Machine Learning Infused Distributed Optimization for Coordinating Virtual Power Plant Assets

TL;DR

The paper tackles the scalability and privacy limitations of centralized DER coordination in Virtual Power Plants by introducing LOOP-MAC, a machine-learning augmented distributed optimization framework. LOOP-MAC replaces the ADMM primal-dual updates with agent-specific neural approximators that predict optimal local decisions while enforcing hard local constraints via a gauge map, and it employs look-ahead recurrent training to handle temporal dependencies. The approach demonstrates up to 500x faster per-iteration performance and faster convergence than traditional ADMM, while maintaining feasibility with respect to coupled constraints, in a three-agent VPP case study. This method enables real-time, privacy-preserving coordination of diverse DERs and can significantly accelerate look-ahead dispatch in wholesale energy markets, aligning with FERC Order 2222 goals and practical real-time operation needs.

Abstract

Amid the increasing interest in the deployment of Distributed Energy Resources (DERs), the Virtual Power Plant (VPP) has emerged as a pivotal tool for aggregating diverse DERs and facilitating their participation in wholesale energy markets. These VPP deployments have been fueled by the Federal Energy Regulatory Commission's Order 2222, which makes DERs and VPPs competitive across market segments. However, the diversity and decentralized nature of DERs present significant challenges to the scalable coordination of VPP assets. To address efficiency and speed bottlenecks, this paper presents a novel machine learning-assisted distributed optimization to coordinate VPP assets. Our method, named LOOP-MAC(Learning to Optimize the Optimization Process for Multi-agent Coordination), adopts a multi-agent coordination perspective where each VPP agent manages multiple DERs and utilizes neural network approximators to expedite the solution search. The LOOP-MAC method employs a gauge map to guarantee strict compliance with local constraints, effectively reducing the need for additional post-processing steps. Our results highlight the advantages of LOOP-MAC, showcasing accelerated solution times per iteration and significantly reduced convergence times. The LOOP-MAC method outperforms conventional centralized and distributed optimization methods in optimization tasks that require repetitive and sequential execution.

Figures (8)

• Figure 1: Examples of agents controlled by a VPP.
• Figure 2: Comparison between ADMM and the proposed $\mathcal{LOOP-MAC}$ method. The ADMM approach is comprised of two key components: the dual update and the primal optimization. The dual update guides the consensus protocol, while the primal optimization leverages this estimation to adjust the local decision-making process through iterative solvers. In contrast, our $\mathcal{LOOP-MAC}$ method replaces these two procedures with a single-step data infusion and mapping input parameters and other agents' values to agent-level optimized values, directly.
• Figure 3: The structure and building blocks of Neural Approximator $\xi^i$.
• Figure 4: ADMM convergence values present a time-series prediction challenge, with outputs from one iteration feeding into the next. Our training approach uses a look-ahead format, enabling recurrent joint training of neural approximators, integrating prior outputs as current inputs.
• Figure 5: Our proposed $\mathcal{LOOP-MAC}$ method to coordinate DERs.