Food security is a problem that cannot be ignored on international scale.The growing population poses challenges to the demand of food.Rice accounts for a relatively high proportion of our country’s total agricultural output.Therefore,as an important part of agricultural research,the prediction of rice yield is self-evident in the development and construction of our country’s agriculture and food security ensuring.In order to improve the accuracy of rice yield prediction,reduce the complexity of the prediction model and improve its generalization,this paper analyzes the rice yield data in a certain area of Guangxi for several years,and focuses on the feature dimension reduction and neural network model in the field of crop yield prediction.Finally,a principal component analysis(PCA)algorithm and a long short-term memory(LSTM)recurrent neural network based on particle swarm optimization(PSO)were used to construct a yield prediction model,which could more accurately predict the yield of rice.The main work of this paper is as follows:(1)In order to solve the problems of too many rice planting condition features,information redundancy and dimension inconsistency among the features,an improved PCA algorithm is proposed.The data preprocessing stage of the algorithm can eliminate the feature dimension without losing its correlation coefficient and variation coefficient.Finally,through experiments comparing the IPCA algorithm proposed in this paper and the traditional PCA algorithm,the results show that the IPCA algorithm has stronger dimensionality reduction ability and can retain more information contained in the original data.(2)This paper studies the problem of crop yield prediction and establishes a PSO-based LSTM rice yield prediction model.Aiming at the problem that the neural network depends on the initialization parameters,this paper uses the PSO algorithm to optimize the training parameters of the LSTM.Meanwhile,an individual adaptive adjustment strategy,an asynchronous nonlinear decreasing strategy and a reverse escape strategy are proposed to improve the PSO algorithm.The performance of the PSO algorithm has been well improved.Finally,experiments are designed to verify that the IPSO algorithm has better optimization performance and stability.(3)Use IPSO to determine the training parameters of the LSTM,use the IPCA algorithm to reduce the dimension of the original data,and use the obtained principal components as the input of the neural network to form the final prediction model.Design experiments to compare the model proposed in this paper with other classic models.The experimental results show that the model proposed in this paper has high prediction accuracy and high converge speed,and has a strong ability to jump out of the local optimal solution. |