In order to detect and diagnose faulty components in analog circuits more efficiently, we propose to use the adaptive wolf pack algorithm to optimize the extreme learning machine(ELM). The method includes the adaptive genetic algorithm which effectively selects feature parameters to generate optimal feature subsets. They are then used to construct the samples which are input into the ELM network to classify the faults. Given that the connection weights between the input layer and hidden layer in the ELM network, and the deviation of the hidden layer can affect the learning speed and classification accuracy, we apply our method to optimize them and select the corresponding optimal value, thus improving the training stability of the ELM network and the success rate of fault diagnosis. The specific realization process of these methods is given through the diagnosis of two typical analog circuits, and their fault diagnosis rates are over 99%. Simulation results show that the method has good accuracy and stability for fault diagnosis of analog circuits.