(61q) Management of Multi-Microgrid System with 2D CNN Forecasting Model and End-Effect Mitigation Using Value Function

Microgrid (MG) systems are becoming popular as they facilitate the incorporation of variable renewable energy (VRE) sources into local energy systems with low carbon emissions. However, the intermittent and uncertain nature of these sources poses significant challenges in the management as they result in highly irregular and unpredictable electricity generation [1]. To address these challenges, this study proposes the development of machine learning-based forecasting models for VRE sources and electricity price considering their nonstationary and periodic trending behavior. These models can help in effectively managing grids and storage devices. For this, we propose a framework for the optimal management of a multi-microgrid system mitigating the end-effect.

Accurate forecasting of VRE sources and electricity price is crucial for maximizing the economic benefits in MG systems through operational decisions such as electricity production schedules, transaction, and storage [2, 3]. With the rise of electricity markets, accurate forecasting has become increasingly important, and numerous methods have been developed in recent years. Especially, deep learning methods such as recurrent neural network (RNN) and convolutional neural network (CNN) have attracted the attention of many energy industries and academic entities. These data-driven forecasting models share a fundamental assumption that the future trends will have similar statistical patterns to the historical data [3]. However, previous studies have not considered the cyclic nature of the hourly and daily variations of VRE sources and electricity price. This study proposes a novel approach to make multi-day-ahead forecasts from past data using a 2D CNN model and its variant (combined with RNN) that can handle the hourly and daily correlations simultaneously.

Multi-microgrid (MMG) system has emerged as another promising solution to alleviate the challenges caused by uncertainties as it entails flexibility to trade energy between individual MGs [4, 5]. However, excessive installation of electric cable connections across the MMG network can be prohibitively expensive and is often not justified [6]. This study proposes a two-stage stochastic programming (2SSP) approach as a tool to balance a tradeoff between design- and operation-level decisions for MMG under stochastic uncertainty. It considers the capital cost of installing electric cables and the operating cost of the MMG. In the 2SSP, the choice of optimization horizon gets limited to a short, finite length due to the rapidly increasing problem size given the multi-scale nature of the problem. This limitation on the horizon length manifests as the storage being drained empty at the end of each horizon, known as the end-effect, which has a negative impact on the long-term management of the energy system and storage maintenance. To mitigate the end-effect, we propose several realistic valuation methods for the terminal energy in the storage at the end of a horizon. These include a benchmark method using a parameter based on the forecasted electricity price during the following period and an advanced method based on the value function approximation using reinforcement learning (RL) that can be combined with the 2SSP formulation. The 2SSP model is integrated with Markov Decision Process (MDP) through the value of terminal energy level in the storage at the end of each day to mitigate the end-effect and account for the longer-term value of stored electricity. This integration enables the value function to be recursively updated using observations sampled from the 2D CNN-based forecasting model and 2SSP decision-making model [7]. The proposed approach is compared with conventional methods used in previous studies and the benchmark method based on the forecasted electricity price. The results indicate that the proposed framework can ensure cost-effective and reliable energy supply from MMG systems by mitigating the end-effect.

References

[1] M. T. Kelley, R. Baldick, and M. Baldea, “Demand response scheduling under uncertainty: Chance-constrained framework and application to an air separation unit,” AIChE Journal, vol. 66, no. 9, pp. e16273, 2020.

[2] B. Uniejewski, J. Nowotarskim, and R. Weron, “Automated variable selection and shrinkage for day-ahead electricity price forecasting,” Energies, vol. 9, no. 8, pp. 621, 2016.

[3] T. Hong, P. Pinson, Y. Wang, R. Weron, D. Yang, and H. Zareipour, “Energy forecasting: A review and outlook,” IEEE Open Access Journal of Power and Energy, vol. 7, pp. 376–388, 2020.

[4] I. Zenginis, J. S. Vardakas, C. Echave, M. Morato, J. Abadal, and C. V. Verikoukis, “Cooperation in microgrids through power exchange: An optimal sizing and operation approach,” Applied Energy, vol. 203, pp. 972-981, 2017.

[5] A. Jafari, H. G. Ganjehlou, T. Khalili, and A. Bidram, “A fair electricity market strategy for energy management and reliability enhancement of islanded multi-microgrids,” Applied Energy, vol. 270, pp. 115170, 2020.

[6] D. Han and J. H. Lee, “Two-stage stochastic programming formulation for optimal design and operation of multi-microgrid system using data-based modeling of renewable energy sources,” Applied Energy, vol. 291, pp. 116830, 2021.

[7] J. Shin, J. H. Lee, and M. J. Realff, “Operational planning and optimal sizing of microgrid considering multi-scale wind uncertainty,” Applied Energy, vol. 195, pp. 616-633, 2017.