Abstract: The loop pipe apparatus are used to simulate the different conditions of the actual pipeline and denoise the original signal by the wavelet method. Kernel⁃based Principal Component Analysis (KPCA) is used to extract the time⁃frequency domain eigenvalues of the leaked signals, and the final input vector of the neural network is obtained. Because the traditional BP neural network is easy to fall into local minimum when it is used to identify working conditions, the BP neural network is optimized by genetic algorithm (GA) and particle swarm optimization (PSO). Compared with the traditional BP neural network,the result show that the two optimized BP neural networks have stronger ability to identify leakage working conditions. Finally, from the two aspects of test accuracy and training time, two different optimization algorithms are compared and their different application situations are proposed.

Key words: BP neural network, Condition recognition, KPCA, Genetic algorithm, Particle swarm optimization (PSO) algorithm

摘要： 　利用环道实验装置模拟实际管道的不同工况，应用小波分析对原始信号降噪，并利用基于核的主成分分析方法（KPCA）提取处理后泄漏信号的时频域特征值，得到神经网络最终输入向量。由于传统BP神经网络在进行工况识别时容易陷入局部极小值，因此利用遗传算法（GA）和粒子群算法（PSO）对BP神经网络进行优化。结果表明，两种优化后的神经网络相较传统BP神经网络具有更强的识别泄漏工况能力。最后从测试准确度和训练时间两个方面，对两种不同优化算法进行对比并提出其不同的适用情况。

关键词: BP神经网络, 　工况识别, 　KPCA, 　遗传算法, 　粒子群算法