Using ChebConv and B-Spline GNN models for Solving Unit Commitment and Economic Dispatch in a day ahead Energy Trading Market based on ERCOT Nodal Model

Abstract

Spectral Convolutions and B-Spline Graph Neural Network techniques have been used in past to learn embeddings in various complex, multidimensional structured knowledge graphs like genetics, social networks, geometric shapes and more. Spectral graphs provide a way to apply fast and localized filters on graph data. B-Spline kernels provides a way to keep the computation time independent by due to the local support property of B-spline basis functions. This thesis aims at using each of these models to test their viability for solving the Unit Commitment (UC) and Economic Dispatch (ED) problem for the energy market. There have been multiple attempts of solving the UC ED problem from linear regression to neural networks and complex mathematics models. Everyone has their own set of advantages and disadvantages. Currently, industry uses Power System Optimizer (PSO), a MILP based solution which is extremely precise, but is extremely reluctant to scale in both time and compute. Some models fail to precisely represent the complex structure of the networks. This thesis focuses to use Graph Neural Network (GNN) which gives us the ability to represent the complex structure of the energy network and learn the energy market and use tools that will help to scale the modes for larger datasets.