人工神经网络在溶解度预测方面的应用

溶解度作为一项重要的物化指标,一直是化学学科的研究重点。然而,通过实验测量获得数据耗时费力,因此,科研人员建立了多种理论方法来进行估算,其中,人工神经网络因其能够关联复杂的多变量情况而受到广泛关注。本文综述了人工神经网络在物质溶解度预测方面的应用,介绍了应用最广泛的3种神经网络(BP神经网络、小波神经网络、径向基神经网络)的模型结构、预测方法和预测优势,探讨了神经网络的不足以及改进方法。文章最后对神经网络在物质溶解度预测方面的发展前景进行了展望。与其他方法相比,人工神经网络技术在物质溶解度预测方面具有预测结果精确度高、操作简单等特点,具有广阔的应用前景,但输入变量选择、隐含层节点数确定、避免局部最优等问题还需逐步建立系统的理论指导。

Prediction of Solubility using Artificial Neural Network

Solubility as one of the most important physicochemical properties of pharmaceuticals and chemical materials has always been the research priority in chemistry discipline, while experimental studies are very expensive and time consuming, so many researchers have tried to estimate the thermodynamic property by theoretical methods especially artificial neural network which received high attention because of its capability to correlate most nonlinear multi-variable phenomena with any complexity. In this paper, a review of application of artificial neural networks in predicting solubilities of compounds is given. Topological structure, prediction methodology and advantages of three of the most popular artificial neural networks (back-propagation neural network (BP), wavelet neural network (WNN), and radial basis function network (RBF)) are elucidated. Drawbacks of these neural networks and methods used to improve are also discussed. Finally, this paper also gives the prospect of prediction of solubility using artificial neural network. Compared with other theoretical methods, artificial neural networks have high accuracy in predicting solubility of compounds, and are easily to be utilized. Artifitial neural network has a broad application prospect in chemistry, while theoretical guidance on the selection of input variables, determination of appropriate number of neurons in hidden layers and avoidance of local optimum should be established gradually.