91 results about "Local outlier factor" patented technology

The invention relates to the field of topological structure identification of a low-voltage power distribution network and particularly relates to a low-voltage transformer district user phase identification method based on smart meter voltage curve similarity analysis. The method comprises a step of extracting a transformer district transformer and the smart meter voltage sequence data of a userfrom power consumption information collection system, a step of calculating a DTW distance between different user voltage sequences and calculating a local outlier factor of each user based on the DTWdistance and judging whether the connection relationship of the transformer district of the user is correct or not, and a step of calculating the DTW distance between each user of a correct transformer district transformer connection relationship and A, B and C phase voltage sequences of the transformer district transformer, wherein the user phase is a phase with a smallest DTW distance in the A,B and C phases. According to the method, the low-voltage transformer district user phase identification can be carried out in an online way, the manual on-site patrol is not needed, and the problemsof low recognition accuracy, low work efficiency and high cost of the low-voltage transformer district user phase identification are effectively solved.

The invention discloses a phase identification method for low-voltage district consumers based on voltage curve clustering analysis. The phase identification method comprises the following steps: first, extracting data of a district transformer with relatively-serious three-phase imbalance and voltage sequences of smart electric meters of the consumers using the district transformer from an electric information acquiring system; second, calculating the discrete Frechet distance between voltage curves of different consumers, calculating a local outlier factor of each consumer based on the discrete Frechet distance and further judging whether the connection relation of the district transformer of the consumers is correct or not; third, clustering the district consumers into three different consumer groups for a consumer set with the correct connection relation of the district transformer according to a K-medoids algorithm based on the discrete Frechet distance between the voltage curvesof different consumers, and further realizing accurate identification of the low-voltage district consumers. The method has the advantages that phase identification of the low-voltage district consumers can be carried out online without artificial field survey, and the problems of low accurate rate, low working efficiency and high cost of the phase identification of the low-voltage district consumers are effectively solved.

The invention provides a cloud application fault diagnosis system based on statistical monitoring. The cloud application fault diagnosis system based on statistical monitoring comprises a monitoring agent, a running status tracker and a fault detection locator, wherein the monitoring agent is used for collecting monitoring information when a cloud application runs; the running status tracker is used for abstracting the running status of the system to be local outlier factors and correlation coefficients; the fault detection locator is used for analyzing the running status of the system according to monitoring data provided by the running status tracker so as to detect a fault and locate reasons of the fault. According to the method, the running status of the system is described from utilization and performance of system resources according to the monitoring data through the local outlier factor and kernel canonical correlation analysis method, the system fault is detected according to a control chart, and abnormal measurements are located through a feature selection method. The method has the advantages that related knowledge of the application such as a software system structure and parameter estimation is not needed, and the method is simple, easy to implement and wide in application range; various faults of the cloud application can be automatically detected without manual intervention, and the abnormal degrees of the measurements can be quantized.

The invention discloses a wind turbine generator state-of-health online monitoring and fault diagnosis method based on SOM-MQEs and an SFCM. The method comprises the steps of firstly, processing obtained real-time data according to a local outlier factor algorithm, a partial least squares method and a Laplacian Eigenmap dimension reduction technology, and extracting important characteristic parameters influencing the state of health of wind turbine generators; secondly, inputting the characteristic parameters to an SOM-MQE state-of-health evaluation model, calculating health decay indexes of the wind turbine generators, and evaluating the state of health of the wind turbine generators; finally, utilizing a fuzzy c-means soft clustering algorithm for clustering analysis on operating data ofwind turbine generators with abnormal states to determine the fault types of the wind turbine generators. By means of the method, the state of health of the wind turbine generators can be monitored accurately in real time, faulty parts are accurately positioned, the accuracy of detecting the state abnormality of the wind turbine generators reaches 99.9% or so, and a guiding idea is provided for corresponding maintenance by maintenance personnel aiming at the real-time operating condition of a draught fan.

The present invention provides a method of identifying outlier traffic data, characterized in that the method firstly collects the traffic data, calculating data average local outlier factor, then judging the outlier data by one of the following two norms: the maximal m data of the average local outlier factor is outlier data, or the average local outlier factor more than the preset threshold is outlier data, finally the identified outlier factor is corrected by deleting or adopting filter method, or analyzing the concealed message contained in the outlier data. The method effectivily detect outlier data of boundary and interior, the effect precedes the outlier detecting method based on the statistics.

The invention provides a batch process online fault detection method of a dynamic multi-direction local outlier factor algorithm, relating to a batch process fault detection method. Firstly, three-dimensional data is expanded into two-dimensional in the sliding window of a training sample, and the standard processing is carried out. Then k neighbors of a training set (i) are found in each window, and a local outlier factor algorithm is used to calculate a reachability distance and a local reachability density to obtain an LOF statistical amount. The control limit of the LOF statistical amount at that time is calculated through nuclear density estimation. K neighbors of new time data are founded in the training set, and the LOF statistical amount at that time is calculated by using a local outlier factor algorithm. If the statistical amount exceeds a control limit, the data sample at that time is failed, otherwise, the data sample is normal. If a test indicates that a system is failed, the staff needs to identify a situation timely and eliminate danger. According to the method, the process monitoring can be carried out effectively, and a fault detection effect is improved.

The invention discloses a multi-modal process fault detection method based on a local neighbor standardization matrix, and relates to an industrial process fault detection method. The method enables historical data at a normal state to serve as a training set of modeling data, and carries out multi-mode process modeling and fault detection through employing a local neighbor standardization matrix method. The method comprises the steps: carrying out the preprocessing of unequal-length batch data through employing a local weighting algorithm, determining the maximum retainable length of the unequal-length batch data in the training set, and reconstructing lost data points of the unequal-length batch data through weighing and k-neighbor information; constructing a main local neighbor standardization matrix for an equal-length training set, carrying out modal clustering through employing a K-means algorithm, and eliminating off-cluster samples at all modals through employing a local off-cluster factor method. The method can prevent information loss from affecting the modal clustering effect of a multi-modal process, eliminates off-cluster points, and enables the fault diagnosis result of a multi-modal intermittent process to be more accurate through the construction of the main local neighbor standardization matrix.

The invention belongs to the technical field of electric power detection, particularly relates to the technical field of electric power electricity stealing detection, and particularly relates to a method for detecting electricity stealing on the basis of electricity utilization behavior patterns of electric energy users and application. The method and the application have the advantages that themethod starts from electricity utilization behavior characteristics of the users, factors such as behavior habit, climate and seasons are considered according to electricity utilization energy consumption characteristics of the different electric power users, and different user energy consumption time-sharing models are built; the problem of incapability of accurately determining electricity stealing behavior of users under the consideration of reasonable electricity utilization behavior change of the electric energy users after the anomaly of electric energy measurement data of the users is detected on the basis of K-means clustering algorithms and LOF (local outlier factor) algorithms can be solved by the aid of the method and the application; the method is combined with electricity utilization information acquisition systems for measuring electric energy loss, the electricity stealing probability is introduced according to energy consumption formulas in districts, the electric energy electricity stealing probability can be computed, and accordingly different types of electricity stealing behavior of the electric energy users can be reliably monitored.

The invention provides a network traffic anomaly detection method and device, which relate to the technical field of information security and are used for solving the problem of low accuracy and efficiency of network traffic anomaly detection. The method comprises the following steps of: analyzing network traffic anomaly detection samples by using a preset clustering analysis algorithm and obtaining the distribution characteristics of data points for clustering analysis; eliminating dense data points in the clustering analysis result to obtain a preliminary anomaly detection data set; A presetclustering algorithm addinga referrer field to each data point to point the data point to the nearest neighboring data point;according to the preset outlier analysis algorithm, obtaining the first outlier set and the outlier number of the preliminary outlier detection data set, and adjusting the distance parameter value according to the preset outlier analysis algorithm, the number of outlier points and the number of local outlier factors to obtain a second outlier set, and taking the intersection of the first outlier set and the second outlier set as the network traffic anomaly detection result. The invention is used for detecting network traffic anomaly.

The invention discloses a data classification method based on improved local abnormal factor detection. The method comprises the steps of outlier factor detection; similarity measurement; selection ofan initial clustering center point, and screening of data with relatively small outlier factors as a candidate set of the initial clustering center through a local outlier factor detection LOF algorithm for adaptively adjusting k distance parameters; and iterative optimization on the clustering center. In the iteration stage of the optimized clustering center, the outlier factor between the datais standardized by using the outlier standardization, so that the value range of the new outlier factor new _ ri is greater than or equal to 1. According to the invention, the accuracy of cluster center positioning and cluster division is improved.

The invention discloses a method for identifying outlier data in effective parking lot occupancy. The method includes steps of 1), determining a time sequence of the effective parking lot occupancy of a parking place; 2), carrying out N-scale wavelet decomposition and reconstruction for the time sequence of the effective parking lot occupancy by a wavelet function to obtain N+1 reconstructed time sequences, and forming a data set D by the N+1 reconstructed time sequences; 3), computing weighted projection vectors of all data points in the data set D with N+1 dimensions to form a weighted data set DW; 4), computing a mean local outlier factor of the weighted data set DW by an outlier data mining algorithm based on density; and 5), judging outlier points on the basis of the mean outlier factor of the weighted data set DW. The outlier data of the effective parking lot occupancy are identified by the outlier data mining algorithm on the basis of wavelet analysis and local information entropy weighting according to the characteristics of periodicity and volatility of the time sequence of the effective parking lot occupancy, the misjudgment rate is reduced, and the reliability is improved.

The invention relates to the electricity-stealing-prevention technical field of low-voltage transformer areas and provides a local outlier factor-based low-voltage electricity-stealing user positioning method. According to the method, firstly, k transformer areas most similar to a monitored transformer area are searched based on feature attributes which affect the line loss of transformer areas, namely k nearest transformer areas are searched. Secondly, whether the line loss rate of the monitored transformer area is abnormal or not is analyzed and judged according to the line loss rates of k nearest transformer areas. If the line loss rate of the monitored transformer area within a certain period of time is abnormal, a discrete Frechet distance among the load curves of all users within theabove period of time in the monitored transformer area is calculated. Finally, based on the discrete Frechet distance among the load curves of users, the local outlier factor of the load curve of each user in the monitored transformer area is calculated. The larger the local outlier factor, the larger the electricity-stealing probability of the user. Based on the method, the rank ordering resultof the electricity-stealing probabilities of all users in the monitored transformer area with the abnormal line loss rate is outputted. Therefore, most of electricity-stealing users can be detected and found out only through detecting users ranked in the top of the rank ordering result. As a result, the electricity-stealing prevention working efficiency is greatly improved.

The embodiment of the invention provides an outlier data identification method and system in catalytic cracking device data. The method comprises the following steps that: according to preset sampling time, obtaining the original attribute value of catalytic cracking production data, and obtaining the time series set of the original attribute value; using a sliding time window to reconstruct the time series set, and obtaining the sub-series set of the original attribute value, wherein the sub-series set comprises a plurality of sub series; obtaining the interactive weight vector of the sub-series set, and constructing a weighted series set according to the interactive weight vector and the sub-series set, wherein the weighted series set comprises a plurality of weighted sub series; and obtaining the local outlier factor of each weighted sub series in the weighted series set, and identifying the outlier data in the catalytic cracking device data according to the local outlier factor. The system is used for executing the above method. In the embodiment of the invention, outlier data identification accuracy in the catalytic cracking device data and the reliability of the catalytic cracking device data are improved.

The invention discloses a method for outlier data mining based on parallel computation. By utilization of a local sparse degree on attribute dimensions, a relevant subspace is defined again, and thus distribution characteristics on various local data sets can be effectively depicted; by utilization of the probability density of the local data sets, a local outlier factor calculation formula is given, the degree of data objects disobeying local data set distribution characteristics is effectively embodied, and N data objects with a maximum outlier degree are selected to be defined as local outlier data. According to the method for outlier data mining based on parallel computation, a sparsity factor and an outlier factor are calculated through a Map; when values of the factors are subjected to total ordering, one Map is used to sample the factors, a function for determining what node each (K2,V2) is distributed to is implemented, and thus the problems that the data distribution is uneven and data are locally ordered and globally disorderly are effectively solved.

The invention discloses an access traffic anomaly detection method and device based on LOF and isolated forest, and the method comprises the steps: traffic preprocessing: preprocessing access traffic data to obtain a traffic data set, the preprocessing comprising traffic extraction, traffic cleaning and traffic normalization; machine learning model training: taking the traffic data set as model input to carry out machine learning training, respectively using a local outlier factor (LOF) detection algorithm and an isolated forest algorithm to carry out multiple iteration training, and obtaining two groups of trained anomaly detection models, namely N anomaly detection models, and storing the anomaly detection models; and performing combined intelligent analysis: performing target flow detection by using the two groups of anomaly detection models trained in the step 2, and performing result screening by using a bagging Bagging algorithm. According to the method, a local outlier factor LOF detection algorithm is combined with an isolated forest to perform conjoint analysis, anomaly detection is performed on the collected access traffic category, whether the traffic is abnormal or not is judged, and the security of the system is ensured.

The invention discloses an SQL injection attack detection method based on a dynamic clonal selection algorithm. The method comprises the steps of extracting SQL statements submitted by a client browser and submitting the SQL statements to a rule base detection module for SQLIAs (SQL Injection Attacks) rapid mode matching; submitting the statements with unsuccessful mode matching to a dynamic detection module for deeper detection, introducing a local outlier factor as a fitness function for optimization of the distance between detectors before the dynamic detection module is operated, and constructing efficient detectors to identify normal data and abnormal data; operating the dynamic detection module to carry out detection; calling a clonal selection algorithm to carry out learning and identification; and updating system modules according to a detection result. The problem that the detection performance of an existing WAF (Web Application Firewall) is reduced and unknown features and various deformation attacks cannot be identified is solved.

The invention discloses a method for monitoring and evaluating the actual state of an MOV based on outlier detection. The method comprises the steps of: carrying out the sampling of various actual parameters at different time points at the moment of the operation of the MOV for several times, and forming an original data set; performing principal component analysis on the original data set to reduce the dimension of the collected features, and achieving visualization on the two-dimensional or three-dimensional level; then, calculating a distance between two adjacent points in the data set after PCA, a kth distance, a kth distance neighborhood, a reachable distance and a local reachable density according to a local abnormal factor; and finally, calculating a local outlier factor to obtain an outlier detection evaluation result of any sampling data of the MOV. According to the invention, the executable difficulty of traditional online monitoring is reduced, the method for monitoring andevaluating the actual state of the MOV based on outlier detection has the higher industrial application value, enhances the effective way of monitoring and maintaining the lightning protection deviceby people, and avoids the risk conditions of damage, fire, explosion and the like of an electrical system caused by abnormal operation or degradation of an MOV device.

The invention discloses a method for extracting the center of a structured light strip of an asphalt pavement image. The method comprises steps of dividing an image through employing a light strip gray threshold, cutting an ROI (region of interest) of the light strip of the image, and carrying out the subsequent extraction algorithm in the ROI; adopting an unbiased extraction method of a Steger curve stripe center to obtain a light bar sub-pixel center point; a local outlier factor (LOF) detection method being used for the light bar center point extracted by the Steger, outliers being found and removed by calculating the local density of each point, and the accurate light bar center point being obtained. According to the structured light strip center extraction method for the asphalt pavement image, the light strip center extraction effect is obviously improved, interference is removed for the characteristics of noise points, unnecessary convolution calculation is avoided, calculationefficiency is high, and the method has high universality and robustness and is suitable for large-scale popularization and application. The method is applied to pavement surface three-dimensional scanning and apparent reconstruction, and extraction precision and recognition efficiency of the apparent information of the asphalt pavement are improved.

The invention discloses a deviation feature-based outlier mining method. The method comprises The following steps of: (1) dividing each dimensionality of a data set into h uniformly-spaced intervals so as to divide the whole data set into hd grids; (2) associating each data point with the grids, and if one grid does not comprise a data point, not considering the grid; (3) for each grid in a spaceformed via the division, solving a centroid of the grid and calculating a local outlier factor of the centroid; and (4) calculating a local outlier factor of each data object, wherein the local outlier factors of the objects in the data set are equal to the outlier factors of the centroids of the grids. According to the method, when outliers in the data set are detected, the data space is dividedinto the grids by adoption of an F-LOF detection algorithm, and the local outlier factors of the data points are calculated on the basis of the centroids of the grids, so that the calculation time isshortened, the detection efficiency is improved and the superiority is embodied.

The invention relates to an intelligent fault detection method for an acquisition operation and maintenance system. Data mining and analysis of the acquisition data of the acquisition operation and maintenance system is carried out, and the fault detection of acquisition equipment is carried out based on a Gaussian kernel density local outlier factor algorithm. The collection operation and maintenance system intelligent fault detection method comprises the steps of feature extraction, principal component analysis, local outlier factor calculation and the like. The method comprises the steps: feature extraction is performed on collected operation and maintenance system data; each intelligent collection device is mapped to a two-dimensional plane through principal component analysis, data visualization is achieved, local outlier factors are convenient to calculate, then the local outlier factors of each intelligent collection device are calculated on the basis of a Gaussian kernel density local outlier factor algorithm, and finally suspected fault probability sorting of all the intelligent collection devices is output. The method is less influenced by local data distribution, is suitable for the condition that a data set lacks training samples, provides reference for acquisition, operation and maintenance detection of an electric power company, improves the hit rate of field inspection, can save a large amount of manpower and material resources, and has great economic benefits.

The invention relates to an intelligent diagnosis method for a fault reason of a capacitor voltage transformer. The method comprises the following steps: acquiring data to generate a training set D; the training set D is key value pair data mapped from an input data set X to an output data set Y, the input data set X is error evaluation value data, and the output data set Y is transformer fault reason data; based on an LOF (Local Outlier Factor) algorithm, carrying out preprocessing of abnormal error data elimination on an input data set X in the training set; on the basis of the pre-processed training set, RNN model training of WOA (Whale Optimization Algorithm) parameter adjustment and optimization is carried out, and on the basis of the pre-processed training set, RNN model training of WOA (Whale Optimization Algorithm) parameter adjustment and optimization is carried out; based on the trained RNN model, fault cause prediction of the capacitor voltage transformer is carried out; and operation and maintenance personnel can find out the out-of-tolerance reason of the mutual inductor in time and carry out targeted maintenance, so that the field operation and maintenance efficiency is improved, the workload of the operation and maintenance personnel is reduced, and the operation and maintenance cost is remarkably reduced.

The invention discloses an intelligent electric meter fault rate estimation method and system under multi-environmental stress. The method comprises the steps of obtaining historical fault rate data of intelligent electric meters in different types of environmental regions, performing noise point detection based on a weighted local outlier factor of a weighted Euclidean distance, cleaning noise data, and obtaining a historical sample set; determining the form of a kernel function, and optimizing hyper-parameters of the kernel function; establishing a Gaussian process regression model, and training and testing the Gaussian process regression model by using the historical sample set; and inputting fault sample data of the to-be-tested intelligent electric meter, eliminating noise data based on the weighted local outlier factor of the weighted Euclidean distance, and obtaining the reliability of the to-be-tested intelligent electric meter through the trained Gaussian process regression model. According to the method, the change trend of the fault rate of the intelligent electric meter along with time under multi-environment stress can be effectively evaluated, and the reliability of the fault rate can be accurately solved.

The invention relates to a lithium battery pack consistency detection method and device based on a local outlier factor. The method comprises the following steps: acquiring voltage curves of a plurality of battery cells in a battery pack; calculating the distance between the voltage curve of each battery cell and the voltage curves of other battery cells by using the Euclidean distance to obtain a plurality of battery cell distances corresponding to each battery cell; calculating a local outlier factor score of each battery cell in the battery pack according to the plurality of battery cell distances corresponding to each battery cell; and based on the local outlier factor score of each battery cell, determining the battery cells inconsistent with other battery cells in the battery pack. According to the invention, the outlier cells of a lithium battery pack are detected by using local outlier factors, noise data can be effectively filtered, and the accuracy and effectiveness of abnormal cell detection are improved.

The invention particularly relates to a data quality evaluation method for a power grid dispatching automation system based on big data, which comprises the following steps: acquiring data of a dispatching system, classifying the acquired data according to equipment, and constructing data of each equipment into a sample space X; establishing a sample space X, carrying out KNN algorithm classification on all sample points p in the sample space X, and outputting samples which may be outliers in the sample space; according to the established sample space X, carrying out a local outlier factor algorithm LOF on all sample points p in the sample space X, and outputting samples which may be outliers in the sample space X; according to the established sample space X, outputting samples which may be outliers in the sample space X through an INFLO algorithm on the samples in the sample space X; comparing the results obtained in the step 2, and determining that the sample is an abnormal point ifthe occurrence frequency of the same sample in the three results is greater than or equal to two; outputting data of the anomalous points.

The invention relates to the technical field of water affair network pipe network detection, in particular to a water affair network pipe network pipe exploding detecting method based on local outlier factors. The method includes the steps that S1: detection data of each detection point in a to-be-detected pipe network at the current moment and historical detection data at the same moment in the past few days are collected; S2: according to the detection data of each detection point at the current moment and the historical detection data, the outlier factors of the detection data of each detection point at the current moment are calculated; S3: the space adjacent relationship of the detection points is acquired, and according to the outlier factors of every two adjacent detection points at the current moment, the pipe exploding probability of pipe segments between every two adjacent detection points is calculated; and S4, whether the pipe exploding probability of each pipe segment is larger than a preset threshold value or not is judged, if the pipe exploding probability of each pipe segment is larger than the threshold value, it is judged that pipe exploding happens to the pipe segment, and otherwise, it is judged that no pipe exploding happens. Due to the method, it is not needed to know about labels of data, practicability is achieved, and the pipe network pipe exploding detecting feasibility is higher.