Grain Yield Combination Prediction Based On Neural Network Optimized By Genetic Algorithm

The CPC Central Committee, the State Council issued the "Several opinions on deepening reform and innovation to speed up the construction of agricultural modernization” in the beginning of the year 2015".Comments pointed out that the grain production capacity must be enhanced. Thus it can be seen, the grain yield is one of the standards to measure the economic strength of a country. It is an inexhaustible motive force that can guarantee the people have ample food and clothing. And it is the important guarantee to realize the transition from traditional agriculture to modern agriculture. Although China’s grain output steady growth year after year, but we are still faced with many difficulties, such as: land use, land salinization, weather disaster and so on. These are potential factors that may result in food production.China has rich experience in dealing with the food crisis since ancient times the corresponding. But relative to the actual conditions of population, relatively large food consumption, the relative scarcity of arable land resources, It becomes urgent to solve the problems in agricultural research that protect the stability and sustainable development of agriculture and food security. Therefore, according to the change law of the development of grain yield in the present study to predict its development tendency is not only crucial to formulate food policy, but also crucial to make decision to control grain yield system. It has important practical significance for the protection of national food security.In this paper Firstly, the concept, characteristics, the network structure and learning method of neutral network are analyzed in this paper. Secondly, the BP neural network, RBF neural network and GRNN neural network is studied on the basis of these. In the actual application process, BP neural network has two problems. One is that it has slow convergence speed, even no convergence. The other is that selection of initial weights, threshold and network structure is random and the selection of initial point is not necessarily global problem and iterative results are not necessarily optimal problem.RBF neural network has two problems. One is that determination of the hidden layer unit is local and the optimal hidden layer units can’t be guaranteed to select. The other is that the number of hidden layer is usually fixed and those are selected through experience. Large time is consumed. GRNN neural network has two problems. One is that selection of radial basis function centers and widths, the connection weights from the hidden layer to theoutput layer has great effect to function approximation ability of the neural network. The other is that the conventional GRNN learning rule is very easy to make the results converge to a local minimum, even no convergence. Genetic algorithm is a random global mimic natural biological evolution of search and optimization method. It is an efficient, parallel, global search method. It can automatically obtain and accumulate knowledge about the search space during the searching process, and adaptively control the search process in order to obtain the optimal solution. In this paper, genetic algorithm is introduced to optimize BP neural network, RBF neural network and GRNN neural network in order to improve the prediction performance of the neural network.In traditional combination prediction methods, the different weighted average coefficients are given according to the different individual forecasting methods. For an individual prediction method, the weighted average coefficients are the same in different sample interval time. However, an individual prediction method has different performance in different sample interval time. Namely it has high prediction accuracy at a certain time point, and it has low prediction accuracy at another time point. So the existing combination forecasting methods are inconsistent with the actual. Combination forecasting model based on IOWA operator introduces IOWA operator, the weights is given according to prediction accuracy in different sample interval time, and the quadratic sum of error is taken as criterion to build combination model. Therefore, the results of GA-BP, GA-RBF and GA-GRNN are mixed by using combination forecasting model based on IOWA operator in this paper in order to improve the prediction accuracy further.The experimental results show that this method can effectively improve the prediction accuracy of the grain yield.In addition, the advantages and disadvantages of several mixed programming of C# and MATLAB methods are analyzed in this paper. Then grain predicting system is developed. C# is used as the front-end development environment to design system interface, display and output the results. MATLAB R2010 a is used as the back-end computing and graphics drawing tool for the design and development. Prediction of grain flow is simplified by this system and the manual calculation is reduced effectively. The system has a wide application prospect.