Predictor Comparison

This script analyzes the predictive power of different sets of predictors on day-ahead electricity price forecasting

Contents

Import data and generate all predictors

Load pre-trained networks

Baseline performance with all predictors

Compute performance of best neural network with all predictors

Price, Load & Fuel

Do the hour, weekday, holiday and temperature variables provide any predictive power when the load is known?

Selected Predictors:
CurrentLoad          
PrevWeekSameHourLoad 
prevDaySameHourLoad  
prev24HrAveLoad      
PrevWeekSameHourPrice
prevDaySameHourPrice 
prev24HrAvePrice     
prevDayNGPrice       
prevWeekAveNGPrice   

Baseline error with 14 predictors: MAE = $5.21, MAPE = 6.57%
Error with selected  9 predictors: MAE = $5.69, MAPE = 7.11%

No Load Information

How accurate is the prediction if the load information is not known?

Selected Predictors:
DryBulb              
DewPoint             
Hour                 
Weekday              
IsWorkingDay         
PrevWeekSameHourPrice
prevDaySameHourPrice 
prev24HrAvePrice     
prevDayNGPrice       
prevWeekAveNGPrice   

Baseline error with 14 predictors: MAE = $5.21, MAPE = 6.57%
Error with selected 10 predictors: MAE = $5.98, MAPE = 7.31%

No Current Load Information

How accurate is the prediction if the real-time load is not known?

Selected Predictors:
DryBulb              
DewPoint             
Hour                 
Weekday              
IsWorkingDay         
PrevWeekSameHourLoad 
prevDaySameHourLoad  
prev24HrAveLoad      
PrevWeekSameHourPrice
prevDaySameHourPrice 
prev24HrAvePrice     
prevDayNGPrice       
prevWeekAveNGPrice   

Baseline error with 14 predictors: MAE = $5.21, MAPE = 6.57%
Error with selected 13 predictors: MAE = $5.70, MAPE = 7.15%