Optimal Scheduling of Isolated Microgrids Using Automated Reinforcement Learning-based Multi-period Forecasting (2108.06764v1)

Abstract: In order to reduce the negative impact of the uncertainty of load and renewable energies outputs on microgrid operation, an optimal scheduling model is proposed for isolated microgrids by using automated reinforcement learning-based multi-period forecasting of renewable power generations and loads. Firstly, a prioritized experience replay automated reinforcement learning (PER-AutoRL) is designed to simplify the deployment of deep reinforcement learning (DRL)-based forecasting model in a customized manner, the single-step multi-period forecasting method based on PER-AutoRL is proposed for the first time to address the error accumulation issue suffered by existing multi-step forecasting methods, then the prediction values obtained by the proposed forecasting method are revised via the error distribution to improve the prediction accuracy; secondly, a scheduling model considering demand response is constructed to minimize the total microgrid operating costs, where the revised forecasting values are used as the dispatch basis, and a spinning reserve chance constraint is set according to the error distribution; finally, by transforming the original scheduling model into a readily solvable mixed integer linear programming via the sequence operation theory (SOT), the transformed model is solved by using CPLEX solver. The simulation results show that compared with the traditional scheduling model without forecasting, this approach manages to significantly reduce the system operating costs by improving the prediction accuracy.

• The paper presents a novel automated reinforcement learning (AutoRL) framework to improve forecasting and optimize scheduling in isolated microgrids, aiming to reduce operating costs under renewable and load uncertainty.
• A key innovation is the PER-AutoRL model, which automates the selection of optimal deep reinforcement learning architectures and hyperparameters for enhanced multi-period single-step forecasting.
• Empirical results demonstrate that the proposed AutoRL-based approach significantly reduces microgrid operating costs compared to conventional methods by effectively handling the stochastic nature of renewable energy and load.

Optimal Scheduling of Isolated Microgrids with Automated Reinforcement Learning

The paper presents a rigorous approach to optimizing the scheduling of isolated microgrids using an automated reinforcement learning (AutoRL) framework. This research addresses the inherent uncertainty in load and renewable energy outputs that challenge effective microgrid operations. The primary contribution is leveraging an automated reinforcement learning-based multi-period forecasting model to enhance the prediction accuracy of renewable power generation and load variability, thus improving microgrid economic operations.

Key Innovations

1. PER-AutoRL Model: A prioritized experience replay automated reinforcement learning (PER-AutoRL) model is introduced. Unlike conventional machine learning models, this framework can independently determine optimal model architectures and hyperparameters, streamlining the complex process of deploying deep reinforcement learning (DRL) models.
2. Multi-period Single-step Forecasting: The paper proposes a novel single-step multi-period forecasting approach, a significant departure from traditional multi-step methods known for error accumulation over successive periods. By treating each forecasted period independently, the presented model alleviates cumulative prediction errors, thus increasing reliability.
3. Error Distribution and Revision: A t location-scale (TLS) distribution models the uncertainty of forecasting errors. By accounting for the probabilistic nature of these errors, forecasting results are revised to minimize systemic biases, fortifying the reliability of the dispatching basis.
4. Integrated Scheduling Model: The paper constructs a scheduling model integrating demand response and the revised forecasts, structured to minimize operating costs. Incorporating sequence operation theory (SOT), the model translates into a mixed-integer linear program solvable by optimization solvers such as the CPLEX.

Empirical Analysis

The empirical results presented demonstrate that the adoption of the PER-AutoRL framework significantly reduces system operating costs when compared to conventional scheduling models sans forecasting. Simulation outcomes indicate that the model effectively handles the stochastic nature of renewable outputs and load demands, underscoring the efficacy of accurate prediction methods in reducing microgrid economic burdens.

Discussion and Implications

While the research provides robust evidence supporting the proposed method’s benefits, it also highlights the intricate balance between forecast accuracy and the computational feasibility of large-scale deployments. The proposed method's potential application in broader integrated energy systems, tackling energy management issues, and enhancing operational flexibility under uncertainty, marks exciting future directions.

This paper contributes substantially to the literature on microgrid operation optimization under uncertainty conditions, offering a compelling case for integrating advanced AI techniques into energy management practices. As AI technologies continue to evolve, the methods discussed hold promise for more adaptable and efficient microgrid systems and broader implications for sustainable energy integration in power systems globally.