Performance Comparison of Deep Learning Models for Short Term Load Forecasting

Abstract

In today's power industry, the ever-changing load environment makes accurate load forecasts even more important than ever. Even, day ahead load forecasting is crucial for the successful and efficient operation of the grid. With the absence of good reliable forecasts of daily load fluctuations, systems within the grid can become overloaded, resulting in a failure of the system, thus establishing the need for sound forecasting methods. As a solution to this challenge, different types of deep learning algorithms namely, Long Short-Term Memory, Bidirectional Long Short-Term Memory, Convolutional Neural Network based LSTM, Convolutional Neural Network based BiLSTM are applied for short-term load forecasting in this paper. The efficacy of the various deep learning techniques for short term load prediction is compared. The technique which produces the least value of root mean squared error between real and estimated load, is the most preferred one. The comparative analysis shows that Convolutional Neural Network based BiLSTM technique is a better option for short term load forecasting.