Energy consumption prediction based on Facebook Prophet - Gated Recurrent Unit Method

Abstract

Electric energy consumption prediction is an essential and complex task in modern commercial medium industries. It plays a significant role in including the energy expenditure while drawing up the annual financial budget of the company. In this project, Electric energy consumption prediction model utilizing the combination of FbProphet and GRU has been developed. Raw data (from IGTF-dataset) of hourly energy consumption was collected from Igara Growers Tea Factory and refined using the MinMax scaler for effective training. Then a hybrid model with the integration of Fbprophet and a five-layer gated recurrent unit (GRU) was developed and trained exploiting the transfer learning technique. The trained model was evaluated using the root mean square error (RMSE), mean square error (MSE), and mean absolute error (MAE) metrics. The experimental results show that, in comparison with other existing models, this hybrid model provides more accurate results with relatively lower errors that is (RMSE = 0.0103), (MSE = 0.0001) and (MAE = 0.0076). This project will guide the industry in the proper allocation of resources by helping to estimate relatively the correct amount to be spent on power in the stipulated duration of time in the future up to one year period. Further work should be done on the prediction of Uganda’s future energy tariffs based on the available past quarterly tariff data as determined by the energy regulatory bodies so as to avoid the task of having to input the current tariff to predict future expenditure on electrical energy by the industry.