Performance analysis and optimization of packed-bed TES systems based on ensemble learning method

Most existing studies that focused on the performance analysis and optimization of packed bed thermal energy storage systems (PBTESS) have ignored some impact factors, which decreased the accuracy of the system performance prediction. In this work, the Latin Hypercube Sampling (LHS) and numerical simulation were used to construct the training database, and a full-parameter prediction model of PBTESS was constructed with the help of the LightGBM (LGBM), the naive Bayes optimization algorithm was then used to further optimize the performance of PBTESS. The prediction model considered all parameters of the system and the rationality was verified through the numerical simulation. The results showed that LGBM was effective in predicting actual heat release time, actual heat storage, and actual utilization rate of the material with R-square of 0.960, 0.962, 0.956. Meanwhile, the trained prediction model could analyze the effects of all parameters on the performance of PBTESS by the characteristic importance method. The optimization results showed that actual heat release time, actual heat storage, and actual utilization rate of the material of PBTESS were improved by 410%, 14.11%, and 39.86%.