优化的神经网络在模拟电路故障诊断中的应用

Optimized BP neural networks based on PSO for analog circuit fault diagnosis

在检测和诊断模拟电路的故障中,为了解决传统的BP神经网络初始权值和阈值是随机选取的,容易陷入局部最小值,出现不收敛或收敛速度慢的缺点,文中提出了采用粒子群算法来优化传统BP网络的方法。首先利用粒子群算法优化BP神经网络的初始权值和阈值,然后再由改进的BP算法进行训练仿真。通过和在单一改进BP算法下训练的诊断仿真结果进行对比,可以发现通过粒子群算法优化后的神经网络改善了网络不收敛的缺陷,在训练速度以及诊断正确率上都有所提高。

In the detection and diagnosis of faults in analog circuits, of the traditional BP neural network is selected at random, it is easy to as the initial values and thresholds fall into local minimum, and tends to the fact that there is no convergence or slow convergence, the method based on particle swarm optimization （PSO） and back-propagation （BP） neural networks is proposed. At first, optimizing the initial weights and thresholds of BP neural network, and then training and simulating with the improved BP algorithm. Comparing the simulation results with those only with a single improvement in BP algorithm, we can find that the neural network improved the convergence defect after being optimizing by genetic algorithm, and the speed and accuracy of diagnosis are also improved.