Prediction of Vessel Traffic Volume in Ports Based on Improved Fuzzy Neural Network

Guilian Su,Tian Liang,Mei Wang

doi:10.1109/access.2020.2982904

Abstract

The accurate prediction of vessel traffic volume (VTV) is very helpful to rational scheduling of port resources and reducing vessel accidents. However, the traditional VTV prediction methods face problems like the overfitting of historical data and prediction inaccuracy. To solve these problems, this paper improved the fuzzy neural network (FNN) with quantum genetic algorithm (QGA). Firstly, the basic principles of neural network (NN), fuzzy inference and the QGA were introduced in turn. Then, the weights of the FNN were optimized by the QGA. On this basis, the authors established a VTV prediction model based on the improved FNN. Finally, the established model was applied to predict the VTV in an actual port of China, in comparison with several classic NNs. The VTV data were collected based on the length and gross tonnage of vessels. The results show that the improved FNN outperformed the contrastive methods in the accuracy of VTV prediction, thanks to the optimization by the QGA. The proposed method enjoys a great application potential in the prediction of the VTV in ports.

Highlights

Thanks to the boom of economy and international trade, the number of vessels has increased significantly, and the vessel traffic volume (VTV) in China’s coastal ports has multiplied
To achieve the desired output, the five-layer fuzzy neural network (FNN) is trained by the nonlinear parameters of the membership function in the first layer and the linear parameters randomly inputted to the fourth layer
PREDICTION RESULTS AND ANALYSIS As mentioned above, the proposed quantum genetic algorithm (QGA)-FNN was compared with the backpropagation neural network (BPNN), RBFNN and FNN

Summary

INTRODUCTION

Thanks to the boom of economy and international trade, the number of vessels has increased significantly, and the vessel traffic volume (VTV) in China’s coastal ports has multiplied. G. Su et al.: Prediction of VTV in Ports Based on Improved FNN methods (DDPMs) [16], [17]. Many studies have shown that, fuzzy neural network (FNN) usually outperforms model-driven linear methods (e.g. RAPM and TSPM) in VTV prediction. (1) To the best of our knowledge, this is the first attempt to predict the VTV in ports through the combination of the QGA and the FNN. For many nonlinear complex systems, it is possible to identify the input and output, but extremely difficult to set up a model or functions for the systems. To achieve the desired output, the five-layer FNN is trained by the nonlinear parameters of the membership function in the first layer and the linear parameters randomly inputted to the fourth layer

QGA-BASED TRAINING OF THE FNN

DATA PROCESSING

Findings

CONCLUSION