Integration of Artificial Neural Networks and Noise Filtering for Forecasting Annual Electricity Loads

A. Ghanbari and S.F. Ghaderi (Iran)


Noise Filtering, Artificial Neural Networks, Forecasting, Model Selection


Nowadays Artificial Intelligence (AI) based techniques are becoming more and more widespread to solve complicated practical problems in various fields, because of their flexibility, symbolic reasoning, and explanation capabilities. Meanwhile, electricity demand forecasting is known as one of the most important challenges in managing supply and demand of electricity. In this study we present an intelligent multistage procedure for forecasting annual electricity loads. Firstly, we collect the most influential factors and apply noise filtering on our raw data. At the next stage we employ feature selection technique to select the best model out of collected factors. Subsequently selected filtered factors will be fed into Artificial Neural Networks (ANN) in order to build annual load forecasting model (called Filtered-ANN). For investigating and validating influence of noise filtering on improvement of ANN, we compare results of the integrated approach with Simple-ANN model which is built using raw data. We carry out the comparisons by means of paired t-test to check if the improvement is statistically significant or not. Results show that the Filtered-ANN significantly outperforms Simple-ANN and we may consider noise filtering as an effective concept to be integrated with AI approaches such as ANN in order to improve accuracy of forecasts.

Important Links:

Go Back