Price Forecasting Of Aquatic Products Based On BP Artificial Neural Network Model

Aquatic products play an important role in China’s fishery economy and international trade. As a specific manifestation of the cost of production and the relationship between supply and demand, aquatic products prices are relevant not only with production, sale and economic benefit, but also with the national macro-economic policies. Its changes attract the extensive attentions in aquatic products market. Therefore the price prediction of aquatic products could be used as a basis for farmers to make production decisions, help them to grasp the initiative in economic decision-making and get profits. Furthermore, the price prediction could provide the scientific information support for the government to formulate policies that may improve efficiency of resources allocation, mitigate price fluctuations of aquatic products, and finally realize the aim of sustained, rapid and healthy development of aquaculture.Artificial neural network is a new cross-disciplinary approach, which has been widely applied to practical problems of prediction in recent years. Especially with its nonlinear systems, artificial neural network prediction has great perspective in future. Thus we proposed a novel method to predict the prices of aquatic products using artificial neural network.Based on the application principles of BP neural network, this study proposed the steps in prediction, analyzed the feasibility and explored the key technologies in the establishment of BP neural network based prediction model. These technologies included sample selection, pre-processing, selection of input and output unites, identification of implicit nodes, initial weight and threshold selection, activation function, and the selection of training algorithm and parameters. Ultimately we established a6:10:1three-layered neural network prediction model, and implemented the price forecast of aquatic products using neural network theory and MATLAB6.5.In this study, Crucian carp was selected as a predict species. Using time, geographic factors, environmental and economic conditions as variables at input layer, the price as the output, we did model training and simulation with the input of samples. As the results, the relative error is less than1%between the prices of actual market the values predicted by the trained network. The results not only showed the good accuracy and precision of the neural network model established in our study, verified the feasibility of the application of the neural network in the price forecast of aquatic products, but also provided a valuable new method to do market expectation of aquatic products.