Research Of Soil Suitability Evaluation Method Based On BP Neural Network

Soil suitability evaluation is aimed at revealing the soil suitability and restriction for different agricultural purposes, which provides the foundation for determining the most appropriate way of land use. The appropriateness of soil and its effect on crop cultivation is a basic condition for land remediation and resettlement in resettlement areas, related to the major issues such as the development direction of farming, farming patterns, the development level of agricultural production, the number and manner of resettlement. An in-depth investigation on soil types and traits, and evaluation of suitability can provide a scientific basis for land use in resettlement areas. At present, the traditional methods for soil suitability evaluation in practice have some limitations. Therefore, this article tries to introduce the artificial neural network to evaluate soil suitability, establish a soil suitability evaluation model by BP algorithm and use it for an actual soil suitability evaluation in resettlement area so as to provide an objective and accurate method for soil suitability evaluation.In this paper, the soil suitability evaluation model based on BP artificial neural network has been improved in three areas: fristly the improved L-M (Levenberg-Marquardt) learning rule has been used in this model as a training function of BP neural network, as the result the training error converges to the minimum during less iterative cycle times, the speed of network training have been increased; secondly, the expert sample (training sample) has been expanded of the network model, it means that there generated 15 groups of training samples randomly between the critical value in each class based on the training samples that composed of national soil quality standards, so as to improved the robustness and accuracy of identification of the model; the third, this model have an optimization for hidden layer nodes by the optimization algorithm based on the theory of Golden Section, which could quickly determining the optimal number of nodes in hidden layer, and enhancing the performance of the network model. In this paper, the finalize improved structure of BP network model is 9-7-1, the training mean square error is 0.00033. This model was used in the soil suitability evaluation in DuLuoXi Hydropower Station Resettlement, and compared it with the evaluation results based on partial least-squares regression method and experience index method, the results shows that the experience index method is greater influenced by man-made subjective factors, and the evaluation result is significantly differ with the other two evaluation methods. The model error and evaluation error of BP neural network model are both less than partial least-squares regression method, it means that the soil suitability evaluation based on BP neural network model has a higher precision. The study shows that the soil suitability evaluation model based on BP artificial neural network established in this paper provides a simple, objective and practical method for the soil evaluation work.