Nonlinear Autoregressive Neural Network and SARIMA Model for Forecasting Peak Electricity Demand in the Kurdistan Region

With the great economic and social development witnessed by the Kurdistan Region of Iraq, the demand for electrical energy has increased significantly, causing an imbalance as the current generation and distribution infrastructure struggles to keep pace. The researchers aim to study and address this issue by forecast peak demand to support future planning in the sector to ensure efficient electricity supply during peak times and avoid overloading the network, using artificial neural networks methods that are characterized by their ability to learn and adapt to complex data and traditional methods (Box-Jenkins method) known for their accuracy in analysis. We will analyze the time series data of monthly peak electricity demand for ten years in the Kurdistan Region (January 2014 to July 2024) with a total of 127 observations. The results of the Box-Jenkins method identified the SARIMA (1,0,1)(0,1,1)12 model as the most suitable for time series analysis, as it showed high predictive accuracy and outperformed other models in terms of RMSE, AIC and SBIC for forecasting electricity demand increase, While the results of the nonlinear autoregressive neural network (NARNN) model, which was structured by adjusting the hidden layer neurons and delay numbers through trial and error, showed that the optimal NARNN model (1:12,10) achieved the lowest RMSE, MAE, MAPE, and R values when compared to the other models tested. When comparing the performance of the NAR neural network with the SARIMA method, the SARIMA (1,0,1)(0,1,1)12 model showed superiority over the NAR neural network in terms of accuracy, making it the better choice for forecasting peak electricity consumption and working to reduce the gap between demand and production in the future.