In the era of information big data, the dependence degree on analog circuits is getting more severe, which results in the requirement for diagnosis accuracy of analog circuits grow with every passing day. However, analog circuits are very difficult to diagnosis, as a result, it is the bottleneck of electronic system diagnosis and maintenance. In this paper, an IHHO-SVM combining AVMD and PE and manifold learning is put forward. Firstly, adaptive variational modal decomposition AVMD is used to obtain IMF signals from observable signals of circuit under test, which could not only suppress noises disturbance, but also adaptively determine the number of IMF signals and improve the decomposition accuracy. Then IMF signals are computed with permutation entropy (PE) to construct fault features in order to fully reflect the local characteristic of IMF signal at different time span. Based on all these works, t-distributed stochastic neighbor embeddings(t-SNE) is combined to realize dimensionality reduction while remaining excellent discrimination power of fault features vectors, with the new feature vector formed at last. Finally, Harris Hawks algorithm is combined to optimize the support vector machine, which is called HHO-SVM here, for fault classification. The simulation tests show that the algorithm proposed in this paper has an excellent accuracy of 100%.