On The Optimization Of Neural Network For Grain Yield Prediction By Using PSO Algorithm

Grain yield is one of the standards to measure a country’s economic strength, whichplays a positive role in promoting countries’ economic and social development. It providesthe guarantee for all ethnic groups with inexhaustible driving force. It is also the key torealize transition from traditional agriculture to modern agriculture. Through the ages,agriculture has been accompanied with the survival and development of human being.Whether the traditional grain or new crops by people carefully nurtured, all of which havemade an important contribution on promoting innovation and progress of agriculturescience and technology. Since the People’s Republic of China was founded, with thedevelopment of agricultural scientific and technological achievements, the agriculture andthe rural economy development steady and fastly. As we all know, China is a largeagricultural country. People have a lot of experience in responses the food crisis. With thelargest population, it is the largest country in grain yield and consumption. As relativescarcity of arable land resources, safeguarding the stable and sustainable development ofagriculture become a focus of agricultural research. Therefore, scientific forecasting ofgrain yield is of great guiding significance to agricultural development.This thesis takes an overview on the present situation of grain yield in China firstly.Then it briefly expounded the basic principle of Artificial Neural Network, the BackPropagation Neural Network and Particle Swarm Optimization. Furthermore, the originaldata of relevant factors affecting grain yield were standardized, and the simulationexperiment were established. We found that the Back Propagation Neural Network has theslow convergence and easy fall into local minimal and other defects of the BP neuralnetwork in the training process. Therefore, an improved particle swarm optimizationalgorithm was introduced to optimize the performance of the BP neural network. Itimproves the acceleration coefficients and inertia weight of the PSO algorithm. Comparedwith the traditional BP algorithm and particle swarm optimization, the improved particleswarm optimization algorithm can not only quickly converge to the neural network learningobjectives, but also improve the accuracy of the prediction model. It also shows that it isfeasible and effective to predict the food production with this method.