Energy storage battery pack equalization control method

Abstract

The invention discloses an energy storage battery pack equalization control method, which belongs to the field of energy storage battery packs, and comprises the following steps: S1, constructing a multi-physics field-aging coupling model; s2, based on a multi-physical field-aging coupling model, space-time-frequency domain joint state estimation and unbalanced pattern recognition are realized; s3, formulating optimal equalization topology and control parameters according to space-time-frequency domain joint state estimation and an unbalanced mode recognition result; s4, establishing an electric-thermal-fluid multi-physics field coupling control model, and optimizing equalization current regulation and control, liquid cooling flow velocity and cooling fan rotating speed based on the optimal equalization topology and control parameters; and S5, optimizing the full life cycle of the digital twin drive. By adopting the equalization control method for the energy storage battery pack, the efficient coordination of the energy storage battery pack from dynamic modeling, accurate state estimation and intelligent equalization decision to full life cycle management is realized, and the equalization efficiency and the system reliability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage battery packs, and particularly to an energy storage battery pack equalization control method. Background Art

[0002] With the exponential growth of the market scales of electric vehicles (EVs) and renewable energy storage systems (ESSs), energy storage units, which are the core components of the above applications, have also been more widely used. The safety, durability, and energy efficiency of energy storage units directly determine the upper limit of system performance. However, the internal imbalance in battery packs caused by manufacturing process deviations, working environment differences, and usage condition changes has become a key bottleneck restricting their large-scale application.

[0003] To solve the above problems, the prior art has conducted research in the following aspects: 1. Battery modeling and state estimation technology: Existing battery management systems (BMSs) mainly use equivalent circuit models (ECMs) to describe the dynamic behavior of batteries. Such models simulate the voltage response characteristics of batteries through RC networks and have the advantages of low computational complexity and good real-time performance. However, they ignore the microstructure evolution of electrode materials and the electrolyte diffusion process, resulting in significant limitations in describing battery aging mechanisms (such as SEI film growth and active material loss). In terms of state estimation, extended Kalman filtering (EKF) and its derivative algorithms (such as UKF, PF) are the current mainstream methods. However, in the face of strong disturbance scenarios such as sudden temperature changes and large current shocks, due to noise model mismatch and linearization error accumulation, the estimation accuracy drops significantly.

[0004] 2. Battery equalization control technology: Traditional equalization control strategies can be divided into two categories: passive equalization and active equalization. Passive equalization consumes the energy of high-voltage single cells through resistor discharge. It has a simple structure but low efficiency (usually less than 80%); active equalization uses energy storage components such as capacitors, inductors, or transformers to achieve energy transfer, and the efficiency can reach more than 90%. However, it has problems such as complex topology and high cost. Moreover, existing active equalization schemes mainly control based on voltage or state of charge deviation, ignoring the multi-physical field coupling effects such as temperature field distribution and aging rate differences, resulting in a significant reduction in the equalization effect as the battery ages.

[0005] It can be seen that the existing technologies have the following defects: 1. Insufficient multi-physical field coupling modeling: The existing battery models fail to effectively integrate the coupling effects of multi-physical fields such as electrochemistry, heat conduction, and fluid dynamics, resulting in a sharp decline in model accuracy under complex working conditions (such as fast charging and low-temperature environments), and being unable to accurately predict the internal state evolution of the battery. 2. Poor robustness of state estimation: Traditional state estimation algorithms rely on accurate battery models and noise statistical characteristics, and are vulnerable to sensor errors, model uncertainties, and external disturbances in practical applications, leading to divergent estimation results and affecting battery safety and life. 3. Insufficient adaptability of the equalization control strategy: Existing equalization strategies are mostly designed based on static parameters and cannot adapt to the parameter drift and uneven pattern evolution during battery aging, resulting in a significant reduction in equalization efficiency as the number of cycles increases. 4. Lack of system collaborative control: There is a lack of a collaborative optimization mechanism among the power management, thermal management, and equalization control subsystems, resulting in high system energy consumption and slow response speed, and it is difficult to meet the real-time control requirements under high-dynamic working conditions. 5. Weak full-life cycle management ability: Existing BMS mainly focuses on the management of the battery usage stage, lacks a full-life cycle data closed-loop from manufacturing, use to retirement, and cannot achieve accurate traceability of battery states and maximize the value of cascade utilization. Summary of the Invention

[0006] The purpose of the present invention is to provide an equalization control method for an energy storage battery pack to solve the above technical problems.

[0007] To achieve the above purpose, the present invention provides an equalization control method for an energy storage battery pack, including the following steps: S1. Multi-physical field-aging coupling modeling and parameter identification: Based on the pseudo-two-dimensional electrochemistry theory, heat conduction theory, and data-driven algorithm, a multi-physical field-aging coupling model is constructed to depict the lithium-ion diffusion kinetics, interface impedance evolution law, and thermal-electric interaction characteristics; S2. Space-time-frequency domain joint state estimation: Based on the multi-physical field-aging coupling model, a composite estimation framework integrating the extended Kalman filter, wavelet transform, and three-dimensional convolutional neural network is used. Through the collaborative processing of time-domain dynamic tracking, frequency-domain feature extraction, and spatial-domain pattern recognition, space-time-frequency domain joint state estimation and uneven pattern recognition are realized; S3. Model predictive-reinforcement learning hybrid equalization decision-making: A hybrid decision-making architecture of model predictive control and deep reinforcement learning is constructed. Using the constructed hybrid decision-making architecture, according to the space-time-frequency domain joint state estimation and uneven pattern recognition results, the optimal equalization topology and control parameters are formulated; S4. Electro-thermal-fluid collaborative control: An electro-thermal-fluid multi-physical field coupling control model is established, and based on the optimal equalization topology and control parameters, the electro-thermal-fluid multi-physical field coupling control model is used to collaboratively optimize the equalization current regulation and the liquid cooling flow rate and the cooling fan speed; S5. Digital Twin-Driven Full-Life Cycle Optimization: Construct a Digital Twin covering battery manufacturing and sorting, the usage stage management involved in Steps S1 - S4, and retired cascade utilization, and achieve dynamic calibration of the model through a physical information fusion algorithm to generate cross-stage equilibrium strategies.

[0008] Therefore, by adopting the above-mentioned method for balancing control of an energy storage battery pack, the present invention has the following beneficial effects: 1. Improvement in the accuracy of multi-physical field modeling: The P2D (pseudo-two-dimensional) electrochemistry-thermal coupling model constructed based on the pseudo-two-dimensional electrochemistry theory and the heat conduction equation, combined with the aging prediction of a gated recurrent unit neural network, can dynamically depict the characteristics of lithium-ion diffusion, interfacial impedance evolution, and thermal-electric interaction. The modeling error is reduced by more than 40% compared with the traditional equivalent circuit model; 2. Enhancement of the robustness of state estimation: Combining the extended Kalman filter, wavelet transform, and three-dimensional convolutional neural network, and dynamically adjusting the noise covariance through the high-frequency energy ratio. Under conditions such as sudden temperature changes (±20°C) and large current impacts (2C), the state of charge estimation error is stabilized within ±1.5%, which is a 50% improvement compared with the traditional EKF algorithm; 3. Adaptive optimization of the balancing strategy: The hybrid architecture of model predictive control and deep reinforcement learning, with the balancing time, energy loss, and temperature fluctuation as the optimization objectives, dynamically adjusts the weight coefficients and prediction horizons, enabling the balancing efficiency to reach more than 92%, a 30% improvement compared with the traditional passive balancing, and the efficiency decay rate is reduced by 60% as the battery ages (after 500 cycles); 4. Electro-thermal-fluid collaborative temperature control: Through the coupled model of computational fluid dynamics and heat transfer, combined with fuzzy logic control of the coolant flow rate and fan speed, the maximum temperature difference of the battery pack is controlled within ±2°C, a 50% reduction compared with the traditional independent thermal management scheme, and the energy consumption is reduced by 25% at the same time; 5. Closed-loop management of the full life cycle: The Digital Twin covers manufacturing and sorting, usage management, and retired cascade utilization. The sorting matching degree is optimized to 95% through the genetic algorithm, the capacity prediction error at retirement is <3%, and the full life cycle cost is reduced by 18% compared with the traditional scheme.

[0009] Next, through the drawings and embodiments, the technical solutions of the present invention will be further described in detail. Description of the Drawings

[0010] Figure 1 It is a flowchart of a method for balancing control of an energy storage battery pack of the present invention. Detailed Embodiments

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer and more understandable, the following further elaborates on the embodiments of the present invention in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not used to limit the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of this application. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout.

[0012] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0013] The following elaborates in detail on the embodiments of the present invention in conjunction with the accompanying drawings.

[0014] As Figure 1 shown, a method for balancing control of an energy storage battery pack includes the following steps: S1. Multi-physics - aging coupling modeling and parameter identification: Based on the pseudo-two-dimensional electrochemical theory, heat conduction theory, and data-driven algorithms, a multi-physics - aging coupling model is constructed to characterize the lithium-ion diffusion kinetics, interface impedance evolution law, and thermal - electrical interaction characteristics. Step S1 specifically includes the following steps: S11. Based on the pseudo-two-dimensional electrochemical theory, integrating the Fourier heat conduction equation, a P2D electrochemical - thermal coupling model is constructed. S111. Electrochemical modeling of porous electrodes: A two-dimensional radially symmetric model of electrode particles is constructed, and based on the porous electrode theory, the geometric structure parameters and physical property parameters of the energy storage battery pack are defined. The geometric structure parameters include the electrode particle radius , electrode porosity and electrode thickness ; the material parameters include the solid-phase diffusion coefficient , electrolyte conductivity and initial lithium-ion concentration ; S112. According to Fick's second law, diffusion equations describing the diffusion of lithium ions in the solid phase and electrolyte inside the electrode particles of the energy storage battery pack are established: ; ; In the formula, represents the solid-phase lithium-ion concentration; represents time; represents the gradient operator; represents the electrolyte phase potential; S113. Use the Butler-Volmer equation as a description tool for the electro-chemical reaction kinetics at the electrode-electrolyte interface, and combine the solution results of the two-dimensional radial symmetry model to calculate the interfacial reaction current density and the overpotential at the electrode interface : ; In the formula, represents the exchange current density; and represent the anodic and cathodic transfer coefficients respectively; represents the Faraday constant; represents the gas constant; represents the absolute temperature; represents the overpotential at the electrode interface; S114. Calculate the total heat generation power of the energy storage battery pack : ; Among them, ; ; ; In the formula, , and represent the ohmic heat generation power, polarization heat generation, and side reaction heat generation of the energy storage battery pack respectively; represents the charge and discharge current; represents the ohmic internal resistance of the energy storage battery pack; represents the overpotential of the side reaction; S115. Considering the total heat generation power of the energy storage battery pack, establish a heat conduction equation describing the temperature change of the battery according to Fourier's law of heat conduction and the principle of energy conservation: ; In the formula, represents the mass of the energy storage battery pack; represents the specific heat capacity; represents the temperature of the energy storage battery pack; represents the environmental reference temperature; represents the equivalent thermal resistance; represents the heat dissipation surface area; In step S115, the energy storage battery pack is disassembled into three-dimensional grid units according to geometric structures and physical characteristics, and the number of three-dimensional grid units is set. As follows: ; In the formula, , and respectively represent the length, width, and height of the energy storage battery pack; , and represent the side lengths of the three-dimensional grid units in the x-axis, y-axis, and z-axis directions.

[0015] S12. Combining the cycle life data of the battery during actual operation, using the time series data processing ability of the gated recurrent unit neural network, construct a data-driven battery aging prediction model; S121. Aging data collection and preprocessing to obtain a standardized aging data set: Collect the full-life cycle cycle data of the energy storage battery pack under various charge-discharge conditions and various environmental temperatures, where the full-life cycle cycle data includes the number of cycles , depth of charge and discharge , operating temperature , terminal voltage , charge-discharge current , capacity and test internal resistance ; S122. Build a gated recurrent unit neural network model and use the set feature vector and label vector to train the gated recurrent unit neural network model; among them, represents the capacity attenuation rate, represents the internal resistance growth rate; represents the environmental temperature at the location of the energy storage battery pack; S13. Couple the battery aging prediction model and the P2D electrochemistry-thermal coupling model: Feed back the capacity attenuation rate and internal resistance growth rate predicted by the trained battery aging prediction model to the P2D electrochemistry-thermal coupling model, and dynamically update the electrochemical parameters in the P2D electrochemistry-thermal coupling model.

[0016] S2. Spatiotemporal-frequency domain joint state estimation: Based on the multi-physics field-aging coupling model, fuse the composite estimation framework of the extended Kalman filter, wavelet transform, and three-dimensional convolutional neural network, and through the coordinated processing of time-domain dynamic tracking, frequency-domain feature extraction, and spatial-domain pattern recognition, realize spatiotemporal-frequency domain joint state estimation and unbalanced pattern recognition; Step S2 specifically includes the following steps: S21. Construct an extended Kalman filter state space model; S211. Define the state vector , output vector , where represents the state of charge, represents the internal polarization voltage of the energy storage battery pack; S212. Based on the battery equivalent circuit model, construct the state transition equation and the observation equation: ; ; In the formula, and respectively represent the state vectors of the th iteration and the th iteration; represents the state transition function; represents the observation function; represents the output vector of the th iteration; represents the charge and discharge current of the th iteration; represents the operating temperature of the th iteration; and respectively represent the process noise and the measurement noise; represents the charge and discharge current of the th iteration; S213. Design and conduct constant current charge and discharge and pulse charge and discharge experiments, and collect the voltage, current, and temperature data of the energy storage battery pack under different working conditions. Use the least squares method to determine the initial state transition matrix and the observation matrix , and initialize the process noise covariance matrix and the measurement noise covariance matrix ; In step S213, the initial state transition matrix and the observation matrix , and the initial process noise covariance matrix and the measurement noise covariance matrix are calculated as follows: ; ; ; ; In the formula, represents the sampling time interval; Represents the state of charge time constant; Represents the polarization voltage time constant; Represents the slope of the open circuit voltage - state of charge curve; Represents the state of charge estimation error variance; Represents the voltage measurement error variance; Represents the temperature measurement error variance.

[0017] S22, Wavelet transform signal preprocessing; S221. Perform discrete wavelet transform on the collected terminal voltage signal of the energy storage battery pack to decompose the signal into different frequency sub - bands, obtaining low - frequency components and high - frequency components; S222. Based on the decomposed high - frequency components, calculate the high - frequency energy ratio : ; In the formula, and respectively represent the total energy of the high - frequency components and the total energy of the signal; Represents the th high - frequency component; Represents the decomposition layer number of the high - frequency components; S223. Based on the high - frequency energy ratio combined with the dynamic weight coefficient adjust the measurement noise covariance matrix of the Kalman filter state - space model: ; In the formula, Represents the adjusted measurement noise covariance matrix; Among them, ; In the formula, and respectively represent the maximum weight coefficient and the minimum weight coefficient; Represents the mean value within the sliding window; Represents the high - frequency energy ratio threshold; Represents the historical maximum high - frequency energy ratio; S24. Perform time update and measurement iterative update on the Kalman filter state - space model: Time update: ; ; Measurement update: ; ; ; In the formula, represents the prior estimate of the state vector at the -th iteration, based on all the observation information at the -th iteration; represents the posterior estimate of the state vector at the -th iteration and all the observation information before it; ; represents the posterior estimate of the state vector at the -th iteration and all the observation information before it; ; represents the charge and discharge current at the -th iteration; represents the working temperature at the -th iteration; represents the prior estimate error covariance matrix at the -th iteration, based on the prior estimate at the -th iteration; represents the error covariance matrix of the posterior estimate at the -th iteration; represents the state transition matrix at the -th iteration; represents the covariance matrix of the process noise; represents the updated Kalman gain; represents the updated observation matrix; represents the battery measurement value at the -th iteration; represents the identity matrix; represents the transpose of the state transition matrix at the -th iteration; S25. Collect the voltages of each single cell of the energy storage battery pack and the working temperature, and after normalization, organize them into a tensor format to obtain the preprocessed tensor data ; where represents the single cell number, represents the time step; represents the batch size; represents the number of single cells; represents the time step; represents two channels of voltage and temperature; represents the set of real numbers; S26. Construct a three-dimensional convolutional neural network, and use the labeled unbalanced data of the battery pack to train the network and tensor data to train the three-dimensional convolutional neural network; The three-dimensional convolutional neural network described in step S26 includes multiple 3D convolutional layers, 3D max pooling layers, and fully connected layers. Among them, the 3D convolutional layer: uses a convolutional kernel of size 3 3 ×3 to extract spatio-temporal features; the pooling size of the 3D max pooling layer is 2 2 ×2 to reduce the feature dimension; the fully connected layer: is used for classification output.

[0018] S27. Convert the preprocessed real-time acquired energy storage battery pack data into tensor data, input it into the trained three-dimensional convolutional neural network, use the three-dimensional convolutional neural network to identify the unbalanced pattern of the battery pack, and output the unbalanced feature vector .

[0019] S3. Model prediction - reinforcement learning hybrid balancing decision: Construct a hybrid decision-making architecture of model predictive control and deep reinforcement learning. Use the constructed hybrid decision-making architecture to formulate the optimal balancing topology and control parameters according to the spatio-temporal - frequency domain joint state estimation and the unbalanced pattern recognition result; Step S3 specifically includes the following steps: S31. Based on the model predictive control theory, with the equalization time, energy loss, temperature fluctuation, and unbalanced degree of the energy storage battery pack as the optimization objectives, combined with the multi-physics - aging coupling model constructed in S1 and the spatio-temporal - frequency domain joint state estimated in S2, construct an optimization problem data model, and perform rolling optimization within a finite time domain to generate the optimal control quantity; S311. With the equalization time of the energy storage battery pack , energy loss , temperature fluctuation , and unbalanced degree as the optimization objectives, construct the following objective function : ; In the formula, , , , and all represent weight coefficients, and ; represents the reference equalization time; represents the safe temperature threshold; Denote the two-norm of the imbalance feature vector; S312. Based on the multi-physics-aging coupling model constructed in S1, establish an MPC prediction model by combining the battery state estimated in S2, and predict the voltages, capacities, and temperatures of each cell in the energy storage battery pack within a future finite time domain through the state space equation: ; ; ; wherein, denotes the state vector of the th iteration prediction; denotes the state vector of the th iteration prediction; and respectively denote the equilibrium control inputs of the th iteration and the th iteration; denotes the working temperature at the th iteration; denotes the output vector of the th iteration prediction; S313. Considering battery safety and hardware limitations, add constraint conditions: State constraint: ; Control constraint: ; Safety constraint: ; wherein, and respectively denote the allowable minimum and maximum state of charge; denotes the state of charge of the th cell battery; and respectively denote the lower and upper limits of the allowable working voltage of the cell battery; denotes the terminal voltage of the th cell battery; and respectively denote the lower and upper limits of the equilibrium control input; denotes the equilibrium control input amount for the th cell battery; S314. Use the sequential quadratic programming or interior point method to solve the mathematical model of the optimization problem to obtain the optimal control quantity in the current cycle: ; S32. Use the deep reinforcement learning algorithm to implement the optimal control quantity in the current cycleInteract with the energy storage battery pack environment to dynamically adjust the weight coefficients and prediction horizons of the MPC prediction model; S321. Define the state space of deep reinforcement learning , action space and reward function : State space: ; Action space: ; Reward function: ; In the formula, , , and respectively represent the adjustment amounts of the weight coefficients , , and ; represents the adjustment amount of the finite time domain ; , , and all represent the weight coefficients of the reward function, and ; represents the benchmark objective function value; S322. Use the operation data of the MPC prediction model to train the deep reinforcement learning network, use the trained deep reinforcement learning network to update the MPC prediction model parameters, and feedback the MPC prediction model parameters to step S311 to form a closed-loop optimization. The MPC prediction model parameter update formula is as follows: ; In the formula, represents the weight coefficient of the updated objective function, and ; represents the weight coefficient of the original objective function; represents the adjustment amount of the weight coefficient of the objective function; represents the updated finite time domain ; S33. Combine the rolling optimization of the MPC prediction model and the parameter adaptability of deep reinforcement learning to generate and execute a dynamic balancing strategy, and realize energy transfer or dissipation through a hardware circuit to complete battery pack balancing; S331. According to the optimal control quantity output by the MPC prediction model, customize the following balancing strategy: When is positive, use a bidirectional DC-DC converter for energy transfer; When When it is negative, resistive dissipation is performed; S332. Collect the equalization current, voltage, and operating temperature data, feedback it to step S2 to update the state parameters, and input the equalization effect into the deep reinforcement learning network to update the equalization strategy.

[0020] S4. Electro-thermal-fluid collaborative control: Establish an electro-thermal-fluid multi-physical field coupling control model, and based on the optimal equalization topology and control parameters, use the electro-thermal-fluid multi-physical field coupling control model to collaboratively optimize the equalization current regulation and the liquid cooling flow rate and the rotation speed of the cooling fan; Step S4 specifically includes the following steps: S41. Based on the computational fluid dynamics and heat transfer theories, construct a heat conduction-convection coupling model for the battery pack, and analyze the interaction between the coolant flow characteristics and the battery heat generation distribution; S411. Construct a fluid mechanics model: Establish a three-dimensional geometric model of the battery pack flow channel, define the cross-sectional shape of the flow channel, the pipe diameter and the fluid flow velocity , and use the N-S equation to describe the turbulent flow of the coolant, and solve the flow field distribution by the finite element method: ; ; In the formula, represents the fluid density; represents the dynamic viscosity; represents the acceleration due to gravity; S412. Based on the total heat generation power in step S114, establish a heat conduction equation: ; In the formula, , and respectively represent the material density, specific heat capacity, and thermal conductivity of the energy storage battery pack; S42. Adopt the conjugate heat transfer method to calculate the convective heat transfer coefficient between the battery and the coolant, and couple and solve the temperature field: ; Among them, ; And set the following convective heat transfer boundary conditions: ; In the formula, represents the fluid thermal conductivity; represents the Nusselt number; represents the Reynolds number; represents the Prandtl number; represents the pump speed of the cooling fan; Represents the fluid temperature; S43. Based on model predictive control and fuzzy logic, construct an electro-thermal-fluid multi-objective collaborative control algorithm to achieve an optimal balance of battery temperature uniformity, energy consumption, and fluid flow rate; S431. Define a multi-objective function that includes temperature deviation, fluid energy consumption, and the impact of electrical performance : ; Wherein, , and both represent the safety threshold weights, and ; Represents the maximum temperature difference of the energy storage battery pack; Represents the pump power consumption of the cooling fan; Represents the reference power consumption; Represents the state of charge deviation caused by the temperature difference; Represents the rated state of charge; S432. Construct fuzzy control rules: Define the input variables as the temperature deviation and the temperature change rate , and set up a fuzzy rule base based on the input change amount, and then use the Mamdani method for fuzzy inference and defuzzification to obtain the fluid flow rate adjustment amount and the pump speed adjustment amount ; S44. Based on the heat conduction-convection coupling model in step S41 and the fuzzy control rules in step S43, drive the fluid system and the electrical equalization system to work together, and optimize the control parameters through real-time feedback.

[0021] In this embodiment, the following feedback optimization strategy is set: When , increase and start active equalization; When , decrease and switch to passive equalization.

[0022] Wherein, Represents the temperature deviation threshold.

[0023] S5. Digital twin-driven full-life cycle optimization: Construct a digital twin that covers battery manufacturing sorting, the usage stage management involved in steps S1 - S4, and retired echelon utilization, and achieve dynamic calibration of the model through a physical information fusion algorithm to generate a cross-stage equalization strategy.

[0024] Step S5 specifically includes the following steps: S51. Integrate battery manufacturing and sorting data, multi - physical field characteristics during the usage stage, and the law of performance degradation during retirement to construct a digital twin model covering the entire life cycle of the battery; S511. Modeling during the manufacturing and sorting stage: Collect single - cell battery parameters , denote the open - circuit voltage, and based on the Monte Carlo method, simulate manufacturing deviations to construct a parameter distribution model of single - cell batteries, and optimize the sorting strategy through the genetic algorithm: ; In the formula, denote the sorting matching degree; and respectively denote the standard deviation and the mean value of the single - cell battery capacity; S512. Multi - physical field modeling during the usage stage: Integrate the P2D electrochemical - thermal coupling model described in step S1, the state estimation process described in step S2, the equalization control process described in step S3, and the hydrodynamic equation in step S4 to establish a digital twin during the usage stage; S513. Modeling for retired battery cascade utilization: Based on the aging prediction model described in step S1.2, predict the capacity at retirement : ; In the formula, denote the rated capacity; denote the attenuation coefficient; S52. Through the multi - source data fusion algorithm, interact the real - time monitoring data with the digital twin model, and use the federated learning algorithm to achieve dynamic calibration of the full - life - cycle model; S53. Cross - stage mapping: Establish the mapping relationship between manufacturing parameters and performance indicators during the usage stage : ; In the formula, denote the mapping function; denote the degree of aging;[[ID=5). Meanwhile, align the aging trajectories of different energy - storage battery packs through the dynamic time warping algorithm; S54. Based on the full - life - cycle digital twin model, define a multi - objective function including manufacturing cost, usage efficiency, and retirement value : ; In the formula, , and all denote weight coefficients, and ; and respectively represent the manufacturing cost and the benchmark cost; and respectively represent the residual value at retirement and the benchmark value; S55. Based on the multi-objective function , use reinforcement learning to optimize the digital twin model; S55. Map the cross-stage strategy to the manufacturing sorting rules, the control parameters in the usage stage, and the retirement decision logic, execute them respectively, and collect the execution data to iteratively update the cross-stage strategy parameters through the gradient descent method.

[0025] Simulation experiment

[0026] Simulation tools: MATLAB / Simulink + PyTorch (for the three-dimensional convolutional neural network part). Battery pack model: a lithium iron phosphate battery pack with 20 series and 10 parallel connections (single-cell capacity 20 Ah, rated voltage 3.2 V). Operating conditions: Fast charging condition: charge at 1.5C to 80% SoC, and then equalize at 0.5C; Low-temperature condition: under a -10°C environment, perform 1C charge and discharge cycles; Hybrid condition: NEDC cycle (including acceleration, constant speed, and deceleration stages) + random temperature perturbation (5°C to 40°C).

[0027] Comparison schemes: Comparison 1: Traditional passive equalization (resistive dissipation); Comparison 2: Active equalization based on the ECM model (inductive transfer); Comparison 3: Independent power management + thermal management scheme.

[0028] Table 1 Experimental results

[0029] As can be seen from Table 1, 1. Equalization efficiency and speed: Through model prediction and reinforcement learning, the present application dynamically optimizes the equalization topology. Under the fast charging condition, the equalization time is shortened by 56.8%, and the energy loss is reduced by 63.6%, proving that the hybrid decision-making architecture has better adaptability to complex working conditions than traditional single strategies. 2. Temperature uniformity: By adopting electro-thermal-fluid collaborative control, the present application reduces the temperature difference by 67.9% under the low-temperature condition and reduces the SoC deviation by 50% - 68% under the hybrid condition, indicating that the multi-physics field coupling model effectively solves the problem of aggravated imbalance caused by uneven temperature. For traditional schemes, due to the lack of a collaborative mechanism, the temperature difference and deviation are significantly higher. 3. Aging robustness: After 500 cycles, the equalization efficiency of this scheme still remains at 89.5%, which is 14.4% - 24.2% higher than that of traditional schemes, attributed to the parameter dynamic calibration driven by the digital twin. For traditional schemes, due to the lack of consideration of the aging coupling effect, the efficiency decay is obvious. 4. Advantages in the whole life cycle: By combining manufacturing sorting optimization and retirement prediction, this scheme reduces the total life cycle cost of the battery pack by 18% in the simulation, verifying the improvement of the whole life cycle management on the system economy. For the comparison schemes, due to the lack of cross-stage strategy mapping, the cost optimization is limited.

[0030] In summary, this application is significantly superior to traditional methods in terms of balanced efficiency, temperature control, anti-aging ability, and full-life cycle management, thus demonstrating the effectiveness of this application.

[0031] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for balancing control of an energy storage battery pack, characterized in that: It includes the following steps: S1. Multi-physical field-aging coupling modeling and parameter identification: Based on the pseudo-two-dimensional electrochemical theory, heat conduction theory, and data-driven algorithms, a multi-physical field-aging coupling model is constructed to characterize the lithium-ion diffusion kinetics, interface impedance evolution law, and thermoelectric interaction characteristics; S2. Spatiotemporal-frequency domain joint state estimation: Based on the multi-physical field-aging coupling model, a composite estimation framework integrating the extended Kalman filter, wavelet transform, and three-dimensional convolutional neural network is used. Through the collaborative processing of time-domain dynamic tracking, frequency-domain feature extraction, and spatial-domain pattern recognition, spatiotemporal-frequency domain joint state estimation and unbalanced pattern recognition are realized; S3. Model prediction-reinforcement learning hybrid balancing decision-making: A hybrid decision-making architecture of model predictive control and deep reinforcement learning is constructed. Using the constructed hybrid decision-making architecture, according to the spatiotemporal-frequency domain joint state estimation and unbalanced pattern recognition results, the optimal balancing topology and control parameters are formulated; S4. Electro-thermal-fluid collaborative control: An electro-thermal-fluid multi-physical field coupling control model is established, and based on the optimal balancing topology and control parameters, the electro-thermal-fluid multi-physical field coupling control model is used to collaboratively optimize the balancing current regulation and the liquid cooling flow rate and the rotation speed of the cooling fan; S5. Digital twin-driven full-life cycle optimization: A digital twin covering battery manufacturing sorting, the usage stage management involved in steps S1-S4, and retired cascade utilization is constructed, and through the physical information fusion algorithm, the model is dynamically calibrated to generate cross-stage balancing strategies.

2. The method for balancing control of an energy storage battery pack according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11. Based on the pseudo-two-dimensional electrochemical theory, the Fourier heat conduction equation is integrated to construct a P2D electrochemical-thermal coupling model; S111. Electrochemical Modeling of Porous Electrodes: Construct a two-dimensional radially symmetric model of electrode particles, and define the geometric structure parameters and physical property parameters of the energy storage battery pack based on the porous electrode theory. The geometric structure parameters include the electrode particle radius , the electrode porosity , and the electrode thickness ; the material parameters include the solid-phase diffusion coefficient , the electrolyte conductivity , and the initial lithium-ion concentration . S112. According to Fick's second law, a diffusion equation describing the lithium ions in the solid phase inside the electrode particles and the electrolyte of the energy storage battery pack is established: ； ； In the formula, represents the solid-phase lithium-ion concentration; represents time; represents the gradient operator; represents the electrolytic liquid-phase potential; S113. Use the Butler-Volmer equation as a description tool for the electro-chemical reaction kinetics at the electrode-electrolyte interface, and calculate the interfacial reaction current density by combining the solution results of the two-dimensional radial symmetry model. and the overpotential at the electrode interface : ； In the formula, represents the exchange current density; and represent the anodic and cathodic transfer coefficients respectively; represents the Faraday constant; represents the gas constant; represents the absolute temperature; represents the overpotential at the electrode interface; S114. Calculate the total heat generation power of the energy storage battery pack : ； Among them, ； ； ； In the formula, , and respectively represent the ohmic heat generation power, polarization heat generation, and side reaction heat generation of the energy storage battery pack; represents the charge and discharge current; represents the ohmic internal resistance of the energy storage battery pack; represents the side reaction overpotential; S115. Considering the total heat generation power of the energy storage battery pack, according to Fourier's heat conduction law and the principle of energy conservation, a heat conduction equation describing the battery temperature change is established: ； In the formula, represents the mass of the energy storage battery pack; represents the specific heat capacity; represents the temperature of the energy storage battery pack; represents the ambient reference temperature; represents the equivalent thermal resistance; represents the heat dissipation surface area; S12. Combining the cycle life data of the battery during actual operation, using the time-series data processing ability of the gated recurrent unit neural network, a data-driven battery aging prediction model is constructed; S121. Aging data collection and preprocessing to obtain a standardized aging data set: Collect the full-life cycle cycle data of the energy storage battery pack under various charge and discharge conditions and various ambient temperatures, where the full-life cycle cycle data includes the number of cycles , depth of charge and discharge , operating temperature , terminal voltage , charge and discharge current , capacity and test internal resistance ; S122. Build a gated recurrent unit neural network model and use the set feature vector and the label vector to train the gated recurrent unit neural network model; where represents the capacity attenuation rate represents the internal resistance growth rate; represents the ambient temperature at the location of the energy storage battery pack S13. Coupling the battery aging prediction model and the P2D electrochemistry-thermal coupling model: The capacity attenuation rate and the internal resistance growth rate predicted by the trained battery aging prediction model are fed back to the P2D electrochemistry-thermal coupling model to dynamically update the electrochemical parameters in the P2D electrochemistry-thermal coupling model.

3. A method for balancing control of an energy storage battery pack according to claim 2, characterized in that: In step S115, the energy storage battery pack is disassembled into three-dimensional grid units according to geometric structures and physical characteristics, and the number of three-dimensional grid units is set as follows: ； Wherein, , and respectively represent the length, width and height of the energy storage battery pack; , and represent the side lengths of the three-dimensional grid unit in the x-axis, y-axis, and z-axis directions.

4. A method for balancing control of an energy storage battery pack according to claim 3, characterized in that: Step S2 specifically includes the following steps: S21. Construct an extended Kalman filter state space model; S211. Define a state vector according to the electrochemical characteristics of the energy storage battery pack , and an output vector , where represents the state of charge, and represents the internal polarization voltage of the energy storage battery pack; S212. Based on the battery equivalent circuit model, a state transition equation and an observation equation are constructed: ； ； In the formula, and respectively represent the state vectors of the -th iteration and the -th iteration; represents the state transition function; represents the observation function; represents the output vector of the -th iteration; represents the charge-discharge current of the -th iteration; represents the operating temperature at the -th iteration; and respectively represent the process noise and the measurement noise; represents the charge-discharge current of the -th iteration; S213. Design and conduct constant current charge and discharge and pulse charge and discharge experiments, collect the voltage, current, and temperature data of the energy storage battery pack under different working conditions, and use the least squares method to determine the initialized state transition matrix and the observation matrix , and initialize the process noise covariance matrix and the measurement noise covariance matrix ; S22. Wavelet transform signal preprocessing; S221. Perform discrete wavelet transform on the collected terminal voltage signal of the energy storage battery pack to decompose the signal into different frequency sub-bands and obtain the low-frequency component and the high-frequency component; S222. Calculate the high-frequency energy ratio based on the decomposed high-frequency components : ； In the formula, and represent the total energy of the high-frequency components and the total energy of the signal, respectively; represents the th high-frequency component; represents the decomposition level of the high-frequency components; S223. Based on the high-frequency energy ratio Combined with the dynamic weight coefficient Adjust the measurement noise covariance matrix of the Kalman filter state space model: ； In the formula, represents the adjusted measurement noise covariance matrix; Among them, ； In the formula, and respectively represent the maximum weight coefficient and the minimum weight coefficient; represents the mean value within the sliding window; represents the high-frequency energy proportion threshold; represents the historical maximum high-frequency energy proportion; S24. Perform time update and measurement iterative update on the Kalman filter state space model: Time update: ; ； Measurement update: ; ； ； Wherein, represents the prior estimate of the state vector at the -th iteration, based on all the observation information at the -th iteration; represents the posterior estimate of the state vector at the -th iteration, after considering all the observation information up to and including the -th iteration; represents the posterior estimate of the state vector at the -th iteration, after considering all the observation information up to and including the -th iteration; represents the charge and discharge current at the -th iteration; represents the operating temperature at the -th iteration; represents the prior estimate error covariance matrix at the -th iteration, based on the -th iteration; represents the error covariance matrix of the posterior estimate after the -th iteration; represents the error covariance matrix of the posterior estimate after the -th iteration; represents the state transition matrix at the -th iteration; represents the covariance matrix of the process noise; represents the updated Kalman gain; represents the updated observation matrix; represents the updated measurement noise covariance matrix; represents the battery measurement value at the -th iteration; represents the identity matrix; represents the -th iteration of the state transition matrix transposed; represents the -th iteration of the updated observation matrix transposed; S25. Collect the voltages of each single cell in the energy storage battery pack and the operating temperature , and after normalization, organize them into a tensor format to obtain the preprocessed tensor data ; where represents the single cell number represents the time step represents the batch size represents the number of single cells represents the number of time steps represents two channels of voltage and temperature represents the set of real numbers S26. Construct a three-dimensional convolutional neural network and use the labeled unbalanced battery pack data to train the network and tensor data Train the three-dimensional convolutional neural network; S27. Convert the data of the energy storage battery pack collected in real time after preprocessing into tensor data, input it into the trained three-dimensional convolutional neural network, use the three-dimensional convolutional neural network to identify the imbalance pattern of the battery pack, and output the imbalance feature vector .

5. A method for balancing control of an energy storage battery pack according to claim 4, characterized in that: In step S213, the initialized state transition matrix and the observation matrix are initialized, and the process noise covariance matrix and the measurement noise covariance matrix are calculated as follows: ； ； ； ； Wherein, represents the sampling time interval; represents the state of charge time constant; represents the polarization voltage time constant; represents the slope of the open circuit voltage - state of charge curve; represents the state of charge estimation error variance; represents the voltage measurement error variance; represents the temperature measurement error variance.

6. A method for balancing control of an energy storage battery pack according to claim 5, characterized in that: The 3D convolutional neural network described in step S26 includes multiple 3D convolutional layers, 3D max pooling layers, and fully connected layers. Among them, the 3D convolutional layer: uses a convolutional kernel with a size of 3 3 × 3 to extract spatio-temporal features; The pooling size of the 3D max pooling layer is 2 2 2 to reduce the feature dimension; Fully connected layer: Used for classification output.

7. A method for balancing control of an energy storage battery pack according to claim 6, characterized in that: Step S3 specifically includes the following steps: S31. Based on the model predictive control theory, with the balancing time, energy loss, temperature fluctuation, and unbalanced degree of the energy storage battery pack as the optimization objectives, combined with the multi-physical field-aging coupling model constructed in S1 and the spatiotemporal-frequency domain joint state estimated in S2, an optimization problem data model is constructed, and rolling optimization is carried out within a finite time domain to generate the optimal control quantity; S311. Taking the equalization time of the energy storage battery pack , energy loss , temperature fluctuation and the degree of imbalance as the optimization objectives, construct the following objective function : ； In the formula, , , and all represent weight coefficients, and ; represents the reference equilibrium time; represents the safety temperature threshold; represents the two-norm of the unbalanced feature vector; S312. Based on the multi-physical field-aging coupling model constructed from S1, a MPC prediction model is established by combining the battery state estimated from S2, and the voltages, capacities, and temperatures of each cell in the energy storage battery pack within a future finite time domain are predicted through the state space equation. of the energy storage battery pack within a future finite time domain: ； ； In the formula, represents the state vector of the n-th iteration prediction; represents the state vector of the m-th iteration prediction; and respectively represent the equalization control inputs of the n-th iteration and the m-th iteration; represents the working temperature at the n-th iteration; represents the output vector of the S313. Considering battery safety and hardware limitations, add constraint conditions: State constraint: ; Control Constraint: ; Safety constraint: ; In the formula, and respectively represent the minimum and maximum allowable state of charge; represents the th state of charge of a single battery cell; and respectively represent the lower and upper limits of the allowable operating voltage of a single battery cell; represents the th terminal voltage of a single battery cell; and respectively represent the lower and upper limits of the equalization control input; represents the equalization control input amount for the th single battery cell; S314. Use sequential quadratic programming or interior point method to solve the mathematical model of the optimization problem, and obtain the optimal control quantity for the current period. ; S32. Use the deep reinforcement learning algorithm to achieve the optimal control quantity in the current cycle Interact and learn with the energy storage battery pack environment, and dynamically adjust the weight coefficients and prediction horizons of the MPC prediction model; S321. Define the state space of deep reinforcement learning , the action space and the reward function : State space: ; Action space: ; Reward function: ; In the formula, , , and respectively represent the adjustment amounts of the weight coefficients , , and ; represents the adjustment amount of the finite time domain ; , , and all represent the weight coefficients of the reward function, and ; represents the reference objective function value; S322. Train a deep reinforcement learning network using the operating data of the MPC prediction model, update the parameters of the MPC prediction model using the trained deep reinforcement learning network, and feedback the MPC prediction model parameters to step S311 to form a closed-loop optimization. The MPC prediction model parameter update formula is as follows: ； In the formula, represents the weight coefficient of the updated objective function, and ; represents the weight coefficient of the original objective function; represents the adjustment amount of the weight coefficient of the objective function; represents the updated finite time domain ; S33. Combine the rolling optimization of the MPC prediction model with the parameter adaptability of deep reinforcement learning to generate and execute a dynamic balancing strategy, and achieve energy transfer or dissipation through a hardware circuit to complete battery pack balancing; S331. Customize the following equilibrium strategy according to the optimal control quantity output by the MPC prediction model , as follows: When is positive, energy transfer is performed using a bidirectional DC-DC converter; When is negative, resistive dissipation is performed; S332. Collect the balanced current, voltage, and working temperature data and feedback it to step S2 to update the state parameters, and input the balancing effect into the deep reinforcement learning network to update the balancing strategy.

8. A method for balancing control of an energy storage battery pack according to claim 7, characterized in that: Step S4 Specifically, it includes the following steps: S41. Based on the computational fluid dynamics and heat transfer theory, construct a heat conduction-convection coupling model for the battery pack to analyze the interaction between the coolant flow characteristics and the battery heat generation distribution; S411. Construct a hydrodynamic model: Establish a three-dimensional geometric model of the battery pack flow channel, define the cross-sectional shape of the flow channel, the pipe diameter and the fluid flow velocity , and use the N-S equation to describe the turbulent flow of the coolant, and solve the flow field distribution by the finite element method: ； ； In the formula, represents the fluid density; represents the dynamic viscosity; represents the acceleration due to gravity; S412. Total heat production power based on step S114 Establish a heat conduction equation: ； In the formula, , and respectively represent the material density, specific heat capacity, and thermal conductivity of the energy storage battery pack; S42. Use the conjugate heat transfer method to calculate the convective heat transfer coefficient between the battery and the coolant , and solve the temperature field by coupling ； Among them, ; And set the following convective heat transfer boundary conditions: ; In the formula, represents the fluid thermal conductivity; represents the Nusselt number; represents the Reynolds number; represents the Prandtl number; represents the pump speed of the cooling fan; represents the fluid temperature; S43. Based on model predictive control and fuzzy logic, construct an electro-thermal-fluid multi-objective collaborative control algorithm to achieve an optimal balance of battery temperature uniformity, energy consumption, and fluid flow rate; S431. Define a multi-objective function that includes temperature deviation, fluid energy consumption, and electrical performance impact : ； In the formula, , and all represent the safety threshold weights, and ; represents the maximum temperature difference of the energy storage battery pack; represents the pump power consumption of the cooling fan; represents the reference power consumption; represents the state of charge deviation caused by the temperature difference; represents the rated state of charge; S432. Construct fuzzy control rules: Define the input variables as the temperature deviation and the temperature change rate , set up a fuzzy rule base based on the input change amount, and then use the Mamdani method for fuzzy inference and defuzzification to obtain the fluid flow rate adjustment amount and the pump speed adjustment amount ; S44. Based on the heat conduction-convection coupling model in step S41 and the fuzzy control rules in step S43, drive the fluid system and the electrical balancing system to work together, and optimize the control parameters through real-time feedback.

9. A method for balancing control of an energy storage battery pack according to claim 8, characterized in that: Step S5 specifically includes the following steps: S51. Integrate the battery manufacturing sorting data, multi-physical field characteristics during use, and the law of performance degradation during retirement to construct a digital twin model covering the entire life cycle of the battery; S511. Manufacturing and sorting stage modeling: Collecting parameters of single cells , representing the open-circuit voltage, simulating manufacturing deviations based on the Monte Carlo method, constructing a parameter distribution model of single cells, and optimizing the sorting strategy through the genetic algorithm: ； In the formula, represents the sorting matching degree; and respectively represent the standard deviation of the single cell capacity and the mean value of the single cell capacity; S512. Multi-physical field modeling during use: Integrate the P2D electrochemistry-thermal coupling model described in step S1, the state estimation process described in step S2, the balancing control process described in step S3, and the fluid dynamics equation in step S4 to establish a digital twin during use; S513, Decommissioning cascade utilization modeling: Based on the aging prediction model described in step S1.2, predict the capacity at retirement : ； Wherein, represents the rated capacity; represents the attenuation coefficient; S52. Through a multi-source data fusion algorithm, interact the real-time monitoring data with the digital twin model, and use the federated learning algorithm to achieve dynamic calibration of the full life cycle model; S53. Cross-stage mapping: Establish manufacturing parameters and mapping relationships with performance indicators in the usage stage : ； In the formula, represents a mapping function; represents the degree of aging; At the same time, align the aging trajectories of different energy storage battery packs through the dynamic time warping algorithm; S54. Define a multi-objective function including manufacturing cost, usage efficiency, and retirement value based on the digital twin model of the whole life cycle : ； In the formula, , and all represent weight coefficients, and ; and respectively represent the manufacturing cost and the benchmark cost; and respectively represent the remaining value at retirement and the benchmark value; S55. Based on multiple objective functions , optimize the digital twin model using reinforcement learning; S55. Map the cross-stage strategy to the manufacturing sorting rules, control parameters during use, and retirement decision logic, execute them respectively, and collect the execution data to iteratively update the cross-stage strategy parameters through the gradient descent method.

Images