Using Graph Convolutional Network and Message Passing Neural Networks for Solving Unit Commitment and Economic Dispatch in a day ahead Energy Trading Market based on ERCOT Nodal Model.

Abstract

Various machine learning applications will pre-process graphical representations into a vector of real values which in turn loses information regarding graph structure. Graph Neural Networks (GNNs) are a combination of an information diffusion mechanism and neural networks, which represent a set of transition functions and a set of output functions. Graph Convolution Network (GCN) is based on the optimized variant of CNN which operates on graph and is a scalable approach for semi-supervised learning on structured graph data. Message Passing Neural Networks (MPNNs) summaries the cohesions between many of the existing Neural Network models for structured graph data. This thesis proves the viability of semi-supervised learning GCN model and supervised learning MPNNs to solve the crucial problems like the Unit Commitment (UC) and Economic Dispatch (ED) for the energy market. Power System Optimizer (PSO), a MILP based solution which simulates energy market accurately, but is extremely reluctant to scale in both time and compute. This thesis aims at representing the complex structure of the energy network using GNN and training the models to simulate the market with increased flexibility to scale in time and compute