BP神经网络的发展及其在化学化工中的应用

人工神经网络（ANN）由于本身具有极强的非线性映射能力、容错性、自学习能力得到广泛的应用。基于反向传播算法（BP）的神经网络作为ANN重要组成部分，在涉及多种非线性因素建模时，相对于传统的反应机理建模显示出巨大的优势。虽然神经网络的发展几经繁荣与冷落，但目前在不同领域已经获得成功的应用。本文概述了BP神经网络的映射原理、缺点以及相应的改进方法，介绍其在催化剂设计、动力学模拟、理化特性估算、过程控制与优化、化学合成与反应性能预测的应用现状，展示了使用不同优化方法的改进模型在实验设计与优化方面取得的成果。最后指出未来BP神经网络的发展要进一步结合数据深度挖掘与机器学习等技术，为今后化学化工领域的研究提供强有力的工具。

Development of BP neural network and its application in chemistry and chemical engineering

Artificial neural network (ANN) is in universe application because of intrinsic favorable nonlinear mapping ability, fault tolerance and self-learning ability. Backpropagation (BP) neural network, an important part of ANN, has great advantages over traditional reaction mechanism modeling which deals with nonlinear multi-factor system. Currently it has been successfully applied in varying fields after experiencing intermittent prosperous and fading development historically. Herein the principle of mapping process, the shortcomings and the corresponding improvement methods of BP neural network are briefly summarized, and its applications in catalyst designing, kinetic simulation, physical and chemical properties prediction, process control and optimization, chemical synthesis and reaction performance prediction were introduced in detail. The prediction accuracy and efficiency in experiment design and process optimization based on BP neural network could be enhanced remarkably by improved algorithm. Finally, it was pointed out that BP neural network, further combined with data depth mining and machine learning techniques, could be a powerful tool for future research in chemical field.