融合机器学习与 SHAP 值算法的居民需求响应个体异质性因素挖掘与应用研究

Based on the large-scale residential electricity demand response (EDR) experiment and household survey data, this study uses machine learning and SHAP (Shapley additive explanatory) value algorithm to identify and analyze the influencing factors of residents’ participation in demand response from the whole and individual levels. Our study found that whether residents participate in EDR activities is the result of the joint action of external incentives, family structure, electricity use habits, electricity use knowledge, and the value and polarity of the effect is varied in heterogeneity. Among them, external incentives such as telemarketing have the greatest impact on customers’ participation in demand response, and the effect is more obvious among older and more educated groups; customers with a baseline electricity consumption of about 1 kW·h during the response period have a higher tendency to participate; households with a stronger awareness of energy conservation and higher conditions for saving electricity have a higher probability of participation. At the same time, according to the interaction and decomposition properties of SHAP value, classified marketing for users in subsequent EDR activities can save 93.9% of marketing costs and increase the number of participants by 46.4%. This study has carried out a more detailed analysis and research on the heterogeneity of different groups, providing an important support for more accurate and intelligent EDR to China’s new power system.