Long term load forecast thorugh artificial neural network and different forecasting methods: Zonguldak case

In this thesis, three-year long term electric energy load forecasting of Zonguldak province was conducted by using Regression, Back-propagation Artificial Neural Networks (BANN) and Radial Based Artificial Neural Networks (RBANN) forecasting methods. Temperature (℃), population, total square meters of partially completed buildings or added structures, gross domestic product per capita, number of employees in mining and quarrying business, total electricity consumption per capita (kWh) were used as independent variables in all models developed by forecasting methods. In this context, 6 different models per year and 2 different models per month, a total of 8 different models were established and analysed by forecasting methods. In regression analysis, dependent and independent variables including the years 2007-2017, were selected with multilinear and single regression methods in all models. In BANN method, %70 of the values between 2007 and 2017 years were randomly selected and used as training data in the models developed on MATLAB programme. In RBANN method, one or two hidden layers were used. On the other hand in BANN method, %70 of the values between 2007-2017 years were used as training data in the models developed on MATLAB programme. R² (quadratic error) method was used as the forecasting performance model in the trained models. The three-year energy consumption forecast data, found by forecasting methods, were compared among themselves. As a result of these comparisons, the highest R² value of 0.9983 was obtained with RBANN and MHY1 models as the best estimation result.