Electronic atomization rod battery intelligent temperature control management optimization method and system

Abstract

The invention discloses an electronic atomization rod battery intelligent temperature control management optimization method and system, and the method comprises the steps: determining a heat distribution initial state through analyzing a difference value between battery temperature monitoring data and a safety threshold value, dynamically adjusting capacitor design parameters through employing a real-time improvement algorithm, and carrying out the intelligent temperature control management of a battery according to the optimized capacitor configuration. Evaluating the change trend of the response speed of the battery in a high-load scene, if the requirement of the early warning system is not met, carrying out secondary correction on heat distribution, determining a temperature control strategy based on the adjusted heat distribution model, calculating the heat accumulation trend by utilizing a heat management efficiency evaluation algorithm, and when the heat management efficiency is lower than a preset standard, carrying out secondary correction on the heat distribution; and abnormal points are extracted from the temperature data, and a new capacitor design adjustment scheme is obtained. Finally, temperature control output parameters are determined through iterative optimization, a reaction window of an early warning system is expanded, the stable thermal management operation state of the battery system is achieved, the safety and performance of the battery are effectively improved, and a basis is provided for production optimization of the electronic atomization rod battery.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and particularly to an intelligent temperature control management optimization method and system for an electronic atomizing rod battery. Background Art

[0002] As an emerging consumer electronic product, the intelligence level of the atomizing rod battery management technology of electronic cigarettes directly determines the safety of the product and the user experience. In this field, temperature control management is not only a key link to ensure the stable operation of the device, but also an indispensable core technology to prevent potential risks caused by overheating. With the continuous growth of the market demand for high-performance electronic atomizing rods, the battery system needs to effectively control the temperature while meeting the power output to extend the service life and improve the safety. However, the current research and applications still face many bottlenecks that need to be urgently broken through.

[0003] Existing battery temperature control methods mostly rely on traditional heat dissipation designs or simple temperature monitoring mechanisms, and these solutions often show limitations when dealing with complex usage scenarios. For example, conventional designs are difficult to quickly respond to temperature changes under high-load operation, resulting in heat accumulation exceeding the safety threshold, and even possibly causing device failures or safety hazards. In addition, many solutions are difficult to achieve a balance between power output and temperature control, either sacrificing performance for stability or increasing the overheating risk due to excessive pursuit of power. These defects indicate that there are obvious deficiencies in the accuracy of dynamic temperature control and the system efficiency of the existing technologies.

[0004] In the intelligent temperature control management of electronic atomizing rod batteries, the core challenges focus on how to improve the real-time performance of temperature monitoring and the response ability of the early warning system. Specifically, the unreasonable design of the battery capacitance directly affects the response speed of the system to temperature changes. When the temperature approaches the safety threshold of 60°C, the failure to timely adjust the heat distribution becomes one of the technical bottlenecks. At the same time, if the capacitance parameters cannot balance the power demand and temperature stability, it will lead to temperature control disorders in high-load scenarios. These factors have not been effectively solved, resulting in low thermal management efficiency of the system in a dynamic environment, and thus giving rise to unique optimization problems.

[0005] Therefore, how to improve the accuracy of temperature monitoring by optimizing the battery capacitance design and reduce heat accumulation during high-load use to extend the reaction window of the early warning system has become a key issue in the intelligent temperature control management of electronic atomizing rod batteries. The solution to this problem not only requires balancing the contradiction between power output and temperature control, but also ensuring the stability and safety of the system in complex scenarios. Summary of the Invention

[0006] In order to solve the problems raised in the above background art, in the first aspect of the present invention, an intelligent temperature control management optimization method for an electronic atomizing rod battery is provided, including:

[0007] S1. Obtain multi-point data collected by the temperature sensor during battery operation, and determine the initial state of heat distribution by analyzing the difference between the temperature monitoring data and the safety threshold;

[0008] S2. According to the initial state of heat distribution, dynamically adjust the capacitor design parameters using a real-time improvement algorithm to obtain optimized capacitor configuration data;

[0009] S3. Based on the optimized capacitor configuration data, obtain the change trend of the battery response speed under high-load scenarios, and determine whether it meets the reaction window requirements of the warning system;

[0010] S4. If the change trend of the response speed is lower than the response speed change threshold, perform secondary correction on the heat distribution data to obtain an adjusted heat distribution model;

[0011] S5. According to the adjusted heat distribution model, obtain the real-time balance parameters of power output and temperature control, and determine the temperature control strategy under the current operating state;

[0012] S6. Through the temperature control strategy, calculate the heat accumulation trend of the battery system using a heat management efficiency evaluation algorithm to obtain a quantitative value of heat management efficiency;

[0013] S7. If the quantitative value of heat management efficiency is lower than the preset standard, extract abnormal points from the temperature monitoring data, obtain a new capacitor design adjustment plan, and determine whether iterative optimization is required;

[0014] S8. According to the iteratively optimized capacitor design adjustment plan, obtain the dynamic change data of heat distribution under high-load scenarios, and determine the final temperature control output parameters;

[0015] S9. Through the final temperature control output parameters, expand the reaction window of the warning system using a real-time improvement algorithm to obtain a stable heat management operating state.

[0016] Optionally, step S3, based on the optimized capacitor configuration data, obtaining the change trend of the battery response speed under high-load scenarios, and determining whether it meets the reaction window requirements of the warning system, includes:

[0017] Step S31. Obtain the battery response data under high-load scenarios through the preset capacitor configuration data to obtain the original sequence of response speed;

[0018] Step S32. Extract the speed change characteristics from the original sequence of response speed to determine the time series of speed change;

[0019] Step S33. Process the time series of speed change using a sliding window method to obtain the feature vector of the change trend;

[0020] Step S34, obtain the early warning threshold range of the reaction window through the preset early warning system requirements, and determine whether the change trend exceeds the early warning threshold range;

[0021] Step S35, if the change trend exceeds the early warning threshold range, classify the trend through the support vector machine in scikit-learn to determine the probability value of the abnormal state;

[0022] Step S36, according to the probability value of the abnormal state, use the weighted average method to fuse the response speed and the change trend to obtain a comprehensive evaluation index;

[0023] Step S37, judge whether the battery response meets the requirements of the reaction window according to the matching degree between the comprehensive evaluation index and the system requirements.

[0024] Optionally, in step S35, if the change trend exceeds the early warning threshold range, classify the trend through the support vector machine in scikit-learn to determine the probability value of the abnormal state, including:

[0025] Step S351, obtain the change trend from the time series and generate an initial sequence of the trend;

[0026] Step S352, use the sliding window method to process the initial sequence of the trend, calculate the statistical features within the window, and generate a vector representation after feature extraction;

[0027] Step S353, if the vector representation after feature extraction exceeds the preset vector threshold range, use the support vector machine to classify the vector representation to obtain the probability value of the abnormal state;

[0028] Step S354, according to the probability value of the abnormal state, use the weighted average method to fuse the change trend and the vector representation after feature extraction to generate a score for comprehensive evaluation;

[0029] Step S355, judge whether there is an abnormal state by comparing the score of the comprehensive evaluation with the score threshold range;

[0030] Step S356, after obtaining the judgment result of the abnormal state, use the context information of the time series to smooth the result to generate the final trend state;

[0031] Step S357, for the final trend state, use the logistic regression method to perform secondary verification on the abnormal state to determine the final classification result.

[0032] Optionally, in step S36, according to the probability value of the abnormal state, use the weighted average method to fuse the response speed and the change trend to obtain a comprehensive evaluation index, including:

[0033] Step S361: Obtain a local sequence by means of a sliding window method according to the change trend, and obtain a sequence segment representation.

[0034] Step S362: For the sequence segment representation, calculate the mean and variance within the segment by using a statistical method to obtain distribution characteristic values.

[0035] Step S363: If the distribution characteristic values exceed the preset distribution characteristic threshold range, then perform weighted fusion on the moving speed of the sliding window and the distribution characteristic values to obtain a corrected probability estimate.

[0036] Step S364: According to the corrected probability estimate, calculate a comprehensive score by using a weighted average method to determine whether an abnormal state exists.

[0037] Step S365: For the judgment result, obtain adjacent segment data of the time series to obtain a smoothed state trend.

[0038] Step S366: According to the smoothed state trend, verify the abnormal state by using a preset state trend threshold to determine the final classification result.

[0039] Optionally, in step S4, if the change trend of the response speed is lower than the response speed change threshold, then perform secondary correction on the heat distribution data to obtain an adjusted heat distribution model, including:

[0040] Step S41: If the change trend of the response speed is lower than the response speed change threshold, then judge the relationship between the change trend and the response speed change threshold by using a preset comparison rule to obtain a determination result of the speed change.

[0041] Step S42: According to the determination result, obtain the original data of the heat distribution, and use a data cleaning tool to perform denoising and missing value processing on the original data to obtain processed heat data.

[0042] Step S43: For the processed heat data, perform a secondary correction operation, fit a linear regression model by using the least squares method, and adjust the data deviation to obtain corrected heat data.

[0043] Step S44: Through the corrected heat data, construct an initial distribution model, and use the K-means clustering algorithm to determine the characteristics of the heat distribution to obtain a preliminary distribution model.

[0044] Step S45: If there is a deviation between the characteristics of the preliminary distribution model and the change trend, then perform parameter adjustment on the deviation part to obtain an optimized distribution model.

[0045] Step S46: According to the optimized distribution model, combined with the determination result of the speed change, generate an adjusted heat distribution model.

[0046] Step S47: Calculate the correlation coefficient between the model generation result and the response speed change trend through the adjusted heat distribution model. If the correlation coefficient is greater than the preset value, the final model is obtained.

[0047] Optionally, in step S44, an initial distribution model is constructed through the corrected heat data, and the K-means clustering algorithm is used to determine the characteristics of the heat distribution to obtain a preliminary distribution model, including:

[0048] Step S441: Judge the distribution uniformity of the corrected data through the heat distribution threshold, and output a distribution evaluation result containing discrete values;

[0049] Step S442: Extract the coordinate points exceeding the preset temperature threshold from the distribution evaluation result as heat distribution characteristic points;

[0050] Step S443: Use the K-means algorithm to group the characteristic point coordinates. The input data is the normalized three-dimensional coordinate set, and the output is the grouped data containing the cluster center positions;

[0051] Step S444: Calculate the Euclidean distance between each cluster center and the preset reference point according to the grouped data, establish a mapping relationship table of distance-temperature weights, and output distribution pattern data containing distance coefficients;

[0052] Step S445: When the deviation between the distance coefficient in the distribution pattern and the actual temperature gradient exceeds 5%, use the gradient descent method to update the weight parameters of the mapping relationship table, and output the optimized mapping relationship table;

[0053] Step S446: Smooth the optimized distribution data using Gaussian filtering, with a filter kernel size of 3×3, and output the smoothed distribution data with reduced standard deviation;

[0054] Step S447: On the basis of the smoothed data, use bilinear interpolation to supplement the temperature values of the missing grid points, and use mirror filling for the boundary conditions, and output the complete temperature field matrix;

[0055] Step S448: Reconstruct the three-dimensional thermal map according to the complete temperature field matrix, and use the cluster center positions of the characteristic points as key nodes to generate the final distribution structure.

[0056] Optionally, in step S5, according to the adjusted heat distribution model, obtain the real-time balance parameters of power output and temperature control, and determine the temperature control strategy under the current operating state, including:

[0057] Step S51: Establish a three-dimensional heat distribution model through COMSOL Multiphysics, input real-time temperature sensor data, and output the heat flux density distribution in each region;

[0058] Step S52: Import the heat flux density and the power meter reading into the least squares fitting module to obtain the power-temperature weight coefficient matrix as the real-time balance parameter;

[0059] Step S53: Extract the region with the highest power sensitivity to temperature according to the weight coefficient in the real-time balance parameter;

[0060] Step S54: If the power change amount in this region exceeds the power change threshold, modify the boundary condition constraint value at the corresponding position in COMSOL to generate an updated temperature control instruction set;

[0061] Step S55: Send the updated control instruction set to the PLC controller, and at the same time collect the 10-second interval data of the temperature sensor;

[0062] Step S56: Use the moving average algorithm to smooth the data sequence and then calculate the first derivative to obtain the temperature change rate of each region;

[0063] Step S57: Select the three regions with the largest absolute value of the change rate, adjust their thermal conductivity coefficients in COMSOL, and recalculate the heat flux density distribution;

[0064] Step S58: Input the adjusted heat flux density distribution into the least squares fitting module again to update the power-temperature weight coefficient matrix;

[0065] Step S59: Continuously compare the Euclidean distance between the new matrix and the historical matrix. When the distance is less than 1, it is determined that the system enters the equilibrium state.

[0066] Optionally, in step S51, a three-dimensional heat distribution model is established through COMSOL Multiphysics, real-time temperature sensor data is input, and the heat flux density distribution of each region is output, specifically including:

[0067] Step S511: Match the heat flux density and the region distribution through a pre-established mapping table to obtain the initial heat flux value of each region;

[0068] Step S512: According to the initial heat flux value, obtain real-time data from the sensor, and use the linear interpolation method to process the real-time data to obtain a smoothed heat flux density sequence;

[0069] Step S513: For the smoothed heat flux density sequence, calculate the mean and variance of each region as time series features, and determine the feature change trend;

[0070] Step S514: If the feature change trend exceeds the feature change threshold, use the mesh refinement technology to adjust the mesh density of the corresponding region in the three-dimensional model, and output the updated heat flux distribution;

[0071] Step S515: Using the updated heat flux distribution, calculate the dynamic weight coefficients of each region by the entropy weight method to obtain the distribution parameters after weight adjustment.

[0072] Step S516: Process the distribution parameters and sensor data by the weighted average method, and judge the heat flux stability of each region according to the standard deviation.

[0073] Step S517: For the regions with low heat flux stability, adjust the model boundary conditions and output the optimized heat distribution result.

[0074] Optionally, in step S54, if the power change amount in this region exceeds the power change threshold, modify the boundary condition constraint value at the corresponding position in COMSOL to generate an updated temperature control instruction set, including:

[0075] Step S541: If the absolute value of the power change amount in the region exceeds the power change threshold, compare the power change with the power change threshold to determine the set of regional coordinates that need to be adjusted.

[0076] Step S542: For the set of regional coordinates, obtain the monitoring data sequence containing timestamps and power values from the real-time database.

[0077] Step S543: Process the monitoring data sequence by cubic spline interpolation of the SciPy library and output the smoothed power two-dimensional array.

[0078] Step S544: Calculate the correction value of the heat conduction coefficient at the boundary of each region according to the smoothed power array, and update the material property parameters of the COMSOL model by the finite difference method.

[0079] Step S545: Run the COMSOL fluid dynamics module with the new parameters and output the temperature distribution matrix.

[0080] Step S546: For the temperature distribution matrix, calculate the temperature stability index at the position of each sensor by the inverse distance weighted formula.

[0081] Step S547: When the stability index of any region is greater than 7, adjust the cooling coefficient constraint value of the corresponding region by 1 as the step size.

[0082] Step S548: Repeat the COMSOL simulation and stability calculation until the termination condition is met.

[0083] Step S549: Finally, generate an instruction array containing the region number and the compressor speed according to the preset power-cooling parameter mapping table.

[0084] In the second aspect of the present invention, there is provided an intelligent temperature control management optimization system for an e-cigarette rod battery, which performs intelligent temperature control management optimization on the e-cigarette rod battery by using the method described above. The system includes:

[0085] A temperature data acquisition and analysis module, which is used to obtain multi-point data collected by a temperature sensor during battery operation, and determine the initial state of heat distribution by analyzing the difference between the temperature monitoring data and the safety threshold;

[0086] A capacitance parameter dynamic adjustment module, which is used to dynamically adjust the capacitance design parameters by using a real-time improvement algorithm according to the initial state of heat distribution, and obtain optimized capacitance configuration data;

[0087] A response speed evaluation module, which is used to obtain the change trend of the battery's response speed under high-load scenarios through the optimized capacitance configuration data, and determine whether it meets the reaction window requirements of the early warning system;

[0088] A heat distribution correction module, which is used to perform secondary correction on the heat distribution data if the change trend of the response speed is lower than the response speed change threshold, and obtain an adjusted heat distribution model;

[0089] A temperature control strategy determination module, which is used to obtain the real-time balance parameters of power output and temperature control according to the adjusted heat distribution model, and determine the temperature control strategy under the current operating state;

[0090] A thermal management efficiency evaluation module, which is used to calculate the heat accumulation trend of the battery system by using a thermal management efficiency evaluation algorithm through the temperature control strategy, and obtain a quantitative value of the thermal management efficiency;

[0091] A capacitance design iterative optimization module, which is used to extract abnormal points from the temperature monitoring data and obtain a new capacitance design adjustment plan if the quantitative value of the thermal management efficiency is lower than the preset standard, and determine whether iterative optimization is required;

[0092] A dynamic change data acquisition module, which is used to obtain the dynamic change data of heat distribution under high-load scenarios according to the iterative optimized capacitance design adjustment plan, and determine the final temperature control output parameters;

[0093] A thermal management operation state stability module, which is used to expand the reaction window of the early warning system by using a real-time improvement algorithm through the final temperature control output parameters, and obtain a stable thermal management operation state.

[0094] The technical solutions provided by the embodiments of the present invention have the following beneficial effects:

[0095] An intelligent temperature control management optimization method and system for an e-cigarette rod battery provided by the present invention determine the initial state of heat distribution by analyzing the difference between the battery temperature monitoring data and the safety threshold, and adopt a real-time improvement algorithm to dynamically adjust the capacitor design parameters. According to the optimized capacitor configuration, the change trend of the battery response speed under high-load scenarios is evaluated. If the requirements of the warning system are not met, secondary correction of the heat distribution is performed. Based on the adjusted heat distribution model, a temperature control strategy is determined, and the heat accumulation trend is calculated using a heat management efficiency evaluation algorithm. When the heat management efficiency is lower than the preset standard, abnormal points are extracted from the temperature data to obtain a new capacitor design adjustment plan. Finally, the temperature control output parameters are determined through iterative optimization, and the reaction window of the warning system is extended to achieve a stable heat management operation state of the battery system, effectively improving the safety and performance of the battery and providing a reliable basis for the production optimization of e-cigarette rod batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 It is a flowchart of an intelligent temperature control management optimization method for an e-cigarette rod battery of the present invention.

[0097] Figure 2 It is a schematic structural diagram of an intelligent temperature control management optimization system for an e-cigarette rod battery of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0098] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail below with reference to the drawings and specific embodiments.

[0099] Please refer to Figure 1 , in the first aspect of the present invention, an intelligent temperature control management optimization method for an e-cigarette rod battery is provided, including:

[0100] S1. Obtain multi-point data collected by a temperature sensor during battery operation, and determine the initial state of heat distribution by analyzing the difference between the temperature monitoring data and the safety threshold.

[0101] Optionally, this step further includes:

[0102] Step S11. Obtain multi-point data collected by the temperature sensor during the battery operation period to obtain temperature monitoring values.

[0103] Step S12. Calculate the temperature deviation data of each sensor point according to the difference between the temperature monitoring value and the preset safety threshold.

[0104] Step S13. Use the temperature deviation data to analyze the change trend of heat distribution through linear regression to obtain distribution characteristics.

[0105] Step S14, if the distribution feature exceeds the preset range, use the K-means clustering algorithm to divide the abnormal area to obtain abnormal distribution points.

[0106] Step S15, according to the position information of the abnormal distribution points, use the bilinear interpolation method to generate the global map of the heat distribution and determine the abnormal expansion range.

[0107] Step S16, by comparing the global map with the heat distribution in the initial state, calculate the dynamic characteristics of the distribution change based on the Euclidean distance.

[0108] Step S17, for the dynamic characteristics, use the multiple linear regression algorithm to predict the future trend of the heat distribution and obtain the prediction result.

[0109] Exemplarily, obtain the multi-point data collected by the temperature sensor during the battery operation period to obtain the temperature monitoring values.

[0110] For example, in a battery management system, temperature sensors are distributed at multiple key positions of the battery pack, such as the surface of the battery cells and the heat dissipation channels, and data is collected every 5 seconds. Assume that during a certain operation period, sensor A records 35°C and sensor B records 38°C. These data are the temperature monitoring values and provide the basis for subsequent analysis. According to the difference between the temperature monitoring value and the preset safety threshold, calculate the temperature deviation data of each sensor point.

[0111] Exemplarily, assume that the preset safety threshold is 40°C. The deviation of sensor A is 35 - 40 = -5°C, and that of sensor B is 38 - 40 = -2°C. This deviation data reflects the degree of deviation of each point from the safety standard and provides the basis for judging abnormalities. Use the temperature deviation data to obtain the distribution characteristics by linearly regressing the change trend of the heat distribution.

[0112] In a possible implementation, collect the deviation data for 10 consecutive minutes and plot the time-deviation curve. If the deviation of sensor A gradually rises from -5°C to -1°C, linear regression can fit an upward trend, indicating that the heat accumulation in this area is accelerating. Distribution features such as the slope quantify this change. If the distribution feature exceeds the preset range, use the K-means clustering algorithm to divide the abnormal area to obtain abnormal distribution points.

[0113] Specifically, assume that the slope threshold is 0.5 and the slope of the sensor A area is 0.8, which exceeds the range. K-means clustering divides the battery pack into a normal area and an abnormal area, and the abnormal distribution points may be concentrated on the surface of the battery cells near A. This division helps to quickly locate the problem area. According to the position information of the abnormal distribution points, use the bilinear interpolation method to generate the global map of the heat distribution and determine the abnormal expansion range.

[0114] In one embodiment, the abnormal point A (35 °C) and the adjacent point B (38 °C) are 2 cm apart. Through interpolation calculation, the temperature in the intermediate region is approximately 36.5 °C. The global map shows that the abnormality spreads from A to B, with a range of approximately 5 square centimeters. This improves the visualization accuracy of the heat distribution. By comparing the heat distribution of the global map with the initial state, the dynamic characteristics of the distribution change are calculated based on the Euclidean distance.

[0115] Preferably, at the initial state, the temperature at point A is 30 °C and now it is 35 °C, with a change distance of 5 °C; the temperature at point B rises from 32 °C to 38 °C, with a distance of 6 °C. The dynamic characteristics indicate that the unevenness of the heat distribution intensifies, providing data support for prediction. For the dynamic characteristics, a multiple linear regression algorithm is used to predict the future trend of the heat distribution and obtain the prediction result.

[0116] It can be understood that by combining multiple variables such as temperature, time, and deviation, it is predicted that the temperature at point A may reach 40 °C after 10 minutes, approaching the safety threshold. This prediction result helps with early warning, optimizing the battery heat dissipation design, and improving safety.

[0117] It should be noted that the above method forms a complete chain from data collection to prediction, and the logic between each step is tight.

[0118] For example, the deviation data supports regression analysis, and the clustering result assists in interpolating to generate the global map. The final prediction improves the system reliability. This progressive approach not only accurately locates abnormalities but also prevents potential risks and ensures the stability of battery operation.

[0119] S2. According to the initial state of the heat distribution, a real-time improvement algorithm is used to dynamically adjust the capacitor design parameters to obtain optimized capacitor configuration data.

[0120] Optionally, this step further includes:

[0121] Step S21: Obtain a temperature matrix as the initial heat distribution data set, where the temperature matrix includes at least one temperature value;

[0122] Step S22: Extract the mean, variance, and range from the temperature matrix as eigenvalue;

[0123] Step S23: Input the eigenvalue into the K-means algorithm with a preset clustering number of 3 to obtain a state classification result, where the state classification result includes a set of class labels;

[0124] Step S24: For each class of state data in the set of class labels, use the PID control algorithm to calculate the capacitor parameter adjustment amount to obtain a set of adjusted capacitor values, where the proportional coefficient of the PID control algorithm is 5 and the integral time constant is 2 seconds;

[0125] Step S25, calculate the Euclidean distance between the capacitance values before and after adjustment. When the Euclidean distance exceeds the preset Euclidean distance threshold of 3, optimize the capacitance parameters using the gradient descent method. The learning rate of the gradient descent method is 0.1, and the upper limit of the number of iterations is 100 times. Output the candidate parameter set;

[0126] Step S26, perform a stability test on the candidate parameter set using a sliding window. The window width of the sliding window is 5 samples. When the parameter fluctuation amplitude within 3 consecutive windows is less than 1, determine the candidate parameter set as stable configuration data;

[0127] Step S27, verify the real-time performance of the stable configuration data through the FPGA hardware platform. When the processing delay is less than 10 milliseconds, generate the final configuration parameters;

[0128] Step S28, store the adjustment timestamp, parameter change curve, and verification result into the SQLite database to generate a business data record table.

[0129] Exemplarily, when obtaining the temperature matrix as the initial heat distribution data set, it can be understood that the temperature matrix is usually composed of multi-point data collected by multiple sensors during the operation of the battery.

[0130] For example, in a battery pack, 10 temperature sensors are arranged at different positions of the battery cells, and the temperature values are recorded once per second to form a 10×60 matrix, representing the heat distribution within 1 minute. Each element in the matrix is a specific temperature value, such as 32°C or 39°C, providing basic data for subsequent analysis.

[0131] Exemplarily, when extracting the mean, variance, and range from the temperature matrix as eigenvalues, the average value of all temperature values can be calculated first to reflect the overall heat level.

[0132] For example, the mean of the 60 data in the matrix may be 36°C. The variance measures the degree of dispersion of the temperature. If the variance is large, such as 5, it may indicate that the temperature fluctuates significantly in some areas. The range is the difference between the highest value and the lowest value, such as 42°C minus 30°C to get 12°C, highlighting the non-uniformity of the heat distribution. These eigenvalues describe the heat characteristics from different perspectives.

[0133] Specifically, when inputting the eigenvalues into the K-means algorithm with a preset number of clusters of 3, the algorithm will classify the data into three states according to the mean, variance, and range, such as the low-temperature area, medium-temperature area, and high-temperature area.

[0134] In a possible implementation, the mean of the low-temperature area is about 32°C and the variance is small; the mean of the high-temperature area may reach 40°C and the variance is large. The obtained class label set assigns a classification mark to each group of data for subsequent processing.

[0135] Preferably, when the PID control algorithm is adopted for each type of status data in the category label set, the PID calculates the adjustment amount of the capacitance parameter through the proportional, integral, and differential links.

[0136] For example, in the high-temperature area, it may be necessary to increase the capacitance value to enhance heat dissipation. The PID calculates the adjustment amount to be 0.2 microfarads according to the proportional coefficient of 5 and the integral time constant of 2 seconds. This method can dynamically adjust the parameters to adapt to different heat states.

[0137] In one embodiment, when calculating the Euclidean distance between the capacitance values before and after adjustment, if the capacitance value before adjustment is 1 microfarad and the capacitance value after adjustment is 1.2 microfarads, the distance is 0.2. If it exceeds the preset Euclidean distance threshold of 3, it indicates that the change is too large and further optimization is required.

[0138] It can be understood that when using the gradient descent method, a learning rate of 0.1 and 100 iterations can gradually approach better parameters.

[0139] For example, after iteration, the capacitance value is optimized from 1.2 microfarads to 1.15 microfarads, reducing the risk of over-adjustment. When using a sliding window for stability verification.

[0140] Specifically, a window width of 5 samples means that 5 consecutive parameter values are checked each time. If the fluctuation within 3 consecutive windows is less than 1, such as from 1.15 microfarads to 1.16 microfarads, it indicates that the parameters are stable. This method ensures the reliability of the configuration.

[0141] For example, when verifying the real-time performance through the FPGA hardware platform, if the processing delay is 8 milliseconds, which is lower than the requirement of 10 milliseconds, the generated final configuration parameters can be directly applied to the battery management system. This hardware implementation improves the response speed.

[0142] In a possible implementation manner, when storing the adjustment timestamp, the parameter change curve, and the verification result in the SQLite database, the timestamp records the time when the adjustment occurs, such as 2025-03-26 10:00:00, the parameter change curve shows the change process of the capacitance value from 1 microfarad to 1.15 microfarads, and the verification result records the delay and stability data. This recording method is convenient for tracking and analysis.

[0143] S3. Through the optimized capacitance configuration data, obtain the change trend of the battery response speed under high-load scenarios, and determine whether it meets the reaction window requirements of the warning system.

[0144] Optionally, this step further includes:

[0145] Step S31. Through the preset capacitance configuration data, obtain the battery response data under high-load scenarios, and obtain the original sequence of the response speed.

[0146] Step S32: Extract the speed change features from the original sequence of response speed to determine the time series of speed change.

[0147] Step S33: Process the time series of speed change using the sliding window method to obtain the feature vector of the change trend.

[0148] Step S34: Obtain the warning threshold range of the reaction window through the preset warning system requirements, and determine whether the change trend exceeds the warning threshold range.

[0149] Step S35: If the change trend exceeds the warning threshold range, classify the trend using the support vector machine in scikit - learn to determine the probability value of the abnormal state.

[0150] Step S36: According to the probability value of the abnormal state, fuse the response speed and the change trend using the weighted average method to obtain the comprehensive evaluation index.

[0151] Step S37: Determine whether the battery response meets the requirements of the reaction window according to the matching degree between the comprehensive evaluation index and the system requirements.

[0152] Exemplarily, when obtaining the battery response data under high - load scenarios through the preset capacitance configuration data, it can be understood that such data usually reflects the performance of the battery under extreme working conditions.

[0153] Exemplarily, when a battery pack operates in a high - load charging mode, the original sequence of response speed collected may be the change value of the current output recorded per second, such as rising from 100 amperes to 120 amperes and lasting for 60 seconds, forming a sequence with a length of 60. Such data can intuitively reflect the dynamic response ability of the battery.

[0154] Specifically, when extracting the speed change features from the original sequence of response speed, the key points in the sequence can be concerned.

[0155] For example, take the difference between two adjacent seconds to obtain the time series of speed change, such as changing from 2 amperes / second to 5 amperes / second in a certain period. This change reflects the agility of the battery response.

[0156] In a possible implementation, if the change suddenly increases within a certain period, it may indicate local overload.

[0157] Preferably, when processing the time series of speed change using the sliding window method, the window width is set to 5 seconds and the step size is 1 second.

[0158] Exemplarily, if the change values within a certain window are 2, 3, 4, 5, and 6 amperes per second respectively, the feature vector may be a combination of the average value of 4 and the maximum value of 6. This way can capture the smoothness and abruptness of the trend.

[0159] It should be noted that when obtaining the warning threshold range of the response window through the preset warning system requirements, the warning threshold is usually based on the system's requirements for response time.

[0160] For example, the normal range of speed change is set to be from 1 to 5 amperes per second, and exceeding this range is regarded as abnormal.

[0161] In one embodiment, if the average value of the feature vector of a certain window reaches 7, it exceeds the warning threshold and triggers subsequent analysis.

[0162] It can be understood that if the change trend exceeds the warning threshold range, when classifying the trend through a support vector machine, this method is good at distinguishing normal and abnormal states.

[0163] For example, after inputting the feature vector, the model may output that the probability of the abnormal state is 0.85.

[0164] In one embodiment, if the probabilities of three consecutive windows all exceed 0.8, it is confirmed that the possibility of abnormality is relatively high.

[0165] Specifically, when fusing the response speed and the change trend by using a weighted average method according to the probability value of the abnormal state, a weight of 0.6 can be given to the probability and a weight of 0.4 can be given to the speed change trend.

[0166] For example, at a certain moment, the probability is 0.85 and the trend value is 6. The comprehensive index may be 0.85×0.6 + 6×0.4 = 2.91. This kind of fusion can comprehensively evaluate the battery state.

[0167] In one possible implementation manner, when judging whether the battery response meets the requirements according to the matching degree between the comprehensive evaluation index and the system requirements, if the system requirement index is lower than 3 and the actual value is 2.91, it is considered that the requirements are met.

[0168] For example, in a certain high-load scenario, the index is always lower than the warning threshold, indicating that the battery configuration is reasonable and can quickly respond to the change of demand.

[0169] Optionally, in step S35, if the change trend exceeds the warning threshold range, then classifying the trend through the support vector machine in scikit-learn to determine the probability value of the abnormal state, further includes:

[0170] Step S351, obtaining the change trend from the time series and generating an initial sequence of the trend.

[0171] Step S352: Process the initial sequence of the trend using a sliding window method, calculate the statistical features within the window, and generate a vector representation after feature extraction.

[0172] Step S353: If the vector representation after feature extraction exceeds the preset vector threshold range, use a support vector machine to classify the vector representation and obtain the probability value of the abnormal state.

[0173] Step S354: According to the probability value of the abnormal state, use a weighted average method to fuse the change trend and the vector representation after feature extraction to generate a score for comprehensive evaluation.

[0174] Step S355: By comparing the score for comprehensive evaluation with the preset score threshold range, determine whether there is an abnormal state.

[0175] Step S356: After obtaining the judgment result of the abnormal state, use the context information of the time series to smooth the result and generate the final trend state.

[0176] Step S357: For the final trend state, use a logistic regression method to perform a secondary verification of the abnormal state to determine the final classification result.

[0177] Exemplarily, obtaining the change trend from the time series and generating the initial sequence is the first step in analyzing the response dynamics of the battery under high-load scenarios.

[0178] Exemplarily, the voltage or current data of the battery under a specific load can be arranged in chronological order to form a continuous trend sequence. For example, the voltage value is recorded every 0.1 second to obtain a sequence containing hundreds of data points for subsequent analysis. When using the sliding window method to process the initial sequence, it can be understood that the sequence is scanned one by one through a time window of a fixed length to extract statistical features.

[0179] In a possible implementation, assume that the window length is set to 1 second and the step size is 0.2 second, and calculate the average value, maximum value, and standard deviation within each window.

[0180] For example, the data within a window is [3.5, 3.6, 3.7, 3.8, 3.9] volts, its average value is 3.7 volts, and the standard deviation is approximately 0.16. This method helps to capture the regularity of local changes, and the generated feature vector can better reflect the dynamic characteristics of the trend. If the feature vector exceeds the preset vector threshold range, a support vector machine needs to be used for classification.

[0181] It should be noted that the vector threshold range may be set based on historical data. For example, when the standard deviation exceeds 0.2, it is regarded as abnormal.

[0182] Specifically, the support vector machine maps the feature vectors to a high-dimensional space by training normal and abnormal samples and calculates their abnormal probabilities. For example, if the probability value of a vector is 0.75, it indicates a relatively high possibility of abnormality. This classification method can effectively distinguish between normal and abnormal states. According to the abnormal state probability value, a weighted average method is used to fuse the change trend and the feature vector to generate a comprehensive evaluation score.

[0183] In one embodiment, assume that the trend weight is 0.4 and the feature vector weight is 0.6. If the trend value is 3.8 and the feature value is 0.75, then the comprehensive score is 3.8×0.4 + 0.75×0.6 = 1.97. This fusion method balances the dynamic trend and the static feature and provides a more comprehensive evaluation basis.

[0184] Preferably, when judging the abnormal state by comparing the comprehensive score with the score threshold range, the score threshold can be set to 2.0.

[0185] For example, if the score is 1.97 and it does not exceed the score threshold, the state is considered normal; if the score is 2.3, it is regarded as abnormal. This judgment is intuitive and facilitates the system to respond quickly. After obtaining the abnormal state judgment result, the time series context information is used to smooth the result, which can eliminate the influence of isolated noise.

[0186] For example, if a certain point is judged to be abnormal, but the previous and next 5 points are all normal, it can be smoothed to the normal state. This method improves the stability of the result and avoids misjudgment. For the final trend state, logistic regression is used for secondary verification to further confirm the classification result.

[0187] In one embodiment, logistic regression can combine the context features (such as the average value in the previous 5 seconds) with the current score to output the final probability.

[0188] For example, the comprehensive score is 2.3, the context average value is 3.5, and the logistic regression outputs an abnormal probability of 0.85 to confirm the abnormal state. This secondary verification enhances the reliability of the result and helps to improve the accuracy of the early warning system.

[0189] Optionally, step S36, according to the probability value of the abnormal state, using a weighted average method to fuse the response speed and the change trend to obtain a comprehensive evaluation index, further includes:

[0190] Step S361, according to the change trend, use the sliding window method to obtain the local sequence and get the sequence segmented representation.

[0191] Step S362, for the sequence segmented representation, use the statistical method to calculate the mean and variance within the segment to obtain the distribution feature value.

[0192] Step S363, if the distribution eigenvalue exceeds the preset distribution eigenvalue threshold range, then perform weighted fusion on the moving speed of the sliding window and the distribution eigenvalue to obtain a corrected probability estimate.

[0193] Step S364, according to the corrected probability estimate, use the weighted average method to calculate the comprehensive score and determine whether there is an abnormal state.

[0194] Step S365, for the judgment result, obtain adjacent segmented data of the time series to get a smoothed state trend.

[0195] Step S366, according to the smoothed state trend, use the preset state trend threshold to verify the abnormal state and determine the final classification result.

[0196] Exemplarily, according to the change trend, use the sliding window method to obtain the local sequence to get the sequence segmented representation.

[0197] It can be understood that this method splits the continuous time series into multiple small segments according to certain rules to analyze the local characteristics more carefully.

[0198] Exemplarily, in the high-load scenario of the battery, it is assumed that the current data is collected every 0.2 seconds, the sliding window length is set to 1 second, and the step size is 0.2 seconds. A complete sequence may be split into multiple segments with a length of 1 second. For example, a certain segment of data is [2.1, 2.3, 2.5, 2.4, 2.2] amperes. This segmentation method can clearly reflect the fluctuation of the current in a short time. For the sequence segmented representation, use the statistical method to calculate the mean and variance within the segment to obtain the distribution eigenvalue.

[0199] Specifically, the mean reflects the central tendency of the data, and the variance describes the degree of dispersion of the data.

[0200] In a possible implementation, for the segment of [2.1, 2.3, 2.5, 2.4, 2.2] amperes mentioned above, the mean is about 2.3 amperes and the variance is small, indicating stable fluctuations. For another segment of data such as [2.0, 2.8, 2.1, 2.7, 2.4] amperes, the variance will be significantly larger. This distribution eigenvalue can intuitively reveal the stability of the local sequence. If the distribution eigenvalue exceeds the preset distribution eigenvalue threshold range, then perform weighted fusion on the moving speed of the sliding window and the distribution eigenvalue to obtain a corrected probability estimate.

[0201] It should be noted that the moving speed of the sliding window (such as a step size of 0.2 seconds or 0.5 seconds) will affect the sensitivity of feature extraction.

[0202] Preferably, if the step size is small, a lower weight, such as 0.3, can be assigned to the moving speed during fusion, while the weight of the distribution eigenvalue is 0.7. For example, if the mean value of a certain section is 2.3 amperes and the variance exceeds the distribution feature threshold, combined with a fast moving window, the correction probability may be relatively high, indicating a more obvious abnormal tendency. This kind of fusion can more accurately estimate potential anomalies. According to the corrected probability estimate, a weighted average method is used to calculate the comprehensive score to determine whether there is an abnormal state.

[0203] In one embodiment, assume that the probability estimate is 0.8, the trend value is set to 2.5, and the weights are 0.6 and 0.4 respectively. The score may be close to 1.72. Comparing with the distribution feature threshold of 1.5, an abnormal situation can be preliminarily judged. This method balances multiple factors to ensure a more comprehensive judgment.

[0204] For the judgment result, adjacent segmented data of the time series are obtained to obtain the smoothed state trend.

[0205] For example, if a certain section is determined to be abnormal, but the mean values of the previous and subsequent sections are stably around 2.2 amperes, it can be considered that this abnormality may be noise interference, and the state tends to be normal after smoothing. This kind of smoothing processing can effectively reduce misjudgment and improve consistency. According to the smoothed state trend, a preset state trend threshold is used to verify the abnormal state to determine the final classification result.

[0206] In a possible implementation, the state trend threshold is set to 1.8. The comprehensive score of a certain section after smoothing is 1.9. Combining with the mean value of the previous few sections of 2.1 amperes, the final verification still confirms the abnormality. In another case, if the score drops to 1.6, it is verified as normal. This secondary verification method improves the reliability of classification and helps the system to more accurately identify abnormal battery states.

[0207] S4. If the change trend of the response speed is lower than the preset response speed change threshold, then perform secondary correction on the heat distribution data to obtain an adjusted heat distribution model.

[0208] Optionally, this step further includes:

[0209] Step S41. If the change trend of the response speed is lower than the preset response speed change threshold, then judge the relationship between the change trend and the response speed change threshold through a preset comparison rule to obtain a determination result of the speed change.

[0210] Step S42. According to the determination result, obtain the original data of the heat distribution, and use a data cleaning tool to perform denoising and missing value processing on the original data to obtain processed heat data.

[0211] Step S43. For the processed heat data, perform a secondary correction operation, fit a linear regression model by the least squares method, and adjust the data deviation to obtain corrected heat data.

[0212] Step S44: Construct an initial distribution model based on the calibrated heat data, and use the K-means clustering algorithm to determine the characteristics of the heat distribution, obtaining a preliminary distribution model.

[0213] Step S45: If there is a deviation between the characteristics of the preliminary distribution model and the change trend, then adjust the parameters for the deviation part to obtain an optimized distribution model.

[0214] Step S46: Generate an adjusted heat distribution model based on the optimized distribution model and in combination with the determination result of the speed change.

[0215] Step S47: Calculate the correlation coefficient between the model generation result and the response speed change trend through the adjusted heat distribution model. If the correlation coefficient is greater than the preset value, then obtain the final model.

[0216] Exemplarily, when the change trend of the response speed is lower than the preset response speed change threshold, it is necessary to judge its relationship with the response speed change threshold through a comparison rule.

[0217] For example, in a high-load battery scenario, assume that the preset response speed change threshold is 3 amperes per second, and the actual change trend is 1.5 amperes per second. In one possible implementation, the comparison rule can be defined as "a state of low speed when it is 50% lower than the response speed change threshold", so the determination result is "low speed".

[0218] Exemplarily, from another perspective, if the change trends are 1.2, 1.5, and 1.8 amperes per second respectively in different time periods and are all lower than the response speed change threshold, it can be inferred that the battery is in a stable but slow-reacting state. This determination provides a basic basis for subsequent analysis.

[0219] It can be understood that when obtaining the original data of the heat distribution according to the determination result, the heat data usually comes from sensor acquisition.

[0220] For example, during the operation of the battery, the temperature sequence in a certain area may be 35, 36, and 38 degrees Celsius.

[0221] Preferably, when processed by a data cleaning tool, first remove outliers, such as noise points that suddenly jump to 50 degrees Celsius, and then fill in the missing values with the mean to obtain a smoothed sequence such as 35, 36, and 37 degrees Celsius.

[0222] Specifically, this kind of denoising can avoid interfering with subsequent modeling and ensure the reliability of the data. When performing secondary calibration on the processed heat data, the least squares method for fitting a linear regression model can adjust the deviation.

[0223] In one embodiment, assume that a certain segment of data is 35, 36, and 37 degrees Celsius, but the expected trend should be more linear. After fitting, the corrected values are 35.2, 36.1, and 36.9 degrees Celsius. This method can reduce measurement errors and improve data consistency.

[0224] It should be noted that the corrected data can better reflect the true law of heat change. When constructing the initial distribution model with the corrected heat data, the K-means clustering algorithm can identify features.

[0225] For example, the heat data of a certain battery pack is divided into two categories: a low-temperature area of 30-35 degrees Celsius and a high-temperature area of 40-45 degrees Celsius. The preliminary model after clustering shows the concentrated area of heat distribution. This method intuitively reflects the heat aggregation characteristics.

[0226] In a possible implementation, if the data points are evenly distributed, the feature is reflected as the average heat value. If there is a deviation between the features of the preliminary distribution model and the change trend, the parameters need to be adjusted.

[0227] Exemplarily, assume that the model predicts the high-temperature area to be 40-45 degrees Celsius, but the speed change trend indicates that it should be 42-47 degrees Celsius. Then, adjust the clustering center to the new range, and the optimized distribution model is closer to the actual situation.

[0228] Preferably, this adjustment can improve the adaptability of the model to dynamic scenarios.

[0229] Specifically, when generating the adjusted heat distribution model according to the optimized distribution model combined with the determination result, the low-speed state can be associated with the heat distribution.

[0230] For example, at a low speed, the heat is concentrated near 35 degrees Celsius, and this feature is highlighted in the adjusted model. This integration can more accurately describe the battery operating state.

[0231] It can be understood that when calculating the correlation coefficient through the adjusted heat distribution model, the correlation coefficient reflects the matching degree between the model and the speed change trend.

[0232] In one embodiment, if the correlation coefficient is 0.9, which is greater than the preset value of 0.85, the final model can be confirmed to be valid.

[0233] For example, in a certain scenario, the heat distribution is highly consistent with the speed change trend, indicating that the model can reliably predict the battery behavior. This high correlation provides an important reference for system optimization.

[0234] Optionally, step S44, by using the corrected heat data, constructs an initial distribution model, uses the K-means clustering algorithm to determine the characteristics of the heat distribution, and obtains a preliminary distribution model, further includes

[0235] Step S441, determine the distribution uniformity of the corrected data through a preset heat distribution threshold, and output a distribution evaluation result including discrete values.

[0236] Step S442, extract the coordinate points exceeding the preset temperature threshold from the distribution evaluation result as heat distribution feature points.

[0237] Step S443, use the K-means algorithm to group the feature point coordinates. The input data is a set of normalized three-dimensional coordinates, and the output is grouped data including the cluster center positions.

[0238] Step S444, calculate the Euclidean distance between each cluster center and the preset reference point according to the grouped data, establish a mapping relationship table of distance-temperature weights, and output distribution pattern data including distance coefficients.

[0239] Step S445, when the deviation between the distance coefficient in the distribution pattern and the actual temperature gradient exceeds 5%, use the gradient descent method to update the weight parameters of the mapping relationship table, and output the optimized mapping relationship table.

[0240] Step S446, perform smoothing processing on the optimized distribution data using Gaussian filtering. The filter kernel size is 3×3, and output the smoothed distribution data with reduced standard deviation.

[0241] Step S447, on the basis of the smoothed data, use bilinear interpolation to supplement the temperature values of the missing grid points. The boundary condition uses mirror filling, and output the complete temperature field matrix.

[0242] Step S448, reconstruct the three-dimensional thermal map according to the complete temperature field matrix, and use the cluster center positions of the feature points as key nodes to generate the final distribution structure.

[0243] Exemplarily, when judging the distribution uniformity of the corrected data, the discrete degree of the data can be evaluated through a preset heat distribution threshold. Suppose the temperature value range in a certain heat distribution area is between 20 and 80 degrees Celsius, and the preset heat distribution threshold is 10 degrees Celsius. If the maximum temperature difference in a certain area exceeds the heat distribution threshold, it is considered that the distribution is uneven. In a possible implementation, the temperature mean and variance of each grid point can be statistically calculated, and an evaluation result including discrete values is output. For example, the variance of a certain area is 15, exceeding the standard, indicating that the distribution needs to be optimized. This method intuitively reflects the fluctuation of the data and helps with subsequent adjustments.

[0244] In one embodiment, when extracting feature points from the evaluation result, the temperature threshold can be set to 60 degrees Celsius, and the points exceeding this value are used as representatives of the heat concentration area.

[0245] For example, in a certain distribution, the temperatures at coordinates (3, 5, 2) and (4, 6, 3) are 65 and 70 degrees Celsius respectively, and these two points are selected as feature points. This screening method can quickly locate the key areas for subsequent analysis.

[0246] Specifically, when using the K-means algorithm for grouping, the input three-dimensional coordinates can be normalized to between 0 and 1.

[0247] For example, the original coordinate (3, 5, 2) may become (0.3, 0.5, 0.2) after normalization. After grouping, assuming three clusters are obtained, and the cluster center positions are (0.4, 0.6, 0.3), etc., the output data clearly divides the heat concentration areas. This grouping can effectively reveal the distribution law.

[0248] It should be noted that when calculating the Euclidean distance between the cluster center and the reference point, the reference point can be set as the coordinate origin (0, 0, 0).

[0249] For example, the distance between a certain cluster center (0.4, 0.6, 0.3) and the origin is approximately 0.78. Through the mapping table, this distance can be associated with the temperature weight, such as the closer the distance, the higher the weight. This mapping relationship provides a basis for subsequent modeling.

[0250] Preferably, when the distance coefficient deviates significantly from the temperature gradient, the gradient descent method can be used for optimization.

[0251] For example, the initial weight is set to 1. If the actual gradient shows a deviation of 8%, the weight can be adjusted to 1.2 through multiple iterations to reduce the deviation to less than 5%. This dynamic adjustment enhances the adaptability of the model.

[0252] For example, when using Gaussian filtering to smooth the data, a 3×3 kernel can effectively reduce the influence of noise. Suppose the temperature at a certain point is 70 degrees Celsius, and the surrounding points cause fluctuations due to noise. After smoothing, the standard deviation decreases from 5 to 2, and the distribution becomes more stable. This processing improves the readability of the data.

[0253] It can be understood that when bilinear interpolation is used to supplement missing points, if there is no data at a certain grid point, the temperature weighted calculation of the four adjacent points can be performed.

[0254] For example, the temperatures of the adjacent points are 50, 60, 55, and 65 degrees Celsius, and the interpolation result may be 58 degrees Celsius. The boundary mirror filling ensures the continuity at the edges. This method guarantees the integrity of the temperature field.

[0255] In one embodiment, when reconstructing the three-dimensional thermal map, the cluster center of the feature points can be used as the key node.

[0256] For example, a certain cluster center (0.4, 0.6, 0.3) is marked as the core of the high-temperature area, and the thermal map intuitively shows its transition with the surrounding areas. This structured presentation is convenient for understanding the overall picture of the heat distribution.

[0257] S5. According to the adjusted heat distribution model, obtain the real-time balance parameters of power output and temperature control, and determine the temperature control strategy under the current operating state.

[0258] Optionally, this step further includes:

[0259] Step S51. Establish a three-dimensional heat distribution model through COMSOL Multiphysics, input real-time temperature sensor data, and output the heat flux density distribution in each region.

[0260] Step S52. Import the heat flux density and power meter readings into the least squares fitting module to obtain the power-temperature weight coefficient matrix as the real-time balance parameter.

[0261] Step S53. According to the weight coefficient in the real-time balance parameter, extract the region with the highest sensitivity of power to temperature.

[0262] Step S54. If the power change amount in this region exceeds the power change threshold, modify the boundary condition constraint value at the corresponding position in COMSOL to generate an updated temperature control instruction set.

[0263] Step S55. Send the updated control instruction set to the PLC controller, and at the same time collect the 10-second interval data of the temperature sensor.

[0264] Step S56. Use the moving average algorithm to smooth the data sequence and then calculate the first derivative to obtain the temperature change rate in each region.

[0265] Step S57. Select the three regions with the largest absolute value of the change rate, adjust their thermal conductivity coefficients in COMSOL, and recalculate the heat flux density distribution.

[0266] Step S58. Input the adjusted heat flux density distribution into the least squares fitting module again to update the power-temperature weight coefficient matrix.

[0267] Step S59. Continuously compare the Euclidean distance between the new matrix and the historical matrix. When the distance is less than 1, it is determined that the system enters the balanced state.

[0268] Specifically, when establishing a three-dimensional heat distribution model through COMSOL Multiphysics, it can be understood as a modeling method based on finite element analysis. The core is to visualize the heat distribution of the physical field in three dimensions.

[0269] For example, in the scenario of an industrial heat exchanger, assuming the diameter of the heat exchanger pipe is 50 mm, the length is 2 m, and the external ambient temperature is 25 °C, the real-time temperature sensor data can be input into the model. The sensors are arranged at the pipe inlet, middle, and outlet, recording temperatures of 80 °C, 65 °C, and 50 °C respectively. After the model calculation, the heat flux density distribution in each region is output. The heat flux density at the inlet may reach 1200 W / m 2 , which drops to 800 W / m in the middle 2 , and is 500 W / m at the outlet 2 . This distribution reflects the attenuation law of heat transfer along the pipe.

[0270] Specifically, when importing the heat flux density and the power meter reading into the least squares fitting module, the purpose is to establish the correlation between power and temperature.

[0271] For example, the power meter shows that the power at the inlet is 500 W, 350 W in the middle, and 200 W at the outlet. Through fitting, a power-temperature weight coefficient matrix can be obtained. Assuming the inlet weight coefficient is 0.8, 0.6 in the middle, and 0.4 at the outlet. This indicates that the inlet area is more sensitive to temperature changes.

[0272] Preferably, after extracting the region with the highest sensitivity of power to temperature, if the change in inlet power is 50 W, exceeding the preset power change threshold of 30 W, then the boundary conditions in COMSOL need to be adjusted. For example, change the inlet heat flux boundary value from a fixed 1200 W / m 2 to a dynamic input value to adapt to real-time changes.

[0273] In a possible implementation, after generating the updated temperature control instruction set, it is sent to the PLC controller, and at the same time, temperature data is collected at 10-second intervals.

[0274] For example, the inlet temperature changes from 80 °C to 82 °C, the middle rises from 65 °C to 66 °C, and the outlet remains at 50 °C.

[0275] After smoothing using the moving average algorithm and calculating the first derivative, the inlet temperature change rate is 0.2 °C / s, 0.1 °C / s in the middle, and 0 °C / s at the outlet.

[0276] Select the three regions with the largest change rate to adjust the heat conduction coefficient. Assuming the heat conduction coefficient of the inlet pipe material is adjusted from 50 W / (m·K) to 55 W / (m·K) to accelerate heat transfer and improve the system response efficiency.

[0277] Exemplarily, after recalculating the heat flux density distribution, the inlet heat flux density may rise to 1250 W / m 2 , 820 W / m in the middle 2 , and 510 W / m at the outlet 2It is input into the least squares fitting module again to update the weight coefficient matrix. The inlet coefficient may become 0.85, the middle part is 0.62, and the outlet is 0.41. The Euclidean distance between the new matrix and the historical matrix is compared in a loop. Assuming the initial distance is 1.5, it drops to 0.9 after two iterations, indicating that the system is gradually approaching equilibrium. This method can effectively improve the control accuracy of heat distribution.

[0278] It should be noted that the process of adjusting the heat conduction coefficient and boundary conditions is actually an optimization of the dynamic characteristics of the system.

[0279] For example, if the temperature change rate in the inlet area is too high, it may cause local overheating. By increasing the heat conduction coefficient, heat diffusion can be accelerated to avoid equipment damage.

[0280] In one embodiment, when the Euclidean distance is less than 1, the system enters the equilibrium state, the heat flux density distribution tends to be stable, and the real-time performance of the temperature control instruction set is guaranteed. This method not only improves the flexibility of heat management but also extends the service life of the equipment.

[0281] Optionally, in step S51, a three-dimensional heat distribution model is established by COMSOL Multiphysics, real-time temperature sensor data is input, and the heat flux density distribution of each region is output. It further includes:

[0282] Step S511, match the heat flux density and regional distribution through a pre-established mapping table to obtain the initial heat flux values of each region.

[0283] Step S512, according to the initial heat flux values, obtain real-time data from the sensor, and use the linear interpolation method to process the real-time data to obtain a smoothed heat flux density sequence.

[0284] Step S513, for the smoothed heat flux density sequence, calculate the mean and variance of each region as time series features to determine the feature change trend.

[0285] Step S514, if the feature change trend exceeds the preset feature change threshold, then use the mesh refinement technology in the three-dimensional model to adjust the mesh density of the corresponding region and output the updated heat flux distribution.

[0286] Step S515, through the updated heat flux distribution, calculate the dynamic weight coefficients of each region by using the entropy weight method to obtain the distribution parameters after weight adjustment.

[0287] Step S516, process the distribution parameters and sensor data by using the weighted average method, and judge the heat flux stability of each region according to the standard deviation.

[0288] Step S517, for the regions with low heat flux stability, adjust the model boundary conditions and output the optimized heat distribution result.

[0289] Exemplarily, by matching the heat flux density with the regional distribution through a pre-established mapping table, the initial heat flux values of each region are obtained. The core of this process lies in quickly locating the heat flux reference.

[0290] For example, in an industrial heat exchanger system, assume that the initial heat flux density of a certain region is 50 W / m 2 , and its corresponding physical location, such as the inlet of the pipe near the heat source, can be quickly determined through the mapping table.

[0291] Exemplarily, this method can simplify the complex three-dimensional distribution into an operable initial data set, laying a foundation for subsequent analysis. According to the initial heat flux values, real-time data is obtained from sensors, and the linear interpolation method is used to process the real-time data to obtain a smoothed heat flux density sequence. The key to this technology lies in eliminating data noise.

[0292] It can be understood that the raw data collected by sensors may fluctuate due to environmental interference. For example, a jump from 48 W / m 2 to 52 W / m 2 caused by a sudden temperature change. Through linear interpolation, these points can be smoothed into a continuous 50 W / m 2 trend line, which is convenient for subsequent analysis.

[0293] Preferably, this smoothing process can improve the reliability of the data. For the smoothed heat flux density sequence, the mean and variance of each region are calculated as time series features to determine the trend of feature changes.

[0294] For example, in the heat flux density sequence of a certain region for 10 consecutive minutes, the mean may be 51 W / m 2 , and the variance is 2, indicating that the heat flux is relatively stable. If the variance of another region reaches 10, it indicates that there may be abnormal fluctuations. This feature extraction helps to quickly identify potential imbalance points in the system. If the trend of feature changes exceeds the preset feature change threshold, the grid refinement technology is used in the three-dimensional model to adjust the grid density of the corresponding region, and the updated heat flux distribution is output.

[0295] Specifically, assume that the feature change threshold is a variance of 5. When the variance of a certain region is 10, the grid of this region can be refined from a 1 cm resolution to 0.5 cm. This refinement can more accurately capture local heat flux changes, such as finding the phenomenon of heat flux concentration at a certain pipe corner, thereby optimizing the model accuracy. Through the updated heat flux distribution, the entropy weight method is used to calculate the dynamic weight coefficients of each region to obtain the distribution parameters after weight adjustment.

[0296] In a possible implementation, if the heat flux in a certain area changes frequently, its entropy value is relatively high, and the weight may be adjusted to 0.4, while the weight of the stable area is 0.2. This dynamic allocation can highlight the attention to key areas and improve the system response ability. The weighted average method is used to process the distribution parameters and sensor data, and the heat flux stability of each area is judged according to the standard deviation.

[0297] For example, the weighted average heat flux of a certain area is 52 W / m 2 , and the standard deviation is 1.5, indicating relatively high stability; while the standard deviation of another area is 3, which requires further attention. This method provides data support for subsequent adjustments by quantifying stability. For the area with low heat flux stability, the model boundary conditions are adjusted to output the optimized heat distribution result.

[0298] In one embodiment, if the standard deviation of a certain area is 3, the thermal resistance value in the boundary conditions can be adjusted from 0.1 m 2 ·K / W to 0.05 m 2 ·K / W to enhance the heat transfer efficiency. This adjustment can effectively reduce the risk of local overheating and ensure the stable operation of the overall system.

[0299] It should be noted that the implementation of each step above is closely centered around the real-time optimization of the heat flux distribution.

[0300] For example, from the mapping of the initial value to the adjustment of the boundary conditions, the whole process forms a closed-loop feedback mechanism. The advantage of this method is that it can quickly respond to heat flux changes and at the same time maintain a high degree of consistency between the model and the actual operating state. Through gradual refinement and trade-off, the accuracy and stability of heat management can be significantly improved.

[0301] Optionally, in step S54, if the power change amount in this area exceeds the power change threshold, modify the boundary condition constraint value at the corresponding position in COMSOL to generate an updated temperature control instruction set, which further includes:

[0302] Step S541, if the absolute value of the power change amount in the area exceeds the preset power change threshold, compare the power change with the power change threshold to determine the set of area coordinates that need to be adjusted.

[0303] Step S542, for the set of area coordinates, obtain the monitoring data sequence including the time stamp and power value from the real-time database.

[0304] Step S543, use cubic spline interpolation of the SciPy library to process the monitoring data sequence and output the smoothed two-dimensional power array.

[0305] Step S544, calculate the correction value of the heat conduction coefficient of each area boundary according to the smoothed power array, and update the material property parameters of the COMSOL model by the finite difference method.

[0306] Step S545: Run the COMSOL fluid dynamics module with the new parameters and output the temperature distribution matrix.

[0307] Step S546: For the temperature distribution matrix, use the inverse distance weighted formula to calculate the temperature stability index at each sensor position.

[0308] Step S547: When the stability index of any region is greater than 7, adjust the cooling coefficient constraint value of the corresponding region in steps of 1.

[0309] Step S548: Repeat the COMSOL simulation and stability calculation until the termination condition is met.

[0310] Step S549: Finally, generate an instruction array containing the region number and the compressor speed according to the preset power-cooling parameter mapping table.

[0311] Exemplarily, when the absolute value of the regional power change exceeds the preset power change threshold, it is necessary to quickly locate the affected region.

[0312] For example, in an industrial heat exchanger system, assume the power change threshold is set to 10W. If the power of a certain region jumps from 100W to 115W, with a change of 15W, exceeding the power change threshold, the coordinates of this region are included in the adjustment set.

[0313] Specifically, its position can be determined through a coordinate grid system, such as (3, 5, 2), which is convenient for subsequent processing.

[0314] It can be understood that this comparison can quickly screen out the regions that need attention. When obtaining the monitoring data sequence from the real-time database for the region coordinate set, the data usually includes timestamps and power values.

[0315] In a possible implementation, assume that a certain region has recorded power values within 5 minutes: 100W, 102W, 105W, 108W, 107W, with a timestamp per second. This sequence reflects the dynamic change trend of power.

[0316] Exemplarily, the database can quickly extract this data through the time index, providing a basis for subsequent interpolation. When using cubic spline interpolation of the SciPy library to process the data, the purpose is to generate a smooth power curve.

[0317] For example, the above sequence may have minor fluctuations due to noise. Through interpolation, a smooth two-dimensional array can be generated, such as the power value per second smoothly transitioning to 100W, 101.5W, 104W, 107W, 107.5W.

[0318] Preferably, this method can retain the trend characteristics while reducing the impact of mutations. When calculating the correction value of the thermal conductivity coefficient based on the smoothed power array, it can be deduced through the power difference at the regional boundary.

[0319] Specifically, if the powers on both sides of a certain boundary are 107W and 102W respectively, with a difference of 5W, it can be inferred that the thermal conductivity coefficient needs to be increased by 0.02W / m·K.

[0320] It should be noted that this correction value is then used to update the material properties of the COMSOL model through the finite difference method to ensure the consistency between the simulation and the actual situation. After running the COMSOL fluid mechanics module, the output temperature distribution matrix reflects the temperature of each region.

[0321] For example, the temperature distribution in a certain region may be [350K, 352K, 355K], and in another region it may be [340K, 345K, 348K].

[0322] In one embodiment, this matrix can visually display the heat transfer situation. When calculating the temperature stability index using the inverse distance weighted formula, the temperatures at the sensor positions are comprehensively evaluated.

[0323] For example, if the temperatures near a certain sensor are 352K, 355K, and 353K respectively, and the distance weights are 0.5, 0.3, and 0.2 respectively, the calculated stability index is 6.5.

[0324] Exemplarily, if the index in another region reaches 8, exceeding the index threshold of 7, the cooling coefficient needs to be adjusted. When adjusting the constraint value of the cooling coefficient, it increases or decreases in steps of 1.

[0325] For example, if the stability index of a certain region is 8 and the current cooling coefficient is 3, it can be adjusted to 4 to enhance the cooling effect.

[0326] It can be understood that this iterative adjustment is carried out through repeated simulations until the index drops below 7. When finally generating the instruction array according to the power-cooling parameter mapping table, assuming the mapping table stipulates that 100W corresponds to a rotational speed of 2000rpm and 110W corresponds to 2200rpm. For a power of 107W in a certain region, the rotational speed can be deduced to be approximately 2180rpm.

[0327] In one possible implementation, the instruction array is in the form of [(Region 1, 2180rpm), (Region 2, 2000rpm)], which is directly used for compressor control. This method can quickly respond to power changes and improve the system efficiency.

[0328] S6. Through the temperature control strategy, use the thermal management efficiency evaluation algorithm to calculate the heat accumulation trend of the battery system, and obtain the quantified value of the thermal management efficiency.

[0329] Optionally, this step further includes:

[0330] Step S61: Collect the surface temperature data of the battery cells through a temperature sensor, and use the Pandas library to calculate the integral result of the temperature values within the sliding window as the heat accumulation value.

[0331] Step S62: Apply the Savitzky-Golay filter of Scipy to the heat accumulation value sequence to extract the first derivative as the change trend feature.

[0332] Step S63: Input the trend feature matrix into a pre-trained LIBSVM model, which uses a radial basis function kernel to judge the heat distribution balance.

[0333] Step S64: When the output is in an unbalanced state, adjust the proportional coefficient of the PID controller according to the current temperature range difference to generate a new fan speed control instruction.

[0334] Step S65: According to the operation result of the new instruction, calculate the ratio of the standard deviation of all cell temperatures to the system energy consumption as the efficiency index.

[0335] Step S66: If this index exceeds the preset index threshold of 85, store the current PID parameters in the policy database as the optimized operation mode; otherwise, repeat the adjustment process until the standard is met.

[0336] Exemplarily, after collecting the surface temperature data of the battery cells through a temperature sensor.

[0337] Exemplarily, the Pandas library can be used to process the temperature data within 10 consecutive seconds. Assuming 1 data point is collected per second, a total of 10 data points, such as 28.5, 28.7, 29.0, etc., and the sliding window is set to 5 seconds, calculate the temperature integral during this period to obtain the heat accumulation value. The core of this method is to convert discrete temperature points into a continuous heat trend for subsequent analysis.

[0338] In a possible implementation, if the temperature within the window gradually increases, the integral result may be 143.5, reflecting that heat is accumulating and providing a data basis for subsequent judgment.

[0339] It should be noted that after the heat accumulation value sequence passes through the Savitzky-Golay filter, the first derivative can be extracted as the change trend feature.

[0340] For example, after filtering and smoothing, assuming the derivative of a certain sequence changes from 0.02 to 0.15, it indicates that the temperature rising speed has accelerated. This feature extraction method can effectively capture the dynamic characteristics of heat changes and provide a basis for balance judgment.

[0341] Specifically, if the derivative value in a certain battery cell area continues to be too high, it may indicate local overheating and further attention is required.

[0342] In one embodiment, when inputting the trend feature matrix into the LIBSVM model, it can be understood that the model is trained with historical data to learn to identify the pattern of heat distribution.

[0343] Preferably, due to its strong adaptability to non-linear relationships, the radial basis function kernel can determine whether the temperature change in a certain area is abnormal.

[0344] For example, if the input features show that the derivative of a certain cell is 0.2 while others are 0.05, the model may output "unbalanced", indicating that adjustment is needed. When it is determined to be in an unbalanced state.

[0345] For example, if the temperature range difference reaches 5 degrees, the proportional coefficient of the PID controller can be adjusted.

[0346] In one embodiment, if the original coefficient is 0.8, it can be increased to 1.2, and the fan speed is increased from 2000 revolutions per minute to 2500 revolutions per minute to enhance heat dissipation. This dynamic adjustment can quickly respond to changes in heat distribution and maintain system stability. After running according to the new instruction, calculate the ratio of the standard deviation of all cell temperatures to the energy consumption as the efficiency index.

[0347] For example, if the standard deviation drops from 2.5 to 1.8 and the energy consumption ratio is 80, it indicates that the heat dissipation effect has improved but not reached the standard.

[0348] Preferably, by repeatedly adjusting the PID parameters until the standard deviation is 1.2 and the ratio is 86, the system efficiency is significantly improved. If the efficiency index meets the standard, for example, reaches 87, then the current PID parameters are stored in the strategy database.

[0349] In one possible implementation, parameter combinations such as a proportional coefficient of 1.5 and an integral time of 10 seconds can be used as an optimization mode for subsequent calls. This storage mechanism can accumulate experience and improve the adaptability for long-term operation. If it does not meet the standard, continue to adjust until the requirements are met. This iterative method ensures the continuous optimization of the thermal management strategy.

[0350] S7, if the quantified value of the thermal management efficiency is lower than the preset standard, extract the abnormal points from the temperature monitoring data, obtain a new capacitor design adjustment plan, and determine whether iterative optimization is required.

[0351] Optionally, this step further includes:

[0352] Step S71, when the quantified value of the thermal management efficiency is lower than the preset standard, extract the continuous 24-hour sampling data from the temperature monitoring database, and identify the data points that exceed three times the standard deviation of the mean through the 3σ principle to form an abnormal point set.

[0353] Step S72: Normalize the set of abnormal points, extract the voltage volatility and temperature rise rate as characteristic parameters, input them into a pre-trained support vector classifier, and output the classification result of capacitor overloading or underloading.

[0354] Step S73: Match the capacitor parameter adjustment comparison table according to the classification result, and generate a preliminary adjustment plan for increasing the heat dissipation area or reducing the capacitance value.

[0355] Step S74: Import the preliminary adjustment plan into ANSYS simulation software, run the thermodynamic simulation to obtain new temperature distribution data, and calculate the ratio of the maximum temperature difference before and after adjustment.

[0356] Step S75: If the ratio does not reach the preset improvement threshold of 15%, extract the heat dissipation layer thickness and dielectric constant from the simulation report as optimization variables, and use the gradient descent method to iteratively update the design plan.

[0357] Step S76: In the iteratively optimized design plan, select the efficiency data of the past 30 days as the independent variable and the adjustment plan number as the dependent variable, and use the least squares method to fit a linear equation to predict the efficiency improvement value in the next cycle.

[0358] Step S77: When the predicted value exceeds the preset standard by 5%, output the final design plan.

[0359] Step S78: Extract the dielectric constant and equivalent series resistance parameters from the final plan, calculate the Pearson correlation coefficient with the measured data of the infrared thermal imager, and determine that the optimization is completed when the correlation coefficient is greater than 9.

[0360] Exemplarily, when extracting the continuous 24-hour sampling data from the temperature monitoring database, it can be understood that the temperature fluctuations during the operation of the battery system are captured through long-term monitoring.

[0361] For example, assuming that the sampling is done once per minute, a total of 1440 data points, and the temperature range is between 25 degrees and 35 degrees. When using the 3σ principle to identify abnormal points, first calculate the mean and standard deviation.

[0362] Exemplarily, if the mean is 30 degrees and the standard deviation is 1.5 degrees, then the data exceeding 34.5 degrees or below 25.5 degrees are classified into the set of abnormal points.

[0363] In one embodiment, it may be found that there are 10 points exceeding 35 degrees from 2 am to 3 am, indicating insufficient heat dissipation at night. These abnormal points provide key clues for subsequent analysis.

[0364] Specifically, after normalizing the set of abnormal points and extracting the voltage volatility and temperature rise rate as characteristic parameters, the voltage volatility can be defined as the percentage of voltage change per unit time, and the temperature rise rate is the speed of temperature increase.

[0365] For example, the voltage of a certain abnormal point drops from 4.2 volts to 4.1 volts over 10 minutes, with a volatility of 2.38%, while the temperature rises from 34 degrees to 36 degrees, and the temperature rise rate is 0.2 degrees per minute.

[0366] In one possible implementation, these characteristic parameters are input into a support vector classifier. The model learns through historical data, and the output result may be "capacitor overload", indicating insufficient heat dissipation capacity. This classification lays the foundation for subsequent adjustments.

[0367] Preferably, when matching and adjusting the comparison table according to the classification result, a comparison table can be preset, including heat dissipation area and capacitance value adjustment suggestions.

[0368] For example, if it is determined as "overload", the comparison table may suggest increasing the heat dissipation area from 100 square centimeters to 120 square centimeters, or reducing the capacitance value from 50 microfarads to 45 microfarads.

[0369] It should be noted that this preliminary scheme aims to quickly respond to abnormal states and provide a feasible direction. After importing it into ANSYS simulation software, running a thermodynamic simulation may show that the maximum temperature difference before adjustment is 6 degrees, and it drops to 4.8 degrees after adjustment, with a 20% decrease in the ratio. This simulation verifies the feasibility of the scheme.

[0370] Exemplarily, if the ratio does not reach the 15% improvement threshold, when extracting the heat dissipation layer thickness and dielectric constant as optimization variables from the simulation report, the heat dissipation layer thickness can be adjusted from 2 millimeters to 2.5 millimeters, and the dielectric constant can be reduced from 4.0 to 3.8. When using the gradient descent method for iterative update, it can be understood that the optimal solution is gradually approached through multiple adjustments.

[0371] For example, after three iterations, the temperature difference may drop from 4.8 degrees to 4.2 degrees. This method improves the accuracy of the design.

[0372] Specifically, in the iteratively optimized scheme, when fitting a linear equation using the efficiency data of the past 30 days, efficiency data such as 82, 84, 85, etc. can be used as independent variables, and scheme numbers such as A1, A2, A3 can be used as dependent variables.

[0373] For example, the fitting result may predict that the efficiency of the next cycle is 87.5%, exceeding the preset standard by 5%. This prediction provides data support for decision-making.

[0374] Preferably, after outputting the final scheme, when calculating the Pearson correlation coefficient between the dielectric constant and equivalent series resistance parameters and the measured data of the infrared thermal imager, if the coefficient reaches 0.95, it indicates that the optimization result is highly consistent with the actual situation. This verification ensures the reliability of the scheme.

[0375] S8. According to the adjusted capacitance design scheme after iterative optimization, obtain the dynamic change data of the heat distribution under high-load scenarios, and determine the final temperature control output parameters.

[0376] Optionally, this step further includes:

[0377] Step S81. Through the adjusted capacitance design scheme, obtain the heat distribution data under high-load scenarios.

[0378] Step S82. Extract the distribution characteristics from the heat distribution data, use the gradient descent method to iteratively optimize the heat distribution, and determine the dynamic change data.

[0379] Step S83. According to the dynamic change data, calculate the change trend and judge the preliminary range of temperature control.

[0380] Step S84. If the change trend exceeds the preset change trend threshold, adjust the high-load conditions and re-obtain the optimized distribution characteristics.

[0381] Step S85. According to the optimized distribution characteristics, use the support vector machine to process the dynamic change data and determine the output parameters of temperature control.

[0382] Step S86. According to the output parameters, obtain the final adjustment scheme of temperature control and judge whether the heat distribution is stable.

[0383] Step S87. If the heat distribution is stable, verify the output parameters of temperature control through the change trend to obtain the final result.

[0384] Exemplarily, when obtaining the heat distribution data under high-load scenarios through the adjusted capacitance design scheme, it can be understood as collecting the key information reflecting the heat distribution under the high-load state of the battery system operation.

[0385] For example, in the scenarios of electric vehicle acceleration or climbing, the battery may bear continuous high-current output.

[0386] In one embodiment, assuming that the system operating power is 80 kilowatts and the duration is 15 minutes, the heat distribution data can be collected by a thermocouple array arranged on the surface of the battery module, and the data points obtained may show that the temperature in the edge area is higher than that in the central area.

[0387] Specifically, when extracting the distribution characteristics from the heat distribution data, the concentration degree and gradient change of heat can be concerned.

[0388] In one possible implementation, the distribution characteristics include the position of the highest temperature point and the heat diffusion radius.

[0389] For example, the collected data may show that the highest temperature point is at the top of the battery pack, with a temperature of 38 degrees and a heat diffusion radius of 5 centimeters.

[0390] Preferably, when using the gradient descent method to iteratively optimize the heat distribution, the temperature peak is gradually reduced by adjusting initial parameters such as the heat dissipation wind speed or the thickness of the heat conducting material.

[0391] Exemplarily, after adjusting the wind speed from 2 m / s to 2.5 m / s, the peak temperature may drop from 38 degrees to 36 degrees. This method can effectively smooth the heat distribution.

[0392] It should be noted that when calculating the change trend based on dynamically changing data, the trend reflects the evolution law of temperature over time or load.

[0393] For example, assuming the temperature is recorded every 5 minutes, and the consecutive three recordings are 34 degrees, 35 degrees, and 36 degrees, the change trend shows a linear increase. When determining the initial range of temperature control, an upper limit can be set based on this trend, such as not exceeding 37 degrees. This range provides a basis for subsequent adjustments. If the change trend exceeds the preset change trend threshold, such as the temperature rising rate exceeding 0.3 degrees per minute, it is necessary to adjust the high-load conditions.

[0394] In one embodiment, the input power can be reduced to 70 kW. After re-collecting the data, the temperature rising rate may drop to 0.2 degrees per minute. After extracting the optimized distribution characteristics, for example, the heat concentration area is reduced to 3 centimeters, which supports subsequent analysis.

[0395] Specifically, when using a support vector machine to process dynamically changing data, the temperature change rate and the heat distribution radius can be used as input features to train a model to distinguish between "controllable" and "out-of-control" states.

[0396] In one possible implementation, the model output result may indicate that the temperature is controllable under the current parameters, and the output parameters such as the fan speed are 3000 revolutions per minute. This classification result provides direct guidance for the adjustment scheme.

[0397] Preferably, when obtaining the final adjustment scheme for temperature control according to the output parameters, the adjustment can be combined with the fan speed and the number of heat sinks.

[0398] For example, when the rotation speed is set to 3000 revolutions per minute and the number of heat sinks is increased from 10 to 12, when judging whether the heat distribution is stable, the uniformity of the temperature field can be observed through an infrared thermal imager.

[0399] Exemplarily, after adjustment, the maximum temperature difference drops from 5 degrees to 3 degrees, indicating that the distribution tends to be stable. This scheme improves the durability of the system. If the heat distribution is stable, when verifying the output parameters through the change trend, it can be understood as using the actual operation data to test the effect of the scheme.

[0400] For example, after running continuously for 1 hour, the temperature stabilizes within 35 degrees, which is consistent with the expected trend. The final result shows that the adjustment plan is feasible. This verification ensures the reliability of long-term operation.

[0401] S9. Through the final temperature control output parameters, use a real-time improvement algorithm to expand the reaction window of the warning system to obtain a stable thermal management operating state.

[0402] Optionally, this step further includes:

[0403] Step S91: Obtain temperature control data through sensors and process it using the Kalman filter algorithm to obtain an adjusted parameter optimization result.

[0404] Step S92: Extract the mean, variance, and trend characteristics from the adjusted parameter optimization result, compare the characteristic values with the preset parameter thresholds, determine whether the thermal management state reaches the thermal management state threshold, and determine a preliminary basis for stable state.

[0405] Step S93: For the preliminary basis of stable state, obtain the current operating data of the warning system and process it through the sliding window algorithm to obtain an expanded reaction window range.

[0406] Step S94: According to the expanded reaction window range, calculate the system response time within the window, determine whether it meets the maximum response time limit specified in the real-time monitoring requirements, and obtain optimized system response data.

[0407] Step S95: Through the optimized system response data, use the linear regression algorithm to analyze the change trend of the thermal management state to obtain a stable prediction result of the management operation.

[0408] Step S96: Extract the key indicators of temperature deviation and fluctuation frequency from the stable prediction result of the management operation, determine whether the temperature control is consistent with the parameter optimization, and determine the final thermal management operating state.

[0409] Step S97: According to the final thermal management operating state, obtain the feedback data of real-time monitoring, adjust the reaction window through the sliding window algorithm, and obtain continuously optimized system operating parameters.

[0410] Exemplarily, when obtaining temperature control data through sensors, it can be understood that during the operation of the battery system, sensors arranged at key positions are used to collect temperature information in real time.

[0411] Exemplarily, the sensor can be a thermistor, installed on the surface and inside of the battery module, and the collected data may show that the temperature of a certain area is 40 degrees.

[0412] Specifically, when using the Kalman filter algorithm to process these data, the core lies in smoothing the noise impact through prediction and update steps.

[0413] In one embodiment, the initial temperature data may fluctuate between 39 and 41 degrees due to environmental interference. After filtering, it is adjusted to a stable output of 40 degrees. This method can improve the reliability of the data and provide an accurate basis for subsequent optimization.

[0414] It should be noted that when extracting the mean, variance, and trend features from the optimized results of the adjusted parameters, the mean reflects the average level of the temperature, the variance indicates the degree of fluctuation, and the trend feature reveals the direction of change.

[0415] For example, the mean may be 40 degrees, the variance is 0.5, and the trend shows that the temperature is rising slowly.

[0416] Preferably, when comparing these features with the preset parameter thresholds, the parameter thresholds can be set such that the mean does not exceed 42 degrees and the variance is less than 1. In one possible implementation, if the current mean is 40 degrees and the variance is 0.5, it is determined that the thermal management state is initially stable. This comparison provides a quantitative basis for state assessment. For the initial basis of stable state, when obtaining the current operating data of the warning system, it can be understood as extracting real-time temperature and power information from the system log.

[0417] In one embodiment, the data may show that the power is 75 kilowatts and the temperature is 41 degrees.

[0418] Specifically, when using the sliding window algorithm for processing, the window size can be set to 10 minutes and the step size is 1 minute. The obtained data sequence may reflect that the temperature rises from 40 degrees to 41 degrees. The extended reaction window range is thus determined to be 12 minutes. This processing method can capture short-term fluctuations and provide a more generous reaction time for early warning.

[0419] For example, when calculating the system response time based on the extended reaction window range, the time from temperature over - standard to fan startup within the window may be 30 seconds.

[0420] Preferably, when determining whether the real - time monitoring requirements are met, the maximum response time limit is set to 40 seconds. Since the current 30 seconds is less than 40 seconds, it indicates that the system response data can be optimized. This judgment ensures the real - time nature of the monitoring.

[0421] It should be noted that when analyzing the change trend of the optimized system response data using the linear regression algorithm, the algorithm predicts the future temperature trend through historical data.

[0422] In a possible implementation, the temperature data for the past 30 minutes are 39 degrees, 40 degrees, and 41 degrees, and the prediction result may show that the temperature will reach 42 degrees after 1 hour. This prediction provides a reference for the stability of the management operation.

[0423] Specifically, when extracting the temperature deviation and the fluctuation frequency from the stable prediction result, the deviation may be 1 degree and the fluctuation frequency is once every 10 minutes.

[0424] For example, when judging whether the temperature control and parameter optimization are consistent, if the target deviation is less than 1.5 degrees and the frequency is lower than once every 8 minutes, the current result meets the expectation. This provides a key indicator for determining the final thermal management state.

[0425] Exemplarily, when obtaining the real-time monitoring feedback data according to the final thermal management operation state, the feedback may show that the temperature is stable within 41 degrees.

[0426] In one embodiment, when adjusting the reaction window again by the sliding window algorithm, the window is shortened from 12 minutes to 10 minutes, and the system operation parameter such as the fan speed is adjusted to 3200 revolutions per minute. This continuous optimization ensures the adaptability and efficiency of the system.

[0427] Please refer to Figure 2 , the second aspect of the present invention provides an intelligent temperature control management optimization system for an e-cigarette rod battery, which performs intelligent temperature control management optimization on the e-cigarette rod battery by using the method described above. The system mainly includes:

[0428] A temperature data acquisition and analysis module, which is used to obtain multi-point data collected by a temperature sensor during the operation of the battery, and determine the initial state of the heat distribution by analyzing the difference between the temperature monitoring data and the preset 60°C safety threshold;

[0429] A capacitance parameter dynamic adjustment module, which is used to dynamically adjust the capacitance design parameters by using a real-time improvement algorithm according to the initial state of the heat distribution, and obtain the optimized capacitance configuration data;

[0430] A response speed evaluation module, which is used to obtain the change trend of the response speed of the battery under a high-load scenario through the optimized capacitance configuration data, and judge whether it meets the reaction window requirements of the early warning system;

[0431] A heat distribution correction module, which is used to perform secondary correction on the heat distribution data if the change trend of the response speed is lower than the response speed change threshold, and obtain an adjusted heat distribution model;

[0432] A temperature control strategy determination module, which is used to obtain the real-time balance parameters of power output and temperature control according to the adjusted heat distribution model, and determine the temperature control strategy under the current operating state;

[0433] The thermal management efficiency evaluation module is used to calculate the heat accumulation trend of the battery system through the temperature control strategy and the thermal management efficiency evaluation algorithm to obtain the quantitative value of the thermal management efficiency;

[0434] The capacitor design iterative optimization module is used to extract abnormal points from the temperature monitoring data if the quantified value of the thermal management efficiency is lower than the preset standard, obtain a new capacitor design adjustment plan, and determine whether iterative optimization is required;

[0435] The dynamic change data acquisition module is used to obtain the dynamic change data of heat distribution under high load scenarios according to the capacitor design adjustment plan after iterative optimization, and determine the final temperature control output parameters;

[0436] The thermal management operation state stabilization module is used to expand the response window of the early warning system through the final temperature control output parameter and adopt the real-time improvement algorithm to obtain a stable thermal management operation state.

[0437] The present invention provides an electronic atomizer rod battery intelligent temperature control management optimization method and system, which determines the initial state of heat distribution by analyzing the difference between battery temperature monitoring data and the safety threshold, and dynamically adjusts the capacitor design parameters using a real-time improvement algorithm. According to the optimized capacitor configuration, the battery response speed change trend under high-load scenarios is evaluated. If the early warning system requirements are not met, a secondary correction of the heat distribution is performed. Based on the adjusted heat distribution model, the temperature control strategy is determined, and the heat accumulation trend is calculated using a thermal management efficiency evaluation algorithm. When the thermal management efficiency is lower than the preset standard, abnormal points are extracted from the temperature data to obtain a new capacitor design adjustment plan. Finally, the temperature control output parameters are determined through iterative optimization, and the early warning system reaction window is expanded to achieve a stable thermal management operation state of the battery system, effectively improving the safety and performance of the battery, and providing a reliable basis for the production optimization of electronic atomizer rod batteries.

[0438] It should be noted that the above examples are only some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and there are many variations. All variations that can be directly derived or associated with the content disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims

1. An intelligent temperature control management optimization method for an electronic atomizing rod battery, characterized in that, The method includes: S1. Obtain multi-point data collected by a temperature sensor during battery operation, and determine the initial state of heat distribution by analyzing the difference between the temperature monitoring data and the safety threshold; S2. According to the initial state of heat distribution, dynamically adjust the capacitor design parameters using a real-time improvement algorithm to obtain optimized capacitor configuration data; S3. Based on the optimized capacitor configuration data, obtain the change trend of the battery response speed under high-load scenarios, and determine whether it meets the reaction window requirements of the warning system; S4. If the change trend of the response speed is lower than the response speed change threshold, perform secondary correction on the heat distribution data to obtain an adjusted heat distribution model; S5. According to the adjusted heat distribution model, obtain the real-time balance parameters of power output and temperature control, and determine the temperature control strategy under the current operating state; S6. Through the temperature control strategy, use a heat management efficiency evaluation algorithm to calculate the heat accumulation trend of the battery system to obtain a quantitative value of heat management efficiency; S7. If the quantitative value of heat management efficiency is lower than the preset standard, extract abnormal points from the temperature monitoring data, obtain a new capacitor design adjustment plan, and determine whether iterative optimization is required; S8. According to the iteratively optimized capacitor design adjustment plan, obtain the dynamic change data of heat distribution under high-load scenarios, and determine the final temperature control output parameters; S9. Through the final temperature control output parameters, use a real-time improvement algorithm to expand the reaction window of the warning system to obtain a stable heat management operating state.

2. The method according to claim 1, characterized in that, The step S3, based on the optimized capacitor configuration data, obtaining the change trend of the battery response speed under high-load scenarios, and determining whether it meets the reaction window requirements of the warning system, includes: Step S31. Obtain the battery response data under high-load scenarios through the preset capacitor configuration data to obtain the original sequence of the response speed; Step S32. Extract the speed change characteristics from the original sequence of the response speed to determine the time series of speed changes; Step S33. Use a sliding window method to process the time series of speed changes to obtain the feature vector of the change trend; Step S34. Through the preset warning system requirements, obtain the warning threshold range of the reaction window, and determine whether the change trend exceeds the warning threshold range; Step S35. If the change trend exceeds the warning threshold range, classify the trend using a support vector machine in scikit-learn to determine the probability value of the abnormal state; Step S36. According to the probability value of the abnormal state, use a weighted average method to fuse the response speed and the change trend to obtain a comprehensive evaluation index; Step S37. Judge whether the battery response meets the requirements of the reaction window according to the matching degree between the comprehensive evaluation index and the system requirements.

3. The method according to claim 2, characterized in that, The step S35, if the change trend exceeds the warning threshold range, classifying the trend using a support vector machine in scikit-learn to determine the probability value of the abnormal state, includes: Step S351. Obtain the change trend from the time series to generate the initial sequence of the trend; Step S352: Process the initial sequence of the trend using the sliding window method, calculate the statistical features within the window, and generate the vector representation after feature extraction; Step S353: If the vector representation after feature extraction exceeds the preset vector threshold range, use the support vector machine to classify the vector representation and obtain the probability value of the abnormal state; Step S354: According to the probability value of the abnormal state, use the weighted average method to fuse the change trend and the vector representation after feature extraction to generate the comprehensive evaluation score; Step S355: Determine whether there is an abnormal state by comparing the comprehensive evaluation score with the score threshold range; Step S356: After obtaining the judgment result of the abnormal state, use the context information of the time series to smooth the result and generate the final trend state; Step S357: For the final trend state, use the logistic regression method to perform a secondary verification of the abnormal state to determine the final classification result.

4. The method according to claim 3, wherein In the said Step S36, according to the probability value of the abnormal state, use the weighted average method to fuse the response speed and the change trend to obtain the comprehensive evaluation index, including: Step S361: According to the change trend, obtain the local sequence through the sliding window method to get the sequence segmented representation; Step S362: For the sequence segmented representation, use the statistical method to calculate the mean and variance within the segment to obtain the distribution characteristic value; Step S363: If the distribution characteristic value exceeds the preset distribution characteristic threshold range, then perform weighted fusion on the moving speed of the sliding window and the distribution characteristic value to obtain the corrected probability estimate; Step S364: According to the corrected probability estimate, use the weighted average method to calculate the comprehensive score and judge whether there is an abnormal state; Step S365: For the judgment result, obtain the adjacent segmented data of the time series to get the smoothed state trend; Step S366: According to the smoothed state trend, use the preset state trend threshold to verify the abnormal state to determine the final classification result.

5. The method according to claim 1, characterized in that, In the said Step S4, if the change trend of the response speed is lower than the response speed change threshold, then perform secondary correction on the heat distribution data to obtain the adjusted heat distribution model, including: Step S41: If the change trend of the response speed is lower than the response speed change threshold, then judge the relationship between the change trend and the response speed change threshold through the preset comparison rule to obtain the determination result of the speed change; Step S42: According to the determination result, obtain the original data of the heat distribution, and use the data cleaning tool to perform denoising and missing value processing on the original data to obtain the processed heat data; Step S43: For the processed heat data, perform the secondary correction operation, fit the linear regression model through the least squares method, and adjust the data deviation to obtain the corrected heat data; Step S44: Through the corrected heat data, construct the initial distribution model, and use the K-means clustering algorithm to determine the characteristics of the heat distribution to obtain the preliminary distribution model; Step S45: If there is a deviation between the characteristics of the preliminary distribution model and the change trend, then perform parameter adjustment on the deviation part to obtain the optimized distribution model; Step S46: Generate an adjusted heat distribution model based on the optimized distribution model and in combination with the determination result of the speed change. Step S47: Calculate the correlation coefficient between the model generation result and the response speed change trend through the adjusted heat distribution model. If the correlation coefficient is greater than the preset value, obtain the final model.

6. The method according to claim 5, wherein In the said Step S44: Construct an initial distribution model through the calibrated heat data, and use the K-means clustering algorithm to determine the characteristics of the heat distribution to obtain a preliminary distribution model, including: Step S441: Judge the distribution uniformity of the calibrated data through the heat distribution threshold, and output a distribution evaluation result containing discrete values. Step S442: Extract the coordinate points exceeding the preset temperature threshold from the distribution evaluation result as heat distribution feature points. Step S443: Use the K-means algorithm to group the feature point coordinates. The input data is the normalized three-dimensional coordinate set, and the output is the grouped data containing the cluster center positions. Step S444: Calculate the Euclidean distance between each cluster center and the preset reference point according to the grouped data, establish a mapping relationship table of distance-temperature weights, and output distribution pattern data containing distance coefficients. Step S445: When the deviation between the distance coefficient and the actual temperature gradient in the distribution pattern exceeds 5%, use the gradient descent method to update the weight parameters of the mapping relationship table, and output the optimized mapping relationship table. Step S446: Use Gaussian filtering to smooth the optimized distribution data. The filter kernel size is 3×3, and output the smoothed distribution data with reduced standard deviation. Step S447: On the basis of the smoothed data, use bilinear interpolation to supplement the temperature values of the missing grid points, and use mirror filling for the boundary conditions, and output the complete temperature field matrix. Step S448: Reconstruct a three-dimensional thermal map according to the complete temperature field matrix, and use the cluster center positions of the feature points as key nodes to generate the final distribution structure.

7. The method according to claim 1, characterized in that, In the said Step S5: Obtain the real-time balance parameters of power output and temperature control according to the adjusted heat distribution model, and determine the temperature control strategy under the current operating state, including: Step S51: Establish a three-dimensional heat distribution model through COMSOL Multiphysics, input the real-time temperature sensor data, and output the heat flux density distribution in each region. Step S52: Import the heat flux density and the power meter reading into the least squares fitting module to obtain the power-temperature weight coefficient matrix as the real-time balance parameter. Step S53: Extract the region with the highest power sensitivity to temperature according to the weight coefficient in the real-time balance parameter. Step S54: If the power change amount in this region exceeds the power change threshold, modify the boundary condition constraint value at the corresponding position in COMSOL to generate an updated temperature control instruction set. Step S55: Send the updated control instruction set to the PLC controller, and at the same time collect the 10-second interval data of the temperature sensor. Step S56: Use the moving average algorithm to smooth the data sequence and then calculate the first derivative to obtain the temperature change rate in each region. Step S57: Select the three regions with the largest absolute value of the change rate, adjust their heat conduction coefficients in COMSOL, and recalculate the heat flux density distribution. Step S58: Input the adjusted heat flux density distribution into the least squares fitting module again to update the power-temperature weight coefficient matrix. Step S59: Continuously compare the Euclidean distance between the new matrix and the historical matrix. When the distance is less than 1, it is determined that the system enters the equilibrium state.

8. The method according to claim 7, characterized in that, It also includes: Step S51: Establish a three-dimensional heat distribution model through COMSOL Multiphysics, input real-time temperature sensor data, and output the heat flux density distribution of each region, specifically including: Step S511: Match the heat flux density and the regional distribution through a pre-established mapping table to obtain the initial heat flux values of each region. Step S512: According to the initial heat flux values, obtain real-time data from the sensors, and use the linear interpolation method to process the real-time data to obtain a smoothed heat flux density sequence. Step S513: For the smoothed heat flux density sequence, calculate the mean and variance of each region as time series features, and determine the feature change trend. Step S514: If the feature change trend exceeds the feature change threshold, use the mesh refinement technology in the three-dimensional model to adjust the mesh density of the corresponding region, and output the updated heat flux distribution. Step S515: Through the updated heat flux distribution, calculate the dynamic weight coefficient of each region using the entropy weight method to obtain the distribution parameters after weight adjustment. Step S516: Use the weighted average method to process the distribution parameters and sensor data, and judge the heat flux stability of each region according to the standard deviation. Step S517: For the regions with low heat flux stability, adjust the model boundary conditions and output the optimized heat distribution result.

9. The method according to claim 8, wherein In the said Step S54, if the power change amount in this region exceeds the power change threshold, modify the boundary condition constraint value at the corresponding position in COMSOL to generate an updated temperature control instruction set, including: Step S541: If the absolute value of the power change amount in the region exceeds the power change threshold, compare the power change with the power change threshold to determine the set of regional coordinate values that need to be adjusted. Step S542: For the set of regional coordinate values, obtain the monitoring data sequence containing timestamps and power values from the real-time database. Step S543: Use the cubic spline interpolation of the SciPy library to process the monitoring data sequence and output a smoothed two-dimensional power array. Step S544: Calculate the correction value of the heat conduction coefficient at the boundary of each region according to the smoothed power array, and update the material property parameters of the COMSOL model through the finite difference method. Step S545: Run the COMSOL fluid dynamics module with the new parameters and output the temperature distribution matrix. Step S546: For the temperature distribution matrix, calculate the temperature stability index at the position of each sensor using the inverse distance weighting formula. Step S547: When the stability index of any region is greater than 7, adjust the cooling coefficient constraint value of the corresponding region in steps of 1. Step S548: Repeat the COMSOL simulation and stability calculation until the termination condition is met. Step S549: Finally, generate an instruction array containing the region numbers and the compressor speeds according to the preset power-cooling parameter mapping table.

10. An intelligent temperature control management optimization system for an electronic atomizing rod battery, characterized in that, The intelligent temperature control management of the e-cigarette rod battery is optimized by using the method described in any one of claims 1-9. The system includes: A temperature data acquisition and analysis module, which is used to obtain multi-point data collected by a temperature sensor during battery operation, and determine the initial state of heat distribution by analyzing the difference between the temperature monitoring data and the safety threshold; A capacitance parameter dynamic adjustment module, which is used to dynamically adjust the capacitance design parameters by using a real-time improvement algorithm according to the initial state of heat distribution, and obtain the optimized capacitance configuration data; A response speed evaluation module, which is used to obtain the change trend of the battery response speed under high-load scenarios through the optimized capacitance configuration data, and judge whether it meets the reaction window requirements of the early warning system; A heat distribution correction module, which is used to perform secondary correction on the heat distribution data if the change trend of the response speed is lower than the response speed change threshold, and obtain the adjusted heat distribution model; A temperature control strategy determination module, which is used to obtain the real-time balance parameters of power output and temperature control according to the adjusted heat distribution model, and determine the temperature control strategy under the current operating state; A thermal management efficiency evaluation module, which is used to calculate the heat accumulation trend of the battery system by using a thermal management efficiency evaluation algorithm through the temperature control strategy, and obtain the quantitative value of the thermal management efficiency; A capacitance design iterative optimization module, which is used to extract abnormal points from the temperature monitoring data and obtain a new capacitance design adjustment plan if the quantitative value of the thermal management efficiency is lower than the preset standard, and judge whether iterative optimization is required; A dynamic change data acquisition module, which is used to obtain the dynamic change data of heat distribution under high-load scenarios according to the iterative optimized capacitance design adjustment plan, and determine the final temperature control output parameters; A thermal management operation state stability module, which is used to expand the reaction window of the early warning system by using a real-time improvement algorithm through the final temperature control output parameters, and obtain a stable thermal management operation state.

Images