Research On The Ship Motion Pattern Based On Inertial Sensor

Ship motion pattern recognition is an important part of the ship’s intelligent perception system,and it has important application value in ship intelligent control and decision-making,remote monitoring,and danger early warning.This paper takes the inertial data generated by the continuous motion of the ship as the research object,and aims to realize the recognition of the five basic motion patterns of the ship.A series of researches have been carried out on the collection,preprocessing,segmentation and feature extraction of ship motion data and ship motion pattern recognition algorithm.The main contents include:(1)In order to improve the accuracy of ship motion pattern recognition,this paper designs a ship motion pattern recognition framework,which adds a data segmentation module on the basis of existing research.To meet the demand for ship motion analysis of time series,this paper builds a high-performance ship motion data acquisition module based on a nine-axis IMU composed of a three-axis accelerometer,a three-axis gyroscope,and a three-axis geomagnetic device.(2)To improve the quality of identification data,This paper analyze the components of the collected ship motion data and filter out the interference components in it.First,the original signal is subjected to outlier detection using a probabilistic model,and the outlier points are interpolated according to the correlation between the data;Secondly,use the mean method to reduce the zero drift error of the gyroscope;Select the ellipse fitting method to calibrate the geomagnet;According to the energy distribution characteristics of the gravitational component,this paper uses the third-order or fourth-order elliptic low-pass filter to eliminate the gravitational component in the three-axis acceleration signal;Finally,for the random noise in the original signal,an improved EMD denoising method is proposed,which uses wavelet threshold method to filter the high-frequency components decomposed by EMD,retaining the useful information in it,making up for the disadvantage of EMD denoising method of directly discarding the high-frequency components leading to incomplete signal information,and providing data support for the subsequent ship motion pattern recognition.(3)A data segmentation algorithm based on hidden Markov model is designed,which is based on hidden Markov theory,combined with segmented regression model and linear logistic transformation,and uses expectation maximization algorithm to find the model parameters in order to realize the segmentation of continuous ship motion data into different time windows,each window contains one independent and complete motion data and represents different ship motion patterns.After that,the variation characteristics of the motion parameters of each motion pattern segment of the obtained ship are analyzed,the mapping relationship between the ship motion parameters and the motion patterns is mined,and a variety of time-domain features that can characterize the ship motion patterns,such as mean,variance,quadratic spacing and variation difference,are extracted according to the mapping relationship between them.(4)A new algorithm for identifying ship motion patterns is proposed in this paper.This algorithm,referred here as binary tree support vector machines,The algorithm uses the maximum cut problem theory to find the maximum distance between samples to realizes the unsupervised construction of binary tree,and uses support vector machine for identification and classification at decision nodes.And also use the particle swarm algorithm to optimize the recognition model,and the effectiveness of the proposed method is verified by using the ship motion data collected from real ships as samplesThe experimental results show that the proposed recognition method only needs to train 5 sub-classifiers for the recognition of the six motion modes of the ship’s acceleration,deceleration,uniform speed,left turn,right turn and stationary,and the average recognition accuracy of the recognition reaches 96%.The effectiveness and superiority of the proposed method are shown.