A Wavelet Transform-Based Hybrid Short Term Load Forecasting Method for Managing the Costs of EV Charging Stations and Parking Lots

Subject Areas : Power

Yashar Khanchoupani ¹ , Mojtaba Beiraghi ² , Reza Ghanizadeh ³

1 - Department of electrical engineering, Islamic Azad University, Urmia Branch, Urmia, Iran
2 - Department of electrical engineering, Islamic Azad University, Urmia Branch, Urmia, Iran
3 - Department of electrical engineering, Islamic Azad University, Urmia Branch, Urmia, Iran

Received: 2023-07-20 Accepted : 2023-11-18 Published : 2023-12-01

Keywords: performance, RBF network, RLS Algorithm, Short Term Load Forecasting, hybrid learning,

Abstract :

Advanced data processing methods are emerging on a daily basis and they can be used in many tasks. Short-term load forecasting (STLF) is an essential task for power distribution systems. Specially, for the electric vehicle (EV) charging stations and parking lots the STLF can give precise time tables in which the electrical power demand is at the lowest and the electric prices are low too. In this paper, we developed a STLF system that works based on wavelet transform of the load data collected from the previous 3 months. Next, we perform the feature selection using the Gram-Schmidt (GS) procedure. The selected features are then fed to the Radial basis function (RBF) network in order to predict the power usage one day ahead. The learning algorithm that is used for the RBF network is the hybrid "k-means, RLS" algorithm. It is indicated that the proposed RBF network works more accurately or at least equal to the previously presented support vector machine (SVM) predictor. Also, the computational complexity of the RBF network is much less than SVM and consequently, the time consumption of the presented system is far less than recently proposed methods. The numerical simulation results based on the real-world load data of the Urmia city showed that the proposed method based on the wavelet transform features and the hybrid RBF is more efficient than the previous SVM both in accuracy (more than 29%) and computational complexity (more than 30%).