Machine learning models in short term electricity consumption forecasting with data preprocessing techniques: The case of İzmir

In recent years, with the increase in human population, the demand for electrical energy has also increased. In addition to the increase in the demand for electrical energy, the gradual decrease in fossil resources used in electricity generation has increased the importance of accurate estimation of electricity consumption demand. In this study, short-term electricity consumption forecasting models are developed using hourly electricity consumption data of Izmir province between 01.01.2017 and 31.12.2021. Twelve different models have been created with artificial neural network models by using different feature extraction methods in order to increase the success rate. The results obtained from these models are compared with the ARIMA method, which is one of the traditional methods. As a result of the study, the actual values and the prediction value obtained from the model were compared in terms of success evaluation criteria such as MAPE, RMSE, MAE etc. and the model that gives the most accurate prediction result was tried to be determined. When the data obtained as a result of the research were analysed, it was determined that the best performance was achieved in the model in which variable mode decomposition method and entropy method were used together. According to the MAPE (Mean Absolute Percentage Error) success evaluation criterion, this model achieved a value of 0.4394 and showed a superior performance. In addition, the coefficient of determination (R²) value of 0.997 according to the success criterion shows that the model has a high accuracy rate. These results show that this method is more effective and reliable compared to other methods. These results are compared with the ARIMA model, which is one of the traditional methods. As a result of the study, the prediction value obtained from the model created with the actual values were compared in terms of success evaluation criteria such as MAPE, RMSE, MAE etc. and the model that gives the most accurate prediction result was tried to be determined. When the data obtained as a result of the research were analysed, it was determined that the best performance was achieved in the model in which variable mode decomposition method and entropy method were used together. This model showed a superior performance by reaching a value of 0.4394 according to the MAPE (Mean Absolute Percentage Error) success evaluation criterion. In addition, the coefficient of determination (R²) value of 0.997 according to the success criterion shows that the model has a high accuracy rate. These results show that this method is more effective and reliable compared to other methods. These results are compared with the ARIMA model, which is one of the traditional methods. In the load forecasting study using the ARIMA model, the MAPE value was found to be 23.13. This comparison clearly shows that the model using the variable mode decomposition method and entropy method together works with a much lower error rate and produces more accurate forecasts compared to the ARIMA model. This proves that modern methods are superior to traditional methods and provide more reliable results.