Dynamic control method of battery thermal management system based on deep learning

Abstract

The present invention relates to the field of batteries, and more specifically to a deep learning-based dynamic control method for a battery thermal management system. The method comprises: training a preset neural network based on historically collected battery management data and obtaining an error distribution during the training process; developing a fuzzy control model; inputting the real-time battery temperature into the preset neural network to obtain a battery temperature prediction value; obtaining a battery temperature prediction value distribution based on the error distribution and the battery temperature prediction value; and integrating the real-time coolant flow rate and the real-time fan speed based on the error distribution to obtain a coolant flow correction value and a fan speed correction value, thereby completing the control of the battery thermal management. The technical solution of the present invention can reduce the battery temperature error range, improve the accuracy of the battery temperature detection results, and thereby improve the accuracy of the thermal management system control.

Description

Technical Field

[0001] The present invention relates to the field of batteries. More specifically, the present invention relates to a dynamic control method for a battery thermal management system based on deep learning. Background Art

[0002] As a core component of electric vehicles, batteries' performance directly determines the vehicle's range, safety, and overall user experience. Thermal management has always been a significant challenge in battery systems. Batteries generate significant heat during charging and discharging, and battery temperatures can rise significantly under high-power loads. If battery temperature cannot be effectively controlled, it can impact performance and lifespan, and may even lead to safety hazards such as fires and explosions. Therefore, the design and optimization of battery thermal management systems has become an integral part of electric vehicle technology.

[0003] Current battery thermal management technologies typically utilize liquid or air cooling systems, which absorb and remove heat generated by the battery by regulating the flow of a cooling medium. However, in practice, traditional thermal management systems often suffer from issues such as delayed regulation, uneven heat distribution, and delayed system response, resulting in the inability to accurately control the battery temperature under different operating conditions.

[0004] There is a certain error in the predicted temperature, which means that the battery thermal management system designed based on the predicted temperature may not be able to respond to actual temperature changes completely accurately. Existing battery thermal management systems are usually designed based on the predicted temperature, but do not take into account the impact of prediction errors. This will lead to the risk of overheating or overcooling of the battery during actual operation, thereby affecting battery performance, life, and even safety. Summary of the Invention

[0005] To address the technical issue of inaccurate control of a battery temperature management system due to a failure to account for prediction errors, the present invention provides a deep learning-based dynamic control method for a battery thermal management system. The method includes: training a preset neural network based on historically collected battery management data and obtaining an error distribution during the training process; developing a fuzzy control model, including: using battery temperature, predicted temperature, and ambient temperature as control inputs, and coolant flow rate and fan speed as control outputs; developing control rules based on the control inputs and outputs to complete the construction of the fuzzy control model; inputting the real-time battery temperature into the preset neural network to obtain a predicted battery temperature value; obtaining a predicted battery temperature value distribution based on the error distribution and the predicted battery temperature value; inputting the real-time battery temperature, the predicted battery temperature value distribution, and the real-time ambient temperature into the fuzzy control model; outputting the real-time coolant flow rate and real-time fan speed; integrating the real-time coolant flow rate and real-time fan speed based on the error distribution to obtain a corrected coolant flow rate and a corrected fan speed value, thereby completing the control of battery thermal management.

[0006] By combining historical battery management data with real-time monitoring information, the battery's thermal management system is precisely controlled to ensure stable operation within the optimal temperature range. By predicting battery temperature through a neural network and finely adjusting coolant flow and fan speed based on error distribution, the system dynamically adapts to temperature fluctuations under different battery operating conditions. This not only responds to temperature changes in real time, preventing battery performance degradation or safety issues caused by overheating or hypothermia, but also adjusts system operating parameters based on environmental changes, optimizing energy consumption and improving battery life, efficiency, and safety.

[0007] Preferably, the preset neural network is a BP network, the input of the BP network is battery management data, the battery management data includes the battery current, voltage, soc value and ambient temperature, the output of the BP network is the battery temperature prediction value, and the loss function of the model is mean square error loss.

[0008] Preferably, the fuzzy control model includes: a battery temperature function, a predicted temperature function, an ambient temperature function, a coolant flow function and a fan speed function, wherein the battery temperature function includes a low temperature membership function, a suitable temperature membership function and a high temperature membership function.

[0009] Preferably, the low temperature membership function satisfies the relationship:

[0010] , represents the low temperature membership function, Indicates temperature, Indicates the preset first temperature threshold.

[0011] Preferably, the suitable temperature membership function satisfies the relationship:

[0012] , represents the suitable temperature membership function, Indicates temperature, Indicates the preset first temperature threshold, Indicates the preset second temperature threshold, Indicates the preset third temperature threshold, Indicates the preset fourth temperature threshold.

[0013] Preferably, the high temperature membership function satisfies the relationship:

[0014] , represents the high temperature membership function, Indicates temperature, Indicates the preset fourth temperature threshold, Indicates the preset fifth temperature threshold, Indicates the preset sixth temperature threshold.

[0015] Preferably, the error distribution includes: obtaining all loss values of the preset neural network during the training process, for any loss value, taking the ratio of the number of times the loss value occurs to the total number of training times as the loss value probability, traversing to obtain the loss value probability of each loss value, and obtaining the error distribution.

[0016] The error distribution reflects the error characteristics and stability of the neural network at different training stages, providing a more accurate basis for error control in subsequent temperature predictions. By effectively utilizing the error distribution, targeted adjustments can be made for different training losses, thereby optimizing the accuracy and stability of battery temperature predictions. This enhances the battery thermal management system's adaptability to temperature fluctuations, ensuring that the battery always remains within the ideal operating temperature range, avoiding overheating or low-temperature operation due to prediction errors, and improving the overall battery performance and safety.

[0017] Preferably, the error distribution includes: obtaining a training set of a preset neural network, performing clustering according to fluctuations of the training set to obtain a plurality of clusters, and one cluster corresponding to one error distribution.

[0018] Clustering can be used to identify error characteristics under different operating conditions. In practice, the most appropriate error distribution can be automatically matched to the real-time battery temperature, optimizing the adjustment of coolant flow and fan speed. This not only enables the battery thermal management system to more accurately respond to temperature changes in various dynamic operating conditions, avoiding temperature control instability caused by error fluctuations, but also effectively improves the operating efficiency and safety of the battery.

[0019] Preferably, integrating the real-time coolant flow rate and the real-time fan speed according to the error distribution to obtain the coolant flow rate correction value and the fan speed correction value includes: using the loss value probability of each loss value in the error distribution as a weight, integrating the control data of the real-time coolant flow rate and the real-time fan speed to obtain the control correction value, the control correction value including the coolant flow rate correction value and the fan speed correction value; , represents the control correction value, Indicates the loss value The probability of loss value, Indicates the loss value Corresponding real-time coolant flow and real-time fan speed control data.

[0020] The weighted adjustment method based on error distribution enables the system to more accurately control the addition of coolant and the response of the fan speed system when facing temperature fluctuations and changes in operating conditions, ensuring that the battery operates within the optimal operating temperature range, avoiding extreme conditions such as overheating or low temperatures, and improving the operating stability and safety of the battery.

[0021] Preferably, the integrating the real-time coolant flow and the real-time fan speed according to the error distribution to obtain the coolant flow correction value and the fan speed correction value includes: matching the real-time battery temperature to the cluster, using the error distribution of the cluster including the real-time battery temperature as a weight, and integrating the control data of the real-time coolant flow and the real-time fan speed to obtain the coolant flow correction value and the fan speed correction value.

[0022] The ability to respond to battery temperature changes in real time allows the cooling system to dynamically adjust coolant flow and fan speed based on current temperature conditions and error characteristics, achieving more precise temperature management. This not only optimizes battery heat dissipation and prevents performance degradation due to overheating or overcooling, but also improves overall system reliability and energy efficiency, ensuring that the battery maintains an optimal operating temperature under various operating conditions.

[0023] Beneficial effects of the present invention:

[0024] The present invention uses historical data to train a neural network, which can accurately predict battery temperature and dynamically adjust the coolant flow rate and fan speed based on real-time temperature, environmental conditions, and temperature prediction distribution to ensure that the battery operates within the optimal temperature range. Through the fuzzy control model, it is possible to accurately adjust the battery temperature and make reasonable responses under different temperature conditions. In addition, the introduction of error distribution enables the system to make corrections based on the probability of loss values during historical training, thereby further improving the control effect and reducing the risk of temperature fluctuations and overheating. The present invention not only improves the accuracy and response speed of battery thermal management, but also enhances the adaptability and stability of the system under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flowchart of a dynamic control method for a battery thermal management system based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0027] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0028] Reference Figure 1 The dynamic control method of the battery thermal management system based on deep learning includes steps S1 to S3, which are specifically as follows:

[0029] S1: Train a preset neural network based on historically collected battery management data and obtain the error distribution during the training process.

[0030] In one embodiment, historical battery data is collected, including relevant data such as current, voltage, SOC (State of Charge), ambient temperature, and battery temperature. During the data preprocessing stage, this data is first normalized to eliminate differences between different dimensions and ensure that each feature has a balanced impact on model training. Next, a neural network based on backpropagation (BP) is selected as the prediction model. The BP network takes as input the battery current, voltage, SOC value, and ambient temperature, and outputs the predicted battery temperature.

[0031] To optimize the model's training performance, we use mean squared error loss as the loss function and update the model parameters using a gradient descent algorithm. During training, the model's loss value drops below a set threshold or reaches the maximum number of training iterations. For example, training stops when the number of training iterations reaches 150 or the loss value falls below 0.0001.

[0032] After the training is completed, the loss value of the BP network during each training is counted to obtain the error distribution.

[0033] S2: Formulate a fuzzy control model.

[0034] In one embodiment, the battery temperature, predicted temperature and ambient temperature are used as control inputs, and the coolant flow rate and fan speed are used as control outputs. Control rules are formulated based on the control inputs and control outputs. For example, if the predicted temperature is high, the coolant flow rate and fan speed are increased; if the ambient temperature is high and the predicted temperature is moderate, the coolant flow rate and fan speed are not adjusted.

[0035] In the fuzzy control model, both the control input and the control output have membership functions, including battery temperature function, predicted temperature function, ambient temperature function, coolant flow function and fan speed function. The battery temperature function includes low temperature membership function, suitable temperature membership function and high temperature membership function.

[0036] The low temperature membership function satisfies the relationship:

[0037] , represents the low temperature membership function, Indicates temperature, Indicates the preset first temperature threshold.

[0038] The suitable temperature membership function satisfies the relationship:

[0039] , represents the suitable temperature membership function, Indicates temperature, Indicates the preset first temperature threshold, Indicates the preset second temperature threshold, Indicates the preset third temperature threshold, Indicates the preset fourth temperature threshold.

[0040] The high temperature membership function satisfies the relationship:

[0041] , represents the high temperature membership function, Indicates temperature, Indicates the preset fourth temperature threshold, Indicates the preset fifth temperature threshold, Indicates the preset sixth temperature threshold.

[0042] The temperature range of the battery is well known to those skilled in the art, so the first temperature threshold, the second temperature threshold, the third temperature threshold, the fourth temperature threshold, the fifth temperature threshold and the sixth temperature threshold can all be set by those skilled in the art.

[0043] The predicted temperature function, ambient temperature function, coolant flow function, and fan speed function can be directly constructed using existing membership functions.

[0044] By defining membership functions for different temperature ranges, the model can more accurately reflect battery conditions within different temperature ranges. Low-temperature, optimum-temperature, and high-temperature membership functions enable the battery management system to respond appropriately to different temperature ranges, avoiding overcooling or insufficient cooling. The low-temperature range ensures the battery remains unaffected by low temperatures by properly adjusting coolant flow and fan speed. The optimum-temperature range ensures the battery maintains stable performance within its optimal operating temperature range. The high-temperature range automatically adjusts based on preset thresholds to prevent overheating and extend battery life.

[0045] Fuzzy control can address the complexity of temperature prediction errors and external environmental changes during actual operation, improving system robustness and stability. Furthermore, the use of fuzzy control methods can avoid the sudden changes that may occur in traditional control methods, reduce unnecessary system fluctuations, optimize battery thermal management, and ensure battery safety and performance under various operating conditions.

[0046] S3: Input the real-time battery temperature into the preset neural network to obtain the battery temperature prediction value, obtain the battery temperature prediction value distribution based on the error distribution and the battery temperature prediction value, input the real-time battery temperature, the battery temperature prediction value distribution and the real-time ambient temperature into the fuzzy control model, output the real-time coolant flow and real-time fan speed, integrate the real-time coolant flow and real-time fan speed according to the error distribution to obtain the coolant flow correction value and the fan speed correction value, and complete the control of the battery thermal management.

[0047] In one embodiment, based on all the loss values obtained in step S1, the ratio of the number of times the loss value occurs to the total number of training times is used as the loss value probability, and the loss value probability of each loss value is traversed to obtain the error distribution.

[0048] By analyzing the distribution of loss values during neural network training, we can gain a deeper understanding of the model's error characteristics and stability at different training stages. Obtaining the error distribution makes model performance evaluation more intuitive and can guide subsequent parameter adjustment and optimization processes, thereby improving the model's generalization ability and practical application effects.

[0049] The real-time battery temperature, the distribution of predicted battery temperature values, and the real-time ambient temperature are input into the fuzzy control model, and the real-time coolant flow rate and the real-time fan speed are output. The loss value probability of each loss value in the error distribution is used as a weight, and the control data of the real-time coolant flow rate and the real-time fan speed are integrated to obtain the control correction value, which includes the coolant flow correction value and the fan speed correction value.

[0050] , represents the control correction value, Indicates the loss value The probability of loss value, Indicates the loss value Corresponding real-time coolant flow and real-time fan speed control data.

[0051] By using the probability of each loss value in the error distribution as a weight, we perform a weighted integration of the control data for coolant flow and fan speed. This more accurately reflects the error characteristics during training, allowing for real-time, refined adjustments to the cooling system based on actual operating conditions. This approach enables the model to provide appropriate control corrections under various operating conditions, optimizing coolant flow and fan speed to ensure the battery operates within a safe and efficient operating temperature range.

[0052] The battery temperature is controlled according to the obtained coolant flow correction value and fan speed correction value.

[0053] In another embodiment, a training set of a preset neural network is obtained, and clustering is performed according to fluctuations of the training set to obtain a plurality of clusters, where one cluster corresponds to one error distribution.

[0054] The real-time battery temperature is matched to the cluster, the error distribution of the cluster containing the real-time battery temperature is used as the weight, and the control data of the real-time coolant flow and the real-time fan speed are integrated to obtain the coolant flow correction value and the fan speed correction value.

[0055] By clustering the fluctuations of a preset neural network training set, a corresponding error distribution is established for each cluster, and the most appropriate cluster is matched based on the real-time battery temperature. This method allows for more precise dynamic adjustment of coolant flow and fan speed based on changes in battery temperature. By using the cluster error distribution as weight for control data integration, real-time correction can be made based on error characteristics under different operating conditions, optimizing the performance of the battery cooling system.

[0056] It should be noted that those skilled in the art may make various modifications and improvements without departing from the scope of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be based on the appended claims.

Claims

1. A dynamic control method for a battery thermal management system based on deep learning, characterized in that: include: Train a preset neural network based on historically collected battery management data and obtain the error distribution during the training process; The error distribution includes: obtaining all loss values of the preset neural network during the training process, and for any loss value, taking the ratio of the number of times the loss value appears to the total number of training times as the loss value probability, traversing to obtain the loss value probability of each loss value, and obtaining the error distribution; Develop a fuzzy control model, including: using battery temperature, predicted temperature, and ambient temperature as control inputs, and coolant flow and fan speed as control outputs. Develop control rules based on the control inputs and control outputs to complete the construction of the fuzzy control model. Inputting the real-time battery temperature into a preset neural network to obtain a battery temperature prediction value, obtaining a battery temperature prediction value distribution based on the error distribution and the battery temperature prediction value, inputting the real-time battery temperature, the battery temperature prediction value distribution, and the real-time ambient temperature into a fuzzy control model, and outputting a real-time coolant flow rate and a real-time fan speed; integrating the real-time coolant flow rate and the real-time fan speed according to the error distribution to obtain a coolant flow correction value and a fan speed correction value, including: using the loss value probability of each loss value in the error distribution as a weight, integrating control data of the real-time coolant flow rate and the real-time fan speed to obtain a control correction value, the control correction value including a coolant flow correction value and a fan speed correction value; , represents the control correction value, Indicates the loss value The probability of loss value, Indicates the loss value The corresponding real-time coolant flow and real-time fan speed control data complete the control of battery thermal management.

2. The method for dynamic control of a battery thermal management system based on deep learning according to claim 1, characterized in that: The preset neural network is a BP network, the input of the BP network is battery management data, the battery management data includes the battery current, voltage, soc value and ambient temperature, the output of the BP network is the battery temperature prediction value, and the loss function of the model is mean square error loss.

3. The method for dynamic control of a battery thermal management system based on deep learning according to claim 1, characterized in that: The fuzzy control model includes: Battery temperature function, predicted temperature function, ambient temperature function, coolant flow function and fan speed function, wherein the battery temperature function includes a low temperature membership function, a suitable temperature membership function and a high temperature membership function.

4. The method for dynamic control of a battery thermal management system based on deep learning according to claim 3, characterized in that: The low temperature membership function satisfies the relationship: , represents the low temperature membership function, Indicates temperature, Indicates the preset first temperature threshold.

5. The method for dynamic control of a battery thermal management system based on deep learning according to claim 3, characterized in that: The suitable temperature membership function satisfies the relationship: , represents the suitable temperature membership function, Indicates temperature, Indicates the preset first temperature threshold, Indicates the preset second temperature threshold, Indicates the preset third temperature threshold, Indicates the preset fourth temperature threshold.

6. The method for dynamic control of a battery thermal management system based on deep learning according to claim 3, characterized in that: The high temperature membership function satisfies the relationship: , represents the high temperature membership function, Indicates temperature, Indicates the preset fourth temperature threshold, Indicates the preset fifth temperature threshold, Indicates the preset sixth temperature threshold.

Images