Short Term Missing Data Prediction of Power System Using Discrete Cosine Transform and Simple Continued Fraction Compressed Long Short Term Memory Network

Short term load forecasting/ missing data prediction is a crucial part of smart grid for its operation and control. Long Short Term Memory (LSTM) Network is a widely used technique for time series data prediction which encodes features in a compressed form. However, if LSTM is fed with compressed form of feature its prediction can be made more accurate for data with more length. In this research work, thus the LSTM is fed with data compressed in two stages by using Discrete Cosine Transform (DCT) and Simple Continuous Fraction. A power consumption dataset from PWD, West Bengal is used to evaluate the proposed technique. A comparative study with other existing methods has also been done.