Deep Learning Networks for Short Term Load Forecasting

Accurate forecasting of electrical loads is essential for a power system to operate reliably and economically. Load forecasting is necessary for optimal power generation. Accurate load forecasting results in efficient and economic utilization of electrical infrastructure. However, forecasting load is difficult because load changes regularly and is affected by a number of other factors like temperature, pressure, humidity, season, etc. The amount of data collected at various points of a smart grid is of high volume, variety, and veracity and have the characteristics of big data. Deep learning models are extensively used in short term load forecasting in order to manage the high volume of data. This paper focuses on application of deep learning model to forecast electrical load using Long Short Term Memory (LSTM) and Recurrent Neural Network (RNN). The evaluation metrics R-Squared (R2), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) are used to compare these models. The Dayton Power and Light Company’s estimated energy consumption in Megawatts (MW) is the data set used to implement the deep learning models for short-term load forecasting. Adam optimization using TensorFlow framework is used in developing these models. The developed model is suitable for load forecasting in real time data also.