Addressing hybrid uncertainties in energy management: A synergistic Planner-Learner approach

The rapidly increasing penetration of renewable generation and the diversification of residential loads have brought significant challenges to home energy management (HEM). Traditional HEM approaches suffer from high risk, excessive conservatism, model dependency, and dimensionality issues, making them fail to handle heterogeneous uncertainties. To overcome these issues, this study proposes Planner-Learner, a day-ahead HEM approach addressing compound uncertainties arising from environment and user-behavior. Environmental uncertainties are modeled via kernel density estimation and a Copula-based joint distribution, then managed by a model-based Planner enhanced with a Negative SoftMax mechanism that dynamically updates scenario probabilities. User-behavior uncertainties are assigned to a reinforcement learning-based Learner with pruned state and action spaces via explicit uncertainty classification to mitigate the curse of dimensionality. Planner and Learner alternate in a closed-loop process: the solution of Planner initializes the Learner, whose refined policy is fed back to update the Planner until convergence. Case studies based on real-world data demonstrate that the proposed approach reduces electricity costs by up to 17.24%, nearly eliminates thermal and travel comfort constraint violations, maintains energy storage system reliability, and minimizes curtailment of renewable energy source, outperforming benchmark approaches.