Approach for forecasting smart customer demand with significant energy demand variability

Load forecasting in an emerging smart grid has become a challenging task. This paper presents an innovative approach to forecast highly variable smart customer load using smart meter energy consumption data. The smart meter data is systematically linearized by applying extended k-means clustering approach, smoothing the linearized load profiles and then linearizing the load profiles using Taylor series linearization process. Case studies are presented using real world smart meter data and then applying the proposed approach and artificial neural network. Four different scenarios are considered for forecasting and the results showed that, in case of high variability in smart customer energy demand, the accuracy of forecasting using linearized profiles is higher than using original non-linear profiles as the source of forecasting. The forecasting process was repeated several times to verify the robustness of the approach and the results justify the accuracy of the forecast further with the proposed approach.