From Dense to Sparse: Event Response for Enhanced Residential Load Forecasting

TL;DR

This paper addresses residential load forecasting (RLF) by leveraging event-related sparse knowledge that resides in appliance usage patterns, which is overlooked by traditional dense-data approaches. It introduces ERKG, a two-part framework: a Multivariate State Predictor (MSP) that forecasts future appliance operational states to capture electricity usage events, and a knowledge-guided mechanism that uses event-response signals to regularize RLF training as a plug-in module. The method demonstrates robust improvements across three public datasets and three strong baselines, with average MAE gains around 8–9% and notable gains in longer-horizon forecasts. By focusing learning on event-related regularities and down-weighting noise, ERKG provides practical enhancements to existing forecasting systems and can be integrated with diverse RLF models without altering inference. The work offers a path toward more accurate, appliance-aware forecasting and suggests avenues for extending the approach with covariates, unlabeled event modeling, and efficiency improvements.

Abstract

Residential load forecasting (RLF) is crucial for resource scheduling in power systems. Most existing methods utilize all given load records (dense data) to indiscriminately extract the dependencies between historical and future time series. However, there exist important regular patterns residing in the event-related associations among different appliances (sparse knowledge), which have yet been ignored. In this paper, we propose an Event-Response Knowledge Guided approach (ERKG) for RLF by incorporating the estimation of electricity usage events for different appliances, mining event-related sparse knowledge from the load series. With ERKG, the event-response estimation enables portraying the electricity consumption behaviors of residents, revealing regular variations in appliance operational states. To be specific, ERKG consists of knowledge extraction and guidance: i) a forecasting model is designed for the electricity usage events by estimating appliance operational states, aiming to extract the event-related sparse knowledge; ii) a novel knowledge-guided mechanism is established by fusing such state estimates of the appliance events into the RLF model, which can give particular focuses on the patterns of users' electricity consumption behaviors. Notably, ERKG can flexibly serve as a plug-in module to boost the capability of existing forecasting models by leveraging event response. In numerical experiments, extensive comparisons and ablation studies have verified the effectiveness of our ERKG, e.g., over 8% MAE can be reduced on the tested state-of-the-art forecasting models.

Figures (6)

• Figure 1: The illustration of the proposed approach ERKG: Learning event-related sparse knowledge from dense data, enabling model focus on the electricity consumption behaviors in reality, i.e., regular and uncertain patterns. Specifically, (A): The original dense load series consist of regular and uncertain patterns, as well as noise, which impact predictive performances. (B): By learning the intrinsic relationship between future electricity usage events and historical load records, ERKG extracts sparse knowledge, i.e., event-related sparse knowledge, which is represented by class probabilities of appliance operational states. (C): The use of sparse knowledge guides the prediction model to focus on regular and uncertain patterns, reducing noise fitting. In practice, we utilize appliance operational states class probabilities as weights to regularize the training loss of the prediction model.
• Figure 2: Two learning paradigms. (A): Existing methods directly predict future loads using historical load values. (B): Our event forecasting method estimates appliance operational states using historical load values and models the state changes of appliances to learn event-related sparse knowledge.
• Figure 3: Left Penal: The paradigm of existing methods, e.g., Crossformerzhang2022crossformer, ETSformerwoo2022etsformer and TSMixerekambaram2023tsmixer. They directly model series value-level relationship between history and future. Right Penal: The overview of the proposed ERKG Approach. It can enhance the performance of existing methods through the estimation of electricity usage events for different appliances
• Figure 4: The proposed electricity usage events forecasting model.
• Figure 5: Prediction results on UK-Daleh1: We used four prediction sequences to demonstrate the enhanced load forecasting capability of our method under both changing and unchanging appliance operational states.