Research On Recurrent Neural Network Optimization Algorithm For Cotton Yield Prediction

The cotton industry is an important pillar supporting China’s agricultural economy,and the level of cotton production affects the development of China’s textile and other light industries to a certain extent.Accurate prediction of cotton yield is important for guiding agricultural development,rational planning of planting strategies,reducing waste of water and fertiliser resources and predicting cotton prices.The yield of cotton is affected by many factors.This paper analyses the factors affecting cotton yield in three dimensions: meteorology,soil and remote sensing image information.Cotton yield prediction was studied in an experimental field at the University of Missouri in the Mississippi River Basin.To address the problem that remote sensing information prediction and regression model prediction cannot effectively handle complex and high-dimensional input information and are less stable,this paper develops a neural network model based on Back-Propagation(BP)and two recurrent neural network models: Long and Short Term Memory(LSTM)and Gated Recurrent Unit(GRU).Experiments show that all three neural network models can effectively handle high-dimensional data and predict cotton yield quantitatively,with the model errors in descending order of BP,GRU and LSTM,verifying the feasibility of subsequent neural network models to predict cotton yield.In this work,two recurrent neural networks are integrated with an attention mechanism to develop a cotton yield prediction model with an attention mechanism LSTM＿Attention＿GRU(LAG)to address the shortcomings of traditional neural network models in processing the weight information of cotton yield influencing factors.The LAG model with the Encoder-Decoder framework is developed to use the LSTM as the encoding layer of cotton time series information,and the attention mechanism is used to assign weights to the output encoding information of the LSTM,and finally the GRU neural network decodes the weight information and outputs the prediction results.The accuracy and efficiency of the LAG model for cotton yield prediction was verified to be much better than the above three traditional neural network models.In order to solve the problem of difficult selection of hyperparameters for the LSTM coding layer in the LAG model,this work determines the optimal hyperparameters for the LAG neural network model by improving the Chimp Optimization Algorithm(Ch OA),and finally establishes a LAG neural network model based on the improved Chimp Optimization Algorithm Improved Chimp Optimization AlgorithmLSTM＿Attention＿GRU(ICOA-LAG).The model improves the global optimization capability and convergence speed of the chimp optimization algorithm by improving the chaos mapping method,convergence factor,backward learning strategy,and Coase Gaussian variation in the chimp optimization algorithm,and finally determines the optimal hyperparameters of the ICOA-LAG model.The experimental results show that the accuracy and efficiency of the ICOA-LAG model have been further improved to meet the demand for accurate cotton yield prediction.