor is used for online parameter identification.

5. The method of claim 1, wherein, In step S3, the improvement consists in introducing a decaying memory factor in the covariance prediction step of the unscented Kalman filter algorithm , and the predicted covariance is updated in the following way : ； wherein, denotes the prior prediction covariance matrix based on the state transition equation and the unscented transformation, is the process noise covariance matrix, k is the discrete time step.

6. The method of claim 1, wherein, Steps S1, S2 and S3 are cooperatively executed based on the same real-time data stream, and the current, voltage and temperature data generated during the preheating process are directly used for model parameter updating and SOC estimation.

7. A lithium battery low temperature preheating and state collaborative estimation device, characterized in that, The device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method of any one of claims 1-6 when executing the computer program.

8. An electric device, characterized by The device comprises a lithium battery low-temperature preheating and state cooperative estimation device according to claim 7.

Images