基于粒子群优化BP神经网络的脉象识别方法

Pulse recognition method based on PSO-BP neural network

针对传统脉诊存在易受主观因素影响、诊断结果可靠性不高等问题,提出基于粒子群优化BP神经网络的脉象识别方法。粒子群算法中评判粒子好坏的适应度函数采用神经网络的输出误差,以此获得最优粒子的位置向量,并把其值作为BP神经网络的初始权值和阈值。在Matlab中建立基于BP算法、PSO-BP算法和GA-BP算法的三种ANN模型用于脉象信号的识别。实验结果表明,在识别脉象时,优化后的算法降低了传统BP神经网络的输出误差,提高了识别精度,PSO-BP算法明显改善了传统BP神经网络的泛化能力。

Since the traditional pulse diagnosis is easily affected by subjective factors,and its result has low reliability,a pulse recognition method based on PSO-BP neural network is put forward.The fitness function judging the particles in PSO algorithm adopts the neural network to output the error,so as to obtain the position vector of the optimal particle.The value of the position vector is taken as the initial weight and threshold of the BP neural network.The ANN model based on BP algorithm,PSO-BP algorithm and GA-BP algorithm was established in Matlab to recognize the pulse signal.The experimental results show that the optimized algorithm can reduce the output error of the traditional BP neural network and improve the recognition accuracy for pulse recognition.The PSO-BP algorithm can improve the generalization ability of the traditional BP neural network significantly.