DAILY LOAD FORECASTING WITH HOURLY TEMPERATURES
Load forecasting at the daily resolution is generally performed by aggregating the predicted hourly load. Long-term daily load forecasting is essential for resource planning in the power system and price evaluation of energy contracts. Short-term daily load forecasting is required for balancing operation of the grid and trading strategies of the day-ahead energy market. Another relevance of daily load forecasting is the disaggregation of the monthly consumption data into daily load curves to determine the supplier obligation. Most of the published study on daily load forecasting is focused on daily peak load forecasting. While the other two modules of daily load, that are, the daily energy and daily minimum load have not been covered extensively. In this research, we have modeled hourly temperatures of a day to forecast daily load directly. The study delves into the hourly temperature data to find the best subsets that influence the daily load using the daily load series. This kind of study on multi-frequency series is unique. The study also finds that the daily load is strongly influenced by human activity pattern and hence, temperatures of specific hours of a day are more significant than the highest or the lowest temperature of the day. The proposed model uses Multiple Linear Regression (MLR) technique to model the methodology on two real case studies. The study also employs two MLR based benchmark models; one is the hourly load forecasting model that frames the recency effect of temperatures using the big data approach. The aggregation of predicted hourly loads gives the daily load forecasts. The other benchmark is a direct daily load forecasting model that is based on Tao’s vanilla model using maximum and minimum temperatures of a day. Since daily load forecasting finds its application in long-term as well as short-term, the proposed model is evaluated for a year ahead, as well as, one day ahead forecasting. The research empirically demonstrates that the proposed model, using the groups of hourly temperatures and daily load series, performs reasonably well in comparison to benchmarks models in ex-post forecasting while in ex-ante forecasting the proposed methodology emerges out to be the most robust model.