Based On Improved LSTM Fertilizer Price Forecast

Among all the means of agricultural production,chemical fertilizer has always occupied a pivotal position,which has the effect of increasing the output of crops and improving soil fertility,and occupies a large proportion in the expenditure of agricultural production.Excessive fluctuations in fertilizer prices will reduce the enthusiasm of using fertilizer,thus affecting the output of crops,soil fertility and the quality of agricultural products.This will not only reduce the income of farmers,but also adversely affect China's agricultural production and reduce the competitiveness of China's agricultural products in the international economy.Therefore,accurately predicting the price of fertilizers can effectively avoid large fertilizer price fluctuations,reduce farmers' expenditures,and stabilize China's grain market.On this basis,the efficiency of the long-and short-term memory neural network algorithm is improved.This paper firstly studies the influencing factors of fertilizer price changes and the problems faced by fertilizer price forecasting.By comparing traditional linear forecasting methods,classical neural network algorithms and support vector machine algorithms,the recurrent neural network(RNN)model of the memory(LSTM)unit is proposed.This model can effectively utilize the ability of sequence data to rely on long-distance information,fully exploit the hidden rules behind sequence data,and perform short-term price time series prediction.Based on this,the long-term and short-term memory neural network algorithm is improved by the quasi-Newton method.Secondly,in the experimental simulation stage,dhe urea fertilizer price data of Jiangsu Province in the past ten years was taken as the experimental object,the data of the experimental data was cleaned,and the time series problem was transformed into a supervised problem.Then,based on the improved long-and short-term memory neural network model.the mean square error(MSE)is used as the evaluation index,and the data is regression fitted.Then the training model is optimized for the established model.Finally,the optimized model will be used in the fertilizer price information system to predict the price of fertilizer trading in the short term.Through empirical analysis,the paper finds that the mean square error obtained in the test set based on the improved LSTM model is lower than that of the BP neural network and the support vector machine on the test set,indicating the feasibility of the improved LSTM model.Sex.At the same time,the construction of the fertilizer price information system can also allow the fertilizer industry practitioners,farmers and the government to intuitively observe the trend of fertilizer prices and provide them with certain decision support.