1.School of Automation, Wuhan University of Technology, Wuhan 430070, China;2.Zhejiang Electric Power Design Institute Co., Ltd., China Energy Engineering Group, Hangzhou 310012, China;3.Guangdong Key Laboratory of Electric Power Equipment Reliability, Electric Power Research Institute of Guangdong Power Grid Co., Ltd., Guangzhou 510080, China

Wind disaster may lead to large-scale power failure in distribution networks. Effective prediction of user number in power outage for distribution networks can provide auxiliary guidance for emergency repair. This paper proposes a prediction model for user number in power outage for distribution networks based on high-efficient data dimensionality reduction. First, based on the data-driven method and considering all characteristic variables, i.e., global variables, a prediction model for user number in power outage for distribution networks based on random forest algorithm is constructed. Then, the relationship between various characteristic variables and response variables affecting power outage users of distribution networks during typhoon is analyzed in detail by using partial dependence plots. The characteristic dimension is reduced and the secondary modeling forms a more efficient prediction model for user number in power outage for distribution networks after dimensionality reduction. Compared with the prediction model considering global variables, the prediction model for user number in power outage for distribution networks after characteristic dimensionality reduction reduces the burden of data collection to a certain extent, improves the calculation efficiency, and provides an effective basis for disaster prevention and mitigation of power grid.