Research On Improved Permutation Entropy Algorithm For Abnormal Change Detection

Abnormal change is a common phenomenon in nature and human society in which the individual is separated from the original existence state by the change of the environment or the state of the organism itself.To a large extent,the causes of abnormal changes are the result of long-term effects and cannot be intuitively discovered,and so its consequences are often unexpected.Therefore,the detection of abnormal changes has great practical significance for preventing the occurrence of unknown risks.PE algorithm,which was proposed by Bandt and Pompe,is a method of complexity measure of time series.It is widely used by scholars because of its simple concept and good robustness to noise.Many studies have shown that PE algorithm has significant advantages in anomaly detection of signal.However,with the deepening of the research,the shortcomings of PE and its derived algorithm are increasingly apparent.In order to improve the performance of PE algorithm in detecting abnormal changes,three PE optimization algorithms are proposed in this paper from macroscopic and microscopic aspects.The main work and innovations are as follows:1.PPE algorithm which introduce phase information into PE is proposed.Using logistic map,we analyze the difference of PE and PPE in detecting the chaos window,and then the influence of data length,embedding dimension,and time delay are also be studied.Relay on the AR model,the performance of PPE algorithm when the series have different length and SNR are discussed.Finally,we apply PPE algorithm to analyze the ECG and engineering signals.The simulation results demonstrate that PPE can improve the detection effect of abnormal changes compared with PE.Increasing embedding dimension can increase the capability of detecting abnormal changes using PPE algorithm.Furthermore,PPE is not sensitive to data length when embedding dimension is less than or equal to 5 and it is more susceptible to noise than PE.The results from actual signals show that PPE can be used as an effective analytical tool in the field of biomedical and engineering signal processing.2.Based on the formula of two-index entropy,we propose generalized phase permutation entropy algorithm,verify the advantage of PPEq,δ in detecting the dynamical changes of system,analyze the influence of q,δ selection on dynamical change of system,and explore the effect of data length and noise for PPEq,δ.Finally,PPEq,δ is applied to analyze abnormal ECG signals.The conclusion can be obtained that PPEq,δ has a more significant effect on the detection of the same dynamical changes when the value of q,δ are the same.Whether use in the logistic map or in detecting abnormal ECG signal dynamical changes,when q>0 and q>0,the effect of PEq,δ and PPEq,δ can be improved by decreasing q value,increasing δ value or simultaneously changing both value.Furthermore,the change in data length has no effect for PEq,δ and PPEq,δ and both of them are robust to noise.3.An improved PE algorithm named CPE is proposed to optimize the trivial problem existing in PE based on the secondary partitioning.The performance of CPE algorithm for dynamical change detection is analyzed using synthetic signal,logistic map and Lorenz map.The detection ability of CPE algorithm in different SNR is studied and the algorithm complexity is discussed.The simulation results show that CPE can accurately capture minor feature information and amplify the detection results of dynamical change compared with PE,WPE and AAPE,but it has less robustness to noise and requires a higher computation cost than the others.Finally,we use CPE to analyze the rolling bearing fault signals.The simulation of actual signal illustrates that CPE performs better in detecting abnormal pulse of the rolling bearing when the embedding dimension is small.