Smart Grid Consumer Behavioral Model using Machine Learning

The data set of the power consumption embeds useful information about the consumer behavior. The behavioral models identify the consumer groups that have similar behavior. These behavioral models help the utility for numerous applications. Though the data set is same, different applications extract different behavioral models with the help of relevant features. The rationale for the choice of relevant features is described. Here, two applications are considered - demand response and rational tariff design. Clustering of voluminous data becomes a challenge due to its volatile nature. For dimensionality reduction, a representative curve is proposed which imbibes all the features that are considered for the analysis. Seasonal variations are also taken into account. Machine learning algorithms like k-means, expectation maximization and self-organizing maps are applied. The proposed method is tested on a practical system with 789 consumers. A performance index is used to evaluate various algorithms. The results obtained are discussed.