Distributed battery energy storage in modern power systems: Artificial-intelligence-driven optimization for utility-substation autonomy expansion and residual load flattening

High photovoltaic (PV) penetration in isolated power systems can create midday reverse power flows, renewable-energy curtailment, transformer/line loading concerns and steep evening ramping requirements. This paper proposes and validates a decentralized, forecast-in-the-loop operating framework for distributed Battery Energy Storage Systems (BESS) installed at HV/MV utility substations, through a Convex-Heuristic-Artificial Intelligence (Linear Programming-Genetic Algorithm-Deep Neural Networks) synergy and workflow. The objective is the optimum decentralized BESS dispatch profile on a multi-bus transmission system for mitigating grid imports and PV exports at substation level. Unlike generic feeder smoothing or cost-minimization studies, the proposed objective explicitly targets the daily near-zero net demand at the substation boundary, thereby contributing to the increase of local renewable self-consumption while indirectly flattening the residual load seen by conventional generators at system level. The framework is tested using real 30-min net-load data from 50 HV/MV substations of the islanded power system in Cyprus, with detailed assessment of five substations exhibiting the largest PV surplus. For the assumed 20 MW / 80 MWh BESS at each selected substation, annual import energy is reduced by approximately 15–60%, annual reverse-flow/export energy by approximately 60–90% and the aggregate daily mean ramp rate of conventional generation by about 16%. The results demonstrate that decentralized substation-level BESS operation can materially increase local demand autonomy and mitigate duck-curve effects, resulting to an easier adaptation of Distributed Renewable Generation to traditional and isolated power systems.