Genetic k-means-algorithm-based classification of direct load-control curves

To improve the load factors of a power system, direct load control (DLC) of air conditioners is one effective approach adopted by utilities. However, the performance of a DLC programme should be evaluated periodically to validate the preferential tariff provided to the customers involved. Prior to evaluating the performance, the DLC curves recorded should be categorised properly. The only further factor evaluated is the category of the DLC curves that comply with the required control pattern. Therefore, developing efficient approaches for classifying DLC curves is needed. In the paper, with the aim of reducing the number of variables for classification and enhancing the classification effectiveness, autoregression moving-averaging (ARMA) modelling techniques are employed to extract the features of the DLC curves. To ensure the adequacy of the ARMA models used for the DLC curves, the Akaike information criterion is assessed. Based on the features extracted, the genetic k-means algorithm is then adopted for classification owing to its ability to partition given global data optimally into a specified number of clusters. Through the proposed approaches, categories are derived of the DLC curves complying and noncomplying with the control pattern. The results obtained from the comparisons with the artificial-neural-network approach show that the clusters divided using the proposed approach exhibit very high classification rates for the practical data on Taiwan Power Company DLC programmes.