Electricity demand forecasting based on feature extraction and optimized backpropagation neural network

• Published in: e-Prime Vol. 6; p. 100293
• Main Authors: Ofori-Ntow Jnr, Eric, Ziggah, Yao Yevenyo
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-12-2023; Elsevier
• Subjects: Backpropagation neural network; Electricity demand forecasting; Hybrid model; Particle swarm optimization; Electricity demand forecasting; Hybrid model; Backpropagation neural network; Particle swarm optimization
• ISSN: 2772-6711; 2772-6711
• DOI: 10.1016/j.prime.2023.100293

•A new hybrid model including feature selection technique, PSO, and BPNN is developed for forecasting electricity demand data.•The originality of the data is preserved after the feature selection as compared to what is in the literature.•The comparison stage involves seven (7) other models which include hybrid and standalone models.•The general stability of the proposed model throughout the training, testing, and validation stages is established.
As the global population is growing at a high rate, so is the electricity demand also increasing at a faster rate. This exerts pressure on electricity-generating plants and maintenance engineers because of the variability in demand. Avoiding disruption in the supply to meet demand requires forecasting what the future of demand will look like to be able to plan adequately towards it. This study, therefore, develops a new forecasting model using feature extraction (FE) where statistical information of the hourly demand data is extracted which serves as input variables for Backpropagation neural network (BPNN) optimized by particle swarm optimization (PSO) for electricity demand forecasting in Ghana. The model known as FE-PSO-BPNN is compared to other seven models such as Radial Basis Function (RBFNN), Random Forest (RF), Gradient Boosting Machine (GBM), Multivariate Adaptive Regression Splines (MARS), BPNN, and PSO-RBFNN where FE selects the input variables for all models. Electricity demand data from Ghana Grid Company from the period including 1st September 2018 to 30th November 2019 is used for the testing of the model's performance. Evaluation criteria such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Scatter Index (SI) were used. The proposed model is more powerful in forecasting electricity demand than the others as it has RMSE (0.5344), MAE (3.3845), MAPE (0.1773), and SI (0.0003). The model is expected to be a better option for electricity sector managers when considering demand forecasting.