Abstract: The application of the convention BP neural network in the coronary intervention recrudescence prediction might lend to complex architecture for the neural networks and false diagnosis.A new combined model，which is based on niche technology，gene expression programming and BP neural network，is proposed in this paper.Under considering the shortcomings of that BP neural network easily fall into local minima，the method uses niche technology and gene expression programming to optimize the structure and original weight distribution of BP networks.The neural network is accurately trained with gradient descent algorithm.The combined model is applied to the prediction of the 2-year prediction of coronary intervention.Compared the result with convention BP neural network，it shows that the new method speeds up the convergence and improves the prediction accuracy.

Key words: BP neural network, gene expression programming, niche, coronary intervention, prediction

摘要： 为了克服BP神经网络在冠心病介入术后复发预测时存在网络结构复杂且易出现误判断的不足，提出一种小生境技术、基因表达式编程与BP神经网络相结合的冠心病介入术后复发预测的组合模型构造方法。该方法首先利用小生境技术和基因表达式编程的方法对BP神经网络的权值、阈值和网络结构进行优化，解决由于BP神经网络易陷入局部最优的缺陷;然后用梯度下降法对优化后的BP神经网络进行精确调整。将此方法应用于冠心病介入术后2年复发预测中，结果表明优化后的BP神经网络比未优化的BP神经网络具有较好的收敛性能，且预测精度更高。

关键词: BP神经网络, 基因表达式编程, 小生境, 冠心病介入术, 预测