Optimization and forecasting algorithms for converter dominated distribution networks using blockchain and AI

Abstract

Integration of power electronic converter-based distributed energy resources (DERs) in electric power distribution networks is growing exponentially with the recent interest in reducing carbon emissions from fossil fuel-based generation. As the contribution of renewable energy sources in the DER mix continues to increase, so does the incorporation of battery energy storage systems and other controllable loads to compensate for the high variability and uncertainty in the generation from renewable DERs and grid demand. Strategies for increasing the contribution of renewable energy sources and using reserves to accommodate for variations and uncertainty in generation and load include distributed optimal power flow (OPF) methods and improved forecasting. This work proposes a co-optimization of power flow and flexibility reserves, executed on a private blockchain for security, solved using a parameterized deterministic method based on semi-distributed architecture and alternating direction method of multipliers (ADMM) based distributed architecture that addresses uncertainty and enhances the flexibility of the distribution network. However, ADMM guarantees convergence only for strictly convex problems and hence a relax-and-fix heuristic algorithm is proposed in co-ordination with ADMM to solve the OPF problem, which is non-convex in nature. Also, an accurate short-term load forecasting algorithm is essential to reduce the uncertainty in the dispatch results using the OPF algorithm. In this work, a short-term residential load forecasting algorithm is proposed using a two-stage stacked long short-term memory network-based recurrent neural network and Hampel filter to address this issue. All the proposed algorithms are tested using different case studies. Results demonstrate that the proposed algorithms reduce the impact of uncertainty in the distribution network, automate scheduling flexibility reserve and minimize its cost, reduce the OPF execution time using a distributed architecture, and produce residential load forecast with a significantly lower prediction error.