39results about How to "Improve anomaly detection performance" patented technology

This application relates to a graph neural network-based abnormality detection method for electric power Internet of Things equipment, which includes the following steps: S1: using data collection tools to collect traffic data and business data of different electric power Internet of Things devices; S2: analyzing the collected data Koopman analysis; S3: Construct the graph structure of the power Internet of Things; S4: Construct a graph neural network model using graph models as input, and use graph convolution and graph attention networks to update node feature states; S5: Use K-Means clustering Anomaly detection is performed on nodes at a certain moment. The present invention captures the nonlinear dynamic characteristics of the power Internet of Things data by introducing Koopman analysis; introduces the graph convolutional neural network, extracts the spatial characteristics of the power Internet of Things, and fuses the attributes of the device nodes themselves and the neighborhood device nodes in the topological structure of the Power Internet of Things The information realizes the anomaly detection of the power Internet of things, effectively improving the accuracy and stability of the detection.

The invention provides an abnormal detection method for industrial control network signals based on a deep learning structure, and relates to the technical field of abnormal value detection in industrial control network data. The present invention aims to solve the problem that it is difficult to detect a small amount of abnormal values ​​because it needs to be artificially defined for distinguishing normal data and abnormal values ​​in the existing methods. Select part of the data from the industrial control network data as a training sample, perform data normalization and standardization operations on the training sample, obtain the normalized calibrated data, and use the data enhancement algorithm to add some false positives to the normalized calibrated data The sample values ​​form the detected data; the normal data and the detected data are input into an autoencoder compression network for training, and the trained data are obtained respectively; the data are input into the comparison network and calculated by the deep neural network to obtain The distance between the normal data and the detected data, using a classifier to determine the abnormal value in the detected data according to the distance. It is used for signal anomaly detection.

The invention relates to robust anomaly and change detection utilizing sparse decomposition. The present disclosure describes systems, non-transitory computer-readable media, and methods for determining latent components of a metrics time series and identifying anomalous data within the metrics time series based on one or both of spikes / dips and level changes from the latent components satisfying significance thresholds. To identify such latent components, in some cases, the disclosed systems account for a range of value types by intelligently subjecting real values to a latent-component constraint for decomposing the time series and intelligently excluding non-real values from the latent-component constraint. The disclosed systems can further identify significant anomalous data values from latent components of the metrics time series by jointly determining whether one or both of a subseries of a spike-component series and a level change from a level-component series satisfy significance thresholds.

The invention provides a feed shaft diagnostic device and a feed shaft diagnostic method for a machine tool. On the basis of a control signal used in a feed shaft of a normal machine tool, abnormality detection of a rolling guide mechanism in consideration of the influence of a weight change of a workpiece can be performed. The feed axis diagnostic device is provided with a feed axis diagnostic unit (11) that detects an abnormality in a rolling guide mechanism (10) in a machine tool that machines a workpiece while driving a moving body (9) along a guide surface of the rolling guide mechanism (10) by means of a ball screw (8). A feed shaft diagnosis unit (11) acquires a feed speed during non-machining and a load applied to a ball screw (8) at the feed speed, and detects an abnormality on the basis of an approximation function calculated on the basis of the relationship between a plurality of feed speeds acquired under a plurality of different feed speed conditions within a predetermined period and the load.

The invention discloses a monitoring video anomaly detection method based on multi-region variable-scale 3D-HOF. Firstly, the monitoring video is obtained as an input, and the video is partitioned, and then the variable-scale 3D-HOF features and optical flow direction information entropy in each partition are extracted. , and combined into the final detection features, and finally use the sparse combination learning algorithm to learn an initial sparse combination set in each partition, judge whether the new data is abnormal through the reconstruction error, and use the normal data to update the sparse combination set online. The application of the present invention not only solves the problem of perspective deformation in surveillance video, but also makes full use of the difference of motion information in different optical flow amplitude intervals to obtain more accurate motion speed information. The invention is suitable for abnormal detection of monitoring video, has low calculation complexity, accurate detection result and good algorithm robustness. The invention has wide application in the technical field of video analysis.