[Paper Review] Long-Term Energy Management for Microgrid with Hybrid Hydrogen-Battery Energy Storage: A Prediction-Free Coordinated Optimization Framework

Ning Qi, Kaidi Huang+2 more|arXiv (Cornell University)|Jul 31, 2024

Microgrid Control and Optimization4 citations

TL;DR

This paper proposes a prediction-free two-stage optimization framework for long-term microgrid energy management using hybrid hydrogen-battery storage. It uses kernel regression to update hydrogen SoC references online and an adaptive virtual-queue-based online convex optimization (OCO) algorithm with expert-tracking and penalty terms, achieving sublinear dynamic regret without prediction. The method reduces operational costs by ~30% and load loss by ~80% compared to existing approaches on Elia and North China datasets.

ABSTRACT

This paper studies the long-term energy management of a microgrid coordinating hybrid hydrogen-battery energy storage. We develop an approximate semi-empirical hydrogen storage model to accurately capture the power-dependent efficiency of hydrogen storage. We introduce a prediction-free two-stage coordinated optimization framework, which generates the annual state-of-charge (SoC) reference for hydrogen storage offline. During online operation, it updates the SoC reference online using kernel regression and makes operation decisions based on the proposed adaptive virtual-queue-based online convex optimization (OCO) algorithm. We innovatively incorporate penalty terms for long-term pattern tracking and expert-tracking for step size updates. We provide theoretical proof to show that the proposed OCO algorithm achieves a sublinear bound of dynamic regret without using prediction information. Numerical studies based on the Elia and North China datasets show that the proposed framework significantly outperforms the existing online optimization approaches by reducing the operational costs and loss of load by around 30% and 80%, respectively. These benefits can be further enhanced with optimized settings for the penalty coefficient and step size of OCO, as well as more historical references.

Motivation & Objective

• Address the challenge of long-term microgrid energy management under seasonal uncertainties and limited prediction availability.
• Develop a coordinated optimization framework that integrates hybrid hydrogen-battery energy storage for both short-term and seasonal balancing.
• Enable real-time operation without relying on forecast data by using historical and AI-generated scenarios for SoC reference generation.
• Ensure robustness and sublinear dynamic regret in online decision-making through adaptive virtual-queue and expert-tracking mechanisms.
• Achieve cost-effective and reliable microgrid operation across diverse climatic and load conditions using real-world datasets.

Proposed method

• Proposes an approximate semi-empirical hydrogen storage model that captures power-dependent efficiency in hydrogen systems.
• Introduces a two-stage framework: offline generation of hydrogen SoC reference using historical and AI-generated scenarios, followed by online adaptation via kernel regression.
• Employs an adaptive virtual-queue-based online convex optimization (OCO) algorithm with dynamic step size updates informed by expert-tracking and long-term pattern tracking penalties.
• Incorporates penalty terms for both long-term SoC pattern tracking and expert performance tracking to improve convergence and robustness.
• Theoretical analysis proves sublinear dynamic regret bound without prediction, leveraging convexity and bounded gradient assumptions.
• Uses a time-varying step size policy based on a weighted combination of expert performance and historical tracking error, ensuring stability and adaptability.

Figure 1: Schematic diagram of hydrogen storage system.

Experimental results

Research questions

• RQ1 Can a prediction-free online optimization framework effectively manage long-term microgrid operations with hybrid hydrogen-battery storage under seasonal variability?
• RQ2 How does the integration of kernel regression for SoC reference updating improve real-time adaptability compared to static or forecast-based methods?
• RQ3 To what extent does expert-tracking and pattern-penalty regularization enhance the dynamic regret performance of online convex optimization in microgrid dispatch?
• RQ4 What is the impact of penalty coefficient and step size tuning on operational cost and load loss reduction in real-world microgrid scenarios?
• RQ5 How does the proposed framework compare to existing online and offline optimization methods in terms of cost, reliability, and scalability on real datasets?

Key findings

• The proposed framework reduces annual operational costs by approximately 30% compared to existing online optimization methods on the Elia and North China datasets.
• Loss of load is reduced by around 80% compared to baseline methods, demonstrating significant improvement in supply reliability.
• The method achieves sublinear dynamic regret bound without using prediction, proven theoretically through convex analysis and bounded gradient assumptions.
• Numerical results show that performance gains are further enhanced with optimized penalty coefficients and step sizes, particularly when leveraging more historical and AI-generated reference scenarios.
• The kernel regression-based SoC reference update mechanism enables accurate online adaptation, outperforming MPC-based and static reference methods in dynamic and uncertain environments.
• On the North China dataset, the framework achieves a cost of $5.30×10⁵ and a load loss of 80.60 MWh, significantly outperforming M1 (cost: $11.74×10⁵, loss: 208.85 MWh) and M4 (cost: $23.24×10⁵, loss: 2104.74 MWh).

Figure 2: Approximation of alkaline electrolyzer properties at $90\,^{\circ}\mathrm{C}$ , 10 bar: (a) hydrogen production, (b) efficiency.