Products and Solutions

AGORA Load Forecast

New Forecast Methodology Most load forecast software is based on one of two methodologies: ARIMA (Auto-Regressive Integrated Moving Average) or FFNN (Feed Forward Neural Networks). While ARIMA methods can provide good forecast results in the short term, constant maintenance is required and it is difficult to incorporate a priori knowledge into the forecast. FFNN uses dynamic regression to evolve a forecast, but requires frequent re-training and resists the incorporation of new load tendencies into the forecast.

Mixed-Technology Forecast Methodology for Power System Demand Improves Forecast Results, Reduces Maintenance Requirements

The AGORA Load Forecast software utilizes two different methods concurrently to address these issues and improve forecast results.	Forecast Display Screen
AGORA Load Forecast – Combines a phase-space projection method and a symbolic forecasting method into one software system, overcoming the shortcomings of previously-used forecast methods. The AGORA Load Forecast: · Takes into account normal and special days, as well as exceptions  · Doesn’t require time-consuming and costly re-training of a neural network. · Incorporates a priori knowledge · Runs as a standalone application on a PC, or can be integrated into EMS applications	State-Space Representation of Historical Data
Phase-Space Projection – The first of the two forecasting methods used by AGORA reduces all of the known information about historical load records into a two-dimensional plot, or state-space. Each data point in this plot represents a value of load, temperature, time of day, and type of day. A bottle-neck neural network is used to forecast load based on this historical data. The reliability of the forecast can be determined, based on the local point density. Forecast error can be determined by path divergence and convergence.
	Symbolic Forecasting – The second forecasting methodology uses specific  information about type of day and exceptions to modify the forecast.  Day of the week (weekday vs. weekend, Monday vs. Friday) day of the month,  and month of the year are defined to adjust the forecast based on seasonality and type of load day. Exceptions can be defined so that holidays or other  untypical load days don’t skew the forecast results.