Two-Stage Machine Learning Framework for Simultaneous Forecasting of Price-Load in the Smart Grid

Abstract

In this paper, the electricity load and price patterns of consumers are forecasted using a two-stage forecasting framework. The electricity usage statistics of the consumers are recorded through smart meters and based on the historical load and price patterns the proposed model forecasts the future loads and prices used for further bidding purposes. A hybrid two stage forecasting framework combining the variational mode decomposition (VMD) method, echo state neural network (ESNN) and differential evolution (DE) algorithm is proposed. The training of the hybrid forecasting framework is done by decomposing the load and price time-series data using the VMD. The decomposed data are then used for training the ESNN. Differential evolution algorithm is used to tune the ESNN. Initially, the price and load data are used separately to train the ESNN, and in the second stage, both the data are used along with the forecasted output of the previous stage are used to train the ESNN. The proposed forecasting framework is experimented on 3 smart gird data derived from Smart Meter Energy Consumption Data in London Households of UK Power Networks (UKPN), for demonstration purpose.