Multi-scale multi-physics energy storage emergency space twin modeling method

Abstract

The application relates to the technical field of energy storage safety, and discloses a multi-scale multi-physical-field energy storage emergency space twin modeling method, which comprises the following steps: presetting macro low-dimensional and micro high-dimensional models in an edge computing device, measuring residual computing power, setting a concurrent threshold, calculating space-time gradient characteristics by using a macro model based on annular buffer data to mark a candidate region, calculating a thermodynamic urgency index, sorting the candidate region, and screening out an activated region, calling historical data to drive a micro model to perform super-real-time operation, completing internal state initialization, establishing a heterogeneous model physical boundary bidirectional coupling by using flux impedance matching, outputting a fault evolution prediction result, and generating a fixed-point emergency control instruction. By means of thermodynamic urgency scheduling and boundary coupling technology, the application solves the high-fidelity simulation problem under the condition of limited computing power on the edge side, and improves the prediction accuracy and response speed of the multi-scale multi-physical-field energy storage emergency space twin modeling.

Description

Technical Field

[0001] This invention relates to the field of energy storage safety technology, specifically a multi-scale, multi-physics field energy storage emergency space twin modeling method. Background Technology

[0002] Industrial-grade energy storage power stations typically contain a large number of high-energy-density battery modules. Their multi-scale, multi-physics-field energy storage emergency space refers to a cross-scale physical coupling environment encompassing everything from the macroscopic thermal flux field distribution at the container level to the microscopic electrochemical reaction dynamics at the cell level. Twin modeling of this environment aims to construct a high-fidelity virtual model capable of mapping the operating state of physical entities in real time, providing decision support for safety assessment and emergency shutdown under extreme operating conditions.

[0003] Existing technologies primarily rely on battery management systems to collect voltage and surface temperature data, using lookup tables or equivalent circuit models for basic state estimation and over-limit protection, or employing cloud-based digital twin systems for remote centralized monitoring. Some applications typically deploy simplified models at the edge to reduce reliance on communication bandwidth, while high-precision simulation models containing complex physicochemical mechanisms are usually only used for offline design verification or post-fault inversion.

[0004] However, relying solely on cloud processing is insufficient to meet the real-time requirements of millisecond-level low latency in emergency response to sudden failures, while the limited computing resources of edge devices cannot support the concurrent operation of high-dimensional mechanism models across the entire site. Existing monitoring methods lack a mechanism for dynamically scheduling computing power based on the urgency of risks, making it difficult to balance the breadth of full-domain patrols with the depth of local fault analysis. Furthermore, traditional models typically neglect the bidirectional physical boundary coupling between the macroscopic environmental thermal field and the microscopic internal state, and lack the ability to reconstruct the temporal state based on historical data sequences, leading to biases in model initialization and making it difficult to accurately reveal the dynamic evolution mechanism of thermal runaway under extreme disaster scenarios.

[0005] Therefore, this invention proposes a multi-scale, multi-physics field energy storage emergency space twin modeling method to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a multi-scale, multi-physics field energy storage emergency space twin modeling method. This method solves the problems of existing energy storage system edge-side monitoring being unable to achieve both full-domain coverage and local high-fidelity simulation under limited computing power, as well as the low accuracy of thermal runaway evolution prediction due to the lack of macro-micro physical boundary coupling and temporal state reconstruction.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a multi-scale, multi-physics field energy storage emergency space twin modeling method, comprising the following steps:

[0008] In the edge computing device, a macroscopic low-dimensional model and a microscopic high-dimensional model are pre-set. The remaining computing power resources of the edge computing device under stable operation are measured, and a threshold for the number of concurrent loads of the microscopic high-dimensional model is set based on the remaining computing power resources.

[0009] A circular data buffer is established in memory to store voltage, current and temperature data collected by the sensor network. The spatiotemporal gradient feature values ​​of each sensor node are calculated using the macroscopic low-dimensional model. Regions whose spatiotemporal gradient feature values ​​exceed a preset safety threshold are marked as candidate regions.

[0010] Calculate the thermodynamic urgency index for each candidate region, prioritize all candidate regions based on the thermodynamic urgency index, and select the candidate regions with the highest ranking and whose total number does not exceed the concurrency threshold as the activation regions.

[0011] The historical data segment corresponding to the activated region is retrieved from the circular data buffer, and the historical data segment is used as the input sequence to perform the ultra-real-time accelerated operation of the micro high-dimensional model, thereby completing the state initialization of the micro high-dimensional model.

[0012] A bidirectional coupling relationship is established between the macroscopic low-dimensional model and the microscopic high-dimensional model at the physical boundary. The coupled microscopic high-dimensional model is used to output the fault evolution prediction results and generate fixed-point emergency control commands for the activated region.

[0013] Preferably, the steps of pre-setting the macroscopic low-dimensional model and the microscopic high-dimensional model include:

[0014] A macroscopic low-dimensional model based on a lumped parameter thermal network is constructed. The energy storage battery array is discretized into a finite number of thermal nodes using the macroscopic low-dimensional model, and the temperature dynamic response of the thermal nodes is described using state-space equations.

[0015] A microscopic high-dimensional model based on electrochemical thermal coupling is constructed, and the solid-phase lithium ion concentration and local potential distribution inside the microscopic high-dimensional model are described by the nonlinear discrete state equation after being processed by the order reduction technique.

[0016] Preferably, the step of setting a threshold for the number of concurrent loads of the microscopic high-dimensional model includes:

[0017] Determine the upper limit of the total computing power of the edge computing device;

[0018] The baseline load of the edge computing device while running the basic software and the macroscopic low-dimensional model was determined.

[0019] Measure the unit load during the runtime of a single instance of the aforementioned microscopic high-dimensional model;

[0020] The total computing power limit is multiplied by the safety redundancy coefficient and then subtracted from the baseline load. The difference is then divided by the unit load and rounded down to obtain the concurrency threshold.

[0021] Preferably, the step of calculating the spatiotemporal gradient feature values ​​of each sensor node includes:

[0022] Obtain the temperature measurement values ​​of the sensor node at the current time and the previous sampling time, and calculate the rate of temperature evolution over time;

[0023] The neighborhood set of the sensor node is obtained based on the topology determined by the macroscopic low-dimensional model, and the degree of thermal anomaly between the sensor node and its neighboring nodes in the neighborhood set is calculated.

[0024] The spatiotemporal gradient feature value is obtained by weighting and summing the product of the evolution rate and the time term weight coefficient, and the product of the thermal anomaly degree and the spatial term weight coefficient.

[0025] Preferably, the step of calculating the thermodynamic urgency index for each of the candidate regions includes:

[0026] Acquire pre-set spatial geometric layout data of the entire energy storage site, wherein the spatial geometric layout data includes the three-dimensional coordinates of key facilities;

[0027] The acceleration of the evolution of physical parameters within the candidate region is calculated using the second-order finite difference method;

[0028] Calculate the Euclidean spatial distance between the candidate region and the key facility;

[0029] The thermodynamic urgency index is obtained by assigning weighting coefficients to the acceleration and the Euclidean space distance respectively and then performing a weighted calculation.

[0030] Preferably, after the step of selecting the candidate regions that rank highly and whose total number does not exceed the concurrency threshold as activation regions, the method further includes:

[0031] For the remaining candidate regions that were not selected as the activation region, the edge computing device continues to run the macroscopic low-dimensional model;

[0032] A positive empirical correction coefficient is superimposed on the output of the macroscopic low-dimensional model to obtain a corrected temperature estimate, and the temperature estimate is used to perform monitoring in the next cycle.

[0033] Preferably, the step of completing the state initialization of the microscopic high-dimensional model includes:

[0034] The current and ambient temperature in the historical data segment are used as time-varying boundary conditions.

[0035] The microscopic high-dimensional model is driven to evolve on a virtual time axis using a differential equation solver with adaptive step size control.

[0036] This allows the internal electrochemical state variables and temperature distribution of the microscopic high-dimensional model to evolve from their nominal initial values ​​to estimated physical states at the current moment.

[0037] Preferably, the step of establishing the bidirectional coupling relationship between the macroscopic low-dimensional model and the microscopic high-dimensional model at the physical boundary includes:

[0038] The boundary coupling temperature is calculated iteratively using the flux impedance matching algorithm.

[0039] Construct an energy balance residual equation and calculate the difference between the internal heat flux fed back by the microscopic high-dimensional model and the environmental heat flux applied by the macroscopic low-dimensional model;

[0040] The boundary coupling temperature is corrected using the difference and the effective thermal admittance coefficient until the difference is less than a preset heat flux balance threshold.

[0041] Preferably, the step of outputting fault evolution prediction results using the coupled microscopic high-dimensional model includes:

[0042] Using the current state vector of the coupled model as the starting condition, the microscopic high-dimensional model is driven to perform advanced simulation and deduction using the load demand prediction curve within a preset future time window;

[0043] Calculate and output a dynamic hazard vector, which includes a thermal stability component characterizing the degree of thermal runaway approach, a chemical stability component characterizing the integral of the side reaction rate, and a mechanical stress component characterizing diffusion-induced stress.

[0044] Preferably, the step of generating a targeted emergency control command for the activated area includes:

[0045] Calculate the weighted magnitude of the dynamic hazard vector;

[0046] Based on the weighted modulus, a current allowable ratio coefficient is generated for the electrical branch where the activated region is located, and a cooling gain of the thermal management controller is generated for the flow channel where the activated region is located.

[0047] The allowable current ratio and the cooling gain are encapsulated into a control message and sent to the battery management system and the thermal management controller.

[0048] This invention provides a multi-scale, multi-physics field energy storage emergency space twin modeling method. It has the following beneficial effects:

[0049] 1. This invention achieves real-time, multi-scale monitoring of the entire large-scale energy storage system by constructing a heterogeneous model library containing macroscopic low-dimensional and microscopic high-dimensional levels and dynamically allocating edge computing resources based on a thermodynamic urgency index. Under limited hardware computing power, this method can maintain macroscopic monitoring of the entire site's spatial thermal environment while simultaneously loading high-precision models for in-depth local analysis of identified high-risk areas. This hierarchical strategy effectively balances the computational contradiction between overall coverage and local accuracy, ensuring the engineering usability of twin modeling technology in industrial settings and solving the computing power bottleneck problem in concurrent simulation of large-scale battery arrays.

[0050] 2. This invention achieves real-time interaction and deep data fusion between physical entity sensor data and the internal state of the virtual model by establishing a circular data buffer and performing retrospective time-domain state reconstruction. This method uses historical measured data sequences to drive the microscopic model in ultra-real-time evolution, eliminating simulation initialization bias caused by a lack of historical memory and ensuring that the digital twin can accurately reproduce the solid-phase lithium concentration and potential distribution inside the battery. This time-series data-based calibration mechanism significantly improves the model's accuracy in identifying hidden faults such as early internal short circuits, guaranteeing the physical authenticity of the microscopic state estimation.

[0051] 3. This invention reveals the dynamic failure mechanism of battery thermal runaway under extreme disaster scenarios by performing heterogeneous model boundary coupling and advanced strategy deduction. The method utilizes a flux impedance matching algorithm to achieve energy conservation at the boundary between the macroscopic environmental thermal field and the microscopic electrochemical reaction heat, and accurately predicts the evolution path of the fault. Based on this, the generated emergency control commands can precisely block the energy accumulation at the fault source before physical damage occurs, overcoming the hysteresis of traditional single-threshold protection and effectively reducing the risk of thermal runaway propagation and cascading failures in the energy system. Attached Figure Description

[0052] Figure 1 This is a flowchart of the multi-scale multi-physics field energy storage emergency space twin modeling method of the present invention;

[0053] Figure 2 This is a schematic diagram of the thermodynamic state reconstruction and fault evolution timing of the activated region in this invention. Detailed Implementation

[0054] The technical solutions in 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.

[0055] This invention provides a multi-scale, multi-physics field energy storage emergency space twin modeling method. This method operates in an industrial-grade energy storage system environment, which mainly includes an energy storage battery array, a battery management system, and an edge computing device that executes the core logic of this method.

[0056] The energy storage battery array, composed of numerous individual batteries connected in series and parallel, serves as the physical carrier for energy storage. The battery management system (BMS), acting as the underlying monitoring unit, is responsible for collecting physical signals from each node and providing basic protection. Edge computing devices establish communication connections with the BMS and thermal management controller via bidirectional industrial fieldbuses (such as CAN bus, Modbus, or industrial Ethernet). They can acquire real-time voltage, current, and temperature data from various sensor nodes within the energy storage battery array and have the authority to issue control commands to the BMS and thermal management controller.

[0057] Edge computing devices are equipped with a central processing unit and a memory. The memory stores computer program instructions, which, when executed by the central processing unit, implement the steps of the method.

[0058] Please see the appendix Figure 1 A multi-scale, multi-physics field energy storage emergency space twin modeling method includes the following steps:

[0059] S10, two types of physical models are pre-set in the storage space of the edge computing device: a macroscopic low-dimensional model based on a lumped parameter thermal network and a microscopic high-dimensional model based on electrochemical thermal coupling; the single computation cost of the macroscopic low-dimensional model is lower than that of the microscopic high-dimensional model; at the same time, the remaining computing power resources of the edge computing device under stable operation are measured, and the system is set with the threshold of the number of concurrent microscopic high-dimensional models allowed to be loaded at the same time.

[0060] S20: A circular buffer for cyclically storing data is opened in the memory of the edge computing device, and the voltage, current and temperature data collected by the sensor network are written into the circular buffer in real time; the circular buffer is configured to always retain the original data sequence within a preset historical period before the current moment; at the same time, the energy storage space is fully cruised based on a macroscopic low-dimensional model, and the spatiotemporal gradient feature values ​​of each sensor node are calculated in real time; the area where the spatiotemporal gradient feature value exceeds the preset safety threshold is marked as a candidate area;

[0061] S30, when a candidate region is detected, calculate the thermodynamic urgency index for each candidate region; the thermodynamic urgency index comprehensively considers the acceleration of the deterioration of physical parameters within the candidate region and the spatial distance between the candidate region and the critical facilities; prioritize all candidate regions according to the thermodynamic urgency index; select the candidate regions with the highest priority ranking and whose total number does not exceed the maximum concurrent number threshold as the activation regions; for the remaining candidate regions that are not selected as activation regions, maintain the operation of the macroscopic low-dimensional model and superimpose empirical correction coefficients;

[0062] S40, for the determined activation region, retrieve the historical data segment corresponding to the activation region from the circular buffer; call the microscopic high-dimensional model and use the historical data segment as the input sequence to perform ultra-real-time accelerated calculation in the background processing unit; through this accelerated calculation process, the internal electrochemical state variables and temperature distribution state of the microscopic high-dimensional model evolve from the nominal initial value to the real physical state at the current moment, and complete the state initialization of the microscopic high-dimensional model.

[0063] S50: In the activated region after initialization, establish a two-way coupling relationship between the macroscopic low-dimensional model and the microscopic high-dimensional model at the physical boundary; use the flux impedance matching algorithm to iteratively calculate the boundary coupling temperature until the environmental heat flux applied by the macroscopic low-dimensional model and the internal heat flux fed back by the microscopic high-dimensional model reach energy balance; output the fault evolution prediction results and dynamic hazard vector based on the coupled microscopic high-dimensional model, and generate fixed-point emergency control instructions for the activated region accordingly.

[0064] Please see the appendix Figure 1 In step S10, the specific edge-side deployment process of this embodiment first pre-configures a hierarchical physical model library within the non-volatile storage space of the edge computing device. This hierarchical physical model library follows a strategy of "macro-level global monitoring and micro-level local in-depth investigation," containing two types of mathematical models with significantly different spatiotemporal resolutions and computational complexities: a macro-level low-dimensional model based on a lumped parameter thermal network (LPTN), and a micro-level high-dimensional model based on electrochemical thermal coupling. This heterogeneous model architecture aims to resolve the inherent contradiction between the limited computing power at the industrial edge and the high-fidelity simulation requirements for energy storage safety. Furthermore, in this embodiment, the non-volatile storage space also pre-configures the global spatial geometric layout data of the energy storage station. This spatial geometric layout data maps the three-dimensional coordinate relationships of key facilities such as energy storage battery arrays, PCS converters, and fire protection pipe networks in physical space, thereby providing a geometric calculation benchmark for assessing the potential risks of fault areas to surrounding facilities in subsequent steps.

[0065] In this embodiment, a macroscopic low-dimensional model is constructed as a tool for describing the energy flow of the entire system. Based on the principle of heat conduction similarity, this macroscopic low-dimensional model discretizes the physical entity of the energy storage container, simplifying key components such as battery modules, electrical busbars, and cooling channels into a finite number of hot nodes. For any hot node in the system... Based on the law of conservation of energy, a first-order ordinary differential equation is established to describe its dynamic temperature response over time. In edge computing devices, the operating state of this macroscopic low-dimensional model is described by the following state-space equation:

[0066] ;

[0067] In the formula, The global node temperature vector of the spatial thermal network of the energy storage battery array, with dimension... The total number of battery modules and environmental nodes inside the container is usually controlled in the hundreds to maintain low-dimensionality characteristics. It is a diagonal heat capacity matrix, whose diagonal elements represent the product of specific heat capacity and mass of each physical node. Since the heat capacity of a physical entity is always positive, this diagonal heat capacity matrix has non-singularity and invertibility. It is a symmetric positive definite thermal conductivity matrix, which characterizes the thermal conductivity, convection and radiation heat transfer coefficients between nodes. This symmetric positive definite thermal conductivity matrix is ​​usually a sparse matrix. The internal heat generation power vector is related to the current vector. The functions mainly cover Joule heating and polarization heating; To be affected by ambient temperature The driven environmental boundary heat flux vector. Based on the above construction method, the single computational overhead of the macroscopic low-dimensional model is extremely low. Edge computing devices only need to call the sparse matrix solver to achieve millisecond-level global state updates, thereby meeting the requirements of all-time, blind-spot-free cruise monitoring.

[0068] In contrast, high-dimensional microscopic models focus on the accurate reproduction of the microscopic state inside the battery cell. While traditional quasi-2D models can accurately describe the diffusion process of lithium ions in the solid-liquid phase, they involve solving complex partial differential equations. Therefore, this embodiment employs an electrochemical-thermal coupling model processed using order reduction techniques (such as orthogonal decomposition of lithium ions (POD) or discrete empirical interpolation (DEIM). This model retains key microscopic variables that characterize the initial stage of an internal short circuit, such as the lithium ion concentration or local overpotential on the negative electrode solid phase surface. Its nonlinear discrete state equation is expressed as:

[0069] ;

[0070] ;

[0071] In the formula, for The microstate vector at time t, its dimension Although the order has been reduced, to ensure the accuracy of the description of electrochemical reactions, the number of nodes is still much larger than that of the macroscopic low-dimensional model (i.e., ); This is an input vector containing the cell current and the surface boundary temperature; , , , It is a coefficient matrix; This is a nonlinear correction term used to describe the nonlinear characteristics of electrochemical reaction rates as defined by the Butler-Volmer equation. Due to the inclusion of numerous microscopic state variables and nonlinear iterative calculations, the computational cost of a single run for the microscopic high-dimensional model is significantly higher than that of the macroscopic low-dimensional model.

[0072] Given the scarcity of edge computing resources, it is essential to quantify and calibrate the computing power boundary of the system. This process is not simply a matter of configuring static parameters, but rather requires determining the remaining computing power resources of the edge computing devices under stable operating conditions. Specifically, this involves first determining the baseline load when only the industrial-grade energy storage system kernel, basic communication processes, and a global macroscopic low-dimensional model are running. This metric is typically quantified using CPU time slice utilization or floating-point operations per second (FLOPS), representing the resident computing power consumption required to maintain basic system cruise monitoring. Subsequently, the peak resource consumption of a single microscopic high-dimensional model instance within the worst-case execution time (WCET) is determined through stress testing and denoted as unit load. .

[0073] Based on the above measured data, the system sets a threshold for the number of concurrent loads of microscopic high-dimensional models. The calculation logic is as follows:

[0074] ;

[0075] In the formula, This refers to the maximum total computing power nominally specified by the edge computing device hardware. For safety redundancy, the value range in this embodiment is set to 0.80 to 0.90, which aims to reserve buffer resources to cope with sudden network interruption or kernel state switching of industrial-grade energy storage systems and prevent the system from crashing under full load. It is a very small positive number (e.g., 10). -6 (), used to prevent abnormal values ​​where the denominator is zero; This indicates a floor function (rounding down). This is the concurrency threshold. As a hard constraint boundary for system resource scheduling, it ensures that under extreme conditions of concurrent failures in multiple regions, the system can perform priority-based scheduling and allocation based on the determined resource reserves, avoiding computational crashes caused by excessive stacking of model instances.

[0076] See attached document Figure 1 In step S20, this embodiment further performs dynamic data management configuration in the volatile memory (RAM) of the edge computing device. The volatile memory used here is physically independent of the non-volatile storage space used in step S10 for pre-setting the physical model library; the former is dedicated to meeting the data flow requirements of high-frequency read and write operations. Specifically, the system allocates a contiguous logical address space in the heap area of ​​memory to construct a circular data buffer. The logical structure of this buffer is not a non-linear linked list, but rather a closed-loop mechanism based on automatic looping of read and write pointers. Its physical capacity is strictly defined by the product of the sampling frequency of the sensor network and a preset historical time period.

[0077] Preset historical time period The value of this time period is not arbitrarily set, but determined based on the physical evolution of electrochemical thermal runaway. As a preferred approach, this time period is set to cover 3 to 5 times the adiabatic temperature rise time constant of a single cell (typically on the order of hundreds to thousands of seconds). This configuration ensures that the circular data buffer always retains a complete and continuous sequence of original data prior to the current moment. Whenever a new data frame arrives, the write pointer increments along the buffer address; once it reaches the end of the address space, it automatically jumps back to the beginning address to overwrite the oldest historical data. This mechanism ensures that the system, with limited memory resources, can continuously hold a window of historical data reflecting the latest physical state, providing the necessary boundary condition inputs for potential backtracking calculations of the microscopic model in subsequent steps.

[0078] While writing the voltage, current, and temperature data collected by the sensor network (i.e., a hardware set deployed inside the energy storage battery array, including distributed thermistors NTC, Hall current sensors, and voltage probes) into the circular data buffer in real time, the system uses the macroscopic low-dimensional model pre-defined in step S10 to perform a full-domain cruise of the energy storage space. This full-domain cruise does not mean solving all the differential equations of the macroscopic low-dimensional model, but rather focuses on utilizing the predefined physical topology in the macroscopic low-dimensional model, i.e., the thermal resistance connection relationship of each sensor node in the real physical space. The system reads the latest telemetry values ​​of each sensor node in real time and, combined with the adjacency matrix defined in the macroscopic low-dimensional model, calculates the spatiotemporal gradient eigenvalues ​​of each sensor node in real time. The spatiotemporal gradient feature value is a composite physical quantity designed to detect early signs of faults by fusing the temperature rise rate in the time dimension with the heat flux anomaly in the spatial dimension. In this embodiment, the spatiotemporal gradient feature value... The calculation logic is defined by the following formula:

[0079] ;

[0080] In the formula, and They represent the first Temperature measurements of each sensor node at the current time and at the previous sampling time; The sampling time interval; To prevent extremely small values ​​with a denominator of zero (e.g., 10) -6 ); First item ( The second term () characterizes the rate of temperature evolution over time and is used to identify sudden temperature rises. ) characterizes the degree of thermal anomalies in spatial dimensions, where For nodes determined by the topology of the macroscopic low-dimensional model The neighborhood set, This represents the total number of neighboring nodes within the set. Pre-defined nodes in macroscopic low-dimensional models With nodes The equivalent thermal resistance between them, this parameter has been ensured to be a non-zero positive value through parameter identification during model construction. (Symbol) Represents the linear rectified function (ReLU), i.e., only when the node... Temperature higher than neighboring nodes The time is included in the accumulation, which aims to eliminate false gradient contributions to the tested node caused by the heat conduction of neighboring nodes. and These are the time term weighting coefficient and the space term weighting coefficient, respectively. Their values ​​are determined based on the statistical noise level of the battery module under normal operating conditions, and are used to normalize physical quantities of different dimensions.

[0081] Based on the calculated feature values, the system executes dynamic threshold judgment logic. When any node... The calculation results show that the spatiotemporal gradient eigenvalues ​​exceed the preset safety threshold. When this occurs, it indicates that the region to which the node belongs exhibits an internal heat accumulation phenomenon that is significantly different from the ambient temperature rise. The system then maps the node back to a specific physical battery module or electrical partition and marks the physical region as a candidate region. The specific marking operation includes generating a marking information structure containing a unique identifier (ID) for the region, the absolute time of the trigger threshold, and the current gradient peak value.

[0082] See attached document Figure 1In step S30, this embodiment of the invention performs dynamic computing power scheduling based on thermodynamic urgency, based on the candidate region queue identified in step S20. This is because the computing resources of edge computing devices are limited by the concurrency threshold determined in step S10. In order to maximize computational efficiency in industrial-grade energy storage systems, an evaluation mechanism that can quantify the urgency of physical risks is introduced to determine which areas should be given priority access to high-precision microscopic simulation resources.

[0083] When the industrial-grade energy storage system detects the existence of a candidate region, it immediately executes risk quantification logic to calculate the thermodynamic urgency index for each candidate region. In this calculation process, the edge computing device first retrieves pre-set global spatial geometric layout data of the energy storage station. This data accurately maps the three-dimensional coordinates of key facilities such as PCS converters and fire-fighting pipeline valves in physical space. This thermodynamic urgency index does not solely rely on the current static temperature amplitude but is constructed as a dynamic, composite physical metric, aiming to comprehensively consider the acceleration of physical parameter evolution within the candidate region and the spatial distance between the candidate region and the key facilities. In this embodiment, the first... Thermodynamic urgency index of candidate regions The calculation is performed based on the following discretization formula:

[0084] ;

[0085] In the formula, , and They respectively represent the positions located at the th The measured temperature values ​​of sensor nodes within each candidate region at the current time, the previous sampling time, and the sampling time before that; The sampling period of the sensor network; the first term of the formula ( The second derivative of temperature with respect to time, i.e., the acceleration of physical parameter deterioration, is calculated using the second-order difference approximation. The physical motivation for introducing the acceleration term lies in the fact that the thermal runaway evolution of lithium-ion batteries follows the Arrhenius law of dynamics, often exhibiting significant self-accelerating nonlinear characteristics before the critical point. Compared to the simple first-order temperature rise rate, the second-order acceleration can more sensitively identify whether the battery module has crossed the linear temperature rise region and entered the irreversible exponential growth stage. This represents the linear rectification operator (RelU), which sets the term to zero when the calculation result is negative (indicating that the rate of temperature rise is converging or decreasing), thereby avoiding misclassification of areas undergoing cooling as high-risk targets.

[0086] The second term of the formula ( This characterizes the secondary disaster risk implied by the spatial distance between the candidate area and critical facilities. Among them, The total number of key facilities pre-installed in the energy storage power station (including but not limited to PCS converter cabinets, high-voltage DC combiner boxes, and fire suppression pipeline valves); For the first The importance weight of critical facilities is pre-set based on the scope of cascading failures or the amount of economic loss that may be caused by damage to the facilities (the value range is normalized to 0 to 1). Candidate region Geometric center and key facilities Euclidean distance between them; It is a non-zero safety buffer distance constant (e.g., 0.1 meters) designed to prevent numerical anomalies with a denominator of zero when the fault point is directly located on the coordinates of a critical facility; The distance attenuation index (usually set to 2) is used to simulate the physical law of energy attenuation with distance in thermal radiation or explosive shock waves. and These are acceleration weighting coefficients and position weighting coefficients, respectively, used to establish a scheduling balance between "suppressing the source of failure" and "protecting important assets".

[0087] After completing the above calculations, the system prioritizes all candidate regions based on the thermodynamic urgency index and selects the candidate regions with the highest priority ranking and whose total number does not exceed the concurrency threshold as the activation regions. The specific sorting and selection logic is as follows: all candidate regions are sorted according to their corresponding thermodynamic urgency index. The resources are sorted in descending order of their thermodynamic urgency indices to generate a dynamic scheduling sequence list. Then, a truncation and filtering logic is executed, starting from the top of the dynamic scheduling sequence list and proceeding in descending order of thermodynamic urgency index, performing activation determination based on resource boundaries. Specifically, the total number of candidate regions is compared with a concurrency threshold. The comparison process is as follows: If the total number of candidate regions is less than or equal to the concurrency threshold, all candidate regions in the dynamic scheduling sequence list are selected as active regions; if the total number of candidate regions is greater than the concurrency threshold, only regions with a sorting order of 1 to 2 in the dynamic scheduling sequence list are selected. Candidate regions within the specified interval are selected as activation regions. This selection process rigorously ensures that the total number of regions ultimately loaded with microscopic computing tasks never exceeds the concurrency threshold, and that the selected regions are all priority targets with the highest current physical risk.

[0088] For remaining candidate regions that, although marked as candidates, were not selected as activation regions due to their relatively low urgency index ranking, the system employs a "conservative defense" downgrade monitoring strategy. Edge computing devices do not load computationally intensive microscopic high-dimensional models onto these regions to free up computing power for high-risk targets. Instead, the system maintains the operation of a macroscopic low-dimensional model with empirical correction coefficients superimposed. Considering that the macroscopic low-dimensional model is a linearized approximation based on a lumped-parameter thermal network, it may underestimate local hotspots. Therefore, this embodiment superimposes empirical correction coefficients at the output of the macroscopic low-dimensional model. Corrected temperature estimate Defined as:

[0089] ;

[0090] In the formula, This is the original computational output of the macroscopic low-dimensional model; The safety margin factor is a positive value, and its value is typically set between 0.05 and 0.10 (i.e., an upward adjustment of 5% to 10%). This adjustment factor... As a safety margin, the system is forced to maintain a higher level of vigilance in low-computing-power monitoring mode. If the corrected monitoring value triggers a higher-level alarm again, or if the urgency index of the area rises in the next round of calculation, the area will have the opportunity to be upgraded to an active area in subsequent scheduling cycles.

[0091] See attached document Figure 1 In step S40, this embodiment performs retrospective time-domain state reconstruction on the activated region. Given that microscopic high-dimensional models (such as electrochemical P2D models or three-dimensional thermofluid models) consist of nonlinear partial differential equations (PDEs) and a system of algebraic equations, their operation depends on the initial assignment of internal hidden state variables. These internal hidden state variables specifically include the solid-phase lithium-ion concentration gradient inside the electrode particles, the local potential distribution of the electrolyte, and the non-uniform temperature field inside the battery cell core. In this embodiment, the system does not directly use the current time... The instantaneous observations of external voltage, current and surface temperature force the internal differential state of the microscopic high-dimensional model to be set to a static equilibrium state (i.e., zero gradient state). Instead, through the following time-domain reconstruction logic, the internal electrochemical polarization state and thermal distribution state that match the current dynamic operating conditions are established, thereby providing the initial boundary conditions that conform to physical continuity for solving the model.

[0092] To eliminate this initialization bias and achieve seamless synchronization between the state of the microscopic high-dimensional model and the real physical process, this embodiment implements a time-backtracking-based numerical integration strategy. For a given active region, the system uses a direct memory access (DMA) mechanism to retrieve the corresponding historical data segment from a circular buffer. The duration of this historical data segment corresponds to the preset historical time period set in step S20. (For example, it can be set to 300 to 600 seconds), which includes the time from the start time. Up to the current moment Complete time-series data. During retrieval, the system performs integrity checks on data segments. If data frame loss due to communication packet loss is detected, a third-order spline interpolation algorithm is used to fill in the missing points, ensuring the continuity and differentiability of the input sequence on the time axis and meeting the input requirements of the high-order differential equation solver.

[0093] After obtaining the verified data sequence, the system invokes a microscopic high-dimensional model and uses this historical data fragment as the input sequence. It then utilizes a background processing unit (such as an embedded GPU's computing core or a dedicated DSP coprocessor) in the edge computing device, independent of the main monitoring thread, to perform ultra-real-time accelerated computation. This ultra-real-time acceleration refers to the simulation computation proceeding significantly faster than the ebb and flow of physical time. The system employs a rigid differential equation solver with adaptive step-size control (such as the BDF backward difference formula or a higher-order Runge-Kutta method) to process the historical current... and ambient temperature As a time-varying boundary condition, it drives the rapid evolution of the microscopic high-dimensional model on a virtual time axis. In this embodiment, the dynamic process of this state evolution is explicitly described by the following integral equation:

[0094] ;

[0095] In the formula, For microscopic high-dimensional models in virtual time The full state vector at any given time is typically tens to hundreds of dimensions, containing the solid-phase lithium concentration distribution. Liquid phase potential and internal temperature field wait; Based on historical moments The initial equilibrium vector is derived from the voltage and temperature conditions (assuming uniform internal concentration and no thermal gradient). The input excitation vector retrieved from the circular buffer contains the current. and boundary temperature ; The nonlinear differential-algebraic equation operators defined for microscopic high-dimensional models characterize the electrochemical reaction kinetics and heat transfer laws. This is a pre-set set of electrochemical and thermodynamic parameters (such as the solid-phase diffusion coefficient of the electrode active material, the ionic conductivity of the electrolyte, the electrochemical reaction rate constant, the average specific heat capacity of the cell, and the anisotropic thermal conductivity, etc.).

[0096] This accelerated computation process allows the internal electrochemical state variables and temperature distribution of the microscopic high-dimensional model to evolve from their nominal initial values ​​to their current actual physical state. Specifically, under the continuous stimulation of historical operating data, the concentration polarization, ohmic heat generation, and heat conduction processes within the microscopic high-dimensional model are rapidly "reenacted" within an extremely short computation time. As the integration process progresses to the current moment... The state vector inside the microscopic high-dimensional model It is no longer a nominal balance value set by humans. Instead, it includes the true physical state estimate of the cumulative effects of historical operating conditions, thus completing the state initialization of the microscopic high-dimensional model.

[0097] To ensure the reconstruction process does not block real-time monitoring tasks, this embodiment sets strict constraints on the computing power of the background processing unit. Specifically, the computing speedup ratio... The following inequality conditions must be met:

[0098] ;

[0099] In the formula, The actual physical time required for the background processing unit to complete the above integration calculation; The system's preset time synchronization safety factor (preferably set to [value] in this embodiment) The significance of this constraint lies in the requirement to ensure that within an extremely short time... The past was completed within (e.g., milliseconds to seconds). Process backtracking at the minute level ensures that the computational lag time is negligible when the microscopic high-dimensional model switches to real-time monitoring mode, thereby achieving "disturbance-free switching" between the model's internal state and the battery's actual physicochemical state, providing an accurate initial value benchmark for subsequent fault prediction.

[0100] See attached document Figure 1In step S50, this embodiment performs heterogeneous model boundary coupling and emergency strategy simulation. Before step S50, although the microscopic high-dimensional model has completed internal state initialization, it is still in an isolated state in the numerical computation space. Since the activated region does not exist in isolation in physical entities, but is embedded in the overall energy storage system thermal environment represented by the macroscopic low-dimensional model, it is necessary to solve the problem of energy exchange consistency between the two at the spatial geometric interface. If a one-way transfer method is simply adopted, that is, only the ambient temperature calculated by the macroscopic low-dimensional model is imposed on the microscopic high-dimensional model, or the reverse thermal impact of the local hot spots generated by the internal reaction of the microscopic high-dimensional model on the macroscopic environment is ignored, it will lead to numerical discontinuity in the system-level thermal flow field calculation, thereby affecting the physical reliability of the subsequent safety strategy simulation.

[0101] In this embodiment, the system establishes a bidirectional coupling relationship between the macroscopic low-dimensional model and the microscopic high-dimensional model at the physical boundary in the activated region after initialization. Here, "activated region after initialization" specifically refers to a particular battery region where historical data has been loaded and the internal electrochemical state and temperature distribution reconstruction has been completed in step S40. The physical boundary is defined as the geometric outer surface of a single cell or battery module. At this interface, the macroscopic low-dimensional model is viewed as the connection surface of various thermal network nodes, responsible for describing the convective heat transfer effect of the external cooling medium (such as liquid cooling plate fluid or air-cooled airflow) on the boundary and the environmental radiation effect; while the microscopic high-dimensional model is viewed as the boundary condition of the partial differential equation solution domain, responsible for describing the Joule heat and reaction heat generated by the internal electrochemical reaction and their conduction to the surface. To eliminate numerical calculation conflicts caused by the different spatiotemporal resolutions and stiffness differences between the two heterogeneous models, the system introduces an iterative solution mechanism based on thermal impedance characteristics.

[0102] Specifically, the system uses a flux impedance matching algorithm to iteratively calculate the boundary coupling temperature. The principle of this flux-impedance matching algorithm is to analogize the thermal resistance effect in the heat transfer process to electrical impedance. By dynamically matching the heat flux response characteristics on both the inner (microscopic high-dimensional model side) and outer (macroscopic low-dimensional model side) sides, it seeks the interface equilibrium temperature that can simultaneously satisfy the energy conservation constraints and thermodynamic continuity on both sides. Within each simulation time step, the system constructs the following energy balance residual equation and solves it using a modified Newton-Raphson iteration method:

[0103] ;

[0104] ;

[0105] In the formula, Indicates the first Boundary coupling temperature estimate at the next iteration (unit: ); It is the heat flux residual function; To assume the boundary temperature is Under these conditions, the internal heat flux (unit: W / m²) flowing towards the boundary is calculated by the microscopic high-dimensional model based on its internal temperature gradient. 2 ); Under the same conditions, the macroscopic low-dimensional model calculates the environmental heat flux (i.e. heat dissipation flux) flowing from the boundary to the environment based on the external cooling fluid state.

[0106] In the formula The effective thermal admittance coefficient characterizes the sensitivity of the system's heat flux at the boundary to temperature changes (i.e., thermal stiffness). To prevent numerical divergence due to the denominator approaching zero during iteration, this embodiment... The calculation introduces regularization:

[0107] ;

[0108] In the formula, Non-zero local minimum constant (e.g., 10) -6 This ensures the numerical stability of the division operation. The adaptive relaxation factor has a value range set to [value range missing]. The system is based on the residual The rate of change of this factor is dynamically adjusted to achieve a balance between convergence speed and numerical stability.

[0109] Through the above iterative process, the system continuously corrects the boundary temperature. , driving residual The heat flux decreases monotonically until the environmental heat flux applied by the macroscopic low-dimensional model and the internal heat flux fed back by the microscopic high-dimensional model reach energy equilibrium. The numerical determination of this equilibrium state is given by the following convergence criterion:

[0110] ;

[0111] In the formula, The preset heat flux balance threshold (e.g., 5W / m) 2 The threshold value was set with reference to the measurement noise level of the system temperature sensor and the control accuracy requirements of thermal management. When the iteration error meets this condition, it indicates that the microscopic heat generation and macroscopic heat dissipation are physically precisely conserved at the boundary. The boundary temperature determined at this point is... This represents the actual interface temperature at that moment. The system then uses this as a fixed boundary condition, substituting it into both the macroscopic low-dimensional model and the microscopic high-dimensional model for state updates at the next time step. This achieves seamless fusion of multi-scale models at the thermodynamic level. This mechanism ensures that even under abnormal operating conditions where the battery experiences rapid temperature rise, there will be no gaps in energy calculations between the heterogeneous models, providing a reliable foundation consistent with physical laws for subsequent fault evolution inferences based on this coupled model.

[0112] See attached document Figure 1 In step S50, this embodiment also performs emergency strategy simulation and graded response. At this stage, the coupled model invoked by the system specifically refers to the joint simulation entity (i.e., a combination of macroscopic low-dimensional models and microscopic high-dimensional models) established through the aforementioned boundary coupling process, capable of simultaneously characterizing system-level thermal environment interaction and microscopic physicochemical reactions within the battery cell. Based on the high-fidelity physical boundary conditions of this coupled model, the system is no longer limited to passive monitoring of the current state, but instead enters a proactive prediction stage of the fault evolution trend within a short future window. This allows for the generation of intervention strategies for specific areas before microscopic physical risks evolve into macroscopic uncontrollable accidents.

[0113] In this embodiment, the system outputs fault evolution prediction results based on the coupled microscopic high-dimensional model. Specifically, the edge computing device utilizes a background processing unit to calculate the fault evolution prediction results at the current time. State vector of the coupled model As a starting condition, the preset future time window issued by the BMS will be used. The load demand forecast curve (e.g., within the next 30 to 60 seconds) serves as input, driving the coupled model (simultaneously running the macroscopic and microscopic parts) to perform advanced simulation. Because the high-dimensional microscopic model integrates sub-reaction models describing the kinetics of lithium plating, the heat generation from the decomposition of the solid electrolyte interphase (SEI) membrane, and aging and runaway mechanisms such as electrolyte drying, this simulation process can quantitatively assess the future electrochemical and thermodynamic states within the activated region under the continued operation of the current conditions. The evolutionary trajectory within a given period of time.

[0114] To transform the complex multiphysics simulation results into standardized metrics applicable to hierarchical decision-making, the system calculates and outputs dynamic hazard vectors. Unlike existing alarm mechanisms that rely solely on a single temperature or voltage threshold, this dynamic hazard vector is a normalized multidimensional numerical set used to comprehensively characterize the critical approximation degree of the activation region across three key physical dimensions: thermodynamics, electrochemistry, and mechanics. In this embodiment, the definition of the dynamic hazard vector and the calculation logic of its components are given by the following formula:

[0115] ;

[0116] In the formula, The thermal stability component characterizes the highest temperature of the core. (Distinguished from surface temperature) Approaching the critical thermal runaway trigger temperature within the prediction window The degree of. The preset safe operating temperature limit (e.g., 45°C). Set the critical thermal runaway trigger temperature (e.g., 130°C) and set This is to ensure that the denominator is positive. As a chemical stability component, it is determined by the rate of side reactions within the prediction window. The result is obtained by time integration and normalization. The side reaction rate (e.g., lithium deposition rate at the negative electrode, mol / (m)) 3 ·s)). Permissible cumulative side reaction threshold (unit: mol / m) 3 ), representing the total amount of byproducts that the system can tolerate without causing permanent capacity loss or short-circuit risk. The mechanical stress component characterizes the internal stress state of active particles caused by the rapid insertion and extraction of lithium ions. The diffusion-induced stress is calculated by the microscopic model based on the lithium concentration gradient inside the particle. The yield strength of the electrode material represents the physical limit at which the material undergoes mechanical fracture or pulverization. If the calculated result of any of the above components is less than 0, it is truncated to 0 to ensure that the vector represents the positive risk level.

[0117] After obtaining the quantified dynamic hazard vector, the system generates targeted emergency control commands for the activated areas based on the weighted modulus of this dynamic hazard vector. This process is achieved through a pre-set nonlinear risk-response mapping function, the technical purpose of which is to apply precise suppression measures only to the identified high-risk areas while ensuring that the overall power output of the energy storage system is as stable as possible. The targeted emergency control commands mainly include the allowable current proportionality coefficient for the electrical branch where a specific module or unit is located. and directional cooling gain for thermal management systems The thermal management system (or cooling system) refers to the collection of hardware actuators configured in an energy storage system to regulate battery temperature, including but not limited to liquid-cooled circulating pumps, refrigeration compressors, cooling fans, and flow channel solenoid valves. The specific control command generation logic follows the following S-shaped decay and saturation gain relationship:

[0118] ;

[0119] ;

[0120] In the formula, The allowable proportionality coefficient of the current applied to the corresponding electrical branch in the activation region (within the range of (0, 1]), this coefficient varies with the magnitude of the weighted hazard vector. The increase in frequency decreases along a Sigmoid curve, achieving a smooth transition from normal operation mode to complete disconnection mode and avoiding grid shock caused by hard disconnection; It is a diagonal weight matrix used to adjust the attention weights of the three risk dimensions of thermal, electrical and mechanical according to the characteristics of different battery chemical systems (for example, ternary lithium batteries can be given a higher weight for thermal risk). To control the response sensitivity coefficient (preferably a value of 5 to 10), which is used to adjust the steepness of the control intervention; Set as the risk warning threshold (e.g., 0.8). The target power gain of the cooling system (corresponding to the liquid cooling pump speed or fan duty cycle). and These are the maximum physical capacity limit and basic settings of the cooling system, respectively. This is the thermal risk feedback gain coefficient.

[0121] The final edge computing device will use the above calculations. and The command is encapsulated as a high-priority CAN bus control message and sent point-to-point to the Battery Management System (BMS) and Thermal Management Controller. Upon receiving this command, the BMS does not need to trigger a system-wide shutdown. Instead, it only performs corresponding power limiting or relay disconnection operations on the electrical branch where the activated area is located. Simultaneously, the Thermal Management System performs maximum power cooling on the flow channel in that area. This tiered response mechanism ensures that the energy storage system can achieve maximum safety with minimal system performance loss when facing localized micro-risks.

[0122] To verify the effectiveness of the multi-scale, multi-physics field energy storage emergency space twin modeling method of the present invention, this embodiment applies it to an industrial-grade energy storage system environment with a rated capacity of 2MWh. This system environment mainly includes an energy storage battery array, a battery management system, and an edge computing device executing the core logic of this method. The energy storage battery array consists of 10 battery clusters connected in parallel, each battery cluster containing 16 battery modules connected in series, for a total of 160 battery modules in the system. Each battery module has 4 temperature sensor nodes deployed within it. The edge computing device uses an embedded controller based on an industrial-grade ARM architecture, establishing communication connections with the battery management system and thermal management controller via a CAN bus. It is capable of acquiring sensor data in real time and has the authority to issue control commands.

[0123] 1. Construct a hierarchical physical model library and perform hardware resource calibration.

[0124] A hierarchical physical model library is pre-installed in the non-volatile storage space of the edge computing device.

[0125] Macroscopic low-dimensional model: Construct a global model based on lumped parameter thermal network, discretize the energy storage battery array into 320 thermal nodes (each battery module is simplified to 2 nodes, core and surface), and include environmental nodes.

[0126] Microscopic high-dimensional model: A pre-built single-cell battery model based on electrochemical thermal coupling is provided, which includes 35 microscopic state variables such as solid-phase lithium-ion concentration, liquid-phase potential and local temperature field.

[0127] The system measures the remaining computing power resources of edge computing devices under stable operating conditions. The total computing power limit of the edge computing devices is known. 100 GFLOPS. The baseline load when running the operating system, communication protocol stack, and global macroscopic low-dimensional model. The unit load of a single microscopic high-dimensional model instance is 15 GFLOPS. The measured value was 12 GFLOPS. A safety redundancy factor was set. It is 0.85, a very small positive number. Neglected. Calculate the concurrency threshold according to the formula described in step S10. :

[0128] ;

[0129] Therefore, the system sets the threshold for the number of concurrent loadings of microscopic high-dimensional models to be 5.

[0130] 2. Establish a circular data buffer and perform global gradient monitoring.

[0131] Edge computing devices establish a circular data buffer in memory and configure preset historical time periods. The interval is 600 seconds. Voltage, current, and temperature data collected by the sensor network are written to this buffer in real time, with old data being cyclically overwritten by new data.

[0132] The system performs a full-domain cruise of the energy storage space based on a macroscopic low-dimensional model. During operation, the sensor node of battery module No. 5 in cluster 3 (hereinafter referred to as Region A) detected a temperature increase from 35.0℃ to 38.5℃ within 10 seconds. The system calculated the spatiotemporal gradient characteristic value of this node, and the result showed that its value exceeded the preset safety threshold. The edge computing device marked Region A as a candidate region. At the same time, battery module No. 12 in cluster 7 (hereinafter referred to as Region B) experienced a slow temperature rise to 40.0℃ due to poor local heat dissipation, and its spatiotemporal gradient characteristic value also exceeded the preset safety threshold, so it was marked as a candidate region.

[0133] 3. Calculate the thermodynamic urgency index of region A and region B using a dynamic computing power scheduling system based on thermodynamic urgency.

[0134] Region A: The second derivative of temperature with respect to time (acceleration term) is positive, and according to the preset spatial geometry layout data, the spatial distance between Region A and the key facility (high-voltage DC combiner box) is... The distance is only 0.5 meters, which results in a high value for the location weight term.

[0135] Region B: Although the current temperature is high, the temperature acceleration term is close to zero, and it is far from critical facilities. Calculation results show that the thermodynamic urgency index of Region A is higher than that of Region B. The system prioritizes candidate regions, with Region A ranked first. Since the total number of candidate regions (2) does not exceed the concurrent threshold (5), the system selects both Region A and Region B as active regions. (Note: If the total number of candidate regions exceeds 5, the system will only select the top 5 regions as active regions, and the remaining regions will continue to run the macroscopic low-dimensional model with empirical correction coefficients.)

[0136] 4. Perform retrospective temporal state reconstruction on the activated region.

[0137] For active region A, the system retrieves the historical data segment of the past 600 seconds corresponding to that region from the circular data buffer.

[0138] The edge computing device invokes the microscopic high-dimensional model, using this historical data fragment as input, and performs ultra-real-time accelerated computation in the background processing unit. The computation takes 0.5 seconds, meeting the time synchronization requirements. Through this computation, the internal electrochemical state variables (such as the lithium-ion concentration on the negative electrode surface) and temperature distribution of the microscopic high-dimensional model evolve from nominal initial values ​​to the current actual physical state, completing the state initialization of the microscopic high-dimensional model. The calculation results show that there is an abnormal side reaction heat generation within region A.

[0139] 5. Performing Heterogeneous Model Boundary Coupling and Emergency Strategy Simulation: In the initialized activation region A, the system establishes a bidirectional coupling relationship between the macroscopic low-dimensional model and the microscopic high-dimensional model at the physical boundary. Iterative calculations are performed using a flux-impedance matching algorithm until the difference between the environmental heat flux applied by the macroscopic low-dimensional model and the internal heat flux fed back by the microscopic high-dimensional model is less than the heat flux balance threshold, thus determining the boundary coupling temperature.

[0140] Based on the coupled microscopic high-dimensional model, the system predicts the failure evolution over the next 60 seconds. The prediction shows that if the current operating conditions are maintained, the core temperature in region A will exceed the critical thermal runaway trigger temperature after 45 seconds. The system outputs a dynamic hazard vector, where the thermal stability component approaches 1.

[0141] Based on this, the edge computing device generates a targeted emergency control command:

[0142] For the electrical branch in area A, the allowable current proportional coefficient is set to 0 (i.e., the branch is disconnected).

[0143] For the thermal management controller, the cooling gain is set to the maximum value. The edge computing device sends instructions to the battery management system and the thermal management controller. The battery management system disconnects the third cluster of relays, and the thermal management controller turns on the corresponding area's fan at full speed, thereby preventing the fault evolution before physical damage occurs.

[0144] See attached document Figure 2 ,Should Figure 2 This demonstrates the temperature evolution timeline and state reconstruction effect of the target object selected as the activation area in this embodiment during the entire process of emergency space twin modeling.

[0145] Should Figure 2 China has established a system based on the current time. The relative time coordinate system is based on the position of the vertical line with the horizontal coordinate of 0 in the figure. Figure 2 The vertical solid line divides the timeline into two parts: the left side ( ) represents the state tracing process based on historical data, the right region ( This characterizes the future state deduction process based on model coupling.

[0146] To the left of the vertical solid line, within the historical time period, the gray solid line curve with circular markers represents the sequence of measured temperature data collected by the sensor network, showing a non-linear upward trend in the temperature of the activated region over time. The black solid line curve represents the state trajectory of the microscopic high-dimensional model during background operation. As shown in the figure, the black solid line initially deviates from the measured data, but with continuous correction from the input historical data, it quickly converges and conforms to the gray solid line. At that moment, the two curves completely overlapped, indicating that the microscopic high-dimensional model had been successfully initialized and accurately reproduced the current physical entity state.

[0147] Within the prediction period to the right of the vertical solid line, the system simulated two scenarios based on the current coupling state. The black dashed curve illustrates the fault evolution path under uncontrolled conditions, showing that without intervention, the temperature will rise sharply according to an exponential law and rapidly exceed the temperature marked as the critical thermal runaway trigger temperature. The critical thermal runaway threshold is shown in the black dashed-dot curve marked with a triangle. The controlled evolution path after applying a fixed-point emergency control command. The curve shows that, under the combined effect of current cutoff and powerful cooling, the temperature rise is immediately halted and reverses, remaining consistently at the upper limit of the safe operating temperature range. Within the safe operating range of ), this comparison verifies that the method can effectively prevent further evolution of thermal runaway.

[0148] 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 multi-scale, multi-physics field energy storage emergency space twin modeling method, characterized in that, Includes the following steps: In the edge computing device, a macroscopic low-dimensional model and a microscopic high-dimensional model are pre-set. The remaining computing power resources of the edge computing device under stable operation are measured, and a threshold for the number of concurrent loads of the microscopic high-dimensional model is set based on the remaining computing power resources. A circular data buffer is established in memory to store voltage, current and temperature data collected by the sensor network. The spatiotemporal gradient feature values ​​of each sensor node are calculated using the macroscopic low-dimensional model. Regions whose spatiotemporal gradient feature values ​​exceed a preset safety threshold are marked as candidate regions. Calculate the thermodynamic urgency index for each candidate region, prioritize all candidate regions based on the thermodynamic urgency index, and select the candidate regions with the highest ranking and whose total number does not exceed the concurrency threshold as the activation regions. The historical data segment corresponding to the activated region is retrieved from the circular data buffer, and the historical data segment is used as the input sequence to perform the ultra-real-time accelerated operation of the micro high-dimensional model, thereby completing the state initialization of the micro high-dimensional model. A bidirectional coupling relationship is established between the macroscopic low-dimensional model and the microscopic high-dimensional model at the physical boundary. The coupled microscopic high-dimensional model is used to output the fault evolution prediction results and generate fixed-point emergency control commands for the activated region.

2. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 1, characterized in that, The steps of setting the pre-defined macroscopic low-dimensional model and microscopic high-dimensional model include: A macroscopic low-dimensional model based on a lumped parameter thermal network is constructed. The energy storage battery array is discretized into a finite number of thermal nodes using the macroscopic low-dimensional model, and the temperature dynamic response of the thermal nodes is described using state-space equations. A microscopic high-dimensional model based on electrochemical thermal coupling is constructed, and the solid-phase lithium ion concentration and local potential distribution inside the microscopic high-dimensional model are described by the nonlinear discrete state equation after being processed by the order reduction technique.

3. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 1, characterized in that, The step of setting a threshold for the number of concurrent loads of the microscopic high-dimensional model includes: Determine the upper limit of the total computing power of the edge computing device; The baseline load of the edge computing device while running the basic software and the macroscopic low-dimensional model was determined. Measure the unit load during the runtime of a single instance of the aforementioned microscopic high-dimensional model; The total computing power limit is multiplied by the safety redundancy coefficient and then subtracted from the baseline load. The difference is then divided by the unit load and rounded down to obtain the concurrency threshold.

4. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 1, characterized in that, The steps for calculating the spatiotemporal gradient feature values ​​of each sensor node include: Obtain the temperature measurement values ​​of the sensor node at the current time and the previous sampling time, and calculate the rate of temperature evolution over time; The neighborhood set of the sensor node is obtained based on the topology determined by the macroscopic low-dimensional model, and the degree of thermal anomaly between the sensor node and its neighboring nodes in the neighborhood set is calculated. The spatiotemporal gradient feature value is obtained by weighting and summing the product of the evolution rate and the time term weight coefficient, and the product of the thermal anomaly degree and the spatial term weight coefficient.

5. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 1, characterized in that, The step of calculating the thermodynamic urgency index for each of the candidate regions includes: Acquire pre-set spatial geometric layout data of the entire energy storage site, wherein the spatial geometric layout data includes the three-dimensional coordinates of key facilities; The acceleration of the evolution of physical parameters within the candidate region is calculated using the second-order finite difference method; Calculate the Euclidean spatial distance between the candidate region and the key facility; The thermodynamic urgency index is obtained by assigning weighting coefficients to the acceleration and the Euclidean space distance respectively and then performing a weighted calculation.

6. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 1, characterized in that, After the step of selecting the candidate regions that rank highly and whose total number does not exceed the concurrency threshold as activation regions, the method further includes: For the remaining candidate regions that were not selected as the activation region, the edge computing device continues to run the macroscopic low-dimensional model; A positive empirical correction coefficient is superimposed on the output of the macroscopic low-dimensional model to obtain a corrected temperature estimate, and the temperature estimate is used to perform monitoring in the next cycle.

7. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 1, characterized in that, The steps for completing the state initialization of the microscopic high-dimensional model include: The current and ambient temperature in the historical data segment are used as time-varying boundary conditions. The microscopic high-dimensional model is driven to evolve on a virtual time axis using a differential equation solver with adaptive step size control. This allows the internal electrochemical state variables and temperature distribution of the microscopic high-dimensional model to evolve from their nominal initial values ​​to estimated physical states at the current moment.

8. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 1, characterized in that, The steps for establishing the bidirectional coupling relationship between the macroscopic low-dimensional model and the microscopic high-dimensional model at the physical boundary include: The boundary coupling temperature is calculated iteratively using the flux impedance matching algorithm. Construct an energy balance residual equation and calculate the difference between the internal heat flux fed back by the microscopic high-dimensional model and the environmental heat flux applied by the macroscopic low-dimensional model; The boundary coupling temperature is corrected using the difference and the effective thermal admittance coefficient until the difference is less than a preset heat flux balance threshold.

9. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 1, characterized in that, The steps for outputting fault evolution prediction results using the coupled microscopic high-dimensional model include: Using the current state vector of the coupled model as the starting condition, the microscopic high-dimensional model is driven to perform advanced simulation and deduction using the load demand prediction curve within a preset future time window; Calculate and output a dynamic hazard vector, which includes a thermal stability component characterizing the degree of thermal runaway approach, a chemical stability component characterizing the integral of the side reaction rate, and a mechanical stress component characterizing diffusion-induced stress.

10. The multi-scale, multi-physics field energy storage emergency space twin modeling method according to claim 9, characterized in that, The step of generating a targeted emergency control command for the activated area includes: Calculate the weighted magnitude of the dynamic hazard vector; Based on the weighted modulus, a current allowable ratio coefficient is generated for the electrical branch where the activated region is located, and a cooling gain of the thermal management controller is generated for the flow channel where the activated region is located. The allowable current ratio and the cooling gain are encapsulated into a control message and sent to the battery management system and the thermal management controller.

Images