Research On Compressed Sensing About Temperature Field Of Power Battery Pack Based On Machine Learning

In recent years,with the strong support of the country for the new energy industry,the application scenarios of power batteries are becoming more and more extensive.However,the thermal field imbalance is still one of the bottlenecks restricting the performance improvement of power batteries,so the battery thermal management system is attracting more and more attention from professionals.Based on thermodynamic temperature field model because of it need a lot of certainty material parameters,and easily affected by environmental temperature,and as with the aging process of the battery,a lot of physical and chemical parameters will change,these factors are seriously limits the application range of this method;however,it is also difficult to achieve a complete layout that completely depends on the sensor network in engineering applications.In order to solve the above problems,this paper explores the machine learning method to solve the problem of temperature field compression perception,and research the two-dimensional unsteady temperature field online prediction method of 18650 power lithium battery pack under the condition of low sensor compression rate.In order to simulate the real working environment,the adjustable power load is used in the experiment,the temperature field of the power battery pack is completely sampled by the calibrated 128-channel thermocouple sensor and the sampled data is taken as the prediction target,the ElaticNet,KNN,ERT,DNN and LSTM algorithms in machine learning were respectively used to input sparse sampling data under different sensor compression rates for offline prediction and online verification.The differences between the predicted temperature field data and the fully sampled temperature field data were analyzed and evaluated,and the results show that in the experimental environment:(1)With the continuous increase of the compression rate of the sensor,the overall prediction error becomes smaller and smaller,however when the compression ratio of the sensor increases to a certain extent,the prediction error decreases gradually.(2)Besides the linear algorithm ElasticNet,the rest of the algorithm is available from the sensor in compression ratio as low as 5%,and when different monomer battery at the same time under the condition of the highest temperature difference is as high as 35 ?,still can rebuild the maximum error is less than 1 ?,the average error is less than 0.1 ? temperature field.(3)Compared with the temperature field prediction effect of DNN and LSTM,ERT and KNN algorithm has higher accuracy and stability.To sum up,this study provides theoretical basis for temperature monitoring under low sensor compression rate and experimental basis for performance optimization of the sensing part of the thermal management system about power batteries,which has a good originality and application prospect in the field of new energy.