Scenario-Based Hybrid Differential Evolution for Local Energy Community Scheduling: A Multi-Objective Synthetic Benchmark with Shared Storage
Year: 2026
Type: article
Abstract: This study compares multi-objective search configurations for day-ahead scheduling of photovoltaic generation, shiftable loads, and a shared battery in a synthetic local energy community. The objectives are mean scenario cost, mean cost in the worst 10% of scenarios (empirical Conditional Value-at-Risk (CvaR) at confidence 0.90), and relative dispersion of a source-defined household score. That score credits exported net load at both retail and feed-in prices; it is a proxy, not household billing or an allocation of battery benefits. Scenario-based Hybrid Differential Evolution (S-HDE) combines Generalized Differential Evolution 3 (GDE3)-style selection, a scenario curriculum, and periodic battery pattern search. Ten 20-household instances, each with 30 scenarios, are compared across six configurations and five seeds (300 reported runs), with a common ceiling of 4,400 schedule evaluations but unequal attained counts. Non-dominated Sorting Genetic Algorithm II (NSGA-II) has the best reported mean Hypervolume (HV) (0.2407 ± 0.0382; mean rank 1.30), followed by S-HDE (0.2298 ± 0.0261; rank 2.80). Printed per-instance means reproduce the Friedman result (χ² = 41.257, p = 8.32 × 10⁻⁸). Reported normal-approximation Wilcoxon comparisons favor NSGA-II over S-HDE (Holm-adjusted p = 0.02531); neither ablation differs detectably from S-HDE. The findings describe the reported in-sample benchmark. Original code, objective definitions, front archives, and evaluation ledgers are needed to verify computational provenance; corrected household billing would define a different optimization experiment.
Source: Systemic Analytics
DOI: 10.31181/sa442026100
Language: English
Topic: Smart Grid Energy Management
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7235221601
