SqueezeNet for the forecasting of the energy demand using a combined version of the sewing training-based optimization algorithm

Noradin Ghadimi; Elnazossadat Yasoubi; Ehsan Akbari; Mohammad Hosein Sabzalian; Hamzah Ali Alkhazaleh; Mojtaba Ghadamyari

doi:10.1016/j.heliyon.2023.e16827

SqueezeNet for the forecasting of the energy demand using a combined version of the sewing training-based optimization algorithm

Heliyon. 2023 Jun 3;9(6):e16827. doi: 10.1016/j.heliyon.2023.e16827. eCollection 2023 Jun.

Authors

Noradin Ghadimi^{1

2}, Elnazossadat Yasoubi³, Ehsan Akbari⁴, Mohammad Hosein Sabzalian⁵, Hamzah Ali Alkhazaleh⁶, Mojtaba Ghadamyari⁷

Affiliations

¹ Young Researchers and Elite Club, Ardabil Branch, Islamic Azad University, Ardabil, Iran.
² College of Technical Engineering, The Islamic University, Najaf, Iraq.
³ Department of Electrical Engineering, West Tehran Branch, Islamic Azad University, Tehran, Iran.
⁴ Department of Electrical Engineering, Mazandaran University of Science and Technology, Babol, Iran.
⁵ Smart Grid Laboratory (LabREI), Department of Systems and Energy, School of Electrical and Computer Engineering (FEEC), University of Campinas (UNICAMP), Campinas, Brazil.
⁶ College of Engineering and IT, University of Dubai, Academic City, 14143, Dubai, United Arab Emirates.
⁷ Department of Electrical Engineering, Shahid Beheshti University, 48512 Tehran, Tehran, Iran.

Abstract

With the introduction of various loads and dispersed production units to the system in recent years, the significance of precise forecasting for short, long, and medium loads have already been recognized. It is important to analyze the power system's performance in real-time and the appropriate response to changes in the electric load to make the best use of energy systems. Electric load forecasting for a long period in the time domain enables energy producers to increase grid stability, reduce equipment failures and production unit outages, and guarantee the dependability of electricity output. In this study, SqueezeNet is first used to obtain the required power demand forecast at the user end. The structure of the SqueezeNet is then enhanced using a customized version of the Sewing Training-Based Optimizer. A comparison between the results of the suggested method and those of some other published techniques is then implemented after the method has been applied to a typical case study with three different types of demands-short, long, and medium-term. A time window has been set up to collect the objective and input data from the customer at intervals of 20 min, allowing for highly effective neural network training. The results showed that the proposed method with 0.48, 0.49, and 0.53 MSE for Forecasting the short-term, medium-term, and long-term electricity provided the best results with the highest accuracy. The outcomes show that employing the suggested technique is a viable option for energy consumption forecasting.

Keywords: Combined sewing training-based optimization; Energy demand forecasting; SqueezeNet.