A Data Filtration and Deep Learning Approach for Electrical Load Forecasting

The advent of AI and Machine Learning has optimized electrical load forecasting which is an important aspect in the power sector for proper planning and maintenance of power systems. Accurate load prediction is generally challenging due to the fact that the load used by the end user lies completely at the discretion of the user but still it is possible to get fair estimates of the average load conditions using surveys and prediction mechanisms. Any information related to pattern to be followed by connected Electrical Load will helps any electric utility organization to make important decisions regarding purchasing and generating electric power, unit commitment decisions, load switching, reduce spinning reserve capacity and infrastructure development. Hence load forecasting is viewed as field of research to develop a model so that efficient and reliable operation of power system could be carried out. The paper also presents a short summary of previous techniques pertaining to the adopted methodology. The proposed algorithm uses the discrete wavelet transform for data pre-processing i.e. removing sudden spikes or irregularities in the electrical load data. The regression back propagation algorithm is used for the training purpose. It is found that the proposed system attains a Mean Absolute Percentage Error (MAPE) of around 2.4%.