英語翻譯
Several simulations were performed to forecast the hourly
Ontario Energy Price using Matlab neural network and fuzzy
logic tools and considering predispatch demand,capacity short-
falls,outages,and imports as inputs.
The length of the training sets are 14 or 28 days because the
demand has a seasonal period of about three weeks and to avoid
overtraining during the learning process that may occur with
longer sets,while the checking and testing sets are both 1 day,
each day consisting of 24 sets of data.Therefore,the output is
a vector of 24 elements or hours.
The results presented in this paper were obtained using his-
torical data from May 1 to September 22,and their performance
was measured mainly using a mean absolute percentage error
(MAPS) defined by the following expression: