Research On Price Prediction Of Agricultural Products Based On HTM And Seq2Seq

China is a large agricultural production country.The change trend of agricultural product prices affects the happiness index of residents to a certain extent,and the fluctuation of agricultural product prices is affected by the supply and demand of agricultural product markets.When agricultural practitioners are asymmetric with regard to the agricultural product market information,they are more likely to produce blind planting behaviors,which often leads to supply-side imbalances and disorder,and planting enthusiasm falls sharply.If a certain price prediction mechanism can be established so that agricultural practitioners can know the price trend of agricultural products in advance,they can better guide agricultural planting,protect the interests of practitioners,and promote the transformation of agricultural production to a modern model,thereby reducing the burden of government macro-control.With the development of emerging information technologies such as big data,Internet of Things and artificial intelligence,it has provided strong support for the construction of modern agriculture.The main forms of precision agriculture oriented to the production process and e-commerce platform for crop sales are driving the changes in the agricultural development path,but how to cut prices and guide agricultural planting has not been extensively researched and applied.Therefore,agricultural product price prediction is a question worth exploring.This article focuses on the application of deep learning in crop price prediction.The main research contents are as follows:(1)Agricultural product price prediction is one of the applications of time series.This article starts with the application of time series prediction in agriculture and consults related literature.Generally,time series prediction research is carried out by constructing models,and the most commonly used model is currently a machine learning model.Among them,the integrated learning model constructed by the machine learning model can solve more complicated time series models.(2)In view of the high-dimensional and decentralized problems characterized by agricultural production data,this paper first builds a web crawling system to obtain agricultural data sets of different dimensions,and introduces mutual information methodto preprocess the data.At the same time,in order to improve the applicability of the data and indirectly improve the accuracy of the model,the usability of the hierarchical real-time network in time series is discussed,and the hierarchical real-time network is used for anomaly detection.The analysis of agricultural product price time series data and quantitative experiment analysis prove the effectiveness of the proposed method,effectively reduce the abnormal value of agricultural product time series,and improve the accuracy of the model.(3)In view of the fact that agricultural product data fluctuates frequently and it is difficult to capture the law of data,this paper proposes an integrated learning method to capture the agricultural product price fluctuation law and predict the future price of agricultural products.This article combines the characteristics of Seq2 Seq and the adjustable input and output lengths to facilitate multi-day forecasting.In addition,in order to obtain the features of the time series from a global perspective,based on the Seq2 Seq model,the CNN model was introduced,the Encoder part of the model was changed,and the features extracted by RNN and CNN were recombined to build an integrated learning model PRCNN,Make it more consistent with the prediction of agricultural time series data.Because the COCOB optimizer does not need to set the learning rate and the fitting speed is faster,the COCOB optimizer is introduced to speed up the fitting of the model to the data and reduce the training time.(4)In addition,by constructing a variety of models with better performance,such as LSTM and GRU,etc.,a comparative test is performed to verify the performance of the new model on agricultural product time series data sets,and the RMSE,MAE,and MAPE of the results of different models are compared.The results show that the constructed integrated learning model has better performance in agricultural product price prediction and better generalization ability.