根据土壤质量定量评价指标分级体系生成足够多代表性好的神以网络训练和检验用的样本。建立神经网络模型时 ,利用删减或扩张准则确定神经网络最佳拓扑结构 ,避免“过拟合”现象 ,利用检验样本监控在训练过程中不发生“过学习”现象 ,使建立的土壤质量的综合评价与预测模型具有较好的泛化能力和预测能力。对三江平原地区主要耕作土壤质量的综合评价与预测结果表明 ,神经网络方法能较好地应用于土壤质量综合评价与预测 ,比加权综合指数法能更精细地评价与预测土壤的变化趋。 According to the grading system of quantitative evaluation index of soil quality, enough representative samples of good God training and testing are generated. When establishing neural network model, the optimal topology of neural network is determined by using the truncation or expansion criterion to avoid “over-fitting” phenomenon. By using the test sample monitoring, no “over-study” phenomenon occurs in the training process and the established soil quality Comprehensive evaluation and forecasting model has good generalization ability and forecasting ability. The results of comprehensive evaluation and prediction on the quality of main tillage soils in the Sanjiang Plain show that the neural network method can be applied to the comprehensive evaluation and prediction of soil quality better than the weighted comprehensive index method to evaluate and predict the trend of soil finer.