A heat pipe-assisted immersion battery pack thermal management system

Abstract

This invention discloses a heat pipe-assisted immersion battery pack thermal management system, belonging to the field of battery control technology. It includes a sealed outer shell, inside which a battery pack is installed. A gap is provided between adjacent battery groups within the battery pack, and a heat dissipation pipe is inserted into the gap. The lower half of the heat dissipation pipe is located inside the sealed outer shell, while its upper half passes through the sealed outer shell and extends to the outside. A fan is installed on the outside of the sealed outer shell near the heat dissipation pipe. This invention significantly improves the heat dissipation efficiency of the battery pack by combining heat pipes, immersion cooling, and intelligent control strategies. It combines the advantages of dynamic and static immersion cooling, avoiding temperature gradient problems caused by uneven coolant flow, and enhances the heat dissipation effect through heat pipes and a wet-core structure, ensuring uniform temperature distribution within the battery pack and effectively extending battery life.

Description

Technical Field

[0001] This invention relates to the field of battery control technology, and more specifically, to a heat pipe-assisted immersion battery pack thermal management system. Background Technology

[0002] During charging and discharging, power batteries generate a significant amount of heat. If this heat cannot be dissipated effectively and promptly, the battery temperature will rise. When the battery temperature exceeds a certain range, it not only accelerates battery aging and shortens its lifespan but also triggers a series of safety issues, such as thermal runaway, fire, and even explosion. Furthermore, temperature differences between individual cells within the battery pack affect battery consistency and overall performance, thereby reducing the reliability and driving range of electric vehicles.

[0003] In existing technologies, immersion cooling is generally used to effectively dissipate heat from batteries. Immersion cooling technology is divided into two categories: dynamic and static, each with different characteristics and problems.

[0004] Dynamic immersion cooling systems achieve efficient heat dissipation through circulating coolant. The coolant circulates around or inside the battery pack, quickly removing heat generated by the battery and exhibiting high heat dissipation efficiency. However, due to the flow characteristics of the coolant, uneven flow can easily occur during actual operation, leading to significant temperature differences in different parts of the battery pack and creating temperature gradients. The existence of temperature gradients affects battery consistency and overall performance, reducing the reliability and stability of the thermal management system.

[0005] Static immersion cooling completely submerges the battery pack in coolant, relying on natural convection and heat conduction to maintain temperature uniformity. This method effectively avoids temperature gradients caused by uneven coolant flow, as seen in dynamic cooling systems, ensuring relatively consistent temperatures across the battery pack. However, the static cooling system has relatively low heat dissipation efficiency, especially during high-load operation. Natural convection of the coolant struggles to dissipate large amounts of heat quickly and effectively, leading to increased battery temperature and impacting battery performance and lifespan.

[0006] To address the aforementioned contradictions and problems in current power battery thermal management technologies, this invention proposes a heat pipe-assisted immersion battery pack thermal management system. Summary of the Invention

[0007] To address the problems existing in the prior art, the present invention aims to provide a heat pipe-assisted immersion battery pack thermal management system. This system can significantly improve the heat dissipation efficiency of the battery pack by combining heat pipes, immersion cooling, and intelligent control strategies. It also combines the advantages of dynamic and static immersion cooling, avoiding the temperature gradient problem caused by uneven coolant flow, and enhancing the heat dissipation effect through heat pipes and wet core structure. This ensures the uniformity of temperature distribution within the battery pack and effectively extends the battery life.

[0008] To solve the above problems, the present invention adopts the following technical solution.

[0009] A heat pipe-assisted immersion battery pack thermal management system includes a sealed housing, inside which a battery pack is installed. A gap is provided between adjacent battery packs, and a heat dissipation pipe is inserted into the gap. The lower half of the heat dissipation pipe is located inside the sealed housing, and its upper half passes through the sealed housing and extends to the outside. A fan is installed on the outside of the sealed housing near the heat dissipation pipe. A wet core is installed on the upper surface of the sealed housing at the end of the heat dissipation pipe away from the fan. The sealed housing contains coolant, and a water pump is also installed inside the sealed housing. The water pump is used to draw coolant from the sealed housing and inject it into the wet core. The coolant is discharged from the lower end of the wet core and flows back into the sealed housing.

[0010] It also includes,

[0011] The sensing and assessment module is used to collect battery temperature parameters, coolant state parameters, and environmental parameters in real time, and generate multi-dimensional feature vectors; it calculates the battery thermal risk index based on the collected data, which is generated by a weighted combination of the battery's maximum temperature, temperature rise rate, and state of charge.

[0012] The predictive modeling module is used to input multi-dimensional feature vectors into the time series prediction model; predict the spatial distribution of battery temperature field in future periods based on historical temperature data; and calibrate the prediction accuracy by embedding a loss function into the thermodynamic equation.

[0013] The decision optimization module executes a hierarchical control strategy based on the predicted temperature field and thermal risk index; it uses an evolutionary algorithm to solve for the energy-efficient optimal control command that satisfies the temperature constraints.

[0014] The execution monitoring module is used to convert control commands into actuator drive signals; monitor the execution status and actual battery temperature data in real time through the closed-loop controller; monitor execution deviation in real time, and activate the fault tolerance mechanism when abnormal flow is detected.

[0015] The learning evolution module updates system parameters based on feedback execution status and actual temperature data through a reinforcement learning mechanism, with the objective function balancing temperature stability and energy efficiency; and periodically calibrates the time series prediction model and control parameters.

[0016] Furthermore, the perception assessment module also includes:

[0017] Determine the data acquisition sensor configuration for the battery and cooling system, including setting corresponding sensors on the battery cells, coolant pipelines and the exterior of the sealed housing to monitor the battery cell temperature, coolant state parameters, environmental parameters and obtain the battery state of charge.

[0018] Key data features are extracted based on the sensor configuration, including the highest temperature of individual battery cells, the rate of temperature rise, the temperature difference of coolant, the flow fluctuation, the external temperature and humidity of the sealed casing, and the state of charge of the battery, and integrated to form a feature information group.

[0019] Based on the characteristic information group, different weights are assigned to the maximum temperature of the battery cell, the rate of temperature rise, and the state of charge of the battery, and a thermal risk index reflecting the thermal risk of the battery is calculated in a comprehensive manner.

[0020] Furthermore, the predictive modeling module also includes:

[0021] Retrieve historical temperature datasets associated with the current feature information group, including records of temperature changes in individual battery cells and temperature field distribution patterns, construct a continuous time series in chronological order, and label the time nodes;

[0022] A time series prediction model based on time dependence is constructed. This model integrates the temperature correlation characteristics of individual battery cells and takes the coolant state and environmental parameters as auxiliary variables as inputs. It learns the temperature field evolution pattern by learning the historical temperature change trend.

[0023] Input the time series into the prediction model, predict the temperature change trend of each battery cell and the spatial distribution of the battery pack temperature field in the future period based on the current feature information, and mark the time nodes of the prediction results.

[0024] Based on thermodynamic principles, a verification standard for the prediction results is established. The model parameters are adjusted by comparing the consistency between the prediction results and the heat conduction law until the prediction results meet the thermodynamic constraints.

[0025] Furthermore, the decision optimization module also includes:

[0026] Establish a thermal risk classification threshold system, including determining the thermal risk index threshold and temperature field prediction index threshold based on the battery safety temperature range and historical fault data;

[0027] The real-time thermal risk index and temperature field prediction results are compared with the threshold system. Based on the degree of risk index exceeding the standard and the range of temperature field violation areas, low, medium and high control level signals are matched and output.

[0028] In response to control level signals, the system invokes preset basic control rules to generate corresponding level heat dissipation actuator parameter adjustment schemes, including fan and water pump linkage control strategies.

[0029] Under the constraints of parameter adjustment scheme, the optimal control instruction set that meets the temperature safety constraints and has the minimum total power consumption is selected through simulation and energy consumption evaluation of multiple sets of control instructions.

[0030] Furthermore, the monitoring module includes:

[0031] Convert the control commands of the heat dissipation actuator into drive electrical signals that match its signal type, including converting fan speed commands into pulse modulation signals and water pump flow commands into voltage signals;

[0032] A drive signal is sent to the heat dissipation actuator to start operation. At the same time, the actual operating status of the actuator, battery temperature, and coolant parameters are collected in real time by sensors to form a closed-loop monitoring dataset synchronized with the target parameters and time stamps.

[0033] The deviation between the actual operating parameters of the actuator and the target parameters is calculated, and an operational anomaly is judged based on the preset deviation threshold, including determining an abnormal flow rate if the flow rate deviation exceeds the limit.

[0034] In response to abnormal flow, the fault tolerance mechanism is activated, increasing the water pump power to the preset safety limit and coordinating with the fan to enhance heat dissipation. If the abnormality persists, an alarm is triggered and the abnormal time and deviation data are recorded to ensure that the battery temperature meets the thermal management safety constraints.

[0035] Furthermore, the learning evolution module also includes;

[0036] The deviation parameters between the actual operating state of the actuator and the optimal control command are compared, and the spatial distribution difference between the actual battery temperature data and the predicted temperature field is analyzed; real-time energy consumption data and temperature fluctuation characteristics of the system are collected simultaneously.

[0037] Based on the deviation parameters and fluctuation characteristics, a reinforcement learning reward rule is set with temperature stability and energy efficiency as the core optimization objectives, and the reward value is dynamically allocated according to the degree of objective achievement.

[0038] Furthermore, the learning evolution module also includes:

[0039] The deviation data and reward rules are input into the reinforcement learning mechanism. Through iterative optimization, the hot risk judgment threshold, the feature weights of the time series prediction model, and the actuator response coefficients in the control strategy are adjusted so that the system converges to the reward maximization state.

[0040] The temperature field prediction model is periodically retrained and the control parameters are calibrated. The model structure and threshold range are dynamically adjusted based on the actual operating results, and the optimization results are deployed to the next round of control process.

[0041] Furthermore, the wet core is made of multiple sets of horizontally and vertically connected metal tubes that are interwoven, and the internal parts of the multiple sets of metal tubes are interconnected, with space left between the interwoven metal tubes for liquid flow.

[0042] Furthermore, a diverter box is fixedly connected to the upper end of the wet core, and a manifold box is fixedly connected to its lower end. Both the diverter box and the manifold box are hollow, and their interiors are interconnected through a metal tube on the wet core. A guide hole is provided on the upper surface of the sealing shell below the manifold box. The interior of the manifold box is interconnected with the interior of the sealing shell through the guide hole. A guide pipe is fixedly connected to the outlet of the water pump. The end of the guide pipe away from the water pump passes through the sealing shell and connects to the diverter box, and the interior of the guide pipe is interconnected with the interior of the diverter box.

[0043] Furthermore, the upper and lower ends of the heat dissipation pipe are sealed, and the inside of the heat dissipation pipe is filled with evaporating liquid. Metal fins are fixedly installed on the upper half of the heat dissipation pipe, and a protective cover is fixedly installed on the upper end of the sealed outer shell. The wet core, fan and the upper half of the heat dissipation pipe are all set inside the protective cover.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] (1) The heat pipe-assisted immersion battery pack thermal management system proposed in this solution significantly improves the heat dissipation efficiency of the battery pack by combining heat pipe, wet core, immersion cooling and intelligent control strategy. Combining the advantages of dynamic and static immersion cooling, it avoids the temperature gradient problem caused by uneven coolant flow and enhances the heat dissipation effect through heat pipe and wet core structure, ensuring the uniformity of temperature distribution in the battery pack and effectively extending the battery life.

[0046] (2) This solution forms a closed-loop intelligent control system through built-in perception assessment, prediction modeling, decision optimization and execution monitoring modules. By collecting battery temperature, coolant status and environmental parameters in real time, it can accurately calculate the battery thermal risk index and predict the battery temperature field distribution in the future. The decision optimization module executes a graded control strategy based on the prediction results to ensure that the most appropriate heat dissipation measures are taken under different thermal risk levels, which not only ensures battery safety but also improves energy efficiency.

[0047] (3) By introducing learning evolution, this scheme enables the system to have the ability to learn and continuously optimize itself. Through the reinforcement learning mechanism, the system can dynamically adjust the control strategy and model parameters according to the actual operation effect, and continuously optimize the heat dissipation performance and energy consumption efficiency. This adaptive capability enables the system to cope with various complex working conditions, ensuring that the battery pack always maintains the best thermal management state during long-term operation, which greatly improves the reliability and stability of the system. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0049] Figure 1 This is an external view of the overall structure of the present invention;

[0050] Figure 2 This is a structural diagram of the upper surface component of the sealed outer shell of the present invention;

[0051] Figure 3 This is a schematic diagram of the heat dissipation pipe of the present invention;

[0052] Figure 4 This is a schematic diagram of the internal structure of the sealed outer shell of the present invention;

[0053] Figure 5 This is a structural diagram of the sealed outer casing of the present invention;

[0054] Figure 6 This is a schematic diagram of the battery pack structure of the present invention;

[0055] Figure 7 This is a schematic diagram of a heat pipe-assisted immersion battery pack thermal management system according to the present invention.

[0056] Explanation of the labels in the diagram:

[0057] 1. Sealed outer casing; 2. Positioning groove; 3. Battery pack; 4. Heat sink; 5. Metal fins; 6. Fan; 7. Water pump; 8. Guide pipe; 9. Diverter box; 10. Wet core; 11. Combinator box; 12. Injection port; 13. Protective cover; 14. Guide hole; 15. Pressure plate. Detailed Implementation

[0058] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0059] Please see Figures 1 to 7A heat pipe-assisted immersion battery pack thermal management system includes a sealed housing 1, inside which a battery pack 3 is installed. A gap is provided between adjacent battery packs in the battery pack 3, and a heat dissipation pipe 4 is inserted into the gap. The lower half of the heat dissipation pipe 4 is located inside the sealed housing 1, and its upper half passes through the sealed housing 1 and extends to the outside. A fan 6 is installed on the outside of the sealed housing 1 near the heat dissipation pipe 4. A wet core 10 is installed on the upper surface of the sealed housing 1 at the end of the heat dissipation pipe 4 away from the fan 6. The wet core 10 is made of multiple sets of horizontally and vertically connected metal tubes that are interconnected internally. Space is left between the intersecting metal tubes for liquid flow. The sealed housing 1 contains coolant, and a water pump 7 is also installed inside the sealed housing 1. The water pump 7 is used to draw coolant from the sealed housing 1 and inject it into the wet core 10. The coolant is discharged from the lower end of the wet core 10 and flows back into the sealed housing 1.

[0060] The upper end of the wet core 10 is fixedly connected to a diverter box 9, and the lower end is fixedly connected to a manifold box 11. Both the diverter box 9 and the manifold box 11 are hollow and their interiors are interconnected through a metal tube on the wet core 10. The upper surface of the sealing shell 1 is provided with a guide hole 14 below the manifold box 11. The interior of the manifold box 11 is interconnected with the interior of the sealing shell 1 through the guide hole 14. A guide pipe 8 is fixedly connected to the outlet of the water pump 7. The end of the guide pipe 8 away from the water pump 7 passes through the sealing shell 1 and is connected to the diverter box 9. The interior of the guide pipe 8 is interconnected with the interior of the diverter box 9. The upper and lower ends of the heat dissipation pipe 4 are sealed, and the interior of the heat dissipation pipe 4 is filled with evaporating liquid. Metal fins 5 are fixedly installed on the upper half of the heat dissipation pipe 4. A protective cover 13 is fixedly installed on the upper end of the sealing shell 1. The wet core 10, the fan 6, and the upper half of the heat dissipation pipe 4 are all located inside the protective cover 13.

[0061] By adopting the above technical solution, the battery pack 3 generates heat when outputting current and charging. As the heat accumulates, the temperature of the battery pack 3 becomes too high, affecting its operation. The heat from the battery pack 3 is transferred to the coolant inside the sealed casing 1. After the coolant heats up, it heats the heat dissipation pipe 4. The heat dissipation pipe 4 can be made of copper or aluminum, which allows for good heat conduction. After the lower half of the heat dissipation pipe 4 is heated, the evaporator inside it vaporizes. The rising vapor transfers heat to the upper half of the heat dissipation pipe 4, which is then cooled. The evaporator inside cools down and returns to the lower half. The upper half of the heat dissipation pipe 4 is equipped with metal fins 5, which increase the contact area with the air, thereby improving the heat dissipation of the heat dissipation pipe 4. The accelerated heat dissipation of the heat dissipation pipe 4 also accelerates the heat dissipation of the coolant, thus accelerating the heat dissipation of the battery pack 3.

[0062] A temperature sensor, such as a thermistor, is installed inside the sealed housing 1 to detect the temperature of the coolant. When the coolant temperature is higher than a first preset temperature, the fan 6 is activated. The fan 6 blows air, accelerating the airflow around the metal fins 5, thereby accelerating the cooling of the metal fins 5. By accelerating the cooling of the metal fins 5, the cooling of the coolant is accelerated, thus ensuring that the battery pack 3 is maintained within a stable temperature range. If the coolant temperature is higher than a second preset temperature, the water pump 7 will be activated. The water pump 7 will draw coolant from the sealed housing 1, and then the coolant will flow through the guide pipe 8, finally flowing into the distribution box 9. The coolant will then flow into the metal tube of the wet core 10. Because multiple sets of metal tubes are staggered and interconnected, the coolant will flow into each metal tube as it flows downwards, finally exiting from the lower end of the wet core 10 and flowing into the manifold 11. The coolant in the manifold 11 flows into the sealed housing 1 through the guide hole 14, thus realizing the circulation of the coolant. When the coolant circulates, it works in conjunction with the air cooling of the fan 6. Air passes through the gaps in the wet core 10. This airflow, combined with the rapidly flowing coolant inside the metal tube, can quickly cool the coolant, thus effectively cooling the battery pack 3 and achieving thermal management of the battery pack 3.

[0063] The side wall of the sealed housing 1 has a liquid injection port 12, through which coolant can be injected into the sealed housing 1. The liquid injection port 12 can be sealed by a sealing cap to prevent coolant from spilling out. The inner bottom wall of the sealed housing 1 has a positioning groove 2. The battery pack 3 is composed of multiple individual battery cells, and each battery cell is installed in its corresponding positioning groove 2. A pressure plate 15 is also provided at the upper end of the battery cell, so that the upper and lower ends of the battery cell are subjected to stable force while preventing coolant from splashing, thereby ensuring that the battery pack 3 is stably installed in the sealed housing 1.

[0064] In some embodiments, the system further includes a sensing and evaluation module for real-time acquisition of battery temperature parameters, coolant state parameters, and environmental parameters, and for generating a multi-dimensional feature vector; and for calculating a battery thermal risk index based on the acquired data, which is generated by a weighted combination of the battery's maximum temperature, temperature rise rate, and state of charge.

[0065] The predictive modeling module is used to input multi-dimensional feature vectors into the time series prediction model; predict the spatial distribution of battery temperature field in future periods based on historical temperature data; and calibrate the prediction accuracy by embedding a loss function into the thermodynamic equation.

[0066] The decision optimization module executes a hierarchical control strategy based on the predicted temperature field and thermal risk index; it uses an evolutionary algorithm to solve for the energy-efficient optimal control command that satisfies the temperature constraints.

[0067] The execution monitoring module is used to convert control commands into actuator drive signals; monitor the execution status and actual battery temperature data in real time through the closed-loop controller; monitor execution deviation in real time, and activate the fault tolerance mechanism when abnormal flow is detected.

[0068] The learning evolution module updates system parameters based on feedback execution status and actual temperature data through a reinforcement learning mechanism, with the objective function balancing temperature stability and energy efficiency; and periodically calibrates the time series prediction model and control parameters.

[0069] The perception assessment module also includes:

[0070] Determine the data acquisition sensor configuration for the battery and cooling system, including setting corresponding sensing devices on the battery cells, coolant pipelines and the exterior of the sealed housing 1 to monitor the battery cell temperature, coolant state parameters, environmental parameters and obtain the battery state of charge.

[0071] Key data features are extracted based on the sensor configuration, including the highest temperature of the battery cell, the rate of temperature rise, the temperature difference of the coolant, the flow fluctuation, the external temperature and humidity of the sealed casing 1, and the state of charge of the battery, and integrated to form a feature information group.

[0072] Based on the characteristic information group, different weights are assigned to the maximum temperature of the battery cell, the rate of temperature rise, and the state of charge of the battery, and a thermal risk index reflecting the thermal risk of the battery is calculated in a comprehensive manner.

[0073] By adopting the above technical solution, at the battery cell level, thermocouples or NTC thermistors are used as temperature sensing devices. At least one sensing unit is tightly attached to the surface of each battery cell. Thermal grease is used to fill the gap between the sensing unit and the cell casing to reduce thermal resistance, ensuring that the temperature monitoring error is controlled within ±0.5℃. The signal output of the sensing unit is aggregated to the data acquisition module through a shielded cable, realizing independent real-time monitoring of the temperature of each cell.

[0074] The sensing configuration of the coolant pipeline adopts a distributed layout: temperature sensors are installed at the inlet section where the coolant flows into the sealed housing 1 and at the outlet section where it flows out of the sealed housing 1, with a distance between them not less than 5 times the pipeline diameter to avoid local flow field interference, and are used to capture the heat exchange temperature difference between the coolant and the battery pack 3; an electromagnetic flowmeter is installed in the straight section of the middle section of the guide pipe 8, with a measurement accuracy of not less than ±2%, and the flow field distribution is optimized by the rectification structure in the pipeline to reduce the impact of flow fluctuations on measurement stability; temperature sensing points are set at the upper and lower ends of the heat dissipation pipe 4, with a distance between the two points not less than 1 / 3 of the effective heat exchange section length of the heat dissipation pipe, and are used to monitor the phase change heat transfer effect of the evaporator.

[0075] The sensing configuration outside the sealed housing 1 includes an integrated temperature and humidity sensor mounted on the windward side of the top of the housing. Its measurement range covers -20℃ to 60℃ and 0 to 100%RH (humidity), and its sampling frequency is synchronized with the internal sensing devices. The battery state of charge (SOC) is calculated via the communication interface of the battery management system (BMS), and data exchange is achieved using CAN bus or RS485 protocol, ensuring that the SOC data update delay does not exceed 1 second. All sensor power supply modules employ an isolated design to avoid data drift caused by electromagnetic interference.

[0076] Based on the above sensing configuration, the raw monitoring data is analyzed in real time, and the temperature data of all battery cells are compared hourly to filter out the instantaneous maximum value and mark the corresponding cell number and occurrence time to form the highest temperature sequence.

[0077] Using a 10-second time window, the ratio of the change in the highest temperature within each window to the time is calculated. Instantaneous fluctuations are filtered out using a moving average algorithm to obtain a smooth temperature rise rate curve.

[0078] The temperature sensor reading difference between the inlet section of the coolant flowing into the sealed housing 1 and the outlet section of the coolant flowing out of the sealed housing 1 is calculated in real time. When the difference is less than a preset threshold, an abnormal flag is triggered, indicating a decrease in heat exchange efficiency.

[0079] Spectral analysis is performed on the instantaneous flow data output by the flow meter to extract the standard deviation of the flow rate within 1 minute and quantify the flow stability in the pipeline.

[0080] Based on the temperature and humidity integrated sensor installed on the outside of the sealed housing 1, the real-time data of the external sensor is directly read, and the hourly average value is used as the environmental reference parameter.

[0081] After obtaining the raw SOC data from the BMS, it is corrected in conjunction with the current charge / discharge rate to eliminate SOC jumps caused by instantaneous current surges.

[0082] After being timestamped, the aforementioned feature parameters are packaged into feature information groups at fixed intervals (e.g., every 2 seconds). Each group of data contains 7 key features and corresponding collection time stamps, providing input for subsequent thermal risk assessment.

[0083] The thermal risk index based on feature information groups is calculated using a weighted fusion mechanism. For example, the highest temperature of a single battery cell is used as the core parameter and given the highest weight because it directly reflects the critical state of thermal runaway. When the battery approaches the upper limit of its safe temperature, the weight of this parameter is dynamically increased.

[0084] The rate of temperature rise is assigned the second highest weight to assess the rate of heat accumulation. When the rate exceeds a preset safety threshold, its contribution to thermal risk is amplified through weight adjustment.

[0085] The battery's state of charge (SOC) is assigned a basic weight, and its correlation with thermal stability (e.g., increased thermal sensitivity at high SOC) is considered. The weight is appropriately increased when the SOC is above 80% or below 20%. Through this weighting, the normalized eigenvalues ​​of the three factors are combined to generate a thermal risk index ranging from 0 to 100. An index below 30 is considered low risk, 30-70 medium risk, and above 70 high risk. The index value is dynamically updated in real-time with changes in the characteristic parameters, providing a quantitative basis for the cooling system's control strategy.

[0086] In some embodiments, the predictive modeling module further includes:

[0087] Retrieve historical temperature datasets associated with the current feature information group, including records of temperature changes in individual battery cells and temperature field distribution patterns, construct a continuous time series in chronological order, and label the time nodes;

[0088] A time series prediction model based on time dependence is constructed. This model integrates the temperature correlation characteristics of individual battery cells and takes the coolant state and environmental parameters as auxiliary variables as inputs. It learns the temperature field evolution pattern by learning the historical temperature change trend.

[0089] Input the time series into the time series prediction model, predict the temperature change trend of each battery cell and the spatial distribution of the temperature field of battery pack 3 in the future period based on the current feature information, and mark the time nodes of the prediction results.

[0090] Based on thermodynamic principles, a verification standard for the prediction results is established. The model parameters are adjusted by comparing the consistency between the prediction results and the heat conduction law until the prediction results meet the thermodynamic constraints.

[0091] By adopting the above technical solution, based on the current feature information group (including parameters such as the highest temperature of individual battery cells, coolant temperature difference, and ambient temperature and humidity), related datasets can be matched in the historical database through data retrieval. The retrieval logic adopts multi-dimensional feature matching: using the current ambient temperature and humidity and the battery state of charge (SOC) as the primary retrieval conditions, historical operating periods with a similarity higher than 85% are selected; then, secondary filtering is performed through the coolant flow fluctuation range and the initial temperature distribution pattern, finally determining historical samples that are highly consistent with the current operating conditions.

[0092] The selected historical data is linearly interpolated along the time axis to fill gaps caused by data acquisition intervals, ensuring that the time resolution is consistent with the current monitoring frequency (e.g., 2 seconds / time). A sliding time window (1 hour long) is used to segment the historical temperature data, extracting records of individual battery cell temperature changes (including the temperature time-series curve for each cell) and temperature field distribution patterns (e.g., high-temperature region migration paths, temperature gradient change characteristics) for each time period. This data is then concatenated with the current feature information group in chronological order to form a continuous time series. Each data point is labeled with a time node accurate to milliseconds to ensure temporal continuity and timestamp alignment.

[0093] The time series prediction model adopts a spatiotemporal fusion architecture. The core layer is a sequence modeling module that handles time dependencies. It captures the short-term temporal characteristics of temperature data through a sliding time window (the window size is set to 30 sampling points), while embedding a long-term memory mechanism to learn the temperature evolution pattern on a multi-hour scale.

[0094] The model input layer adopts a multi-channel design. The main channel inputs the time series of battery cell temperature and encodes the spatial correlation between battery cells through the adjacency matrix (such as the thermal conductivity coefficient of adjacent cells and the positional distance weight) to realize the quantitative modeling of the temperature correlation characteristics of cells. The auxiliary channel simultaneously inputs coolant state parameters (flow fluctuation, inlet and outlet temperature difference) and environmental parameters (temperature, humidity, atmospheric pressure). The feature mapping layer transforms non-temperature parameters into temperature field-related influencing factors (such as the correction coefficient of coolant flow rate change on temperature gradient).

[0095] During the model training phase, historical time series data are used for iterative optimization. The temperature field evolution data of the past 24 hours are used as training samples. The internal weight parameters are adjusted through the backpropagation mechanism, so that the model can gradually learn the temperature field evolution pattern under different operating conditions (such as the temperature accumulation effect under high SOC state and the temperature response lag characteristics after sudden changes in coolant flow). Finally, a time series prediction model that can capture both temporal dynamics and spatial correlation is formed.

[0096] The completed continuous time series is input into the time series prediction model to initiate a multi-step prediction process. The model first encodes the current feature information group and extracts key parameters (such as the current location of the high-temperature unit and the heat exchange efficiency of the coolant) as the initial conditions for prediction. Then, based on the temperature evolution patterns learned from history, the model generates point-by-point prediction results for the next 10-30 minutes using a time series extrapolation algorithm.

[0097] The predicted output comprises two dimensions: first, the temperature variation trend of each individual battery cell, which is formed by creating individual cell temperature curves from the predicted temperature values ​​at continuous time points, marking the possible temperature peaks and their arrival times; second, the spatial distribution of the overall temperature field of the battery pack, which uses a grid-based method (grid precision matched to the cell arrangement density) to characterize the spatial temperature gradient, and color mapping to encode the distribution range of high-temperature regions (>40℃), medium-temperature regions (25-40℃), and low-temperature regions (<25℃), and annotates the predicted expansion rate of each region. All prediction results are labeled with corresponding time points in chronological order, consistent with the timestamp format of historical data.

[0098] Based on the fundamental principles of thermodynamics, a standard for verifying prediction results was established. The core verification items include: heat conduction directionality (the trend of heat transfer from high-temperature cells to low-temperature cells), heat conservation (the balance between the heat carried away by the coolant and the heat generated by the battery), and phase change heat transfer law (the matching between the temperature difference between the upper and lower ends of heat pipe 4 and the phase change rate of the evaporator).

[0099] The predicted results are compared item by item with the above standards: if a temperature inversion phenomenon occurs, where heat is transferred from a low-temperature unit to a high-temperature unit, or if the coolant flow rate increases but the predicted temperature does not show a decreasing trend, it is determined to violate thermodynamic laws. At this time, the model parameter calibration mechanism is activated: the heat transfer coefficient in the spatial correlation matrix is ​​corrected, and the weight consistent with the actual heat transfer direction is increased; the influence factors of auxiliary variables (such as coolant flow rate) are adjusted to strengthen their constraint on the temperature field prediction. Through multiple rounds of iterative verification, the conformity of the prediction results with all verification items reaches more than 95%, ensuring the physical rationality of the prediction results.

[0100] In some embodiments, the decision optimization module further includes:

[0101] Establish a thermal risk classification threshold system, including determining the thermal risk index threshold and temperature field prediction index threshold based on the battery safety temperature range and historical fault data;

[0102] The real-time thermal risk index and temperature field prediction results are compared with the threshold system. Based on the degree of risk index exceeding the standard and the range of temperature field violation areas, low, medium and high control level signals are matched and output.

[0103] In response to control level signals, the preset basic control rules are invoked to generate corresponding level heat dissipation actuator parameter adjustment schemes, including the linkage control strategy of fan 6 and water pump 7;

[0104] Under the constraints of parameter adjustment scheme, the optimal control instruction set that meets the temperature safety constraints and has the minimum total power consumption is selected through simulation and energy consumption evaluation of multiple sets of control instructions.

[0105] By adopting the above technical solution, the construction of the thermal risk classification threshold system is based on the fusion analysis of battery safety operation boundaries and historical fault data. First, based on the material characteristics and safety standards of battery cells, the core temperature parameter thresholds are determined: 80% of the battery's highest tolerable temperature (e.g., 50℃) (40℃) is taken as the critical value for temperature field prediction; when the area of ​​the high-temperature region (>40℃) exceeds 10%, it is judged as a temperature field violation; when the temperature rise rate exceeds 2℃ / min, it is marked as a rapid temperature rise risk.

[0106] Meanwhile, statistical modeling was performed on historical fault data to extract the thermal risk index characteristics before the fault occurred: by analyzing 50 thermal risk warning events in the past 3 months, it was determined that the fault probability was less than 0.1% when the thermal risk index was 30, and the fault probability rose sharply to 15% when it was 70. Based on this, 30 was set as the first threshold (low-medium risk critical point) and 70 was set as the second threshold (medium-high risk critical point).

[0107] The final threshold system includes two judgment dimensions: thermal risk index threshold (30, 70) and temperature field index threshold (10% of high temperature area and 2℃ / min temperature rise rate). The two are jointly judged through logic and calculation to ensure that the threshold setting covers both sudden risks (such as a sudden increase in the index) and cumulative risks (such as the expansion of high temperature area).

[0108] The control level matching adopts a dynamic comparison mechanism between real-time data and the threshold system, which collects the thermal risk index and predicted temperature field results in real time, and then makes a judgment:

[0109] Low risk level determination: When the thermal risk index is consistently below 30, and the proportion of high-temperature areas in the temperature field prediction results is <5% and the temperature rise rate is <0.5℃ / min, it is determined to be low risk, and a low-level control signal is output.

[0110] Medium risk level determination: When the heat risk index fluctuates between 30 and 70, or the high temperature area accounts for 5%-10% and the temperature rise rate is 0.5-2℃ / min (any one of the conditions is met), it is determined to be medium risk, and a medium level control signal is output.

[0111] High-risk level determination: When the heat risk index exceeds 70, or the proportion of high-temperature areas is >10% and the temperature rise rate is >2℃ / min (any one of the conditions is met), it is determined to be high-risk, and a high-level control signal is output.

[0112] The judgment process incorporates a lag compensation mechanism, ensuring that there is no delay when switching from medium risk to high risk, and that switching from high risk to medium risk requires the indicator to fall back to within the threshold for 30 seconds, thus avoiding actuator wear caused by frequent switching.

[0113] The working principle of generating parameter adjustment schemes for heat dissipation actuators is as follows: the parameter adjustment scheme is based on the control-level signals calling a preset rule base, which is pre-generated and fixed using orthogonal experimental design.

[0114] Low-risk solution: Maintain current operating parameters, with fan 6 operating at 30% of its rated speed, water pump 7 flow rate controlled at 2L / min, and actuator status locked via pulse width modulation (PWM) signal;

[0115] Medium-risk solution: Activate the synergistic enhancement strategy, increase the speed of fan 6 to 60% of the rated value, and increase the flow rate of water pump 7 to 3L / min in a 1:1.5 ratio, using a combination of air cooling and liquid cooling for heat dissipation;

[0116] High-risk solution: Implement full-load cooling, with fan 6 running at 100% rated speed and water pump 7 increasing flow rate to the maximum value of 5L / min.

[0117] When generating the solution, the following constraints are embedded: the flow rate of water pump 7 does not exceed the upper limit of pipeline pressure (8L / min), and the speed of fan 6 is not lower than the minimum stable speed (10% of the rated value) to ensure that the actuator operates within a safe range.

[0118] The selection of the optimal instruction involves multi-objective optimization within the range of parameter tuning options:

[0119] The instruction generation is based on the parameter range of the scheme (such as the fan 6 speed of 50%-70% and the water pump 7 flow rate of 2.5-3.5L / min in medium risk). 20 sets of candidate instructions are generated by uniform sampling. Each set includes specific parameters such as the PWM duty cycle of fan 6 and the voltage of water pump 7.

[0120] Simulation verification: Input the candidate command into the thermodynamic balance model to simulate the temperature field change in the next 10 minutes and verify whether the safety constraint of "maximum temperature ≤ 40℃, high temperature area cleared" is met.

[0121] The thermal balance model can be constructed using a neural network. The model is first input with battery status (current temperature, charging / discharging power), heat dissipation parameters (fan speed, water pump flow rate), and ambient temperature. The output is the temperature index for the next 10 minutes (average temperature every 2 minutes, and final maximum cell temperature). Historical operating data covering multiple operating conditions is selected, normalized, and divided into training and validation sets. These sets are then sliced ​​into input-output pairs by time. A shallow feedforward structure is used, containing an input layer (5 neurons), one hidden layer (32 neurons), and an output layer (6 neurons). The parameter scale is controlled to ensure rapid computation. The model is trained using an appropriate loss function (emphasizing penalties for maximum temperature deviation) and an optimizer, combined with an early stopping mechanism to avoid overfitting. It is then tested and calibrated using new operating condition data to ensure that the maximum temperature prediction error for over 95% of the samples is ≤2℃. The model is then solidified into a lightweight file and integrated into the simulation process. After inputting candidate control commands, the model quickly outputs temperature prediction results, directly verifying whether safety constraints are met.

[0122] Energy consumption assessment: Through power monitoring, calculate the real-time power consumption of each set of commands (fan 6 power consumption is proportional to the square of the speed, water pump 7 power consumption is linearly related to the flow rate), and accumulate the total energy consumption for 10 minutes;

[0123] Optimal selection: Using the Pareto optimality principle, the set with the lowest total energy consumption is selected from the candidate instructions that meet the safety constraints, and this set is taken as the optimal control instruction set for the final output. The instruction format conforms to the frame structure requirements of the actuator communication protocol (such as Modbus-RTU). The selection process iterates every 2 minutes to ensure that the control instructions can dynamically adapt to changes in battery state.

[0124] The specific principle for selecting the set with the lowest total energy consumption using the Pareto optimality rule is as follows: First, candidate instructions that meet the safety constraints (temperature meets the standard) are screened to form a set of feasible solutions; for each feasible solution, it is evaluated from two dimensions: "safety redundancy" (the difference between temperature and safety threshold) and "total energy consumption".

[0125] Pairwise, feasible solutions are compared, and "dominated solutions" (i.e., solutions with lower energy consumption and high safety redundancy) are eliminated. The remaining solutions form the "Pareto optimal solution set". The instruction with the lowest total energy consumption is selected from the optimal solution set. If the energy consumption is the same, the instruction with the highest safety redundancy is selected as the final optimal instruction.

[0126] In some embodiments, the monitoring module includes:

[0127] Convert the control commands of the heat dissipation actuator into drive electrical signals that match its signal type, including converting the fan speed command of 6 into a pulse modulation signal and the water pump flow command of 7 into a voltage signal;

[0128] A drive signal is sent to the heat dissipation actuator to start operation. At the same time, the actual operating status of the actuator, battery temperature, and coolant parameters are collected in real time by sensors to form a closed-loop monitoring dataset synchronized with the target parameters and time stamps.

[0129] The deviation between the actual operating parameters of the actuator and the target parameters is calculated, and an operational anomaly is judged based on the preset deviation threshold, including determining an abnormal flow rate if the flow rate deviation exceeds the limit.

[0130] In response to abnormal flow, the fault tolerance mechanism is activated, increasing the power of water pump 7 to the preset safety limit and enhancing the heat dissipation capacity of fan 6. If the abnormality persists, an alarm is triggered and the abnormal time and deviation data are recorded to ensure that the battery temperature meets the thermal management safety constraints.

[0131] By adopting the above technical solution, the fan 6 speed command conversion uses pulse width modulation (PWM) technology. The control module maps the target speed percentage (e.g., 30%, 60%) to the corresponding duty cycle signal. For example, the rated speed corresponds to a 50Hz PWM signal, and 30% speed corresponds to a 15% duty cycle. A 0-5V square wave signal is output through a dedicated PWM generation chip, and the steepness of the signal rise and fall edges is controlled within 100ns to ensure that the response delay of the fan 6 motor is ≤50ms.

[0132] The flow command conversion of water pump 7 is achieved through a digital-to-analog converter (DAC) module, which linearly converts the flow percentage (e.g., 2L / min corresponds to 40% of the rated flow) into a 0-12V DC voltage signal. The mapping relationship between voltage and flow is calibrated using the factory calibration curve of water pump 7. For example, 12V corresponds to a maximum flow of 5L / min, and 6V corresponds to 2.5L / min. The conversion accuracy is controlled within ±0.1V to ensure that the flow deviation is ≤5%.

[0133] During the conversion process, a signal isolation circuit is embedded, and an optocoupler is used to achieve electrical isolation between the control module and the actuator, suppressing signal distortion caused by common-mode interference and ensuring the stability of the drive signal. The actuator of this invention is a hardware device that directly implements heat dissipation, including controlling the fan 6 and the water pump 7. Its specific functions are as follows: the control module sends a drive signal to regulate the operation of the actuator, and the actual state of the actuator is fed back to the control module through a sensor, realizing dynamic coordination of "command-execution-monitoring-adjustment".

[0134] After the drive signal is sent, the multi-dimensional real-time monitoring link is activated:

[0135] Regarding actuator status acquisition, a Hall effect speed sensor is installed at fan 6, which outputs 2 pulse signals per revolution. The actual speed is calculated by a counter module (sampling frequency 1kHz). A flow meter is installed at the outlet of water pump 7, which outputs a 4-20mA current signal. This signal is converted into a 0-5V signal by a current-to-voltage conversion circuit to reflect the actual flow rate in real time (sampling frequency 500Hz).

[0136] Regarding the acquisition of battery and coolant parameters, the signals from the battery cell temperature sensor (sampling frequency 10Hz), the coolant inlet and outlet temperature sensor (sampling frequency 5Hz), and the pipeline pressure sensor (sampling frequency 1Hz) are first aggregated by a multiplexer, and then synchronously converted into digital signals by a 16-bit ADC module.

[0137] All acquired signals are timestamped by a local high-precision clock module (accuracy ±10ppm) and packaged into datasets at 20ms intervals. Each data packet contains the actual speed of fan 6, the actual flow rate of water pump 7, the temperature of each battery cell, the temperature difference of coolant, and the pressure value, ensuring that the time deviation between the target parameter (command value) and the actual parameter is ≤1ms, forming a complete closed-loop feedback link.

[0138] The deviation calculation adopts the instantaneous difference analysis method between the real-time value and the target value. For example, the deviation rate between the actual speed and the commanded speed of fan 6 is (actual value - target value) / target value × 100%. The flow deviation rate of water pump 7 adopts the same calculation logic.

[0139] The preset deviation threshold is dynamically adjusted according to the actuator characteristics: the normal deviation threshold for fan 6 is ±10% (relaxed to ±15% in the low-speed range), and the normal flow deviation threshold for water pump 7 is ±8% (relaxed to ±12% when the flow rate is ≤2L / min). If the deviation rate exceeds the threshold for three consecutive sampling cycles (60ms), and sensor malfunction is ruled out, it is determined to be an abnormal operation.

[0140] When the flow deviation rate of water pump 7 is greater than 20%, or the absolute value of flow is less than 0.5 L / min (lasting for 1 second), and the pipeline pressure is greater than or equal to 1.2 times the rated value, it is determined to be a pipeline blockage; if the flow deviation rate is greater than 20% and the pressure is less than or equal to 0.8 times the rated value, it is determined to be a power failure of water pump 7. Both situations are classified as abnormal flow.

[0141] After an abnormal traffic event is triggered, the tiered fault tolerance process is initiated:

[0142] Primary compensation: Immediately increase the water pump drive voltage to the safe upper limit (12V) and continue for 2 seconds to try to break through the blockage or overcome insufficient power. At the same time, increase the PWM duty cycle of fan 6 by 20% (not exceeding the rated value) to compensate for the decrease in liquid cooling efficiency by enhancing air cooling.

[0143] Status monitoring: If the flow rate recovers to within the deviation threshold (≤±12%) within 2 seconds, it will be gradually restored to the original command value at a rate of 5% / s to avoid system fluctuations caused by sudden parameter changes.

[0144] Level 2 response: If the abnormality continues for more than 5 seconds, the current limiting protection will be activated, and at the same time, fan 6 will run at 100% rated speed to forcibly maintain the heat dissipation capacity.

[0145] Alarm and Recording: After an anomaly persists for 10 seconds, a fault code (e.g., 0x00A3 representing abnormal flow) is sent via the CAN bus to activate the alarm device. Data such as the anomaly occurrence time, maximum deviation, and highest battery temperature are written to non-volatile memory to ensure fault traceability. Throughout the fault-tolerant process, real-time temperature monitoring ensures the battery temperature does not exceed 45°C; if it approaches the threshold, an emergency shutdown protection is triggered. This alarm device can be a buzzer installed in the sealed housing 1. Activating the alarm device charges the buzzer, causing it to sound and trigger the alarm.

[0146] In some embodiments, the learning evolution module further includes;

[0147] The deviation parameters between the actual operating state of the actuator and the optimal control command are compared, and the spatial distribution difference between the actual battery temperature data and the predicted temperature field is analyzed; real-time energy consumption data and temperature fluctuation characteristics of the system are collected simultaneously.

[0148] Based on the deviation parameters and fluctuation characteristics, a reinforcement learning reward rule is set with temperature stability and energy efficiency as the core optimization objectives, and the reward value is dynamically allocated according to the degree of objective achievement.

[0149] The deviation data and reward rules are input into the reinforcement learning mechanism. Through iterative optimization, the hot risk judgment threshold, the feature weights of the time series prediction model, and the actuator response coefficients in the control strategy are adjusted so that the system converges to the reward maximization state.

[0150] The time series prediction model is periodically retrained and the control parameters are calibrated. The model structure and threshold range are dynamically adjusted according to the actual operating results, and the optimization results are deployed to the next round of control process.

[0151] By adopting the above technical solution, the deviation rate between the actual speed of fan 6 and the optimal command (calculated every 100ms) is first averaged to generate a speed deviation curve; the instantaneous difference between the actual flow rate and the commanded flow rate of water pump 7 is integrated to obtain the cumulative flow deviation. Both types of deviation parameters are stored according to actuator type, and the duration of the deviation and the time of peak occurrence are marked.

[0152] Battery pack 3 was divided into 10×10 spatial units using a grid-based method. The actual temperature (collected by distributed sensors) of each unit was compared with the corresponding predicted temperature field, and the absolute value of the temperature difference and the mean square error of the unit were calculated. The overlap between the actual high-temperature region (>40℃) and the predicted high-temperature region was marked. If the overlap rate was <60%, it was judged as a significant deviation in spatial distribution. The cell number and duration of the deviation region were recorded simultaneously.

[0153] System energy consumption data is sampled at 500Hz using a power sensor, and the average power consumption per minute is calculated and compared with historical values ​​for the same period. Temperature fluctuation characteristics are quantified by calculating the standard deviation of the difference between the highest and lowest temperatures of battery pack 3 (sliding window of 1 minute) to capture instantaneous rises and falls. All data are aligned with timestamps to form a four-dimensional dataset containing deviation parameters, temperature differences, energy consumption, and fluctuation characteristics.

[0154] The working principle of setting reinforcement learning reward rules is that the reward rules are dynamically generated based on a dual-objective optimization system:

[0155] Define a temperature stability reward, with the core indicators being "temperature fluctuation range ≤ 3℃" and "actual temperature deviation from target temperature (35±2℃) ≤ 2℃". When both conditions are met, the basic reward is linearly allocated according to the fluctuation range (the smaller the fluctuation, the higher the reward); if the temperature of a single unit exceeds 40℃, a penalty mechanism is triggered (30% of the basic reward is deducted); if there is no temperature exceedance for 5 consecutive minutes, an additional stability reward is given.

[0156] Define an energy efficiency reward, with the average energy consumption under the best historical operating conditions as the baseline. For every 10% lower actual energy consumption than the baseline, the reward value increases by 15%. If the energy consumption is higher than the baseline but the temperature stability is significantly better than the historical level (fluctuation amplitude reduced by 50%), the reward compensation mechanism can be activated (maximum compensation of 20%).

[0157] Under high-risk operating conditions (thermal risk index > 50), the temperature stability reward weight automatically increases to 70%, while the energy consumption weight decreases to 30%; the opposite applies to low-risk operating conditions. Reward values ​​are calculated on a minute-by-minute basis, and a reward decay coefficient (0.95 / minute) is used to suppress short-term speculative behavior and encourage long-term optimization.

[0158] The specific principle of reinforcement learning parameter optimization is to achieve system optimization through iterative adjustments:

[0159] Define a state space with a four-dimensional dataset (deviation parameters, temperature differences, energy consumption, and fluctuation characteristics) as input states, covering actuator response deviations (such as fan speed lag), time series prediction model errors (such as deviation rate in high-temperature areas), and system dynamic characteristics (such as the rate of change of energy consumption with temperature).

[0160] The design action space includes three types of adjustable parameters: thermal risk judgment threshold (the first threshold is finely adjusted in the range of 25-35, and the second threshold is finely adjusted in the range of 65-75), time series prediction model feature weights (such as the weight coefficient of coolant flow characteristics is adjusted by ±10%), and actuator response coefficients (such as the delay compensation coefficient of the PWM signal of fan 6).

[0161] Iterative optimization process: Each learning cycle lasts 10 minutes. The current state and reward value are input into the reinforcement learning mechanism, and the action parameters are adjusted using the policy gradient method. If the reward value increases after adjusting a certain parameter, the adjustment is retained and the exploration step size is increased; if the reward value decreases, the step size is decreased and fine-tuning is performed in the opposite direction. Through 100-200 iterations, the system converges to the optimal state where the reward value is stable at over 80%.

[0162] Set a retraining cycle, and retrain the temperature field time series prediction model every 24 hours based on 100,000+ operational data (including extreme working condition samples) accumulated on that day. New samples are prioritized to supplement high error working conditions with a temperature deviation rate >5%. The model's ability to fit complex working conditions is improved by increasing the number of neurons in the hidden layer (dynamically adjusted to 32-64).

[0163] The control parameters are calibrated weekly, and the thermal risk judgment threshold is calibrated weekly. The average reward value under different thresholds is calculated using a sliding window (data from the past 7 days), and the threshold combination that maximizes the reward is selected. The actuator response coefficient is updated regularly through impulse response testing. For example, a step command is sent to fan 6 every 3 days, and the response coefficient is corrected according to the rise time of the actual speed.

[0164] The optimized model and parameters are deployed through a canary release mechanism. They are first tested for 2 hours in 5% of the battery modules to verify that the temperature stability and energy consumption indicators meet the standards. Then, they are fully updated to the next round of control process to ensure the safety and continuity of the iteration process.

[0165] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. A heat pipe-assisted immersion battery pack thermal management system, comprising a sealed outer casing (1), characterized in that: A battery pack (3) is installed inside the sealed housing (1). There is a gap between two adjacent battery packs in the battery pack (3). A heat dissipation pipe (4) is inserted in the gap. The lower half of the heat dissipation pipe (4) is located inside the sealed housing (1), and its upper half passes through the sealed housing (1) and extends to the outside. A fan (6) is installed on the outside of the sealed housing (1) near the heat dissipation pipe (4). A wet core (10) is installed on the upper surface of the sealed housing (1) at the end of the heat dissipation pipe (4) away from the fan (6). The sealed housing (1) is filled with coolant. A water pump (7) is also installed inside the sealed housing (1). The water pump (7) is used to draw coolant from the sealed housing (1) and inject it into the wet core (10). The coolant is discharged from the lower end of the wet core (10) and flows back into the sealed housing (1). return include, The perception and evaluation module is used to collect battery temperature parameters, coolant state parameters and environmental parameters in real time, and generate multi-dimensional feature vectors. The battery thermal risk index is calculated based on the collected data. This index is generated by a weighted combination of the battery's maximum temperature, temperature rise rate, and state of charge. The predictive modeling module is used to input multi-dimensional feature vectors into the time series prediction model; predict the spatial distribution of battery temperature field in future periods based on historical temperature data; and calibrate the prediction accuracy by embedding a loss function into the thermodynamic equation. The decision optimization module executes a hierarchical control strategy based on the predicted temperature field and thermal risk index; it uses an evolutionary algorithm to solve for the energy-efficient optimal control command that satisfies the temperature constraints. The execution monitoring module is used to convert control commands into actuator drive signals; monitor the execution status and actual battery temperature data in real time through the closed-loop controller; monitor execution deviation in real time, and activate the fault tolerance mechanism when abnormal flow is detected. The learning evolution module learns from the feedback execution status and actual temperature data; The system parameters are updated by reinforcement learning mechanism, and the objective function takes into account both temperature stability and energy efficiency. Regularly calibrate time series forecasting models and control parameters; also includes: The deviation parameters between the actual operating state of the actuator and the optimal control command are compared, and the spatial distribution difference between the actual battery temperature data and the predicted temperature field is analyzed. Real-time energy consumption data and temperature fluctuation characteristics of the system are collected simultaneously. Based on the deviation parameters and fluctuation characteristics, a reinforcement learning reward rule with temperature stability and energy efficiency as the core optimization objectives is set, and the reward value is dynamically allocated according to the degree of objective achievement. The deviation data and reward rule are input into the reinforcement learning mechanism, and the thermal risk judgment threshold, the feature weights of the time series prediction model, and the actuator response coefficient in the control strategy are adjusted through iterative optimization, so that the system converges to the reward maximization state. The time series prediction model is periodically retrained and the control parameters are calibrated. The model structure and threshold range are dynamically adjusted according to the actual operating effect, and the optimization results are deployed to the next round of control process.

2. The heat pipe-assisted immersion battery pack thermal management system according to claim 1, characterized in that, The perception assessment module also includes: Determine the data acquisition sensor configuration for the battery and cooling system, including setting corresponding sensing devices on the outside of the battery cell, coolant pipeline and sealed housing (1) to monitor the battery cell temperature, coolant state parameters, environmental parameters and obtain the battery state of charge. Based on the sensor configuration, key data features are extracted, including the highest temperature of the battery cell, the rate of temperature rise, the temperature difference of the coolant, the flow fluctuation, the external temperature and humidity of the sealed shell (1) and the state of charge of the battery, and integrated to form a feature information group. Based on the characteristic information group, different weights are assigned to the maximum temperature of the battery cell, the rate of temperature rise, and the state of charge of the battery, and a thermal risk index reflecting the thermal risk of the battery is calculated in a comprehensive manner.

3. The heat pipe-assisted immersion battery pack thermal management system according to claim 2, characterized in that, The predictive modeling module also includes: Retrieve historical temperature datasets associated with the current feature information group, including records of temperature changes in individual battery cells and temperature field distribution patterns, construct a continuous time series in chronological order, and label the time nodes; A time series prediction model based on time dependence is constructed. This model integrates the temperature correlation characteristics of individual battery cells and takes the coolant state and environmental parameters as auxiliary variables as inputs. It learns the temperature field evolution pattern by learning the historical temperature change trend. Input the time series into the time series prediction model, predict the temperature change trend of each battery cell and the spatial distribution of the temperature field of the battery pack (3) in the future period based on the current feature information, and mark the time nodes of the prediction results; Based on thermodynamic principles, a verification standard for the prediction results is established. The model parameters are adjusted by comparing the consistency between the prediction results and the heat conduction law until the prediction results meet the thermodynamic constraints.

4. The heat pipe-assisted immersion battery pack thermal management system according to claim 3, characterized in that, The decision optimization module also includes: Establish a thermal risk classification threshold system, including determining the thermal risk index threshold and temperature field prediction index threshold based on the battery safety temperature range and historical fault data; The real-time thermal risk index and temperature field prediction results are compared with the threshold system. Based on the degree of risk index exceeding the standard and the range of temperature field violation areas, low, medium and high control level signals are matched and output. The system responds to control level signals by calling preset basic control rules and generating corresponding level heat dissipation actuator parameter adjustment schemes, including the linkage control strategy of fan (6) and water pump (7); Under the constraints of parameter adjustment scheme, the optimal control instruction set that meets temperature safety constraints and minimizes total power consumption is selected through simulation and energy consumption evaluation of multiple sets of control instructions.

5. A heat pipe-assisted immersion battery pack thermal management system according to claim 4, characterized in that, The execution monitoring module includes: Convert the control commands of the heat dissipation actuator into drive electrical signals that match its signal type, including converting the fan (6) speed command into a pulse modulation signal and the water pump (7) flow command into a voltage signal; A drive signal is sent to the heat dissipation actuator to start operation. At the same time, the actual operating status of the actuator, battery temperature, and coolant parameters are collected in real time by sensors to form a closed-loop monitoring dataset synchronized with the target parameters and time stamps. The deviation between the actual operating parameters of the actuator and the target parameters is calculated, and an operational anomaly is judged based on the preset deviation threshold, including determining an abnormal flow rate if the flow rate deviation exceeds the limit. In response to abnormal flow, the fault tolerance mechanism is activated, the power of the water pump (7) is increased to the preset safety limit and the heat dissipation capacity of the fan (6) is enhanced in conjunction; if the abnormality continues, an alarm is triggered and the abnormal time and deviation data are recorded to ensure that the battery temperature meets the thermal management safety constraints.

6. A heat pipe-assisted immersion battery pack thermal management system according to claim 1, characterized in that: The wet core (10) is made of multiple sets of horizontally and vertically connected metal tubes that are interwoven with each other, and the internal parts of the multiple sets of metal tubes are interconnected, with space left between the interwoven metal tubes for liquid flow.

7. A heat pipe-assisted immersion battery pack thermal management system according to claim 6, characterized in that: The upper end of the wet core (10) is fixedly connected to a diversion box (9), and the lower end is fixedly connected to a manifold box (11). Both the diversion box (9) and the manifold box (11) are hollow, and their interiors are connected to each other through a metal tube on the wet core (10). The upper surface of the sealing shell (1) is provided with a guide hole (14) below the manifold box (11). The interior of the manifold box (11) is connected to the interior of the sealing shell (1) through the guide hole (14). The outlet of the water pump (7) is fixedly connected to a guide pipe (8). The end of the guide pipe (8) away from the water pump (7) passes through the sealing shell (1) and is connected to the diversion box (9). The interior of the guide pipe (8) is connected to the interior of the diversion box (9).

8. A heat pipe-assisted immersion battery pack thermal management system according to claim 7, characterized in that: The upper and lower ends of the heat dissipation pipe (4) are sealed, and the inside of the heat dissipation pipe (4) is filled with evaporating liquid. Metal fins (5) are fixedly installed on the upper half of the heat dissipation pipe (4), and a protective cover (13) is fixedly installed on the upper end of the sealed outer shell (1). The wet core (10), the fan (6) and the upper half of the heat dissipation pipe (4) are all set inside the protective cover (13).

Images