PeakFocus: Bridging Peak Localization and Intensity Regression via a Unified Multi-Scale Framework for Electricity Load Forecasting

Electricity load peak forecasting (ELPF), simultaneously predicting peak timing and intensity, is a prerequisite for effective grid scheduling and risk management. However, existing methods face three limitations. First, they adopt a two-stage predict-then-locate paradigm, which severs the link between temporal localization and intensity regression. Second, they still struggle with the multi-scale representation conflict, leading to peak misjudgment and timing misalignment. Third, the lack of explicit peak timing context during intensity regression causes intensity smoothing because predictions are dominated by global smoothing trends. To address these limitations, we propose PeakFocus, a unified framework for ELPF. (i) A Unified Peak-Aware Pipeline (UPAP) utilizes a triple hybrid loss to jointly supervise temporal localization and intensity regression, alongside a tolerance-based evaluation protocol. (ii) A Multi-Scale Mixing Peak Locator (MSM-PL) exploits coarse-grained features to mitigate peak misjudgment caused by local fluctuations, and injects them into fine-grained features via a cascade mechanism to resolve timing misalignment. (iii) A Location-Aware Decoder (LAD) injects peak timing context into the intensity regression process, providing explicit guidance to counteract intensity smoothing and improve peak intensity estimation. Extensive experiments on the public Electricity (ELC) dataset and our industrial-scale World Large-scale Electricity Load (WLEL) dataset show that PeakFocus outperforms baselines in both timing precision and intensity estimation.

한 줄 요약

**[Load Forecasting / Multi-Scale]** PeakFocus가 전력 부하 peak 시점·강도 동시 예측 — UPAP triple hybrid loss·MSM-PL multi-scale peak locator·LAD location-aware decoder가 기존 predict-then-locate 패러다임의 단절·multi-scale 충돌·intensity smoothing 해결.

핵심 기여도

전력 부하 peak forecasting(ELPF)이 peak 시점과 강도를 동시 예측해 효과적 grid scheduling·risk management의 prerequisite지만 기존 방법에 3 한계가 있음을 진단 — (1) two-stage predict-then-locate 패러다임이 temporal localization과 intensity regression 사이 link 단절, (2) multi-scale representation 충돌로 peak 오판·timing 오정렬, (3) intensity regression 중 explicit peak timing context 부재로 예측이 global smoothing trend에 dominated되며 intensity smoothing 발생.
PeakFocus 제안 — ELPF 통합 프레임워크.
Unified Peak-Aware Pipeline(UPAP)이 triple hybrid loss를 활용해 temporal localization·intensity regression을 jointly supervise, tolerance-based 평가 protocol 동반.
Multi-Scale Mixing Peak Locator(MSM-PL)가 coarse-grained feature로 local 변동에 의한 peak 오판 완화·cascade 메커니즘으로 fine-grained feature에 주입해 timing 오정렬 해결; Location-Aware Decoder(LAD)가 peak timing context를 intensity regression에 주입해 explicit 가이드로 intensity smoothing 상쇄·peak intensity 추정 개선.

핵심 아이디어

전력 부하 peak forecasting의 세 본질적 한계(단절된 paradigm·multi-scale 충돌·intensity smoothing)는 localization·regression의 joint supervision, multi-scale cascade, location-aware decoding의 통합 디자인으로 동시 해결 가능하다.

기술적 접근법

**방법론**: PeakFocus — UPAP + MSM-PL + LAD 통합 ELPF 프레임워크.
**핵심 기법**: (1) UPAP의 triple hybrid loss로 localization·regression joint supervision·tolerance 기반 평가, (2) MSM-PL의 coarse feature가 local 변동 완화·cascade로 fine feature에 주입해 timing 오정렬 해결, (3) LAD가 peak timing context를 intensity regression에 주입해 smoothing 상쇄, (4) Public ELC 데이터셋·산업급 WLEL 데이터셋 실험, (5) Timing precision·intensity 추정 양면 평가.

주요 결과

Public Electricity(ELC) 데이터셋에서 baseline 능가.
산업급 WLEL(World Large-scale Electricity Load) 데이터셋에서 baseline 능가.
Timing precision·intensity 추정 양면에서 일관 개선.
Predict-then-locate 패러다임 대비 unified pipeline의 효과 정량 입증.

의의 및 한계

**의의**: ELPF의 unified 프레임워크 제시, 3 한계 동시 해결의 명확한 design rationale, 산업급 데이터셋 검증, peak-aware loss·decoder 패턴의 일반화 가능성. **한계**: 전력 부하 도메인 중심으로 다른 시계열 peak 예측 일반화 추가 검증, WLEL 데이터 공개 여부, 매우 long-horizon forecasting의 cascade 깊이 효과는 후속.

실용적 활용

전력 grid 운영·peak shaving·수급 계획.
일반 시계열 peak·extreme event 예측.
Energy management·demand response 시스템.