Perception Method of Voltage Spatial-Temporal Distribution for EV Enriched Distribution Network: A Priority Adaptive DNN Enhancement Approach

Abstract

With rapid growth of electric vehicles (EVs) and the integration of large-scale distributed energy sources in distribution networks (DNs), their stochastic and disorderly integration presents a significant challenge to real-time voltage perception. Accordingly, this paper puts forth a priority adaptive deep neural network (DNN) enhancement approach for voltage real-time perception, with the objective of establishing the spatial-temporal mapping relationship between bus voltage of DN with renewable energy, load, and EV data. Firstly, a DNN model is proposed as a solution to the problem of inaccurate voltage perception in EV-enriched DNs caused by the limitations of fuzzy power flow models. Then, to address the issue of the deep learning (DL) method's inability to achieve precise results in extreme circumstances, a power flow model is integrated with the DL method through iteration to support the voltage perception model in making well-informed decisions. And, in order to tackle the difficulties presented by large-scale, complex, non-convex, non-linear problems on resource-constrained devices, a mix precision DNN quantization method is proposed to enable the real-time, high-precision sensing of bus voltages in EV-rich DNs. Finally, the effectiveness of the proposed method is demonstrated by a 141-bus DN test case.