Machine Learning Approach Electric Appliance Consumption and Peak Demand Forecasting of Residential Customers Using Smart Meter Data

Abstract

Electric consumption forecasting using smart meter dataset is one of the aspects in which machine learning approach is highly applied. Forecasting peak demand and electric appliance consumption requires detailed analysis of smart meter data through classification and clustering methods. Forecasting of electrical appliance and Peak demand is necessary action and a significant part in electric power system planning and development. However, due to variability of household consumption level demand and appliance consumption demand, deep and detail analysis of customers’ smart meter data is required in order to identify critical attributes and the source of variation between the consumption level of appliance, as well as customers demand. This paper focuses on forecasting levels of electric appliance consumption and peak demand with the life style of residential customer’s using data obtained from Irish and Umass repository. Further on, customers life style is analyzed from the results of customer peak demand forecast. Supervised and unsupervised machine learning algorithm called CLARA clustering, support vector machine (SVM) and artificial neural network are applied as in order to achieve forecast the appliance consumption level and peak demand. Mean electric appliance consumption values are calculated from daily, weekly, monthly and total consumption for each appliance from 1 year smart data of 1 min time interval for electric appliance consumption forecasting of individual households. For the customers’ peak demand consumption, only mean of weekly consumption of aggregated households is computed together. The forecasting of customers electric consumption using SVM provides outcome of 99.6% accuracy which is much better than the previous works in the same field of study. The obtained result shows that the implemented methodologies and algorithms are applied at their best level of performance.