Hybrid Models Based on Error Correction to Predict Thermoelectric Power Plants Fuel Consumption

Abstract

Monitoring and controlling operating parameters in thermoelectric power plants (TPPs) is a relevant activity to achieve the optimal functioning of the electric system. Due to system complexity, the adjustment of operating parameters is often performed based on technical team know-how and a simplified analytical model, which can result in a suboptimal regime. Therefore, time series forecasting methods are promising alternatives to describe the observed behavior of energy generation, fuel consumption, and greenhouse gas emissions. For this purpose, different linear and non-linear models are applied to predict the next minute of fuel consumption related to an engine used in a Brazilian TPP powered by diesel and heavy fuel oil (HFO). This work uses error correction-based approaches that address linear models combined with Artificial Neural Networks (ANN) to predict the time series. The combination of ARIMA optimized by Particle Swarm Optimization (PSO) and Multilayer Perceptron (MLP) reduced the error rate by 32.97% when compared with the best linear model. The hybrid model developed in this work can be used as an auxiliary tool for maintenance and operation planning of TPPs engines when a short-term horizon is considered.