AI intelligent control thermal management system

Abstract

This invention proposes an AI-powered intelligent control thermal management system. A data acquisition module collects real-time environmental, battery, cabinet, and fire protection data from the energy storage system to construct a multi-dimensional raw data set. A data preprocessing module performs noise reduction, normalization, and data completion on the raw data set to obtain a standardized data set. A digital twin modeling module constructs a digital twin model of the energy storage system based on the standardized data set, integrating the physical mechanisms of heat conduction, convection heat transfer, and battery heat generation. A topology self-learning module continuously trains the standardized data set and the output data of the digital twin model using deep learning algorithms to generate a large-scale energy storage thermal management model. This invention effectively extends battery cycle life, aims to maximize energy efficiency ratio, and optimizes temperature control strategies through a genetic algorithm, avoiding energy waste caused by the start-stop control methods of existing technologies.

Description

Technical Field

[0001] This invention relates to the field of thermal management technology for energy storage systems, and in particular to an AI-powered intelligent control thermal management system. Background Technology

[0002] With the rapid development of the new energy industry, energy storage systems, as core equipment for energy storage and dispatch, directly determine the reliability of new energy power supply through their operational stability and safety. Thermal management systems, as a key subsystem of energy storage systems, bear the important responsibility of maintaining the operating temperature of batteries and core components within their optimal range. Their temperature control accuracy and energy efficiency directly affect the cycle life, charge / discharge efficiency, and safety performance of the energy storage system.

[0003] Currently, mainstream energy storage thermal management solutions on the market generally adopt a single-point independent design. The thermal management unit (such as liquid cooling unit or air-cooled module) is supplied as an independent sub-component. Its control logic only establishes a simple communication connection with the energy management system (EMS) of the energy storage cabinet and executes mechanical action responses. Specifically, the control process of the existing technology is as follows: the EMS sends a start command to the thermal management unit according to the preset target water temperature of the battery (usually 20-25℃). After the thermal management unit starts, it continues to run until the water temperature reaches the target value, and then automatically stops. When the water temperature exceeds the target value range, the EMS sends a start command again, and the thermal management unit repeats the above cycle.

[0004] However, existing thermal management units only receive a single target water temperature signal from the EMS, failing to establish data interaction and coupling with environmental systems (temperature, humidity, dust, etc.), battery systems (SOC, voltage, current, cell temperature uniformity, etc.), the entire cabinet system (heat dissipation duct resistance, cabinet sealing, power consumption of other heat-generating components, etc.), and fire protection systems (fire warning signals, extinguishing medium status, etc.). This results in the control logic lacking a comprehensive understanding of the overall system's operating status. The thermal management unit acts only as a passive execution unit, lacking proactive situational awareness and intelligent decision-making capabilities. It cannot adjust its temperature control strategy based on environmental changes, battery operating status fluctuations, and dynamic changes in the overall cabinet heat load. It can only perform start-stop control based on fixed thresholds, leading to low temperature control accuracy (typically a deviation of ±3℃ or more), and frequent start-stop operations result in energy waste.

[0005] Secondly, existing technologies do not consider energy efficiency optimization targets under different operating scenarios. For example, when the ambient temperature is low, energy consumption can be reduced by combining natural cooling and active cooling, but existing technologies will still activate the active cooling unit. When the battery is in a low SOC and low heat state, excessive cooling will lead to unnecessary energy loss, thereby reducing the overall energy efficiency ratio of the entire energy storage system. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the defects of the existing technology. The present invention proposes an AI intelligent control thermal management system.

[0007] To address the issue that existing technologies' thermal management units only receive a single target water temperature signal from the EMS, failing to establish data interaction and coupling with the environmental system, battery system, overall system, and fire protection system (fire warning signals, extinguishing medium status, etc.), resulting in a lack of comprehensive perception of the entire system's operating status in the control logic, the thermal management unit acts only as a passive execution unit, lacking proactive situational awareness and intelligent decision-making capabilities. It cannot adjust temperature control strategies based on environmental changes, battery operating status fluctuations, and dynamic changes in the overall heat load of the storage unit, and can only perform start-stop control based on fixed thresholds, leading to low temperature control accuracy and energy waste due to frequent start-stop cycles. Furthermore, existing technologies do not consider energy efficiency optimization targets under different operating scenarios; in low SOC and low heat generation states of the battery, excessive cooling leads to unnecessary energy loss, thereby reducing the overall energy efficiency ratio of the entire energy storage system. The technical solution adopted in this invention is: An AI-powered intelligent thermal management method includes the following steps: S1: Data acquisition steps, real-time acquisition of environmental data, battery data, cabinet data and fire protection data of energy storage system, to build a multi-dimensional raw data set; S2: Data preprocessing step, which involves denoising, normalizing and data completion of the original dataset to obtain a standardized dataset; S3: Digital twin model construction steps: Based on the standardized data set, integrate the physical mechanisms of heat conduction, convection heat transfer and battery heat generation to construct a digital twin model of the energy storage system. The digital twin model maps the thermal characteristics of the energy storage system and the coupling relationship of each subsystem. S4: Topology self-learning step, using deep learning algorithms to continuously train the standardized dataset and the output data of the digital twin model to generate a large-scale energy storage thermal management model, which has the self-learning ability to dynamically adjust model parameters; S5: Temperature control demand prediction step, using the energy storage thermal management big model, inputting real-time standardized data, to predict the target temperature range, heat load change trend and potential thermal risks of the energy storage system within a future preset time window; S6: Temperature control strategy formulation steps. With the goal of maximizing energy efficiency ratio, the optimal temperature control strategy is formulated by combining the target temperature range, heat load change trend and potential thermal risks through optimization algorithm. The optimal temperature control strategy includes the operating power of the liquid cooling unit, coolant flow rate, cooling fan speed and start / stop timing. S7: Temperature control execution and feedback adjustment steps: According to the optimal temperature control strategy, control commands are sent to the liquid chiller unit to drive the liquid chiller unit to perform temperature control operations; at the same time, feedback data after temperature control is collected in real time, and the feedback data is input into the energy storage thermal management big model to dynamically adjust the model parameters and the optimal temperature control strategy to achieve closed-loop control.

[0008] Preferably, in step S1, the environmental data includes ambient temperature, ambient humidity, dust concentration, wind speed and solar radiation intensity. The environmental data is collected by a multi-dimensional sensor array deployed outside the energy storage cabinet, with a collection frequency of 1-10Hz. The battery data includes the SOC value of the battery pack, the voltage of individual cells, the temperature of individual cells, the charging and discharging current, and the battery heat generation rate. The battery data is uploaded in real time through the battery management system (BMS). The cabinet data includes the internal temperature field distribution, heat dissipation duct resistance, cabinet sealing coefficient, and overall cabinet heat dissipation efficiency. The cabinet data is collected by temperature sensors, pressure sensors, and flow sensors deployed inside the cabinet. The fire protection data includes fire warning status, extinguishing medium pressure, and fire protection system operation status, and the fire protection data is uploaded in real time through the fire alarm system.

[0009] Preferably, in step S2, the denoising process employs either a moving average method or a wavelet transform method; when using the moving average method, the denoising formula is: in, Let t be the number of environments after denoising. Dynamic average window length , The data collection time interval For ti The original environmental data at that moment; The normalization process uses the min-max normalization method, and the normalization formula is: in For the normalized dataset The minimum value of the data. The maximum value of the original data; The data completion process uses linear interpolation, and the completion formula is as follows: in, The data completed at time t. Effective data acquisition time before and after time t Valid data.

[0010] Preferably, in step S3, the digital twin model is a multiphysics coupling model, including a heat conduction sub-model, a convection heat transfer sub-model, and a battery heat generation sub-model. The heat conduction model is based on Fourier's law, and the formula is: That Density of cabinet and battery materials The specific heat capacity of the material, where T is the temperature. For time, The thermal conductivity of the material Internal heat source intensity; The convective heat transfer sub-model is constructed based on Newton's law of cooling, which is: in, For convective heat transfer, The convective heat transfer coefficient is... Heat exchange area, The wall temperature, Fluid temperature; The battery thermal generation sub-model is constructed based on Joule's law and the polarization thermal generation model, and the formula is: That Heat generation rate per unit volume of battery Charging and discharging current, Battery internal resistance, To balance the battery's potential, This is the battery open-circuit voltage. This refers to the battery terminal voltage. Temperature coefficient of equilibrium potential.

[0011] Preferably, in step S4, the deep learning algorithm is a CNN-LSTM hybrid neural network algorithm, and the topology self-learning step includes: S41: Divide the standardized dataset into a training set, a validation set, and a test set in a 7:2:1 ratio; S42: Construct a CNN-LSTM hybrid neural network, which includes an input layer, a CNN feature extraction layer, an LSTM temporal modeling layer, a fully connected layer, and an output layer. The number of neurons in the input layer is consistent with the dimension of the standardized data. The CNN feature extraction layer includes 2-4 convolutional blocks, the LSTM temporal modeling layer includes 1-3 LSTM units, the fully connected layer includes 1-2 hidden layers, and the output layer has 3 neurons. S43: Using the mean squared error (MSE) as the loss function, the Adam optimizer is used to train the CNN-LSTM hybrid neural network. Dropout regularization is used during training to prevent overfitting. The loss function formula is: Where M is the number of training samples. For the j-th sample, Let be the predicted value for the j-th sample; S44: Adjust the hyperparameters of the neural network using the validation set. If the loss function value on the validation set does not decrease for 5 consecutive epochs, halve the learning rate. S45: Use the test set to evaluate the performance of the trained neural network. When the mean absolute error (MAE) is ≤0.5℃ and MAE is ≤5%, the neural network is determined to be a large-scale energy storage thermal management model; otherwise, return to step S43 to retrain. S46: Every 24 hours, add the newly collected standardized data to the training set and repeat steps S43-S45 to achieve dynamic updates of the large-scale energy storage thermal management model.

[0012] Preferably, in step S5, the preset time window is 5-30 minutes, and the potential thermal risk is determined by the thermal risk assessment index R. The formula for calculating the thermal risk assessment index R is: in, To predict temperature, For the optimal operating temperature of the battery, This is the upper limit of the battery's safe temperature. To predict heat load, For rated heat load, For the maximum allowable heat load, This is the fire early warning status coefficient. coefficient, The values ​​range from 0.2 to 0.5; when R ≥ 0.7, it is considered high risk; when 0.3 ≤ R < 0.7, it is considered medium risk; when R < 0.3, it is considered low risk.

[0013] Preferably, in step S6, the optimization algorithm is a genetic algorithm, and the temperature control strategy formulation step includes: S61: Define the optimization variable vector X=[P,Q,V], where P is the operating power of the liquid chiller unit, Q is the coolant flow rate, and V is the cooling fan speed; S62: Define the objective function for optimization: in Cooling capacity of liquid chiller Total power consumption of the chiller unit and cooling fans; S63: Set constraints: in Battery temperature Minimum allowable operating temperature of the pool Maximum permissible operating temperature of the pool These are the minimum and maximum operating power of the liquid-cooled unit, respectively. These are the minimum and maximum flow rates of the coolant. These are the minimum and maximum speeds of the cooling fan, respectively. S64: Use a genetic algorithm to solve the above optimization problem and obtain the optimal variable vector X*=[P*,Q*,V*]. Based on the optimal variable vector, formulate the optimal temperature control strategy.

[0014] Preferably, in step S7, the feedback data includes the battery temperature after temperature control, the inlet and outlet temperatures of the coolant, the operating power of the liquid cooler unit, and the speed of the cooling fan; the closed-loop control uses a PID adjustment algorithm to correct the temperature control strategy, and the PID adjustment formula is: in, For control quantity correction value, This is the proportionality coefficient. Integral coefficient, These are the differential coefficients. Let t be the temperature deviation at time t.

[0015] An AI-powered intelligent control thermal management system, characterized in that it comprises: The data acquisition module is used to collect environmental data, battery data, cabinet data and fire protection data of the energy storage system in real time, and to build a multi-dimensional raw data set; The data preprocessing module is used to perform noise reduction, normalization, and data completion on the original dataset to obtain a standardized dataset. The digital twin modeling module is used to construct a digital twin model of the energy storage system based on the standardized data set and by integrating the physical mechanisms of heat conduction, convection heat transfer and battery heat generation. The topology self-learning module is used to continuously train the standardized dataset and the output data of the digital twin model using deep learning algorithms to generate a large-scale energy storage thermal management model. The temperature control demand prediction module is used to use the large energy storage thermal management model to input real-time standardized data and predict the target temperature range, heat load change trend and potential thermal risks of the energy storage system within a future preset time window. The temperature control strategy formulation module is used to formulate the optimal temperature control strategy with the goal of maximizing the energy efficiency ratio, combined with the target temperature range, heat load change trend and potential thermal risks, through optimization algorithms. The temperature control execution module is used to send control commands to the liquid chiller unit according to the optimal temperature control strategy, and drive the liquid chiller unit to perform temperature control operations. The feedback adjustment module is used to collect feedback data after temperature control in real time, input the feedback data into the large energy storage thermal management model, dynamically adjust the model parameters and the optimal temperature control strategy, and realize closed-loop control. The data acquisition module, data preprocessing module, digital twin modeling module, topology self-learning module, temperature control demand prediction module, temperature control strategy formulation module, temperature control execution module, and feedback adjustment module are all integrated into the thermal management controller. The thermal management controller establishes data interaction with the liquid chiller, BMS, fire alarm system, and sensor array through a communication bus.

[0016] Preferably, the data acquisition module includes an environmental sensor array, a battery data acquisition unit, a cabinet data acquisition unit, and a fire protection data acquisition unit; The environmental sensor array includes a temperature sensor, a humidity sensor, a dust sensor, a wind speed sensor, and a solar radiation intensity sensor. The battery data acquisition unit is connected to the BMS to acquire data such as the SOC value of the battery pack, the voltage of individual cells, and the temperature of individual cells in real time. The cabinet data acquisition unit includes a temperature sensor, a pressure sensor, and a flow sensor inside the cabinet. The fire data acquisition unit is connected to the fire alarm system to acquire data such as fire early warning status and fire extinguishing medium pressure in real time. The communication bus uses CAN485 communication or Ethernet bus.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention, through multi-system data coupling, precise digital twin modeling, and AI prediction algorithms, can predict battery heat load changes in advance and dynamically adjust the temperature control strategy to keep battery temperature fluctuations within ±0.5℃, which is far superior to the ±3℃ temperature control accuracy of existing technologies. This ensures that the battery always operates in the optimal temperature range, effectively extending battery cycle life. With the goal of maximizing energy efficiency, the invention optimizes the temperature control strategy through a genetic algorithm, avoiding energy waste caused by the start-stop control of existing technologies.

[0018] This invention breaks the decoupling of subsystems in existing technologies, achieving real-time interaction and deep coupling of data from multiple systems such as environment, battery, cabinet, and fire protection. This enables thermal management decisions to comprehensively consider the overall system operation, avoiding the biased decision-making caused by single data-driven approaches and improving system operational stability. The large-scale energy storage thermal management model of this invention possesses continuous self-learning capabilities, dynamically optimizing model parameters based on long-term operating data of the energy storage system. It adapts to conditions such as battery degradation, aging of heat dissipation components, and changes in the operating environment, maintaining system temperature control accuracy and energy efficiency ratio without manual intervention, significantly reducing operation and maintenance costs. Attached Figure Description

[0019] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein: Figure 1 This is a diagram showing the overall architecture of the AI ​​intelligent control thermal management system of the present invention. Figure 2 This is a flowchart of the AI ​​intelligent control thermal management method of the present invention. Detailed Implementation

[0020] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0021] Specific embodiments of the present invention are described below with reference to the accompanying drawings.

[0022] Please see Figures 1-2 This embodiment proposes an AI-powered intelligent control thermal management system, including: The data acquisition module is used to collect environmental data, battery data, cabinet data and fire protection data of the energy storage system in real time, and to build a multi-dimensional raw data set; The data acquisition module includes an environmental sensor array, a battery data acquisition unit, a cabinet data acquisition unit, and a fire protection data acquisition unit; The environmental sensor array includes temperature sensors, humidity sensors, dust sensors, wind speed sensors, and solar radiation intensity sensors; the battery data acquisition unit communicates with the BMS to acquire data such as the SOC value of the battery pack, individual battery voltage, and individual battery temperature in real time; the cabinet data acquisition unit includes internal temperature sensors, pressure sensors, and flow sensors; the fire protection data acquisition unit communicates with the fire alarm system to acquire data such as fire warning status and extinguishing medium pressure in real time; the communication bus adopts CAN485 communication or Ethernet bus, with a data transmission rate ≥100Mbps.

[0023] The data preprocessing module is used to perform noise reduction, normalization, and data completion on the original dataset to obtain a standardized dataset.

[0024] The digital twin modeling module is used to construct a digital twin model of an energy storage system based on a standardized dataset, integrating the physical mechanisms of heat conduction, convection heat transfer, and battery heat generation.

[0025] The topology self-learning module is used to continuously train the standardized dataset and the output data of the digital twin model using deep learning algorithms to generate a large-scale energy storage thermal management model.

[0026] The temperature control demand prediction module is used to predict the target temperature range, heat load change trend and potential thermal risks of the energy storage system within a preset time window by using a large-scale energy storage thermal management model and inputting real-time standardized data.

[0027] The temperature control strategy formulation module is used to formulate the optimal temperature control strategy with the goal of maximizing the energy efficiency ratio, taking into account the target temperature range, the trend of heat load change and potential thermal risks, through optimization algorithms.

[0028] The temperature control execution module is used to send control commands to the liquid chiller unit according to the optimal temperature control strategy, and drive the liquid chiller unit to perform temperature control operations.

[0029] The feedback adjustment module is used to collect feedback data after temperature control in real time, input the feedback data into the large-scale energy storage thermal management model, dynamically adjust the model parameters and the optimal temperature control strategy, and realize closed-loop control.

[0030] The data acquisition module, data preprocessing module, digital twin modeling module, topology self-learning module, temperature control demand prediction module, temperature control strategy formulation module, temperature control execution module, and feedback adjustment module are all integrated into the thermal management controller. The thermal management controller establishes data interaction with the liquid chiller, BMS, fire alarm system, and sensor array through the communication bus.

[0031] An AI-powered intelligent control thermal management method, wherein the executing entity is a thermal management controller, includes the following steps: Data acquisition steps: This controller acquires environmental data, battery data, cabinet data, and fire protection data of the energy storage system in real time to construct a multi-dimensional raw data set; In the data acquisition process, environmental data includes ambient temperature (E_T), ambient humidity (E_H), dust concentration (E_D), wind speed (E_V), and solar radiation intensity (E_S). This environmental data is collected through a multi-dimensional sensor array deployed outside the energy storage cabinet, with a collection frequency of 1-10Hz. Battery data includes the battery pack's SOC value (B_S), individual cell voltage (B_V), individual cell temperature (B_T), charge / discharge current (B_I), and battery heat generation rate (B_Q). This battery data is uploaded to the controller in real-time through the Battery Management System (BMS). Cabinet-wide data includes the internal temperature field distribution (C_T), heat dissipation duct resistance (C_R), cabinet sealing coefficient (C_S), and overall cabinet heat dissipation efficiency (C_E). This cabinet-wide data is collected through temperature sensors, pressure sensors, and flow sensors deployed inside the cabinet. Fire protection data includes fire warning status (F_W), extinguishing medium pressure (F_P), and fire protection system operating status (F_S). This fire protection data is uploaded to the controller in real-time through the fire alarm system.

[0032] Data preprocessing steps: This controller performs noise reduction, normalization, and data completion on the original dataset to obtain a standardized dataset; In the data preprocessing step, denoising is performed using either the moving average method or the wavelet transform method; when using the moving average method, the denoising formula is: in, Let t be the number of environments after denoising. Dynamic average window length , The data collection time interval For ti The original environmental data at that moment; The normalization process uses the min-max normalization method, and the normalization formula is: in For the normalized dataset The minimum value of the data. The maximum value of the original data; The data completion process uses linear interpolation, and the completion formula is as follows: in, The data completed at time t. Effective data acquisition time before and after time t Valid data.

[0033] Digital twin model construction steps: Based on a standardized dataset, this controller integrates the physical mechanisms of heat conduction, convection heat transfer, and battery heat generation to construct a digital twin model of the energy storage system. The digital twin model maps the thermal characteristics of the energy storage system and the coupling relationships of each subsystem. In the digital twin model construction process, the digital twin model is a multi-physics coupled model, including a heat conduction sub-model, a convection heat transfer sub-model, and a battery heat generation sub-model. The heat conduction model is based on Fourier's law, and the formula is: That Density of cabinet and battery materials (kg / m³) The specific heat capacity of the material (J / (kg·K)), where T is the temperature (K). For time (s), Thermal conductivity of the material (W / (m·K)) Internal heat source intensity (W / m³); The convective heat transfer sub-model is constructed based on Newton's law of cooling, which is: in, For convective heat transfer (W). The convective heat transfer coefficient is (W / (m²·K)). Heat exchange area (m²) The wall temperature is (K). Fluid temperature (K); The battery thermal generator model is constructed based on Joule's law and the polarization thermal generation model, and the formula is: That Heat generation rate per unit volume of battery (W / m³). Charging and discharging current (A) Battery internal resistance (Ω). The battery equilibrium potential (V). This is the battery open-circuit voltage (V). This is the battery terminal voltage (V). Temperature coefficient of equilibrium potential (V / K).

[0034] Topology self-learning steps: This controller uses deep learning algorithms to continuously train the standardized dataset and the output data of the digital twin model to generate a large-scale energy storage thermal management model. The large-scale energy storage thermal management model has the self-learning ability to dynamically adjust model parameters. In the topology self-learning step, the deep learning algorithm is a CNN-LSTM hybrid neural network algorithm, and the topology self-learning steps include: This controller divides the standardized dataset into training, validation, and test sets in a 7:2:1 ratio. This controller constructs a CNN-LSTM hybrid neural network, which includes an input layer, a CNN feature extraction layer, an LSTM temporal modeling layer, a fully connected layer, and an output layer. The number of neurons in the input layer is consistent with the dimension of the standardized data. The CNN feature extraction layer includes 2-4 convolutional blocks (each convolutional block contains a convolutional layer, a batch normalization layer, and a ReLU activation function). The LSTM temporal modeling layer includes 1-3 LSTM units (each LSTM unit has 64-256 hidden layer neurons). The fully connected layer includes 1-2 hidden layers (each hidden layer has 32-128 neurons). The output layer has 3 neurons (outputting the target temperature prediction, heat load prediction, and heat risk level, respectively). This controller uses mean squared error (MSE) as the loss function and employs the Adam optimizer to train the CNN-LSTM hybrid neural network. Dropout regularization (with a dropout rate of 0.1-0.3) is used during training to prevent overfitting. The loss function formula is as follows: Where M is the number of training samples. For the j-th sample, Let be the predicted value for the j-th sample; This controller adjusts the hyperparameters of the neural network (including learning rate, kernel size, and number of LSTM units) using the validation set. When the loss function value on the validation set does not decrease for 5 consecutive epochs, the learning rate is halved. This controller uses a test set to evaluate the performance of the trained neural network. When the mean absolute error (MAE) is ≤0.5℃ (temperature prediction) and MAE is ≤5% (heat load prediction), the neural network is determined to be a large-scale energy storage thermal management model; otherwise, it returns to step S43 for retraining. Every 24 hours, this controller adds newly collected standardized data to the training set and repeats steps S43-S45 to achieve dynamic updates of the large-scale energy storage thermal management model.

[0035] Temperature demand prediction steps: This controller uses a large-scale energy storage thermal management model, inputs real-time standardized data, and predicts the target temperature range, heat load change trend, and potential thermal risks of the energy storage system within a preset time window in the future. In the temperature control demand prediction step, the preset time window is 5-30 minutes. Potential thermal risks are assessed using the thermal risk assessment index R. The formula for calculating the thermal risk assessment index R is: in, To predict temperature (K). The optimal operating temperature (K) for the battery. This is the upper limit of the battery's safe temperature (K). To predict heat load (W) Rated heat load (W). The maximum allowable heat load (W) This is the fire warning status coefficient (F_W=0 indicates no warning, F_W=1 indicates a warning). coefficient( (The values ​​range from 0.2 to 0.5). When R ≥ 0.7, it is considered high risk; when 0.3 ≤ R < 0.7, it is considered medium risk; and when R < 0.3, it is considered low risk.

[0036] Temperature control strategy formulation steps: This controller aims to maximize the energy efficiency ratio. Combining the target temperature range, heat load change trend and potential thermal risks, it formulates the optimal temperature control strategy through optimization algorithm. The optimal temperature control strategy includes the operating power of the liquid chiller, coolant flow rate, cooling fan speed and start / stop timing. In the temperature control strategy formulation process, the optimization algorithm is a genetic algorithm, and the temperature control strategy formulation steps include: This controller defines an optimization variable vector X=[P,Q,V], where P is the operating power of the liquid chiller unit (kW), Q is the coolant flow rate (L / min), and V is the cooling fan speed (r / min). This controller is configured with an optimization objective function (maximizing energy efficiency ratio): in Cooling capacity (kW) of liquid chiller unit Total power consumption of the chiller and cooling fan (kW); This controller has the following constraints: in Battery temperature (K) Minimum permissible operating temperature of the pool (K) Maximum permissible operating temperature of the pool (K). These are the minimum and maximum operating power (kW) of the liquid-cooled unit, respectively. These are the minimum and maximum flow rates (L / min) of the coolant. These are the minimum and maximum speeds (r / min) of the cooling fan, respectively. This controller uses a genetic algorithm to solve the above optimization problem and obtains the optimal variable vector X*=[P*,Q*,V*]. Based on the optimal variable vector, the optimal temperature control strategy is formulated.

[0037] Temperature control execution and feedback adjustment steps: This controller sends control commands to the liquid chiller unit according to the optimal temperature control strategy, driving the liquid chiller unit to perform temperature control operations; at the same time, this controller collects feedback data after temperature control in real time, inputs the feedback data into the energy storage thermal management big model, dynamically adjusts the model parameters and the optimal temperature control strategy, and realizes closed-loop control; In the temperature control execution and feedback adjustment steps, the feedback data includes the battery temperature after temperature control, the coolant inlet and outlet temperatures, the liquid cooler unit operating power, and the cooling fan speed; the closed-loop control uses a PID control algorithm to correct the temperature control strategy, and the PID control formula is: in, For control quantity correction value, This is the proportionality coefficient (ranging from 0.1 to 1.0). Integral coefficient (range 0.01-0.1). These are the differential coefficients (ranging from 0.001 to 0.01). The temperature deviation at time t ( (where t is the actual battery temperature at time t).

[0038] System hardware configuration In this embodiment, the AI ​​intelligent control thermal management system is applied to a 5MWh containerized energy storage power station, and its hardware configuration is as follows: Thermal management controller: It adopts an industrial-grade embedded controller with an ARM Cortex-A9 quad-core processor as the main chip, with a main frequency of 1.2GHz, 2GB of memory, and 32GB of storage capacity; it integrates 2 CAN485 communication interfaces, 2 Ethernet interfaces, and 4 RS485 interfaces, supporting a data transmission rate of 100Mbps; the operating temperature range is -45℃ to 85℃, which meets the harsh working environment requirements of energy storage power stations.

[0039] Data acquisition module: Environmental sensor array: Temperature sensor (measurement range -45℃~85℃, accuracy ±0.1℃), humidity sensor (measurement range 0~100%RH, accuracy ±2%RH), dust sensor (measurement range 0~1000μg / m³, accuracy ±5%), wind speed sensor (measurement range 0~30m / s, accuracy ±0.1m / s), and solar radiation intensity sensor (measurement range 0~2000W / m², accuracy ±5%) are deployed on the outside of the container, with a sampling frequency of 5Hz.

[0040] Battery data acquisition unit: Communicates with the battery management system (BMS) via CAN485 to acquire in real time the SOC value (accuracy ±1%), individual cell voltage (measurement range 2.5~4.2V, accuracy ±0.001V), individual cell temperature (measurement range -30℃~80℃, accuracy ±0.1℃), charge / discharge current (measurement range -500A~500A, accuracy ±0.5%FS), and battery heat generation rate (calculation accuracy ±3%) of 16 battery packs.

[0041] The container's data acquisition unit consists of 20 temperature sensors (measurement range -30℃ to 80℃, accuracy ±0.1℃) evenly deployed inside the container to collect the internal temperature field distribution; 4 pressure sensors (measurement range 0 to 1MPa, accuracy ±0.5%FS) to collect the resistance of the cooling duct; and 2 flow sensors (measurement range 0 to 100L / min, accuracy ±1%FS) to collect the coolant flow rate. The overall container sealing coefficient is calculated from the air leakage rate measured by the pressure sensors.

[0042] Fire data acquisition unit: Communicates with the fire alarm system via RS485 interface to obtain real-time fire warning status (0=no warning, 1=warning), extinguishing medium pressure (measurement range 0~10MPa, accuracy ±0.5%FS) and fire system operation status (0=normal, 1=fault).

[0043] Temperature control module: adopts liquid-cooled unit (cooling capacity 50kW, operating power 5~20kW), coolant is 50% ethylene glycol aqueous solution; cooling fan is variable frequency fan (speed range 500~3000r / min, power 0.5~2kW); liquid-cooled unit communicates with thermal management controller via CAN485, receives control commands and feeds back operating status.

[0044] Communication bus: CAN485 communication (1Mbps transmission rate) is used to connect the thermal management controller with the BMS, liquid cooling unit, and fire alarm system; Ethernet bus (100Mbps transmission rate) is used to connect the thermal management controller with the sensor array to ensure the real-time performance and stability of data transmission.

[0045] In this embodiment, data preprocessing is performed according to the following steps: Denoising: A moving average method is used, with a moving average window length of N=5, to denoise time-series data such as ambient temperature and battery temperature. The denoising formula is as follows: in (Sampling frequency 5Hz) This is the raw ambient temperature data at time t-0.2i.

[0046] Normalization: The min-max normalization method is used to normalize all data. For example, the ambient temperature range is -45℃ to 85℃, and the normalization formula is: Normalized data is mapped to the [0,1] interval, which facilitates neural network training.

[0047] Data completion processing: When data is missing at a certain moment (e.g., due to sensor malfunction), linear interpolation is used to complete it. For example, if the battery temperature data at t=10s is missing, and the temperature at t=9.8s is 25.0℃ and the temperature at t=10.2s is 25.2℃, then the completion formula is: Digital Twin Model Construction In this embodiment, the parameters of the digital twin model are set as follows: Thermal conductivity model: The energy storage cabinet is made of Q235 steel, and is dense. =7850kg / m³, specific heat capacity c=460J / (kg·K), thermal conductivity k=50W / (m·K); the battery casing material is aluminum alloy, density ρ=2700kg / m³, specific heat capacity c=900J / (kg·K), thermal conductivity k=200W / (m·K); the internal heat source intensity q includes the battery heat generation rate and the heat dissipation power of other electronic components (set to 500W).

[0048] Convection heat transfer sub-model: The convective heat transfer coefficient of the coolant (50% ethylene glycol aqueous solution) is h = 1000 W / (m²·K), and the heat transfer area is A = 10 m²; the convective heat transfer coefficient of air is h = 30 W / (m²·K), and the heat transfer area is A = 20 m².

[0049] Battery thermal generator model: Battery internal resistance R = 0.005Ω, temperature coefficient of equilibrium potential. Open circuit voltage (When SOC=50%), terminal voltage Data is collected in real time by BMS.

[0050] The above model was solved using COMSOL Multiphysics software to obtain parameters such as the temperature field distribution and heat flux density of the energy storage system, thus realizing real-time mapping between the physical world and the virtual world.

[0051] Topology self-learning and large-scale thermal management model training In this embodiment, a large-scale energy storage thermal management model is constructed using a CNN-LSTM hybrid neural network, with the following specific parameters: Neural network structure: Input layer: Number of neurons = 28 (5-dimensional environmental data + 10-dimensional battery data + 8-dimensional cabinet data + 5-dimensional fire protection data); CNN feature extraction layer: 2 convolutional blocks, the first convolutional block (3×3 kernel size, 32 kernels, stride 1, ReLU activation function), the second convolutional block (3×3 kernel size, 64 kernels, stride 1, ReLU activation function), each convolutional block is followed by a max pooling layer (2×2 pooling kernel size, stride 2). LSTM temporal modeling layer: 1 LSTM unit, number of hidden layer neurons = 128, dropout rate = 0.2; Fully connected layer: 1 hidden layer, 64 neurons, ReLU activation function; Output layer: Number of neurons = 3 (target temperature prediction, heat load prediction, heat risk level).

[0052] Training parameters: Training set sample size = 70,000 (historical 3 months of running data, collection frequency 5Hz, calculated based on 86,400 data points per day); The number of samples in the validation set is 20,000, and the number of samples in the test set is 10,000. Loss function: Mean Squared Error (MSE); Optimizer: Adam optimizer, initial learning rate = 0.001; Training epoch=100, batch size=64.

[0053] Model evaluation: After training, the mean absolute error (MAE) of temperature prediction on the test set was 0.3℃, the MAE of heat load prediction was 3%, and the accuracy of heat risk level prediction was 98%, which meets the design requirements.

[0054] Topology self-learning: Every 24 hours, the 86,400 data points collected that day are added to the training set, the model is retrained, and the weights and biases of the neural network are adjusted to achieve dynamic updates of the model.

[0055] Temperature control demand forecasting and strategy formulation Temperature control demand prediction: Set a preset time window of 15 minutes, and use the weighting coefficient of the thermal risk assessment index. , Optimal battery operating temperature Rated heat load .

[0056] For example, when real-time standardized data is input into the model, the predicted battery temperature is obtained after 15 minutes. heat negative Thermal risk assessment index: It was determined to be low risk.

[0057] Temperature control strategy formulation: The optimal temperature control strategy is solved by a genetic algorithm, and the optimized variable vector is X=[P,Q,V], where the operating power of the liquid chiller unit P ranges from 5 to 20kW, the coolant flow rate Q ranges from 20 to 80L / min, and the cooling fan speed V ranges from 1000 to 3000r / min.

[0058] The objective function to be optimized is: That The cooling capacity of the liquid chiller (positively correlated with P and Q, fitting formula) $P_f$ represents the power consumption of the cooling fan (positively correlated with V, according to the fitting formula). ).

[0059] The constraints are: The optimal variable vector X*=[8kW, 40L / min, 1500r / min] was obtained by solving the problem using a genetic algorithm, corresponding to an energy efficiency ratio (COP) of 4.2. At this point, the liquid-cooled unit's cooling capacity is [value missing]. Fan function Total merit This satisfies the goal of maximizing energy efficiency ratio.

[0060] Closed-loop control execution Temperature control execution: The thermal management controller sends control commands to the liquid chiller unit according to the optimal temperature control strategy: operating power 8kW, coolant flow rate 40L / min; and sends control commands to the cooling fan: speed 1500r / min.

[0061] Feedback Adjustment: The thermal management controller collects feedback data in real time after temperature control. If t=5 minutes, the actual battery temperature... Temperature bias .

[0062] The PID control algorithm is used to correct the control quantity, and the proportional coefficient is set. Differential coefficients ,but: Assuming the integral term is -0.2 and the differential term is -0.02, then , correspondingly adjust the operating power of the liquid cooling unit to 8 - 0.5 = 7.5 kW, keep the coolant flow rate at 40 L / min, and keep the fan speed at 1500 r / min.

[0063] Model update: Input the feedback data (actual temperature, flow rate, power, etc.) into the large - scale energy storage thermal management model, and adjust the weight parameters of the model to ensure the accuracy of the next prediction and decision.

[0064] Thermal risk warning and handling When the thermal risk assessment index R ≥ 0.7 (high risk) is predicted by the model, for example, when predicting the battery temperature , then: At this time, the thermal management controller immediately takes the following measures: Adjust the temperature control strategy: Increase the operating power of the liquid cooling unit to 20 kW, increase the coolant flow rate to 80 L / min, and increase the fan speed to 3000 r / min to maximize the cooling capacity; Trigger an early warning signal: Send a thermal risk warning message to the EMS of the energy storage power station through the communication bus, including the predicted temperature, heat load, risk level, etc.; Link with the fire protection system: If the actual temperature continues to rise within 10 minutes, send a trigger signal to the fire alarm system to start the fire - fighting plan.

[0065] Use examples Application scenarios In this embodiment, the AI intelligent control thermal management system is applied to a 5MWh container - type energy storage power station supporting a certain new - energy power station. This energy storage power station uses lithium iron phosphate batteries, with a rated charge - discharge current of 150 A, an operating temperature range of - 30°C to 55°C, and an optimal operating temperature of 20 - 25°C; the area where the energy storage power station is located has a temperate continental climate, with a maximum ambient temperature of 45°C in summer, a minimum ambient temperature of - 30°C in winter, large fluctuations in ambient temperature in spring and autumn, and there are dust - laden weather conditions.

[0066] System deployment and debugging Hardware deployment: According to the hardware configuration requirements of the present invention, deploy an environmental sensor array outside the energy storage container, and deploy cabinet temperature sensors, pressure sensors, and flow sensors inside; install the thermal management controller in the control room of the container, connect to the BMS, liquid cooling unit, and fire alarm system through CAN485 communication, and connect to the sensor array through an Ethernet bus; install the liquid cooling unit at the end of the container, and arrange the coolant pipes along the battery packs to ensure uniform heat exchange.

[0067] Software debugging: The control program of this invention is burned into the thermal management controller, including modules such as data acquisition, preprocessing, digital twin modeling, topology self-learning, prediction, strategy formulation and closed-loop control; import the operating data of the past 3 months (including environmental, battery, cabinet and fire protection data under different seasons and charging and discharging conditions) to perform initial training on the large-scale energy storage thermal management model; debug the communication interface to ensure normal data transmission between devices; set the optimal battery operating temperature of 25℃, the upper limit of safe temperature of 45℃, and the warning threshold and other parameters.

[0068] Trial operation: Conduct a one-month trial operation, during which the system's operating status will be monitored: the completeness and accuracy of data acquisition, the accuracy of model prediction, the rationality of temperature control strategy, and the response speed of closed-loop adjustment, etc.; based on the trial operation data, optimize PID parameters, genetic algorithm weight coefficients, etc., to ensure that the system meets the design specifications.

[0069] Actual operating effect After six months of actual operation, the AI ​​intelligent control thermal management system of this invention has achieved the following results: Temperature control accuracy: The battery temperature remains stable between 24.5~25.5℃ with a fluctuation range of ±0.5℃, which is far superior to existing technologies (the original system uses traditional EMS control with a temperature fluctuation of ±3℃); even under high summer temperatures (ambient temperature 45℃) or low winter temperatures (ambient temperature -30℃), it can still maintain a stable temperature range without exceeding the temperature limit.

[0070] Energy Efficiency Ratio: The average energy efficiency ratio of the thermal management system reaches 4.0, which is 42.9% higher than the original traditional system (energy efficiency ratio 2.8); under high temperature conditions in summer, the energy efficiency ratio is 3.5, and under low temperature conditions in winter, the energy efficiency ratio reaches 5.2 (by combining natural cooling and active cooling to reduce energy consumption); within 6 months, the total power consumption of the thermal management system is 8640kWh, which is 3360kWh less than the original system (12000kWh), showing significant energy-saving effect.

[0071] Early warning and safety: Three potential thermal risks occurred during operation: one was due to increased battery charging heat during the high-temperature period in summer, and the other was due to sand and dust blocking the heat dissipation air duct, resulting in reduced heat dissipation efficiency. The system predicted the risks 8 to 10 minutes in advance and successfully avoided temperature runaway by adjusting the temperature control strategy (increasing cooling power and fan speed) or triggering early warnings (prompting maintenance personnel to clean up the sand and dust). No safety accidents occurred.

[0072] Adaptive capability: As the battery degrades (the battery SOC degradation rate is 3% after 6 months), the large-scale energy storage thermal management model dynamically adjusts the parameters of the battery thermal generation sub-model through topology self-learning to ensure that the temperature control accuracy is not affected; when sandstorms cause increased resistance in the heat dissipation duct, the system automatically increases the fan speed to compensate for the decrease in heat dissipation efficiency without manual intervention.

[0073] Operation and maintenance costs: Due to the system's self-learning and self-adaptive capabilities and stable operation, only two routine inspections were conducted within 6 months, and no equipment failures or repairs occurred, resulting in a 60% reduction in operation and maintenance costs compared to the original system.

[0074] The comparison with existing technologies is shown in the table below: As can be seen from the comparison, the AI ​​intelligent control thermal management system of the present invention is significantly superior to the existing technology in terms of temperature control accuracy, energy efficiency ratio, safety, and operation and maintenance costs, and can effectively meet the needs of energy storage power stations to develop towards high capacity, high density, and long life.

[0075] Summarize This invention achieves intelligent, precise, and efficient thermal management of energy storage systems through key technologies such as multi-dimensional data acquisition and preprocessing, digital twin model construction, topology self-learning large model training, AI intelligent prediction, optimal temperature control strategy formulation, and closed-loop control execution. This invention overcomes the system decoupling and data silo problems of existing technologies, enabling the thermal management system to possess proactive situational awareness, intelligent decision-making, early warning, and adaptive adjustment capabilities. It significantly improves temperature control accuracy and energy efficiency ratio, reduces the incidence of thermal risks and operation and maintenance costs, and provides reliable technical support for the safe, stable, and efficient operation of energy storage systems.

[0076] The technical solution of this invention is not only applicable to containerized energy storage power stations, but can also be flexibly adapted to other energy storage scenarios such as distributed energy storage devices and power battery packs, possessing broad application prospects and promotional value. In the future, the computational efficiency of the AI ​​algorithm can be further optimized to improve the prediction accuracy of the model under extreme operating conditions, while expanding the linkage with other subsystems of the energy storage system (such as charge and discharge control and energy dispatch) to achieve coordinated optimization of the entire system.

[0077] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. An AI-powered intelligent control thermal management method, characterized in that, The method includes the following steps: S1: Data acquisition steps, real-time acquisition of environmental data, battery data, cabinet data and fire protection data of energy storage system, to build a multi-dimensional raw data set; S2: Data preprocessing step, which involves denoising, normalizing and data completion of the original dataset to obtain a standardized dataset; S3: Digital twin model construction steps: Based on the standardized data set, integrate the physical mechanisms of heat conduction, convection heat transfer and battery heat generation to construct a digital twin model of the energy storage system. The digital twin model maps the thermal characteristics of the energy storage system and the coupling relationship of each subsystem. S4: Topology self-learning step, using deep learning algorithms to continuously train the standardized dataset and the output data of the digital twin model to generate a large-scale energy storage thermal management model, which has the self-learning ability to dynamically adjust model parameters; S5: Temperature control demand prediction step, using the energy storage thermal management big model, inputting real-time standardized data, to predict the target temperature range, heat load change trend and potential thermal risks of the energy storage system within a future preset time window; S6: Temperature control strategy formulation steps. With the goal of maximizing energy efficiency ratio, the optimal temperature control strategy is formulated by combining the target temperature range, heat load change trend and potential thermal risks through optimization algorithm. The optimal temperature control strategy includes the operating power of the liquid cooling unit, coolant flow rate, cooling fan speed and start / stop timing. S7: Temperature control execution and feedback adjustment steps: According to the optimal temperature control strategy, control commands are sent to the liquid chiller unit to drive the liquid chiller unit to perform temperature control operations; at the same time, feedback data after temperature control is collected in real time, and the feedback data is input into the energy storage thermal management big model to dynamically adjust the model parameters and the optimal temperature control strategy to achieve closed-loop control.

2. The AI ​​intelligent control thermal management method according to claim 1, characterized in that, In step S1, the environmental data includes ambient temperature, ambient humidity, dust concentration, wind speed and solar radiation intensity. The environmental data is collected by a multi-dimensional sensor array deployed outside the energy storage cabinet, with a collection frequency of 1-10Hz. The battery data includes the SOC value of the battery pack, the voltage of individual cells, the temperature of individual cells, the charging and discharging current, and the battery heat generation rate. The battery data is uploaded in real time through the battery management system (BMS). The cabinet data includes the internal temperature field distribution, heat dissipation duct resistance, cabinet sealing coefficient, and overall cabinet heat dissipation efficiency. The cabinet data is collected by temperature sensors, pressure sensors, and flow sensors deployed inside the cabinet. The fire protection data includes fire warning status, extinguishing medium pressure, and fire protection system operation status, and the fire protection data is uploaded in real time through the fire alarm system.

3. The AI ​​intelligent control thermal management method according to claim 1, characterized in that, In step S2, the denoising process employs either the moving average method or the wavelet transform method; when the moving average method is used, the denoising formula is: in, Let t be the number of environments after denoising. Dynamic average window length , The data collection time interval For ti The original environmental data at that moment; The normalization process uses the min-max normalization method, and the normalization formula is: in For the normalized dataset The minimum value of the data. The maximum value of the original data; The data completion process uses linear interpolation, and the completion formula is as follows: in, The data completed at time t. Effective data acquisition time before and after time t Valid data.

4. The AI ​​intelligent control thermal management method according to claim 1, characterized in that, In step S3, the digital twin model is a multiphysics coupling model, including a heat conduction sub-model, a convection heat transfer sub-model, and a battery heat generation sub-model. The heat conduction model is based on Fourier's law, and the formula is: That Density of cabinet and battery materials The specific heat capacity of the material, where T is the temperature. For time, The thermal conductivity of the material Internal heat source intensity; The convective heat transfer sub-model is constructed based on Newton's law of cooling, which is: in, For convective heat transfer, The convective heat transfer coefficient is... Heat exchange area, The wall temperature, Fluid temperature; The battery thermal generation sub-model is constructed based on Joule's law and the polarization thermal generation model, and the formula is: That Heat generation rate per unit volume of battery Charging and discharging current, Battery internal resistance, To balance the battery's potential, This is the battery open-circuit voltage. This refers to the battery terminal voltage. Temperature coefficient of equilibrium potential.

5. The AI ​​intelligent control thermal management method according to claim 1, characterized in that, In step S4, the deep learning algorithm is a CNN-LSTM hybrid neural network algorithm, and the topology self-learning step includes: S41: Divide the standardized dataset into a training set, a validation set, and a test set in a 7:2:1 ratio; S42: Construct a CNN-LSTM hybrid neural network, which includes an input layer, a CNN feature extraction layer, an LSTM temporal modeling layer, a fully connected layer, and an output layer. The number of neurons in the input layer is consistent with the dimension of the standardized data. The CNN feature extraction layer includes 2-4 convolutional blocks, the LSTM temporal modeling layer includes 1-3 LSTM units, the fully connected layer includes 1-2 hidden layers, and the output layer has 3 neurons. S43: Using the mean squared error (MSE) as the loss function, the Adam optimizer is used to train the CNN-LSTM hybrid neural network. Dropout regularization is used during training to prevent overfitting. The loss function formula is: Where M is the number of training samples. For the j-th sample, Let be the predicted value for the j-th sample; S44: Adjust the hyperparameters of the neural network using the validation set. If the loss function value on the validation set does not decrease for 5 consecutive epochs, halve the learning rate. S45: Use the test set to evaluate the performance of the trained neural network. When the mean absolute error (MAE) is ≤0.5℃ and MAE is ≤5%, the neural network is determined to be a large-scale energy storage thermal management model; otherwise, return to step S43 to retrain. S46: Every 24 hours, add the newly collected standardized data to the training set and repeat steps S43-S45 to achieve dynamic updates of the large-scale energy storage thermal management model.

6. The AI ​​intelligent control thermal management method according to claim 1, characterized in that, In step S5, the preset time window is 5-30 minutes, and the potential thermal risk is determined by the thermal risk assessment index R. The formula for calculating the thermal risk assessment index R is: in, To predict temperature, For the optimal operating temperature of the battery, This is the upper limit of the battery's safe temperature. To predict heat load, For rated heat load, For the maximum allowable heat load, This is the fire early warning status coefficient. coefficient, The values ​​range from 0.2 to 0.5; when R ≥ 0.7, it is considered high risk; when 0.3 ≤ R < 0.7, it is considered medium risk; when R < 0.3, it is considered low risk.

7. The AI ​​intelligent control thermal management method according to claim 1, characterized in that, In step S6, the optimization algorithm is a genetic algorithm, and the temperature control strategy formulation steps include: S61: Define the optimization variable vector X=[P,Q,V], where P is the operating power of the liquid chiller unit, Q is the coolant flow rate, and V is the cooling fan speed; S62: Define the objective function for optimization: in Cooling capacity of liquid chiller Total power consumption of the chiller unit and cooling fans; S63: Set constraints: in Battery temperature Minimum allowable operating temperature of the pool Maximum permissible operating temperature of the pool These are the minimum and maximum operating power of the liquid-cooled unit, respectively. These are the minimum and maximum flow rates of the coolant. These are the minimum and maximum speeds of the cooling fan, respectively. S64: Use a genetic algorithm to solve the above optimization problem and obtain the optimal variable vector X*=[P*,Q*,V*]. Based on the optimal variable vector, formulate the optimal temperature control strategy.

8. The AI ​​intelligent control thermal management method according to claim 1, characterized in that, In step S7, the feedback data includes the battery temperature after temperature control, the inlet and outlet temperatures of the coolant, the operating power of the liquid cooler unit, and the speed of the cooling fan; the closed-loop control uses a PID adjustment algorithm to correct the temperature control strategy, and the PID adjustment formula is: in, For control quantity correction value, This is the proportionality coefficient. Integral coefficient, The differential coefficients are... Let t be the temperature deviation at time t.

9. An AI-powered intelligent control thermal management system, characterized in that, include: The data acquisition module is used to collect environmental data, battery data, cabinet data and fire protection data of the energy storage system in real time, and to build a multi-dimensional raw data set; The data preprocessing module is used to perform noise reduction, normalization, and data completion on the original dataset to obtain a standardized dataset. The digital twin modeling module is used to construct a digital twin model of the energy storage system based on the standardized data set and by integrating the physical mechanisms of heat conduction, convection heat transfer and battery heat generation. The topology self-learning module is used to continuously train the standardized dataset and the output data of the digital twin model using deep learning algorithms to generate a large-scale energy storage thermal management model. The temperature control demand prediction module is used to use the large energy storage thermal management model to input real-time standardized data and predict the target temperature range, heat load change trend and potential thermal risks of the energy storage system within a future preset time window. The temperature control strategy formulation module is used to formulate the optimal temperature control strategy with the goal of maximizing the energy efficiency ratio, combined with the target temperature range, heat load change trend and potential thermal risks, through an optimization algorithm. The temperature control execution module is used to send control commands to the liquid chiller unit according to the optimal temperature control strategy, and drive the liquid chiller unit to perform temperature control operations. The feedback adjustment module is used to collect feedback data after temperature control in real time, input the feedback data into the large energy storage thermal management model, dynamically adjust the model parameters and the optimal temperature control strategy, and realize closed-loop control. The data acquisition module, data preprocessing module, digital twin modeling module, topology self-learning module, temperature control demand prediction module, temperature control strategy formulation module, temperature control execution module, and feedback adjustment module are all integrated into the thermal management controller. The thermal management controller establishes data interaction with the liquid chiller, BMS, fire alarm system, and sensor array through a communication bus.

10. The AI ​​intelligent control thermal management system according to claim 9, characterized in that, The data acquisition module includes an environmental sensor array, a battery data acquisition unit, a cabinet data acquisition unit, and a fire protection data acquisition unit. The environmental sensor array includes a temperature sensor, a humidity sensor, a dust sensor, a wind speed sensor, and a solar radiation intensity sensor. The battery data acquisition unit is connected to the BMS to acquire data such as the SOC value of the battery pack, the voltage of individual cells, and the temperature of individual cells in real time. The cabinet data acquisition unit includes a temperature sensor, a pressure sensor, and a flow sensor inside the cabinet. The fire data acquisition unit is connected to the fire alarm system to acquire data such as fire early warning status and fire extinguishing medium pressure in real time. The communication bus uses CAN485 communication or Ethernet bus.

Images