Application of Deep Sleep Optimizer for strategic placement and sizing of distributed generation in power distribution systems

Abstract

Strategic siting and sizing of renewable-dominated distributed generators (DGs) is pivotal to curbing technical losses and safeguarding voltage quality in modern distribution networks. This paper presents a fully scripted, direct Python-PowerFactory framework that eliminates the file-exchange overhead of conventional MATLAB co-simulation and embeds the recently conceived Deep-Sleep Optimizer (DSO) a meta-heuristic that alternates between ‘‘deep-sleep’’ global exploration and ‘‘wake’’ local exploitation. The proposed workflow is evaluated on the IEEE-33 radial feeder and rigorously benchmarked against the Genetic Algorithm (GA) and TeachingLearning-Based Optimization (TLBO). The multi-objective problem minimizes real-power losses and maximizes the minimum voltage-stability index subject to generator, voltage, and thermal constraints. For fair comparison all solvers use identical population sizes and iteration budgets. DSO reaches steady convergence in fewer than 20 iterations, whereas TLBO and GA require roughly 40 and 70 iterations, respectively. In the single-DG case DSO and TLBO each halve active losses (to ≈91 kW, −56%) and raise the weakest node to 0.983pu, outperforming GA’s 0.978pu. With two DGs, DSO yields the lowest loss (55.9 kW, −73%) using only 2.72 MW of capacity and completes in 528 s one third of TLBO’s runtime. The most demanding three-DG scenario underscores DSO’s superiority: losses plummet from 210 kW to 28.9 kW (−86%), the weakest-bus voltage climbs to 0.998pu, no bus exceeds 1.013pu, and the installed capacity (3.84 MW) is 30% lower than GA’s requirement while runtime (666 s) is 81 % shorter than TLBO’s. DSO also trims reactive losses by 81 % and produces the flattest voltage profile (standard deviation < 0.005 pu) without breaching the 0.95-1.05 pu statutory band. Because both DSO and TLBO are parameter-free, the framework removes the need for heuristic tuning, making it attractive for day-to-day planning. Overall, the results establish DSO as a high-performance, low-maintenance optimization engine that unlocks deeper loss cuts and tighter voltage regulation than established methods at a computational cost compatible with routine distribution-planning studies. Future work will extend the methodology to time-series optimization with stochastic photovoltaic and load profiles, meshed feeders, and full techno-economic assessment.