A novel method for jointly monitoring household flexibility resources considering multiple correlations

Abstract

Although demand response is vital for managing distribution networks with substantial renewable energy, accurately monitoring behind-the-meter flexibility resources remains challenging for developing precise household response plans. To address this, we propose a novel method for jointly monitoring household flexibility resources. The unified monitoring framework comprehensively explores the correlations between environmental factors and flexibility resources, as well as among these resources. It better addresses the complex usage behaviors of adjustable appliances and the high variability of rooftop photovoltaics while reducing computational complexity in monitoring multiple resources. The proposed multi-task learning model integrates a synergistic mechanism of shared feature extraction and task-specific adaptation. It first employs a transformer-based module to capture cross-task temporal features that integrate electrical and environmental correlations. A gated network then adaptively selects relevant features for each resource, which are processed by tower networks to capture long-term patterns and short-term variations. Additionally, we introduce quantile values of flexibility resources as monitoring targets to guide the model in learning power distribution, enabling more precise demand response plans. Experiments demonstrate that our method outperforms existing methods in monitoring various flexibility resources within and across households on public datasets. Furthermore, ablation experiments and model complexity analysis highlight the effectiveness of our method.