Work

An interpretable multi-objective adaptive optimization method for frequency regulation capability demand interval prediction

Year: 2026

Type: article

Abstract: ABSTRACT This paper proposes an interpretable, multi-objective, adaptive interval prediction method tailored for dispatch decision-making, to address the challenges of poor economic performance in traditional interval prediction methods, and the inflexibility and lack of interpretability in existing probabilistic models for planning frequency regulation reserves in high-penetration renewable power systems. First, ridge regression is used to make preliminary estimates of frequency regulation capacity and mileage. Next, a volatility-aware adaptive optimiser is designed to dynamically adjust intervals by minimising total interval width, thereby enhancing reserve cost-effectiveness directly. Case studies demonstrate that the proposed method achieves 100% coverage at the 95% confidence level, reducing the average interval width by 41.46%-44.06% compared to mainstream benchmark models. This improvement is statistically significant (p-value <0.01). The methodology reveals the influence mechanisms of key features through end-to-end interpretability analysis, validating the role of high coverage in ensuring system frequency stability via simulation. This paper provides dispatchers with a reliable, economical and transparent decision-support tool for reserve management.

Source: Results in Engineering

DOI: 10.1016/j.rineng.2026.109715

Authors: Xin Yan, Xiaobin Zhao, Naiwei Tu

Institution: Liaoning Technical University

Language: English

Topic: Electric Power System Optimization

Subfield: Electrical and Electronic Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $1,940