Research On Similarity Search Algorithm For Multidimensional Flying Data

With the occurrence of accident in aviation safety, the safety situation of civil aviation is not optimistic. In order to improve the reliability of scientific management in aviation safety and reduce the accident rate, the relevant flight data are collected and analyzed to identify potentially unsafe factors. It is to determine flight safety problems and provide the basis for the establishment of relevant preventive measures.High dimensionality of QAR and the uncertain relevance among them which makes the method to do the similarity search for time series in the low dimensionality is no longer applicable in such situation. Taking into account the specificity of the civil aviation industry,with the similarity search for QAR to ascertain the plane faults requires a special definition of the similarity. In this paper, expertise and Analytic Hierarchy Process algorithm are combined to be used to calculate the weightiness of different dimensionalities for the plane faults.Translate the QAR Data with the symbolic method, and then build a k-d tree index, which makes it possible to do the similarity search on multidimensional QAR data subsequences.Shape and distance are used together to define Similarity.The main research contents are summarized as follows: First, introduce the Analytical Hierarchy Process algorithms into the fault detection which used to determine the importance degree of each dimension attribute. Build matrix, calculate the weight of each dimension and detect the consistency, eliminate the influences of human factors with bias correction. Second,establish a database based on the theme of the flight phase, using piecewise aggregate approximation algorithm for data pre-processing, so that keep the sampling frequency of each dimension consistency. Third, similarity is defined using a combination of shape and distance.Translate the multi-dimensional subsequences into symbols. Build an index for the subsequences on the symbols with the k-d tree. Each node in the k-d tree contains a pointer that point to the list constituted by the information nodes which have the same characteristic value. Lastly, search the fault models in the k-d tree and get the original data by the list as a candidate set. Calculate the weighted distance in this set to find the fault subsequence. The high precision and the low cost have been proved by the experiments in this paper.