基于BP神经网络的TRIZ理论冲突矩阵

使用Matlab ANN工具箱，将传统TRIZ理论的核心——冲突矩阵变换为一个以工程参数及其冲突为输入层，以冲突解决原理为输出层，并包括一个隐含层的三层BP人工神经网络模型。该模型基于具有较快收敛速度的改进BP算法——弹性反馈学习算法RPROP，利用已有的发明实例和问题解决方案作为学习样本，使模型的建立及修正更加容易，新问题的解决更具有针对性，为企业应用TRIZ提供一种较为简便的实用方法。根据问题的规模和计算需求确定了网络模型的详细参数，具体给出了由实例到样本的抽象过程。最终运行的结果显示了较好的收敛性和较快的收敛速度，证明了该方法的可行性。

Contradiction Matrix of TRIZ Principle Based on BP Network

TRIZ principle is a valid method for innovative design. However, the traditional TRIZ based on statistics can not embody the nonlinear relationship between engineering parameters and inventive principles very well. And it's also difficult to be applied by enterprises. Using the ANN Toolbox of Matlab as a powerful modeling tool, the traditional contradiction matrix of TRIZ is transformed into a three-layer BP ANN model, which takes engineering parameters and their contradictions as input layer, solving principles of contradictions as output layer, besides, a hidden layer also included. Based on the improved BP algorithm RPROP with high convergence speed, the model takes the existing invents and problem-solving methods as study samples. So it is easier to construct or modify it, and more pertinent when applied to a new problem. It provide a easy and practical way for enterprises to use TRIZ. According to size and computing requirement of the problem, model parameters are determined in detail. Abstracting process from example to sample is also provided. The result shows good convergence with satisfying running speed of the model, which validates the proposed method.