An improved Hilbert-Huang transform method for stochastic signals

A non-stationary stochastic signal like the ground acceleration in civil engineering can be analyzed through the Hilbert-Huang transform (HHT) with an empirical mode decomposition (EMD) first. EMD methods often generate both upper and lower envelopes of the signal and the intrinsic mode functions (IMFs) using cubic spline interpolation. Unfortunately, there exist issues like the overshoot and undershoot of the envelopes, as well as the mode mixing. The mode mixing also results from the endpoint effects of EMD and the Hilbert spectrum itself. This study uses the piecewise cubic Hermite interpolation in the Newton form combined with the least squares support vector machines whose parameters are optimized by the particle swarm optimization algorithm to address the aforementioned issues. In order to verify the effectiveness and accuracy of the proposed method, this study uses stochastic ground acceleration signals. Three evaluation indexes are proposed for the verification. It turned out that the proposed method has better performance than the traditional HHT method.