Essays on Residential Electricity Consumption and Heuristics to Solve the Unit Commitment Problem

Abstract

This dissertation is divided into three chapters. The first chapter presents an analysis of residential electricity consumption in Mexico, focusing on identifying socio-economic and geographical factors that affect electricity use and associated expenditures in Mexican households. This research aims to estimate the electricity consumption and received subsidies for each Mexican household surveyed in the National Survey on Energy Consumption in Private Homes (ENCEVI) and the National Household Income and Expenditure Survey (ENIGH) in 2018. We expand these two datasets into a comprehensive source of electricity consumption data for 100,000 households representing all of Mexico. The analysis also uses information from the Mexican Utility Federal Electricity Commission (CFE, Comisión Federal de Electricidad) to merge these two datasets. The estimations from this dataset suggest that large consumers receive large subsidies, with the top 20% households with the highest electricity consumption receiving 50% of the subsidies. We also find that geography andsocio-economic status determine electricity consumption. Households in the extremely hot region consume 51.5% ˘ 4.5% more than those in the lowest income decile. These factors reveal that Mexicans in the highest income brackets benefit the most from subsidies; the average subsidy per household for those in the two highest income brackets is almost twice the average subsidy for those in the lowest bracket.The second and third chapters focus on developing metaheuristic methods to solve the Deterministic and Stochastic Unit Commitment Problem (UCP) faster without compromising solution accuracy. The unit commitment problem (UCP) is an optimization problem that determines the optimal operation schedule of electric power generating units while meeting load and operational constraints. The computational complexity of this Mixed-Integer Linear Programming (MILP) algorithm presents significant challenges for traditional optimization techniques. This research presents a novel method for improving the computational efficiency of the UCP without compromising its accuracy. The approach leverages the fact that the UCP is routinely solved repeatedly for similar instances, which enables "learning" from historical solutions. In a two-step data-driven process, we first identify patterns in past UCP solutions by analyzing the entropy of the commitment binary variables and then use these patterns to fix commitment variables in new problems. The results of an SLA application to a system with the same power generation felt as the PJM system show that the method can find a UCP solution of the same quality in half the time. The method uses entropy as a metric to quantify the persistence of the commitment status of power plants, making it transparent and comprehensible to power system operators. The SLA method is a novel approach to improving the computational efficiency of the UCP, and it has the potential to enhance the efficiency and scalability of UCP optimization methods. The findings of this research are expected to contribute to a more efficient integration of renewable assets and enhance grid reliability.