基于RBF神经网络预测方法的碱渣土工程特性分析

为了缓解纯碱行业生产快速发展与碱渣贮放的矛盾,国内碱厂主要采用贮渣池加高作为解决手段.对待建碱渣坝土样的工程特性进行预测分析,能够为坝体的后续加固加高提供科学依据.碱渣土具有复杂的物理力学性质,应用优化识别预测方法能够更好地揭示数据中各指标间的非线性关系,因此采用RBF神经网络的数据拟合能力,建立碱渣土工程特性的预测模型.通过工程算例的应用效果表明,RBF神经网络具有可调参数少、网络性能稳定和预测精度高等优点,分析了输入样本数目和宽度向量对RBF神经网络预测性能的影响,可为碱渣土工程特性的初步预测分析提供有效的决策依据.

Study of Engineering Characteristics of Slag Soil Based on RBF Neural Network Forecasting Method

In order to relieve a contradiction between the rapid development of soda industry and sludge stor- age, heightening residue storage tank was applied as a mean of resolution in many soda companies. If we ana- lyze the engineering characteristics of slag soil, the analysis results can provide the scientific basis for raising dam. Because slag soil has the complex physico-mechanical properties, the optimal identification forecasting method should better reveal nonlinear relationship of data. In this paper a forecasting model is established with RBF neural network; it has an excellent data fitting ability. Through case studies show that RBF neural network has advantages of few adjustable parameters, network stability and good aecuraey. In this paper, the effects of number of input samples and width vector on the prediction ability of RBF neural network are ana- lyzed. Therefore, based on RBF neural network about optimal prediction for engineering charaeteristies of slag soil is feasible.