The stochastic fluctuation of high-penetration distributed photovoltaic (PV) in distribution networks leads to complex power flow states

exacerbating the frequent voltage violation frequency and network losses. To overcome the issues

this paper proposes a hybrid knowledge-data driven adaptive voltage/var control strategy optimization of PV inverters in distribution networks to achieve efficient voltage control by combining domain knowledge and data-driven learning. First

through historical/forecasted power data of PV and loads

power flow and optimal power flow are both calculated to construct the knowledgeable dataset of PV voltage control. Then

an intelligent PV voltage controller based on fuzzy control theory is designed and the inference knowledge from PV voltage and PV active power to PV Var decision is embedded into the controller. Next

on the basis of the constructed dataset

a data-driven gradient descent algorithm is presented to optimize the parameters of PV voltage controller

so that the accurate map from PV voltage and active power states to the optimal Var decision can be realized to improve the global control performance. Finally

an online decentralized automatic strategy is developed with the optimized PV controllers

which considers both voltage violation control and network loss optimization

and performs adaptive PV Var control responding to operating states. Different control methods are compared by simulations

which verifies the effectiveness and superiority of the method proposed in this paper.