Study On Rice Yield Prediction Based On ODESSA-BiGRU Model

Rice is one of the most important food crops in China.Accurate prediction of rice yield has a great significance for formulating grain storage,transportation and allocation plans and ensuring the balance of grain supply in all parts of China.However,there is a complex nonlinear relationship between rice yield and various meteorological elements in its growth period,which is difficult to be accurately predicted by traditional prediction methods.Therefore,the rice yield prediction method based on the combination of improved squirrel search algorithm(SSA)and bidirectional gated recurrent unit(BIGRU)model was studied.The main research work completed includes:(1)Aiming at the disadvantages of low accuracy and early maturity of SSA,an improved SSA based on opposition based learning and differential evolution(ODESSA)is proposed.Firstly,the algorithm uses opposition based learning to generate its opposition population for the initial population to increase the diversity of the initial population.Then,in the process of squirrel search,the mutation,crossover and selection mechanism in differential evolution algorithm is used to expand the squirrel search range,so as to enhance the global exploration ability of the algorithm.Finally,opposition based learning is used to generate the opposition solutions of all search individuals,which further enhances the global exploration ability and solution accuracy of the algorithm.The performance of ODESSA is compared with other three intelligent algorithms(SSA,PSOSSA,and ISSA)on 12 benchmark functions.The experimental results show that ODESSA obtains 10 optimal results(two of which are parallel optimal)among the 12 benchmark functions.In order to verify the improvement effect,the performance of ODESSA is compared with only using opposition based learning improved SSA(OSSA)and only using differential evolution improved SSA(DESSA)on 12 benchmark functions.The results show that ODESSA is better than OSSA and DESS A,has the best optimization results,and the convergence speed is faster than SSA.(2)Using ODESSA to optimize the training process of BIGRU model,a new prediction model(ODESSA-BIGRU)combining ODESSA and BIGRU is proposed.The model trains the model parameters by using ODESSA instead of the gradient descent method used in the traditional BIGRU training parameters.ODESSA has a larger search range and stronger optimization ability,which improves the deficiency that the gradient descent method is easy to fall into the local optimal solution,so that the BIGRU model can train better parameters and obtain better performance.Finally,the model is applied to the complex problem of rice yield prediction.Using the early rice yield of Nanning over the years and the meteorological data of the same period as the experimental data,the ODESSA-BIGRU model is trained to obtain the final prediction model.The performance of this model is compared with that of other four artificial neural networks.The experimental results show that ODESSA-BIGRU model has higher prediction accuracy than other models,and provides a new method for rice yield prediction.