Enhanced Short Term Load Forecasting Using Artificial Variables

With the ongoing development in the field of smart meters, the electricity consumption data is readily available and can be forecasted as well as analyzed for better planning of grid such as load scheduling, demand-side management, etc. In this paper, machine learning algorithms are utilized to develop a hybrid model for forecasting electricity consumption for a shorter period. The development is made possible by combining the simple algorithms as children and one of the best suitable algorithm as a parent. The parent algorithm combines the output of all the children to generate the final forecast. This hybrid model is labeled as “Blended DKFD” (Decision tree, K-Nearest Neighbor, Feedforward neural network, and Deep feedforward neural network) model. This model not only retains the advantages of children algorithms but also eliminates their drawbacks. In addition to hybridization, the inclusion of intentionally processed variables called Artificial Variables elevates the accuracy of the prediction. The study presented in this article aims at simplification of the development of prediction models, such that any interested body can develop a model suitable for their application cheaply, which will be capable of high accuracy and reliability. The Blended DKFD model is tested on the IEEE 39-bus system for different datasets considering artificial variables, historical data, and environmental factors. Finally, from the simulation results, it has been proved that the proposed model exhibits a low standard deviation with high accuracy depending on the children's algorithms and various variables used to build prediction models.