ISELF - Load Forecasting
The forecast of electricity demands is an essential information for all operation stages of an electric power system. Aware of this necessity, Cepel's Department of Energy Optimization and the Environment (DEA) has been developing a set of models for load forecasting, ranging from real time to weekly load forecasting.
Iself model - a computational system for real time load forecasting.
This system was designed to work in real-time and provide forecasts up to 48 hours ahead, with a temporal resolution of 10, 15, 30 and 60 minutes, updated each 60 minutes. In order to achieve this result, the forecast method adopted involves the use of three distinct methods: fuzzy logic, polynomial neural networks (GMDH) and spline interpolation. The forecast system also has a module for automatic treatment of historical load (filter) time series data, with the aim of preparing data for forecast model adjustment, filling data gaps or correcting outliers and discontinuities present in load records. The ISELF program provides:
• Automatic treatment of load records, including outlier removal and filling of data gaps (FIgure 1).
• Hourly load forecasts up to 48 hours ahead with temporal resolution of 10, 15, 30 and 60 minutes (hourly updated) (FIgures 2 and 3)
Daily peak load and minimum load forecasts up to three days ahead Computational system designed to work in real-time.
Figure 1. Raw load data and fiited data after the removal of outliers and gap filling
FIgure 2. load forecast
FIgure 3. load forecast
PrevCargaDESSEM –daily load forecasting with half-hourly time resolution up to one week ahead.
The hourly/half-hourly load forecasts up to 7 days ahead are fundamental information for determination of the unit commitment dispatch. In this case, the load forecasts should be updated on a daily basis. In this sense, in order to automate the daily load forecasting, Cepel developed PrevcargaDESSEM, an R package with functions that are able to automatic perform load/temperature data cleansing and load forecasting based on Support Vector Machine (SVM) technique. The load forecasts depends on the load past values, temperature forecasts and calendar variables (weekday, holydays, months, .,,).
PrevCargaPMO - weekly and monthly load forecasting up to two months ahead.
The weekly/monthly load forecasts up to 2 months ahead are fundamental information for the hydrothermal operation planning. In this case, the load forecasts should be updated on a weekly basis. In this sense, in order to automate the weekly load forecasting, Cepel developed PrevcargaPMO, an R package with functions that are able to perform automatic load/temperature data cleansing and load forecasting, also based on SVM technique. The load forecasts depend on the load past values, temperature forecasts and calendar variables (weekday, holidays, months, etc.).
Contact the responsible area via email: