Using dynamic linear models and Kalman filter for modeling and forecasting electricity load in Erbil City

الملخص EN

In this paper we constructed four models for hourly (4*24 models for the 24 hours are 96 models of the day) and daily (4*7 models for the 7 days are 7 models of the week) electricity load using dynamic linear models (DLM) with various parameters.

The Bayesian method and Kalman filter where used to estimate the parameters of four load models (Simple DLM, Trend DLM, Trend seasonal DLM and Regression DLM) and one step ahead forecasting.

In order to select the best and most efficient model for estimating and forecasting the electricity load in Erbil City, the four models were compared using (mean absolute error, mean absolute percentage error and root mean square error.

The results presented in this paper based on real measured data.

R-programming language and Microsoft Excel were used for data analyses.

The result shows that Regression-DLM has best estimation and forecasting results compared to other models using the accuracy criteria, and the Kalman filter algorithm is a well-established technique and is suitable for estimating the parameters of load models that are used in this work as dynamic linear equations that include load signal with uncorrelated Gaussian white noise.