Machine Learning Methods for Temperature Prediction of Autonomous Underwater Vehicles’ Battery Pack

Battery pack layout is of great significance to enhance the thermal behavior of autonomous underwater vehicles(AUVs). Because battery pack layout is a high-dimensional and nonlinear problem, there is few research on this topic at present. In order to more accurately predict the maximum temperature (MT) and temperature difference (TD) for different battery pack layouts, two machine learning surrogate models were proposed in this paper, including support vector machine (SVM) and the feed-forward fully-connected neural network (FFN). Tens of thousands of battery pack layout scheme databases were obtained through the finite element method. Then, the machine learning based methods were used to predict the MT and TD of the battery pack. The simulation results of this paper showed that both FFN and SVM have low mean absolute percentage error (MAPE) and mean square error (MSE), which means FFN and SVM can accurately predict the temperature. Meanwhile, it can be found that SVM has more advantage in small-scale problem. The methods in this paper can provide guidance for temperature prediction of AUV’s battery pack layout.