Data-Driven Computational Algorithms for Predicting Electricity Consumption Missing Values: A Comparative Study

Data availability is a significant issue and barrier for modeling and analyzing low voltage networks. This paper develops, implements, and compares several prediction algorithms for finding missing values in energy usage for commercial consumers. Four predictive machine learning models, such as random forest regression, linear regression, multi-layer perceptron, and decision trees, are utilized in this paper. Four commercial users from a regional city in Australia are selected as a dataset based on 30-minute intervals. Firstly, the obtained data is analyzed and pre-processed and then utilized for model training and testing. RMSE and MAE measures are used to compare the effectiveness of each machine learning model. This paper concludes that the multi-layer perceptron model provides better performance than that of random forest, decision tree, and linear regression. The RMSE and MAE of MLP model are 4.7472 and 4.2103, respectively, when using individual users as a training set.