An immersion type energy storage thermal control digital twin modeling simulation method

Abstract

The application relates to the technical field of energy storage system simulation digital twinning, and particularly discloses a submerged energy storage thermal control digital twinning modeling simulation method. The method comprises the following steps: constructing fluid thermal heterogeneous data index based on a flow field topology, analyzing a cooling medium physical flow path to establish a mapping relationship between storage and heat transfer logic; defining a fluid thermal composite linked list structure, and physically and continuously arranging battery module thermal data on an upstream and downstream of a same flow channel in a storage medium; establishing a thermal flow directed graph mapping mechanism, abstracting a complex flow channel into a database graph structure, and configuring node attributes and dynamic thermal dependence weights; and performing flow simulation deduction based on the flow field topology index, calling physically continuous thermal propagation link data to a cache to complete real-time calculation. The application eliminates the delay caused by random addressing, and improves data throughput and modeling fidelity.

Description

Technical Field

[0001] This invention belongs to the field of digital twin simulation of energy storage systems, and specifically relates to a digital twin modeling and simulation method for thermal control of immersion energy storage. Background Technology

[0002] With the continued acceleration of the global energy transition, large-scale energy storage systems are playing an increasingly prominent role in power peak shaving and renewable energy consumption. Immersion liquid cooling technology, with its excellent heat dissipation efficiency and temperature control consistency, has become an important technical approach to ensure the thermal safety of high power density energy storage systems. Digital twin technology, through high-fidelity digital mapping of physical entities, can perceive the internal thermal behavior logic of the system in real time and predict the trend of thermal runaway, playing a key role in the intelligent operation and maintenance of energy storage systems.

[0003] Digital twin modeling for thermal control in immersion energy storage aims to deeply integrate the flow field characteristics of the cooling medium with the thermal response features of the battery module to reconstruct the complex thermal management interaction process. This technical direction requires the simulation model to be able to process multi-source high-frequency sensor data and utilize fluid dynamics and heat transfer algorithms to provide real-time feedback on the temperature rise state inside the flow channel. By constructing a high-precision virtual simulation space, the system can provide decision support for operating condition changes under different discharge rates.

[0004] Existing technologies mostly employ traditional index structures based on spatial geometric coordinates to store thermal parameters, failing to characterize the inherent flow field topology of immersion liquid cooling systems. The upstream and downstream nodes of the flow channel, which possess strong thermal dependence at the physical level, are often discontinuously distributed in the data storage layer. This leads to a large number of cross-page searches and random access operations when the simulation engine extracts specific heat propagation links. This mismatch between data organization logic and fluid thermodynamic characteristics limits the update rate of digital twin models in complex evolution scenarios, making it difficult to meet the real-time and accuracy requirements of simulation results under highly dynamic operating conditions. Summary of the Invention

[0005] The purpose of this invention is to provide a digital twin modeling and simulation method for immersion energy storage thermal control, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a digital twin modeling and simulation method for immersion energy storage thermal control, comprising the following specific steps: Step 1: Construct a fluid thermal heterogeneous data index based on flow field topology. By analyzing the physical flow path of the cooling medium inside the energy storage chamber, establish a mapping relationship between data storage logic and physical heat transfer logic. Step 2: Define the fluid thermal composite linked list structure, taking the coolant flow path as the logical main line, and arrange the thermal data nodes of the battery module that are physically located upstream and downstream of the same flow channel in the storage medium in a physical continuous manner. Step 3: Establish a directed graph mapping mechanism for heat flow, abstract the complex flow channels in the submerged energy storage system into a graph structure in the database, and configure temperature attributes, flow rate attributes and heat dependence weights for each data node; Step 4: Perform flow simulation based on flow field topology index. When the digital twin model updates its thermal state, the system directly retrieves the physically continuous heat propagation link data to the cache unit to complete the real-time calculation of the thermal control state.

[0007] Preferably, in step 1, the process of constructing a fluid thermal heterogeneous data index based on flow field topology includes the partitioning of the physical space of the energy storage system into flow channels. Specifically, the system divides the immersion liquid-cooled space within the energy storage chamber into several independent flow channel units, each containing a preset number of battery modules. The system generates a complete physical flow path descriptor by identifying the inlet location of the cooling medium entering the energy storage chamber, the order in which it flows through each battery module, and the outlet location where it finally flows out of the energy storage chamber. This fluid thermal heterogeneous data index abandons the traditional three-dimensional spatial geometric coordinate index mode and instead adopts a topology index mode with the flow path descriptor as the core weight.

[0008] In this indexing mode, nodes 1 and 2, which may be spatially far apart but are on the same flow path, are assigned adjacent logical addresses in the indexing logic. This approach ensures that the order of data retrieval is highly consistent with the physical process of fluid flow, resolving the orthogonality failure problem between data storage logic and physical heat transfer logic.

[0009] Preferably, in step 2, defining the fluid thermal composite linked list structure involves reorganizing the underlying physical pages of the storage medium. The fluid thermal composite linked list includes a header node and several subsequent nodes. The header node stores the initial state parameters of the coolant inlet, including initial temperature, initial pressure, and initial mass flow rate. Each subsequent node corresponds to a discretized flow channel region or a specific battery module heating surface.

[0010] At the storage level, the system allocates a contiguous address space to store node data belonging to the same flow path. Each node internally encapsulates a logical pointer to a physical downstream node, allowing the simulation engine to directly address the next affected node in a step-by-step manner after completing the calculation of the current node. This physically continuous arrangement eliminates random input / output operations during the simulation calculation process, transforming discrete random retrieval into efficient sequential streaming reads, thereby improving the data throughput of the digital twin model.

[0011] Preferably, in step 3, the specific implementation of establishing the directed heat flow graph mapping mechanism includes digital modeling of heat propagation characteristics. The directed heat flow graph consists of a set of nodes and a set of directed edges. Each element in the node set represents the thermal state of a minimum simulation unit. The set of directed edges represents the direction of heat energy migration with the fluid. The system assigns a thermal dependency weight to each directed edge, which is used to characterize the degree of influence of upstream node thermal changes on the thermal state of downstream nodes.

[0012] The thermal dependence weights are determined by the coolant's specific heat capacity at constant pressure, flow path cross-sectional area, physical displacement between nodes, and real-time flow velocity. The system describes the calculation process using pure textual logic: the thermal dependence weight of the current node equals the product of the coolant flow rate and the time required to flow through the node, divided by the total heat capacity of the fluid medium contained within that node. Through this dynamic weight setting, the directed heat flow graph can reflect the impact of flow field fluctuations on the thermal control state in real time, providing a structured data representation for complex fluid thermodynamic processes.

[0013] Preferably, in step 4, the specific logic for performing the streaming simulation includes the generation of prefetch instructions. When the digital twin system detects a change in the operating condition of the physical entity, the simulation engine initiates a thermal runaway simulation task. The system first locates the initial affected node in the directed heat flow graph and, based on the physical contiguous address provided by the fluid thermal composite linked list, loads all the data of that node and a predetermined number of its downstream nodes into the on-chip cache of the central processing unit at once. During the calculation process, the system updates the temperature attributes of each node sequentially according to the pointing order defined in the directed heat flow graph. For each node update, the system extracts the temperature value of its predecessor node, multiplies it by the corresponding thermal dependency weight, and adds the heat generation rate of the battery module inside the current node to obtain the predicted temperature of the current node in the next simulation step. Since the data is linearly arranged in the cache, the computing unit does not need to wait for the response from the memory bus.

[0014] Preferably, step 1 further includes dynamic reconstruction of the flow field topology. During the operation of the submerged energy storage system, if the power of the circulating pump is adjusted or local blockage occurs inside the flow channel, the physical flow field will change. The system monitors data from pressure and flow sensors distributed at key locations in the flow channel and corrects the flow path descriptor in real time. Based on the feedback values ​​from the sensors, the system recalculates the topological connections between nodes and silently updates the logical pointer of the fluid-thermal composite list in the background. This dynamic reconstruction mechanism ensures that the digital twin model can always accurately map the real flow field characteristics of the physical entity under complex and changing operating conditions.

[0015] Preferably, in step 2, each node of the fluid thermal composite linked list further includes a preset health status flag. This health status flag is compared with the real-time temperature collected by the sensor and a preset safety threshold. If the temperature of the battery module corresponding to a node exceeds a predetermined range, the flag will flip. When the simulation engine performs streaming reads, if it detects an abnormal flag, it will automatically trigger a deep simulation mode to increase the calculation accuracy and sampling frequency of that node and its associated nodes.

[0016] Preferably, in step 3, the update frequency of the heat-dependent weights is synchronized with the sampling period of the physical system. The system has a preset update time threshold; when the rate of change of the physical sensor values ​​exceeds a certain range, the system immediately initiates a weight recalibration task. The calibration process involves a logical deduction of the Reynolds number of the flow field, that is, determining whether a transition from laminar to turbulent flow has occurred based on the ratio of the product of fluid viscosity, flow velocity, and channel characteristic length to viscous force. If a flow transition occurs, the system will automatically adjust the convective heat transfer coefficient term in the heat-dependent weights to ensure the prediction accuracy of the simulation model under different flow states.

[0017] Preferably, in step 4, the system also incorporates a multi-level buffer management mechanism. When loading heat propagation link data, the system not only loads the nodes of the current flow channel but also pre-loads key node data of adjacent flow channels based on the lateral coupling relationships in the directed heat flow graph. This lateral coupling typically occurs on the heat conduction path between battery modules. The system determines the lateral coupling weight by calculating the reciprocal of the lateral thermal resistance.

[0018] During the simulation, the temperature update of the current node depends not only on the convective effects of the upstream fluid but also on the heat conduction contributions from adjacent nodes. Through this comprehensive data pre-fetching and integrated calculation, the digital twin model can fully reproduce the complex 3D thermal field evolution process inside the energy storage compartment.

[0019] Preferably, the present invention also includes a historical data alignment module. This historical data alignment module periodically compares the predicted values ​​obtained from simulation with the measured values ​​fed back by physical sensors. If the deviation between the two exceeds a preset fault tolerance range, the system will initiate a feedback adjustment program. The feedback adjustment program corrects the heat dependence weights in the directed heat flow graph or adjusts the heat capacity parameters in the fluid thermal complex chain. This closed-loop calibration mechanism ensures the consistency and reliability of the digital twin model during long-term operation.

[0020] Preferably, when executing the immersion energy storage thermal control digital twin modeling and simulation method, the system employs a priority-based scheduling algorithm. For battery branches under high-load charging and discharging conditions, the system assigns higher scheduling priority to their associated data channels, increasing their weight in memory swapping and computing resource allocation. This differentiated processing strategy ensures that, with limited computing resources, the thermal control simulation of key risk areas can be updated in the most timely manner.

[0021] Preferably, the method also involves modeling the aging characteristics of the cooling medium. As operating time increases, the physical properties of the submerged coolant, such as viscosity and thermal conductivity, will slowly shift. The system periodically fine-tunes the global physical properties in the fluid-thermal composite chain using a built-in chemical property evolution model. This periodic fine-tuning logic is based on the cumulative operating time of the coolant, the number of cycles at the highest temperature experienced, and a preset aging degradation coefficient. This long-term performance consideration allows the digital twin model to cover the entire lifecycle of the energy storage system, providing a scientific basis for the optimization and upgrading of the thermal management system.

[0022] Preferably, the simulation process is performed within a distributed computing architecture. In this architecture, different partitions of the energy storage module are assigned to different processing units, and these units exchange edge data of the directed heat flux graph via a high-speed backplane bus. To ensure timing consistency, the system employs a synchronization protocol based on a global logical clock. Each processing unit sends a synchronization signal to the central management unit after completing the simulation step for its current partition. Only after all partitions have completed the calculation of their current step and exchanged boundary heat flux data does the system proceed to the next simulation moment. This parallel processing scheme shortens the simulation time for ultra-large-scale energy storage power plants.

[0023] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention improves data retrieval efficiency. By constructing a data storage structure that prioritizes flow field topology, the physical heat transfer logic is deeply coupled with the underlying storage logic, eliminating frequent random addressing in simulation calculations and reducing data read latency.

[0024] 2. This invention enhances the real-time performance of digital twin models. The combination of fluid-thermal composite linked lists and streaming readout mechanisms enables the system to respond quickly to changes in physical conditions, supporting high-frequency, large-scale thermal control state prediction tasks.

[0025] 3. This invention improves modeling accuracy under complex operating conditions. By introducing a directed heat flow graph and dynamic thermal dependence weights, it can accurately characterize the complex fluid-thermal interaction relationships in immersion liquid cooling systems, exhibiting extremely high simulation fidelity in scenarios with unsteady flow field evolution.

[0026] 4. This invention optimizes hardware resource utilization. The data organization method based on physical continuity improves cache hit rate and reduces the pressure on memory bandwidth and input / output systems, making it possible to run large-scale energy storage system digital twins on ordinary hardware platforms.

[0027] 5. This invention possesses excellent dynamic adaptability. Through dynamic index reconstruction and feedback calibration mechanisms, the system can autonomously adapt to changes in physical flow channels and drifts in equipment parameters, ensuring the robustness of the simulation model throughout its entire lifecycle. Attached Figure Description

[0028] Figure 1 The flowchart is provided according to the present invention; Figure 2 This is a schematic diagram of data flow according to the present invention; Figure 3 This is a flowchart illustrating the logical flow of establishing a fluid thermal heterogeneous data index and a fluid thermal composite linked list structure based on flow field topology according to the present invention. Figure 4 This is a schematic diagram of the core principle framework for performing flow simulation deduction based on the directed graph mapping mechanism of heat flow and the configuration of heat dependence weights according to the present invention; Figure 5 This is a flowchart illustrating the logical flow framework for dynamic reconstruction of flow field topology and closed-loop calibration of model parameters based on physical sensor feedback according to the present invention. Detailed Implementation

[0029] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 In the immersion energy storage thermal control digital twin modeling and simulation system, the core data processing logic is based on the deep decoupling and logical reconstruction of the physical flow field and thermodynamic state. This embodiment provides an immersion energy storage thermal control digital twin modeling and simulation method, the execution of which strictly follows these steps: In the method, step 1 involves constructing a fluid thermal heterogeneous data index based on flow field topology. This is achieved by analyzing the physical flow path of the cooling medium inside the energy storage chamber, establishing a mapping relationship between data storage logic and physical heat transfer logic. The process of constructing the fluid thermal heterogeneous data index based on flow field topology includes dividing the physical space of the energy storage system into flow channels. The system divides the immersion liquid-cooled space inside the energy storage chamber into several independent flow channel units, each containing a predetermined number of battery modules. The system generates a complete physical flow path descriptor by identifying the inlet location of the cooling medium entering the energy storage chamber, the order in which it flows through each battery module, and the final outlet location of the cooling medium exiting the energy storage chamber.

[0030] The fluid thermal heterogeneous data index abandons the traditional three-dimensional spatial geometric coordinate indexing mode and instead adopts a topological indexing mode with flow path descriptors as the core weights. In this indexing mode, nodes 1 and 2, which may be spatially far apart but are on the same flow path, are assigned adjacent logical addresses in the indexing logic. This approach ensures that the data retrieval order is highly consistent with the physical process of fluid flow, resolving the orthogonality failure problem between data storage logic and physical heat transfer logic.

[0031] During step 1, the system also dynamically reconstructs the flow field topology. During the operation of the submerged energy storage system, if the power of the circulating pump is adjusted or local blockage occurs inside the flow channel, the physical flow field will change. The system monitors data from pressure and flow sensors located at key positions in the flow channel and corrects the flow path descriptor in real time. Based on the sensor feedback values, the system recalculates the topological connections between nodes and silently updates the logical pointer of the fluid thermal composite list in the background.

[0032] This dynamic reconstruction mechanism ensures that the digital twin model can accurately map the real flow field characteristics of the physical entity under complex and ever-changing operating conditions. To achieve accurate topology mapping, the system's internal flow field identification module polls the values ​​of each pressure monitoring point every 100 milliseconds. When the pressure difference between two adjacent monitoring points exceeds a preset disturbance threshold, the system determines that the flow channel state has changed and initiates a topology verification program. This topology verification program calculates the velocity distribution of the fluid under the current pressure gradient, re-identifies the main flow path of the coolant, and incrementally updates the affected flow path descriptors.

[0033] In the method, step 2 defines a fluid-thermal composite linked list structure. Using the coolant flow path as the logical main line, the thermal data nodes of battery modules located upstream and downstream of the same flow channel at the physical level are physically arranged continuously in the storage medium. Defining the fluid-thermal composite linked list structure involves reorganizing the underlying physical pages of the storage medium. The fluid-thermal composite linked list includes a header node and several subsequent nodes. The header node stores the initial state parameters of the coolant inlet, including initial temperature, initial pressure, and initial mass flow rate. Each subsequent node corresponds to a discretized flow channel region or a specific heated surface of the battery module.

[0034] At the storage level, the system allocates a contiguous address space to store node data belonging to the same flow path. Each node internally encapsulates a logical pointer to a physical downstream node, allowing the simulation engine to directly address the next affected node in a step-by-step manner after completing the calculation of the current node. This physically continuous arrangement eliminates random input / output operations during the simulation calculation process, transforming discrete random retrieval into efficient sequential streaming reads, thereby improving the data throughput of the digital twin model.

[0035] Specifically, regarding the encapsulation format of data nodes, each node, in addition to containing basic temperature and flow rate fields, also includes a preset health status flag. This health status flag compares the real-time temperature collected by the sensor with a preset safety threshold. If the temperature of the battery module corresponding to a node exceeds the predetermined range, the flag will flip its state.

[0036] During streaming reads, if the simulation engine detects an anomaly in the flag bit, it will automatically trigger a deep simulation mode, increasing the computational accuracy and sampling frequency of the node and its associated nodes. The node also stores physical parameters such as the local Reynolds number, Nusselt number, and convective heat transfer area of ​​the cooling medium in that region. During the memory allocation phase, the system employs memory alignment technology to ensure that the starting address of each node is located at the starting boundary of a computer memory page, utilizing the hardware bus bandwidth.

[0037] In the method, step 3 establishes a directed heat flow graph mapping mechanism, abstracting the complex flow channels within the submerged energy storage system into a graph structure in a database, and configuring temperature attributes, flow velocity attributes, and thermal dependency weights for each data node. The specific implementation of establishing the directed heat flow graph mapping mechanism includes digital modeling of heat propagation characteristics. The directed heat flow graph consists of a set of nodes and a set of directed edges. Each element in the node set represents the thermal state of a minimum simulation unit. The set of directed edges represents the direction of thermal energy migration with the fluid. The system assigns a thermal dependency weight to each directed edge, which characterizes the degree of influence of upstream node thermal changes on the thermal state of downstream nodes. The value of the thermal dependency weight is jointly determined by the isobaric specific heat capacity of the coolant, the cross-sectional area of ​​the flow path, the physical displacement between nodes, and the real-time flow velocity.

[0038] In one specific embodiment, the calculation of thermal dependency weights in the directed heat flow graph mapping mechanism is precisely modeled. Specifically, each node in the directed heat flow graph represents a discretized spatial control unit, which contains a portion of the battery module and the cooling medium surrounding it. The thermal dependency weights are used to quantify the instantaneous influence of the thermal state of the upstream node control body on the thermal state of the downstream node control body. The core logic of the weight calculation is based on the principle of energy conservation, combining the time hysteresis effect of fluid flow with heat capacity.

[0039] In the directed heat flow graph, from node... to its downstream nodes The heat dependency weights corresponding to the directed edges are The calculation formula is as follows: ; in, For the nodes that flow through To the node The mass flow rate of the coolant in the flow channel; For coolant from the node flow from center to node The average time required for the center can be calculated from the channel length and the average flow velocity. and These are the equivalent isobaric specific heat capacities of the coolant and battery module materials, respectively. and They are nodes The mass of the coolant contained within the body is controlled to be equivalent to the mass of the battery module. This formula clearly accounts for the hysteresis effect of fluid flow (…). The value is associated with the total heat capacity of the control volume (denominator) and defines the sensitivity of the effect of unit energy input on node temperature.

[0040] When describing the heat dependency weights, the system uses pure textual logic to describe its calculation process: the heat dependency weight of the current node is equal to the product of the coolant flow rate and the time required for the coolant to flow through the node, divided by the total heat capacity of the fluid medium contained in that node. Through this dynamic weight setting, the directed heat flow graph can reflect the impact of flow field fluctuations on the thermal control state in real time. The update frequency of the heat dependency weights is synchronized with the sampling period of the physical system. The system has a preset update time threshold; when the rate of change of the physical sensor values ​​exceeds a certain range, the system immediately initiates a weight recalibration task.

[0041] The calibration process involves a logical deduction of the Reynolds number of the flow field. Specifically, it determines whether a transition from laminar to turbulent flow has occurred based on the ratio of the product of fluid density, velocity, and characteristic channel length to the fluid dynamic viscosity. If a transition occurs, the system automatically adjusts the convective heat transfer coefficient term in the heat dependence weights. When the flow is laminar, the convective heat transfer coefficient is mainly affected by the fluid thermal conductivity and channel geometry; however, when the flow transitions to turbulence, the system adds a turbulence intensity correction term.

[0042] Specifically, the dynamic update process of the heat-dependent weights includes a complete set of flow regime discrimination and parameter correction algorithms. The system first calculates the local Reynolds number within the flow channel in real time and determines the flow state based on the critical Reynolds number threshold. When the flow regime changes, the system uses different empirical correlations to calculate the convective heat transfer coefficient, thereby dynamically adjusting the heat transfer term in the heat-dependent weights. For turbulent flow, the system further introduces a turbulence enhancement factor to correct the basic convective heat transfer coefficient; this factor has a power function relationship with the Reynolds number. The system first calculates the local Reynolds number. : ; in, For fluid density, Characteristic flow velocity, The hydraulic diameter of the flow channel. For fluid dynamic viscosity. When (For example, take) When ), the flow regime is laminar; when At that time, the flow pattern is turbulent.

[0043] In laminar flow, nodes convective heat transfer coefficient Calculated using the Sieder-Tate formula: ; in, For fluid thermal conductivity, For Prandtl numbers, The length of the flow channel. Let be the fluid viscosity at the wall temperature.

[0044] Under turbulent conditions, the system introduces a turbulence intensity correction term, which is calculated using the Gnielinski formula. : ; in, The coefficient of friction. The heat-dependent weight. The heat transfer terms in the data will be dynamically updated to match the new data. The relevant values ​​enable adaptive adjustment of the weights according to the flow regime.

[0045] In the method described, step 4 involves performing a streaming simulation based on the flow field topology index. When the digital twin model updates its thermal state, the system directly retrieves physically continuous heat propagation link data to the cache unit to complete real-time calculation of the thermal control state. The specific logic for performing the streaming simulation includes the generation of prefetch instructions. When the digital twin system detects a change in the operating conditions of the physical entity, the simulation engine initiates a thermal runaway simulation task.

[0046] The system first locates the initial affected node in the directed heat flow graph and, based on the physical contiguous addresses provided by the fluid thermal composite linked list, loads all data of that node and a predetermined number of its downstream nodes into the on-chip cache of the central processing unit at once. During the calculation, the system updates the temperature attributes of each node sequentially according to the pointing order defined in the directed heat flow graph. For each node update, the system extracts the temperature value of its predecessor node, multiplies it by the corresponding thermal dependency weight, and adds the heat generation rate of the battery module inside the current node to obtain the predicted temperature of the current node in the next simulation step. Because the data is linearly arranged in the cache, the computing unit does not need to wait for a response from the memory bus.

[0047] In a preferred embodiment, the node temperature updates in the flow simulation are discretized using an explicit Euler method. The rate of temperature change at each node is determined by the heat flow brought in by the upstream fluid, the heat generated by the local cell, and the lateral heat conduction to adjacent nodes. Within each time step, the simulation engine sequentially calculates the next temperature value for each node according to the topological order defined by the directed heat flow graph, thereby simulating the time evolution of the thermal state. Let the simulation step size be... ,node exist Temperature of a moment The calculation is as follows: ; in, For nodes The temperature at the current moment; For nodes Total heat capacity of the control volume; upstream node The temperature at the current moment; As a reference for coolant inlet temperature; For nodes The heat generation rate of the internal battery module can be determined based on the current. and internal resistance Depend on calculate; For nodes The effective heat dissipation area. This formula constitutes the core iterative operator for simulation derivation, clarifying all the components of temperature update and their calculation methods.

[0048] In step 4, the system also employs a multi-level buffer management mechanism. When loading heat propagation link data, the system not only loads the nodes of the current flow channel but also pre-loads key node data from adjacent flow channels based on the lateral coupling relationships in the directed heat flow graph. This lateral coupling typically occurs along the heat conduction path between battery modules. The system determines the lateral coupling weight by calculating the reciprocal of the lateral thermal resistance. The magnitude of the lateral thermal resistance is determined by the thermal conductivity, thickness, and contact area of ​​the insulating layer material between adjacent battery modules.

[0049] During the simulation, the temperature update of the current node depends not only on the convective effects of the upstream fluid but also on the heat conduction contribution from adjacent nodes. The system calculates the temperature difference between the current node and its adjacent nodes, multiplies this temperature difference by the lateral coupling weight, and calculates the heat exchange based on the simulation step length. Through this comprehensive data prefetching and integrated calculation, the digital twin model can fully reproduce the complex 3D thermal field evolution process inside the energy storage compartment.

[0050] In the multi-level buffer management mechanism, the lateral coupling weight is calculated based on a one-dimensional flat plate thermal conductivity model. The system treats the insulating layer between adjacent battery modules as a flat wall, whose thermal resistance is determined by the material's geometry and thermal properties. The lateral coupling weight is the reciprocal of this thermal resistance, characterizing the heat conduction capacity per unit temperature difference. This weight is directly used in the calculation of the lateral heat conduction term in the node temperature update equation. Adjacent nodes and Lateral thermal resistance between The calculation formula is: ; in, For the thickness of the insulation layer, The thermal conductivity of insulating materials, This represents the contact area between adjacent modules. The lateral coupling weight... That is, the reciprocal of the thermal resistance: ; In the nodal temperature update formula, the lateral heat conduction contribution term should be modified as follows: ; This item will be added to the foregoing. The calculation formula includes the full impact of all heat transfer paths within the parentheses.

[0051] This invention also includes a historical data alignment module. This module periodically compares the predicted values ​​obtained from simulation with the measured values ​​fed back by physical sensors. If the deviation between the two exceeds a preset tolerance range, the system will initiate a feedback adjustment program. The feedback adjustment program corrects the heat dependence weights in the directed heat flow graph or adjusts the heat capacity parameters in the fluid thermal composite chain. For example, when the system observes that the measured temperature rise rate is consistently higher than the simulation prediction, the feedback adjustment program will automatically reduce the local equivalent specific heat capacity parameter in the directed heat flow graph or increase the battery heat generation correction coefficient in the local heat dependence weights until the sum of squared residuals between the simulation curve and the measured curve falls within a preset range. This closed-loop calibration mechanism ensures the consistency and reliability of the digital twin model during long-term operation.

[0052] When executing the immersion energy storage thermal control digital twin modeling and simulation method, the system employs a priority-based scheduling algorithm. For battery branches under high-load charging and discharging conditions, the system assigns higher scheduling priority to their associated data channels, increasing their weight in memory swapping and computing resource allocation. The system monitors the power distribution of the energy storage power station in real time, identifying the top 20% of battery clusters with the highest current density. The data channels corresponding to these high-risk areas are marked with high priority in the fluid thermal composite chain. When the utilization rate of computing resources reaches the 85% warning threshold, the scheduling algorithm forcibly suspends low-priority static channel simulation tasks, prioritizing the streaming computing tasks of high-load channels. This differentiated processing strategy ensures that, with limited computing resources, the thermal control simulation of key risk areas can be updated in the most timely manner.

[0053] The method also involves modeling the aging characteristics of the cooling medium. As operating time increases, the physical properties of the submerged coolant, such as viscosity and thermal conductivity, slowly shift. The system periodically fine-tunes the global physical properties in the fluid-thermal composite chain using a built-in chemical property evolution model. This periodic fine-tuning logic is based on the coolant's cumulative operating time, the number of cycles at the highest temperature experienced, and a preset aging degradation coefficient. The determination of the aging degradation coefficient combines accelerated aging test data from the laboratory with field operation and maintenance data, forming a physical property database that evolves over time. For example, with increasing service life, the dynamic viscosity of the coolant typically exhibits a non-linear growth trend, which leads to an increase in pressure drop within the flow channel, thus affecting the heat dependence weight. The system injects the latest physical property parameters into each node of the fluid-thermal composite chain through an annual physical property calibration task. This long-term performance consideration enables the digital twin model to cover the entire lifecycle of the energy storage system, providing a scientific basis for the optimization and upgrading of the thermal management system.

[0054] The simulation process is performed within a distributed computing architecture. Different partitions of the energy storage module are assigned to different processing units, which exchange edge data of the directed heat flux graph via a high-speed backplane bus. To ensure timing consistency, the system employs a synchronization protocol based on a global logical clock. Each processing unit sends a synchronization signal to the central management unit after completing the simulation step for its current partition. The system only proceeds to the next simulation moment after all partitions have completed the calculation for the current step and exchanged boundary heat flux data. During the distributed exchange process, the system identifies nodes at partition boundaries whose directed edges span different physical processing units. The system establishes a dedicated high-speed exchange channel on the backplane bus to transmit temperature gradient and flow velocity vector information of these cross-partition boundary nodes in real time. This parallel processing scheme shortens the simulation time for ultra-large-scale energy storage power plants.

[0055] Example 2: Based on the immersion energy storage thermal control digital twin modeling and simulation method described in Example 1, this example further discloses the data processing and simulation details under extremely non-uniform flow field conditions. To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0056] In complex energy storage power station environments, extremely non-uniform flow fields can arise due to the local complexity of flow channel design or obstruction by foreign objects. To address this, the fluid thermal heterogeneity data index constructed in step 1 demonstrates high flexibility. The system integrates a flow regime discrimination engine that fine-tunes the flow path descriptor. When a pressure sensor detects a rapid fluctuation in the inlet and outlet pressure difference of an independent flow channel unit, the system determines that a local eddy or dead zone has appeared in that area. The system then dynamically splits the topology index of that area, dividing the originally single flow path node into multiple parallel child nodes to simulate the flow around the dead zone.

[0057] In step 2, the definition of the fluid-thermal composite linked list further expands its descriptive capabilities for localized unsteady flows. For these fission-related child nodes, the system allocates a special shadow address space in the storage medium. The shadow address space is logically parallel to the main channel address space, but remains physically compact. The head node dynamically adds branch pointers pointing to these shadow nodes.

[0058] In the computational logic, the initial state of the shadow node not only references the coolant inlet parameters but also considers the velocity distribution ratio of the main flow channel node at the branch. The system determines the mass flow rate allocated to each child node by calculating the flow resistance coefficient at the branch. This approach can accurately capture local heat accumulation phenomena and prevent the digital twin model from generating prediction biases under complex flow conditions.

[0059] In the directed heat flow graph mapping mechanism of step 3, a nonlinear heat-dependent weight calibration mechanism is introduced for non-uniform flow fields. Because heat transfer in eddy or flow-around regions is no longer simply unidirectional convection, but also includes complex turbulent diffusion components, the system calculates local turbulence intensity and superimposes the heat-dependent weights in multiple dimensions. The system extracts the fluctuating value of local flow velocity and defines the ratio of this value to the average flow velocity as the turbulence intensity. The final value of the heat-dependent weight is equal to the weighted average of the basic convection weight and the turbulent diffusion weight. The calculation logic for the turbulent diffusion weight is: taking the ratio of the fluid turbulent viscosity to the Prandtl number, and then multiplying it by the temperature gradient distribution coefficient between nodes. This refined weight configuration scheme enables the directed heat flow graph to accurately reproduce the heat exchange intensity of each micro-region within the submerged energy storage chamber, improving the accuracy of early warning for local high-temperature hotspots.

[0060] During the streaming simulation in step 4, an adaptive time step management strategy was implemented to address the aforementioned complex operating conditions. When the simulation engine detects a drastic change in the thermal dependency weight of a sub-region in the thermal flux directed graph, it automatically densifies the calculation step size for that region. For example, the global simulation step size is shortened, while the original step size is maintained for other regions in a thermally stable state. To ensure global synchronization, a time alignment bit is set in the head node of the fluid-thermal composite linked list. Only after the high-frequency sampled sub-regions have accumulated a time span equal to the global step size will the system uniformly update the state of all nodes in the linked list. This process fully utilizes the parallel processing capability of the cache, ensuring simulation fidelity under critical transient conditions by allocating different clock domains in the on-chip cache.

[0061] Furthermore, the multi-level buffer management mechanism in this embodiment triggers predictive prefetching under extreme conditions. When the system detects a rapid temperature rise trend on the heat propagation path, the simulation engine predicts the range of nodes that may be affected in the next three simulation steps based on the current heat dependence weight gradient. The system will exceed the preset prefetch quantity limit and push all node data on the potentially risky path into the buffer. This prefetching operation is based on a probabilistic graphical model, where the system calculates the probability of heat propagation to each downstream node, and only when the probability value exceeds the limit will forced prefetching be performed. This intelligent buffer management scheme ensures computational continuity during the thermal runaway propagation process without increasing the bus load.

[0062] Example 3: Based on the foregoing examples, this example details the implementation scheme of the present invention in ultra-large-scale distributed energy storage clusters. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0063] For ultra-large-scale energy storage power stations containing tens of thousands of battery modules, the data scale of digital twin systems is exploding. In step 1, the fluid thermal heterogeneous data index adopts a hierarchical indexing architecture. The system divides the entire power station into multiple energy storage compartment-level indices and branch-level indices. Each level of index follows the principle of prioritizing flow field topology. The energy storage compartment-level index is responsible for describing the distribution topology of coolant in the main pipeline, while the branch-level index delves into the microchannel logic inside each liquid-cooled plate. This hierarchical indexing structure allows the system to flexibly switch simulation accuracy according to the current simulation task requirements. In the macro-energy efficiency assessment mode, the system only loads the energy storage compartment-level index, simplifying each branch to a thermal equivalent node; while in the safety warning mode, the system fully deploys the branch-level index.

[0064] In the fluid-thermal composite linked list defined in step 2, virtual node technology is introduced to handle complex pipeline connection logic. Virtual connection nodes are set up at the interfaces between energy storage compartments or between different battery branches. These nodes do not correspond to physical battery cells, but store the thermodynamic boundary conditions for cross-regional fluid exchange, such as the average temperature of the mixed fluid and total pressure loss. At the underlying storage medium, these virtual nodes serve as the glue points for each level of the linked list, enabling cross-link access to data through a globally unified address space. When the simulation engine scans a virtual node, it triggers a remote procedure call to synchronize the latest flow data from the associated processing unit.

[0065] In the heat flow directed graph described in step 3, a compressed storage algorithm with heat-dependent weights is introduced for large-scale clusters. Since large-scale systems contain numerous units with similar geometric structures and flow channel characteristics, the system identifies these topological similarities and establishes weight templates for similar units. Each data node no longer stores the complete set of weight parameters, but only stores the index value pointing to the weight template and local fine-tuning deviation values. This method reduces the storage overhead of the heat flow directed graph while maintaining accuracy, allowing the digital twin model of ultra-large-scale systems to reside entirely in the server's physical memory, avoiding frequent virtual memory swapping.

[0066] In step 4, when performing distributed streaming simulation, this embodiment employs a dynamic task balancing strategy. The system's internal task monitor continuously evaluates the computational load and data throughput of each processing unit. Due to variations in flow field complexity and battery operating conditions within each partition, computation time differs. When a processing unit experiences severe computational lag due to handling in-depth simulations of localized thermal runaway, the task balancing strategy automatically migrates its less affected downstream flow channel nodes to processing units with lower loads. During the migration, the system uses a high-speed backplane bus to synchronize physical page copies of the fluid-thermal composite linked list and instantly updates the logical pointers in the fluid-thermal composite linked list upon completion of the migration. This dynamic load balancing mechanism eliminates the long-tail effect in distributed computing, ensuring the global real-time performance of the entire energy storage power station's digital twin system.

[0067] Furthermore, in ultra-large-scale clusters, the historical data alignment module evolves into a machine learning-based self-evolving calibration center. This self-evolving calibration center aggregates historical operational data from all sensors across the entire station, constructing high-dimensional feature vectors to describe the evolution of the flow field and thermal control state. When deviations occur between physically measured values ​​and simulated predictions, the calibration center not only corrects the current thermal dependence weights but also uses pattern recognition technology to identify whether such deviations are caused by a common equipment aging or design problem. If it is a common deviation, the system will batch update the parameter settings of all energy storage compartments of the same model. This collective learning capability enables the modeling accuracy of the digital twin system to continuously improve with increasing operating time.

[0068] The system also integrates an environmental coupling simulation module when executing the method. This module considers convective heat transfer between the energy storage chamber shell and the outside atmosphere, as well as the impact of solar radiation on surface temperature. The system transforms external environmental parameters into dynamic boundary conditions for the head nodes of the fluid-thermal composite chain table. For example, it updates the initial return water temperature of the coolant at the radiator end based on real-time meteorological data. By incorporating external environmental factors into the flow field topology indexing system, the digital twin model possesses simulation adaptability across climate zones and in all weather conditions, providing a standardized digital foundation for thermal management of energy storage power plants in different environments worldwide.

[0069] In its implementation, all textual logic expressions aim to accurately characterize complex physical phenomena. For example, when describing the reduction effect of local resistance of coolant at flow channel bends on the heat dependence weight, the system identifies the curvature parameter in the flow path descriptor, calculates the local kinetic energy loss coefficient, and accordingly reduces the positive contribution of flow velocity attributes to the heat dependence weight. This deep deduction based on pure textual logic covers all key physical constraints such as fluid mechanical energy loss, heat exchange rate, and mass conservation, constructing a rigorous digital twin logic closed loop.

[0070] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A digital twin modeling and simulation method for thermal control of immersion energy storage, characterized in that, Includes the following steps: Step 1: Construct a fluid thermal heterogeneous data index based on flow field topology. By analyzing the physical flow path of the cooling medium inside the energy storage chamber, establish a mapping relationship between data storage logic and physical heat transfer logic. Step 2: Define the fluid thermal composite linked list structure, taking the coolant flow path as the logical main line, and arrange the thermal data nodes of the battery module that are physically located upstream and downstream of the same flow channel in the storage medium in a physical continuous manner. Step 3: Establish a directed graph mapping mechanism for heat flow, abstract the complex flow channels in the submerged energy storage system into a graph structure in the database, and configure temperature attributes, flow rate attributes and heat dependence weights for each data node; Step 4: Perform flow simulation based on flow field topology index. When the digital twin model updates its thermal state, the system directly retrieves the physically continuous heat propagation link data to the cache unit to complete the real-time calculation of the thermal control state.

2. The digital twin modeling and simulation method for immersion energy storage thermal control according to claim 1, characterized in that, The construction of the fluid thermal heterogeneous data index based on the flow field topology includes: dividing the immersion liquid cooling space in the energy storage chamber into several independent flow channel units, each of which contains a preset number of battery modules. A physical flow path descriptor is generated by identifying the inlet location of the cooling medium entering the energy storage compartment, the order in which it flows through each battery module, and the outlet location of it exiting the energy storage compartment. A topology indexing pattern with the physical flow path descriptor as the core weight is adopted. The first node and the second node on the same flow path are assigned adjacent logical addresses in the indexing logic, so that the order of data retrieval is consistent with the physical process of fluid flow.

3. The digital twin modeling and simulation method for immersion energy storage thermal control according to claim 2, characterized in that, The construction of the fluid thermal heterogeneous data index based on the flow field topology also includes the dynamic reconstruction of the flow field topology: by monitoring the data of pressure sensors and flow sensors distributed at key locations in the flow channel, the physical flow path descriptor is corrected in real time. Based on the feedback values ​​from the sensors, the topological connection relationship between nodes is recalculated, and the logical pointer of the fluid thermal composite list is updated in the background. The flow field identification module polls the values ​​of each pressure monitoring point at a preset time period. When the pressure difference between two adjacent monitoring points exceeds the preset disturbance threshold, it determines that the flow channel state has changed and starts the topology verification program. The topology verification program re-identifies the main flow path of the coolant by calculating the flow velocity distribution of the fluid under the current pressure gradient, and incrementally updates the physical flow path descriptors that are affected.

4. The digital twin modeling and simulation method for immersion energy storage thermal control according to claim 3, characterized in that, The defined fluid thermal composite linked list structure includes the organization of the underlying physical pages of the storage medium: the fluid thermal composite linked list includes a header node and several subsequent nodes, the header node stores the initial state parameters of the coolant inlet, the initial state parameters include the initial temperature, initial pressure and initial mass flow rate; Each subsequent node corresponds to a discretized flow channel region or a battery module heating surface; At the storage level, the system allocates a contiguous address space to store node data belonging to the same channel. Each node encapsulates a logical pointer to the physical downstream node, enabling the simulation engine to address the next affected node in a step-by-step manner after completing the calculation of the current node. During the storage allocation phase, memory alignment technology is used to ensure that the starting address of each node is located at the starting boundary of the computer's memory page.

5. The immersion energy storage thermal control digital twin modeling and simulation method according to claim 4, characterized in that, Each node of the fluid thermal composite linked list includes: a temperature field, a flow rate field, and a preset health status flag. The health status flag is compared with the real-time temperature collected by the sensor and a preset safety threshold. When the temperature of the battery module corresponding to a certain node exceeds the predetermined range, the health status flag will flip. If the simulation engine detects an abnormality in the health status flag bit during streaming read, it will automatically trigger a deep simulation mode to increase the calculation accuracy and sampling frequency of the node and its associated nodes. The node also stores the local Reynolds number, Nusselt number, and convective heat transfer area of ​​the cooling medium in the corresponding region.

6. The digital twin modeling and simulation method for immersion energy storage thermal control according to claim 5, characterized in that, The mechanism for establishing a directed heat flow graph includes: the directed heat flow graph consists of a set of nodes and a set of directed edges. Each element in the set of nodes represents the thermal state of the smallest simulation unit, and the set of directed edges represents the direction of thermal energy migration with the fluid. Each directed edge is assigned a thermal dependency weight, which is used to characterize the degree of influence of the thermal changes of the upstream node on the thermal state of the downstream node. The value of the heat dependence weight is determined by the specific heat capacity at constant pressure of the coolant, the cross-sectional area of ​​the flow path, the physical displacement between nodes, and the real-time flow velocity. The calculation logic of the heat dependence weight is described as follows: the heat dependence weight of the current node is equal to the product of the coolant flow rate and the time required for the coolant to flow through the node, and then divided by the total heat capacity of the fluid medium contained in the node.

7. The digital twin modeling and simulation method for immersion energy storage thermal control according to claim 6, characterized in that, The update process of the heat-dependent weights includes: when the rate of change of the physical sensor values ​​exceeds the range, initiating a weight recalibration task; The recalibration task involves a textual logical deduction of the Reynolds number of the flow field, that is, judging whether the flow regime has changed from laminar to turbulent based on the ratio of the product of fluid density, flow velocity and flow channel characteristic length to the fluid dynamic viscosity. If the flow regime changes, the system automatically adjusts the convective heat transfer coefficient term in the heat dependence weights; When the flow is in a laminar state, the convective heat transfer coefficient is affected by the fluid thermal conductivity and the flow channel geometry. When the flow state changes to turbulence, the system adds a turbulence intensity correction term, the value of which is directly proportional to the preset exponent of the Reynolds number.

8. The digital twin modeling and simulation method for immersion energy storage thermal control according to claim 7, characterized in that, The execution of the flow field topology index-based streaming simulation includes: when the digital twin system detects a change in the operating conditions of the physical entity, the simulation engine initiates a thermal runaway simulation task and generates prefetch instructions; Based on the physical contiguous address provided by the fluid thermal composite linked list, the starting affected node in the directed heat flow graph is located, and the data of this node and a predetermined number of downstream nodes are loaded into the processor's on-chip cache at once. The temperature attribute of each node is updated sequentially according to the pointing order defined in the heat flow directed graph. When updating each node, the temperature value of its predecessor node is extracted, multiplied by the corresponding thermal dependence weight, and superimposed with the heat generation rate of the battery module inside the current node to obtain the predicted temperature of the current node in the next simulation step.

9. The digital twin modeling and simulation method for immersion energy storage thermal control according to claim 8, characterized in that, The execution of the flow field topology index-based flow simulation also includes configuring a multi-level buffer management mechanism: when loading heat propagation link data, the system preloads key node data of adjacent flow channels according to the lateral coupling relationship in the directed heat flow graph; The lateral coupling occurs on the heat conduction path between battery modules. The lateral coupling weight is determined by calculating the reciprocal of the lateral thermal resistance, which is determined by the thermal conductivity, thickness, and contact area of ​​the insulating layer material between adjacent battery modules. During the simulation, the temperature update of the current node is composed of the convective influence of the upstream fluid and the heat conduction contribution from adjacent nodes. Calculate the temperature difference between the current node and its neighboring nodes, multiply the temperature difference by the lateral coupling weight, and calculate the heat exchange amount based on the time length of the simulation step.

10. The digital twin modeling and simulation method for immersion energy storage thermal control according to claim 9, characterized in that, The method also includes historical data alignment and distributed computing scheduling: the predicted values ​​obtained from simulation are compared with the measured values ​​fed back by physical sensors periodically. If the deviation between the two exceeds the preset fault tolerance range, the feedback adjustment program is started. The feedback regulation program corrects the heat dependence weights in the heat flow directed graph or adjusts the heat capacity parameters in the fluid heat complex chain list; A priority-based scheduling algorithm is adopted. For battery branches in high-load charging and discharging state, their associated data channels are given higher scheduling priority, and their weight in memory swapping and computing resource allocation is increased. In the distributed computing architecture, different partitions of the energy storage compartment are assigned to different processing units, and the processing units exchange edge data of the heat flow directed graph through a high-speed backplane bus. A synchronization protocol based on a global logic clock is adopted. After all processing units have completed the simulation step size of the current partition and exchanged the boundary heat flux data, the simulation will proceed to the next simulation moment.

Images