Abstract
Cumulative responses such as fruit growth and plant growth, as affected by environmental factors, during storage for identification were frequently obtained by repeating the experiments under different environmental conditions. In this case, it is not well-known how many data patterns are necessary for the acceptable identification of a system. This paper explores an effective way for identifying the cumulative responses, which are obtained from a model, using neural networks. For identification, seven patterns of training data and six patterns of testing data were obtained from a model. Each pattern had 9 data for 9 days. The results show that a conventional identification method in which current output is estimated from previous time series of both input and output can be seen unsuitable for identification. An addition of linear data (1, 2, …, N), for example : number of days, as input variable significantly improves the accuracy of identification. Besides, the as input variable significantly improves the accuracy of identification. Besides, the estimated error becomes smaller when the number of data pattern is three or more. This result suggests that three types of data patterns are enough to identify the cumulative responses such as fruit growth.
