Predicting Thermoelectric Power Plants Diesel/Heavy Fuel Oil Engine Fuel Consumption Using Univariate Forecasting and XGBoost Machine Learning Models

Monitoring and controlling thermoelectric power plants (TPPs) operational parameters have become essential to ensure system reliability, especially in emergencies. Due complexity, operating control is often performed based on technical know-how simplified analytical models that can result limited observations. An alternative this task using time series forecasting methods seek generalize characteristics past information. However, the analysis of these techniques large diesel/HFO engines used Brazilian under dispatch regime has not yet been well-explored. Therefore, given complex engine fuel consumption during generation, work aimed investigate patterns generalization abilities when linear nonlinear univariate are a representative database related an engine-driven generator TPP located Pernambuco, Brazil. Fuel predictions artificial neural networks were directly compared XGBoost regressor adaptation perform as with lower computational cost. AR ARIMA applied benchmark, PSO optimizer was model adjustment. In summary, it possible observe ARIMA-PSO had similar performances operations error distributions full-load output normal frequency distribution −0.03 ± 3.55 0.03 3.78 kg/h, respectively. Despite their similarities, achieved better adherence capturing load adjustment periods. On other hand, approaches NAR showed significantly performance, achieving mean absolute reductions 42.37% 30.30%, respectively, best model. modeling 8.7 times computationally faster than training. The at disturbances ramp, shut-down, sudden fluctuations steps, despite being inferior full-load, due its high sensitivity slight variations.