Forecasting Short Term Peak Loads of Distribution Transformer (DT) Using Machine Learning and Computational Statistics—Various Methodologies and Their Pros and Cons

Machine Learning (ML) is a technique that employs computational statistics to learn the pattern of data and from there it tends to predict. Over last few years, abundance of data have been made available in standard format and therefore it is compelling to use ML techniques to achieve better prediction. In several instances we have witnessed 98–99% accuracy in prediction levels using ML techniques. In Utility Industry, one of the key parameter every DISCOM would like to forecast is Peak Load. Since HT (>11 kVA) and EHT (Extra High Tension > 66 kVA) customers in the designated DISCOM region demand continuous power supply, reducing interruption hours and feeder outages are two most important factors draw attentions of Utility Service Providers. It has been found that accurate prediction of Peak Load Demand for a day ahead can remarkably reduce the above two parameters. In fact, knowing the peak load in advance can help in optimizing the load distribution at substation levels. While predicting peak load, we should consider weather (Temperature), time of the day (expressed in HH:MM format), Day type (Weekdays or Weekends) and finally kWH of a DT.