Application of artificial intelligence models to electricity consumption data

In this study, electricity consumption data of Bilecik province of Turkey for Lighting, Residential, Industrial, Agricultural Irrigation and Commercial sectors for the period of 2016 January - 2024 June are analysed. In order to estimate the future trends of electricity consumption, data are collected using the Electricity Market Monthly Official Statistics published by the Energy Market Regulatory Authority (EMRA) and processed by Time Series Analysis. RNN, GRU, LSTM and BiLSTM models were trained at Epoch 25, 50, 100 and 200 on a sectoral basis, and the period 2016-2019 was used as training data and the period 2020-2024 June was used as test data. The performance of the models was evaluated with MAE, RMSE, MSE and MAPE metrics. The findings of the study show that BiLSTM and LSTM models predict with high accuracy in Residential and Commercial sectors, while GRU and BiLSTM models are more successful in Industrial and Agricultural Irrigation sectors. The RNN model generally performed worse than the other models. The study revealed that deep learning models have a strong learning ability in electricity consumption forecasting, but more hyperparameter optimisation is needed in some sectors. As a result, these models have enabled the prediction of electricity consumption of Bilecik province in the coming years on a sectoral basis and provided important predictions in terms of energy management. For future studies, optimisation methods can be used in more complex and larger datasets to improve the performance of the models on a sectoral basis.