Immersed phase change liquid cooling battery thermal load dynamic modeling and prediction method

Abstract

The invention relates to the technical field of phase change liquid cooling energy storage, in particular to an immersed phase change liquid cooling battery thermal load dynamic modeling and prediction method, and the method comprises the steps: dividing the surface of a battery into n independent calculation regions based on the geometric structure and thermal characteristics of a battery module, static thermal resistance coefficients, dynamic thermal capacity coefficients and thermal diffusion weight coefficients of all the areas are simulated and calibrated through experiments and computational fluid mechanics, and a calibrated parameter matrix is stored in a nonvolatile memory of a BMS; the method comprises the following steps: acquiring battery temperature data at high frequency through a multi-source sensor network arranged in each partition, performing Kalman filtering and derivation calculation through a preprocessing module of a data processing layer to obtain a temperature change rate, calling a pre-stored coefficient by a regionalization modeling module, and calculating an instantaneous thermal load of each partition and a total thermal load of the module; a traditional'uniform temperature body 'hypothesis is broken through, a battery module is divided into a plurality of subareas, and the device is particularly suitable for capturing a non-uniform heat phenomenon caused by non-uniform flow field and local bubble generation under an immersed liquid cooling condition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of phase change liquid cooling energy storage, more particularly to a method for dynamic modeling and prediction of battery thermal load of immersed phase change liquid cooling. BACKGROUND

[0002] With the promotion of the "double carbon" strategy and the rapid development of energy storage industry, electrochemical energy storage (such as lithium-ion batteries, supercapacitors, etc.) is widely used in power grid frequency modulation, electric vehicle fast charging and aerospace power systems. In these application scenarios, the battery needs to withstand high-rate charging and discharging, and the instantaneous heat power increases sharply, which easily leads to thermal runaway risk.

[0003] To cope with strong heat shock, a variety of thermal management methods have been proposed. Among them, the immersed phase change liquid cooling technology has become a cutting-edge solution due to its extremely high heat transfer efficiency. This method directly immerses the battery in a dielectric phase change liquid, and uses the liquid boiling to absorb latent heat to achieve heat dissipation. However, this technology brings the following modeling problems:

[0004] Strong nonlinear heat generation behavior: The heat generation of the battery at high rate is not linear, and the internal resistance heat, polarization heat and reaction heat change sharply with temperature, SOC (state of charge) and current. The traditional thermal model based on steady-state assumption has a large error.

[0005] Dynamic mutation of local heat dissipation conditions: In the process of phase change cooling, bubbles are randomly nucleated, grown, detached and converged on the surface of the battery. The presence of bubbles will dramatically change the heat transfer coefficient of the local surface (for example, the heat transfer capacity of the gas film covered area is much lower than that of the liquid wetting area), resulting in a dramatic, rapid and random dynamic change of the heat dissipation conditions on the microscale. The existing model based on limited temperature measurement points or average heat transfer coefficient cannot capture such mutations.

[0006] The current battery management system (BMS) mostly uses simplified thermal models such as average temperature or first-order RC, which cannot reflect the local transient characteristics, resulting in lag or failure of the thermal control strategy.

[0007] Therefore, there is an urgent need for a calculation and prediction method that can accurately and dynamically describe the real thermal state of the battery in the immersed phase change liquid cooling system in real time, laying the foundation for an intelligent thermal management system from "passive response" to "active prediction and adaptive adjustment". SUMMARY

[0008] Therefore, the present application provides a method for dynamic modeling and prediction of battery thermal load of immersed phase change liquid cooling, which aims to solve the above technical problems.

[0009] To achieve the above purpose, the present application adopts the following technical solutions:

[0010] The application discloses a battery thermal load dynamic modeling and prediction method based on an immersion phase change liquid cooling, and comprises the following steps:

[0011] In step S1, based on the geometric structure and thermal characteristics of the battery module, the battery surface is divided into n independent calculation regions, and the static thermal resistance coefficient, dynamic heat capacity coefficient and thermal diffusion weight coefficient of each region are calibrated through experiment and computational fluid dynamics simulation, and the calibrated parameter matrix is stored in the nonvolatile memory of the BMS.

[0012] In step S2, the battery temperature data are collected at a high frequency by a multi-source sensor network arranged in each subregion, the temperature change rate is obtained through Kalman filtering and derivation calculation of a preprocessing module of a data processing layer, the pre-stored coefficient is called by a regional modeling module, and the instantaneous thermal load of each subregion and the total thermal load of the module are calculated.

[0013] In step S3, a long short-term memory network model arranged in the BMS or an edge computing device is called by a prediction module, the historical thermal load data sequence cached in a time series database is taken as input, and the future short-term thermal load prediction value is output.

[0014] In step S4, the thermal load prediction value is sent to a feedforward adjustment cooling liquid flow or system pressure of a thermal management system and uploaded to an energy management system to optimize a charging and discharging power distribution strategy through a control interface provided by an output layer, so that a closed-loop control is formed.

[0015] Further, in step S2, the instantaneous thermal load of each subregion is calculated through the following formula:

[0016]

[0017] In the formula, k 1,i is a static thermal resistance coefficient, T i is a real-time temperature of a subregion, T ref is a reference temperature, is a temperature change rate of the subregion, k 2,i is a dynamic heat capacity coefficient, W i is a thermal diffusion weight coefficient.

[0018] Further, in step S2, the principle of dividing the n independent calculation regions comprises a geometric feature driven subregion and a heat dissipation path driven subregion, wherein the geometric feature driven subregion is divided into different regions according to the physical structure of the battery, and the regions such as long edges, short edges, tab positions and pole nearby parts with complex geometry are divided into different regions, and each region corresponds to at least one temperature sensor; and the heat dissipation path driven subregion is divided into different regions according to the cooling liquid contact surface and flow field distribution, and the regions with large liquid flow rate and flow dead angle are divided into different regions.

[0019] Furthermore, the calibration method for the static thermal resistance coefficient specifically involves applying a constant power P in a constant temperature environment, using a simulated heat source of the same size as the battery cell, placed close to the i-th region. heat Simultaneously, a high-precision thermal imager was used to monitor the steady-state temperature T in the area. steady Through formula k 1,i =P heat / (T steady -T ambient T was calculated to obtain the result. ambient The ambient temperature.

[0020] Furthermore, the calibration method for the thermal diffusion weight coefficient is as follows: a three-dimensional two-phase flow simulation of the complete battery box is performed using CFD software to simulate the flow and heat transfer of liquid and vapor, and a local heating source is set up in the actual liquid cooling tank for experimental auxiliary measurement. The thermal diffusion weight matrix between regions is obtained by calculation.

[0021] Furthermore, the Long Short-Term Memory (LSTM) network model is a two-layer stacked structure trained offline. The first layer LSTM unit is used to capture short-term temperature fluctuations caused by bubble generation and detachment, and the second layer LSTM unit is used to extract medium-term thermal inertia and phase transition hysteresis effects. The model is trained using weighted mean square error as the loss function and assigns higher weights to short-term peak regions.

[0022] Furthermore, the system is coupled to battery cells, modules, and battery packs immersed in phase change coolant, and is communicatively connected to the thermal management system (TMS) and energy management system (EMS), including:

[0023] The sensing layer consists of multiple high-precision sensors, including an infrared matrix sensor, a fiber optic temperature sensor, and an embedded thermocouple for monitoring the surface and internal temperature of the battery, as well as sensors for collecting coolant pressure, flow rate, and level. The sensing layer is connected to the system to obtain electrical parameters such as current, SOC, and SOH. The output of the sensing layer is communicatively connected to the input of the data processing layer.

[0024] The data processing layer input is connected to the sensing layer output, receiving and processing sensor data. The data processing layer includes:

[0025] A preprocessing module for Kalman filtering, outlier rejection and time synchronization processing of raw sensor data is connected to the output end of the preprocessing module, a regional modeling module for managing partition parameter library and performing partition heat load calculation is connected to the regional modeling module, a heat diffusion coupling module for realizing dynamic heat transfer between partitions is connected to the heat diffusion coupling module, an adaptive fusion and smoothing module for smoothing heat load data is connected to the output end of the adaptive fusion and smoothing module and accesses a time series database, a prediction module for running an LSTM prediction model is connected to the output end of the data processing layer, an output layer for receiving real-time heat load values and prediction values, and the output layer output end is connected to the TMS and EMS through a control interface, and is used for outputting local heat load fields of each partition, module total heat load and prediction values.

[0026] Further, the prediction module in the data processing layer is deployed in the memory of the master MCU of the battery management system or the edge computing device connected thereto, the LSTM model is pre-trained and solidified into a model file stored in the non-volatile memory of the device, and is loaded into the cache during real-time operation to realize the prediction function.

[0027] Further, the heat load prediction value generated by the output layer is connected to the controller of the thermal management system through a first data interface and is connected to the decision unit of the energy management system through a second data interface.

[0028] Further, the inter-regional heat diffusion weight coefficient is a dimensionless weight factor obtained through CFD simulation and experimental calibration, and is used to represent the inter-regional heat coupling strength caused by the flow of cooling liquid.

[0029] Compared with the prior art, the present application has the following positive effects:

[0030] 1. Partition dynamic modeling: breaking through the traditional "uniform temperature body" assumption, the battery module is divided into multiple partitions, which is particularly suitable for capturing non-uniform thermal phenomena caused by uneven flow field and local bubble generation under the condition of immersion liquid cooling.

[0031] 2. Multi-physical field coupling: unifying static thermal resistance (steady-state heat conduction), dynamic heat capacity (transient heat storage) and regional heat diffusion (space heat transfer driven by cooling liquid) in one calculation framework, which can fully reflect the complex thermal behavior under the condition of immersion liquid cooling system.

[0032] 3. Optimization modeling for phase change cooling: the diffusion weight obtained through CFD simulation explicitly considers the cooling liquid circulation and bubble phase change characteristics, so that the model can accurately capture the heat dissipation mutation in typical immersion cooling scenarios such as boiling heat transfer and bubble detachment.

[0033] 4. Intelligent predictive thermal management: combined with the LSTM prediction model, the future thermal load is predicted in advance, and the predicted value is input as a feedforward signal into the liquid cooling system and energy management system, so that the immersion liquid cooling system can realize active optimization in terms of air pressure regulation, pump speed control, etc., avoid temperature shock and improve cooling efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only a part of the embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0035] Figure 1 The system workflow diagram of the present application.

[0036] Figure 2 The dynamic thermal load modeling logic flowchart of the present application. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0038] Referring to the drawings Figures 1-2 In order to make the above-mentioned purposes of the present application more obvious and easy to understand, and to embody the characteristics and advantages of the inventive method, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0039] The present application proposes a battery thermal load dynamic modeling and prediction method specially used for immersion phase change liquid cooling energy storage system. The method is based on the geometric structure and thermal characteristics of the battery module, divides the battery surface into several independent calculation regions, and establishes a multi-physical field coupling model considering static thermal resistance effect, dynamic thermal capacity effect and inter-regional heat diffusion effect. Among them, the static thermal resistance coefficient is obtained by experiment calibration, which is used to characterize the steady-state heat dissipation capacity of the battery material; the dynamic thermal capacity coefficient reflects the transient temperature response characteristics of the battery under high-rate charging and discharging; the inter-regional heat diffusion weight is determined by computational fluid dynamics (CFD) simulation, which is used to quantitatively describe the influence of phase change cooling liquid flow on local heat transfer. Through the weighted calculation of the thermal behavior of each region, the local thermal load can be dynamically solved and the total thermal load of the module can be integrated, so as to accurately depict the sudden change of local heat dissipation condition caused by phase change characteristics such as bubble generation and detachment, and significantly improve the modeling accuracy under extreme transient working conditions.

[0040] On this basis, the application further introduces a deep learning-based forward-looking prediction mechanism. The system caches historical thermal load curves through a time series database, and calls a long short-term memory (LSTM) model deployed in a BMS or an edge computing device to realize prediction of future short-term thermal load trends. The prediction value can not only reveal the change trend of the thermal load in advance, but also be used as a feedforward input to optimize the operation strategy of a thermal management system (TMS) and an energy management system (EMS); the EMS can optimize the charging and discharging power distribution in combination with the prediction value to avoid overheating risks caused by limited heat dissipation, and improve energy efficiency and battery life while ensuring safety. Through the fusion of "real-time modeling + trend prediction", the application realizes a leap from passive response to active regulation, and provides technical support for intelligent control of the immersion phase change liquid cooling system.

[0041] The system workflow consists of four levels: first, the data perception and processing layer collects the battery operating state through temperature, pressure, flow and electrical parameters and the like of multiple source sensors, and filters, excludes abnormalities and synchronously processes the original data to ensure the accuracy of the input data. Subsequently, the dynamic thermal load modeling layer models the battery module based on the pre-calibrated static thermal resistance, dynamic thermal capacity and inter-regional thermal diffusion weight, calculates the instantaneous thermal load of each region and the whole in real time, and outputs a stable total thermal load value after smoothing processing, as shown in the thermal model flowchart. Figure 1 Then, the intelligent prediction and strategy optimization layer uses the historical thermal load data stored in the time series database to call the LSTM prediction model deployed in the BMS or the edge computing device to predict the future short-term thermal load trend, and uses the result to optimize the feedforward control of the TMS and the power scheduling of the EMS, so as to realize advanced active regulation. Finally, in the execution and feedback layer, the optimized strategy is implemented into the cooling system and the battery power scheduling through the thermal management and power management actuators, and the battery operating state is fed back to the perception layer through the sensor, forming a closed loop regulation to realize efficient, safe and intelligent battery thermal management. The system workflow diagram is shown in Figure 1 The dynamic thermal load modeling logic flowchart is shown in Figure 2 .

[0042] System architecture and functional level: The battery cells, modules and battery packs measured are immersed in the phase change cooling liquid. The perception layer realizes comprehensive monitoring through multiple source high-precision sensors, including an infrared matrix, a fiber optic temperature sensor, a buried thermocouple (covering the surface and internal key nodes); the cooling liquid pressure, flow and liquid level are collected; the BMS electrical parameters (current, SOC, SOH) can be used for thermal load modeling and prediction assistance.

[0043] The data processing layer includes a pre-processing module that performs Kalman filtering, outlier rejection, and time synchronization processing on the raw sensor data to generate smooth and reliable temperature sequences, a regional modeling module that grids the battery surface into independent calculation regions and manages the static thermal resistance, dynamic thermal capacity, and adjacency weight coefficients of each region through a parameter library, a thermal diffusion coupling module that uses the region adjacency weights obtained through CFD simulation and experimental calibration to realize the coupling calculation of inter-regional dynamic heat transfer, and an adaptive fusion and smoothing module that further smooths the local and total thermal loads through mutation detection, moving average, and exponential filtering to ensure data stability.

[0044] The prediction module uses historical thermal load data sequences to input an LSTM model deployed in the BMS or edge computing to predict future short-term thermal load Q pred (t+Δt), which provides the basis for the feedforward control and strategy optimization of the thermal management system (TMS) and energy management system (EMS), generates the local thermal load field q i (t) and the total thermal load Q(t) of the module / battery pack, and provides a real-time control interface for liquid cooling system flow, air pressure adjustment, and charge / discharge power optimization, realizing the intelligent regulation and control of "real-time modeling + trend prediction" in the output layer.

[0045] Method flow

[0046] System initialization and parameter pre-calibration (offline), which mainly completes the battery geometric modeling, partitioning, and experimental and simulation calibration of key parameters, laying the foundation for real-time operation.

[0047] 1. Battery module partitioning principles and methods

[0048] Geometric feature-driven partitioning: based on the physical structure of the battery (such as the long side, short side, and tab position of a square cell), the battery is divided into multiple regions. Parts with complex local geometry (such as the vicinity of the pole) are separately partitioned to ensure uniformity of the temperature field within the partition. Each region corresponds to at least one temperature sensor

[0049] Heat dissipation path-driven partitioning: considering the cooling liquid contact surface and flow field distribution, regions with different cooling intensities are independently partitioned. Regions with similar thermal characteristics can be appropriately merged. For example, regions with high liquid flow rates and flow rate dead zones should be partitioned separately.

[0050] Scale and precision balance: the number of partitions n is positively correlated with the computational complexity. The typical range is several tens to several hundred partitions (such as a 16x16 array = 256 partitions), which ensures model accuracy while avoiding computational overload.

[0051] 2. Key coefficient calibration method

[0052] a. Static thermal resistance coefficient k 1,i Calibration:

[0053] Physical meaning: Characterize the heat dissipation capability of the i-th region in unit temperature rise (W / ℃) in steady state. It reflects the thermal conductivity of the battery material in this region, the contact with the cooling liquid, etc.

[0054] Calibration method: In a constant temperature environment, use a simulated heat source (such as a thin film heating sheet) the same size as the battery monomer to tightly adhere to the i-th region, and apply a constant power P heat , while monitoring the steady-state temperature T steady of the region with a high-precision thermal imager. Calculate it by the formula k 1,i = P heat / (T steady -T ambient ), T ambient is the ambient temperature. This process needs to be repeated in each region or representative region.

[0055] b. Dynamic thermal capacity coefficient k 2,i calibration:

[0056] Physical meaning: Characterize the thermal inertia of the i-th region, i.e. the heat power required for unit temperature change rate (J / ℃·s or W·s / ℃). It reflects the specific heat capacity and mass of the battery in this region.

[0057] Calibration method: Apply a known step current (such as from 0 to maximum current) to the battery, and simultaneously collect the temperature change curve of the i-th region at high speed. By analyzing the temperature rise rate of the region and the known instantaneous heat generation power, use parameter identification algorithms (such as least squares method) to fit k 2,i .

[0058] c. Thermal diffusion weight coefficient calibration:

[0059] Physical meaning: Quantify the degree of influence of the heat of the j-th region on the i-th region due to the flow of the cooling liquid. It is a dimensionless weight factor.

[0060] Calibration method: Use computational fluid dynamics (CFD) software (such as Fluent, Star-CCM+, etc.) to simulate three-dimensional two-phase flow (liquid, vapor) in the complete battery box. In the simulation, set a certain region as a heat source, observe the temperature response of other regions, and combine experimental assistance to set a local heating source in the actual liquid cooling tank to measure the temperature rise delay and amplitude of adjacent sensors, in order to correct the CFD results. Finally, the thermal diffusion weight matrix between regions can be calculated, which will be pre-stored in the BMS.

[0061] 3. Parameter storage and model initialization

[0062] The parameters are stored in the BMS non-volatile memory in the form of a matrix. The model is loaded into the MCU cache during initialization, ensuring that the real-time calculation delay is less than 10 ms, meeting the high-rate operating condition requirements.

[0063] As shown in the flowchart Figure 1 , the system performs the following steps in the normal operating state in a loop:

[0064] 1. Real-time temperature acquisition and preprocessing:

[0065] The temperature data T i (i = 1, 2, …, n) of all partitions are collected at a high frequency (sampling rate ≥ 1 Hz) by the arranged multi-source sensor network (such as infrared thermal imager, distributed optical fiber temperature sensor, etc.). The raw data collected need to be immediately processed by Kalman filter to eliminate environmental noise and measurement error, obtaining smooth and reliable temperature values.

[0066] 2. Calculate the temperature change rate:

[0067] The derivative operation is performed on the filtered temperature data sequence of each partition, using the difference method: dT i / dt ≈ (T i (t) - T i (t - Δt)) / Δt), and the instantaneous temperature change rate of each region is calculated in real time.

[0068] 3. Dynamic heat load modeling and calculation:

[0069] The central processing unit (such as the high-performance MCU in the BMS) calls the pre-stored coefficients k 1,i , k 2,i , and the heat diffusion weight matrix to perform the following core calculation formula:

[0070] a. Partition-level dynamic equation

[0071] The instantaneous heat load calculation formula of each partition is:

[0072]

[0073] In the formula, k 1,i is the static thermal resistance coefficient, T i is the real-time temperature of the partition, T ref is the reference temperature, is the temperature change rate of the partition, k 2,i is the dynamic heat capacity coefficient, and W i is the heat diffusion weight coefficient.

[0074] b. Total heat load calculation

[0075]

[0076] c. Smoothing and output

[0077] Transient total heat load Q initial A stable and reliable total heat load value Q can be obtained by performing a moving average filter. The Q value is output in real time to the controller, which can be used for flow or air pressure adjustment of the liquid cooling system, etc. to realize advanced and adaptive precise temperature control.

[0078] The heat load prediction method based on adaptive LSTM introduces a prediction module based on deep learning to realize the functions from "real-time response" to "forward-looking prediction".

[0079] 1. History data cache: the system maintains a time series database to continuously store the high-frequency calculated heat load values Q(t) in a certain time window (such as the past 10 minutes), forming a historical heat load curve.

[0080] 2. Prediction model deployment: deploy a pre-trained long short-term memory (LSTM) model in the BMS or connected edge computing device. The network model structure is optimized, the input is a sequence of past time, and the output is the heat load prediction value Q pred (t+Δt) in the future short term.

[0081] Model training: the LSTM model is trained offline in the cloud or using a large amount of historical operation data (including battery data under different working conditions and different health states), until its prediction accuracy meets the requirements, and then it is solidified as a model file and deployed to the terminal.

[0082] Online prediction: when running in real time, the system automatically inputs the cached historical heat load sequence into the LSTM model, and the model outputs the prediction value Q pred of the future short-term heat load. This prediction value contains the trend information of the heat load.

[0083] Explanation of LSTM model

[0084] In order to adapt to the time-varying characteristics of battery heat load under the immersed phase change liquid cooling system, the LSTM model used in the present application is optimized in structure and training mechanism:

[0085] a. Network structure optimization

[0086] Based on the basic LSTM unit, a double-layer stacking structure is introduced: the first layer is specifically used to capture short-term temperature fluctuations, and the second layer is used to extract medium-term thermal inertia and phase change hysteresis effect. A fully connected layer and a normalization layer are introduced in the output layer to ensure that the prediction result can converge stably and avoid divergence of the heat load value.

[0087] b. Input / output time window design

[0088] The input sequence adopts high-frequency thermal load data Q(t) in the past 5-10 minutes, and the time step is consistent with the system sampling frequency (≥1 Hz) to ensure timely response to rapid thermal disturbances such as bubble generation and detachment of the cooling medium. The output prediction window is set to the thermal load curve Q pred (t+Δt) in the future 10-60 seconds, matching the real-time regulation and response requirements of the thermal management system (TMS) and the energy management system (EMS).

[0089] c. Training and loss function optimization

[0090] Model training is based on multi-source experimental data and simulation data covering different charging / discharging rates, SOC intervals, and cooling liquid working conditions to ensure generalization ability. Weighted mean square error (WMSE) is used as the loss function: higher weight is given to short-term peak areas to improve the prediction accuracy of the model under extreme thermal shock conditions. Early stopping mechanism and cross-validation are used during training to prevent overfitting and ensure prediction reliability under different battery health states.

[0091] Application of prediction function (operation strategy optimization)

[0092] 1. Feedforward control of thermal management system (TMS):

[0093] The predicted value Q pred is sent to the thermal management controller in advance. If it is predicted that the thermal load will rise sharply after 20 seconds, the controller can slightly reduce the system pressure (lower the boiling point in advance) or increase the pump speed to prepare sufficient cooling capacity for the upcoming thermal shock, thereby completely avoiding the occurrence of temperature spikes and achieving smooth control.

[0094] 2. Optimization of power scheduling system:

[0095] The predicted value Q pred is uploaded to the energy management system (EMS). The EMS can optimize the battery's operation strategy more intelligently based on future thermal load predictions:

[0096] Active power limitation: If it is predicted that the future thermal load will exceed the upper limit of the system's heat dissipation capacity, the EMS can gradually reduce the charging or discharging rate in advance and smoothly, rather than stopping urgently when the temperature exceeds the limit. This not only ensures safety, but also improves energy utilization efficiency and user experience.

[0097] Lifetime optimization: Avoiding battery operation at high temperature and high thermal load is key to prolonging battery life. Based on prediction, the system can actively choose to perform high-power charging and discharging during periods of low thermal load, optimizing the long-term use mode of the battery.

[0098] Demand response optimization: When participating in grid frequency regulation services, the system can evaluate the thermal safety risks of executing a high-power instruction, thus more accurately and safely participating in scheduling.

[0099] The core of the present application is to propose a "partition-coupling-dynamic" battery thermal load dynamic modeling method suitable for submerged phase change liquid cooling energy storage system. The method divides the battery module into multiple calculation regions, and combines experimental calibration and CFD simulation to obtain the thermal characteristic parameters of each region, and establishes a coupling model considering static thermal resistance, dynamic thermal capacity and inter-regional thermal diffusion effect. During operation, the system collects temperature data in real time and calculates the thermal load of each partition and the whole, realizing high-precision modeling. Further, based on the prediction mechanism of LSTM network, the historical thermal load curve is used to predict the future trend, and the result is input as a feedforward signal into the thermal management system and the energy management system, so as to realize active heat dissipation regulation and power optimization, complete the transformation from "passive response" to "active regulation", and provide technical support for safe, efficient and long-life operation of the battery system.

[0100] Through the system and method, the following positive effects are achieved:

[0101] 1. Partition dynamic modeling: Breakthrough the traditional "uniform temperature body" assumption, divide the battery module into multiple partitions, especially suitable for capturing non-uniform thermal phenomena caused by uneven flow field and local bubble generation under submerged liquid cooling conditions.

[0102] 2. Multi-physical field coupling: Unifies static thermal resistance (steady-state heat conduction), dynamic thermal capacity (transient heat storage) and regional thermal diffusion (cooling liquid driven space heat transfer) in one calculation framework, which can fully reflect the complex thermal behavior under submerged liquid cooling system.

[0103] 3. Optimization modeling for phase change cooling: The diffusion weight obtained by CFD simulation explicitly considers the cooling liquid circulation and bubble phase change characteristics, so that the model can accurately capture the heat dissipation mutation in typical submerged cooling scenarios such as boiling heat transfer and bubble detachment.

[0104] 4. Intelligent predictive thermal management: Combined with LSTM prediction model, the future thermal load is predicted in advance, and the predicted value is input as a feedforward signal into the liquid cooling system and energy management system, so that the submerged liquid cooling system can realize active optimization in terms of air pressure regulation, pump speed control, etc., avoid temperature shock and improve cooling efficiency.

[0105] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can make equivalent replacement or change according to the technical solution and inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

[0106] The various embodiments described in this specification are intended to be illustrative of the invention and do not limit the scope of the invention. Although specific embodiments have been described herein, they are not to be taken as the only embodiments of the invention. Various modifications can be made to the embodiments described and other embodiments can be used without departing from the spirit or scope of the invention. Accordingly, the scope of the invention is to be limited only by the claims.

Claims

1. A method for dynamic modeling and prediction of battery thermal load of immersion phase change liquid cooling, characterized in that, The method comprises the following steps: Step S1, based on the geometric structure and thermal characteristics of the battery module, the battery surface is divided into n independent calculation regions, and the static thermal resistance coefficient, dynamic thermal capacity coefficient and thermal diffusion weight coefficient of each region are calibrated through experimental and computational fluid dynamics simulation, and the calibrated parameter matrix is stored in the non-volatile memory of the BMS; Step S2, through the multi-source sensor network arranged in each subarea, the battery temperature data is collected at high frequency, after the Kalman filter and derivation calculation of the preprocessing module of the data processing layer, the temperature change rate is obtained, and the pre-stored coefficient is called by the regional modeling module to calculate the instantaneous thermal load of each subarea and the total thermal load of the module; Step S3, the long short-term memory network model deployed in the BMS or edge computing device is called by the prediction module, the historical thermal load data sequence cached in the time series database is taken as input, and the future short-term thermal load prediction value is output; Step S4, the thermal load prediction value is sent to the thermal management system for feedforward regulation of coolant flow or system pressure through the control interface provided by the output layer, and is uploaded to the energy management system to optimize the charging and discharging power distribution strategy, forming a closed loop control.

2. The method of claim 1, wherein, In the step S2, the calculation of the instantaneous thermal load of each subarea is realized by the following formula: where k 1,i is the static thermal resistance coefficient, T i is the zoned real-time temperature, T ref is the reference temperature, is the zoned temperature change rate, k 2,i is the dynamic thermal capacitance coefficient, W i is the thermal diffusion weight coefficient.

3. The method of claim 1, wherein, In the step S2, the principle of dividing the n independent calculation regions includes: geometric feature driven partitioning and heat dissipation path driven partitioning, wherein the geometric feature driven partitioning divides the parts with complex geometry such as long side, short side, tab position and pole nearby into different regions according to the physical structure of the battery, and each region corresponds to at least one temperature sensor; the heat dissipation path driven partitioning divides the regions with large liquid flow rate and flow dead angle into different regions according to the liquid contact surface and flow field distribution.

4. The method of claim 1, wherein, The calibration method of the static thermal resistance coefficient is specifically that in a constant temperature environment, a simulated heat source with the same size as the battery monomer is closely attached to the i-th region, and a constant power P is applied heat , and the steady-state temperature T of the region is monitored by a high-precision thermal imager steady , and the static thermal resistance coefficient k 1,i is calculated by the formula P heat / (T steady -T ambient ), wherein T ambient is the ambient temperature.

5. The method of claim 1, wherein, The calibration method of the thermal diffusion weight coefficient is as follows: a three-dimensional two-phase flow simulation is performed on the complete battery box by CFD software to simulate the flow and heat transfer of liquid and vapor, and a local heating source is set in the actual liquid cooling tank for experimental auxiliary measurement, and the thermal diffusion weight matrix between regions is calculated.

6. The method of claim 1, wherein, The long short-term memory network model is a double-layer stacked structure after offline training, wherein the first layer of LSTM unit is used to capture the short-time temperature fluctuation caused by bubble generation and separation, and the second layer of LSTM unit is used to extract the medium-term thermal inertia and phase change hysteresis effect, the model adopts weighted mean square error as the loss function for training, and higher weight is given to the short-time peak area.

7. The method of claim 1, wherein, The system is coupled with the battery monomer, module and battery pack immersed in the phase change cooling liquid, and is in communication connection with the thermal management system TMS and the energy management system EMS, comprising: A perception layer composed of multi-source high-precision sensors, the sensors include infrared matrix sensors, optical fiber temperature sensors and embedded thermocouples for monitoring the temperature of the battery surface and the inside, and sensors for collecting cooling liquid pressure, flow and liquid level, the perception layer is connected with the system to obtain current, SOC and SOH electrical parameters, and the output end of the perception layer is in communication connection with the input end of the data processing layer; The data processing layer input end is connected with the output end of the perception layer, receives and processes sensor data, and the data processing layer comprises: A preprocessing module for Kalman filtering, outlier rejection and time synchronization processing of original sensor data, a regional modeling module connected with the preprocessing module output end for managing partition parameter library and performing partition heat load calculation, a heat diffusion coupling module connected with the regional modeling module for implementing dynamic heat transfer coupling calculation between partitions, an adaptive fusion and smoothing module connected with the heat diffusion coupling module for smoothing heat load data, an output layer connected with the adaptive fusion and smoothing module output end and accessing a time series database, a prediction module for running an LSTM prediction model, and an output layer connected with the data processing layer output end, receiving real-time heat load values and prediction values, the output layer output end being connected with a TMS and an EMS through a control interface for outputting local heat load fields of each partition, module total heat load and prediction values.

8. The method of claim 7, wherein, The prediction module in the data processing layer is deployed in the memory of a master MCU of a battery management system or an edge computing device connected therewith, the LSTM model is pre-trained and solidified into a model file stored in the non-volatile memory of the device, and is loaded into a cache in real-time running to realize the prediction function.

9. The method of claim 7, wherein, The heat load prediction values generated by the output layer are connected with a controller of a thermal management system through a first data interface and connected with a decision unit of an energy management system through a second data interface.

10. The method of claim 7, wherein, The inter-regional heat diffusion weight coefficient is a dimensionless weight factor obtained through CFD simulation and experimental calibration, and is used to represent the inter-regional heat coupling strength caused by the flow of cooling liquid.

Images