A Machine Learned Approach to Identify the Anomalies in Load Pattern of Pakistan

Abstract

The energy crisis of Pakistan is worsening day by day for different explicit and implicit reasons. Load pattern and its use in the studies of power system is a vital area. Here it is used for the identification and analysis of anomalies in the load pattern of Pakistan while using Support Vector Regressor Machine as a Machine Learning tool. The training phase has been provided with retrospective data of electrical load, temperature, and relative humidity so as to predict the future electrical load in the testing phase, based on the then data of temperature and relative humidity. Based on temperature, three groups of electrical load have been opted based on particle swarm optimization clustering namely moderate, cold and hot. The difference curve between the actual load and predicted load illustrated various anomalies in all of the three clusters. The high numbers of anomalies were found in the hot cluster whilst confirming the dependence of load pattern upon the weather parameters. The analysis of the difference curve between the patterns portrays to be deformed enormously. The analysis of the anomalies has also being carried out in terms of contextualized parameters.