Decentralized Methods for Multi-Agent Problems Over Networks
Abstract
Controlling the modern electricity grid is a complex problem. This complexity is further exacerbated by the following facts: i) grids are becoming larger, ii) the increased integration of renewable and other distributed intermittent sources of power are making the supply very variable, iii) the demand is growing and requiring the grid to be operated very close to capacity, and iv) there is a mandate to reduce spinning reserves to reduce losses and the environmental impact. The grid control problem is currently solved centrally where all the load data is communicated to a central server and then the computed power settings are communicated to the generators and VaR supplying units. Due to the increased integration of renewable sources with highly variable supply, there is a need to reduce the latency in the control decisions. Achieving the tight latency constraints in the control decisions would require massive computing capabilities; hence, a huge capital investment. However, the modern grid is smart. Each node of the grid has smart devices that are essentially computers with limited computational, storage and messaging capabilities; but, the computational power of these assets has never been fully utilized to solve grid-scale, computationally intensive problems. One can take advantage of these smart-devices on the grid in a highly distributed way, e.g., the processing power of the NIC (Network Interface Controller) cards in already existing devices in the grid can be used for computation and message passing without additional hardware cost. With these facts in mind, we propose to develop decentralized power flow optimization techniques that can solve the grid control problem in an economical and scalable manner. This approach would make a greater penetration possible while maintaining grid resiliency; it can also strengthen grid security and resiliency, for when computation is distributed to the network, rather than executed in a central server, there is no single point of attack or failure. Our proposed distributed approach to solving the optimal power flow problem leverages several important components: i) convex relaxations for the non-convex problems, ii) bound tightening, and cutting planes for strengthening the convex formulation, iii) a reformulation of the optimization problem as a consensus problem, and iv) a provably fast algorithm for solving the consensus problem. The main technical challenges in the proposed approach include designing a methodology for updating the convex formulation, i.e., bound tightening and adding cutting planes, in an online fashion while solving the problem in a distributed manner, and tracking the optimal solution when the load and supply change over time. We extend the consensus approach to the setting where the individual "node", instead of being a single bus, is a subnetwork, and the consensus needs to be maintained across the lines connecting the subnetworks. This is precisely the setting of a interconnection of micro-grids with very few lines connecting them. Thus, our proposed consensus method would result in a distributed method for controlling a network of micro-grids. The proposed decentralized OPF techniques when incorporated into the grid infrastructure can lead to significant energy saving and emission reduction. Successful completion of this project will enable a highly distributed, secure and operational solution to the grid optimization problem that will help accelerate the adoption of renewable sources. The proposed decentralized OPF methods can also be used to replace or augment traditional voltage/VAR control approaches, and since it will be a software-based technology running locally on the devices, it is cost effective compared to state-of-the-art centralized techniques used. This abstract is publicly releasable
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2018
- Source ID
- W911NF1710298
Entities
People
- Necdet Aybat
Organizations
- Army Contracting Command
- Pennsylvania State University
- United States Army