41 results about "Multidimensional scaling analysis" patented technology

The invention provides an intelligent power distribution network fault identification method. A wide area measurement system is used to upload information, and a multidimensional state monitoring matrix is formed by data preprocessing and data fusion, and then the multidimensional analysis of the matrix is carried out, and the LOF values of the matrix are calculated, and according to the LOF values of the various nodes of the matrix, the fault detection and the fault positioning of the intelligent power distribution network are realized. Wavelet transformation of three-phase voltage of a faulted part is carried out, and the wavelet singular entropy values of the three-phase voltage are used to build a fault characteristic sample database. Pre-classification of faults is carried out by using zero sequence voltage low frequency energy reflecting grounding fault information, and on the basis of the above mentioned pre-classification, an SVW fault type discrimination prediction model is built. The effective fault detection and the effective fault positioning of the intelligent power distribution network are realized, and the reasonable classification of the different fault types of the faulted areas is realized.

An isometric mapping based facial image recognition method belongs to the technical field of image processing and comprises the following steps: processing the input images and expressing the images in the form of vectors; taking column vectors as training samples to constitute a training sample set, and then taking each training sample in the training sample set as a vertex to construct a neighborhood linked undirected weighed graph GX; computing the shortest path between every two training samples according to the neighborhood linked undirected weighed graph GX and establishing the shortest path matrix DG=(dG (i, j)); carrying out direct linear classification processing on the shortest path matrix and obtaining the optimal projection matrix W; and acting the optimal projection matrix W on a test set Dtest to reduce dimensions, thus obtaining the low-dimensional optimized feature samples, and then classifying the test set by the minimum distance method or the K-nearest neighbor method to obtain the face recognition results. The method simultaneously takes the features of the spatial relations between image pixels into consideration, uses direct linear discriminant analysis to replace multi-dimension analysis and obtains higher recognition rates in the experiments of face recognition.

The invention provides an indoor positioning method based on Wi-Fi fingerprints and multi-dimensional scaling analysis. In the method, acquisition of user's behaviors and Wi-Fi signals is performed with crowdsourcing thoughts to replace the present site investigation based on the fingerprint method; a fingerprint model identifies position information of each Wi-Fi signal on a user track, distributes the position information to the corresponding position and acquires fingerprint information of each sampling point; the standard K-near neighbor algorithm is improved and different weights are distributed to each wireless access point, so that the positioning of room stage can be realized; in order to obtain users' absolute coordinate positions and describe user's walking tracks in a room, the multi-dimensional scaling analysis method is improved; the results of the multi-dimensional scaling analysis method is further corrected by using the corner of a corridor and the door of the room as the key points, so that the user's more accurate absolute coordinate positions can be obtained. According to the indoor positioning method based on Wi-Fi fingerprints and multi-dimensional scaling analysis, the site investigation stage is omitted and the indoor positioning precision is improved.

The invention discloses a self-localization method of a sensor network node based on a smartphone. The method uses a general smartphone to replace a custom hardware module to be used as a network node, a plurality of mobile phone nodes orderly transmit linear frequency modulation (LFM) sound signals of 2k-6kHz by means of a loudspeaker and a microphone, and meanwhile, different mobile phone nodes sample the linear frequency modulation sound signals at a fixed frequency (44.1 kHz). Detection is performed on a sampling waveform by means of a generalized correlation method, and in view of a multipath effect, the self-localization method of the sensor network node based on the smartphone provided by the invention adopts a method of combining a threshold value method and time-frequency analysis to effectively inhibit the multipath effect, so that the arrival time of the linear frequency modulation sound signals can be obtained, then distance information between different nodes can be obtained, and finally, the unknown nodes can be located by adopting a multidimensional scaling (MDS) algorithm. The method of the invention does not need clock synchronization between the mobile phone nodes, and since the frequency is fixed, the arrival time of the sound signals can be accurately estimated through a sampling number; besides, the location accuracy is high, the cost is low, the networking is convenient and the prospect is wide.

The invention proposes a method for identifying physical faults and information faults of a distribution network. The basic idea of ​​the present invention is to preprocess and fuse the data uploaded by the intelligent terminal in the distribution network to establish a high-dimensional space-time state monitoring matrix, and after the matrix is ​​subjected to multi-dimensional scale analysis and dimensionality reduction, outlier detection is performed on it. Obtain the LOF value of each node, on this basis, the detection of faults in the smart distribution network can be realized, and then according to the LOF value of the generalized nodes, it can be identified whether the fault is a power fault in the distribution network or a communication node fault; further, combined with multi-dimensional scale The analysis and visualization results of outlier detection enable the location of distribution network fault areas and fault communication nodes. The invention can not only effectively detect the fault of the intelligent distribution network, but also locate the fault area and the fault communication node of the distribution network.

The invention relates to a hydraulic pump fault diagnosis method based on local mean conversion and Softmax. Through analyzing vibration signals collected from a hydraulic pump, the real-time state of the hydraulic pump can be obtained. According to the method, firstly, LMD (local mean decomposition) is used for decomposing the vibration signals into a plurality of PF (product function) components; then, the PF components including fault information are analyzed; feature parameters such as energy and corresponding time domain statistics are extracted; then, MDS (multi-dimension scaling) is used for carrying out feature reduction; after the reduced features are obtained, a trained logistic model is used for carrying out health evaluation on the hydraulic pump; in the health evaluation process of the hydraulic pump, if the fault occurrence is detected, a trained Softmax regression model can diagnose a possible fault mode. The hydraulic pump fault diagnosis method has the advantages that the health state of the hydraulic pump can be effectively evaluated, and the fault diagnosis can be carried out.

The invention discloses a method for classifying, optimizing and analyzing a website based on a user mental model. The method comprises the following steps of: firstly preprocessing log data of the website, wherein the log data contains the data relative to a concept of optimizing of a website classified catalogue issued by a user based on cognition of the user on the website classified catalogue, and extracting the concept from the log data by preprocessing; then determining a co-occurrence relation between the concept issued by the user and the concept in the website classified catalogue by a user mental model classification theory, wherein the concept presents a specific name of the website classified catalogue, such as books and daily articles; converting the co-occurrence relation into a co-occurrence matrix; converting the co-occurrence matrix into a similarity matrix by virtue of a pearson coefficient algorithm; and finally carrying out clustering analysis and multi-dimensional dimensional analysis to analyze similarity and spatiality among concepts of the cognition of the user on the website classified catalogue. Due to the adoption of six steps, decision supports can be provided to the optimizing of the website classified catalogue from a quantified angle based on the user mental model of the website.

The present invention relates to a method for evaluating water quality by using microbial diversity indicators in water sediments. According to the method, the structure of microbiological community in sediments is determined by using the Miseq high throughput sequencing, the relationship between the microorganism and the living environment is researched by using such statistic methods as principal component analysis, and non metric multidimensional scaling analysis, and the PA and BGN are calculated, wherein beta evaluates microbial indicators to determine the water eutrophication level and pollution level. According to the present invention, the microbial community characteristics of different types of water sediments in different pollution types can be objectively and comprehensively obtained by using the high throughput sequencing technology, water bodies and water qualities are evaluated integrally, and the selected microbial evaluation indicators can sensitively reflect the change of the water quality and the adaptability is wide, which can accurately detect all types of water sediments.

The invention provides a RSS data flatting method based on a multi-dimension analysis algorithm, which aims at solving the defects of reducing Radio Map precision and locating precision greatly caused by the reduced RSS data collection, environment noise and increased influence of the collection error on Radio Map in the prior art. The specific process is as follows: 1. arranging m AP in the indoor area to be located, 2. obtaining similarity distance dik and djk, 3. obtaining similarity matrix among different RP, 4. obtaining the similarity distance of noise RP and other RP, replacing the corresponding value in the similarity matrix on the data space in step 3, 5. calculating and obtaining RSS', replacing RSS in step 1; and 6, repeating step 4 and step 5 to realize flatting the RSS data. The flatting method provided in the invention is used for locating indoor area.

The invention discloses a competitive sports teamwork mode analogy method based on chordal graph visualization. The method comprises the following steps: 1) establishing a foundation database of game related information; 2) constructing a chordal graph according to the assisting information of players in a single game live broadcast record, i.e. dividing each player into a scoring sector and an assisting sector, respectively representing a scoring record and an assisting frequency record of each player, connecting the corresponding assisting sectors and scoring sectors of both cooperated parties by a chord so as to obtain a detail record of a game; 3) clustering the players with a multidimensional scaling method according to the historical record data of mutual assisting information of players in a team to obtain the two-dimensional coordinates of m vectors so as to obtain the clustering information of m players. Compared with the prior art, the competitive sports teamwork mode analogy method has the characteristics that chordal graph visualization design is adopted so as to bring convenience for users to better know and analyze game score distribution and player cooperation situations.

The invention provides a fault data processing method and system and a fault prediction method, which are applied to a rolling bearing, and the fault data processing method comprises the following steps: collecting time domain vibration signals of the rolling bearing under different fault conditions; for the time domain vibration signal under each fault condition, carrying out variational mode decomposition on the time domain vibration signal to obtain a plurality of two-dimensional images; according to the two-dimensional image, using a histogram of oriented gradient algorithm to extract and obtain corresponding image features; and carrying out dimension reduction processing on the image features by adopting a multi-dimensional scaling analysis algorithm, and carrying out processing to obtain fault data corresponding to the current fault condition. According to the method, the collected time domain vibration signals are converted into the digital images, the intuition is high, the fault detection problem is converted into the image recognition method, the traditional characteristic parameter operation process is converted into the gray value operation process of the gray image, the characteristic operation time is greatly shortened, and the processing process is simplified.

The invention relates to a transformer oil dielectric loss regression prediction method based on multi-frequency ultrasonic detection, and belongs to the technical field of transformer detection. The method comprises the following steps: S1, carrying out multi-frequency ultrasonic detection; S2, performing multi-dimensional scale analysis (MDS); S3, establishing a back propagation neural network (BPNN); S4, obtaining a global optimal solution by using a particle swarm optimization (PSO) algorithm; and S5, establishing a transformer oil dielectric loss prediction model based on the MDS-PSO-BPNN. According to the invention, the relationship between the ultrasonic characteristic value and the transformer oil dielectric loss is established, so that the transformer fault can be detected by multi-frequency ultrasonic waves, the problems of huge and complex structure and tedious operation of a traditional detection system are solved, and the operation state of the transformer oil can be circularly monitored on line.

The invention belongs to the technical field of primary and middle school education knowledge map analysis, in particular to a primary and secondary education knowledge map analysis system based on co-word and co-citation analysis, to determine the typical primary and secondary education knowledge publications related to knowledge map, the authors of the downloaded typical primary and middle school education knowledge journals are analyzed by co-citation and a co-citation matrix is generated, the co-citation matrix of the journal is analyzed in multi-dimension scale, After clustering analysisand factor analysis, the main academic knowledge and the knowledge map of the hot spots in primary and secondary education are drawn, As that prior art exist, the invention solves the problem that theprior art exists in order to better understand the research hotspot situation of the current domestic and foreign academic circles in the field of primary and secondary education, In order to providereference for the development of teacher education in primary and secondary schools in China, the paper reveals the three major academic groups in this field and their concerns, educational academichot spots and their research groups are very easy to judge the beneficial technical effects.

The invention discloses a product competition relation visual analysis method, device and equipment, and relates to the technical field of computers. The method comprises the steps of obtaining high-dimensional product coordinates and high-dimensional audience user attribute coordinates through a corresponding analysis method according to a preset audience user obtaining strategy of a product andpreset audience user attributes; based on the high-dimensional product coordinates and the high-dimensional audience user attribute coordinates, calculating a distance matrix of each product and eachattribute by adopting an Euclidean distance; carrying out smoothing processing on the distance matrix; performing dimensionality reduction mapping on the smoothed distance matrix into a coordinate matrix in a two-dimensional space by adopting a multi-dimensional scale analysis method; and drawing coordinate points corresponding to the coordinate matrix in a two-dimensional space coordinate graph,and displaying a competitive relationship between products. The technical problem that the coordinate points are excessively concentrated or scattered, and consequently graph display is disordered issolved. The beneficial effects of accurately and effectively displaying the competitive relationship between the products and facilitating the analysis and use of advertisers are achieved.

The invention discloses an indoor WLAN signal plan mapping and positioning method based on multidimensional scaling analysis. The method comprises the following steps of observing the motion path mode of a user in a target area in an off-line stage and acquiring a signal sequence according to the observed motion path mode; using an MDS method to perform dimension reduction on the acquired signal sequence and drawing signal plans corresponding to each path mode; converting the signal plans into gray scale images and performing feature extraction and nerve network training; and in a positioning stage, converting a newly acquired signal sequence into gray scale images and further performing feature extraction, performing mode discrimination on the newly acquired signal sequence by using the nerve network acquired through training and further acquiring a user motion curve and direction by estimating. The method can be applied to a radio communication network environment. The method which is mainly used for indoor wireless local area network positioning solves the problem in a conventional fingerprint positioning method that great manpower and material resources are required for investment.

The invention discloses a track multiscale analysis method used for protein structure prediction. A binary association graph is adopted for representing global folding information of protein conformations; structure difference among different conformations is represented with a Hamming distance; by constructing a global difference information matrix of a search track, the difference information matrix is converted into a two-dimensional plane structure difference information graph by adopting a multidimensional scale analysis method; and through the plane structure difference information graph, a structure relationship of all the conformations in a search process can be obtained, and tracking and viewing of the search track are realized. The method is applied to the protein structure prediction; and by analyzing the search track, tiny structure difference can be accurately captured and information of structural composition modes such as a folding type, a contact mode and the like can be displayed.