Short-term individual residential load forecasting using an enhanced machine learning-based approach based on a feature engineering framework: A comparative study with deep learning methods

Accurate short-term forecasting of the individual residential load is a challenging task due to the nonlinear behavior of the residential customer. Moreover, there are a large number of features that have impact on the energy consumption of the residential load. Recently, deep learning algorithms are widely used for short-term load forecasting (STLF) of residential load. Although deep learning algorithms are capable of achieving promising results due to their ability in feature extraction, machine learning algorithms are also prone to obtain satisfactory results with lower complexity and easier implementation. Identifying the most dominant features which have the highest impact on residential load is a pragmatic measure to boost the accuracy of STLF. But deep learning algorithms use feature extraction, which leads to the loss of data interpretability due to transforming the data. This paper proposes to improve the accuracy of the individual residential STLF using an enhanced machine learning-based approach via a feature-engineering framework. To this end, various datasets and features such as historical load and climate features are collected. Afterward, correlation analysis and outlier detection via the k nearest neighbor algorithm are deployed to implement outlier detection. In the next stage, feature selection algorithms are used to identify the foremost dominant features. Additionally, this paper conducts a comparative study between the proposed approach and state-of-the-art deep learning architectures. Eventually, the isolation forest algorithm is used to verify the effectiveness of the proposed approach by identifying anomalous samples and comparing the results of the proposed approach with those of deep learning algorithms.