Import Commodity Customs Declaration Price Prediction Based On Artificial Neural Network

Import trade plays an irreplaceable role in making up for the shortage of domestic resources and maintaining the development of national economy.Among them,the customs declaration price of imported commodities affects the development of the domestic market.The prediction of the customs declaration price of imported commodities can help the government solve the risks in the market and realize the macrocontrol of the market.Therefore,it is very important to study the prediction of the customs declaration price of imported commodities.With the development of artificial intelligence-related technology,people keep trying to use neural network method to forecast various commodity price data,and put forward a variety of commodity price data forecast models.Considering that there are many disturbance factors in domestic and foreign markets,the import commodity declaration price sequence contains many irregular fluctuations,which hinders the construction of the model.The neural network has a strong ability to solve nonlinear problems,but it is very likely to over-capture the characteristics of irregular fluctuation data in the network training,resulting in low prediction accuracy,and then reduce its ability to predict the customs declaration price of imported commodities.Therefore,this paper firstly uses wavelet transform to decompose the customs declaration price sequence,and then uses artificial neural network to predict and analyze the sub-sequence,so as to reduce the negative impact of irregular fluctuations in the sequence on the neural network model and improve the prediction ability of the model.Finally,the prediction results are obtained by using wavelet reconstruction.BP neural network,Elman neural network and convolutional neural network(CNN)are three neural network prediction methods that can be used to fit time series.Therefore,in this paper,the above three neural network methods are respectively combined with wavelet decomposition and compared with the traditional time series model to carry out empirical analysis on the customs declaration price data of imported commodities,and then verify the universality of the improved prediction model to the prediction and analysis of other time series.The main tasks are as follows:(1)The content studied in this paper is the research topic of digital ecological construction of shipping trade finance in Qingdao based on blockchain,in which the original data set of customs declaration price of imported natural rubber provided by Qingdao Customs Center is selected.In view of the diversity of commodity types,large time span,outliers and irrelevant variables in the original data set,the NRCDP data set is established as the sample set for model research.(2)Aiming at the problem that the customs declaration price in NRCDP data set is difficult to predict due to fluctuations and anomalies,a prediction method based on wavelet decomposition and genetic algorithm optimization of BP neural network group(W-GA-BPG)is proposed,which mainly uses wavelet transform to decompose customs declaration price in NRCDP data set.A set of BP neural network optimized by genetic algorithm was used to predict several decomposition subsequences respectively.Finally,the prediction results of customs declaration price were obtained through wavelet reconstruction,and the mean square error(MSE)and determination coefficient(Rsquare)of the prediction results of future natural rubber customs declaration price were compared and analyzed,so as to establish the optimal model.(3)A prediction method based on the combination of wavelet decomposition and Elman neural network(W-Elman)and a prediction method based on the combination of wavelet decomposition and convolutional neural network(W-CNN)are proposed.The method of W-Elman and W-CNN mainly uses wavelet function to decompose the customs declaration price in the NRCDP data set,and then uses the neural network model of a single structure to predict the sub-sequence.Finally,the predicted sub-sequence is reconstructed by wavelet to obtain the prediction result of customs declaration price.In order to verify the universality of the model,the proposed method is also applied to other time series stock prices and futures prices.The final results show that the three models proposed in this paper have higher accuracy in the prediction of natural rubber customs declaration price,and have a great improvement compared with the classical model.It has been well applied in the research topic of Qingdao shipping trade finance digital ecological construction based on blockchain,and the three models also have certain universality in other time series.