Data-Driven CVR with PV Coordination for MV and LV Distribution Systems

This paper proposes a data-driven method by utilizing residential smart meter data to acquire the aggregated load ZIP parameters. An appliance state identification method based on temporal convolutional network (TCN) and the multihead self-attention mechanism is proposed to identify the device states and then calculate the aggregated ZIP parameters. Considering the different characteristics of LV or MV networks to which the CVR is directly applied, two CVR methods with PV coordination are proposed for LV and MV distribution systems. For the CVR in the LV system, the real-time CVR is implemented with a house-level aggregated ZIP model. For the CVR implementation in the MV system, a two-stage data-driven robust CVR with PV coordination is proposed. The half-hour on-load tap changers (OLTC) and capacitor banks (CBs) are determined ahead via a data-driven Bayesian method. A numerical simulation is performed on a realistic LV distribution network in the UK and the IEEE 33-bus network to verify the effectiveness of the proposed methods.