The Early Warning Model Of Agricultural Sustainability In Counties Based On Improved Artificial Neural Network

Sustainability of agriculture is the extension of the theory of sustainable development to the field of agriculture and rural economic development, and it is also the introspection to conventional economic development model and the model of agricultural modernization. However, the survival and development of the regional agricultural system vulnerable to the interference of the economic-social, resource and environmental factors, so that agricultural development may deviate from the normal orbit and result in a variety of warning conditions. By looking for the warning source and analyzing the warning signs, establishing sensitive and accurate short-term early warning models on the county scale for more appropriate research of early warning of agriculture enables decision-makers to detect and remove the warning conditions timely, to take flexible and effective macro-control measures in process of agriculture. It is of great significance to ensure country-scale sustainable development of agriculture and national food security.The theory of sustainable development and early warning are the important theoretical foundation for the early-warning model of sustainability of agriculture in counties. In this paper, based on the basic theoretical framework of sustainable development, the concept of "early warning of sustainability of agriculture" has been defined. A scientific and easy-to-operate early-warning model architecture for country-scale sustainable agriculture has been established with the artificial neural network improved by the weighted principal component analysis being the core model, by the artful combination of the yellow warning method and the traditional systematic approach. In the area along the Yellow River that is one of the important major grain producing areas located in the Huang-Huai-Hai Plain, five typical counties have been selected as the main case study areas to complete the four key steps of defining the warning condition, looking for the warning source, analyzing the warning sign and forecasting the warning degree. The state, process, trends and their impact factors of the agricultural development in the next five years in five counties were studied in depth.The main conclusions can be summarized as follows: (a) through the coupling of artificial neural networks, statistical early warning methods, and modeling early warning methods, early warning model architecture for sustainability of agriculture on the county scale has good operability. The design of the research ideas especially highlights the coupling of the traditional systematic methods with artificial neural networks. The theory of early warning is as an important theoretical basis for the workflow design. The basic analysis process of the yellow early-warning law, i.e. defining the warning condition, looking for the warning source, analyzing the warning sign and forecasting the warning degree has been adopted. Three-dimensional space for the warning condition of agricultural sustainability, sensitivity analysis of indicators based on the WPCA-NN, Pearson correlation analysis and the warning degree forecast based on the WPCA-NN are the core technology in each step. (b) by changing the value of the initial weight vector, BP algorithm improved by the weighted principal component analysis method not only can reflect the preferences of decision makers on the indicators, but also can obtain a fast convergent and highly accurate neural network model. Weighted principal component analysis was used to change the values of the initial weight vector of the BP neural network to form a three WPCA-NN. Weighted principal component analysis algorithm of the unsupervised learning and linear instructingδlearning rules were used to complete the weight learning from the input layer to the hidden layer and from the hidden layer to the output layer. It not only can reflect the preferences of decision makers on the indicators, but also can avoid the defects of classical BP algorithm, such as low convergence speed and sensitivity to local convergence. (c) classification of economic strength has been made using methods of the weighted principal component analysis, hierarchical cluster analysis, self-organizing feature map network modeling, and global and local spatial heterogeneity analysis. The typical county selection was based primarily on the classification of the economic strength of the 109 counties (cities, districts) along the Yellow River. SOFM neural network modeling is the main technical means, whereas the two pre-processing steps of weighted principal component analysis and hierarchical cluster analysis are critical to ensure a reasonable classification by SOFM network.Five counties, i.e. Kenli, Zhongmou, Gaoqing, and Fengqiu County, were selected, respectively, as the representative of counties of the strongest, stronger, middle, weaker, and weakest economy strength. (d) It is suggested that the empirical analysis of the sustainability of agriculture on the county scale has come up to expectation early warning and conformed to reality. Sensitivity analysis of the indicators reveals that the material conditions of agricultural production and infrastructure conditions, the input levels of land use and the environmental background of agricultural production are the common warning source for sustainable agriculture development in five counties. Although the warning conditions of sustainability of agriculture in five counties from 2009 to 2013 will all occur, they are mainly light and moderate, indicating that the decline of sustainability of agriculture is not very serious. Abnormal fluctuations of resources and environment warning signs have a more direct effect on the warning conditions.The main innovations lie in the following items: (a) the early warning model architecture for sustainability of agriculture on the county scale proposed in this paper is the improvement and further deepening of the existing model architecture; (b) sustainability of agriculture of the Yellow River Basin on the county scale has been studied for the first time; (c) the improved BP algorithm has more explanatory power and applicability for study of the sustainability of agriculture on the county scale.