A Gradient Boosting Classifier to Predict Electric Power Consumption by Nuclear Power Plant

A nuclear power plant (NPP) purchases electricity from the integrated energy system to meet all of its demands during operation. On the basis of anticipated values, electric power is purchased. To adequately meet plant demands, it is crucial to predict the consumption of electricity as precisely as possible. The more exact the forecast, the better the financial outcome. The dataset utilized in this research is a time series of power consumption values in Megawatts (MW) which ranges from 203.2208 MW to 216.4605 MW and 744 samples were collected in total from the observations of electric power consumption (hourly) by NPPs. The entire data needs to be preprocessed and some Gradient Boosting approaches including Stochastic Gradient Boosting (SGBoosting), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) were applied while maintaining the 70%-30% training-testing ratio of the dataset. XGBoost classifier demonstrates 99.81% ac-curacy for predicting power consumption (with 0.982 Coefficient of Determination, 0.0044643 Mean absolute percentage error, and 0.0668 root mean square error), which is better than other models (SGBoost and LightGBM). This approach will be evaluated later with more electric power consumption data by nuclear power plant.