An Outlier Detection Method with Wavelet HMM for UUV Prediction Following

摘要

An inertia algorithm of outlier detection based on wavelet HMM (Hidden Markov Model) is proposed in this paper to handle with the inaccurate original data collected from sensors for UUV predictive following control. The Improved Recursive Wavelet Transform (IRWT) is used to reconstruct the original data and amplify the wavelet coefficients of outliers locally. Wavelet coefficients are updated with historical coefficients of data; therefore, it can be implemented in real-time. A distribution decision function is defined by HMM, which is the basis of pre-outliers detection that obviously different from normal data. The pre-outliers are redetected using inertia algorithm to improve the accuracy of results detected. Original data from lake experiment verify effectiveness and feasibility of the method proposed.