Lithium battery SOC and SOH joint estimation method and system based on iterative denoising

Abstract

The invention provides a lithium battery SOC and SOH joint estimation method and system based on iterative denoising in the technical field of lithium battery management. The method comprises the following steps: S1, acquiring charging fragment data and factory nominal capacity of a kth round of a lithium battery; s2, inputting the charging segment data into an SOH estimation model to obtain an SOH estimation value; s3, calculating the current available capacity based on the SOH estimated value and the factory nominal capacity; s4, correcting the SOC based on the current available capacity through an inverse ampere-hour integral method to obtain a physical correction SOC; s5, inputting the physical correction SOC, the SOH estimation value and k into a residual denoising network to obtain SOC noise data, and correcting the physical correction SOC based on the SOC noise data to obtain a residual correction SOC; and S6, taking the residual error correction SOC as the SOC of the (k-1) th round to carry out iterative denoising, and outputting a final residual error correction SOC and an SOH estimation value after iteration is completed. The method has the advantage that the precision and robustness of SOC and SOH estimation are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of lithium battery management technology, and in particular to a method and system for jointly estimating the SOC and SOH of a lithium battery based on iterative denoising. Background Technology

[0002] In the field of lithium battery management technology, accurate estimation of State of Charge (SOC) and State of Health (SOH) is crucial for ensuring the safe operation of battery systems, extending their lifespan, and optimizing energy dispatch efficiency. SOC characterizes the real-time remaining capacity of a battery, and its commonly used estimation method is the ampere-hour integration method. However, the accuracy of this method heavily depends on the accuracy of the initial SOC and the correct assessment of the current maximum usable capacity of the battery. As the battery ages through cycles, its maximum usable capacity continuously decreases. If a fixed nominal capacity is still used for integration calculations, the SOC estimation results will experience systematic drift, with errors accumulating over time. SOH, as a key parameter reflecting the degree of battery aging, is typically defined as the ratio of the current maximum usable capacity to the factory-statistical capacity. Traditional SOH estimation methods are mostly based on open-circuit voltage plateaus, internal resistance variation patterns, or empirical aging models. However, these methods usually assume that the SOC is known or accurate. In practical applications, the SOC itself has noise or drift, thus forming a coupled estimation problem where SOC and SOH are interdependent. That is, accurate estimation of SOH requires high-precision SOC input, while SOC correction depends on the correct SOH value.

[0003] In recent years, deep learning methods have been gradually applied to the joint estimation of SOC and SOH. For example, recurrent neural networks (RNNs) or temporal convolutional networks (TCNs) are used to directly predict battery state from sensor data such as voltage, current, and temperature. Although such data-driven methods have the ability to capture the nonlinear dynamic characteristics of batteries, they still have obvious limitations: First, most methods model SOC and SOH estimation as a one-way process, lacking a dynamic feedback mechanism and being sensitive to noise in the input SOC. Second, existing methods fail to effectively distinguish the signal-to-noise ratio differences of different features—while direct measurements such as voltage, current, and temperature contain sensor noise, their dynamic responses are generally reliable, whereas SOC becomes the main source of error due to the integral accumulation effect. If all features are denoised indiscriminately, it will not only increase the computational burden but may also weaken the effective information related to aging, thus affecting the accuracy of SOH estimation.

[0004] Meanwhile, diffusion models have demonstrated excellent iterative denoising capabilities in fields such as image and speech generation. They gradually approximate the real data distribution through multi-step noise addition and denoising processes, and control the denoising intensity by using time-step embedding. However, such methods have not yet been effectively explored in battery state estimation. Existing technologies lack a mechanism to guide SOC denoising using SOH as conditional information, and have also failed to construct a closed-loop optimization path of "denoising-feedback-correction," making it difficult to achieve round-by-round suppression and self-correction of estimation errors.

[0005] In summary, the existing technology has the following main shortcomings: 1. SOC estimation is usually based on fixed capacity parameters and fails to dynamically adjust the capacity benchmark as the battery ages, resulting in the continuous accumulation of estimation deviations over long-term use; 2. SOH estimation models often use noisy SOC sequences as input, and the noise interferes with the feature expression, affecting the robustness of the model; 3. Existing joint estimation methods are mostly one-time modeling structures, lacking iterative denoising mechanisms, and cannot achieve a gradual purification process similar to diffusion models; 4. Data-driven methods often ignore the electrochemical and physical constraints of the battery itself, which may generate a SOC trajectory that does not conform to the actual dynamics (e.g., the SOC does not increase monotonically during charging). 5. Without introducing a perception mechanism for the iteration stage, the model cannot distinguish between the "coarse adjustment" and "fine adjustment" stages, which can easily lead to overcorrection or result oscillation.

[0006] Therefore, how to provide a joint estimation method and system for SOC and SOH of lithium batteries based on iterative denoising, so as to improve the accuracy and robustness of SOC and SOH estimation, has become an urgent technical problem to be solved. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method and system for joint estimation of SOC and SOH of lithium battery based on iterative denoising, so as to improve the accuracy and robustness of SOC and SOH estimation.

[0008] In a first aspect, the present invention provides a joint estimation method for SOC and SOH of lithium batteries based on iterative denoising, comprising the following steps: Step S1: Set an iteration round identifier, set the initial value of the iteration round identifier to k, and obtain the charging segment data and factory nominal capacity of the lithium battery in the kth round; the charging segment data includes charging voltage, charging current, charging temperature and SOC; Step S2: Input the charging segment data into the SOH estimation model built based on the Transformer network to obtain the SOH estimate value of the kth round; Step S3: Based on the estimated SOH value and the nominal capacity at the factory, calculate the current available capacity for the kth round; Step S4: Using the inverse ampere-hour integration method, the SOC of the kth round is corrected based on the current available capacity to obtain the physically corrected SOC; Step S5: Input the physical correction SOC, SOH estimate and k into the residual denoising network to obtain the SOC noise data of the kth round. Correct the physical correction SOC based on the SOC noise data to obtain the residual correction SOC. Step S6: Use the residual correction SOC as the SOC of the (k-1)th round, and then construct the charging segment data of the (k-1)th round. Update the value of the iteration round identifier to k-1, perform iterative denoising in the (k-1)th round, until the value of the iteration round identifier is 1, and output the final residual correction SOC and SOH estimate.

[0009] Furthermore, in step S1, the expression for the charging segment data of the kth round is: ; in, This represents the charging segment data in the k-th round; This represents the charging voltage at time t; This represents the charging current at time t; This represents the charging temperature at time t; Represents the SOC at time t in round k; Indicates the sequence length of the charging segment data; The charging voltage, charging current, and charging temperature remain constant throughout each iteration.

[0010] Furthermore, in step S2, the formula for calculating the SOH estimate in the kth round is: ; in, This represents the estimated SOH value in the k-th round; This represents the charging segment data in the k-th round; () represents the SOH estimation model; In step S3, the formula for calculating the current available capacity in the k-th round is: ; in, This represents the current available capacity in round k; This represents the estimated SOH value in the k-th round; Indicates the nominal capacity specified by the manufacturer.

[0011] Furthermore, in step S4, the formula for calculating the physical correction SOC is: ; ; in, Represents the physical correction SOC at time t in round k; Indicates the start time of charging; Indicates the time when charging ends; express Initial SOC at time step; This represents the current available capacity in round k; This represents the integral of the charging current from the start of charging to the current moment.

[0012] Furthermore, in step S5, the formula for calculating the SOC noise data in the kth round is: ; in, This represents the SOC noise data for the k-th round; This represents a residual denoising network; This represents the physical correction SOC in round k; This represents the estimated SOH value in the k-th round; The formula for calculating the residual correction SOC is as follows: ; in, This represents the residual correction SOC in the k-th round, which is the SOC in the charging segment data of the (k-1)-th round. The residual denoising network is built based on a deep learning model. The training process is carried out in conjunction with the SOH estimation model. An end-to-end optimization strategy is adopted. The loss function includes the mean square error between the final SOC estimate and the true SOC label, and the mean square error between the final SOC estimate and the true SOC label. The gradient is backpropagated through K rounds of iterations to simultaneously optimize the model parameters of the SOH estimation model and the residual denoising network. The iteration round k is introduced in the residual denoising network through the timestep embedding mechanism.

[0013] Secondly, the present invention provides a joint estimation system for SOC and SOH of lithium batteries based on iterative denoising, comprising the following modules: An initialization module is used to set an iteration round identifier, set the initial value of the iteration round identifier to k, and obtain the charging segment data of the lithium battery in the kth round and the nominal capacity at the factory; the charging segment data includes charging voltage, charging current, charging temperature and SOC; The SOH estimation module is used to input the charging segment data into the SOH estimation model built based on the Transformer network to obtain the SOH estimate value in the kth round. The current available capacity estimation module is used to calculate the current available capacity in the k-th round based on the SOH estimate and the factory nominal capacity; The SOC physical correction module is used to correct the SOC of the k-th round based on the current available capacity using the inverse ampere-hour integration method to obtain the physically corrected SOC. The SOC residual correction module is used to input the physical correction SOC, SOH estimate and k into the residual denoising network to obtain the SOC noise data of the kth round, and correct the physical correction SOC based on the SOC noise data to obtain the residual correction SOC. The iterative denoising module is used to take the residual corrected SOC as the SOC of the (k-1)th round, and then construct the charging segment data of the (k-1)th round. The value of the iteration round identifier is updated to k-1, and iterative denoising is performed in the (k-1)th round until the value of the iteration round identifier is 1. Finally, the residual corrected SOC and SOH estimate are output.

[0014] Furthermore, in the initialization module, the expression for the charging segment data in the k-th round is: ; in, This represents the charging segment data in the k-th round; This represents the charging voltage at time t; This represents the charging current at time t; This represents the charging temperature at time t; Represents the SOC at time t in round k; Indicates the sequence length of the charging segment data; The charging voltage, charging current, and charging temperature remain constant throughout each iteration.

[0015] Furthermore, in the SOH estimation module, the formula for calculating the SOH estimate in the kth round is: ; in, This represents the estimated SOH value in the k-th round; This represents the charging segment data in the k-th round; () represents the SOH estimation model; In the current available capacity estimation module, the formula for calculating the current available capacity in the k-th round is: ; in, This represents the current available capacity in round k; This represents the estimated SOH value in the k-th round; Indicates the nominal capacity specified by the manufacturer.

[0016] Furthermore, in the SOC physical correction module, the calculation formula for the physically corrected SOC is as follows: ; ; in, Represents the physical correction SOC at time t in round k; Indicates the start time of charging; Indicates the time when charging ends; express Initial SOC at time step; This represents the current available capacity in round k; This represents the integral of the charging current from the start of charging to the current moment.

[0017] Furthermore, in the SOC residual correction module, the calculation formula for the SOC noise data in the kth round is: ; in, This represents the SOC noise data for the k-th round; () represents a residual denoising network; This represents the physical correction SOC in round k; This represents the estimated SOH value in the k-th round; The formula for calculating the residual correction SOC is as follows: ; in, This represents the residual correction SOC in the k-th round, which is the SOC in the charging segment data of the (k-1)-th round. The residual denoising network is built based on a deep learning model. The training process is carried out in conjunction with the SOH estimation model. An end-to-end optimization strategy is adopted. The loss function includes the mean square error between the final SOC estimate and the true SOC label, and the mean square error between the final SOC estimate and the true SOC label. The gradient is backpropagated through K rounds of iterations to simultaneously optimize the model parameters of the SOH estimation model and the residual denoising network. The iteration round k is introduced in the residual denoising network through the timestep embedding mechanism.

[0018] The advantages of this invention are: 1. By setting an iteration round identifier, the initial value of the iteration round identifier is set to k. The charging segment data and the nominal factory capacity of the lithium battery in the kth round are obtained. The charging segment data is input into the SOH estimation model to obtain the SOH estimate for the kth round. Based on the SOH estimate and the nominal factory capacity, the current available capacity of the kth round is calculated. Then, the SOC of the kth round is corrected based on the current available capacity using the inverse ampere-hour integral method to obtain the physically corrected SOC. The physically corrected SOC, the SOH estimate, and k are then input into the residual denoising network to obtain the SOC noise data for the kth round. The physically corrected SOC is corrected based on the SOC noise data to obtain the residual corrected SOC. The residual corrected SOC is then used as the SOC for the (k-1)th round, and the charging segment data for the (k-1)th round is constructed. The value of the iteration round identifier is updated to k-1, and iterative denoising is performed for the (k-1)th round until the value of the iteration round identifier is 1. The final residual corrected SOC is then output. Positive SOC and SOH estimates are constructed by building the entire estimation process into a multi-round iterative process from coarse to fine: In each round, the SOH is first estimated using the current (potentially noisy) SOC sequence to dynamically update the battery capacity benchmark. Then, the inverse ampere-hour integral method is used for preliminary physical correction to ensure that the SOC trajectory conforms to electrochemical laws. Next, a residual denoising network conditioned on the iteration round is used to perform data-driven fine correction of residual noise. Finally, the purified SOC of this round is fed back to the next round as input. This design allows SOH estimation to be based on a SOC sequence that is purified round by round with continuously decreasing noise levels, while SOC correction always depends on the dynamically updated capacity benchmark, thus breaking the error coupling between the two. Through the fusion of physical constraints and data denoising, as well as the model's ability to perceive the iteration stage, the gradual suppression and self-correction of errors are achieved, ultimately greatly improving the accuracy and robustness of SOC and SOH estimation.

[0019] 2. By integrating the Transformer network and the residual denoising network, the SOC and SOH of the lithium battery are estimated simultaneously within a single framework, avoiding the error accumulation and inefficiency problems caused by separate estimation in traditional methods. This joint estimation method can utilize the inherent correlation between SOC and SOH, such as dynamically correcting the capacity parameter in the SOC calculation through the SOH estimate, thereby improving the accuracy and consistency of the overall estimation.

[0020] 3. A multi-round iterative denoising process is adopted, gradually iterating from the kth round to the 1st round. In each round, the SOC noise data is corrected through a residual denoising network, gradually reducing the accumulated error. This iterative design is similar to the idea of ​​a denoising diffusion model, which can effectively handle uncertainties such as sensor noise and model bias, and finally output more stable SOC and SOH estimates. Compared with the single estimation method, iterative denoising improves the robustness and anti-interference ability of the method, and is especially suitable for long-term monitoring of lithium batteries under complex working conditions.

[0021] 4. By combining the physically based inverse ampere-hour integral method with data-driven deep learning models (such as Transformer and residual denoising networks), the inverse ampere-hour integral method provides preliminary SOC correction under physical constraints, ensuring that the results conform to the electrochemical principles of batteries; while the deep learning model learns nonlinear relationships from the data to compensate for the shortcomings of the physical model. This hybrid approach retains the interpretability of model-driven methods while leveraging the adaptability of data-driven methods, reducing the risk of overfitting and improving reliability in real-world scenarios.

[0022] 5. In the SOH estimation stage, a model is built based on the Transformer network, which is good at processing time series data, such as charging voltage, current and temperature series. The self-attention mechanism of the Transformer can capture long-term dependencies, thereby extracting SOH-related features from the charging segment data more accurately. This not only improves the accuracy of SOH estimation, but also enhances the method's ability to predict battery aging trends, demonstrating the effective application of advanced deep learning technology in battery health management.

[0023] 6. The residual denoising network is specifically designed to correct residual noise in SOC estimation and is jointly trained with the SOH estimation model through an end-to-end optimization strategy. The loss function considers the errors of both SOC and SOH, and the gradient is backpropagated through K rounds of iterations to simultaneously optimize all model parameters. This design ensures the collaborative work between different modules, avoids suboptimal problems caused by individual training, and improves the overall efficiency and consistency of the method.

[0024] 7. A timestep embedding mechanism is introduced into the residual denoising network, and the iteration number k is used as the time step information to embed into the model, so that the network can dynamically adapt to the noise characteristics of different iteration stages. This design enhances the method's ability to process sequential data in time, ensures that the denoising process is gradually refined as the iteration progresses, and further improves the smoothness and convergence of the estimation. Attached Figure Description

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] Figure 1This is a flowchart of a joint estimation method for SOC and SOH of lithium batteries based on iterative denoising, according to the present invention.

[0027] Figure 2 This is a schematic diagram of the structure of a lithium battery SOC and SOH joint estimation system based on iterative denoising according to the present invention.

[0028] Figure 3 This is a flowchart illustrating the present invention. Detailed Implementation

[0029] The overall approach of the technical solution in this application is as follows: In each round, the SOH is first estimated using the current SOC sequence to dynamically update the battery capacity benchmark. Then, the inverse ampere-hour integral method is used for preliminary physical correction to ensure that the SOC trajectory conforms to electrochemical laws. Next, a residual denoising network based on the iteration rounds is used to perform data-driven fine correction of residual noise. Finally, the purified SOC of this round is fed back to the next round as input. This design allows SOH estimation to be based on a SOC sequence that is purified round by round with continuously decreasing noise levels, while SOC correction always depends on the dynamically updated capacity benchmark, thereby breaking the error coupling between the two. Through the fusion of physical constraints and data denoising, as well as the model's ability to perceive the iteration stage, the gradual suppression and self-correction of errors are achieved, thereby improving the accuracy and robustness of SOC and SOH estimation.

[0030] Please refer to Figures 1 to 3 As shown, a preferred embodiment of the present invention, a joint estimation method for SOC and SOH of lithium batteries based on iterative denoising, includes the following steps: Step S1: Set an iteration round identifier and set the initial value of the iteration round identifier to k (usually the initial value of k is set to a large integer, such as 5 or 10, to control the noise reduction intensity). Obtain the charging segment data of the lithium battery in the kth round and the nominal capacity at the factory. The charging segment data includes charging voltage, charging current, charging temperature and SOC. These data are collected from the sensors of the battery management system to ensure real-time performance and accuracy. k is used to control the noise reduction intensity to achieve stage adaptability, that is, early iterations (larger k) are coarsely adjusted to eliminate large noise, and later iterations (smaller k) are finely adjusted to optimize details. Step S2: Input the charging segment data into the SOH estimation model built based on the Transformer network to obtain the SOH estimate value of the kth round; Transformer networks utilize self-attention mechanisms to capture long-term dependencies between charging voltage, charging current, and charging temperature sequences, such as identifying the correlation between capacity decay and temperature fluctuations. Transformer outperforms models like RNN or TCN because it can process sequences in parallel, improving estimation efficiency, and is particularly suitable for real-time scenarios such as electric vehicles. Step S3: Based on the estimated SOH value and the nominal capacity at the factory, calculate the current available capacity for the kth round; Step S4: Using the inverse ampere-hour integration method, the SOC of the kth round is corrected based on the current available capacity to obtain the physically corrected SOC; this step is used to eliminate the systematic drift of SOC caused by capacity decay. Step S5: Input the physical correction SOC, SOH estimate and k into the residual denoising network to obtain the SOC noise data of the kth round. Correct the physical correction SOC based on the SOC noise data to obtain the residual correction SOC. Step S6: Use the residual corrected SOC as the SOC of the (k-1)th round, and then construct the charging segment data of the (k-1)th round. Update the value of the iteration round identifier to k-1, and perform iterative denoising for the (k-1)th round until the value of the iteration round identifier is 1. Output the final residual corrected SOC and SOH estimate. This forms a closed-loop feedback system, where the SOH estimate guides the SOC correction, and the corrected SOC feeds back into the SOH estimate, achieving collaborative optimization.

[0031] In step S1, the expression for the charging segment data of the kth round is: ; in, This represents the charging segment data in the k-th round; This represents the charging voltage at time t; This represents the charging current at time t; This represents the charging temperature at time t; Represents the SOC at time t in round k; This indicates the sequence length of the charging segment data, such as the number of sampling points in a single charging process; The charging voltage, charging current, and charging temperature remain constant throughout each iteration to avoid unnecessary disturbances to the original signal and preserve the authenticity of the battery's internal dynamic information.

[0032] In step S2, the formula for calculating the SOH estimate of the kth round is: ; in, This represents the estimated SOH value in the k-th round; This represents the charging segment data in the k-th round; () represents the SOH estimation model; In step S3, the formula for calculating the current available capacity in the k-th round is: ; In step S4, the formula for calculating the physical correction SOC is: ; ; in, Represents the physical correction SOC at time t in round k; Indicates the start time of charging; Indicates the time when charging ends; express Initial SOC at time step; This represents the current available capacity in round k; This represents the integral of the charging current from the start of charging to the current moment.

[0033] The inverse ampere-hour integration method incorporates the current available capacity to ensure that the SOC correction conforms to the battery's electrochemical and physical constraints, avoiding the output of a non-monotonic SOC trajectory.

[0034] The inverse ampere-hour integration method introduces the currently available capacity as the denominator on the basis of current integration to correct for systematic errors caused by capacity decay. For example, during charging, current integration reflects the accumulated charge, and dividing by the dynamic capacity ensures that the state of charge (SOC) increases monotonically, which conforms to electrochemical constraints.

[0035] In step S5, the formula for calculating the SOC noise data in the kth round is: ; in, This represents the SOC noise data for the k-th round; () represents a residual denoising network; This represents the physical correction SOC in round k; This represents the estimated SOH value in the k-th round; The formula for calculating the residual correction SOC is as follows: ; in, This represents the residual correction SOC in the k-th round, which is the SOC in the charging segment data of the (k-1)-th round. The residual denoising network is built based on a deep learning model. The training process is carried out in conjunction with the SOH estimation model. An end-to-end optimization strategy is adopted. The loss function includes the mean square error between the final SOC estimate and the true SOC label, and the mean square error between the final SOC estimate and the true SOC label. The gradient is backpropagated through K rounds of iterations to simultaneously optimize the model parameters of the SOH estimation model and the residual denoising network. The iteration number k is introduced in the residual denoising network through the timestep embedding mechanism, which enables the denoising process to perform bold corrections in the early iterations and fine-tuning in the later iterations, so as to accelerate convergence and suppress oscillations.

[0036] A preferred embodiment of the lithium battery SOC and SOH joint estimation system based on iterative denoising according to the present invention includes the following modules: The initialization module is used to set an iteration round identifier, and set the initial value of the iteration round identifier to k (usually the initial value of k is set to a large integer, such as 5 or 10, to control the noise reduction intensity). It acquires the charging segment data of the lithium battery in the kth round and the factory nominal capacity. The charging segment data includes charging voltage, charging current, charging temperature and SOC. These data are collected from the sensors of the battery management system to ensure real-time performance and accuracy. k is used to control the noise reduction intensity to achieve stage adaptability, that is, early iterations (larger k) are coarsely adjusted to eliminate large noise, and later iterations (smaller k) are finely adjusted to optimize details. The SOH estimation module is used to input the charging segment data into the SOH estimation model built based on the Transformer network to obtain the SOH estimate value in the kth round. Transformer networks utilize self-attention mechanisms to capture long-term dependencies between charging voltage, charging current, and charging temperature sequences, such as identifying the correlation between capacity decay and temperature fluctuations. Transformer outperforms models like RNN or TCN because it can process sequences in parallel, improving estimation efficiency, and is particularly suitable for real-time scenarios such as electric vehicles. The current available capacity estimation module is used to calculate the current available capacity in the k-th round based on the SOH estimate and the factory nominal capacity; The SOC physical correction module is used to correct the SOC of the k-th round based on the current available capacity using the inverse ampere-hour integration method to obtain the physically corrected SOC; this step is used to eliminate the systematic SOC drift caused by capacity decay. The SOC residual correction module is used to input the physical correction SOC, SOH estimate and k into the residual denoising network to obtain the SOC noise data of the kth round, and correct the physical correction SOC based on the SOC noise data to obtain the residual correction SOC. The iterative denoising module uses the residual corrected SOC as the SOC for the (k-1)th round, constructs the charging segment data for the (k-1)th round, updates the iteration round identifier to k-1, performs iterative denoising for the (k-1)th round, until the iteration round identifier is 1, and outputs the final residual corrected SOC and SOH estimate. This forms a closed-loop feedback system, where the SOH estimate guides the SOC correction, and the corrected SOC feeds back into the SOH estimate, achieving collaborative optimization.

[0037] In the initialization module, the expression for the charging segment data in the kth round is: ; in, This represents the charging segment data in the k-th round; This represents the charging voltage at time t; This represents the charging current at time t; This represents the charging temperature at time t; Represents the SOC at time t in round k; This indicates the sequence length of the charging segment data, such as the number of sampling points in a single charging process; The charging voltage, charging current, and charging temperature remain constant throughout each iteration to avoid unnecessary disturbances to the original signal and preserve the authenticity of the battery's internal dynamic information.

[0038] In the SOH estimation module, the formula for calculating the SOH estimate in the kth round is: ; in, This represents the estimated SOH value in the k-th round; This represents the charging segment data in the k-th round; () represents the SOH estimation model; In the current available capacity estimation module, the formula for calculating the current available capacity in the k-th round is: ; in, This represents the current available capacity in round k; This represents the estimated SOH value in the k-th round; Indicates the nominal capacity specified by the manufacturer.

[0039] For example, if the factory-rated capacity is 100Ah and the estimated SOH value is 0.9, then the current usable capacity is 90Ah, which effectively addresses the capacity degradation caused by battery aging.

[0040] In the SOC physical correction module, the calculation formula for the physically corrected SOC is: ; ; in, Represents the physical correction SOC at time t in round k; Indicates the start time of charging; Indicates the time when charging ends; express Initial SOC at time step; This represents the current available capacity in round k; This represents the integral of the charging current from the start of charging to the current moment.

[0041] The inverse ampere-hour integration method incorporates the current available capacity to ensure that the SOC correction conforms to the battery's electrochemical and physical constraints, avoiding the output of a non-monotonic SOC trajectory.

[0042] The inverse ampere-hour integration method introduces the currently available capacity as the denominator on the basis of current integration to correct for systematic errors caused by capacity decay. For example, during charging, current integration reflects the accumulated charge, and dividing by the dynamic capacity ensures that the state of charge (SOC) increases monotonically, which conforms to electrochemical constraints.

[0043] In the SOC residual correction module, the calculation formula for the SOC noise data in the k-th round is: ; in, This represents the SOC noise data for the k-th round; () represents a residual denoising network; This represents the physical correction SOC in round k; This represents the estimated SOH value in the k-th round; The formula for calculating the residual correction SOC is as follows: ; in, This represents the residual correction SOC in the k-th round, which is the SOC in the charging segment data of the (k-1)-th round. The residual denoising network is built based on a deep learning model. The training process is carried out in conjunction with the SOH estimation model. An end-to-end optimization strategy is adopted. The loss function includes the mean square error between the final SOC estimate and the true SOC label, and the mean square error between the final SOC estimate and the true SOC label. The gradient is backpropagated through K rounds of iterations to simultaneously optimize the model parameters of the SOH estimation model and the residual denoising network. The iteration number k is introduced in the residual denoising network through the timestep embedding mechanism, which enables the denoising process to perform bold corrections in the early iterations and fine-tuning in the later iterations, so as to accelerate convergence and suppress oscillations.

[0044] In summary, the advantages of this invention are: 1. By setting an iteration round identifier, the initial value of the iteration round identifier is set to k. The charging segment data and the nominal factory capacity of the lithium battery in the kth round are obtained. The charging segment data is input into the SOH estimation model to obtain the SOH estimate for the kth round. Based on the SOH estimate and the nominal factory capacity, the current available capacity of the kth round is calculated. Then, the SOC of the kth round is corrected based on the current available capacity using the inverse ampere-hour integral method to obtain the physically corrected SOC. The physically corrected SOC, the SOH estimate, and k are then input into the residual denoising network to obtain the SOC noise data for the kth round. The physically corrected SOC is corrected based on the SOC noise data to obtain the residual corrected SOC. The residual corrected SOC is then used as the SOC for the (k-1)th round, and the charging segment data for the (k-1)th round is constructed. The value of the iteration round identifier is updated to k-1, and iterative denoising is performed for the (k-1)th round until the value of the iteration round identifier is 1. The final residual corrected SOC is then output. Positive SOC and SOH estimates are constructed by building the entire estimation process into a multi-round iterative process from coarse to fine: In each round, the SOH is first estimated using the current (potentially noisy) SOC sequence to dynamically update the battery capacity benchmark. Then, the inverse ampere-hour integral method is used for preliminary physical correction to ensure that the SOC trajectory conforms to electrochemical laws. Next, a residual denoising network conditioned on the iteration round is used to perform data-driven fine correction of residual noise. Finally, the purified SOC of this round is fed back to the next round as input. This design allows SOH estimation to be based on a SOC sequence that is purified round by round with continuously decreasing noise levels, while SOC correction always depends on the dynamically updated capacity benchmark, thus breaking the error coupling between the two. Through the fusion of physical constraints and data denoising, as well as the model's ability to perceive the iteration stage, the gradual suppression and self-correction of errors are achieved, ultimately greatly improving the accuracy and robustness of SOC and SOH estimation.

[0045] 2. By integrating the Transformer network and the residual denoising network, the SOC and SOH of the lithium battery are estimated simultaneously within a single framework, avoiding the error accumulation and inefficiency problems caused by separate estimation in traditional methods. This joint estimation method can utilize the inherent correlation between SOC and SOH, such as dynamically correcting the capacity parameter in the SOC calculation through the SOH estimate, thereby improving the accuracy and consistency of the overall estimation.

[0046] 3. A multi-round iterative denoising process is adopted, gradually iterating from the kth round to the 1st round. In each round, the SOC noise data is corrected through a residual denoising network, gradually reducing the accumulated error. This iterative design is similar to the idea of ​​a denoising diffusion model, which can effectively handle uncertainties such as sensor noise and model bias, and finally output more stable SOC and SOH estimates. Compared with the single estimation method, iterative denoising improves the robustness and anti-interference ability of the method, and is especially suitable for long-term monitoring of lithium batteries under complex working conditions.

[0047] 4. By combining the physically based inverse ampere-hour integral method with data-driven deep learning models (such as Transformer and residual denoising networks), the inverse ampere-hour integral method provides preliminary SOC correction under physical constraints, ensuring that the results conform to the electrochemical principles of batteries; while the deep learning model learns nonlinear relationships from the data to compensate for the shortcomings of the physical model. This hybrid approach retains the interpretability of model-driven methods while leveraging the adaptability of data-driven methods, reducing the risk of overfitting and improving reliability in real-world scenarios.

[0048] 5. In the SOH estimation stage, a model is built based on the Transformer network, which is good at processing time series data, such as charging voltage, current and temperature series. The self-attention mechanism of the Transformer can capture long-term dependencies, thereby extracting SOH-related features from the charging segment data more accurately. This not only improves the accuracy of SOH estimation, but also enhances the method's ability to predict battery aging trends, demonstrating the effective application of advanced deep learning technology in battery health management.

[0049] 6. The residual denoising network is specifically designed to correct residual noise in SOC estimation and is jointly trained with the SOH estimation model through an end-to-end optimization strategy. The loss function considers the errors of both SOC and SOH, and the gradient is backpropagated through K rounds of iterations to simultaneously optimize all model parameters. This design ensures the collaborative work between different modules, avoids suboptimal problems caused by individual training, and improves the overall efficiency and consistency of the method.

[0050] 7. A timestep embedding mechanism is introduced into the residual denoising network, and the iteration number k is used as the time step information to embed into the model, so that the network can dynamically adapt to the noise characteristics of different iteration stages. This design enhances the method's ability to process sequential data in time, ensures that the denoising process is gradually refined as the iteration progresses, and further improves the smoothness and convergence of the estimation.

[0051] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A joint estimation method for SOC and SOH of lithium batteries based on iterative denoising, characterized in that: Includes the following steps: Step S1: Set an iteration round identifier, set the initial value of the iteration round identifier to k, and obtain the charging segment data and factory nominal capacity of the lithium battery in the kth round; the charging segment data includes charging voltage, charging current, charging temperature and SOC; Step S2: Input the charging segment data into the SOH estimation model built based on the Transformer network to obtain the SOH estimate value of the kth round; Step S3: Based on the estimated SOH value and the nominal capacity at the factory, calculate the current available capacity for the kth round; Step S4: Using the inverse ampere-hour integration method, the SOC of the kth round is corrected based on the current available capacity to obtain the physically corrected SOC; Step S5: Input the physical correction SOC, SOH estimate and k into the residual denoising network to obtain the SOC noise data of the kth round. Correct the physical correction SOC based on the SOC noise data to obtain the residual correction SOC. Step S6: Use the residual correction SOC as the SOC of the (k-1)th round, and then construct the charging segment data of the (k-1)th round. Update the value of the iteration round identifier to k-1, perform iterative denoising in the (k-1)th round, until the value of the iteration round identifier is 1, and output the final residual correction SOC and SOH estimate.

2. The method for joint estimation of SOC and SOH of lithium batteries based on iterative denoising as described in claim 1, characterized in that: In step S1, the expression for the charging segment data of the kth round is: ; in, This represents the charging segment data in the k-th round; This represents the charging voltage at time t; This represents the charging current at time t; This represents the charging temperature at time t; Represents the SOC at time t in round k; Indicates the sequence length of the charging segment data; The charging voltage, charging current, and charging temperature remain constant throughout each iteration.

3. The method for joint estimation of SOC and SOH of lithium batteries based on iterative denoising as described in claim 1, characterized in that: In step S2, the formula for calculating the SOH estimate of the kth round is: ; in, This represents the estimated SOH value in the k-th round; This represents the charging segment data in the k-th round; () represents the SOH estimation model; In step S3, the formula for calculating the current available capacity in the k-th round is: ; in, This represents the current available capacity in round k; This represents the estimated SOH value in the k-th round; Indicates the nominal capacity specified by the manufacturer.

4. The method for joint estimation of SOC and SOH of lithium batteries based on iterative denoising as described in claim 1, characterized in that: In step S4, the formula for calculating the physical correction SOC is: ; ; in, Represents the physical correction SOC at time t in round k; Indicates the start time of charging; Indicates the time when charging ends; express Initial SOC at time step; This represents the current available capacity in round k; This represents the integral of the charging current from the start of charging to the current moment.

5. The method for joint estimation of SOC and SOH of lithium batteries based on iterative denoising as described in claim 1, characterized in that: In step S5, the formula for calculating the SOC noise data in the kth round is: ; in, This represents the SOC noise data for the k-th round; This represents a residual denoising network; This represents the physical correction SOC in round k; This represents the estimated SOH value in the k-th round; The formula for calculating the residual correction SOC is as follows: ; in, This represents the residual correction SOC in the k-th round, which is the SOC in the charging segment data of the (k-1)-th round. The residual denoising network is built based on a deep learning model. The training process is carried out in conjunction with the SOH estimation model. An end-to-end optimization strategy is adopted. The loss function includes the mean square error between the final SOC estimate and the true SOC label, and the mean square error between the final SOC estimate and the true SOC label. The gradient is backpropagated through K rounds of iterations to simultaneously optimize the model parameters of the SOH estimation model and the residual denoising network. The iteration round k is introduced in the residual denoising network through the timestep embedding mechanism.

6. A joint estimation system for SOC and SOH of lithium batteries based on iterative denoising, characterized in that: Includes the following modules: An initialization module is used to set an iteration round identifier, set the initial value of the iteration round identifier to k, and obtain the charging segment data of the lithium battery in the kth round and the nominal capacity at the factory; the charging segment data includes charging voltage, charging current, charging temperature and SOC; The SOH estimation module is used to input the charging segment data into the SOH estimation model built based on the Transformer network to obtain the SOH estimate value in the kth round. The current available capacity estimation module is used to calculate the current available capacity in the k-th round based on the SOH estimate and the factory nominal capacity; The SOC physical correction module is used to correct the SOC of the k-th round based on the current available capacity using the inverse ampere-hour integration method to obtain the physically corrected SOC. The SOC residual correction module is used to input the physical correction SOC, SOH estimate and k into the residual denoising network to obtain the SOC noise data of the kth round, and correct the physical correction SOC based on the SOC noise data to obtain the residual correction SOC. The iterative denoising module is used to take the residual corrected SOC as the SOC of the (k-1)th round, and then construct the charging segment data of the (k-1)th round. The value of the iteration round identifier is updated to k-1, and iterative denoising is performed in the (k-1)th round until the value of the iteration round identifier is 1. Finally, the residual corrected SOC and SOH estimate are output.

7. The lithium battery SOC and SOH joint estimation system based on iterative denoising as described in claim 6, characterized in that: In the initialization module, the expression for the charging segment data in the kth round is: ; in, This represents the charging segment data in the k-th round; This represents the charging voltage at time t; This represents the charging current at time t; This represents the charging temperature at time t; Represents the SOC at time t in round k; Indicates the sequence length of the charging segment data; The charging voltage, charging current, and charging temperature remain constant throughout each iteration.

8. The lithium battery SOC and SOH joint estimation system based on iterative denoising as described in claim 6, characterized in that: In the SOH estimation module, the formula for calculating the SOH estimate in the kth round is: ; in, This represents the estimated SOH value in the k-th round; This represents the charging segment data in the k-th round; This represents the SOH estimation model; In the current available capacity estimation module, the formula for calculating the current available capacity in the k-th round is: ; in, This represents the current available capacity in round k; This represents the estimated SOH value in the k-th round; Indicates the nominal capacity specified by the manufacturer.

9. A lithium battery SOC and SOH joint estimation system based on iterative denoising as described in claim 6, characterized in that: In the SOC physical correction module, the calculation formula for the physically corrected SOC is: ; ; in, Represents the physical correction SOC at time t in round k; Indicates the start time of charging; Indicates the time when charging ends; express Initial SOC at time step; This represents the current available capacity in round k; This represents the integral of the charging current from the start of charging to the current moment.

10. The lithium battery SOC and SOH joint estimation system based on iterative denoising as described in claim 6, characterized in that: In the SOC residual correction module, the calculation formula for the SOC noise data in the k-th round is: ; in, This represents the SOC noise data for the k-th round; This represents a residual denoising network; This represents the physical correction SOC in round k; This represents the estimated SOH value in the k-th round; The formula for calculating the residual correction SOC is as follows: ; in, This represents the residual correction SOC in the k-th round, which is the SOC in the charging segment data of the (k-1)-th round. The residual denoising network is built based on a deep learning model. The training process is carried out in conjunction with the SOH estimation model. An end-to-end optimization strategy is adopted. The loss function includes the mean square error between the final SOC estimate and the true SOC label, and the mean square error between the final SOC estimate and the true SOC label. The gradient is backpropagated through K rounds of iterations to simultaneously optimize the model parameters of the SOH estimation model and the residual denoising network. The iteration round k is introduced in the residual denoising network through the timestep embedding mechanism.

Images