Lithium ion battery immersed liquid cooling regulation and control method and device, terminal equipment and storage medium

Abstract

The invention discloses a lithium ion battery immersed liquid cooling regulation and control method and device, terminal equipment and a storage medium, and relates to the field of lithium ion batteries, and the method comprises the steps: obtaining the current temperature data, historical temperature data and cooling liquid fluid parameters of a distributed array in a lithium ion battery, and determining the current temperature field distribution and predicted temperature field distribution; constructing a state space, and inputting the state space into the reference decision model, so that the model outputs corresponding actions according to the current state space by taking the minimum temperature difference of the battery module and the minimum cooling energy consumption as awards; according to the corresponding action, the micro electromagnetic valve and the cooling liquid pump are subjected to benchmark regulation and control; acquiring temperature data after reference regulation and control, and determining temperature deviation in combination with a preset temperature requirement; and regulating and controlling according to the temperature deviation. By implementing the method and the device, the stability of global temperature control is improved, so that the problem of temperature interference on other areas due to cooling liquid flow coupling caused by directly adjusting the opening degree of the valve according to local temperature difference in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of lithium-ion batteries, and in particular to a lithium-ion battery immersion liquid cooling control method, device, terminal equipment and storage medium. Background Art

[0002] Lithium-ion batteries, due to their high energy density and long cycle life, are widely used in electric vehicles, energy storage systems, and other fields. However, with the advancement of fast-charging technology for electric vehicles and the increasing demand for high-power energy storage systems, battery module thermal management faces significant challenges. Traditional lithium-ion battery cooling technologies primarily include air cooling, liquid cold plates, and immersion cooling systems. Immersion cooling, due to its direct contact with the battery surface, offers greater heat dissipation efficiency and temperature uniformity.

[0003] Existing immersion liquid cooling systems primarily utilize distributed temperature sensor arrays. Based on the deviation between the real-time temperature at each monitoring point and the preset temperature, these sensors independently adjust the cooling components in the corresponding area to achieve precise local temperature regulation. However, due to the fluid dynamics of the coolant piping system, changes in valve opening at a single node can trigger a chain reaction of coolant flow rate and pressure through pipeline flow coupling, causing the actual cooling efficiency at other monitoring points to deviate from the expected value. Summary of the Invention

[0004] Embodiments of the present invention provide a lithium-ion battery immersion liquid cooling control method, apparatus, terminal device, and storage medium, which can improve the stability of global temperature control, thereby solving the problem of temperature interference caused by the coolant flow coupling to other areas caused by directly adjusting the valve opening based on the local temperature difference in the prior art.

[0005] An embodiment of the present invention provides a method for controlling immersion liquid cooling of a lithium-ion battery, comprising:

[0006] Obtain current temperature data, historical temperature data, and current coolant fluid parameters of a distributed array of lithium-ion batteries;

[0007] determining a current temperature field distribution in the lithium-ion battery based on the current temperature data, and determining a predicted temperature field distribution in the lithium-ion battery based on the historical temperature data and current coolant fluid parameters;

[0008] The current temperature field distribution, predicted temperature field distribution, current coolant fluid parameters, and current pressure change are used as the current lithium-ion battery state space. This state space is then input into a preset benchmark decision model, so that the benchmark decision model outputs a corresponding action space based on the current lithium-ion battery state space, with the minimum battery module temperature difference and the lowest cooling energy consumption as the reward function. The action space includes the micro-solenoid valve opening adjustment and the coolant pump speed adjustment. The initial current pressure change is 0.

[0009] Perform baseline control on the micro solenoid valves and coolant pumps in the entire array of lithium-ion batteries according to the corresponding action space, and obtain temperature data of the distributed array of lithium-ion batteries after baseline control;

[0010] Determine the temperature deviation of each monitoring area in the array based on the temperature data after baseline control and the preset temperature requirements;

[0011] For each monitoring area in the array, the micro solenoid valve and coolant pump are regulated according to the temperature deviation.

[0012] Furthermore, after obtaining the current temperature data, historical temperature data, and current coolant fluid parameters of the distributed array of lithium-ion batteries, the following steps are also included:

[0013] The current temperature data and the historical temperature data of the distributed array in the lithium-ion battery are subjected to outlier elimination, missing value filling and normalization processing to obtain the preprocessed current temperature data and the preprocessed historical temperature data.

[0014] Furthermore, the current temperature field distribution in the lithium-ion battery is determined based on the current temperature data, including:

[0015] Input the current temperature data into a preset resolution reconstruction model, so that the resolution reconstruction model extracts temperature features from the current temperature data through a built-in encoder to obtain a low-dimensional feature vector;

[0016] The low-dimensional feature vector is reconstructed through the built-in decoder to generate the current temperature field distribution in the lithium-ion battery.

[0017] Furthermore, based on the historical temperature data and the current coolant fluid parameters, a predicted temperature field distribution in the lithium-ion battery is determined, including:

[0018] Inputting the historical temperature data and the current coolant fluid parameters into a preset temperature prediction model, so that the temperature prediction model performs feature splicing on the historical temperature data and the current coolant fluid parameters to obtain a fused feature vector;

[0019] Through the built-in forget gate, input gate and output gate, the fused feature vector is screened for time-dependent features to obtain the hidden feature vector;

[0020] Linear mapping is performed on the latent feature vector to generate the predicted temperature field distribution in the lithium-ion battery.

[0021] Furthermore, the micro solenoid valve and the coolant pump are regulated according to the temperature deviation, including:

[0022] Perform proportional differential calculation on the temperature deviation to obtain the adjustment amount of the micro solenoid valve opening;

[0023] According to the adjustment amount of the micro solenoid valve opening, the micro solenoid valve is regulated;

[0024] Determine the adjustment coefficient of the coolant pump speed according to the temperature deviation and the preset temperature difference threshold;

[0025] Determine the adjustment amount of the coolant pump speed according to the adjustment coefficient of the coolant pump speed;

[0026] The coolant pump is regulated according to the adjustment amount of the coolant pump speed.

[0027] Furthermore, the preset baseline decision model is trained in the following way:

[0028] For the baseline decision model to be trained, define the model's state space, action space, and reward function. The state space includes the current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters, and the current pressure change. The action space includes the micro-solenoid valve opening adjustment and the coolant pump speed adjustment. The reward function includes the minimum battery module temperature difference and the lowest cooling energy consumption.

[0029] Construct a policy network based on the state space, action space, and current policy network parameters;

[0030] Initialize the state space and action space to obtain the current state space and current action space;

[0031] The current state space and the current action space are input into the benchmark decision model to be trained, so that the benchmark decision model to be trained determines the updated state space according to the current state space and the current action space through the built-in digital twin simulation module; the reward function value is calculated according to the updated state space; the policy network is updated through the deep Q network algorithm to obtain the updated policy network parameters; the policy optimization loss function is calculated according to the reward function value, the current policy network parameters and the updated policy network parameters, until the policy optimization loss function converges to obtain the preset benchmark decision model.

[0032] Furthermore, the digital twin simulation module includes: battery module three-dimensional heat conduction submodule, coolant flow submodule, battery heat flow coupling boundary submodule, battery heat generation rate calculation submodule and energy conservation submodule;

[0033] The digital twin simulation module is determined by:

[0034] Obtain the battery parameters, coolant parameters, and boundary condition parameters of the lithium-ion battery; battery parameters include: battery density, battery specific heat capacity, battery thermal conductivity, and battery heat generation rate; coolant parameters include: coolant density, coolant dynamic viscosity, coolant specific heat capacity, and coolant thermal conductivity; boundary condition parameters include: convection heat transfer coefficient and external force per unit volume;

[0035] According to the battery parameters and coolant parameters of the lithium-ion battery, the battery module three-dimensional heat conduction submodule, coolant flow submodule, battery heat flow coupling boundary submodule, battery heat generation rate calculation submodule and energy conservation submodule are constructed.

[0036] Based on the above method embodiment, the present invention provides a corresponding device embodiment, including: a battery data acquisition module, a temperature data processing module, an adjustment amount decision module, a reference control module, a temperature difference calculation module, and a fine control module;

[0037] A battery data acquisition module is used to obtain current temperature data, historical temperature data, and current coolant fluid parameters of the distributed array of lithium-ion batteries;

[0038] a temperature data processing module for determining a current temperature field distribution in the lithium-ion battery based on current temperature data, and determining a predicted temperature field distribution in the lithium-ion battery based on historical temperature data and current coolant fluid parameters;

[0039] The adjustment decision module is used to use the current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters, and the current pressure change as the current lithium-ion battery state space, and input the state space into a preset benchmark decision model, so that the benchmark decision model outputs a corresponding action space based on the current lithium-ion battery state space, with the minimum battery module temperature difference and the lowest cooling energy consumption as the reward function; the action space includes: the micro-solenoid valve opening adjustment amount and the coolant pump speed adjustment amount; wherein the current pressure change amount is initially 0;

[0040] A baseline control module is used to perform baseline control on the micro solenoid valves and coolant pumps of the entire array of lithium-ion batteries according to the corresponding action space, and obtain temperature data of the distributed array of lithium-ion batteries after baseline control;

[0041] The temperature difference calculation module is used to determine the temperature deviation of each monitoring area in the array based on the temperature data after baseline control and the preset temperature requirements;

[0042] The fine control module is used to control the micro solenoid valve and coolant pump according to the temperature deviation in each monitoring area in the array.

[0043] Based on the above-mentioned method embodiment, the present invention provides a corresponding terminal device embodiment, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps of the lithium-ion battery immersion liquid cooling control method as described in the present invention are implemented.

[0044] Based on the above-mentioned method embodiment, the present invention provides a corresponding computer-readable storage medium embodiment, including: a stored computer program, which controls the device where the computer-readable storage medium is located to execute the steps of the lithium-ion battery immersion liquid cooling control method as described in the present invention when the computer program is running.

[0045] Compared with the prior art, the beneficial effects of the embodiment of this solution are:

[0046] The present invention obtains current temperature data, historical temperature data, and current coolant fluid parameters for a distributed array of lithium-ion batteries. Based on the current temperature data, the current temperature field distribution in the lithium-ion battery is determined, and based on the historical temperature data and current coolant fluid parameters, a predicted temperature field distribution in the lithium-ion battery is determined. Next, the current temperature field distribution, predicted temperature field distribution, current coolant fluid parameters, and current pressure change are used as the state space of the current lithium-ion battery. This state space is input into a preset benchmark decision model, so that the benchmark decision model outputs a corresponding action space based on the current lithium-ion battery state space, using the reward function of minimizing the battery module temperature difference and minimizing cooling energy consumption as the reward function, thereby reducing energy consumption while meeting heat dissipation requirements. Then, based on the corresponding action space, benchmark control is performed on the micro-solenoid valves and coolant pumps in the entire array of lithium-ion batteries, and temperature data of the distributed array of lithium-ion batteries after benchmark control is obtained. Because the benchmark control has achieved a global preliminary optimization of the coolant flow rate, the temperature of each battery region has approached a reasonable range, and the monitored temperature deviation is minimized. Finally, based on the temperature field after benchmark control, only areas where the local temperature difference exceeds the preset temperature requirements need to be fine-tuned. Compared with the large adjustment of valve openings required by direct temperature difference control, this can effectively reduce the temperature interference of coolant flow coupling on other areas, ensuring the stability of the global temperature field. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 1 is a flow chart of a lithium-ion battery immersion liquid cooling control method provided by one embodiment of the present invention;

[0048] Figure 2 It is a structural schematic diagram of a lithium-ion battery immersion liquid cooling control device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features.

[0051] like Figure 1 As shown, in order to solve the problem in the prior art of directly adjusting the valve opening according to the local temperature difference, which causes the coolant flow coupling to cause temperature interference in other areas, an embodiment of the present invention provides a lithium-ion battery immersion liquid cooling control method, which includes at least the following steps:

[0052] Step S1: obtaining current temperature data, historical temperature data, and current coolant fluid parameters of the distributed array of lithium-ion batteries;

[0053] In step S1, a distributed temperature sensor array is embedded in the lithium-ion battery. Current temperature data from the distributed temperature sensor array is collected at a sampling frequency of no less than 10 Hz, with an analog-to-digital conversion accuracy of no less than 12 bits. Real-time temperature data transmission is achieved using a CAN bus or SPI communication method. In this embodiment, a high-precision temperature sensor with a measurement accuracy of no less than 0.1°C, a response time of less than 0.5 seconds, and an operating temperature range of -20°C to 100°C is selected based on the temperature gradient distribution.

[0054] It should be noted that the cooling area is divided according to the battery module structure, and the distributed temperature sensor array is embedded in the lithium-ion battery to divide it into 8 to 16 independent monitoring areas. The temperature sensor monitoring points in each monitoring area are determined by the following formula:

[0055]

[0056] Among them, N sensor Indicates the total number of sensors; N min Indicates the minimum number of sensors, no less than 12; V module Indicates the volume of the battery module in cm 3 ; V unit Indicates the coverage volume of a single sensor. In this embodiment, the value is 250cm 3 ; Represents the ceiling function.

[0057] The historical temperature data of the lithium-ion battery is obtained from a preset database, for example, the real-time sampled temperature data within the previous three days at the current time node. It should be noted that the sensor of the historical temperature data is the same as the sensor of the current temperature data, that is, the sampling frequency is 10 Hz.

[0058] The current coolant fluid parameters of the lithium-ion battery are synchronously obtained, where the current coolant fluid parameters include flow rate values ​​at several key positions, wherein the key positions are set at flow channel positions corresponding to temperature-sensitive areas in the battery module.

[0059] In a preferred embodiment, after obtaining the current temperature data, historical temperature data, and current coolant fluid parameters of the distributed array of lithium-ion batteries, the method further includes:

[0060] The current temperature data and the historical temperature data of the distributed array in the lithium-ion battery are subjected to outlier elimination, missing value filling and normalization processing to obtain the preprocessed current temperature data and the preprocessed historical temperature data.

[0061] In one embodiment of the present invention, after obtaining the current temperature data, historical temperature data, and current coolant fluid parameters of a distributed lithium-ion battery array, the temperature data needs to be preprocessed to ensure the accuracy of subsequent analysis. Specifically, first, outliers in the temperature data are eliminated using a physical threshold to prevent sudden jumps or distorted data from interfering with thermal state judgment. Single-point anomalies are corrected using neighboring value interpolation, while continuous anomalies are processed using a sliding average filter. Next, linear interpolation is used to fill in missing values ​​caused by communication failures or sensor anomalies. Finally, through minimum-maximum normalization or -Z-Score-standardization, the temperature data is unified to the [0,1] interval or standardized distribution to eliminate the influence of dimensional and numerical fluctuations.

[0062] Step S2: determining a current temperature field distribution in the lithium-ion battery based on the current temperature data, and determining a predicted temperature field distribution in the lithium-ion battery based on the historical temperature data and the current coolant fluid parameters;

[0063] In a preferred embodiment, determining the current temperature field distribution in the lithium-ion battery based on the current temperature data includes:

[0064] Input the current temperature data into a preset resolution reconstruction model, so that the resolution reconstruction model extracts temperature features from the current temperature data through a built-in encoder to obtain a low-dimensional feature vector;

[0065] The low-dimensional feature vector is reconstructed through the built-in decoder to generate the current temperature field distribution in the lithium-ion battery.

[0066] In a preferred embodiment, determining the predicted temperature field distribution in the lithium-ion battery based on historical temperature data and current coolant fluid parameters includes:

[0067] Inputting the historical temperature data and the current coolant fluid parameters into a preset temperature prediction model, so that the temperature prediction model performs feature splicing on the historical temperature data and the current coolant fluid parameters to obtain a fused feature vector;

[0068] Through the built-in forget gate, input gate and output gate, the fused feature vector is screened for time-dependent features to obtain the hidden feature vector;

[0069] Linear mapping is performed on the latent feature vector to generate the predicted temperature field distribution in the lithium-ion battery.

[0070] In step S2, the current temperature data is input into the preset resolution reconstruction model. The resolution reconstruction model adopts a convolutional neural network structure, which consists of two parts: encoder and decoder. The encoder consists of three convolutional layers, and each layer is followed by a ReLU activation function and a batch normalization layer. The convolution operation of the encoder is expressed as:

[0071] z l =σ(W l *z l-1 +b l );

[0072] Among them, z l represents the output of the feature map of the first layer, σ represents the ReLU activation function, which is defined as σ(x) = max(0, x), W l Represents the weight of the convolution kernel of the lth layer, * represents the convolution operator, z l-1 represents the l-1th layer feature map, b l Represents the bias term of layer l. The first convolutional layer uses 32 3×3 convolution kernels with a stride of 2; the second convolutional layer uses 64 3×3 convolution kernels with a stride of 2; and the third convolutional layer uses 128 3×3 convolution kernels with a stride of 2.

[0073] The encoder first extracts features from the discrete data collected by distributed temperature sensors (such as the point temperature of 12 sensors), uses a convolutional neural network (CNN) or self-attention mechanism to capture the spatial correlation and gradient characteristics of the temperature data, and compresses it into a low-dimensional feature vector (such as 128 dimensions), which contains key information about the temperature distribution of the battery module.

[0074] Subsequently, the decoder reconstructs features based on the low-dimensional feature vector through a deconvolution layer or interpolation algorithm, maps the abstract features back to three-dimensional space, and generates a continuous temperature field distribution containing thousands of grid points, achieving high-precision reconstruction from discrete point data to continuous field distribution.

[0075] The decoder consists of three deconvolution layers, and each layer is followed by a ReLU activation function. The deconvolution operation of the decoder is expressed as:

[0076] z l =σ(W l * T z l-1 +b l );

[0077] in,* T Represents the transposed convolution operator. The first deconvolution layer uses 64 4×4 convolution kernels with a stride of 2; the second deconvolution layer uses 32 4×4 convolution kernels with a stride of 2; and the third deconvolution layer uses 1 4×4 convolution kernel with a stride of 2.

[0078] In this embodiment, the input of the resolution reconstruction model is the discrete temperature data collected by the distributed temperature sensor array, that is, the current temperature data, with a dimension of N f ×1, N f is the number of sensors, and the output of the resolution reconstruction model is the reconstructed high-resolution temperature field distribution, that is, the current temperature field distribution, with a dimension of H×W T ×1, H and W T are the height and width of the temperature field, respectively. The total number of network parameters is 235,892, including convolution kernel weights and bias terms.

[0079] It should be noted that the resolution reconstruction model is trained in the following way:

[0080] Obtain a first training data set; each training data in the first training data set includes an actual continuous temperature field distribution and its corresponding discrete temperature data; wherein the number of training data in the first training data set is not less than 10 4 indivual;

[0081] The first training data set is input into the resolution reconstruction model to be trained, so that the resolution reconstruction model to be trained takes the discrete temperature data as input and the predicted continuous temperature field distribution corresponding to the discrete temperature data as output for iterative training. During the training process, the first loss function is calculated according to the predicted continuous temperature field distribution and the actual continuous temperature field distribution by the following formula to quantify the difference between the model prediction and the actual temperature field:

[0082]

[0083] Among them, L(θ) represents the first loss function, θ represents the network parameter set, N τ represents the number of first training data, x i represents the discrete temperature data of the first i-th training data, f θ (xi ) represents the predicted continuous temperature field distribution corresponding to discrete temperature data, y i represents the actual continuous temperature field distribution of the i-th first training data, represents the L2 norm, λ is the regularization coefficient, and its value is 10 -4 .

[0084] Adopt Adam optimizer to update the network parameters:

[0085]

[0086] Among them, θ t represents the parameter of the tth iteration, α represents the learning rate, and its initial value is 10 -3 , represents the first moment estimate of the first loss function with respect to the network parameters, Represents the second-order moment estimate of the first loss function with respect to the network parameters, ∈ represents the smoothing term, and its value is 10 -8 .

[0087] The adaptive characteristics of the Adam algorithm are used to dynamically adjust the network parameters so that the temperature field predicted by the resolution reconstruction model continuously approaches the real distribution, ultimately achieving the convergence of the first loss function and the optimization of the model performance to obtain the preset resolution reconstruction model.

[0088] After reconstructing the current temperature field distribution, the temperature prediction model is used to generate the future temperature field distribution. Specifically, the historical temperature data and the current coolant fluid parameters are input into the preset temperature prediction model. The temperature prediction model is constructed based on a long short-term memory network, consisting of an input layer, two LSTM layers, and an output layer.

[0089] The number of input layer nodes is equal to the number of temperature sensors, which are used to receive historical temperature series data and current coolant fluid parameters, and perform feature splicing to obtain a fused feature vector, which is expressed as:

[0090] X={x t-n+1 , x t-n+2 ,...,x t , v1, v2, ..., v K};

[0091] Among them, X represents the fusion feature vector, x t represents the temperature vector at time t, v1, v2, ..., v K represents the flow velocity value at the Kth key position, n represents the time window length, and its value is 60.

[0092] The first LSTM layer contains 128 hidden units, each of which consists of four components: a forget gate, an input gate, an output gate, and a memory unit. The second LSTM layer also contains 128 hidden units to process the output features of the first layer. The forget gate, input gate, memory unit, and output gate are represented as follows:

[0093] f t =σ g (W f ·[h t-1 , x t ]+b f );

[0094] i t =σ g (W i ·[h t-1 , x t ]+b i );

[0095]

[0096] o t =σ g (W o ·[h t-1 , x t ]+b o );

[0097] h t =o t ⊙tanh(C t );

[0098] Among them, f t Represents the output of the forget gate, i t represents the input gate output, Represents the candidate memory cell state, C t Indicates the state of the memory unit, o t Represents the output of the output gate, h t represents the hidden feature vector, σ g Represents the sigmoid activation function, defined as tanh(·) represents the hyperbolic tangent activation function, W f 、W i 、W C 、W o represents the weight matrix, b f 、b i 、b C 、b o represents the bias vector and ⊙ represents element-wise multiplication.

[0099] The number of nodes in the output layer is equal to the number of grid points after temperature field reconstruction. The LSTM features are mapped to the predicted temperature field through full connection to generate the predicted temperature field distribution:

[0100]

[0101] in, represents the predicted temperature field distribution after k steps, W y represents the output weight matrix, b y Represents the output bias vector.

[0102] The present invention uses an LSTM network and coolant fluid parameters to learn the changing patterns of temperature time series and predict the temperature distribution of battery modules at future moments. The prediction accuracy can achieve a temperature error of less than 1°C, providing a basis for early intervention control.

[0103] It should be noted that the temperature prediction model is trained in the following way:

[0104] Obtain a second training data set; each training data in the second training data set includes historical temperature training data, corresponding historical coolant fluid parameters, and a true value of the temperature field at a future moment corresponding to the historical temperature training data;

[0105] The training data set is input into the temperature prediction model to be trained, so that the temperature prediction model to be trained takes the historical temperature training data and the corresponding historical coolant fluid parameters as input and the temperature field prediction value at the future moment as output for iterative training. During the training process, the second loss function is calculated according to the temperature field prediction value at the future moment and the true value of the temperature field at the future moment by the following formula:

[0106]

[0107] Among them, L MSE Represents the second loss function, namely the mean square error loss, M represents the number of output grid points, represents the predicted value of the temperature field at the future moment of the i-th grid point, Represents the true value of the temperature field at the i-th grid point at the future moment.

[0108] The Adam optimizer is used to update the model parameters such as the LSTM layer weights, bias terms, and fully connected layer parameters of the temperature prediction model based on the second loss function. By continuously iteratively adjusting the parameters, the second loss function is gradually reduced and stabilized, resulting in a preset temperature prediction model that can accurately predict the temperature field distribution of lithium-ion batteries.

[0109] Regarding temperature monitoring, this invention employs a distributed temperature sensor array layout strategy, optimizing the sensor layout based on the thermal characteristics of the battery module to ensure effective coverage of hotspots and areas with significant temperature gradients. A convolutional neural network temperature field reconstruction algorithm is designed for the discrete temperature data provided by the sensors. This algorithm uses an encoder to extract spatial temperature features and a decoder to restore the temperature field resolution, effectively solving the challenge of mapping data from limited measurement points into a continuous temperature field. Furthermore, the long-short-term memory network structure captures the temporal characteristics of temperature changes, predicting future temperature trends and providing a forward-looking basis for control decisions.

[0110] Step S3: The current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters, and the current pressure change are used as the state space of the current lithium-ion battery, and the state space is input into a preset benchmark decision model, so that the benchmark decision model outputs a corresponding action space based on the current lithium-ion battery state space, with the minimum battery module temperature difference and the lowest cooling energy consumption as the reward function; the action space includes: the micro-solenoid valve opening adjustment amount and the coolant pump speed adjustment amount; wherein the current pressure change is initially 0;

[0111] For step S3, the current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters, and the current pressure change are used as the state space of the current lithium-ion battery, and its state vector is expressed as:

[0112] s t =[T1, T2, ..., T M , T t1 , T t2 ,...,T tM , v1, v2, ..., v K ,Δp];

[0113] Among them, s t represents the state vector at time t, T1, T2, ..., T M Represents the current temperature value of M grid points, T t1 , T t2 ,...,T tM Represents the predicted temperature values ​​of M grid points at the next moment (t+1 moment) predicted by the long short-term memory network, v1, v2, ..., v K It should be noted that the pressure change at the initial moment (t=0) is set to 0 because the lithium-ion battery cooling system has not yet started fluid circulation or generated flow resistance.

[0114] The state space is input into the preset benchmark decision model. The policy network in the benchmark decision model is based on the current state space s. t, using the Deep Q Network (DQN) algorithm to select the action a that maximizes the Q value t , the action space includes the micro solenoid valve opening adjustment and the coolant pump speed adjustment:

[0115]

[0116] Among them, a t represents the action vector at time t, Indicates N w The opening adjustment of a micro solenoid valve, Δω pump Indicates the coolant pump speed adjustment. It should be noted that the range of the micro solenoid valve opening adjustment is -10% to +10%, and the range of the coolant pump speed adjustment is -5% to +5%.

[0117] It should be noted that the deep Q network of the present invention includes a state input layer, 4 fully connected hidden layers and an action output layer, wherein the dimension of the state space input layer is equal to the high-resolution temperature field distribution plus the coolant fluid parameter dimension, which is generally 10 3 ~10 4 The first hidden layer contains 512 nodes and uses the ReLU activation function. The second hidden layer contains 256 nodes and uses the ReLU activation function. The third hidden layer contains 128 nodes and uses the ReLU activation function. The fourth hidden layer has a number of nodes equal to the action space dimension, including the micro-solenoid valve opening adjustment and the coolant pump speed adjustment, which are used to generate the Q value estimate for each action. The total number of network parameters is determined according to the specific state space and action space dimensions, and is generally around 10 5 ~10 6 Magnitude.

[0118] Then, the built-in digital twin simulation module is used to simulate the current action a t , the finite element method is used to solve the dynamic changes after the action is executed, and the new state space s is obtained t+1 , and based on the new state space s t+1 , the reward is calculated using the following formula:

[0119] r t =-α·(T max -T min )-β·P sys +γ ξ ·1(T max -T min <T threshold );

[0120] Among them, r t represents the reward obtained at time t, α represents the temperature difference weight coefficient, which is 0.7, T max Indicates the maximum temperature of the battery module, Tmin Indicates the lowest temperature of the battery module, β indicates the energy consumption weight coefficient, which is 0.3, P sys represents the system energy consumption, γ ξ represents the additional reward coefficient, 1(·) represents the indicator function, which is 1 when the condition is met and 0 otherwise, T threshold Indicates the temperature difference threshold, which is set to 3°C.

[0121] It should be noted that the reward function consists of two parts: temperature difference control term and energy consumption control term. The temperature difference control term is calculated by calculating the maximum temperature T of the battery module. max With the minimum temperature T min The temperature uniformity is evaluated if the temperature difference is less than the temperature difference threshold T threshold In addition to the temperature difference control term itself, a positive additional reward is given through the indicator function 1(·) if the temperature difference is equal to or exceeds the temperature difference threshold T threshold , no additional reward is given, guiding the decision model to converge toward the optimal state with a small temperature difference. The energy consumption control item is used to constrain the cooling system's power consumption. The lower the energy consumption, the higher the reward. This avoids excessively increasing pump speed and energy consumption simply to reduce the temperature difference, thus achieving a balance between thermal management efficiency and energy loss.

[0122] The Q value is updated through the DQN algorithm, and the above process is repeated. The policy network in the baseline decision model continuously optimizes the Q value until the Q value converges. At this point, the baseline decision model can choose the action that maximizes the cumulative reward in any state, that is, achieve the minimum temperature difference and the lowest energy consumption.

[0123] In a preferred embodiment, the preset benchmark decision model is trained in the following manner:

[0124] For the baseline decision model to be trained, define the model's state space, action space, and reward function. The state space includes the current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters, and the current pressure change. The action space includes the micro-solenoid valve opening adjustment and the coolant pump speed adjustment. The reward function includes the minimum battery module temperature difference and the lowest cooling energy consumption.

[0125] Construct a policy network based on the state space, action space, and current policy network parameters;

[0126] Initialize the state space and action space to obtain the current state space and current action space;

[0127] The current state space and the current action space are input into the benchmark decision model to be trained, so that the benchmark decision model to be trained determines the updated state space according to the current state space and the current action space through the built-in digital twin simulation module; the reward function value is calculated according to the updated state space; the policy network is updated through the deep Q network algorithm to obtain the updated policy network parameters; the policy optimization loss function is calculated according to the reward function value, the current policy network parameters and the updated policy network parameters, until the policy optimization loss function converges to obtain the preset benchmark decision model.

[0128] In one embodiment of the present invention, the training of the benchmark decision model adopts a reinforcement learning framework. First, the action space and reward function are defined. The state space includes the current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters and the current pressure change, which are used to characterize the battery thermal state and the cooling system operating parameters; the action space is set to the micro-solenoid valve opening adjustment amount and the coolant pump speed adjustment amount as executable control instructions; the reward function includes the minimum temperature difference of the battery module and the lowest cooling energy consumption, and the feedback signal is generated by quantifying the temperature difference and energy consumption indicators.

[0129] Next, a policy network (i.e., deep Q network) is constructed based on the state space, action space dimensions, and current policy network parameters. This network is responsible for the decision-making process of mapping states to actions, initializing the state space and action space, and obtaining the current state and action at the start of training.

[0130] The initialized state and action are input into the training model. The built-in digital twin simulation module simulates the dynamic changes of the system after the action is executed based on physical laws to determine the updated state space. The reward function value is then calculated based on the new state to provide a basis for policy optimization. With the help of the deep Q network algorithm, the reward value and state transition information are used to update the policy network parameters. The Q value function is approximately expressed as:

[0131] Q(s, a; θ)≈Q * (s, a);

[0132] Among them, Q(s, a; θ) represents the parameterized Q value function, θ represents the network parameters, and Q * (s, a) represents the optimal Q-value function.

[0133] Calculate the policy optimization loss function based on the reward function value and the policy network parameters before and after the update:

[0134]

[0135] Among them, L(θ) represents the policy optimization loss function, γ represents the discount factor, which is 0.95, and θ - represents the target network parameters, It represents the mathematical expectation, that is, the average level of the mean square error between the predicted Q value and the target Q value in all possible state transition samples. By minimizing this expectation, the Q value estimation of the policy network gradually approaches the true cumulative reward expectation.

[0136] Finally, through the experience replay mechanism, Sampling batch data, playback buffer size is 10 5 , each training randomly samples a batch of 128 to reduce sample correlation. Then, a target network update mechanism is designed to update the target network parameters every 100 steps to improve training stability. Finally, an exploration-exploitation balance strategy is designed, using the ε-greedy method with an initial exploration rate of 0.8, which decays exponentially to 0.1, balancing the exploration of new actions with the exploitation of known optimal actions:

[0137]

[0138] Among them, η represents the learning rate, which is 10 -4 Through multiple rounds of iterative training until the loss function converges, a benchmark decision-making model with precise decision-making capabilities is ultimately obtained. This model can output optimal control actions based on real-time status, achieving the collaborative optimization goals of minimizing battery module temperature difference and cooling energy consumption.

[0139] In a preferred embodiment, the digital twin simulation module includes: a battery module three-dimensional heat conduction submodule, a coolant flow submodule, a battery heat flow coupling boundary submodule, a battery heat generation rate calculation submodule, and an energy conservation submodule;

[0140] The digital twin simulation module is determined by:

[0141] Obtain the battery parameters, coolant parameters, and boundary condition parameters of the lithium-ion battery; battery parameters include: battery density, battery specific heat capacity, battery thermal conductivity, and battery heat generation rate; coolant parameters include: coolant density, coolant dynamic viscosity, coolant specific heat capacity, and coolant thermal conductivity; boundary condition parameters include: convection heat transfer coefficient and external force per unit volume;

[0142] According to the battery parameters and coolant parameters of the lithium-ion battery, the battery module three-dimensional heat conduction submodule, coolant flow submodule, battery heat flow coupling boundary submodule, battery heat generation rate calculation submodule and energy conservation submodule are constructed.

[0143] In one embodiment of the present invention, the digital twin simulation module of the present invention refers to building a virtual simulation environment for a battery cooling system, integrating heat conduction and fluid dynamics solvers, simulating battery thermal behavior and coolant flow processes, providing a control strategy verification environment, supporting standard operating conditions and extreme operating conditions testing, and generating simulated temperature field data for model training and parameter optimization.

[0144] First, the basic physical parameters of lithium-ion batteries are obtained through experimental measurement or query, including three types of basic information: battery parameters, coolant parameters, and boundary condition parameters. Among them, battery parameters are used to characterize the thermophysical properties of the battery cell, including battery density ρ b 、Battery specific heat capacity c p , battery thermal conductivity k b and battery heat generation rate q g ; Coolant parameters are used to describe the fluid and heat transfer characteristics of the cooling medium, including coolant density ρ f , coolant dynamic viscosity μ, coolant specific heat capacity c p , f and coolant thermal conductivity k f The boundary condition parameters are used to define the energy exchange and force state between the system and the outside world, including the convection heat transfer coefficient h and the external force per unit volume.

[0145] Next, based on the law of conservation of energy and the principles of fluid dynamics, a three-dimensional heat conduction equation for the battery module is established and used as the three-dimensional heat conduction submodule of the battery module in the digital twin simulation module:

[0146]

[0147] Among them, ρ b Indicates battery density, ranging from 2000 to 2500 kg / m 3 , c p Indicates the specific heat capacity of the battery, with a value range of 800 to 1200 J / (kg·K), T represents the temperature variable in K, and t represents the time variable in s, k b Indicates the thermal conductivity of the battery, ranging from 20 to 60 W / (m·K). represents the gradient operator, q g Indicates the battery heat generation rate, in W / m 3 .

[0148] Based on the Navier-Stokes equations, the coolant flow equation is established and used as the coolant flow submodule in the digital twin simulation module:

[0149]

[0150] Among them, ρ f Indicates coolant density, ranging from 800 to 1200 kg / m 3 , represents the fluid velocity vector in m / s, p represents the pressure variable in Pa, and μ represents the coolant dynamic viscosity in the range of 0.4×10 -3 ~4×10- 3 Pa·s, Represents the external force per unit volume, in N / m 3 .

[0151] Establish the fluid and solid thermal coupling boundary conditions and use them as the battery thermal-fluid coupling boundary submodule in the digital twin simulation module:

[0152]

[0153] in, represents the interface normal vector, h represents the convection heat transfer coefficient, and its value range is 50~500W / (m 2 K), T s Indicates the battery surface temperature in K, T f represents the coolant temperature in K. Then, the finite element method is used to discretize the above differential equations, and the number of grid cells is not less than 10 5 The grid size of the hot spot area is not larger than 1mm.

[0154] Establish a battery heat generation rate calculation submodule, which consists of three parts: ohmic heat, activation heat and concentration heat:

[0155]

[0156] Among them, q ohm Indicates ohmic heat, unit is W, q act represents the activation heat in W, q con Indicates concentration heat, unit is W, I indicates charge and discharge current, unit is A, R int represents the internal resistance of the battery in Ω, η represents the polarization overpotential in V, and U represents the open circuit voltage in V.

[0157] Establish energy conservation constraints and use them as energy conservation submodules in the digital twin simulation module:

[0158]

[0159] Among them, c p，f Indicates the specific heat capacity of the coolant, with a value range of 3000~4200 J / (kg·K), k f Indicates the thermal conductivity of the coolant, ranging from 0.1 to 0.6 W / (m·K), S indicates the surface area of ​​the control body, in m 2 , V represents the volume of the control body, the unit is m 3 .

[0160] Finally, deploy the constructed digital twin simulation module to an edge computing processor. Select a high-performance edge computing processor with a clock speed of at least 1.5 GHz, a memory capacity of at least 4 GB, support for a neural network acceleration unit, and a floating-point computing capability of at least 5 TFLOPS. Then, optimize the digital twin simulation module, including pruning, quantization, and compression, to reduce computational complexity.

[0161] Network pruning is determined by:

[0162] W pruned =W⊙M, where M i , j = 1.|W i , j|>τ / ;

[0163] Among them, W represents the original weight matrix, W pruned represents the weight matrix after pruning, M represents the binary mask, and τ represents the pruning threshold, which is determined according to the preset sparsity rate.

[0164] Network quantization is achieved through the following mapping:

[0165]

[0166] Among them, W q Represents the quantized weight, W min 、W max where represents the minimum and maximum weights, respectively, and b represents the quantization bit width, typically 8 bits. Through these optimizations, the size of the digital twin simulation module was reduced to less than 25% of its original size, and the loss in inference accuracy was kept within 3%.

[0167] The present invention constructs a digital twin simulation module through the principles of thermodynamics and fluid mechanics to describe the interaction between heat generation and conduction inside the battery and the flow of coolant, thereby realizing accurate calculation of the battery temperature field and the coolant flow rate field.

[0168] Step S4: performing baseline control on the micro solenoid valves and coolant pumps of the entire array of lithium-ion batteries according to the corresponding action space, and obtaining temperature data of the distributed array of lithium-ion batteries after baseline control;

[0169] For step S4, the motion vector output from step S3 is For the full array of lithium-ion batteries, the opening adjustment of each micro solenoid valve is Precise adjustment corresponding to N w The opening of a micro solenoid valve and the coolant pump speed adjustment Δω pump, adjusting the coolant pump speed and controlling the overall flow rate of the main circuit. At this point, because the action vector is output by reinforcement learning combined with digital twin simulation, it is a global optimal allocation of cooling resources for the entire array with the goals of minimizing temperature differences and minimizing energy consumption. Compared with extensive regulation directly based on real-time temperature differences, it can effectively avoid global flow coupling interference caused by local adjustments. For example, when a temperature increase is detected at a monitoring point, the system, through reinforcement learning optimization, does not directly increase the valve opening in that area. Instead, it simultaneously fine-tunes the valves in adjacent areas (e.g., +5% in that area, -2% in the adjacent area) to maintain overall flow balance.

[0170] After completing the baseline control, the temperature of the lithium-ion battery is collected through distributed temperature sensors, and the temperature data after baseline control is obtained through signal processing.

[0171] Step S5: determining the temperature deviation of each monitoring area in the array according to the temperature data after the benchmark adjustment and the preset temperature requirement;

[0172] Regarding step S5, although the global optimal allocation in step S4 achieves the goal of minimizing the temperature difference, it is also limited by global optimization, and some monitoring areas may still have small temperature differences. Therefore, based on the temperature data after baseline control and the preset temperature requirements, the temperature deviation of each monitoring area in the array is calculated:

[0173]

[0174] Where, ΔT i represents the temperature deviation of the i-th monitoring area, T i represents the average temperature of the i-th monitoring area, Represents the average temperature of the entire lithium-ion battery.

[0175] The present invention uses the difference between the actual temperature and the preset temperature requirement as the temperature deviation, and dynamically calculates the average value of the temperatures of all regions as the preset temperature requirement.

[0176] Step S6: For each monitoring area in the array, the micro electromagnetic valve and the coolant pump are regulated according to the temperature deviation.

[0177] In a preferred embodiment, regulating the micro solenoid valve and the coolant pump according to the temperature deviation includes:

[0178] Perform proportional differential calculation on the temperature deviation to obtain the adjustment amount of the micro solenoid valve opening;

[0179] According to the adjustment amount of the micro solenoid valve opening, the micro solenoid valve is regulated;

[0180] Determine the adjustment coefficient of the coolant pump speed according to the temperature deviation and the preset temperature difference threshold;

[0181] Determine the adjustment amount of the coolant pump speed according to the adjustment coefficient of the coolant pump speed;

[0182] The coolant pump is regulated according to the adjustment amount of the coolant pump speed.

[0183] In step S6, for each monitoring area in the array, the micro-solenoid valve and the coolant pump are regulated according to the temperature deviation. Specifically, the adjustment amount of the micro-solenoid valve opening is obtained by proportionally and differentially calculating the temperature deviation using the following formula:

[0184]

[0185] Where, ΔV i represents the opening adjustment of the micro solenoid valve in the i-th area, K p Indicates the proportional coefficient, the value range is 0.1~0.3, K d Indicates the differential coefficient, the value range is 0.05~0.2, Indicates the rate of change of temperature deviation.

[0186] In addition, to ensure the total flow balance, all valve opening adjustments must meet the following requirements:

[0187]

[0188] According to the adjustment amount of the micro solenoid valve opening, the solenoid valve opening control signal is generated and output after being processed by the compensation algorithm:

[0189] u i (t) = u i (t-1)+ΔV i +K c .V target，i -V actual，i / ;

[0190] Among them, u i (t) represents the control signal at the current moment, u i (t-1) represents the control signal at the previous moment, K c Indicates the compensation coefficient, the value is 0.2, V target，i Indicates the target valve opening, V actual，i Indicates the actual valve opening.

[0191] The present invention calculates the solenoid valve opening adjustment through proportional differential (PD) control. It can directly output the adjustment amount according to the current real-time temperature deviation with the help of the proportional link, quickly offset the existing temperature difference, and use the differential link to predict the temperature change trend, intervene in advance to suppress the potential temperature difference, and effectively shorten the temperature response delay. In addition, a valve response characteristic compensation algorithm is designed to take into account the mechanical hysteresis characteristics of the solenoid valve and issue control instructions 0.1 to 0.3 seconds in advance. Finally, the control instructions are sent to each solenoid valve actuator through the CAN bus or SPI communication interface to accurately control the distribution of coolant in various areas of the battery module and achieve refined temperature control.

[0192] Next, collect the maximum temperature difference and average temperature of the battery module:

[0193] ΔT max =T max -T min ;

[0194]

[0195] Where, ΔT max Indicates the maximum temperature difference of the battery module, T max Indicates the maximum temperature, T min Indicates the minimum temperature, Indicates the average temperature, M T Indicates the number of temperature measurement points. Then set the temperature uniformity target and define the maximum allowable temperature difference threshold ΔT threshold The upper threshold of the average temperature is 3℃ is 40℃.

[0196] According to the proportional control algorithm of the temperature difference threshold, the coolant pump speed adjustment coefficient is calculated:

[0197]

[0198] Finally, the coolant pump speed control signal is calculated as:

[0199] ω pump =min{max{ω base ·k pump , 0.3ω rated},ω rated};

[0200] Among them, ω pump Indicates the actual speed of the coolant pump, ω base Indicates the reference speed, ω rated Indicates rated speed.

[0201] To avoid system oscillation, limit the single pump speed adjustment range:

[0202] |ωpump (t)-ω pump (t-1)|≤0.1·ω rated ;

[0203] The present invention is based on a proportional control algorithm based on a temperature difference threshold. When the maximum temperature difference is greater than the temperature difference threshold, the coolant pump speed increase coefficient is calculated as 1+0.2×(actual temperature difference-temperature difference threshold) / temperature difference threshold, thereby enhancing heat dissipation by increasing the flow rate. When the maximum temperature difference is less than 80% of the temperature difference threshold, the coolant pump speed reduction coefficient is calculated as 1-0.1×(temperature difference threshold×0.8-actual temperature difference) / temperature difference threshold, thereby reducing energy consumption while meeting temperature control requirements. At the same time, through a pump speed smooth transition strategy, the single pump speed adjustment amplitude is limited to no more than 10%, thereby avoiding system oscillation. Finally, the pump speed control signal is calculated, and the coolant pump speed is controlled by a PWM signal or an analog voltage signal.

[0204] In summary, this invention uses reinforcement learning to dynamically control the battery module temperature uniformity and minimize system energy consumption as optimization objectives. It defines a state space encompassing the temperature field distribution and fluid parameters, and an action space encompassing the micro-solenoid valve opening and coolant pump speed. Through interaction between the agent and the environment, the decision-making strategy is continuously optimized. A micro-solenoid valve array control mechanism based on the principle of proportional-differential control precisely adjusts the coolant flow rate in different regions, while a dynamic coolant pump speed adjustment mechanism optimizes system energy consumption based on the overall thermal state of the battery. The two work together to achieve an optimal balance between cooling effectiveness and energy efficiency.

[0205] In order to better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below:

[0206] Researchers conducted an immersion liquid cooling dynamic control experiment on a lithium-ion battery module used in electric vehicles. The module consisted of 96 21700 cylindrical cells arranged in an 8×12 array. Each cell had a rated capacity of 4000mAh, a nominal voltage of 3.7V, a total energy of 142.08Wh, and an internal resistance of 23mΩ. The experiment first established a coupled thermal-fluid dynamic model based on the battery's dimensional, geometric, and thermal characteristics. The parameter settings are shown in Table 1:

[0207] Table 1 Battery thermal-fluid coupling dynamic model parameter settings

[0208] Parameter name Numerical unit Battery density 2450 <![CDATA[kg / m 3 ]]> Battery specific heat capacity 950 J / (kg·K) Battery thermal conductivity 42.3 W / (m·K) Coolant density 876 <![CDATA[kg / m 3 ]]> Coolant specific heat capacity 3580 J / (kg·K) Coolant thermal conductivity 0.145 W / (m·K) Coolant dynamic viscosity <![CDATA[1.5×10 -3 ]]> Pa·s Convective heat transfer coefficient 375 <![CDATA[W / (m 2 ·K)]]>

[0209] Subsequently, 32 temperature sensors were placed on the surface and internal hot spots of the battery module. The sampling frequency was set to 20 Hz, and the measurement accuracy was 0.08°C. The sensor distribution is shown in Table 2:

[0210] Table 2 Statistics of the layout of distributed temperature sensors

[0211]

[0212]

[0213] Based on the large amount of collected temperature data, a convolutional neural network temperature field reconstruction algorithm was trained. The training data contained 10,800 sets of samples, 80% of which were used for training and 20% for validation. The parameter settings during the network training process are shown in Table 3:

[0214] Table 3. Training parameters of convolutional neural network temperature field reconstruction algorithm

[0215] Parameter name Numerical Initial learning rate <![CDATA[1.2×10 -3 ]]> Batch size 64 Number of training rounds 200 Regularization coefficient <![CDATA[8.5×10 -5 ]]> Number of early stop rounds 15 Learning rate decay factor 0.85 Validation set accuracy 96.7％

[0216] Then, a long short-term memory network thermal behavior prediction model was trained. The time window length was set to 60 seconds, the sampling interval was 1 second, and the initialization range of each LSTM gate parameter was [-0.08, 0.08]. The specific parameters are shown in Table 4:

[0217] Table 4 Parameters of the long short-term memory network thermal behavior prediction model

[0218] Parameter name Numerical Forget gate threshold 0.53 Input gate threshold 0.62 Output gate threshold 0.51 Learning rate <![CDATA[9.6×10 -5 ]]> Number of training rounds 320 Temperature prediction error 0.78℃ Forecast time lead time 35 seconds

[0219] Based on the above model, a reinforcement learning dynamic control framework was designed. The battery module was divided into 12 independent control areas. Each area was equipped with a micro solenoid valve. The control framework parameter settings are shown in Table 5:

[0220] Table 5 Parameter settings of reinforcement learning dynamic control framework

[0221] Parameter name Numerical Temperature difference weight coefficient 0.72 Energy consumption weight coefficient 0.28 Temperature difference threshold 2.8℃ Additional bonus coefficient 5.0 Discount Factor 0.94 Playback buffer size 120000 Initial exploration rate 0.82 Final exploration rate 0.12 Target network update frequency 100 steps

[0222] After verification through the digital twin simulation module, the module was deployed to the edge computing processor. The control parameters of the micro-solenoid valve array control mechanism and the coolant pump speed dynamic adjustment mechanism are shown in Table 6:

[0223] Table 6 Control mechanism parameter setting table

[0224] Parameter name Numerical Proportional coefficient 0.26 differential coefficient 0.18 Compensation coefficient 0.22 Solenoid valve opening adjustment range -9.5％～+9.5％ Pump speed adjustment range -4.8％～+4.8％ Temperature difference threshold 2.8℃ Average temperature upper threshold 38.5℃ Maximum adjustment of single pump speed 9.2％

[0225] The experiment tested the battery module under 3C fast-charging conditions, comparing the performance differences between traditional PID control and the proposed method. The test results showed that the proposed method reduced the maximum temperature difference of the battery module by 62.5%, from 8.0°C under traditional control to 3.0°C; reduced system energy consumption by 23.7%, from 112.6W under traditional control to 85.9W; improved the battery module temperature uniformity index by 78.3%, from 0.46 to 0.82; and reduced the coolant flow distribution unevenness by 67.8%, from 0.56 to 0.18.

[0226] Most traditional control methods directly adjust the valve opening according to the local temperature difference. The flow of coolant in the flow channel will produce a coupling effect due to the change in local flow. For example, if the valve opening in a certain area is increased to enhance cooling, the resistance of the branch will be reduced, and the coolant will flow into this area first. The flow of other branches will be forced to reduce, which will cause abnormal temperature fluctuations in other areas. The technical solution of the present invention combines deep learning and reinforcement learning algorithms. Through distributed temperature sensing and high-resolution temperature field reconstruction, it can achieve accurate perception of the temperature distribution inside the battery module; predictive control is achieved by predicting the temperature change trend in advance through the time series prediction model; and precise distribution and flow control of coolant are achieved through the micro-solenoid valve array optimized by reinforcement learning and the dynamic adjustment mechanism of pump speed, which greatly improves the temperature uniformity of the battery module and reduces the energy consumption of the system. This dynamic control method significantly improves the efficiency of the battery thermal management system, extends the battery life, and improves the safety and endurance of electric vehicles.

[0227] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 7 to 9 below.

[0228] Table 7 Variable Explanation Table (Part 1)

[0229]

[0230]

[0231] Table 8 Variable Explanation Table (Part 2)

[0232]

[0233] Table 9 Variable Explanation Table (Part 3)

[0234]

[0235]

[0236] like Figure 2 As shown, based on the above method embodiment, a corresponding device embodiment is provided;

[0237] An embodiment of the present invention provides a lithium-ion battery immersion liquid cooling control device, comprising: a battery data acquisition module, a temperature data processing module, an adjustment amount decision module, a reference control module, a temperature difference calculation module, and a fine control module;

[0238] A battery data acquisition module is used to obtain current temperature data, historical temperature data, and current coolant fluid parameters of the distributed array of lithium-ion batteries;

[0239] a temperature data processing module for determining a current temperature field distribution in the lithium-ion battery based on current temperature data, and determining a predicted temperature field distribution in the lithium-ion battery based on historical temperature data and current coolant fluid parameters;

[0240] The adjustment decision module is used to use the current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters, and the current pressure change as the current lithium-ion battery state space, and input the state space into a preset benchmark decision model, so that the benchmark decision model outputs a corresponding action space based on the current lithium-ion battery state space, with the minimum battery module temperature difference and the lowest cooling energy consumption as the reward function; the action space includes: the micro-solenoid valve opening adjustment amount and the coolant pump speed adjustment amount; wherein the current pressure change amount is initially 0;

[0241] A baseline control module is used to perform baseline control on the micro solenoid valves and coolant pumps of the entire array of lithium-ion batteries according to the corresponding action space, and obtain temperature data of the distributed array of lithium-ion batteries after baseline control;

[0242] The temperature difference calculation module is used to determine the temperature deviation of each monitoring area in the array based on the temperature data after baseline control and the preset temperature requirements;

[0243] The fine control module is used to control the micro solenoid valve and coolant pump according to the temperature deviation in each monitoring area in the array.

[0244] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention, which can implement the lithium-ion battery immersion liquid cooling control method provided by any of the above-mentioned method embodiments of the present invention.

[0245] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Furthermore, in the drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which may be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement the present invention without inventive effort.

[0246] Based on the above-mentioned embodiment of the lithium-ion battery immersion liquid cooling control method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the lithium-ion battery immersion liquid cooling control method of any embodiment of the present invention is implemented.

[0247] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0248] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0249] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0250] Based on the above method embodiment, another embodiment is provided: another embodiment of the present invention provides a computer-readable storage medium, including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the lithium-ion battery immersion liquid cooling control method described in any one of the above method embodiments of the present invention.

[0251] Wherein, the module / unit integrated in the lithium-ion battery immersion liquid cooling control device / terminal equipment, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0252] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A lithium-ion battery immersion liquid cooling control method, characterized in that: include: Obtain current temperature data, historical temperature data, and current coolant fluid parameters of a distributed array of lithium-ion batteries; determining a current temperature field distribution in the lithium-ion battery based on the current temperature data, and determining a predicted temperature field distribution in the lithium-ion battery based on the historical temperature data and current coolant fluid parameters; The current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters, and the current pressure change are used as the current lithium-ion battery state space, and the state space is input into a preset benchmark decision model, so that the benchmark decision model outputs a corresponding action space based on the current lithium-ion battery state space, with the minimum battery module temperature difference and the lowest cooling energy consumption as the reward function; the action space includes: the micro-solenoid valve opening adjustment amount and the coolant pump speed adjustment amount; wherein the current pressure change is initially 0; Perform baseline control on the micro solenoid valves and coolant pumps in the entire array of lithium-ion batteries according to the corresponding action space, and obtain temperature data of the distributed array of lithium-ion batteries after baseline control; Determine the temperature deviation of each monitoring area in the array based on the temperature data after baseline control and the preset temperature requirements; For each monitoring area in the array, the micro solenoid valve and coolant pump are regulated according to the temperature deviation.

2. The lithium-ion battery immersion liquid cooling control method according to claim 1, characterized in that: After obtaining the current temperature data, historical temperature data, and current coolant fluid parameters of the distributed array in the lithium-ion battery, it also includes: The current temperature data and the historical temperature data of the distributed array in the lithium-ion battery are subjected to outlier elimination, missing value filling and normalization processing to obtain the preprocessed current temperature data and the preprocessed historical temperature data.

3. The lithium-ion battery immersion liquid cooling control method according to claim 1, characterized in that: Based on the current temperature data, determine the current temperature field distribution in the lithium-ion battery, including: Input the current temperature data into a preset resolution reconstruction model, so that the resolution reconstruction model extracts temperature features from the current temperature data through a built-in encoder to obtain a low-dimensional feature vector; The low-dimensional feature vector is reconstructed through the built-in decoder to generate the current temperature field distribution in the lithium-ion battery.

4. The lithium-ion battery immersion liquid cooling control method according to claim 1, characterized in that: Determine the predicted temperature field distribution in the lithium-ion battery based on historical temperature data and current coolant fluid parameters, including: Inputting the historical temperature data and the current coolant fluid parameters into a preset temperature prediction model, so that the temperature prediction model performs feature splicing on the historical temperature data and the current coolant fluid parameters to obtain a fused feature vector; Through the built-in forget gate, input gate and output gate, the fused feature vector is screened for time-dependent features to obtain the hidden feature vector; Linear mapping is performed on the latent feature vector to generate the predicted temperature field distribution in the lithium-ion battery.

5. The lithium-ion battery immersion liquid cooling control method according to claim 1, characterized in that: Control the micro solenoid valve and coolant pump according to temperature deviation, including: Perform proportional differential calculation on the temperature deviation to obtain the adjustment amount of the micro solenoid valve opening; According to the adjustment amount of the micro solenoid valve opening, the micro solenoid valve is regulated; Determine the adjustment coefficient of the coolant pump speed according to the temperature deviation and the preset temperature difference threshold; Determine the adjustment amount of the coolant pump speed according to the adjustment coefficient of the coolant pump speed; The coolant pump is regulated according to the adjustment amount of the coolant pump speed.

6. The lithium-ion battery immersion liquid cooling control method according to claim 1, characterized in that: The preset baseline decision model is trained in the following way: For the baseline decision model to be trained, define the model's state space, action space, and reward function; The state space includes: current temperature field distribution, predicted temperature field distribution, current coolant fluid parameters, and current pressure change; the action space includes: micro-solenoid valve opening adjustment and coolant pump speed adjustment; the reward function includes: minimum battery module temperature difference and minimum cooling energy consumption; Construct a policy network based on the state space, action space, and current policy network parameters; Initialize the state space and action space to obtain the current state space and current action space; The current state space and the current action space are input into the benchmark decision model to be trained, so that the benchmark decision model to be trained determines the updated state space according to the current state space and the current action space through the built-in digital twin simulation module; the reward function value is calculated according to the updated state space; the policy network is updated through the deep Q network algorithm to obtain the updated policy network parameters; the policy optimization loss function is calculated according to the reward function value, the current policy network parameters and the updated policy network parameters, until the policy optimization loss function converges to obtain the preset benchmark decision model.

7. The lithium-ion battery immersion liquid cooling control method according to claim 6, characterized in that: The digital twin simulation module includes: a battery module three-dimensional heat conduction submodule, a coolant flow submodule, a battery heat flow coupling boundary submodule, a battery heat generation rate calculation submodule, and an energy conservation submodule; The digital twin simulation module is determined in the following way: Obtaining battery parameters, coolant parameters, and boundary condition parameters of the lithium-ion battery; the battery parameters include: battery density, battery specific heat capacity, battery thermal conductivity, and battery heat generation rate; the coolant parameters include: coolant density, coolant dynamic viscosity, coolant specific heat capacity, and coolant thermal conductivity; the boundary condition parameters include: convection heat transfer coefficient and external force per unit volume; According to the battery parameters and coolant parameters of the lithium-ion battery, the battery module three-dimensional heat conduction submodule, coolant flow submodule, battery heat flow coupling boundary submodule, battery heat generation rate calculation submodule and energy conservation submodule are constructed.

8. A lithium-ion battery immersion liquid cooling control device, characterized in that: include: Battery data acquisition module, temperature data processing module, adjustment amount decision module, benchmark control module, temperature difference calculation module and fine control module; The battery data acquisition module is used to obtain current temperature data, historical temperature data, and current coolant fluid parameters of the distributed array of lithium-ion batteries; The temperature data processing module is used to determine the current temperature field distribution in the lithium-ion battery based on the current temperature data, and to determine the predicted temperature field distribution in the lithium-ion battery based on the historical temperature data and the current coolant fluid parameters; The adjustment amount decision module is used to use the current temperature field distribution, the predicted temperature field distribution, the current coolant fluid parameters, and the current pressure change as the state space of the current lithium-ion battery, and input the state space into a preset benchmark decision model, so that the benchmark decision model outputs a corresponding action space based on the current lithium-ion battery state space, with the minimum battery module temperature difference and the lowest cooling energy consumption as the reward function; the action space includes: the micro-solenoid valve opening adjustment amount and the coolant pump speed adjustment amount; wherein the current pressure change amount is initially 0; The reference control module is used to perform reference control on the micro solenoid valves and coolant pumps of the entire array of lithium-ion batteries according to the corresponding action space, and obtain temperature data of the distributed array of lithium-ion batteries after reference control; The temperature difference calculation module is used to determine the temperature deviation of each monitoring area in the array based on the temperature data after baseline control and the preset temperature requirement; The fine control module is used to control the micro electromagnetic valve and the coolant pump according to the temperature deviation for each monitoring area in the array.

9. A terminal device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the lithium-ion battery immersion liquid cooling control method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that include: A stored computer program, wherein when the computer program is run, the device where the computer-readable storage medium is located is controlled to execute the lithium-ion battery immersion liquid cooling control method according to any one of claims 1 to 7.

Images