Based on the advantages of soft sensor modeling methods with increasing applications in many other fields [5,6], neural networks (NN) have been used to predict the flooding velocity by Piche et al [4]. According to his report,
the prediction performance of NN model for the flooding velocity has an improvement than traditional empirical models.
However, there are still some disadvantages about the NN modeling method. For example, NN is prone to get stuck in
the local minima in its learning and difficult to determine the optimal topology of the network [5,6]. The establishment of the network structure often depends on the experience, which often causes over-fitting during the modeling, and degrades the generalization performance. Another drawback of the NN models is a relatively large amount of samples are needed during training. Consequently, NN models often fail due to the small-sample circumstance [7]. Unfortunately, the flooding velocity data are often very few as they are difficult to obtain actually for a special packed tower. Therefore, it is difficult to obtain a suitable soft sensor model with a good prediction performance for the flooding velocity with limited samples.
Based on the advantages of soft sensor modeling methods with increasing applications in many other fields [5,6], neural networks (NN) have been used to predict the flooding velocity by Piche et al [4]. According to his report,
the prediction performance of NN model for the flooding velocity has an improvement than traditional empirical models.
However, there are still some disadvantages about the NN modeling method. For example, NN is prone to get stuck in
the local minima in its learning and difficult to determine the optimal topology of the network [5,6]. The establishment of the network structure often depends on the experience, which often causes over-fitting during the modeling, and degrades the generalization performance. Another drawback of the NN models is a relatively large amount of samples are needed during training. Consequently, NN models often fail due to the small-sample circumstance [7]. Unfortunately, the flooding velocity data are often very few as they are difficult to obtain actually for a special packed tower. Therefore, it is difficult to obtain a suitable soft sensor model with a good prediction performance for the flooding velocity with limited samples.
การแปล กรุณารอสักครู่..
