Short-term Forecasting of Non-Conforming Net Load Using a Fusion Model with Machine Learning and Deep Learning Methods

Short-term forecasting of non-conforming net load (STFNL) plays a vital role for operating a power system in secure and efficient manner. However, power system load consumption is affected by a variety of external factors and thus includes high levels of volatilities. These volatilities cause STFNL to be a challenging task and inaccurate as more distributed energy resources (DERs) continue to integrate into the power grid. Estimating the average hourly locational distribution of system loads becomes a constant daily challenge to transmission system operators as more non-visible DERs are connected to the distribution system. This paper proposes two commonly used machine-learning and deep learning methods used for load forecasting, i.e., the ensemble bagged and the long short-term memory neural network method. The advantages, features and applications of these methods are used to propose a fusion forecasting model that improves the forecasting accuracy. Additionally, data engineering and preprocessing options are used to increase the accuracy of the proposed model. A comparative study based on real-world transmission grid net load data is performed to verify that the proposed methodology is capable of reaching a relatively higher forecasting accuracy with lower error indices.