In order to forecast 'balancing system marginal price (smf)' of Turkish electricity market, i used load, day-ahead price (sgof) and the difference between up-regulation volume and down-regulation volume in MWh (nth) as inputs. When i give lagged nth (previous day and previous week same hour nth) only,the forecasting performance is bad, however when i give lagged nth with current nth (i.e kth hour nth for predicting the kth hour smf) forecasting performance is really improved successively. The problem is that forecasting nth for a future hour is a challenging task and the performance of forecasting nth is not so good. I want to give a range of current nth instead of one value (maybe probability dist. of nth / monte carlo) in order to forecast a range of smf instead of one value (probability dist.of smf instead of one forecasting value). But i do not know how to create a range of possible nth values as an input of ANN in order to find forecasted smf ranges. Would you please help me about matlab codes? Below i use the script that is modified by your genpredictor script. I hope that I could describe the problem properly and I would be glad to discuss about it with you. Thank you for your contribution and look forward to hear from you.
I noticed for the genPredictors function, some of the inputs to the short term forecasting model is prevDaySameHourLoad and prev24HrAveLoad. It seems that the prev24HrAveLoad uses the current day load and averages it on a 24 hour basis and the prevDaySameHourLoad uses the load from the previous day. I m not sure how these 2 inputs can be used to predict the next day load forecast for the ANN model. For example if today is Sunday and I want to predict the load for tomorrow (Monday) and I m running the load forecast application today at 1 PM, I can't use prevDaySameHourLoad as an input because Sunday is the previous day for the forecasted day Monday and I wouldn't have the 24 hour historical data for Sunday as I m running the program at 1PM on Sunday, so data from 1 PM up to midnight is unknown. Second the prev24HrAveLoad input depends on the actual load data for that day, which for this example would be the load data that I m forecasting for the next day (Monday). I m not sure how these 2 inputs can be used to predict the day ahead load if it requires historical data that you might not have yet.
When running the forecaster, I find that each time I run the codes, I get a different forecast/model for the loads. Is there a way to guarantee a consistent forecast model or is this always the case given that the neural network may be ending in a different local minima. Thank you for your help.
could you please tell me what did you mean by daily5dayHighAve and daily5dayLowAve in the genpredictor function?
% Long Term Forecast Predictors
X = [data.DryBulb data.DewPnt daily5dayHighAve daily5dayLowAve data.Hour dayOfWeek isWorkingDay];