COMPARATIVE PERFORMANCE EVALUATION OF DEMAND AND PREDICTION ALGORITHMS

A brief description is given of the algorithms selected together with the procedures followed in the experiment. A simplified algorithm is proposed and tested and the results discussed. Of the second generation algorithms examined, it was found that the historical average algorithm gave better results than UTCS-2 or UTCS-3 for five minute predictions. In cycle-by-cycle predictions, the moving average version of the simplified algorithm performs better than UTCS-3, and the complete version performed better in all cases. However, the complete version may need to be updated frequently and it needs data collected on at least one previous day, making more than one equation necessary for each day, and a minimum of three equations for each peak and off-peak period. The main advantage of the simplified algorithm is that it does not need extensive historical data, and it could easily be optimised using established control theory methodologies. The authors feel that the emphasis should not be in exact statistical treatment of the data, and that the demand prediction should not be treated as a point process without using more valuable information presently ignored. (TRRL)