AN APPROACH TO CONTROL DEMAND RESPONSE LOAD SHEDDING BASED ON NEURAL NETWORK LOAD FORECASTING

Abstract

This paper describes a Neural Network application to reduce the computational complexities in a smart grid environment by predicting total demand and price for the next operating day. Short term load and price forecasting significantly reduce shortage risks and facilitates costs saving, hence improving the reliability and efficiency of Electrical Service Providers as well as efficiency for consumers. Investigation of the development of electricity price and demand forecasting, with the emergence of demand response programs is required to design this system,. Short Term Load/Price Forecasting (STL/PF) is done for an electricity market that offers Demand Response (DR) Programs. For forecasting, we builtt a robust prediction model using Long Short Time Memory (LSTM) to ensure the generality of the results. The Australian National Electricity Market (ANEM), specifically Queensland State, is used as a subject case study. The objective of the model is forecast total electrical load and price and used it to maximize the utility of the consumer subject to a minimum daily energy usage and minimum energy cost based on maximum and minimum hourly demand levels. Unknown price and electric load is forecast through recurrent neural network with a confidence interval. A simple bidirectional communication device between the power supplier and the consumer enables the functionality of the proposed model. The demand response affects the electricity price while the price affects the demands. Through demand response, the consumer can use electricity efficiently while the supplier can provide power at a lower cost.