Intelligent charging and discharging control method for energy storage cabinet

Abstract

The present application relates to the technical field of smart grid, in particular to a kind of energy storage cabinet intelligent charge-discharge control method.The present application obtains energy storage cabinet operating data, calculates irreversible entropy production rate and cost index based on the second law of thermodynamics and Newton cooling law;Based on the index, a non-cooperative game model containing grid dispatch compliance penalty factor is constructed, and the power allocation weight matrix of Nash equilibrium is solved and cached;When receiving macro instruction, evaluate the instantaneous entropy production rate jump slope and transient thermal elastic allowance;If it breaks the safety boundary, activate the defense mechanism to feed back the virtual limit to the upper layer, the bottom uses thermal elastic allowance to absorb the impact, and the cached matrix is called without communication to redirect high-frequency load to adjacent energy storage cabinet.The present application solves the problem of macro scheduling and physical boundary disconnection, significantly improves the life balance of heterogeneous cluster and the operation resilience under non-steady state condition.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, specifically to a smart charging and discharging control method for energy storage cabinets. Background Technology

[0002] Against the backdrop of the global energy structure transitioning towards cleaner energy, commercial and industrial energy storage systems play a crucial supporting role in peak shaving, demand management, and microgrid regulation. Existing energy storage cabinet charging and discharging control technologies have evolved from early reliance on rule-driven and static timetable control (such as setting fixed thresholds for charging and discharging arbitrage based on grid time-of-use pricing) to single-objective prediction and forward optimization (such as incorporating load forecasting data to maximize economic benefits). However, as energy storage applications develop towards higher frequency, higher capacity, and clustering, existing energy management systems have revealed deep-seated inherent flaws in their scheduling algorithms, primarily manifested in a severe disconnect between macro-level scheduling strategies and underlying physical characteristics.

[0003] First, at the microscopic electrochemical level, when energy storage systems participate in high-frequency bidirectional frequency regulation or smooth out drastically fluctuating microgrid loads, existing control technologies only limit current based on macroscopic state of charge and terminal voltage safety thresholds. This crude scheduling method completely fails to perceive changes in lithium-ion concentration gradients at the microscopic level within the battery. While the system appears to be operating within a "safe threshold," internal electrochemical polarization runaway and latent lithium plating often occur due to high-frequency switching, transforming macroscopic economic scheduling commands into irreversible microscopic physical damage to the battery cells. Second, at the cluster system architecture level, actual operating energy storage power stations generally consist of multiple energy storage cabinets of varying ages connected in parallel. Existing technologies, when dealing with zero-sequence circulating currents between cabinets caused by differences in internal resistance during discharge, generally adopt a passive suppression approach, relying on completely independent temperature control systems for each cabinet for heat dissipation. This "each fighting its own battle" island-style temperature control logic cannot coordinate the extremely uneven temperature distribution between heterogeneous cabinets, leading not only to a surge in ineffective internal friction of the thermal management system but also a high risk of system-level thermal runaway due to the "weakest link" effect. Finally, regarding the robustness of the algorithm model, current mainstream intelligent charging and discharging strategies heavily rely on purely data-driven AI models trained on historical operating data. When encountering real-world non-steady-state scenarios such as sudden extreme power loads, extreme weather, or grid topology reconfiguration, the underlying data distribution undergoes fundamental shifts. Lacking the rigid constraints of underlying physical thermodynamic laws and Kirchhoff's laws, pure black-box AI models are highly susceptible to catastrophic forgetting and predictive failures, subsequently issuing step-power jump commands that exceed the physical limits of the equipment.

[0004] In summary, how to solve the system-level scheduling instability and irreversible damage caused by the disconnect between the idealized expectations of macro-dispatch commands and the real-time physical boundaries of the underlying heterogeneous devices under complex and unsteady operating conditions is a technical problem that urgently needs to be solved in this field.

[0005] Therefore, an intelligent charging and discharging control method for energy storage cabinets is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent charging and discharging control method for energy storage cabinets. This invention acquires energy storage cabinet operating data, calculates irreversible entropy yield and cost index based on the second law of thermodynamics and Newton's law of cooling; constructs a non-cooperative game model with a grid regulation compliance penalty factor based on this index, solves and caches the power allocation weight matrix of the Nash equilibrium; upon receiving macroscopic commands, it evaluates the instantaneous entropy yield jump slope and transient thermoelastic margin; if the safety boundary is breached, a defense mechanism is activated to feed back virtual limiting to the upper layer, the lower layer absorbs the impact using the thermoelastic margin, and redirects high-frequency loads to adjacent energy storage cabinets without communication calls to the cache matrix. This invention solves the problem of the disconnect between macroscopic scheduling and physical boundaries, significantly improving the lifetime equilibrium and operational resilience of heterogeneous clusters under unsteady conditions.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A smart charging and discharging control method for energy storage cabinets, applied to an energy storage cluster comprising multiple heterogeneous energy storage cabinets, includes: The system acquires real-time thermoelectric operation data and coolant flow data of each energy storage cabinet, calculates the local thermodynamic irreversible entropy production rate based on the real-time operation data, and corrects the preset critical threshold of adiabatic temperature rise rate by combining the coolant flow data to generate the entropy increase loss cost index. Each energy storage cabinet obtains the entropy increase loss cost index of adjacent energy storage cabinets, constructs a non-cooperative game model that includes the single-unit lifetime loss target and the global network dispatch compliance penalty factor, solves the power allocation weight matrix that achieves Nash equilibrium, and caches it as a local plan lookup table to generate local fine-tuning instructions. The macro-grid control instructions issued by the energy management system are compared with the local fine-tuning instructions. The instantaneous entropy productivity jump slope when obeying the macro-grid control instructions is evaluated, and the transient thermoelastic margin is calculated based on the corrected critical threshold of the adiabatic temperature rise rate. If the instantaneous entropy production rate jumps above the physical safety boundary threshold, a virtual limiting signal is fed back to the energy management system. At the same time, the transient thermoelastic margin is used to absorb part of the power surge, and the redundant power command exceeding the safety boundary is separated into high-frequency load components. The local contingency plan is called to look up the table and the virtual impedance is dynamically increased. The high-frequency load components are redirected to the adjacent energy storage cabinet for physical absorption without communication triggering the bus shunt due to physical fluctuations.

[0008] Preferably, real-time thermoelectric operation data and coolant flow data of each energy storage cabinet are acquired, and the local thermodynamic irreversible entropy yield is calculated based on the real-time operation data. This includes: real-time acquisition of the surface temperature of the individual cells, the electrode temperature, and the temperature difference of the coolant flowing into and out of the energy storage cabinet; acquisition of the charging and discharging current and calculation of the real-time internal resistance using a preset equivalent circuit model; extraction of the irreversible heat generation rate based on the product of the square of the charging and discharging current and the real-time internal resistance; and, after strictly eliminating reversible reaction thermal interference determined by the reversible entropy change coefficient of the cells, the irreversible heat generation rate is compared with the real-time thermodynamic temperature to obtain the local thermodynamic irreversible entropy yield, which characterizes the degree of irreversible energy degradation inside the cells per unit time due to electrochemical reactions and physical heating.

[0009] Preferably, the preset adiabatic temperature rise rate critical threshold is corrected using Newton's law of cooling by combining coolant flow rate data to generate an entropy increase loss cost index. This includes: obtaining the convective heat transfer coefficient corresponding to the coolant flow rate and the effective heat transfer area between the cell surface and the coolant; obtaining the coolant inlet and outlet temperatures to calculate the effective characteristic temperature along the flow path; calculating the real-time equivalent temperature difference between the cell surface temperature and the effective characteristic temperature along the flow path; determining the real-time heat dissipation power of the current system by multiplying the convective heat transfer coefficient, the effective heat transfer area, and the real-time equivalent temperature difference according to Newton's law of cooling; converting the preset adiabatic temperature rise rate critical threshold into a reference adiabatic heat production power and linearly superimposing it with the real-time heat dissipation power to obtain the corrected safe heat production boundary; calculating the real-time allowable upper limit of entropy production rate based on the safe heat production boundary; and defining the ratio of the local thermodynamic irreversible entropy production rate to the real-time allowable upper limit of entropy production rate as the entropy increase loss cost index.

[0010] Preferably, constructing a non-cooperative game model that includes a single-unit lifetime loss target and a global grid regulation compliance penalty factor specifically includes: setting each energy storage cabinet as an independent game agent; the single-unit lifetime loss target of each agent is defined as minimizing the equivalent power decay rate within a unit step; the equivalent power decay rate is the product of the exponential decay value of the entropy increase loss cost index and the deep stress penalty coefficient, the deep stress penalty coefficient increasing asymmetrically as the state of charge approaches the charging / discharging cutoff interval; introducing a global grid regulation compliance penalty factor into the objective function; the global grid regulation compliance penalty factor is defined as the absolute value of the difference between the actual response power of the single unit and the power allocated by the macro grid regulation command, multiplied by a weighting coefficient dynamically adjusted according to the grid frequency regulation pressure; each agent interacts with the cost index to iteratively adjust the power weight; the iteration is truncated when the marginal benefit of the reduction in the single unit decay rate equals the marginal penalty of deviating from the command, and the output power allocation share is used as the Nash equilibrium solution.

[0011] Preferably, the calculation of the power allocation weight matrix to achieve Nash equilibrium and its caching as a local plan lookup table includes: employing a distributed iterative optimization algorithm to calculate the power allocation share of each agent in different charge state intervals, health state intervals, and the entropy increase loss cost index interval, generating a power allocation weight matrix; storing the power allocation weight matrix in the local controller flash memory of each energy storage cabinet in the form of static structured data, forming the local plan lookup table; the local plan lookup table is associated with dynamic control offsets, and when the communication link is interrupted or local computing is limited, the power allocation weights are retrieved and corrected locally based on real-time parameters.

[0012] Preferably, evaluating the instantaneous entropy production rate jump slope when obeying macro-grid control commands, and calculating the transient thermoelastic margin based on the corrected adiabatic temperature rise rate critical threshold, includes: calculating the power change rate by performing a first derivative calculation on the macro-grid control commands, and extracting the dynamic polarization resistance in the equivalent circuit model that matches the power change rate frequency band; calculating the transient heat generation mutation value using the power change rate and the dynamic polarization resistance, and dividing the transient heat generation mutation value by the real-time absolute temperature to obtain the instantaneous entropy production rate jump slope; obtaining the product of the cell specific heat capacity and the cell mass as the system thermal capacity constant; dividing the corrected safe heat generation boundary by the system thermal capacity constant to obtain the maximum allowable temperature rise rate; integrating the maximum allowable temperature rise rate within a preset time window to obtain the available temperature difference window; and multiplying the system thermal capacity constant by the available temperature difference window to obtain the transient thermoelastic margin characterizing the instantaneous energy absorption extreme value.

[0013] Preferably, redundant power commands exceeding the safety boundary are separated into high-frequency load components. A local pre-set table is consulted, and the virtual impedance is dynamically increased to allow physical fluctuations to trigger bus shunting without communication. The high-frequency load components are then redirected to adjacent energy storage cabinets for physical absorption. This includes: using a digital low-pass filter to perform frequency domain decomposition on the redundant power commands exceeding the safety boundary, defining fluctuation components with frequencies higher than a preset cutoff frequency threshold as high-frequency load components; when the transient thermoelastic margin of the local energy storage cabinet is determined to be lower than a preset safety lower limit threshold, the high-frequency virtual impedance of the local cabinet is instantaneously increased in the control loop based on a local pre-set table lookup to flexibly reject high-frequency load components and induce physical fluctuations on the shared bus; simultaneously, the corresponding adjacent energy storage cabinets in the cluster synchronously sense physical fluctuations based on the same local pre-set table lookup and instantaneously reduce their own high-frequency virtual impedance, guiding the actual high-frequency surge corresponding to the high-frequency load components to automatically shunt and redirect on the physical line into the adjacent energy storage cabinet for physical absorption.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention introduces the local thermodynamic irreversible entropy yield and combines it with a dynamic boundary modified based on Newton's law of cooling. This allows for the quantification of internal physical losses within the battery cell from the fundamental level of energy dissipation. This approach overcomes the limitations of relying solely on electrical parameters such as state of charge or terminal voltage, enabling a deep understanding of the health status of individual energy storage cells under complex temperature control environments. This provides a more valuable physical criterion for subsequent refined power allocation.

[0015] 2. This invention utilizes a non-cooperative game model that incorporates individual lifespan loss and a global compliance penalty factor. This model can comprehensively consider the individual health needs of each energy storage unit and the macro-level dispatch objectives of the power grid. By solving and caching the weight matrix under Nash equilibrium, the system can guide heterogeneous units to achieve Pareto improvement in response share without relying on real-time centralized decision-making. This effectively suppresses the "bottleneck effect" in cluster operation and promotes the stable realization of the economic benefits throughout the entire lifecycle of the energy storage system.

[0016] 3. Based on the calculation of transient thermoelastic margin and the redirection mechanism of high-frequency load components, this invention endows the underlying hardware with physical buffering capabilities when facing sudden power jumps. Without interrupting the macroscopic response, it achieves spatiotemporal reconstruction of the load within the cluster through communication-free local pre-plan invocation. This not only mitigates the transient thermal stress impact on local cells but also reduces the system's dependence on the real-time performance of external communication links, ensuring the safe and continuous operation of the energy storage cluster under non-steady-state conditions. Attached Figure Description

[0017] Figure 1 A flowchart of an intelligent charging and discharging control method for an energy storage cabinet provided in an embodiment of the present invention; Figure 2 This is a flowchart of the underlying flexible unloading and thermoelectric synergistic resistance mechanism provided in an embodiment of the present invention; Figure 3 The flowchart for non-cooperative game theory and local contingency plan generation provided in this embodiment of the invention is as follows. Detailed Implementation

[0018] 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.

[0019] Please see Figures 1 to 3 This invention provides an intelligent charging and discharging control method for energy storage cabinets, the technical solution of which is as follows: A smart charging and discharging control method for energy storage cabinets, applied to an energy storage cluster comprising multiple heterogeneous energy storage cabinets, includes: The system acquires real-time thermoelectric operation data and coolant flow data of each energy storage cabinet, calculates the local thermodynamic irreversible entropy production rate based on the real-time operation data, and corrects the preset critical threshold of adiabatic temperature rise rate by combining the coolant flow data to generate the entropy increase loss cost index. Each energy storage cabinet obtains the entropy increase loss cost index of adjacent energy storage cabinets, constructs a non-cooperative game model that includes the single-unit lifetime loss target and the global network dispatch compliance penalty factor, solves the power allocation weight matrix that achieves Nash equilibrium, and caches it as a local plan lookup table to generate local fine-tuning instructions. The macro-grid control instructions issued by the energy management system are compared with the local fine-tuning instructions. The instantaneous entropy productivity jump slope when obeying the macro-grid control instructions is evaluated, and the transient thermoelastic margin is calculated based on the corrected critical threshold of the adiabatic temperature rise rate. If the instantaneous entropy production rate jumps above the physical safety boundary threshold, a virtual limiting signal is fed back to the energy management system. At the same time, the transient thermoelastic margin is used to absorb part of the power surge, and the redundant power command exceeding the safety boundary is separated into high-frequency load components. The local contingency plan is called to look up the table and the virtual impedance is dynamically increased. The high-frequency load components are redirected to the adjacent energy storage cabinet for physical absorption without communication triggering the bus shunt due to physical fluctuations.

[0020] Example 1: This embodiment applies to a scenario where industrial and commercial microgrid energy storage power stations participate in high-frequency grid regulation. The background problem in this scenario is that the energy storage cluster needs to frequently respond to secondary frequency regulation or millisecond-level power jump commands issued by the grid, and the heterogeneous characteristics of the energy storage cabinets within the cluster are significant. Conventional centralized dispatching only allocates power based on the state of charge. When faced with sudden high-frequency load surges, this often leads to a sharp rise in the internal temperature of some aging energy storage cabinets, resulting in localized polarization heating and latent lithium plating. This not only causes ineffective internal friction in the thermal management system but also increases the risk of system-level thermal runaway and collapse.

[0021] As one embodiment of the present invention, refer to Figure 1 A flowchart of an intelligent charging and discharging control method for an energy storage cabinet, referring to... Figure 2 The flowchart of the underlying flexible unloading and thermoelectric synergistic resistance mechanism is shown below. Figure 3 Non-cooperative game theory and local contingency plan generation sub-flowchart.

[0022] Furthermore, real-time thermoelectric operation data and coolant flow data of each energy storage cabinet are acquired. Based on the real-time operation data, the local thermodynamic irreversible entropy yield is calculated, including: real-time acquisition of the surface temperature of the individual cells, the electrode temperature, and the temperature difference of the coolant flowing into and out of the energy storage cabinet; acquisition of the charging and discharging current and calculation of the real-time internal resistance in combination with a preset equivalent circuit model; extraction of the irreversible heat generation rate based on the product of the square of the charging and discharging current and the real-time internal resistance; and after strictly eliminating the reversible reaction thermal interference determined by the reversible entropy change coefficient of the cell, the irreversible heat generation rate is compared with the real-time thermodynamic temperature to obtain the local thermodynamic irreversible entropy yield, which characterizes the degree of irreversible energy degradation inside the cell due to electrochemical reactions and physical heating per unit time.

[0023] Specifically, at the hardware sensing layer of the energy storage cabinet, a multi-point temperature sensor network deployed on the surface of the battery cells and the positive and negative terminals collects individual cell temperature data in real time at a specific sampling period. Simultaneously, an ultrasonic flow meter acquires the coolant flow rate and temperature difference parameters entering and exiting the liquid cooling plate of the energy storage cabinet. The real-time charging and discharging current collected by a high-precision Hall current sensor is simultaneously read and input into a pre-established second-order reaction equivalent circuit model. The controller uses parameter identification algorithms such as recursive least squares to separate the current ohmic resistance, electrochemical polarization resistance, and concentration polarization resistance of the battery cell online, and then summarizes them into a real-time internal resistance. Subsequently, the controller squares the real-time collected current value and multiplies it by the real-time internal resistance to obtain the real-time irreversible thermal power, including Joule heat and polarization heat. At the same time, to accurately isolate thermodynamic losses, the inherent reversible entropy change coefficient dictionary of the battery cell material is called, and it is multiplied by the charging and discharging current and the current absolute thermodynamic temperature to obtain the dynamic reversible thermal power of the internal chemical reaction of the battery cell. The real-time irreversible thermal power is added to the dynamic reversible thermal power to obtain the real-time total heat generation power, which serves as the benchmark data for subsequent evaluation of the adiabatic temperature rise rate and system heat dissipation requirements. When quantifying the degree of physical degradation, the reversible thermal power component that does not cause physical damage is strictly eliminated, and only the real-time irreversible thermal power value is divided by the absolute thermodynamic temperature value measured by the current sensor network to obtain a continuous scalar flow, namely the local thermodynamic irreversible entropy yield, which is used to continuously reflect the physical degradation rate of the battery cell.

[0024] This invention maps fundamental electrical sensing data to a thermodynamic dimension, constructing a calculation process for local thermodynamic irreversible entropy yield. This scheme enables energy management systems to abstract away apparent voltage fluctuations and directly quantify the dissipation and irreversible degradation of energy within the battery cell. This provides a more objective data foundation that reflects the true physical losses for subsequent refined power allocation, effectively improving the accuracy of state perception.

[0025] In the process of identifying the real-time internal resistance including concentration polarization resistance based on the equivalent circuit model, when the growth slope of the concentration polarization resistance is identified to exceed the steady-state threshold, the control system superimposes an asymmetric high-frequency depolarization perturbation pulse on the base current of the macro-grid adjustment command to actively accelerate the ion relaxation diffusion at the solid-liquid interface through electric field oscillation; subsequently, based on the real ohm and polarization resistance data that fall back after the action of the high-frequency depolarization perturbation pulse, the local thermodynamic irreversible entropy production rate is dynamically refreshed.

[0026] The steady-state threshold for the concentration polarization internal resistance growth slope is determined based on the extreme value of the normal diffusion slope of the Weber impedance in the low-frequency region during standard electrochemical impedance spectroscopy testing of the same batch of cells. In practice, when the real-time internal resistance slope calculated by the controller exceeds the extreme value of this normal diffusion slope for several consecutive sampling cycles, and the depth stress penalty coefficient increases asymmetrically with the state of charge approaching the charge-discharge cutoff interval by 1.5 times to 2.0 times, it can be determined that there is abnormal accumulation of bottom-layer ions, thus precisely triggering the subsequent depolarization intervention mechanism.

[0027] Specifically, when the energy storage cabinet executes macro-grid control commands such as continuous high-rate discharge, a large number of ions accumulate inside the battery cell, causing the identified concentration polarization internal resistance to increase non-linearly and rapidly. Once this growth rate exceeds the preset steady-state threshold, it is determined that a severe concentration polarization barrier is currently in place. At this time, the energy storage converter actively initiates an intervention mechanism, superimposing a millisecond-level asymmetric high-frequency charging perturbation pulse onto the basic discharge DC current without changing the direction of macro-discharge power. The alternating electric field formed by this pulse can oscillate the double layer of the solid-liquid phase interface at the microscale, actively interrupting the directional accumulation of ions and forcibly accelerating the short-range relaxation and diffusion of ions. After the perturbation pulse ends, the terminal voltage and current are collected again. The internal resistance data measured at this time has eliminated the artificially high components of concentration polarization and represents the true ohmic impedance of the battery cell. Then, the true internal resistance data after this decline is used to dynamically update the local thermodynamic irreversible entropy production rate.

[0028] This invention addresses the issues of artificially high internal resistance and misjudgment of state caused by concentration polarization accumulation in energy storage cells during prolonged high-rate operation by introducing active perturbation depolarization pulses at the underlying sensing and identification stages. This technique enables the upper-level energy management system to penetrate the false polarization barrier and obtain the most accurate physical boundary data of the underlying cells, eliminating the distortion in cost index assessment caused by steady-state concentration polarization accumulation. It prevents unnecessary power reduction shutdowns or erroneous triggering of underlying resistance mechanisms, ensuring the accurate implementation of macro-level scheduling expectations.

[0029] Furthermore, by combining coolant flow rate data, the preset adiabatic temperature rise rate critical threshold is corrected using Newton's law of cooling to generate an entropy increase loss cost index. This includes: obtaining the convective heat transfer coefficient corresponding to the coolant flow rate and the effective heat transfer area between the cell surface and the coolant; obtaining the coolant inlet and outlet temperatures to calculate the effective characteristic temperature along the flow path; calculating the real-time equivalent temperature difference between the cell surface temperature and the effective characteristic temperature along the flow path; determining the real-time heat dissipation power of the current system by multiplying the convective heat transfer coefficient, the effective heat transfer area, and the real-time equivalent temperature difference according to Newton's law of cooling; converting the preset adiabatic temperature rise rate critical threshold into a benchmark adiabatic heat generation power and linearly superimposing it with the real-time heat dissipation power to obtain the corrected safe heat generation boundary; calculating the real-time allowable upper limit of entropy production rate based on the safe heat generation boundary; and defining the ratio of the local thermodynamic irreversible entropy production rate to the real-time allowable upper limit of entropy production rate as the entropy increase loss cost index.

[0030] The preset critical threshold for adiabatic temperature rise rate is determined based on the self-generated heat initiation temperature and corresponding temperature rise inflection point slope of the same batch of cells under accelerated calorimeter adiabatic testing conditions. Taking the mainstream industrial and commercial lithium iron phosphate cells used in this embodiment as an example, in order to ensure that an absolute safety red line is drawn before the early characteristics of thermal runaway appear, this critical threshold is usually calibrated and set to a value between the following: the depth stress penalty coefficient increases asymmetrically with the state of charge approaching the charge-discharge cutoff interval by 0.02℃ / min; the depth stress penalty coefficient increases asymmetrically with the state of charge approaching the charge-discharge cutoff interval by 0.05℃ / min; and the depth stress penalty coefficient increases asymmetrically with the state of charge approaching the charge-discharge cutoff interval.

[0031] Specifically, the bottom-level controller of the energy storage cabinet extracts the values ​​from the coolant flow sensor in real time and, combined with the pipe topology of the liquid cooling plate and fluid dynamic constants, maps the current convective heat transfer coefficient. Simultaneously, the system reads the effective heat transfer area parameters between the battery cell casing and the cooling medium. To eliminate temperature rise drift errors caused by heat absorption along the flow path, the system simultaneously collects the inlet and outlet temperatures of the coolant and calculates the effective characteristic temperature along the flow path of the coolant as it flows through the entire heat transfer surface using either a logarithmic or arithmetic mean temperature difference algorithm. Subsequently, the system calculates the real-time equivalent temperature difference between the real-time collected battery cell surface temperature and this effective characteristic temperature along the flow path. The controller uses Newton's law of cooling to multiply the convective heat transfer coefficient, the effective heat transfer area, and this real-time equivalent temperature difference to obtain the real-time heat dissipation power of the system under current operating conditions. To define a safe physical boundary, the pre-set critical threshold for adiabatic temperature rise rate is multiplied by the specific heat capacity and mass parameters of the battery cell to convert it into a baseline adiabatic heat generation power. Subsequently, this baseline adiabatic heat production power is linearly superimposed with the real-time heat dissipation power calculated above to obtain a dynamically broadened safe heat production boundary. The controller then derives the maximum permissible entropy yield scale under this boundary. Finally, the local thermodynamic irreversible entropy yield measured in the previous step is divided by this permissible upper limit; the resulting ratio is the entropy increase loss cost index, a dynamic evaluation indicator that fluctuates in real time with cooling conditions.

[0032] This invention utilizes the physical laws of heat transfer to transform static temperature alarm thresholds into dynamic safety boundaries, achieving deep decoupling between cost assessment and underlying thermal management status. This method can dynamically relax or tighten the available power limits of energy storage units based on the actual heat dissipation capacity of the pumps or fans, avoiding conservative shutdowns caused by traditional fixed thresholds and improving the capacity utilization and environmental adaptability of the energy storage system.

[0033] Furthermore, the construction of a non-cooperative game model that includes a single-unit lifetime loss target and a global grid regulation compliance penalty factor specifically includes: setting each energy storage cabinet as an independent game agent, with the single-unit lifetime loss target of each agent defined as minimizing the equivalent energy decay rate within a unit step; the equivalent energy decay rate is the product of the exponential decay value of the entropy increase loss cost index and the deep stress penalty coefficient, the deep stress penalty coefficient increasing asymmetrically as the state of charge approaches the charging / discharging cutoff interval; a global grid regulation compliance penalty factor is introduced into the objective function, defined as the absolute value of the difference between the actual response power of the single unit and the power allocated by the macro grid regulation command, multiplied by a weighting coefficient dynamically adjusted with the grid frequency regulation pressure; each agent interacts with the cost index to iteratively adjust the power weight, and the iteration is truncated when the marginal benefit of the reduction in the single unit decay rate equals the marginal penalty of deviating from the command, and the output power allocation share is taken as the Nash equilibrium solution.

[0034] The weighting coefficient, which is dynamically adjusted according to the grid frequency regulation pressure, is essentially a hard constraint factor used to adjust the strength of the power grid support in real time. Its value is based on the absolute value and rate of change of the grid frequency deviation observed in real time by the local converter.

[0035] The controller has a pre-defined piecewise function or logical stochastic mapping function: Steady-state range: When the absolute value of the grid frequency deviation is within a preset dead zone (e.g., the deep stress penalty coefficient increases asymmetrically with the state of charge approaching the charge-discharge cutoff range by 0.05Hz), the weighting coefficient takes a minimum benchmark value. At this time, the game model tends to protect the battery life and allows individual cells to make self-generated power fine-tuning based on their own entropy increase cost index.

[0036] Emergency frequency regulation range: When the absolute value of the grid frequency deviation, the depth stress penalty coefficient increases asymmetrically as the state of charge approaches the charging and discharging cutoff range, breaks through the dead zone and continues to increase, or when the absolute value of the rate of change exceeds the preset sudden change threshold, the weighting coefficient increases nonlinearly with the increase of the frequency deviation (such as a quadratic relationship or an exponential relationship).

[0037] Through this dynamic mapping, the greater the pressure on the power grid frequency regulation, the higher the proportion of the penalty cost caused by deviation from the command in the objective function, forcibly guiding each agent into a cooperative state of "sacrificing lifespan and prioritizing grid protection", thereby ensuring the ultimate response to macroscopic commands within the microscopic physical security boundary.

[0038] Specifically, at the cluster network layer, each physical energy storage cabinet is instantiated as a game-theoretic agent with independent decision-making capabilities. Each agent internally runs an objective function aimed at minimizing its equivalent power decay rate within a single control cycle. The specific configuration logic for this decay rate is as follows: extract the dynamically generated entropy increase loss cost index, use it as the base to perform mapping calculations of the exponential decay function, and then multiply the output by a depth stress penalty coefficient. This penalty coefficient is dynamically selected based on the current state of charge (SOC) position within the range. When the SOC is in the normal operating range (e.g., between 20% and 80%), the coefficient changes gradually; when the SOC approaches the charging / discharging limit range, the coefficient exhibits a non-linear, rapidly increasing trend, thus limiting the dispatch output under extreme conditions. Simultaneously, to ensure the cluster's overall ability to follow grid commands, a global grid dispatch compliance penalty factor is forcibly superimposed on the agent's objective function. This factor is calculated by extracting the absolute difference between the actual response power command of the individual unit and the power allocated by the macro-grid dispatch command issued by the upper layer, and multiplying this difference by a dynamic weighting coefficient reflecting the depth of grid frequency drop.

[0039] In the iterative loop, each agent broadcasts and receives the cost index of adjacent cabinets via the local area network, continuously fine-tuning its own power allocation weight variables. The specific truncation judgment logic is as follows: the first-order partial derivative of the power weight variable currently being attempted to be allocated is solved using the equivalent power attenuation rate, and the absolute value of the output derivative is defined as the marginal benefit of the reduction in the individual attenuation rate; simultaneously, the first-order partial derivative of the same power weight variable is solved using the global network regulation compliance penalty factor, and the absolute value of its slope is obtained as the marginal penalty for deviating from the instruction; to eliminate the difference in physical dimensions, a dimensionless normalization transformation operator is introduced, and the marginal benefit and the marginal penalty are mapped to the same numerical reference surface and then compared by difference; when the absolute value of the difference between the two normalized values ​​converges to the preset minimum constant tolerance range, the iterative loop triggers the truncation condition, locking the current weight matrix as the Nash equilibrium solution.

[0040] The preset minimum constant tolerance range is designed to avoid optimization loops caused by truncation errors when the underlying controller performs floating-point calculations. Its lower limit is constrained by the sampling accuracy of the local controller's analog-to-digital converter and the chip's floating-point word length (typically not less than 1e-4, where the deep stress penalty coefficient increases asymmetrically as the state of charge approaches the charging / discharging cutoff interval). Its upper limit depends on the grid's tolerance for power dispatch errors, thus ensuring both the accuracy of Nash equilibrium allocation and the rapid convergence of the distributed iterative optimization algorithm within a finite millisecond control cycle.

[0041] This invention innovatively constructs a non-cooperative game architecture based on marginal benefit truncation, enabling multiple heterogeneous energy storage units to spontaneously achieve Pareto optimal power allocation in the absence of central coordination. This mechanism not only effectively curbs the accelerated degradation of vulnerable units in the cluster due to overload operation, but also takes into account the overall response requirements of grid dispatch through a dynamic penalty mechanism, achieving a balance between micro-level security and macro-level compliance.

[0042] The global network regulation compliance penalty factor introduced in the objective function also includes a nested network topology impedance weight matrix. The control system injects a high-frequency probe carrier through the bottom converter to measure the parasitic inductance values ​​of the physical transmission bus between the local energy storage cabinet and each adjacent energy storage cabinet in real time, so as to generate the network topology impedance weight matrix. When solving the charging and discharging power allocation weight matrix of Nash equilibrium, the power transfer allocation weight is forced to tilt towards the short-distance physical return path with the minimum bus parasitic inductance.

[0043] Specifically, to prevent the power allocation scheme given by the game theory model from causing electromagnetic shocks at the physical execution level, a weak probe carrier of a specific frequency is actively injected into the DC microgrid or AC bus using the high-frequency pulse width modulation module of the underlying energy storage converter. By detecting the voltage response amplitude and phase lag of each adjacent cabinet terminal to this carrier, the parasitic inductance of the connecting copper busbar or cable between the local energy storage cabinet and each adjacent energy storage cabinet is accurately calculated, and these values ​​are summarized to generate a network topology impedance weight matrix. In the iterative solution of the non-cooperative game, this matrix is ​​nested as a pre-filtering damping term in the global network regulation compliance penalty factor. When a game theory algorithm attempts to transfer high-frequency transient power to a distant neighbor cabinet with a lower cost exponent, the penalty factor will be amplified sharply due to the large parasitic inductance of its transmission line. This forces the game theory algorithm to abandon the distant target and instead force the power allocation weight to the near cabinet with the shortest electrical connection distance and the smallest parasitic inductance. This ensures that the output Nash equilibrium solution is not only optimal in terms of lifetime loss, but also has absolute feasibility in terms of physical and electromagnetic safety.

[0044] This invention directly embeds the underlying hardware topology features into the upper-level game theory model, solving the instability problem caused by pure software scheduling algorithms ignoring the boundaries of the underlying electrical hardware during power redistribution. By injecting probe carriers to obtain parasitic inductance, high-frequency commands are forced to flow on the path with the shortest physical distance and the least parasitic inductance, effectively suppressing high-voltage spikes generated on the bus during rapid current redirection within the cluster. This avoids penetrating damage to the insulation system and power electronic devices caused by overvoltage, achieving a deep integration of software computing power and physical electromagnetic laws.

[0045] Furthermore, the power allocation weight matrix that achieves Nash equilibrium is calculated and cached as a local plan lookup table. This includes: using a distributed iterative optimization algorithm to calculate the power allocation share of each agent in different charge state intervals, health state intervals, and the entropy increase loss cost index interval, generating a power allocation weight matrix; storing the power allocation weight matrix in the local controller flash memory of each energy storage cabinet in the form of static structured data, forming the local plan lookup table; the local plan lookup table is associated with dynamic control offsets, and when the communication link is interrupted or local computing is limited, the power allocation weights are retrieved and corrected locally based on real-time parameters.

[0046] Specifically, during idle periods or background operation of the energy storage power station, the cluster master node or edge computing gateway traverses the typical state space that the energy storage cabinets may be in. The charged state interval is discretized into specific step sizes (e.g., 5% per step), and the healthy state interval and the entropy increase loss cost index interval are similarly gradient-divided. At each discrete cross-state point, a distributed iterative optimization algorithm is invoked to pre-calculate the Nash equilibrium power allocation share that each agent should bear when facing a standard step load. This associated data, containing multi-dimensional input parameters and corresponding output weights, is organized into a high-dimensional data matrix structure. Subsequently, this structured matrix is ​​distributed to the local controller of each energy storage cabinet via the underlying communication bus. The local controller uses a file system to permanently store it in non-volatile flash memory, thereby establishing a completely static local pre-plan lookup table. In actual operation, once the inter-cabinet communication delay timeout or upper-level network interruption is detected, the local controller can bypass the complex online calculation process, directly read the current sensor status parameters, and accurately extract the matching preset power weights within a few milliseconds for instruction fine-tuning through multi-dimensional array addressing.

[0047] This invention constructs a low-level emergency addressing mechanism independent of communication links by combining offline computation with local static caching. This method significantly reduces the algorithm computation delay when the microgrid faces sudden operating conditions, ensures that the low-level controller can maintain high-confidence collaborative operation even in extremely harsh environments, and significantly improves the anti-interference capability and scheduling reliability of the entire energy storage power station system.

[0048] Further, the evaluation of the instantaneous entropy production rate jump slope when obeying macro-grid control commands, and the calculation of transient thermoelastic margin based on the corrected adiabatic temperature rise rate critical threshold, includes: calculating the power change rate by performing a first derivative calculation on the macro-grid control commands, and extracting the dynamic polarization resistance in the equivalent circuit model that matches the power change rate frequency band; calculating the transient heat generation mutation value using the power change rate and the dynamic polarization resistance, and dividing the transient heat generation mutation value by the real-time absolute temperature to obtain the instantaneous entropy production rate jump slope; obtaining the product of the cell specific heat capacity and the cell mass as the system thermal capacity constant; dividing the corrected safe heat generation boundary by the system thermal capacity constant to obtain the maximum allowable temperature rise rate; integrating the maximum allowable temperature rise rate within a preset time window to obtain the available temperature difference window; and multiplying the system thermal capacity constant by the available temperature difference window to obtain the transient thermoelastic margin characterizing the instantaneous energy absorption extreme value.

[0049] The length of the preset time window needs to match the macro-schedule update cycle of the upper-level energy management system (for example, the common microgrid secondary frequency regulation command issuance cycle, which is usually set to 2 seconds to 4 seconds as the deep stress penalty coefficient approaches the charge-discharge cutoff interval, in order to assess the maximum temperature rise range that the system can withstand before the next frame command from the power grid arrives).

[0050] Specifically, when the energy management system issues a new power command set, the underlying converter controller performs first-order partial derivative calculations on the received macro-grid dispatch command sequence within a very short time window (e.g., within 100 milliseconds) to accurately quantify the instantaneous jump slope or ramp rate of the power command. Since step loads contain a large number of high-frequency harmonics in the frequency domain, the equivalent circuit model is linked to extract the dynamic polarization resistance parameters corresponding to the frequency band of the power change rate. The controller calculates the extracted command jump amplitude and the dynamic polarization resistance parameters to deduce the predicted value of the impending transient heat generation mutation. To assess the damage rate of this mutation to the system, the predicted heat generation mutation value is divided by the current absolute temperature of the battery cell to obtain the instantaneous entropy productivity jump slope, characterizing the rate of damage deterioration. On the defensive side, the statically configured cell specific heat capacity and single-cell mass parameters in the battery management system are read, and their multiplication establishes the system thermal constant of the current cell cluster. Subsequently, the physical extreme temperature corresponding to the corrected adiabatic temperature rise boundary is obtained, and the real-time measured surface temperature is subtracted from this extreme temperature to obtain the currently available safe temperature difference window. Finally, the controller multiplies the system thermal constant by the available temperature difference window to calculate a specific Joule-level energy value. This value is the transient thermoelastic margin, which characterizes the upper limit of pulse energy that the battery cell can forcibly absorb before the temperature exceeds the safety red line.

[0051] This invention breaks through the limitations of traditional scheduling that only considers static thresholds, and for the first time deeply couples the frequency domain characteristics of power jumps with thermodynamic physical constants in calculation. By accurately quantifying the transient thermoelastic margin, a physical tolerance scale capable of withstanding short-term high-frequency electric shocks is obtained, enabling the energy management system to predict and utilize the thermal inertia of the battery itself to buffer grid impacts, thus optimizing the lead time for safety prevention and control.

[0052] During the process of absorbing the power impact of macro-grid regulation using transient thermal elasticity margin, when it is determined that the transient thermal elasticity margin of the current energy storage cabinet is close to the preset safety lower limit, the frequency converter switching loss transfer strategy is activated: the control system dynamically reduces the switching frequency of the power switching module inside the energy storage converter to reduce the ramp rate of the terminal current, converting the polarization heating stress that should be borne by the battery cell into the voltage fluctuation load of the DC bus capacitor of the energy storage converter, and using the independent liquid cooling heat dissipation system of the energy storage converter to absorb the transient heat load in physical space, so as to dynamically extend the transient thermal elasticity safety time window of the battery cell.

[0053] Specifically, as the energy storage cabinet utilizes its own thermal capacity to withstand high-frequency grid surges, the controller continuously reduces the transient thermal elasticity margin. When this margin drops to a preset extremely low safety limit, it means the cell surface temperature is about to exceed the physical limit for adiabatic temperature rise. At this point, to avoid triggering a hard trip of the circuit breaker, the controller immediately sends a frequency reduction command to the underlying energy storage converter, dynamically and significantly reducing the carrier switching frequency of the power switching modules such as insulated-gate bipolar transistors inside the converter. The sudden drop in switching frequency forces a physical hysteresis in the ramp-up rate of the terminal current output to the cell side, thereby directly reducing the transient polarization heating caused by the drastic current changes inside the cell. At the same time, excess transient energy in the macro-grid dispatch command cannot immediately flow into the cell, but instead accumulates in the form of voltage boost on the large DC bus thin-film capacitor bank inside the converter. Since converter modules are typically equipped with independent and extremely high-velocity forced liquid cooling or air cooling channels, the stored energy and the high heat loss generated by low-frequency switching are quickly dissipated by the converter's cooling system in another physical space within the cabinet.

[0054] This invention breaks through the traditional thermal management barrier where battery cells and converters operate independently in energy storage cabinets, resolving the core contradiction of system avalanche caused by the depletion of the internal thermal buffer capacity of battery cells under extreme operating conditions. Through active frequency reduction technology, the electrochemical polarization heat stress that would otherwise impact the battery cells is laterally shifted to the energy storage converter side, which has greater heat dissipation redundancy, via electromagnetic and capacitor voltage coupling. This provides extremely valuable flexible redirection time for upper-level scheduling, greatly enhancing the system's resilience at extreme limits.

[0055] Furthermore, the redundant power command exceeding the safety boundary is separated into high-frequency load components. The local contingency plan is called to look up the table and the virtual impedance is dynamically increased. The bus shunt is triggered by physical fluctuations without communication. The high-frequency load components are redirected to the adjacent energy storage cabinet for physical absorption. This includes: using a digital low-pass filter to perform frequency domain decomposition on the redundant power command exceeding the safety boundary, and defining the fluctuation components with frequencies higher than the preset cutoff frequency threshold as high-frequency load components. The preset cutoff frequency threshold is dynamically tuned based on the reciprocal of the electrochemical polarization time constant extracted from the equivalent circuit model, so as to accurately remove transient high-frequency abrupt components that cannot be physically followed by the ion diffusion rate at the solid-liquid interface inside the cell. When the transient thermoelastic margin of the local energy storage cabinet is determined to be lower than the preset safety lower limit threshold, the high-frequency virtual impedance of the local cabinet is instantaneously increased in the control loop directly according to the local plan table, so as to flexibly reject high-frequency load components and induce physical fluctuations on the shared bus. The preset safety lower limit threshold of the transient thermoelastic margin is set based on the maximum mechanical and hydraulic delay time required for the liquid cooling system pump to establish a steady-state heat exchange from receiving the full-speed heat dissipation command to the coolant flow rate in the pipeline (in this embodiment, the system calibration delay is from 2 seconds to 3 seconds as the deep stress penalty coefficient asymmetrically increases with the state of charge approaching the charge-discharge cutoff interval). The Joule energy corresponding to this lower limit threshold must be strictly tuned to be greater than the total heat generated by the local battery cell continuously discharging at the maximum allowable power within this delay time window, thereby ensuring that the battery cell will not undergo irreversible physical thermal penetration due to the depletion of its own thermal capacity in the 'thermal hysteresis' blind zone before the cooling system is fully operational.

[0056] Simultaneously, adjacent energy storage cabinets in the cluster synchronously sense physical fluctuations based on the same local plan and instantly reduce their own high-frequency virtual impedance, guiding the real high-frequency surge corresponding to the high-frequency load component to automatically divert and redirect on the physical line into the adjacent energy storage cabinet for physical absorption.

[0057] Specifically, when the assessed rise rate is deemed dangerous, the underlying control system constructs a digital low-pass filter mechanism through its internal digital signal processor. A cutoff frequency threshold for grid commands is set, and the original redundant power commands are processed by the filter. The smoothly changing reference power portion is separated and executed by the control unit, while the rapidly fluctuating portion exceeding the cutoff frequency is precisely stripped away and defined as a high-frequency load component. Subsequently, the control program monitors the locally calculated transient thermoelastic margin value. When it is determined that this margin value has decreased and fallen below the preset safety lower limit threshold, the program directly cuts off the waiting process for communication with the upper-level dispatcher and accesses a pre-cached local contingency plan lookup table at the local control layer via a hardware interface. Based on the lookup result, the local controller no longer sends requests via the network but directly surges the virtual impedance of the high-frequency load component digital command instantaneously in the underlying converter control algorithm. This impedance jump directly causes a small voltage or frequency instantaneous fluctuation corresponding to the load on the underlying shared bus. The adjacent energy storage unit, which is the target of the contingency plan, captures the bus fluctuation signal in real time through its local sensors and confirms its compensation role by combining the same Nash balance table cached within it, and actively lowers its high-frequency virtual impedance. According to Kirchhoff's current law, the high impedance forces the real high-frequency current to be automatically diverted on the bus and instantaneously redirected to the low-impedance adjacent unit for absorption, thus completing the physical spatial translation of the high-frequency load without reducing the total output of the power station.

[0058] This invention constructs a closed-loop protection mechanism of underlying flexible unloading and physical space reconstruction. By separating high-frequency impacts in the time domain and performing instantaneous transfer in space without communication, it can eliminate the risk of thermal runaway of individual aging cabinets without violating the overall power dispatch commitment of the power grid, and significantly enhance the survival resilience of energy storage clusters under harsh operating conditions.

[0059] Example 2: This embodiment provides a specific implementation process of an intelligent charging and discharging control method for energy storage cabinets in a commercial and industrial microgrid cluster scheduling scenario. In this scenario, the energy storage cluster consists of multiple heterogeneous energy storage cabinets with varying physical parameters, service life, and remaining capacity. It needs to respond in real-time to high-frequency modulation commands or power peak-shaving commands issued by the microgrid energy management system. The technical solution of this embodiment aims to address the safety and lifespan degradation issues caused by the disconnect between macroscopic command expectations and the underlying battery physical boundaries through deep perception of underlying physical characteristics and multi-agent game theory.

[0060] As one embodiment of the present invention, refer to Figure 1 A flowchart of an intelligent charging and discharging control method for an energy storage cabinet, referring to... Figure 2 The flowchart of the underlying flexible unloading and thermoelectric synergistic resistance mechanism is shown below. Figure 3 Non-cooperative game theory and local contingency plan generation sub-flowchart.

[0061] During the operation of the energy storage cluster, the local controller of each energy storage cabinet acquires the surface temperature gradient matrix, internal resistance dissipation power, and coolant flow rate data of each individual battery cell in real time through a distributed sensor network. Specifically, a high-precision temperature sensor array is used to monitor the temperature distribution at various feature points on the surface of the battery cell, and the real-time temperature difference of the coolant flowing into and out of the liquid cooling plate is collected simultaneously.

[0062] Based on a second-order reaction equivalent circuit model and combined with charging and discharging current data, a recursive least squares algorithm is used to identify the ohmic internal resistance, electrochemical polarization internal resistance, and concentration polarization internal resistance of the battery cell in real time. To address the concentration polarization phenomenon that occurs in the energy storage cabinet under long-term, high-rate operation, this embodiment introduces an active intervention mechanism based on time-scale decoupling: when the controller detects that the growth slope of the concentration polarization internal resistance exceeds a preset steady-state threshold, the energy storage converter actively superimposes an asymmetric high-frequency depolarization perturbation pulse train onto the base current for executing macro-grid regulation commands. The characteristic frequency of the asymmetric high-frequency depolarization perturbation pulse train is derived in real time by the controller based on the ohmic polarization time constant of the battery cell under the current operating conditions. Its alternating frequency needs to be dynamically maintained within a frequency band slightly higher than the reciprocal of this polarization time constant, thereby generating high-frequency electrostatic oscillations sufficient to break down the ion accumulation barrier without generating macro-level DC heating. The high-frequency alternating electric field of this pulse train can actively agitate the ionic double layer at the solid-liquid interface, accelerating the short-range relaxation diffusion of ions, thereby reducing concentration polarization resistance at the physical level.

[0063] To avoid the computational impact of high-frequency transient oscillations on the convergence of the recursive least squares algorithm, after the perturbation pulse train output ends, the internal resistance identification process is actively suspended, and a relaxation observation window of a preset time length is opened (usually tuned to the physical time constant of the solid-liquid interface double layer). This ensures that by the end of this time window, transient charges exceeding 95% of the threshold value have been physically released, allowing the subsequently extracted terminal voltage to accurately reflect the quasi-steady-state ohmic characteristics of the battery cell. Within the relaxation observation window, the system waits for the high-frequency glitches in the underlying voltage to dissipate and acquires the smooth exponential decline curve of the cell terminal voltage at a conventional sampling frequency. After the terminal voltage enters the quasi-steady-state decay range, the system restarts the parameter identification algorithm to extract the true ohmic and polarization resistance data at this point. Subsequently, based on the true internal resistance data after the pulse action, the irreversible thermal power, including Joule heat and polarization heat, is extracted separately using the product of the square of the real-time current and the true internal resistance. After strictly eliminating the physical interference of reversible reaction heat, this irreversible thermal power is directly divided by the real-time thermodynamic absolute temperature to accurately calculate the local thermodynamic irreversible entropy yield.

[0064] Simultaneously, the system extracts the values ​​from the coolant flow sensor and maps the current convective heat transfer coefficient based on the liquid-cooled plate piping topology, reading the effective heat transfer area. To eliminate temperature rise drift errors caused by fluid heat absorption along the flow path, the system simultaneously collects the inlet and outlet temperatures of the coolant and uses a logarithmic average temperature difference algorithm to calculate the effective characteristic temperature along the flow path of the coolant as it flows through the entire heat transfer surface. Subsequently, the system calculates the real-time equivalent temperature difference between the real-time collected cell surface temperature and this effective characteristic temperature along the flow path. Based on Newton's law of cooling, the system multiplies the convective heat transfer coefficient, the effective heat transfer area, and this real-time equivalent temperature difference to obtain the real-time actual heat dissipation power of the system under the current operating conditions. The system multiplies the preset adiabatic temperature rise rate critical threshold with the cell heat capacity to obtain the baseline adiabatic heat generation power, and linearly superimposes it with the real-time actual heat dissipation power to obtain the corrected safe heat generation boundary. Finally, the ratio of the local thermodynamic irreversible entropy yield to the upper limit of the allowable entropy yield under the corresponding boundary is calculated to dynamically generate the entropy increase loss cost index, which is used to quantify the physical deterioration of a single battery cell.

[0065] Each energy storage cabinet controller interacts in real time with adjacent cabinets via a point-to-point communication protocol, sharing parameters such as entropy increase loss cost index, state of charge, and instantaneous cooling redundancy. Based on this, a thermo-electrical two-layer non-cooperative game model is constructed, with the selfish objective of minimizing individual unit lifetime loss and including a global network regulation compliance penalty factor.

[0066] In the objective function of the game theory model, the equivalent power attenuation rate is configured as an exponential attenuation function with the entropy increase loss cost index as the independent variable, multiplied by a depth stress penalty coefficient. This penalty coefficient exhibits a non-linear, asymmetric increase as the cell's state of charge approaches the charging / discharging cutoff interval. Specifically, it represents either 0% (0% of the cell's state of charge approaching the charging / discharging cutoff interval) or 100% (100% of the cell's state of charge approaching the charging / discharging cutoff interval). Furthermore, to prevent overvoltage surges in power allocation at the physical execution level, a network topology impedance weight matrix is ​​additionally nested in the model. A high-frequency probe carrier is injected through the converter to measure the parasitic inductance of the physical transmission copper busbars between the local cabinet and adjacent cabinets in real time.

[0067] A distributed iterative optimization algorithm is employed to pre-calculate the power allocation weight matrix that enables each cabinet to achieve Nash equilibrium within a multi-dimensional state space composed of different states of charge, health states, and cost indices. The algorithm stops iterating when the lifetime loss benefit gained by any cabinet due to reduced power output is exactly equal to the marginal penalty of compliance it suffers due to deviation from macroscopic commands. The generated weight matrix is ​​stored as a local pre-plan in the controller's non-volatile flash memory. In this way, even if a transient communication link failure occurs during subsequent operation, the local controller can still retrieve the matching power weight allocation value by looking up the table within milliseconds of the current physical parameters, based on the asymmetric increase of the deep stress penalty coefficient as the state of charge approaches the charging / discharging cutoff interval, and generate local fine-tuning commands to maintain cluster lifetime synchronization.

[0068] Upon receiving macro-level grid dispatch instructions from the energy management system, the energy storage cabinet controller compares and differentially calculates these instructions with local fine-tuning instructions. The controller performs first-order derivative calculations on the macro-level instruction sequence to obtain the rate of change of power over time, and uses this to deduce the corresponding dynamic polarization internal resistance heat generation abrupt change value in the equivalent circuit model. By dividing this heat generation abrupt change value by the real-time absolute temperature, the instantaneous entropy production rate jump slope, characterizing the rate of physical damage deterioration, is obtained.

[0069] To assess the system's tolerance to the impact, the controller calculates a transient thermoelastic margin based on the cell's specific heat capacity, mass parameters, and the temperature difference between the current real-time temperature and the extreme temperature point corresponding to the corrected adiabatic temperature rise rate critical threshold. This margin, expressed in joules, represents the upper limit of pulse energy that the cell can rigidly absorb through its own physical heat capacity without triggering hardware over-temperature protection. If the assessment determines that the instantaneous entropy production rate jump slope is within a safe range, the system responds normally to the command; if it is determined that the slope will exceed the preset physical safety boundary threshold, the defense mechanism is automatically triggered.

[0070] When the physical safety boundary is deemed to be at risk of being breached, the underlying flexible unloading and thermoelectric co-resistance mechanism is immediately activated. First, the controller feeds back a virtual limiting signal to the energy management system to satisfy the power conservation rules at the microgrid level and prevent logical conflicts from occurring in the upper-level dispatch due to the failure to meet the command.

[0071] At the execution level, the controller employs a multi-dimensional energy shifting strategy. First, it utilizes the transient thermal elasticity margin to preferentially absorb a portion of the macroscopic power surge within a safe timeframe. When the thermal elasticity margin is determined to be approaching the safe lower limit, a frequency converter switching loss transfer strategy is activated: the controller dynamically reduces the carrier frequency of the power switching module inside the energy storage converter, thereby reducing the ramp rate of the terminal current. This converts the electrochemical polarization heat stress originally borne by the battery cells into the voltage fluctuation load of the converter's DC bus capacitor through electromagnetic coupling, and utilizes the high-flow-rate independent cooling system equipped in the converter to absorb this heat load in another physical space within the cabinet, thereby extending the thermal elasticity time window of the battery cells.

[0072] Second, the controller uses a digital filter to separate redundant power commands exceeding the safety boundary into high-frequency load components. Since the system is now in an emergency defense state, the local controller bypasses the communication handshake process with the central gateway and directly calls the local contingency plan. Because each energy storage cabinet in the cluster has pre-synchronously cached the same power allocation weight matrix, each cabinet has a pre-defined consensus on "who should reject and who should accept." The local cabinet increases the virtual impedance parameter of its controller, creating "high-resistance damping" at the physical level, preventing high-frequency power commands from being converted into local current. Simultaneously, the adjacent cabinet targeting the target adjusts its converter impedance to a "low-resistance path" based on the physical feedback from the busbar sensed by sensors. Relying on the electromagnetic coupling characteristics of the underlying busbar, redundant high-frequency energy, without digital command guidance, is automatically driven into the adjacent cabinet by the law that the depth stress penalty coefficient increases asymmetrically as the state of charge approaches the charging / discharging cutoff interval. The target cabinet uses its remaining transient thermoelastic margin to respond instantaneously to this physical impact.

[0073] This embodiment achieves maximum response within the safety boundary by cooperating with thermoelectric physical states within the underlying cluster while exhibiting limited power output at the upper layer. This process not only avoids the risk of thermal runaway caused by aging cabinets forcibly obeying commands, but also ensures the operational stability and asset security of the energy storage cluster under complex and unsteady conditions.

[0074] 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 intelligent charging and discharging control of energy storage cabinets, applied to an energy storage cluster comprising multiple heterogeneous energy storage cabinets, characterized in that, include: Acquire real-time thermoelectric operation data and coolant flow data of each energy storage cabinet, and calculate the local thermodynamic irreversible entropy production rate based on the real-time operation data; The preset critical threshold for adiabatic temperature rise rate is corrected by combining coolant flow data to generate an entropy increase loss cost index. Each energy storage cabinet obtains the entropy increase loss cost index of adjacent energy storage cabinets, constructs a non-cooperative game model that includes the single-unit lifetime loss target and the global network dispatch compliance penalty factor, solves the power allocation weight matrix that achieves Nash equilibrium, and caches it as a local plan lookup table to generate local fine-tuning instructions. The macro-grid control instructions issued by the energy management system are compared with the local fine-tuning instructions. The instantaneous entropy productivity jump slope when obeying the macro-grid control instructions is evaluated, and the transient thermoelastic margin is calculated based on the corrected critical threshold of the adiabatic temperature rise rate. If the instantaneous entropy production rate jumps above the physical safety boundary threshold, a virtual limiting signal is fed back to the energy management system. At the same time, the transient thermoelastic margin is used to absorb part of the power surge, and the redundant power command exceeding the safety boundary is separated into high-frequency load components. The local contingency plan is called to look up the table and the virtual impedance is dynamically increased. The high-frequency load components are redirected to the adjacent energy storage cabinet for physical absorption without communication triggering the bus shunt due to physical fluctuations.

2. The intelligent charging and discharging control method for an energy storage cabinet according to claim 1, characterized in that, Real-time thermoelectric operation data and coolant flow data of each energy storage cabinet are acquired. Based on the real-time operation data, the local thermodynamic irreversible entropy yield is calculated, including: real-time acquisition of the surface temperature of the individual cells, the electrode temperature, and the temperature difference of the coolant flowing into and out of the energy storage cabinet; acquisition of the charging and discharging current and calculation of the real-time internal resistance in combination with a preset equivalent circuit model; extraction of the irreversible heat generation rate based on the product of the square of the charging and discharging current and the real-time internal resistance; and after strictly eliminating the reversible reaction thermal interference determined by the reversible entropy change coefficient of the cell, the irreversible heat generation rate is compared with the real-time thermodynamic temperature to obtain the local thermodynamic irreversible entropy yield, which characterizes the degree of irreversible energy degradation inside the cell due to electrochemical reactions and physical heating per unit time.

3. The intelligent charging and discharging control method for an energy storage cabinet according to claim 1, characterized in that, The pre-set adiabatic temperature rise rate critical threshold is corrected using Newton's law of cooling by combining coolant flow rate data to generate an entropy increase loss cost index. This includes: obtaining the convective heat transfer coefficient corresponding to the coolant flow rate and the effective heat transfer area between the battery cell surface and the coolant; obtaining the coolant inlet and outlet temperatures to calculate the effective characteristic temperature along the flow path; calculating the real-time equivalent temperature difference between the battery cell surface temperature and the effective characteristic temperature along the flow path; determining the real-time heat dissipation power of the current system by multiplying the convective heat transfer coefficient, the effective heat transfer area, and the real-time equivalent temperature difference according to Newton's law of cooling; converting the pre-set adiabatic temperature rise rate critical threshold into a benchmark adiabatic heat generation power and linearly superimposing it with the real-time heat dissipation power to obtain the corrected safe heat generation boundary; calculating the real-time allowable upper limit of entropy production rate based on the safe heat generation boundary; and defining the ratio of the local thermodynamic irreversible entropy production rate to the real-time allowable upper limit of entropy production rate as the entropy increase loss cost index.

4. The intelligent charging and discharging control method for an energy storage cabinet according to claim 1, characterized in that, The construction of a non-cooperative game model that includes a single-unit lifetime loss target and a global grid regulation compliance penalty factor specifically includes: setting each energy storage cabinet as an independent game agent; the single-unit lifetime loss target of each agent is defined as minimizing the equivalent energy decay rate within a unit step; the equivalent energy decay rate is the product of the exponential decay value of the entropy increase loss cost index and the deep stress penalty coefficient, the deep stress penalty coefficient increasing asymmetrically as the state of charge approaches the charge / discharge cutoff interval; a global grid regulation compliance penalty factor is introduced into the objective function, defined as the absolute value of the difference between the actual response power of the single unit and the power allocated by the macro grid regulation command, multiplied by a weighting coefficient dynamically adjusted according to the grid frequency regulation pressure; each agent interacts with the cost index to iteratively adjust the power weight; the iteration is truncated when the marginal benefit of the reduction in the single unit decay rate equals the marginal penalty of deviation from the command, and the output power allocation share is taken as the Nash equilibrium solution.

5. The intelligent charging and discharging control method for an energy storage cabinet according to claim 1, characterized in that, The calculation of the power allocation weight matrix to achieve Nash equilibrium and its caching as a local plan lookup table includes: employing a distributed iterative optimization algorithm to calculate the power allocation share of each agent in different charge state intervals, health state intervals, and the entropy increase loss cost index interval, generating a power allocation weight matrix; storing the power allocation weight matrix in the local controller flash memory of each energy storage cabinet in the form of static structured data, forming the local plan lookup table; the local plan lookup table is associated with dynamic control offsets, and when the communication link is interrupted or local computing is limited, the power allocation weights are retrieved and corrected locally based on real-time parameters.

6. The intelligent charging and discharging control method for an energy storage cabinet according to claim 1, characterized in that, The evaluation of the instantaneous entropy production rate jump slope when obeying macro-grid control commands, and the calculation of transient thermoelastic margin based on the corrected adiabatic temperature rise rate critical threshold, includes: calculating the power change rate by performing a first derivative calculation on the macro-grid control command, and extracting the dynamic polarization resistance in the equivalent circuit model that matches the power change rate frequency band; calculating the transient heat generation mutation value using the power change rate and the dynamic polarization resistance, and dividing the transient heat generation mutation value by the real-time absolute temperature to obtain the instantaneous entropy production rate jump slope; obtaining the product of the cell specific heat capacity and the cell mass as the system thermal capacity constant; dividing the corrected safe heat generation boundary by the system thermal capacity constant to obtain the maximum allowable temperature rise rate; integrating the maximum allowable temperature rise rate within a preset time window to obtain the available temperature difference window; and multiplying the system thermal capacity constant by the available temperature difference window to obtain the transient thermoelastic margin characterizing the instantaneous energy absorption extreme value.

7. The intelligent charging and discharging control method for an energy storage cabinet according to claim 1, characterized in that, Redundant power commands exceeding the safety boundary are separated into high-frequency load components. A local contingency plan is invoked for lookup, and the virtual impedance is dynamically increased to allow physical fluctuations to trigger bus shunting without communication. The high-frequency load components are then redirected to adjacent energy storage cabinets for physical absorption. This process includes: using a digital low-pass filter to perform frequency domain decomposition on the redundant power commands exceeding the safety boundary, defining fluctuation components with frequencies higher than a preset cutoff frequency threshold as high-frequency load components; when the transient thermoelastic margin of the local energy storage cabinet is determined to be lower than a preset safety lower limit threshold, the high-frequency virtual impedance of the local cabinet is instantaneously increased in the control loop based on a local contingency plan lookup to flexibly reject high-frequency load components and induce physical fluctuations on the shared bus; simultaneously, the corresponding adjacent energy storage cabinets in the cluster synchronously sense physical fluctuations based on the same local contingency plan lookup and instantaneously reduce their own high-frequency virtual impedance, guiding the actual high-frequency surge corresponding to the high-frequency load components to automatically shunt and redirect on the physical lines into the adjacent energy storage cabinets for physical absorption.

Images