基于SHAP与物理引导神经网络的卡诺电池主导因素识别与快速优化

Abstract Carnot battery is an emerging long-duration electricity storage technology, which is expected to be applied on a large scale to promote the consumption of fluctuating renewable energy. However, Carnot battery consists of heat pump, heat storage, and heat engine units, presenting a complex energy flow coupling relationship. Dominant factors deciding power-to-power (PTP) efficiency and their coupling relationships in different scenarios are unclear. Conventional optimization methods are also time-consuming, hindering the optimization design. This paper develops a SHapley Additive exPlanations (SHAP) model to identify the dominant factors of Carnot battery and their coupling relationships. A novel optimization method, named SPGO, integrating SHAP and physics-guided neural network (PGNN) model is further proposed to quickly realize the maximum PTP efficiency and give the corresponding design scheme, applicable to various scenarios. Results show that the evaporation temperatures of heat pump and organic Rankine cycle alternately become the most important dominant factor in different scenarios, and there may be a synergistic effect between them. Moreover, the mean absolute errors of the PGNN-based mapping model for PTP efficiency in the test set, interpolation, and extrapolation scenarios are 15.4 %, 18.8 %, and 30.0 % lower than those of the deep neural network, respectively. Compared with the particle swarm optimization method, the optimization time of proposed SPGO method decreases by 99.3 %, and its relative deviations of maximum PTP efficiency in the dataset, interpolation, and extrapolation scenarios are also less than 1 %, 1 %, and 5 %, respectively. This work provides an important optimization method to promote the performance improvement and popularization of Carnot battery.