Weather Sensitive Residential Load Forecasting Using Neural Networks

Load forecasting is essential for power system planning and operation. Power utility relay on accurate load forecasting to commit generation, schedule outages, and predict potential operational risks. Short-term day-ahead and hour-ahead load forecasting are indispensable functions in energy management, marketing, and distribution systems. Improving load forecasting accuracy is challenging and crucial for power system planning and operation. This paper proposed a load forecasting model using a neural network. The data for the forecast model training was obtained from multiple data sources, including smart meters, week patterns, and weather information. The employed approach and the training model can be adapted to accurately predict the day-ahead and hour-ahead load consumption. In the neural network-based machine learning process, the prediction of the previous time point will be the input of the following forecast to improve the accuracy. The experimental results demonstrated the effectiveness of the proposed method on daily, day-ahead, and hour-ahead load forecasting.