An Application of Prosumer Electric Load Forecasting with Machine Learning-Based Algorithms

Latterly, the concept of prosumer has been commonly used with the ascending penetration of renewable energy sources into current power distribution networks. The consumer term in electricity markets has been evolved into prosumer which can be defined as an individual who consumes and produces electricity. Considering the rising demand for energy, the research of prosumer electric load forecasting has become crucial for maintaining the stability of distribution networks. The prevalent utilisation of machine learning-based methods in electric load forecasting studies demonstrates the efficacy of these approaches in the field of forecasting. This study aims to forecast quarter hour-ahead prosumer net metering consumption using a household dataset from Austin, Texas, USA with the employment of gated recurrent unit (GRU) networks and extreme gradient boosted decision trees (XGBoost) algorithms. Consequently, the results of the study demonstrated that utilising XGBoost algorithm outperformed the GRU networks algorithm for prosumer electric load forecasting in terms of coefficient of determination (R²) and root mean squared error (RMSE) by 90.36% and 0.498 kWh, respectively. Furthermore, it is considered that literature for prosumer electric load forecasting has not been adequately articulated and this paper will not only provide guidance for the young researchers in the field, but also bridge this gap.