Energy storage cabinet multi-cell consistency cooperative control method

Abstract

The present application relates to the technical fields of energy storage system and battery management, in particular to a kind of energy storage cabinet multi-cell consistency collaborative control method.The method comprises: obtaining the data such as irreversible energy loss constant, sampling data and polarization relaxation time constant of cell;In the speed update formula of chaos particle swarm optimization algorithm, introduce the non-symmetrical inertia weight coefficient with power state deviation as switching basis;Energy loss constant is used as check item, and hard constraint penalty is executed to particle triggering heat safety threshold, and the optimal power distribution weight matrix is output;Then, combine polarization relaxation time constant to construct bipartite graph matching model to solve the optimal pairing relationship of balanced node, and generate non-symmetrical pulse width compensation factor to drive bidirectional active equalization circuit according to this.The present application effectively solves the problems of cell local overheating, polarization aggravation and long-term consistency deterioration caused by the existing power distribution algorithm departing from physical boundary.

Description

Technical Field

[0001] This invention relates to the field of energy storage systems and battery management technology, specifically to a method for consistent and coordinated control of multiple battery cells in an energy storage cabinet. Background Technology

[0002] Large-capacity battery energy storage systems, as a key component of power energy storage systems and distribution networks, achieve the storage, conversion, and regulation of electrical energy through charge and discharge control circuitry. In power system frequency regulation and peak shaving applications, battery clusters consist of multiple cells connected in series and parallel. Due to manufacturing process tolerances and differences in operating environments, cells exhibit physical inconsistencies in internal resistance, capacity, and polarization characteristics. These differences can lead to capacity shifts in battery clusters during energy storage and distribution, thereby affecting the output stability and reliability of the power supply circuitry.

[0003] In existing technologies, multi-agent systems and particle swarm optimization (PSO) algorithms are introduced into the power scheduling and energy management of energy storage cabinets to allocate the charging and discharging current of each battery cluster. However, existing PSO algorithms focus on error optimization at the purely mathematical level, neglecting the strong constraints of battery physical characteristics. The energy conversion efficiency during the charging and discharging process of a battery cell exhibits high nonlinearity and definite physical losses during long cycles. When the PSO algorithm iteratively calculates without considering physical boundaries such as the cell's adiabatic temperature rise limit and polarization characteristics, the allocated power scheduling commands can easily cause specific cells to operate in a high polarization range, inducing local abnormal temperature rises or accelerated capacity decay, thereby reducing the operational safety and economy of the power storage system.

[0004] Therefore, there is an urgent need to design a collaborative control method that can directly hard-encode the electrochemical mechanism of the battery cell (including energy loss characteristics and thermal safety boundaries) into the structure of the particle swarm algorithm, so as to achieve precise scheduling and safe control of large-capacity energy storage cabinets in power distribution and energy storage scenarios.

[0005] To address this, a method for consistent collaborative control of multiple battery cells in an energy storage cabinet is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-cell consistency collaborative control method for energy storage cabinets. By deeply integrating the deep physical and electrochemical mechanisms of the cells (such as irreversible energy loss, polarization relaxation time constant, and thermal safety boundary) into the chaotic particle swarm optimization algorithm and the underlying pairing control model, this method solves the problems of local overheating, intensified polarization, and long-term consistency deterioration caused by existing power allocation algorithms deviating from physical boundaries.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for coordinating and controlling the consistency of multiple battery cells in an energy storage cabinet includes: acquiring sampled data and polarization relaxation time constants of battery cell nodes; defining a physical state constraint region based on the sampled data as the search space of a chaotic particle swarm optimization algorithm; initializing particle position vectors representing candidate power allocation weight matrices using chaotic mapping; and acquiring the irreversible energy loss constant of a single charge-discharge cycle of a battery cell.

[0009] An asymmetric inertial weight coefficient based on power state deviation is introduced into the velocity update iteration formula of the chaotic particle swarm optimization algorithm.

[0010] The irreversible energy loss constant is used as the discrete integral weighting coefficient of the fitness function to verify the candidate power allocation weight matrix. For particles that are predicted to cause the battery cell to reach the thermal safety alarm threshold, fitness penalty processing is performed. The particles that pass the verification are perturbed using a chaotic sequence, and the power allocation weight matrix is ​​output.

[0011] Based on the power allocation weight matrix, each cell node is divided into equal discharge nodes and equal charging nodes. A bipartite graph matching model is constructed by combining the polarization relaxation time constant to solve the optimal pairing relationship between equal discharge nodes and equal charging nodes. Based on the difference in polarization relaxation time constant between equal discharge nodes and equal charging nodes in the optimal pairing relationship, an asymmetric pulse width compensation factor is generated to drive the bidirectional active equalization circuit.

[0012] Preferably, the sampling data includes the terminal voltage, charging / discharging current, surface temperature, and state of charge data of each cell node; a second-order spectral density of the cell is established based on the sampling data. Equivalent circuit model, and identification of the circuit using recursive least squares method. The polarization relaxation time constant is calculated from the ohmic internal resistance, polarization internal resistance, and polarization capacitance in the equivalent circuit model; the physical state constraint region is jointly defined by the preset safe operating threshold of the terminal voltage, charging and discharging current, surface temperature, and state of charge data; the initial particle swarm is uniformly distributed within the search space defined by the safe operating threshold using chaotic mapping; the irreversible energy loss constant is obtained by performing standard charge and discharge cycle tests on the same batch of cells in advance, extracting the difference between the total charged energy and the total discharged energy, and calibrating the irreversible energy loss constant by combining the coulomb efficiency decay curves under different aging stages.

[0013] Preferably, the introduction of the asymmetric inertia weight coefficient in the velocity update iteration formula of the chaotic particle swarm optimization algorithm includes:

[0014] The velocity update iteration formula is configured as a vector superposition of inertia retention term, individual cognitive optimization term, and global social optimization term; the difference between the allocated power represented by the corresponding particle and the average allocated power of the system is extracted from the candidate power allocation weight matrix as the power state deviation; the asymmetric inertia weight coefficient is dynamically determined according to the numerical sign of the power state deviation: when the power state deviation is greater than zero, the first inertia weight is selected as the asymmetric inertia weight coefficient to reduce the inertia of the particle in maintaining its original divergence direction; when the power state deviation is not greater than zero, the second inertia weight is selected as the asymmetric inertia weight coefficient to maintain the global exploration capability of the particle, and the first inertia weight is less than the second inertia weight; the asymmetric inertia weight coefficient is introduced as a scaling factor into the inertia retention term and multiplied by the particle's current velocity vector.

[0015] Preferably, the verification of the candidate power allocation weight matrix includes: constructing a fitness evaluation function with the objectives of minimizing the total energy loss during the equalization process and minimizing the cell charge-to-displacement ratio; in the fitness evaluation function, performing discrete integration calculation on the irreversible energy loss constant along a preset equalization time step to output a system energy loss characterization term; extracting the state of charge from the sampled data of all cell nodes to calculate the system state variance as a consistency characterization term; introducing an adaptive weight coefficient based on real-time dynamic adjustment of the charge-to-displacement ratio to perform a weighted summation on the system energy loss characterization term and the consistency characterization term to calculate the comprehensive fitness value of each particle; when the real-time charge-to-displacement ratio is not less than a preset range threshold, setting the weights of the consistency characterization term and the system energy loss characterization term to a first preset weight and a second preset weight, respectively, with the first preset weight being less than the second preset weight; when the real-time charge-to-displacement ratio is less than a preset range threshold, setting the weights of the consistency characterization term and the system energy loss characterization term to a third preset weight and a fourth preset weight, respectively, with the third preset weight being greater than the fourth preset weight.

[0016] Preferably, the fitness penalty processing includes: in each iteration, predicting the temperature rise of each cell within a preset time window based on the candidate power allocation weight matrix and the electrochemical heat generation model; when it is determined that the predicted temperature rise of any cell, when combined with the current node temperature, exceeds the thermal safety alarm threshold, triggering a hard constraint penalty mechanism, marking the corresponding particle as an out-of-bounds infeasible solution, and forcibly assigning a preset penalty constant to the comprehensive fitness value of the corresponding particle in the current iteration step.

[0017] Preferably, obtaining the optimal pairing relationship between the equalization discharge node and the equalization charging node includes: based on the power allocation weight matrix, classifying the cell nodes with negative power allocation as equalization discharge nodes and the cell nodes with positive power allocation as equalization charging nodes; based on the polarization relaxation time constants of the equalization discharge node and the equalization charging node, calculating the difference in polarization response rate between any equalization discharge node and any equalization charging node, and constructing a cross-polarization coupling cost matrix characterizing the degree of transient charge transfer compatibility between the two; and integrating the steady-state energy transfer demand extracted from the power allocation weight matrix with the cross-polarization coupling cost matrix to construct a bipartite graph matching model, and solving for the optimal pairing relationship with the minimum total energy transfer cost.

[0018] Preferably, generating the asymmetric pulse width compensation factor to drive the bidirectional active balancing circuit includes: generating an asymmetric pulse width compensation factor based on the difference in polarization relaxation time constants between the equalization discharge node and the equalization charging node, adjusting the charging and discharging timing of the transient energy storage element in the bidirectional active balancing circuit; injecting the asymmetric pulse width compensation factor into the underlying control loop to generate an interlock drive signal, and alternately controlling the on and off states of the upper and lower bridge arm switches in the bidirectional active balancing circuit through the interlock drive signal to control the bidirectional active balancing circuit.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] 1. This invention obtains the irreversible energy loss constant and polarization relaxation time constant of a single cycle of a battery cell, and directly incorporates the temperature rise prediction value based on the electrochemical heat generation model as a hard constraint penalty term for thermal safety alarm into the particle swarm algorithm. This avoids the problem of specific battery cells deviating from the physical adiabatic boundary caused by pure mathematical optimization, and avoids the risk of thermal runaway caused by local overheating.

[0021] 2. This invention introduces an asymmetric inertial weight coefficient based on power state deviation into the velocity update iteration formula of the particle swarm optimization algorithm. This enables the algorithm to dynamically adjust the divergent inertia and exploration kinetic energy of the particles according to the real-time load state of the battery cell when approaching the optimal solution. It also combines chaotic sequences to perform global perturbation and mutation, which overcomes the defect of traditional algorithms that are prone to getting trapped in local optima and improves the convergence speed and global search accuracy of parameter optimization.

[0022] 3. This invention employs a cross-polarization coupling cost matrix based on the difference in polarization relaxation time constants and a bipartite graph matching model to solve for the optimal pairing relationship, and generates an asymmetric pulse width compensation factor to directly drive the underlying bidirectional active balancing circuit. This mechanism not only deeply adapts the balancing current flow direction to the electrochemical polarization state inside the cell, effectively reducing energy loss in transient charge transfer, but also constitutes an adaptive complete closed loop from top-level digital space optimization to bottom-level physical space control, improving the accuracy of control methods and the overall operating efficiency of the energy storage system. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the overall process of multi-cell consistency and collaborative control of the energy storage cabinet according to the present invention.

[0024] Figure 2 This is the second-order of the present invention. Model and polarization relaxation time constant identification flowchart;

[0025] Figure 3 This is the logic diagram for the asymmetric optimization and temperature rise prediction control of the present invention;

[0026] Figure 4 This is the optimal pairing and active equilibrium control graph based on bipartite graphs of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Example 1:

[0029] Please see Figures 1 to 4 ,in, Figure 1 This is an overall flowchart of the multi-cell consistency and collaborative control of the energy storage cabinet of the present invention. Figure 2 This is the second-order of the present invention. Model and polarization relaxation time constant identification flowchart Figure 3 This is the logic diagram for the asymmetric optimization and temperature rise prediction control of the present invention. Figure 4 As the optimal pairing and active balancing control graph based on bipartite graph of the present invention, the present invention provides a multi-cell consistency cooperative control method for energy storage cabinets, the specific implementation steps of which are as follows:

[0030] Obtain the sampling data and polarization relaxation time constant of the cell node; define the physical state constraint region based on the sampling data as the search space of the chaotic particle swarm optimization algorithm, and initialize the particle position vector representing the candidate power allocation weight matrix using chaotic mapping; obtain the irreversible energy loss constant of a single charge-discharge cycle of the cell.

[0031] An asymmetric inertial weight coefficient based on power state deviation is introduced into the velocity update iteration formula of the chaotic particle swarm optimization algorithm.

[0032] The irreversible energy loss constant is used as the discrete integral weighting coefficient of the fitness function to verify the candidate power allocation weight matrix. For particles that are predicted to cause the battery cell to reach the thermal safety alarm threshold, fitness penalty processing is performed. The particles that pass the verification are perturbed using a chaotic sequence, and the power allocation weight matrix is ​​output.

[0033] Based on the power allocation weight matrix, each cell node is divided into equal discharge nodes and equal charging nodes. A bipartite graph matching model is constructed by combining the polarization relaxation time constant to solve the optimal pairing relationship between equal discharge nodes and equal charging nodes. Based on the difference in polarization relaxation time constant between equal discharge nodes and equal charging nodes in the optimal pairing relationship, an asymmetric pulse width compensation factor is generated to drive the bidirectional active equalization circuit.

[0034] Furthermore, the sampling data includes the terminal voltage, charging / discharging current, surface temperature, and state of charge data of each cell node; a second-order spectral density of the cell is established based on the sampling data. Equivalent circuit model, and identification of the circuit using recursive least squares method. The polarization relaxation time constant is calculated from the ohmic internal resistance, polarization internal resistance, and polarization capacitance in the equivalent circuit model; the physical state constraint region is jointly defined by the preset safe operating threshold of the terminal voltage, charging and discharging current, surface temperature, and state of charge data; the initial particle swarm is uniformly distributed within the search space defined by the safe operating threshold using chaotic mapping; the irreversible energy loss constant is obtained by performing standard charge and discharge cycle tests on the same batch of cells in advance, extracting the difference between the total charged energy and the total discharged energy, and calibrating the irreversible energy loss constant by combining the coulomb efficiency decay curves under different aging stages.

[0035] The second order The equivalent circuit model consists of one ohmic internal resistance connected in series and two parallel circuits. The branch structures respectively characterize the electrochemical polarization and concentration polarization processes. The recursive least squares method, within each sampling period, recursively estimates the model parameters by updating the covariance matrix and gain matrix online based on real-time data of terminal voltage and charge / discharge current. The specific formula is as follows: ; The formula for updating the gain matrix is: ; The formula for updating the covariance matrix is: ; in The parameter estimation vector (including ohmic internal resistance, polarization internal resistance, and polarization capacitance). Sampling time, To observe the voltage, For the regression vector, Here is the gain matrix. Let covariance matrix be the variance matrix. To address the characteristics of sudden changes in charging and discharging power in frequency modulation scenarios, this invention employs an adaptive forgetting factor adjustment mechanism based on the rate of change of current: when the rate of change of current is detected... When the value exceeds the preset threshold for sudden changes in operating conditions, Transient adjustment to 0.95 to accelerate. The algorithm's parameter convergence and historical data forgetting; during the current steady-state period, Restore to To ensure the noise resistance accuracy of parameter identification, thereby ensuring the polarization relaxation time constant. Real-time performance and accuracy. Considering the large time constant characteristic of polarization parameters, this invention introduces asynchronous update logic: when the current change rate is detected to be in the transient adjustment period ( When this occurs, the system activates the polarization parameter observer buffer, and stops monitoring within this window. and Instead of direct physical quantity replacement, a historical stability window-based approach is adopted. The moving average trend is used for compensation calculation. Only when the cumulative number of stationary sampling points is re-covered at least... Only after a certain time span is elapsed is it permissible to synchronize new time constant values ​​to the underlying bipartite graph matching model, thereby eliminating the misleading influence of identification spurious oscillations under short memory windows on pairing decisions. The preset operating condition mutation threshold is a benchmark value for the maximum allowable transient current change rate, calculated based on the maximum rated power ramp-up rate of the energy storage system under primary or secondary frequency regulation scenarios, combined with the total nominal capacity and rated voltage of the battery cluster. In this embodiment, to balance noise immunity accuracy and transient response, this threshold is specifically set as follows: The polarization relaxation time constant The identified polarization resistance and polarization capacitor Calculated, i.e. .

[0036] The preset safe operating threshold is obtained from the cell technical specifications, including: upper limit of terminal voltage. Set as Lower limit Set as Upper limit of charging and discharging current Set as Surface temperature limit Set to 50°C, lower limit Set to -10°C, upper limit of state of charge Set to 90% and lower limit Set to 10%.

[0037] The chaotic mapping specifically adopts The mapping, whose iterative formula is used to generate in A pseudo-random sequence uniformly distributed within the interval is linearly transformed and mapped to a multidimensional search space defined by the safe operation threshold, serving as the position vector of the initial particle swarm to ensure the ergodicity and diversity of the initial population.

[0038] The standard charge-discharge cycle test refers to performing a complete charge-discharge cycle on the battery cell at rated current under constant temperature conditions and recording the energy input. and release energy The irreversible energy loss per cycle is calculated by extracting the difference between the two values. .

[0039] The coulomb efficiency decay curves were obtained by repeating the above tests at different cycle numbers (e.g., the 100th, 300th, and 500th cycles). Specifically, the single irreversible energy loss corresponding to each aging stage was collected. As the dependent variable, and the corresponding number of loops When using variables as independent variables for mathematical fitting, the least squares method is typically employed. The specific mathematical expression for polynomial fitting of order 1 is: ; In the formula, The degree of the polynomial is 2 or 3 in this embodiment. The coefficients are polynomials. By constructing a least-squares objective function: ; The fitting coefficients are then solved by minimizing them. Where, if the order... When the target value is less than the preset precision threshold (e.g.) ), then take To maintain the robustness of long-term extrapolation predictions; otherwise, increase the order. The nonlinear characteristics of the aging curve are captured using a third-order term, ensuring calibration accuracy under complex aging conditions. By fitting this decay curve, the irreversible energy loss constant that dynamically varies with the degree of aging is finally obtained. This constant characterizes the inherent energy loss rate caused by internal resistance growth, side reactions, etc., per unit charge-discharge cycle. It will be directly incorporated into the fitness function of the particle swarm optimization algorithm as a discrete integral weighting coefficient, used to accurately estimate the long-term energy loss of the system under different power allocation strategies during the optimization process.

[0040] In this embodiment, chaotic mapping ensures the ergodicity and diversity of the initial particle population within the search space. Simultaneously, by obtaining the irreversible energy loss constant and the polarization relaxation time constant, a foundation is laid for incorporating physical and thermal safety boundaries into the optimization algorithm, thereby avoiding the safety risks associated with purely mathematical optimization.

[0041] Furthermore, the introduction of the asymmetric inertial weight coefficient in the velocity update iteration formula of the chaotic particle swarm optimization algorithm includes: configuring the velocity update iteration formula as a vector superposition of inertia maintenance term, individual cognitive optimization term, and global social optimization term; extracting the difference between the allocated power represented by the corresponding particle and the average allocated power of the system in the candidate power allocation weight matrix as the power state deviation; dynamically determining the asymmetric inertial weight coefficient according to the numerical sign of the power state deviation: when the power state deviation is greater than zero, selecting the first inertial weight as the asymmetric inertial weight coefficient to reduce the inertia of the particle in maintaining its original divergence direction; when the power state deviation is not greater than zero, selecting the second inertial weight as the asymmetric inertial weight coefficient to maintain the global exploration capability of the particle, wherein the first inertial weight is less than the second inertial weight; introducing the asymmetric inertial weight coefficient as a scaling factor into the inertia maintenance term and multiplying it with the particle's current velocity vector.

[0042] The standard form of the speed update iteration formula is as follows: ,in and Particles In the The velocity and position vector of the next iteration For particles The best historical position The optimal position globally. , As the learning factor, this embodiment takes , , for Random numbers within the interval.

[0043] Elements in the power allocation weight matrix Indicates assignment to the first The first cell Power allocation weights, average power allocation in the system: ; in To be based on the weight vector Calculated The power distribution of each cell, the power state deviation .

[0044] The asymmetric inertia weighting coefficient Defined as: ; First inertial weight Second inertial weight This is based on 128 dimensions in specific energy storage scenarios. Simulation experiments on convergence performance show that, under the constraint of 500 iterations, the convergence performance is verified by grid search. , The combination of the above values ​​within the specified range has the highest global optimal convergence rate.

[0045] The revised speed update formula is as follows: .

[0046] when When the allocated power is higher than average, posing a high risk of overload, a smaller power is used. This aims to weaken the inertial kinetic energy that allows the particle to continue diverging from its current high-power state, reducing its dependence on unfavorable historical velocities and making it more susceptible to being guided by the globally optimal solution (a safer equilibrium solution) and quickly returning to a safe search region; when At that time, use a larger This gives the particle greater kinetic energy, encouraging it (corresponding to low-power or equilibrium solutions) to explore more extensively in order to find a better uniform equilibrium solution.

[0047] This embodiment enables the algorithm to dynamically adjust the particle's optimization kinetic energy based on the real-time load state of the battery cell when approaching the optimal solution. When the power state deviation is greater than zero, its divergent inertia is weakened, prompting it to quickly return to the safe search area and avoid local overheating; when the deviation is not greater than zero, it is given greater exploration kinetic energy, encouraging the particle to find a better uniform equilibrium solution.

[0048] Further, the verification of the candidate power allocation weight matrix includes: constructing a fitness evaluation function with the objectives of minimizing the total energy loss during the equalization process and minimizing the cell charge-to-displacement ratio; in the fitness evaluation function, the irreversible energy loss constant is discretely integrated along a preset equalization time step to output a system energy loss characterization term; the state of charge is extracted from the sampled data of all cell nodes to calculate the system state variance as a consistency characterization term; an adaptive weight coefficient based on real-time dynamic adjustment of the charge-to-displacement ratio is introduced to perform a weighted summation of the system energy loss characterization term and the consistency characterization term to calculate the comprehensive fitness value of each particle; when the real-time charge-to-displacement ratio is not less than a preset range threshold, the weights of the consistency characterization term and the system energy loss characterization term are set to a first preset weight and a second preset weight, respectively, with the first preset weight being less than the second preset weight; when the real-time charge-to-displacement ratio is less than a preset range threshold, the weights of the consistency characterization term and the system energy loss characterization term are set to a third preset weight and a fourth preset weight, respectively, with the third preset weight being greater than the fourth preset weight.

[0049] Specifically, the calculation of the comprehensive fitness value involves using an extreme value normalization method to map the calculated system energy loss characterization term and consistency characterization term to... The dimensionless interval is used to output the normalized energy loss term and the normalized consistency term; then, an adaptive weighting coefficient based on real-time dynamic adjustment of charge difference is introduced to perform a weighted summation of the normalized energy loss term and the normalized consistency term to calculate the comprehensive fitness value of each particle.

[0050] Wherein, the fitness evaluation function Defined as: ; in, Assign particle position vectors corresponding to weight matrices to candidate powers. This indicates mapping variables to The extremum normalization function of the interval, the extremum normalization function The denominator is the maximum value of the corresponding index for all candidate solutions in the current particle swarm, i.e. This contemporary extreme value normalization is performed independently within each iteration and is only used for relative comparison and sorting of particles within the current iteration, and is not used directly across iterations.

[0051] To address the inconsistency in evaluation scales caused by dynamic shifts in the normalization denominator during cross-iteration comparisons, and to avoid logical conflicts arising from the direct addition of physical quantities with different dimensions, this invention introduces a fixed benchmark normalization mechanism based on the maximum value of the initial generation: in the first iteration (initial generation)... When the initial particle swarm is in its initial state, record the original maximum value of the system energy loss corresponding to all candidate schemes. and the original maximum value of the consistency representation term As a fixed normalization benchmark for cross-generational comparisons; in subsequent iterations ( In ), the best position in the particle's history ( ) and the global optimal position ( The updates and comparisons do not use contemporary dynamic normalized values, but instead use the original index values ​​of each particle. and The standard comprehensive fitness value is calculated by dividing each value by the fixed normalized benchmark. The determination is made using the following formula: ; In the formula, and These are adaptive weighting coefficients determined for the current sampling period. This ensures consistency in the evaluation standard scale and dimensional compatibility during intergenerational comparisons. If all particles... If all terms are zero, then the corresponding normalization term is set to zero.

[0052] The system energy loss characterization item The calculation formula is: ; in, The preset total number of equalization time steps, For a single time step, This represents the total number of battery cells. For the first The irreversible energy loss constant of each battery cell In the first Each time step, based on weight Assigned to the The power of each battery cell (positive for charging, negative for discharging). The consistency characterization item. Using the system's state of charge ( Variance calculation: ; in, In weight Under the corresponding power allocation strategy, the predicted first At the end of the equalization cycle of each cell value, For the average of all cells The aforementioned The specific prediction method is: power based on weight allocation. The predicted operating current is calculated based on the real-time terminal voltage of the battery cell. ,in The real-time terminal voltage value of the i-th cell is then used, combined with the real-time available capacity during the polarization relaxation identification process, through the ampere-hour integration method. The discrete integral is updated using the following formula: ; The real-time charge difference Based on all cells in the current sampling period Value calculation. The preset range threshold. Dynamically adjust according to the application scenarios of the battery pack. In frequency modulation scenarios that require fast response and frequent high-power charging and discharging, set a real-time adjustment range threshold: when the power command changes drastically, temporarily relax the consistency requirement to 20% to prioritize power response; in the power stable stage, tighten the consistency requirement to 10% to perform fine balancing.

[0053] The adaptive weight coefficient adjustment rule is as follows: ; in, and The first preset weight and the second preset weight (set to 0.3 and 0.7 respectively) satisfy the following conditions: This indicates that when there are large differences in consistency, priority should be given to optimizing the equilibrium effect. and The third and fourth preset weights are set to 0.7 and 0.3 respectively, satisfying the following conditions: This indicates that energy efficiency is prioritized when consistency is good. In frequency regulation scenarios where the range threshold is dynamically adjusted, the calculation is immediately recalculated based on the new threshold after each threshold update. and The relative sizes of the particles are determined, and corresponding weight combinations are selected accordingly. The smaller the overall fitness value, the better the particle (power allocation scheme).

[0054] This embodiment achieves a dynamic balance between the system's energy loss and consistency characteristics. When cell consistency differences are large, priority is given to optimizing the balancing effect; when consistency is good, priority is given to optimizing energy efficiency. This overcomes the shortcomings of traditional algorithms that ignore the strong constraints of battery physical characteristics, improving the operational safety and economy of the power storage system.

[0055] Furthermore, the fitness penalty processing includes: in each iteration, predicting the temperature rise of each cell within a preset time window based on the candidate power allocation weight matrix and the electrochemical heat generation model; when it is determined that the predicted temperature rise of any cell, when combined with the current node temperature, exceeds the thermal safety alarm threshold, triggering a hard constraint penalty mechanism, marking the corresponding particle as an out-of-bounds infeasible solution, and forcibly assigning a preset penalty constant to the comprehensive fitness value of the corresponding particle in the current iteration step.

[0056] The method for obtaining the temperature rise prediction value is as follows: the allocated power assigned to each cell in the candidate power allocation weight matrix is ​​converted into the corresponding predicted operating current; based on the predicted operating current, the real-time internal resistance of the cell, and the entropy change coefficient corresponding to the real-time state of charge, the dynamic heat generation power of each cell is calculated; the heat dissipation conduction difference equation containing the cell's equivalent heat capacity parameters and surface heat dissipation coefficient is called, and the dynamic heat generation power is used as the heat source input, and the step integral is performed within the future preset time window to calculate the temperature rise prediction value.

[0057] Wherein, the future preset time window length Set to 30 seconds, covering a preset number of equalization control cycles.

[0058] The dynamic heat generation power Taking into account both irreversible Joule heat and reversible reaction entropy heat, the calculation formula is as follows: ; The first term on the right side of the formula represents irreversible Joule heating. To pass the second order The model obtained the first online identification The real-time ohmic internal resistance of each battery cell; the second term is the entropy heat of the reversible reaction. This is the real-time absolute temperature of the battery cell. The cell is in its current real-time state of charge (SFC). The calibration entropy change coefficient under specific conditions. By introducing the reaction entropy heat term, the cell's performance under specific conditions can be effectively corrected. The reversible thermal effect during high-rate charging and discharging within the range ensures the physical accuracy of temperature rise prediction.

[0059] The heat dissipation and conduction difference equation adopts a lumped parameter thermal model, for the first... The temperature change equation for each battery cell is as follows: ; in, This is the equivalent heat capacity of the battery cell. The surface heat dissipation coefficient, For heat dissipation surface area, For ambient temperature, this embodiment uses the differential equation for heat dissipation and conduction to establish a small temperature gradient inside the cell (i.e., ...). number Furthermore, based on the simplified engineering assumption that the independent thermal management duct of the energy storage cabinet can maintain relatively independent convective heat transfer on the surface of each cell and ignore strong thermal crosstalk between adjacent cells (the measured crosstalk temperature difference is less than 1℃), under high frequency modulation conditions, if engineering calculations show that… If necessary, the thermal model is upgraded to a one-dimensional distributed parameter thermal model that includes radial temperature distribution, or an experimentally calibrated modified lumped thermal model is adopted. It is then discretized and solved using the forward Euler method with step integrals. ; in, For thermal simulation steps, For thermal simulation step size, For the first The temperature of the step, This represents the heat generation power corresponding to the step size. From the current temperature... Begin iterative calculation The temperature trajectory over time, taking the highest temperature. .

[0060] The predicted temperature rise The thermal safety alarm threshold The thermal stability of the battery cell's specific chemical system and materials is strictly controlled. Different combinations of positive electrode materials, negative electrode materials, and electrolytes have completely different thermal runaway induction critical points and safe operating temperature ranges. For example, for battery cells using the lithium iron phosphate chemical system, its crystal structure is relatively stable, and its thermal runaway critical temperature is relatively high. This is determined in conjunction with the technical specifications and battery management system provided by the battery cell manufacturer. The highest operating standard typically sets this thermal safety alarm threshold. It is clearly marked as 55℃. The penalty mechanism is: if If so, the particle is marked as invalid, and its overall fitness value is set. ,in The penalty constant is a preset upper limit value. The setting principle is that its value should be at least 100 times greater than the upper limit of the comprehensive fitness value of a normal feasible solution, to ensure that out-of-bounds particles are always in the elimination range during the normalized fitness ranking. Specifically, when the weighted upper limit of the normalized energy loss term and the normalized consistency term is 2, the penalty constant can be set to 200. Any value within the range.

[0061] This embodiment directly incorporates the temperature rise prediction based on the electrochemical heat generation model as a hard constraint penalty term for thermal safety alarms into the particle swarm optimization algorithm, avoiding the problem of specific battery cells deviating from physical adiabatic boundaries caused by pure mathematical optimization. It also avoids the risk of thermal runaway caused by improper scheduling commands leading to localized overheating.

[0062] Furthermore, obtaining the power allocation weight matrix includes: using... The chaotic sequence is generated by mapping. When the variance of the comprehensive fitness value of the particle swarm is less than the population stagnation threshold in a series of preset iterations, it is determined that the current search is trapped in a local optimum. The chaotic sequence is introduced into the position vector of the current global optimum particle to perform a mutation perturbation operation. After the mutation perturbation of the current iteration is completed, it is determined whether the preset termination condition is met: if not, the iteration is returned to continue; if it is met, the optimization is stopped, and the position vector of the global optimum particle is mapped and output as a power allocation weight matrix.

[0063] Among them, the The iterative formula for the mapping is: ,in To control the parameters, the value is fixed in this embodiment. , The population stagnation threshold Based on experience, a value of, for example, 1% of the initial population fitness variance is set. The monitoring method is: calculate continuous... The variance sequence of the overall fitness values ​​of all particles in (set to 10) iterations. If all values ​​in the sequence are less than If so, it is determined that the system is trapped in a local optimum.

[0064] The mutation perturbation operation specifically involves: generating a vector that corresponds to the globally optimal particle position vector. Chaotic sequence vectors of the same dimension ,in Depend on Mapping generation. The perturbation formula is: ; in, The position after the disturbance. The disturbance intensity coefficient is... and These represent the upper and lower bounds for each dimension of the search space. After completing the chaotic mutation perturbation of the globally optimal particle, the historical optimal fitness records of all particles are reset to their current actual fitness values ​​to eliminate the inconsistency between historical memory and the current exploration direction; simultaneously, the current velocity vector of each particle is kept unchanged to maintain the exploration momentum of the population. At the same time, it is necessary to... The components are reconstrained in Within the range.

[0065] The disturbance intensity coefficient The fitness improvement rate is dynamically determined based on the global optimal fitness improvement rate of the current iteration step: when continuous When all values ​​of the variance sequence of the combined fitness values ​​of all particles in the next iteration are less than the population stagnation threshold, it is determined that the population is trapped in a local optimum. Set to upper limit value (Take 0.3) to implement a strong global exploration perturbation; during the normal iteration phase, Set as (Take 0.1) to maintain local fine-grained search.

[0066] The preset termination conditions include: 1. Reaching the maximum number of iterations. (Set to 500 times); 2. The global optimal fitness value is in continuous The improvement in (set to 20) iterations is less than the accuracy threshold. (Set to 0.1%). The algorithm stops when any condition is met, and the final result is... The vector is mapped to the final power allocation weight matrix using a normalization function. .

[0067] This embodiment combines chaotic sequences to globally perturb and mutate particles trapped in local optima, effectively helping the algorithm escape local extremes. This further overcomes the shortcomings of traditional optimization algorithms that are prone to getting trapped in local optima, improving the convergence speed of parameter optimization and the global search accuracy.

[0068] Furthermore, obtaining the optimal pairing relationship between the equalization discharge node and the equalization charging node includes: based on the power allocation weight matrix, classifying the cell nodes with negative power allocation as equalization discharge nodes and the cell nodes with positive power allocation as equalization charging nodes; based on the polarization relaxation time constants of the equalization discharge node and the equalization charging node, calculating the difference in polarization response rate between any equalization discharge node and any equalization charging node, and constructing a cross-polarization coupling cost matrix characterizing the degree of transient charge transfer compatibility between the two; integrating the steady-state energy transfer demand extracted from the power allocation weight matrix with the cross-polarization coupling cost matrix to construct a bipartite graph matching model, and solving to obtain the optimal pairing relationship with the minimum total energy transfer cost.

[0069] Let the set of equal discharge nodes be . Its polarization relaxation time constant is The set of equalization charging nodes is Its polarization relaxation time constant is The difference in polarization response rates The elements of the cross-polarization coupling cost matrix are defined. for: ,in It is a monotonically increasing function, for example , The proportionality coefficient (unit: ) The steady-state energy transfer demand vector represents the additional energy loss rate caused by polarization mismatch. From the power allocation weight matrix Extracted from, for discharge nodes Its required discharge energy is For charging nodes Its required charging energy is Construct a bipartite graph , among which the side The weight (i.e., the total cost of energy transfer). Determined by both energy demand and polarization coupling costs: The bipartite graph matching model aims to find a pairing relationship (i.e., edge set). This ensures that each node is paired at most once, and the total weight of all paired edges is... The goal is to minimize energy transfer while maximizing energy efficiency. This problem can be transformed into a weighted bipartite graph matching problem. Specifically, when faced with... In this case, the system first completes the set dimension by mapping virtual nodes, uniformly reconstructing it into a weighted bipartite graph minimum weight perfect matching problem, and finally outputs the optimal set of pairing relationships. .

[0070] This embodiment employs a cross-polarization coupling cost matrix based on the difference in polarization relaxation time constants and a bipartite graph matching model to solve for the optimal pairing relationship. This approach comprehensively evaluates the compatibility of transient charge transfer and successfully obtains the optimal pairing relationship that minimizes the total energy transfer cost.

[0071] Furthermore, the step of constructing a bipartite graph matching model by combining the polarization relaxation time constant to solve the optimal pairing relationship between equal discharge nodes and equal charging nodes includes: when constructing the bipartite graph matching model, comparing the difference in the number of equal discharge nodes and equal charging nodes in real time; when it is determined that the number of equal discharge nodes and equal charging nodes are not equal, dynamically mapping virtual pairing nodes to the set with fewer nodes to complete the set dimension and reconstruct a complete bipartite graph matching model; extracting the real-time voltage ripple amplitude of the DC bus capacitor of the energy storage cabinet and the energy margin of the system auxiliary buffer circuit as evaluation indicators of the transient charge absorption and release capability of the virtual pairing nodes; based on the evaluation indicators and the polarization relaxation time constant of the unmatched cell nodes, calculating the pseudo-cross-polarization coupling cost between the real nodes and the virtual pairing nodes, and injecting it into the cross-polarization coupling cost matrix for global optimization, guiding the redundant steady-state transfer energy to the DC bus capacitor or auxiliary buffer circuit for safe unloading through the optimal pairing relationship.

[0072] This embodiment solves the problem of local cell idleness caused by the unequal number of charging and discharging cell nodes by introducing a virtual pairing node mapping mechanism based on the DC bus, thereby improving the global balancing efficiency of the energy storage system under complex operating conditions.

[0073] Furthermore, the method of generating an asymmetric pulse width compensation factor to drive the bidirectional active balancing circuit includes: generating an asymmetric pulse width compensation factor based on the difference in polarization relaxation time constants between the equalization discharge node and the equalization charging node to adjust the charging and discharging timing of the transient energy storage element in the bidirectional active balancing circuit; injecting the asymmetric pulse width compensation factor into the underlying control loop to generate an interlock drive signal; and using the interlock drive signal to alternately control the on and off states of the upper and lower bridge arm switches in the bidirectional active balancing circuit to control the bidirectional active balancing circuit.

[0074] The bidirectional active balancing circuit topology includes, but is not limited to, those based on inductors or transformers. , Or a flyback circuit. Assume an optimal pairing relationship. In the middle, discharge node With charging nodes Pairing, the difference in their polarization relaxation time constants .

[0075] The asymmetric pulse width compensation factor according to calculate: ; in The unit is seconds. This is a dimensionless proportionality coefficient, and its typical range is determined by the switching frequency of the specific equalization circuit and the allowable duty cycle adjustment range. Specifically: ; In the formula The maximum allowable duty cycle adjustment range is a dimensionless percentage constant. To balance the switching cycles of the circuit, This represents the maximum polarization relaxation time difference that the system is expected to experience. (Based on the switching frequency...) (Switching cycle) Taking the maximum duty cycle adjustment range of no more than 5% as an example, when The maximum value is hour, The underlying control loop is pulse width modulation (PWM). ) Control loop. For this paired equalization channel, standard signal duty cycle Determined by the demand for energy transfer.

[0076] The compensated drive signal is: upper arm switch drive signal. On-time: ; Lower arm switch drive signal On-time: ; in This is the switching cycle. The interlock drive signal ensures... and The levels will not be high simultaneously; a dead time will be inserted in between. To prevent straight-through. Specifically, when (Slow discharge node polarization response), then Extending the conduction time of the upper bridge arm is equivalent to providing more time during the inductor charging phase, compensating for the charge buildup process at slow-polarization nodes; conversely, it also reduces the charge buildup at slower-polarization nodes. In this way, transient current spikes or voltage overshoots caused by differences in polarization response are compensated, improving charge transfer efficiency and reducing switching stress.

[0077] This embodiment can compensate for transient current spikes or voltage overshoots caused by differences in polarization response, improving charge transfer efficiency and reducing switching stress. This mechanism deeply adapts the equalization current flow direction to the electrochemical polarization state inside the cell, effectively reducing energy loss in transient charge transfer, and forming an adaptive complete closed loop from top-level digital space optimization to bottom-level physical space control, thereby improving the accuracy of control methods and the overall operating efficiency of the energy storage system.

[0078] This embodiment incorporates the obtained irreversible energy loss constant of the battery cell and the thermal safety alarm threshold as hard constraints into a chaotic particle swarm optimization algorithm. Fitness penalties are applied to out-of-bounds particles, avoiding the risk of local overheating and thermal runaway caused by the battery cell deviating from physical boundaries due to pure mathematical optimization. Simultaneously, an asymmetric inertial weight coefficient based on power state deviation is introduced into the algorithm, and perturbations are performed using chaotic sequences to dynamically adjust the divergent inertia and exploratory kinetic energy of the particles. This effectively overcomes the shortcomings of traditional algorithms that easily get trapped in local optima, improving the convergence speed and global search accuracy of parameter optimization. Finally, a bipartite graph matching model is constructed using the polarization relaxation time constant to solve the optimal pairing relationship between charging and discharging nodes. Based on this, an asymmetric pulse width compensation factor is generated to directly drive the bidirectional active balancing circuit, ensuring a deep fit between the balancing current and the electrochemical polarization state. This reduces transient charge transfer energy loss while achieving an adaptive complete closed loop from top-level digital space optimization to bottom-level physical space control, improving the accuracy of control methods and the overall operating efficiency of the energy storage system.

[0079] Example 2: This embodiment uses an energy storage cabinet in a frequency regulation energy storage power station as a specific application scenario to explain in detail the implementation process of the multi-cell consistency and collaborative control method for the energy storage cabinet described in this invention. The energy storage cabinet consists of 128 lithium iron phosphate cells arranged in a 4-parallel, 32-series configuration to form a battery cluster, primarily used for rapid frequency regulation response of the power grid.

[0080] Before the energy storage cabinet was put into operation, all 128 cells underwent standard charge-discharge cycle testing. The test was conducted in a constant temperature environment of 25°C, and the energy input for each cycle was recorded. and release energy Calculate the single irreversible energy loss By combining the coulombic efficiency decay curves of the battery cells at different aging stages (such as 100, 300, and 500 cycles), the irreversible energy loss constant of each battery cell is obtained through fitting. At the same time, a unified thermal safety alarm threshold is set according to the cell specifications. The temperature is 55°C.

[0081] During system operation, the battery management system by The frequency of sampling the terminal voltage of each cell node Charging and discharging current Surface temperature and state of charge Based on this real-time data, a second-order [database] is established and updated online for each cell. Equivalent circuit model. This is achieved through the recursive least squares method (…). Recursively identify the ohmic internal resistance of each battery cell. Polarization internal resistance and polarization capacitor Then, the polarization relaxation time constant of each cell was calculated. .

[0082] In a scenario where 128 battery cells are simultaneously identified online... The computational complexity of each iteration of the algorithm is . ,in For parameter dimensions (in this embodiment) By employing a parallel or batch identification strategy, the 128 battery cells are divided into 8 groups of 16 cells each. The covariance matrix is ​​updated in batches within a single control cycle, ensuring that the identification delay does not exceed [a certain threshold]. ,satisfy Sampling cycle time requirements. Under high-frequency modulation disturbance conditions, if Frequent switching The internal logic will automatically process the 128 battery cells. The output performs 'confidence weight filtering'. Specifically, when When the effective memory window of the algorithm is less than the preset 90-second threshold, the bipartite graph matching model calls... The parameters are fused with historical steady-state values ​​through a first-order hysteresis loop, with the fusion weight increasing linearly with the stationary period. This mechanism effectively compensates for the lack of transient data in frequency regulation scenarios, ensuring that the polarization response rate differences acquired by the bipartite graph matching model at any given time possess physical stability. Meanwhile, the battery management system (BMS) Based on real-time acquired state of charge The lithium iron phosphate cell is retrieved from the pre-calibrated and stored table using a lookup method. Entropy change coefficient at the node Two-dimensional lookup table. Given that lithium iron phosphate batteries... for There exists a plateau period within the interval where the entropy change coefficient is approximately zero. The lookup table is located outside this plateau period. or The calibration is refined to 1% dense nodes, and 5% sparse nodes are used during the plateau period. Linear interpolation is used to calculate the current value during table lookup. The real-time entropy change coefficient is used to eliminate interpolation jump errors between adjacent nodes under high-rate charging and discharging, ensuring the prediction stability of the reversible reaction entropy heat term during the plateau period, and is used as the calculation input for subsequent electrochemical thermal models.

[0083] According to the cell technical specifications, the boundaries of the physical state constraint region (i.e., the search space) are set: terminal voltage. Current Surface temperature °C, These boundaries together constitute the search space for subsequent optimization algorithms.

[0084] When the power grid dispatching system issues a frequency regulation power command (for example, requiring the battery cluster to output power within 10 seconds), When power is reached, the control algorithm is activated. Within the 128-dimensional search space defined by the aforementioned safety boundary, the algorithm utilizes... The chaotic mapping generates 200 initial particles, and the position vector of each particle represents a candidate power allocation weight matrix. .

[0085] In each iteration, the deviation between the power allocation scheme represented by each particle and the average power allocation of the system is calculated. .like (That is, the scheme allocates higher-than-average power to some cells), so a smaller inertia weight is used. (e.g., 0.4), weakening its inertial kinetic energy to maintain a high load state, prompting it to quickly converge to a safe state, and avoiding overheating of the high-load cell; if Then a larger inertia weight is used. (e.g., 0.9) provides greater impetus for exploration and encourages the exploration of more balanced solutions.

[0086] Constructing the fitness function .in, The item uses the aforementioned calibration Discrete integration is performed on the predicted power distribution to estimate the total energy loss; Item calculates all battery cells The variance of the variance is used to assess consistency. Weights and Based on real-time charge-to-electrical difference Dynamic adjustment: when At that time, set Prioritize improving consistency; when At that time, set Prioritize energy efficiency optimization.

[0087] For each candidate solution Based on its allocated power, the temperature rise of each cell in the next 30 seconds is predicted. If the predicted temperature of any cell If the temperature reaches 55°C, the plan is immediately marked as infeasible, and its fitness value is set to a very large penalty constant (e.g., ...). This ensures that it is eliminated in the selection process.

[0088] The algorithm continuously monitors the population fitness variance. If the variance is less than 1% of the initial variance for 10 consecutive iterations, it is determined to be trapped in a local optimum. At this point, it utilizes... Chaotic sequences perturb the position of the currently globally optimal particle, helping the algorithm escape local optima.

[0089] The algorithm terminates when the maximum number of iterations is reached (e.g., 500) or the optimal fitness value improves by less than 0.1% over 20 consecutive iterations. The position vector of the globally optimal particle is decoded, and the final power allocation weight matrix is ​​output. This matrix specifies the exact power each cell should handle under the current frequency modulation command (positive values ​​represent absorbed power, and negative values ​​represent released power).

[0090] according to The 32 cells with negative power allocation were divided into a set of equalization discharge nodes. The 32 battery cells with positive power allocation are divided into a set of equalization charging nodes. (This example assumes that the number of charging and discharging nodes is equal. In actual code and control logic, when encountering many-to-few or few-to-many situations, the system reconstructs a complete bipartite graph model by mapping virtual nodes to the set with fewer nodes, thereby ensuring that all cells with polarization differences can participate in active balancing in the current cycle.)

[0091] For any discharge node (Relaxation time) ) and charging nodes (Relaxation time) ), calculate the cost of its polarization response rate difference ,in This is the loss coefficient, used to convert the difference in polarization relaxation time constant into a dimensionless weight characterizing energy loss. In this embodiment, it is based on the transient charge-discharge heat loss calibration data of the energy storage cabinet under actual frequency regulation conditions. The empirical value range is set to 0.01. Up to 0.05 In this embodiment, .

[0092] The graph is divided into two parts, a discharge node and a charging node, and the edge weights take into account the factors from the discharge node and the charging node. Steady-state energy requirements extracted from , Cost of polarization coupling That is, total cost The Hungarian algorithm is used to solve the minimum weight perfect matching problem in this bipartite graph, yielding the optimal pairing set that minimizes the total energy transfer loss. For example, pairing a "slow" cell with a longer relaxation time with a "fast" cell with a shorter relaxation time can reduce energy loss during transient processes.

[0093] For a 32×32 bipartite graph in this embodiment, the Hungarian algorithm's single-step solution time is... On the level of processor approximately ,satisfy Real-time requirements for control cycle. When the total number of battery cells exceeds 200, the following can be adopted: The tree pre-sorts the polarization relaxation time constants, compressing the number of valid candidate pairs to a minimum. This reduces the computational burden by a significant margin.

[0094] For the optimal pairing relationship Each pair of cells in Calculate the difference in polarization relaxation time constants. Based on the difference Generate asymmetric pulse width compensation factor .in This is a proportionality coefficient. This compensation factor is injected into the underlying bidirectional active equalization circuit (such as an inductor-based equalization circuit). (circuit) Control loop. Specifically: If (If the discharge cell responds more slowly), then the conduction time of the upper bridge arm switch is extended, giving the inductor a longer charging time to compensate for the charge building process of the slow-response cell; if The conduction time of the lower bridge arm is adjusted accordingly. The control circuit generates an interlocked drive signal to precisely control the switching on and off of the switching transistor, achieving efficient and adaptive energy transfer from the discharge node to the charging node.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for coordinated control of multiple battery cells in an energy storage cabinet, characterized in that, include: Obtain the sampling data and polarization relaxation time constant of the cell node; The physical state constraint region is defined based on the sampled data as the search space of the chaotic particle swarm optimization algorithm. The particle position vector representing the candidate power allocation weight matrix is ​​initialized using chaotic mapping. The irreversible energy loss constant of a single charge-discharge cycle of the battery cell is obtained. An asymmetric inertial weight coefficient based on power state deviation is introduced into the velocity update iteration formula of the chaotic particle swarm optimization algorithm. The irreversible energy loss constant is used as the discrete integral weighting coefficient of the fitness function to verify the candidate power allocation weight matrix. For particles that are predicted to cause the battery cell to reach the thermal safety alarm threshold, fitness penalty processing is performed. The particles that pass the verification are perturbed using a chaotic sequence, and the power allocation weight matrix is ​​output. Based on the power allocation weight matrix, each cell node is divided into equal discharge nodes and equal charging nodes; a bipartite graph matching model is constructed by combining the polarization relaxation time constant to solve the optimal pairing relationship between equal discharge nodes and equal charging nodes. Based on the difference in polarization relaxation time constant between the balanced discharge node and the balanced charging node in the optimal pairing relationship, an asymmetric pulse width compensation factor is generated to drive the bidirectional active balancing circuit.

2. The method for multi-cell consistency and collaborative control of an energy storage cabinet according to claim 1, characterized in that, The sampling data includes the terminal voltage, charging and discharging current, surface temperature, and state of charge data of each cell node; a second-order spectral density model of the cell is established based on the sampling data. Equivalent circuit model, and identification of the circuit using recursive least squares method. The polarization relaxation time constant is calculated from the ohmic internal resistance, polarization internal resistance, and polarization capacitance in the equivalent circuit model; the physical state constraint region is jointly defined by the preset safe operating threshold of the terminal voltage, charging and discharging current, surface temperature, and state of charge data; the initial particle swarm is uniformly distributed within the search space defined by the safe operating threshold using chaotic mapping; the irreversible energy loss constant is obtained by performing standard charge and discharge cycle tests on the same batch of cells in advance, extracting the difference between the total charged energy and the total discharged energy, and calibrating the irreversible energy loss constant by combining the coulomb efficiency decay curves under different aging stages.

3. The method for multi-cell consistency and collaborative control of an energy storage cabinet according to claim 1, characterized in that, The introduction of the asymmetric inertia weight coefficient in the velocity update iterative formula of the chaotic particle swarm optimization algorithm includes: The velocity update iteration formula is configured as a vector superposition of inertia retention term, individual cognitive optimization term, and global social optimization term; the difference between the allocated power represented by the corresponding particle and the average allocated power of the system is extracted from the candidate power allocation weight matrix as the power state deviation; the asymmetric inertia weight coefficient is dynamically determined according to the numerical sign of the power state deviation: when the power state deviation is greater than zero, the first inertia weight is selected as the asymmetric inertia weight coefficient to reduce the inertia of the particle in maintaining its original divergence direction; when the power state deviation is not greater than zero, the second inertia weight is selected as the asymmetric inertia weight coefficient to maintain the global exploration capability of the particle, and the first inertia weight is less than the second inertia weight; the asymmetric inertia weight coefficient is introduced as a scaling factor into the inertia retention term and multiplied by the particle's current velocity vector.

4. The method for multi-cell consistency and collaborative control of an energy storage cabinet according to claim 1, characterized in that, The verification of the candidate power allocation weight matrix includes: constructing a fitness evaluation function with the objectives of minimizing the total energy loss during the equalization process and minimizing the cell charge-to-displacement ratio; in the fitness evaluation function, the irreversible energy loss constant is discretely integrated along a preset equalization time step to output a system energy loss characterization term; the state of charge is extracted from the sampled data of all cell nodes to calculate the system state variance as a consistency characterization term; an adaptive weight coefficient based on real-time dynamic adjustment of the charge-to-displacement ratio is introduced to perform a weighted summation of the system energy loss characterization term and the consistency characterization term to calculate the comprehensive fitness value of each particle; when the real-time charge-to-displacement ratio is not less than a preset range threshold, the weights of the consistency characterization term and the system energy loss characterization term are set to a first preset weight and a second preset weight, respectively, with the first preset weight being less than the second preset weight; when the real-time charge-to-displacement ratio is less than a preset range threshold, the weights of the consistency characterization term and the system energy loss characterization term are set to a third preset weight and a fourth preset weight, respectively, with the third preset weight being greater than the fourth preset weight.

5. The method for multi-cell consistency and collaborative control of an energy storage cabinet according to claim 4, characterized in that, The fitness penalty process includes: in each iteration, predicting the temperature rise of each cell within a preset time window based on the candidate power allocation weight matrix and the electrochemical heat generation model; when it is determined that the predicted temperature rise of any cell, when combined with the current node temperature, exceeds the thermal safety alarm threshold, triggering a hard constraint penalty mechanism, marking the corresponding particle as an out-of-bounds infeasible solution, and forcibly assigning a preset penalty constant to the comprehensive fitness value of the corresponding particle in the current iteration step.

6. The method for multi-cell consistency and collaborative control of an energy storage cabinet according to claim 1, characterized in that, Obtaining the optimal pairing relationship between the equalization discharge node and the equalization charging node includes: based on the power allocation weight matrix, classifying the cell nodes with negative power allocation as equalization discharge nodes and the cell nodes with positive power allocation as equalization charging nodes; based on the polarization relaxation time constants of the equalization discharge node and the equalization charging node, calculating the difference in polarization response rate between any equalization discharge node and any equalization charging node, and constructing a cross-polarization coupling cost matrix characterizing the degree of transient charge transfer compatibility between the two; and fusing the steady-state energy transfer demand extracted from the power allocation weight matrix with the cross-polarization coupling cost matrix to construct a bipartite graph matching model, and solving for the optimal pairing relationship with the minimum total energy transfer cost.

7. The method for multi-cell consistency and collaborative control of an energy storage cabinet according to claim 1, characterized in that, The method of generating an asymmetric pulse width compensation factor to drive the bidirectional active balancing circuit includes: generating an asymmetric pulse width compensation factor based on the difference between the polarization relaxation time constants of the equalization discharge node and the equalization charge node, adjusting the charging and discharging timing of the transient energy storage element in the bidirectional active balancing circuit; injecting the asymmetric pulse width compensation factor into the underlying control loop to generate an interlock drive signal, and using the interlock drive signal to alternately control the on and off states of the upper and lower bridge arm switches in the bidirectional active balancing circuit, thereby controlling the bidirectional active balancing circuit.

Images