Lightweight probability forecasting and local control of photovoltaic integrated with energy storage system in active distribution networks

The increase in the use of photovoltaics (PVs) has exacerbated reverse power flow and voltage violations in active distribution networks (ADNs). Photovoltaic integrated with energy storage system (PV-ESS) can effectively alleviate the aforementioned issues by controlling the ESS. To achieve efficient control of ESS, accurate forecasting of PV power is required. Edge computing provides a localized solution for predictive control of PV-ESS. However, there are several challenges with edge computing, including limited computational resources at the edge and high uncertainty of the PV power. Thus, this paper proposes a lightweight probabilistic forecasting and local control method for PV-ESS. First, the complex teacher forecasting model in the cloud is compressed into a compact student model at the edge based on knowledge distillation. Quantile regression and kernel density estimation methods are used to obtain the conditional quantile and probability density function of the PV power. Following this, based on probabilistic forecasting, a local control method of ESS is proposed by constructing linearized chance constraints of the PV power fluctuations at the grid-connected point. Finally, the effectiveness of the proposed method is verified based on a typical PV dataset and the IEEE 33-node system. The results show that the proposed method can realize accurate forecasting of PV power under limited computational resources Furthermore, by smoothing PV power fluctuations through local control of the ESS, it reduces the average voltage deviation by 26.23 % and 17.88 % on different test days.