Power battery charge state estimation method and device and storage medium

Abstract

The invention relates to a power battery state-of-charge estimation method and device and a storage medium, and the method comprises the steps: S1, obtaining the predicted values of an open-circuit voltage and a terminal voltage at a t moment through a second-order RC equivalent circuit model; s2, obtaining a predicted value of the heat production quantity at the moment t by utilizing a heat production equation; s3, obtaining a predicted value of the battery temperature at the t + 1 moment by using the single-state thermal model; s4, obtaining an SOC predicted value at the t + 1 moment by using the first prediction model; and S5, calculating the terminal voltage at the same moment and the difference between the measured value and the predicted value as an observation error, and inputting the observation error from the t-k moment to the t + 1 moment, the battery parameters, the SOC predicted value and the SOC correction value from the t-k moment to the t moment into the time sequence prediction model to obtain the SOC correction value at the t + 1 moment. Compared with the prior art, the SOC prediction precision under the unsteady state working condition can be improved.

Description

Technical Field

[0001] This invention relates to the field of power battery state estimation technology, and in particular to a method, apparatus and storage medium for estimating the state of charge of a power battery. Background Technology

[0002] Lithium-ion batteries, with their higher energy density, longer cycle life, and lower self-discharge rate, play a crucial role in new electric vehicles. As the electrification of construction machinery accelerates, the accuracy of the state-of-charge (SOC) estimation of power batteries directly impacts equipment operational safety and range reliability. Accurate SOC estimation is essential for improving vehicle operational safety, optimizing performance, and extending battery cycle life.

[0003] For example, Chinese patent CN114091282A discloses a lithium-ion battery state estimation method and system based on a fractional-order model. This method includes: constructing a lithium-ion battery coupling model; estimating the parameters of the lithium-ion battery coupling model; and using a two-layer unscented Kalman joint estimation method for state estimation. The output of the Kalman filter estimating the SOC of the lithium-ion battery is used as the input of the Kalman filter estimating the SOH, and the output SOH is used as the input of the Kalman filter estimating the SOC at the next time step. This process is iterated continuously to update the SOC and SOH values ​​in real time. However, this method has the following drawbacks: 1. Although it includes a thermal model, this model primarily describes the heat generation and dissipation processes inside the battery and does not deeply participate in the closed-loop correction of state estimation. In the core algorithm of SOC / SOH estimation, temperature is a static parameter or external input. The model fails to fully reflect the real-time impact of dynamic temperature changes on the battery's internal parameters and to utilize this impact to correct the SOC estimation results. Its thermal coupling is more evident in the model construction stage than in the dynamic feedback of the estimation process.

[0004] 2. Relying on the accuracy of the UKF algorithm and the model itself, it lacks a dedicated mechanism to handle the cumulative errors caused by model mismatch, time-varying parameters, and unmodeled dynamics. Although UKF can handle nonlinearity, it is very sensitive to the accuracy of the system model and noise statistics.

[0005] 3. Its verification may be more based on standard HPPC tests or FUDS conditions, lacking detailed performance verification and comparison under extreme and transient conditions for specific application scenarios (such as construction machinery).

[0006] 4. Fractional-order models are inherently more complex to compute than integer-order models. In addition, the iterative computation of the two-layer UKF places higher demands on the hardware computing power of the battery management system, which may affect real-time performance.

[0007] In summary, among the aforementioned existing technologies, the prediction accuracy of SOC for unsteady operating conditions is relatively low. Summary of the Invention

[0008] The purpose of this invention is to provide a method, device, and storage medium for estimating the state of charge of a power battery in order to overcome the deficiencies of the prior art.

[0009] The objective of this invention can be achieved through the following technical solutions: A method for estimating the state of charge of a power battery, comprising: Step S1: Input the SOC correction value at time t, the predicted value of the battery temperature at time t, and the measured value of the current at time t into the second-order RC equivalent circuit model to obtain the predicted values ​​of the open-circuit voltage and terminal voltage at time t. Step S2: Substitute the predicted values ​​of battery temperature, terminal voltage, and open-circuit voltage at time t, and the measured value of current at time t, into the heat generation equation to obtain the predicted value of heat generation at time t. Step S3: Substitute the predicted heat generation at time t and the measured battery temperature at time t into the single-state thermal model to obtain the predicted battery temperature at time t+1. Step S4: Substitute the measured values ​​of the terminal voltage and current at time t into the first prediction model to obtain the predicted SOC value at time t+1; Step S5: Calculate the difference between the measured and predicted values ​​of the terminal voltage at the same time as the observation error, and input the observation error, battery parameters, and predicted SOC value from time tk to time t+1, as well as the SOC correction value from time tk to time t, into the time series prediction model to obtain the SOC correction value at time t+1.

[0010] The second-order RC equivalent circuit model is as follows: ; ; in: This is the predicted value of the open-circuit voltage. This is the open-circuit voltage prediction function. This is the SOC correction value. This is a predicted value for battery temperature. This is the predicted value of the terminal voltage. The internal resistance of the battery is in ohms. This is the measured value of the current. For the first polarization resistor, This is the second polarization resistor.

[0011] The heat production equation is as follows: ; in: This is a predicted value for heat production. This is the predicted value of the open-circuit voltage. This is the predicted value of the terminal voltage. This is the measured value of the current. This is a predicted value for battery temperature.

[0012] The singlet thermal model is as follows: ; in: For pyrogenic rate, This is the measured value of the battery temperature. This is the measured value of the ambient temperature. This is the equivalent thermal resistance.

[0013] The model parameters of the singlet thermal model include heat generation rate and equivalent thermal resistance; The model parameters of the singlet thermal model are obtained through optimization using an adaptive genetic algorithm, and the fitness function of the optimization process is... f 3 is: ; in: This is the sequence of measured battery temperatures for the kth batch. Let N be the predicted sequence of battery temperatures for the k-th batch, and N be the number of training batches.

[0014] The crossover probability and mutation probability of the adaptive genetic algorithm are: ; ; in: For crossover probability, This is the adjustment coefficient for the first crossover probability. This represents the optimal fitness value for the population. A larger fitness value in the crossover operation. The average fitness value of the population. This is the adjustment coefficient for the second crossover probability. This is the first mutation probability adjustment coefficient. The mutation probability, This represents the fitness value during the mutation operation. This is the second variation probability adjustment coefficient.

[0015] The first prediction model is constructed based on the fractional-order multi-innovation adaptive unscented Kalman filter algorithm.

[0016] The time series prediction model is constructed using a long short-term memory network, and its hyperparameters include the initial learning rate, the number of hidden layer nodes, and the regularization coefficient of the long short-term memory network. The battery parameters include the measured values ​​of current, terminal voltage, and battery temperature. The hyperparameters of the time series prediction model are optimized using an improved gray wolf optimization algorithm, wherein the convergence factor is adjusted by the number of iterations in the improved gray wolf optimization algorithm. ; in: The convergence factor is A random number uniformly distributed within the interval [0,1]. The maximum number of iterations, l This represents the current iteration number.

[0017] A power battery state of charge estimation device includes a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the method described above.

[0018] A storage medium having a program stored thereon, which, when executed, implements the method described above.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. An electrothermal coupling model is constructed based on a second-order RC equivalent circuit model, heat generation equation, and single-state thermal model, considering the electrothermal coupling effect. Subsequently, the state of charge (SOC) and temperature of the power battery are estimated simultaneously. Finally, the difference in terminal voltage is used as the observation error, and the initial SOC prediction value and observation error are used as input parameters. A time series prediction model is employed to dynamically compensate and correct the estimation error. This method effectively improves the accuracy of battery SOC estimation in the battery management system (BMS) of electric bulldozers, providing more reliable technical support for battery state monitoring and safety management.

[0020] 2. By introducing two polarization resistors, the dynamic response characteristics of the battery are simulated more accurately, especially the polarization effect under non-steady-state conditions. The fractional-order model (FOM) used fits the actual voltage better than the integer-order model (IOM), reducing prediction errors. For example, at -20℃, 0℃, and 25℃, the root mean square error (RMSE) and mean absolute error (MAE) of the terminal voltage prediction of FOM are lower than those of IOM, demonstrating the model's superiority in variable environments. Furthermore, the model integrates temperature variables into voltage prediction, providing a foundation for subsequent heat generation and temperature estimation, thereby improving the coherence of the overall SOC estimation.

[0021] 3. The heat generation equation combines Joule heating and reversible thermal effects through a temperature entropy variable, avoiding the bias in heat generation estimation found in traditional methods. Accurate prediction of heat generation is crucial for the temperature model input, ensuring the realism of the electrothermal coupling model. Furthermore, real-time heat generation data provides reliable input for the singlet thermal model, thereby improving the accuracy of subsequent temperature predictions and indirectly enhancing the reliability of SOC estimation.

[0022] 4. The singlet model has a simple structure, is easy to identify parameters and apply in real time, reducing computational complexity while maintaining high accuracy. Under HPPC conditions, the RMSE and MAE of temperature estimation are both below 0.2℃, proving the model's effectiveness under varying loads. Furthermore, the model considers ambient temperature and thermal resistance, enabling unsteady equipment such as bulldozers to maintain stable temperature predictions even in high or low temperature environments.

[0023] 5. AGA dynamically adjusts crossover and mutation probabilities to avoid local optima and ensure that thermal model parameters accurately fit different batches of data. The document states that this optimization minimizes temperature prediction errors, improves model generalization ability, and, through multi-batch training, adapts to load fluctuations in bulldozer operations, reducing estimation bias caused by changes in operating conditions.

[0024] 6. By dynamically adjusting the probability through a fitness threshold, mutations are increased at high fitness to explore new solutions, while stability is maintained at low fitness to accelerate convergence. This improves optimization efficiency and shortens parameter identification time. Furthermore, the adaptive mechanism makes the algorithm suitable for real-time updates of the battery management system, especially under non-steady-state conditions such as heavy-load transient bulldozing (HDTO), enabling rapid response to changes.

[0025] 7. The fractional-order characteristic captures the memory effect of battery diffusion dynamics, and the multi-innovation technology uses historical data to reduce the impact of noise. The SOC estimation RMSE and MAE of FOMIAUKF under three bulldozer operating conditions are lower than those of traditional EKF and AUKF. For example, the RMSE is reduced by more than 45% at 25°C. It performs well under load surge conditions such as HDTO, reducing SOC drift caused by current fluctuations.

[0026] 8. LSTM handles time series dependencies, effectively compensating for SOC prediction errors; IGWO optimizes hyperparameters to avoid overfitting. After correction by AGWO-LSTM, the SOC estimation error is further reduced. For example, under LDCO conditions, RMSE is reduced by 52.3% compared to FOMIAUKF. In addition, the convergence factor introduces randomness, which improves the global search capability of the optimization process, enabling the model to maintain high accuracy under extreme conditions such as low temperature. Attached Figure Description

[0027] Figure 1This is a schematic diagram of the main steps of the method of the present invention; Figure 2 This is a graph showing the voltage and error curves estimated by the thermal model under -20℃ conditions in an embodiment of the present invention. Figure 3 This is a graph showing the voltage and error curves estimated by the thermal model under 0℃ conditions in an embodiment of the present invention. Figure 4 This is a graph showing the voltage and error curves estimated by the thermal model under 25°C conditions in an embodiment of the present invention. Figure 5 These are the voltage RMSE and MAE histograms estimated by the thermal model under conditions of -20℃, 0℃, and 25℃ in the embodiments of this invention; Figure 6 This is a graph showing the temperature and error curves estimated by the thermal model under -20℃ conditions in an embodiment of the present invention. Figure 7 This is a graph showing the temperature and error curves estimated by the thermal model under 0℃ conditions in an embodiment of the present invention. Figure 8 This is a graph showing the temperature and error curves estimated by the thermal model at 25°C in an embodiment of the present invention. Figure 9 These are the temperature RMSE and MAE histograms estimated by the thermal model under the conditions of -20℃, 0℃ and 25℃ in the embodiments of this invention. Detailed Implementation

[0028] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0029] A method for estimating the state of charge of a power battery, such as Figure 1 As shown, it includes: Step S1: Input the SOC correction value at time t, the predicted value of the battery temperature at time t, and the measured value of the current at time t into the second-order RC equivalent circuit model to obtain the predicted values ​​of the open-circuit voltage and terminal voltage at time t. In this embodiment, the Hybrid Pulse Power Performance Testing (HPPC) method is employed to evaluate the dynamic performance of the power battery under pulse charge-discharge conditions, and a parameter identification algorithm is used to identify the parameters of the electrical and thermal models. As an example, the testing cycle of a single power battery cell is divided into an AG phase. The specific process of the HPPC testing scheme is as follows: Before stage A, the battery is kept at rest, and baseline parameters are established. Upon entering stage AB, a 3x discharge pulse is applied. Due to the battery's internal resistance, the voltage drops sharply due to the sudden increase in current. Upon entering stage BC, a 3x discharge is applied for 10 seconds. Upon entering stage CE, the battery is allowed to rest for 40 seconds to stabilize its state. Upon entering stage EG, the battery is charged at a 3x rate for 10 seconds, followed by a 20-second rest period. Then, a 3x discharge is applied to reduce the state of charge (SOC) by 0.1%, and finally, the battery is allowed to rest for 1 hour to ensure complete relaxation. This test procedure accurately characterizes the battery's dynamic response under controlled conditions.

[0030] In this embodiment, a capacity calibration test method is adopted. By using a constant current and constant voltage charging and constant current discharging cycle scheme, the actual capacity of the power battery dependent on temperature can be quantified, laying the foundation for accurately estimating the battery's SOC.

[0031] It's understandable that during the charging and discharging process of a power battery, the terminal voltage is affected by polarization effects and ohmic resistance, making direct measurement difficult. This is because the open-circuit voltage... As a reliable approximation of the terminal voltage, it has a clear functional relationship with the State of Charge (SOC). To present the voltage variation pattern across the entire SOC range, this embodiment uses an eighth-order polynomial equation to represent the open-circuit voltage. Modeling the relationship between SOC and voltage regulation (SOC) allows for the accurate representation of voltage variation patterns across the entire SOC range using fitted curves.

[0032] This embodiment determines the relationship between the temperature-entropy coefficient and the SOC by measuring the equivalent open-circuit voltage under different SOC levels and temperature conditions and fitting the measurement data.

[0033] In this embodiment, based on the electrical and thermal characteristics of lithium-ion single cells under dynamic operating conditions, a second-order RC equivalent circuit model and a single-state thermal model are established. The generated heat, battery temperature, SOC and voltage are used as bridging variables to couple the models, resulting in an electrothermal coupling model.

[0034] Power batteries exhibit nonlinear and strongly coupled electrochemical reaction processes, which poses a challenge to traditional ionic conductor materials in accurately capturing diffusion kinetics and polarization behavior. Fractional calculus, with its inherent memory dependence and nonlocality, provides a better theoretical basis for describing the non-uniform transport and relaxation processes of lithium ions.

[0035] Electrochemical impedance spectroscopy (EIS), as an important electrochemical characterization technique, is commonly used to investigate electrode reaction kinetics, interfacial properties, and material performance. Its spectrum is typically divided into three frequency ranges: high frequency, mid frequency, and low frequency. These three ranges correspond to interfacial polarization effects, electrochemical kinetic processes, and diffusion effects, respectively, and can all be modeled using equivalent circuit elements to characterize the frequency dependence and response characteristics of each process.

[0036] The equivalent circuit model of the second-order RC circuit is as follows: ; ; in: This is the predicted value of the open-circuit voltage. This is the open-circuit voltage prediction function. This is the SOC correction value. This is a predicted value for battery temperature. This is the predicted value of the terminal voltage. The internal resistance of the battery is in ohms. This is the measured value of the current. For the first polarization resistor, This is the second polarization resistor.

[0037] The model parameters of the second-order RC equivalent circuit model can be identified by a genetic algorithm. The optimization objective is to minimize the sum of squared errors between the model open-circuit voltage and the measured terminal voltage.

[0038] Step S2: Substitute the predicted values ​​of battery temperature, terminal voltage, and open-circuit voltage at time t, and the measured value of current at time t, into the heat generation equation to obtain the predicted value of heat generation at time t. The heat generation equation is as follows: ; in: This is a predicted value for heat production. This is the predicted value of the open-circuit voltage. This is the predicted value of the terminal voltage. This is the measured value of the current. This is a predicted value for battery temperature.

[0039] Step S3: Substitute the predicted heat generation at time t and the measured battery temperature at time t into the single-state thermal model to obtain the predicted battery temperature at time t+1. The singlet thermal model is as follows: ; in: For pyrogenic rate, This is the measured value of the battery temperature. This is the measured value of the ambient temperature. This is the equivalent thermal resistance.

[0040] Electrothermal coupling is placed at the core. The technical solution clearly forms a closed loop: current and voltage → heat generation → temperature prediction → temperature feedback to the electrical model to correct parameters → SOC estimation. Temperature is a state variable that is estimated in real time and is directly used to improve the accuracy of the SOC estimation at the next moment, which is crucial for accuracy under non-steady-state conditions (such as high-current charging and discharging, and changes in ambient temperature).

[0041] In this embodiment, the model parameters of the singlet thermal model include the heating rate and equivalent thermal resistance. These parameters are obtained through an adaptive genetic algorithm optimization process, with the fitness function used in the optimization process... f 3 is: ; in: This is the sequence of measured battery temperatures for the kth batch. Let N be the predicted sequence of battery temperatures for the k-th batch, and N be the number of training batches.

[0042] The crossover and mutation probabilities of the adaptive genetic algorithm are: ; ; in: For crossover probability, This is the adjustment coefficient for the first crossover probability. This represents the optimal fitness value for the population. A larger fitness value in the crossover operation. The average fitness value of the population. This is the adjustment coefficient for the second crossover probability. This is the first mutation probability adjustment coefficient. The mutation probability, This represents the fitness value during the mutation operation. This is the second variation probability adjustment coefficient.

[0043] Step S4: Substitute the measured values ​​of the terminal voltage and current at time t into the first prediction model to obtain the predicted SOC value at time t+1; In this embodiment, the first prediction model is constructed based on the fractional-order multi-innovation adaptive unscented Kalman filter algorithm.

[0044] Step S5: Calculate the difference between the measured and predicted values ​​of the terminal voltage at the same time as the observation error, and input the observation error, battery parameters, and predicted SOC value from time tk to time t+1, as well as the SOC correction value from time tk to time t, into the time series prediction model to obtain the SOC correction value at time t+1.

[0045] Temporal correlations are crucial for modeling sequential data. Traditional recurrent neural networks (RNNs) suffer from vanishing or exploding gradients, fundamentally limiting their ability to model long-range dependencies. To address this problem, Long Short-Term Memory (LSTM) networks employ a gating system comprising forget gates, input gates, and output gates, creating a deep structure with selective memory capabilities. The forget gate controls the retention of information from historical states, the input gate regulates the information to be stored in the cell state and stores it as a new candidate value, and the output gate determines the final output.

[0046] An innovative LSTM-based error correction model was introduced, instead of directly using LSTM for SOC prediction. This model takes the observed voltage error, historical SOC values, and battery parameters as time series inputs, dynamically learns and predicts the SOC estimation error, thus providing real-time compensation for the initial FOMIAUKF estimation results. This hybrid "model-driven + data-driven" strategy significantly enhances robustness to complex operating conditions and model uncertainties. After LSTM correction, the SOC estimation error is greatly reduced.

[0047] In this embodiment, the time series prediction model is constructed using a long short-term memory network, and its hyperparameters include the initial learning rate, the number of hidden layer nodes, and the regularization coefficient of the long short-term memory network. Battery parameters include measured values ​​of current, terminal voltage, and battery temperature. The Grey Wolf Optimization Algorithm (GWO) is an efficient metaheuristic optimization method inspired by the social hierarchy and cooperative hunting strategies of grey wolf populations. This algorithm maps the parameters to be optimized to the positions of individual grey wolves within the population. By simulating the dynamic process of α, β, and δ grey wolves guiding the population to hunt prey, parameter optimization can be achieved. The GWO calculates and ranks the fitness value of each grey wolf, where the α grey wolf represents the optimal solution, the second and third ranked individuals are the β and δ grey wolves respectively, and the remaining individuals are the ω grey wolves, representing candidate solutions.

[0048] The group hunting mechanism of gray wolves mainly includes three stages: surrounding the prey, chasing the prey, and attacking the prey. α, β, and δ gray wolves constitute the leadership hierarchy. They use their current optimal position information to guide the population to move towards the potentially optimal area. The remaining individuals in the population dynamically update their own positions based on the position coordinates of these three elite gray wolves.

[0049] Gray wolf packs estimate prey locations using individual α, β, and δ gray wolves, while the remaining wolves continuously update their positions to surround the prey. Once the prey's location is accurately determined and the prey is stationary, the pack launches an attack to capture it.

[0050] The standard gray wolf optimization algorithm has two shortcomings: first, it requires setting a maximum number of iterations, which affects the total computation time; second, it requires defining key parameters to balance the exploration and development capabilities during the optimization process.

[0051] In this embodiment, the hyperparameters of the time series prediction model are optimized using an improved gray wolf optimization algorithm. In this improved gray wolf optimization algorithm, the convergence factor is adjusted by the number of iterations. ; in: The convergence factor is A random number uniformly distributed within the interval [0,1]. The maximum number of iterations, l This represents the current iteration number.

[0052] This algorithm improves its performance and adaptability by adaptively adjusting the convergence factor of the standard Grey Wolf algorithm, thereby enhancing its effectiveness in solving various optimization problems. The algorithm introduces adaptive adjustment of the convergence factor, giving it dynamic adjustment capabilities.

[0053] The performance and convergence speed of a Long Short-Term Memory (LSTM) network are determined by the values ​​of its hyperparameters, including the initial learning rate, the number of hidden layer nodes, and the regularization coefficient. This embodiment uses the Adaptive Gray Wolf Optimization (AGWO) algorithm to determine the optimal hyperparameters of the LSTM network. During the training of the error correction model, the AGWO optimization process for the LSTM network's hyperparameters includes the following steps: Define an optimization parameter space, which includes the range of values ​​for the initial learning rate, the number of hidden layer nodes, and the regularization coefficient of the LSTM network; The convergence characteristics of the gray wolf optimization algorithm are improved by adopting a nonlinear convergence factor, and the algorithm's exploration and development capabilities are dynamically adjusted during the iteration process. When the preset maximum number of iterations is reached or the rate of change of the fitness function is less than a set threshold, the optimal combination of hyperparameters is output. The LSTM network structure is constructed based on the optimal hyperparameter combination.

[0054] In this embodiment, the process of establishing the AGWO-LSTM error correction model includes the following steps: Step 1: Data Processing. First, the selected battery characteristic parameters are normalized. The data is then allocated to training and test sets according to a predetermined ratio. As an example, 70% of the data is allocated to the training set, and the remaining 30% is used as the test set.

[0055] Step 2: Parameter Configuration. Set the gray wolf population size, maximum number of iterations, and population dimension. As an example, the gray wolf population size is 20, the maximum number of iterations is 50, and the population dimension is 4.

[0056] Step 3: Define the fitness function. The fitness value of each gray wolf is calculated using the mean squared error (MSE).

[0057] Step 4: Hyperparameter Optimization. After parameter initialization, the adaptive gray wolf optimization algorithm is used to optimize the hyperparameters of the Long Short-Term Memory network to improve the model's prediction accuracy and generalization ability.

[0058] Step 5: Iterative optimization process. The position of each gray wolf is continuously updated through iterative optimization to determine if the maximum number of iterations or the minimum fitness value has been reached. If the conditions are met, the optimal hyperparameters of the Long Short-Term Memory network are output; otherwise, this step is repeated.

[0059] Step 6: Parameter Assignment and Model Evaluation. Train the Long Short-Term Memory network using the optimized hyperparameters and evaluate its performance.

[0060] Electric bulldozers place more stringent technical requirements on their power battery systems due to their complex operating conditions, highly dynamic operational targets, and rapid fluctuations in cyclic load caused by frequent short-term peak demands. Considering the operational characteristics of electric bulldozers in complex working conditions and variable environments, accurate power battery SOC assessment plays a crucial role in improving operational safety, optimizing performance, and extending battery life.

[0061] A bulldozer's complete operating process typically includes stages such as unloaded forward movement, cutting soil, heavy-load pushing, unloading, and unloaded return, characterized by sudden load increases and periodic changes. To accurately simulate complex working conditions and verify the effectiveness of the aforementioned SOC estimation method, this embodiment designs three typical operating modes. By collecting working condition data under actual operating scenarios, load change data is recorded throughout the entire work cycle. The three typical operating modes include: Lightly cyclical operation (LDCO) is a mode in which the bulldozer operates at 1 / 3 of its bucket capacity, simulating low-intensity leveling, loose material transfer and similar conditions, in which load changes are relatively stable and periodic.

[0062] Standard Continuous Operation (SDCO) mode, in which the bulldozer operates at 2 / 3 of its bucket capacity, represents typical earthmoving and excavation scenarios with moderate and continuous load intensity.

[0063] Heavy Transient Operation (HDTO) is a mode in which the bulldozer operates at full bucket capacity, simulating high-resistance bulldozing and digging, steep slope climbing, hard material removal, and other high-load conditions, characterized by instantaneous load surges and intense impacts.

[0064] To verify the accuracy of the model, experimental data of HPPC discharge conditions under different ambient temperatures were used for verification. Figure 2-5 The terminal voltages and their error estimation results at -20℃, 0℃, and 25℃ are presented respectively. It can be seen that for terminal voltage prediction at different temperatures, both integer-order IOM and fractional-order FOM can accurately fit the actual measured values. However, from the enlarged sub-figures, it can be observed that the simulation results of fractional-order FOM have a higher degree of fit with the actual measured values. Meanwhile, from... Figure 5 As can be seen, the root mean square error (RMSE) and mean absolute error (MAE) of the FOM terminal voltage are significantly better than those of the I0M at all three temperatures. The maximum RMSE and MAE of the integer order IOM are 34.6mV and 20.4mV, respectively, while those of the FOM are 31.9mV and 14.2mV, respectively.

[0065] Figure 6-9 The RMSE and MAE of the thermal model temperature under different ambient temperature conditions are presented. The results show that under HPPC conditions, the maximum RMSE and MAE of the temperature estimate remain below 0.2℃, with the peak value of RMSE being 0.193 and the peak value of MAE being 0.158.

[0066] In this embodiment, the SOC estimation performance of the proposed method is verified based on three typical bulldozer operating conditions: light load cyclic operation (LDCO), heavy load transient operation (HDTO), and standard load continuous operation (SDCO), under ambient temperatures of -20℃, 0℃, and 25℃.

[0067] Under the three operating conditions, the SOC curves obtained by the fractional-order multi-innovation adaptive unscented Kalman filter (FOMIAUKF), the adaptive unscented Kalman filter (AUKF), and the extended Kalman filter (EKF) all showed good agreement with the actual battery SOC curves. However, as can be seen from the enlarged sub-figures, the SOC estimation curve of the FOMIAUKF algorithm had a higher degree of overlap with the actual curve than that of the AUKF and EKF algorithms, indicating that the FOMIAUKF algorithm has better estimation accuracy.

[0068] It is worth noting that among the three algorithms, the EKF algorithm has the highest RMSE and MAE, the AUKF algorithm has intermediate performance, while the FOMIAUKF algorithm achieves the lowest RMSE and MAE under all test conditions. The test results for light-load cyclic operation (LDCO) and heavy-load transient operation (HDTO) at an ambient temperature of 25℃ are as follows: Under lightly loaded cyclic operation (LDCO) conditions, the EKF algorithm has an RMSE of 0.0157 and a MAE of 0.0145, while the AUKF algorithm has an RMSE of 0.0132 and a MAE of 0.0128. The FOMIAUKF algorithm exhibits significantly lower errors, with an RMSE of 0.0086 and a MAE of 0.0063. Compared to the EKF algorithm, the FOMIAUKF algorithm reduces RMSE by 45.2% and MAE by 56.6%.

[0069] Under heavy-load transient operation (HDTO) conditions, the RMSE of the EKF algorithm is 0.0161 and the MAE is 0.0156, while the RMSE of the FOMIAUKF algorithm is 0.0085 and the MAE is 0.0079. Compared with the EKF algorithm, the FOMIAUKF algorithm reduces the RMSE and MAE by 47.2% and 49.4%, respectively.

[0070] Under standard load continuous operation (SDCO) conditions, the FOMIAUKF algorithm has an RMSE of 0.0083 and a MAE of 0.0066. Compared with the EKF algorithm, the FOMIAUKF algorithm reduces the RMSE and MAE by 46.5% and 51.1%, respectively.

[0071] Furthermore, in light-load cyclic operation (LDCO) conditions at ambient temperatures of -20℃ and 0℃, the FOMIAUKF algorithm reduced SOC RMSE by 45.3% and 41.9%, respectively, and MAE by 46.1% and 34.8%, respectively; in heavy-load transient operation (HDTO) conditions, RMSE reduced by 42.7% and 40.8%, respectively, and MAE by 40.6% and 38.5%, respectively; in standard load continuous operation (SDCO) conditions, the RMSE reduction rates at -20℃ and 0℃ ambient temperatures were 41.9% and 43.5%, respectively, and the MAE reduction rates were 45.3% and 40.4%, respectively.

[0072] Based on the above verification results, the SOC estimation method proposed in this embodiment significantly improves the estimation accuracy, the error distribution is more concentrated, and it exhibits better performance under the dynamic operating conditions of bulldozers.

[0073] This embodiment verifies the effectiveness of bulldozer power battery SOC estimation based on the AGWO-LSTM error correction method under three operating conditions: light load cyclic operation (LDCO), heavy load transient operation (HDTO), and standard load continuous operation (SDCO).

[0074] Compared to the FOMIAUKF algorithm, both the Long Short-Term Memory (LSTM) network and the AGWO-LSTM algorithm improve the SOC estimation accuracy. Notably, the estimation curve using the AGWO-LSTM algorithm shows a better fit to the reference SOC curve than the curve using the LSTM network algorithm, indicating that the proposed AGWO-LSTM error correction model has superior accuracy. The second and third rows of the figure show the evaluation metrics for battery SOC estimation under specific operating conditions.

[0075] To visually compare the RMSE and MAE before and after error correction, Table 1 summarizes the SOC estimation performance of the proposed error correction model under three operating conditions. In all scenarios, the AGWO-LSTM error correction model exhibits the smallest deviation from the reference SOC value. Specifically, under the light load cyclic operation (LDCO), heavy load transient operation (HDTO), and standard load continuous operation (SDCO) conditions, the SOC estimation results based on the AGWO-LSTM error correction model significantly improve the estimation accuracy compared to the SOC estimation results based on the FOMIAUKF algorithm. Specifically, the RMSE is reduced by 52.3%, 34.9%, and 34.9%, respectively, and the MAE is reduced by 65.1%, 70.9%, and 59.1%, respectively.

[0076] Experiments were conducted to verify the SOC estimation results and corresponding errors at ambient temperatures of -20℃ and 0℃. Under all three operating conditions, the estimated SOC values ​​showed high consistency with the reference SOC. The histogram comparing the SOC estimation errors further indicated that the RMSE and MAE of the three algorithms decreased progressively. At ambient temperatures of -20℃ and 0℃, compared with other estimation algorithms, the SOC estimation results based on the AGWO-LSTM error correction model achieved the lowest RMSE and MAE under Light Load Cyclic Operation (LDCO), Heavy Load Transient Operation (HDTO), and Standard Load Continuous Operation (SDCO) modes. Furthermore, compared to the FOMIAUKF algorithm, the AGWO-LSTM-based error correction model exhibits the highest error reduction rate under low-temperature conditions. Specifically, under light-load cyclic operation (LDCO) conditions, RMSE and MAE are reduced by 53.4% ​​and 78%, respectively; under heavy-load transient operation (HDTO) conditions, RMSE and MAE are reduced by 35.6% and 77.6%, respectively; and under standard load continuous operation (SDCO) conditions, RMSE and MAE are reduced by 38.7% and 77%, respectively.

[0077] In summary, after optimization with the AGWO-LSTM error correction model, the accuracy of SOC estimation is significantly improved and the error distribution is more concentrated.

[0078] Table 1

[0079] In summary, the proposed power battery state estimation method in this embodiment overcomes the limitations of traditional methods, fully considering fractional-order dynamic characteristics, thermal effects, and real-time error suppression. It constructs a fractional-order equivalent circuit model (FO-ECM) and a corresponding single-state thermal model (STM), employs an adaptive genetic algorithm (AGA), and combines empirical power battery data to complete model parameter identification and verification. Furthermore, by designing three typical bulldozer operation cycles—light-load cyclic operation (LDCO), standard-load continuous operation (SDCO), and heavy-load transient operation (HDTO)—the effectiveness of the proposed FOMIAUKF algorithm estimation algorithm and the AGWO-LSTM error correction model is verified. Experimental results show that the RMSE and MAE of the proposed FOMIAUKF algorithm for SOC estimation are 0.93% and 0.91%, respectively. After compensation by the AGWO-LSTM error correction model, these errors are reduced to 0.58% and 0.27%, respectively, significantly improving the SOC estimation accuracy.

[0080] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for estimating the state of charge of a power battery, characterized in that, include: Step S1: Input the SOC correction value at time t, the predicted value of the battery temperature at time t, and the measured value of the current at time t into the second-order RC equivalent circuit model to obtain the predicted values ​​of the open-circuit voltage and terminal voltage at time t. Step S2: Substitute the predicted values ​​of battery temperature, terminal voltage, and open-circuit voltage at time t, and the measured value of current at time t, into the heat generation equation to obtain the predicted value of heat generation at time t. Step S3: Substitute the predicted heat generation at time t and the measured battery temperature at time t into the single-state thermal model to obtain the predicted battery temperature at time t+1. Step S4: Substitute the measured values ​​of the terminal voltage and current at time t into the first prediction model to obtain the predicted SOC value at time t+1; Step S5: Calculate the difference between the measured and predicted values ​​of the terminal voltage at the same time as the observation error, and input the observation error, battery parameters, and predicted SOC value from time tk to time t+1, as well as the SOC correction value from time tk to time t, into the time series prediction model to obtain the SOC correction value at time t+1.

2. The method for estimating the state of charge of a power battery according to claim 1, characterized in that, The second-order RC equivalent circuit model is as follows: ; ; in: This is the predicted value of the open-circuit voltage. This is the open-circuit voltage prediction function. This is the SOC correction value. This is a predicted value for battery temperature. This is the predicted value of the terminal voltage. The internal resistance of the battery is in ohms. This is the measured value of the current. For the first polarization resistor, This is the second polarization resistor.

3. The method for estimating the state of charge of a power battery according to claim 1, characterized in that, The heat production equation is as follows: ; in: This is a predicted value for heat production. This is the predicted value of the open-circuit voltage. This is the predicted value of the terminal voltage. This is the measured value of the current. This is a predicted value for battery temperature.

4. The method for estimating the state of charge of a power battery according to claim 1, characterized in that, The singlet thermal model is as follows: ; in: For pyrogenic rate, This is the measured value of the battery temperature. This is the measured value of the ambient temperature. This is the equivalent thermal resistance.

5. The method for estimating the state of charge of a power battery according to claim 4, characterized in that, The model parameters of the singlet thermal model include heat generation rate and equivalent thermal resistance; The model parameters of the singlet thermal model are obtained through optimization using an adaptive genetic algorithm, and the fitness function of the optimization process is... f 3 is: ; in: This is the sequence of measured battery temperatures for the kth batch. Let N be the predicted sequence of battery temperatures for the k-th batch, and N be the number of training batches.

6. The method for estimating the state of charge of a power battery according to claim 5, characterized in that, The crossover probability and mutation probability of the adaptive genetic algorithm are: ; ; in: For crossover probability, This is the adjustment coefficient for the first crossover probability. This represents the optimal fitness value for the population. A larger fitness value in the crossover operation. The average fitness value of the population. This is the adjustment coefficient for the second crossover probability. This is the first mutation probability adjustment coefficient. The mutation probability, This represents the fitness value during the mutation operation. This is the second variation probability adjustment coefficient.

7. The method for estimating the state of charge of a power battery according to claim 1, characterized in that, The first prediction model is constructed based on the fractional-order multi-innovation adaptive unscented Kalman filter algorithm.

8. The method for estimating the state of charge of a power battery according to claim 1, characterized in that, The time series prediction model is constructed using a long short-term memory network, and its hyperparameters include the initial learning rate, the number of hidden layer nodes, and the regularization coefficient of the long short-term memory network. The battery parameters include the measured values ​​of current, terminal voltage, and battery temperature. The hyperparameters of the time series prediction model are optimized using an improved gray wolf optimization algorithm, wherein the convergence factor is adjusted by the number of iterations in the improved gray wolf optimization algorithm. ; in: The convergence factor is A random number uniformly distributed within the interval [0,1]. The maximum number of iterations, l This represents the current iteration number.

9. A power battery state-of-charge estimation device, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-8.

10. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method as described in any one of claims 1-8.

Images