CN-121214553-B - Intelligent recognition method and system for illegal behaviors of petrochemical field operation flow

CN121214553BCN 121214553 BCN121214553 BCN 121214553BCN-121214553-B

Abstract

The invention provides a petrochemical field operation flow violation intelligent identification method and system, which relate to the technical field of petrochemical safety production and comprise the following steps: the method comprises the steps of collecting video data for personnel target detection and action recognition, extracting equipment entity images and temperature data, recognizing equipment abnormal states by utilizing a self-adaptive threshold segmentation and local sensitive hash algorithm, constructing a temperature thermal distribution map to recognize a temperature abnormal region, extracting action track data, matching with a standard operation action sequence, calculating operation risk level and generating warning information.

Inventors

LEI HUAWEI
LIU NAN
ZHANG RUILING
YANG HUIHAO
SONG LIN

Assignees

河南油田工程科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20251013

Claims (9)

1. The intelligent recognition method for the illegal behaviors of the petrochemical site operation flow is characterized by comprising the following steps of: acquiring video data of a petrochemical field operation area, carrying out personnel target detection and personnel action recognition to obtain personnel target position information and personnel action information, determining an operation area range, extracting equipment entity images in the operation area range and acquiring temperature data; Performing self-adaptive threshold segmentation on the equipment entity image, constructing an equipment characteristic point cloud image, matching the characteristic point cloud image by utilizing a local sensitive hash algorithm, identifying an equipment abnormal state, constructing temperature data into a temperature thermodynamic distribution diagram in an operation area range, identifying a temperature abnormal area, calculating a temperature abnormal grade, constructing a time sequence combination characteristic according to the equipment abnormal state and the temperature abnormal grade, and calculating a risk degree score of a current operation scene; Extracting action track data in personnel action information, constructing action feature vectors by utilizing track shape descriptors, performing principal component analysis and dimension reduction processing to obtain key action features, matching the key action features with a standard operation action sequence by adopting a sequence alignment algorithm, calculating similarity scores, and grading illegal behaviors by combining the risk degree scores to obtain operation risk grades; when the working risk level exceeds a preset threshold, generating illegal action warning information, and sending the illegal action warning information to a site management terminal; Constructing time sequence combination characteristics according to the abnormal state and the abnormal temperature level of the equipment and calculating the risk degree score of the current operation scene comprises the following steps: acquiring an equipment abnormal state sequence and a temperature abnormal grade sequence, calculating the characteristic association degree of the equipment abnormal state sequence and the temperature abnormal grade sequence at adjacent time points, and carrying out normalization processing on the characteristic association degree to obtain a time sequence dependency matrix; constructing a time sequence causal graph, setting preset monitoring indexes as nodes of the time sequence causal graph, setting causal relations among the monitoring indexes as edges of the time sequence causal graph, calculating probability distribution among the nodes according to historical monitoring data, updating the edges of the time sequence causal graph, analyzing propagation paths of abnormal states in the time sequence causal graph, and calculating risk propagation coefficients based on node coverage areas of the propagation paths; calculating basic risk according to the equipment abnormal state sequence and the temperature abnormal grade sequence, determining self-adaptive weights according to the time sequence dependency matrix, and carrying out weighted fusion on the basic risk and the risk propagation coefficient based on the self-adaptive weights to obtain a risk degree score of a working scene.
2. The method of claim 1, wherein capturing video data of the petrochemical field operation area and performing personnel target detection and personnel action recognition, obtaining personnel target location information and personnel action information and determining an operation area range within which to extract equipment entity images and capture temperature data comprises: Collecting video data of a petrochemical field operation area, and carrying out frame segmentation processing on the video data to obtain a continuous video image sequence; Performing personnel target detection and personnel action recognition on the video image sequence, acquiring personnel target position information and personnel action information, and determining an operation area range according to time sequence change of the personnel target position information; and in the operating area range, extracting an equipment entity image, collecting temperature data of a corresponding area through a temperature sensor array, and establishing a spatial corresponding relation between the equipment entity image and the temperature data.
3. The method of claim 1, wherein the device entity image performs adaptive thresholding, builds a device feature point cloud image and matches the feature point cloud image using a locally sensitive hashing algorithm, identifies a device anomaly, builds temperature data into a thermal profile over an operating region, identifies a temperature anomaly region, and calculates a temperature anomaly level comprising: Collecting equipment entity images, sequentially carrying out Gaussian filtering noise elimination and histogram equalization on the equipment entity images, extracting an equipment target area by utilizing a self-adaptive threshold segmentation algorithm, detecting characteristic points in the equipment target area by adopting a rapid angular point detection algorithm, calculating the local gradient direction of the characteristic points to obtain characteristic descriptors, and constructing an equipment characteristic point cloud picture based on the characteristic points and the characteristic descriptors; Constructing a random projection vector, generating a local sensitive hash function according to the random projection vector, connecting a plurality of local sensitive hash functions in series to form a hash function set, calculating the similarity degree between characteristic points in the characteristic point cloud image of the equipment by utilizing the Ha Xihan set, matching the characteristic point cloud image, calculating a transformation matrix of a matching point pair based on a matching result, obtaining deformation parameters by calculating a norm difference value of the transformation matrix and a unit matrix, and identifying the abnormal state of the equipment according to the deformation parameters; Acquiring temperature data, performing bilinear interpolation processing on the temperature data to obtain temperature distribution data, performing normalization processing on the temperature distribution data to construct a temperature thermal distribution diagram, calculating local density values of all temperature points in the temperature thermal distribution diagram through a Gaussian kernel function, and determining the minimum distance from a current temperature point to other temperature points with local density values larger than the current temperature point for each temperature point; Calculating the product of the local density value and the minimum distance to obtain an abnormality score of the current temperature point, judging a temperature abnormality region by combining a preset abnormality score threshold value, and grading the abnormality score of the temperature abnormality region to obtain a temperature abnormality grade corresponding to the temperature abnormality region.
4. The method of claim 1, wherein extracting motion trajectory data in the personnel motion information, constructing motion feature vectors using trajectory shape descriptors and performing principal component analysis dimension reduction processing, and obtaining key motion features comprises: Acquiring personnel action information, extracting three-dimensional space coordinates of a human body joint point from the personnel action information to form action track data, and carrying out weighted smoothing processing on the action track data to generate smooth track data; Calculating displacement vectors of human body joint points in the smooth track data between adjacent time points, constructing a local coordinate system containing forward direction unit vectors, lateral direction unit vectors and vertical direction unit vectors, and projecting the displacement vectors to the local coordinate system to generate a track shape descriptor; Calculating displacement distance, projection angle and torsion angle according to the track shape descriptor to form a space shape characteristic, and calculating a speed vector, an acceleration vector and an angular speed vector of smooth track data to form an action characteristic vector; Calculating statistics of the space shape features and the motion feature vectors to form time sequence statistical features, constructing a feature correlation matrix, performing principal component analysis and dimension reduction processing on the feature correlation matrix to obtain a feature vector matrix and a feature value diagonal matrix, selecting principal feature vectors according to the accumulated contribution rate, multiplying the principal feature vectors by the time sequence statistical features to generate dimension reduction features, calculating normalized weights of the feature value diagonal matrix, and carrying out weighted summation on the dimension reduction features and the normalized weights to obtain key motion features.
5. The method of claim 1, wherein matching key action features to a standard job action sequence using a sequence alignment algorithm, calculating a similarity score and classifying violations in combination with the risk level score, the obtaining a job risk level comprising: Key action features are obtained, key action feature sequences are generated by combining the key action features at different time points according to time sequences, the key action feature sequences and preset standard operation action sequences are respectively decomposed in a layering mode according to preset layer numbers, and key multi-layer subsequences corresponding to the key action feature sequences and standard multi-layer subsequences corresponding to the standard operation action sequences are generated; Respectively extracting the mean value characteristic, the variance characteristic and the peak value characteristic of each subsequence in the key multi-layer subsequence and the standard multi-layer subsequence, and executing characteristic fusion operation to generate key statistical characteristics corresponding to the key multi-layer subsequence and standard statistical characteristics corresponding to the standard multi-layer subsequence; Constructing an initial layer distance matrix by utilizing the key statistical features and the standard statistical features, performing recursive calculation on the initial layer distance matrix based on a preset fusion weight to generate a distance matrix of each level, and selecting a distance matrix of the last layer as a target distance matrix; And calculating a minimum accumulated distance path based on the target distance matrix to generate an accumulated distance matrix, calculating to obtain a similarity score according to the accumulated distance matrix, the sequence length and a preset scaling factor, performing weighted combination operation on the similarity score and the risk degree score, and grading the illegal behaviors to obtain the operation risk grade.
6. The method of claim 1, wherein generating and transmitting the offensiveness warning information to the field management terminal when the work risk level exceeds a preset threshold value comprises: Receiving operation risk level information, wherein the operation risk level information comprises an operation risk level and a corresponding illegal action type, acquiring a preset threshold value, and comparing the operation risk level with the preset threshold value; When the operation risk level exceeds the preset threshold, generating illegal action warning information according to the operation risk level and the illegal action type and sending the illegal action warning information to a site management terminal corresponding to an illegal action occurrence area, wherein the illegal action warning information comprises the illegal action type, the risk level and early warning prompt contents.
7. A petrochemical field operation flow violation intelligent recognition system for implementing the method of any of the preceding claims 1-6, characterized by comprising: The first unit is used for acquiring video data of a petrochemical field operation area, carrying out personnel target detection and personnel action recognition, obtaining personnel target position information and personnel action information, determining an operation area range, extracting equipment entity images in the operation area range and acquiring temperature data; The second unit is used for carrying out self-adaptive threshold segmentation on the equipment entity image, constructing an equipment characteristic point cloud image, matching the characteristic point cloud image by utilizing a local sensitive hash algorithm, identifying an equipment abnormal state, constructing temperature data into a temperature thermal distribution map in an operation area range, identifying a temperature abnormal area, calculating a temperature abnormal level, constructing a time sequence combination characteristic according to the equipment abnormal state and the temperature abnormal level, and calculating a risk degree score of a current operation scene; The third unit is used for extracting action track data in the personnel action information, constructing action feature vectors by utilizing track shape descriptors, performing principal component analysis and dimension reduction processing to obtain key action features, matching the key action features with a standard operation action sequence by adopting a sequence alignment algorithm, calculating similarity scores, and grading illegal behaviors by combining the risk degree scores to obtain operation risk grades; and the fourth unit is used for generating illegal action warning information when the operation risk level exceeds a preset threshold value and sending the illegal action warning information to the on-site management terminal.
8. An electronic device, comprising: A processor; A memory for storing processor-executable instructions; Wherein the processor is configured to invoke the instructions stored in the memory to perform the method of any of claims 1 to 6.
9. A computer readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the method of any of claims 1 to 6.

Description

Intelligent recognition method and system for illegal behaviors of petrochemical field operation flow Technical Field The invention relates to the technical field of petrochemical safety production, in particular to an intelligent recognition method and system for operation flow violations in a petrochemical site. Background The petrochemical industry is used as a high-risk industry, safety risk management in the production process is always the focus of industry attention, the traditional safety management of petrochemical field operation mainly relies on manual inspection and fixed monitoring equipment, and illegal behaviors are found by inspection or monitoring video inspection of a security officer on site; With the development of information technology, computer vision and artificial intelligence technology are applied to the field of safety supervision, a new technical means is provided for petrochemical field operation safety management, and the novel technical means mainly comprise rule-based abnormal behavior detection, video-monitoring-based personnel behavior analysis and sensor-data-based equipment state monitoring, but the prior art still has the defects of lacking overall analysis of a man-machine interaction process, difficulty in accurately identifying illegal behaviors under a specific equipment operation environment, high false alarm rate or frequent missing report conditions, insufficient risk assessment of an operation environment, incapability of dynamically adjusting the risk level of the illegal behaviors according to environmental factors such as real-time equipment states, temperature changes and the like, incapability of preferentially processing the high-risk illegal behaviors and lacking time sequence analysis capability of an operation action sequence, incapability of comparing and matching detected single actions with a standard operation flow, and difficulty in identifying illegal operations which are similar to normal but have wrong orders or missing key steps; Accordingly, there is a need for a solution to the problems of the prior art. Disclosure of Invention The embodiment of the invention provides an intelligent recognition method and system for the illegal behaviors of a petrochemical field operation flow, which at least can solve part of problems in the prior art. In a first aspect of the embodiment of the present invention, there is provided a method for intelligently identifying violations of a petrochemical field operation process, including: acquiring video data of a petrochemical field operation area, carrying out personnel target detection and personnel action recognition to obtain personnel target position information and personnel action information, determining an operation area range, extracting equipment entity images in the operation area range and acquiring temperature data; Performing self-adaptive threshold segmentation on the equipment entity image, constructing an equipment characteristic point cloud image, matching the characteristic point cloud image by utilizing a local sensitive hash algorithm, identifying an equipment abnormal state, constructing temperature data into a temperature thermodynamic distribution diagram in an operation area range, identifying a temperature abnormal area, calculating a temperature abnormal grade, constructing a time sequence combination characteristic according to the equipment abnormal state and the temperature abnormal grade, and calculating a risk degree score of a current operation scene; Extracting action track data in personnel action information, constructing action feature vectors by utilizing track shape descriptors, performing principal component analysis and dimension reduction processing to obtain key action features, matching the key action features with a standard operation action sequence by adopting a sequence alignment algorithm, calculating similarity scores, and grading illegal behaviors by combining the risk degree scores to obtain operation risk grades; and when the working risk level exceeds a preset threshold, generating illegal action warning information and sending the illegal action warning information to the on-site management terminal. In an alternative embodiment of the present invention, Acquiring video data of a petrochemical field operation area, carrying out personnel target detection and personnel action recognition, obtaining personnel target position information and personnel action information, determining an operation area range, and extracting equipment entity images and acquiring temperature data in the operation area range comprises the following steps: Collecting video data of a petrochemical field operation area, and carrying out frame segmentation processing on the video data to obtain a continuous video image sequence; Performing personnel target detection and personnel action recognition on the video image sequence, acquiring personnel target position information and personn