An intelligent monitoring system for substations based on IoT perception

Abstract

The present invention relates to the field of intelligent monitoring, and specifically to an intelligent monitoring system for a substation based on Internet of Things perception, comprising a sensor, an interface, an equipment operation detection module, an intelligent analysis module, an alarm module and a remote monitoring and control module; the present invention adopts a method based on Internet of Things perception, constructs a point cloud video of a substation with a laser, and realizes recognition and detection of intrusion behavior by performing motion capture on the point cloud video; the present invention uses a dislocated spatiotemporal feature extraction algorithm based on adaptive attention, aggregates local area features and global static features of a point cloud, calculates the integrity of the dislocated spatiotemporal features to coordinate micro-transformations and macro-structural transformations in the point cloud, and adopts trainable basis functions and spline functions to effectively capture the complex multivariate relationship of intrusion behavior in the point cloud, reduce calculation parameters, and improve the recognition efficiency and accuracy of intrusion behavior.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent monitoring, and in particular to an intelligent monitoring system for a substation based on Internet of Things perception. Background Art

[0002] Substations are an important part of the power system, and the stable and safe operation of equipment is of vital importance. Through the monitoring system, the activities of personnel and equipment can be monitored, abnormal intrusion behaviors can be discovered in time, and the possibility of accidents can be reduced; the monitoring system can detect violations by employees or robots, such as unauthorized entry into dangerous areas, unauthorized operation of equipment, etc., to help managers stop improper behavior in time; however, the existing technology can only perform simple monitoring when monitoring substations, and cannot identify and detect intrusion behaviors based on the images obtained by monitoring. When the existing computer vision technology is used for substation monitoring, it is slow to recognize dynamic geometric information in the monitoring, and it is difficult to quickly issue danger alarms. Summary of the invention

[0003] In view of the above problems and to overcome the defects of the prior art, the present invention creatively provides an intelligent monitoring system for substations based on Internet of Things perception. In view of the problem that the prior art can only perform simple monitoring when monitoring substations and cannot identify and detect intrusion behaviors based on the images obtained by monitoring, the present invention adopts a method based on Internet of Things perception to construct a point cloud video of the substation with a laser, and realizes the identification and detection of intrusion behaviors by performing motion capture on the point cloud video; in view of the problem that the existing computer vision technology is slow in recognizing dynamic geometric information in monitoring and difficult to quickly issue danger alarms when applied to substation monitoring, the present invention uses a dislocated spatiotemporal feature extraction algorithm based on adaptive attention to aggregate the local area features and global static features of the point cloud to calculate the dislocated spatiotemporal features. The dislocated spatiotemporal features coordinate the integrity of micro-transformations and macro-structural transformations in the point cloud, and at the same time use trainable basis functions and spline functions to effectively capture the complex multivariate relationships of intrusion behaviors in the point cloud, reduce calculation parameters, and improve the recognition efficiency and accuracy of intrusion behaviors.

[0004] The present invention provides a substation intelligent monitoring system based on IoT perception, including sensors, interfaces, equipment operation detection modules, intelligent analysis modules, alarm modules and remote monitoring and control modules;

[0005] The sensors include a temperature sensor, a humidity sensor, a current sensor and a voltage sensor, which respectively collect data on the temperature, humidity, current and voltage of the substation to obtain operation data, including real-time operation data and historical operation data, collect and construct point cloud video of the substation through laser to obtain point cloud video, and send the operation data and point cloud video to the equipment operation detection module and the intelligent analysis module through the interface respectively;

[0006] The interface connects the equipment operation detection module and the intelligent analysis module to the sensor;

[0007] The equipment operation detection module detects the operation data of the substation, and when there is an abnormality in the real-time operation data, an operation alarm message is issued and transmitted to the alarm module;

[0008] The intelligent analysis module uses an adaptive attention-based staggered spatiotemporal feature extraction algorithm to detect intrusion behavior on the point cloud video. If there is intrusion behavior, an intrusion alarm message is issued and transmitted to the alarm module;

[0009] The alarm module receives operation alarm information and intrusion alarm information and issues an alarm, while saving all alarm information;

[0010] The remote monitoring and control module displays the substation's operating data, point cloud video and all alarm information in real time, allowing users to remotely control substation equipment.

[0011] Furthermore, in the equipment operation detection module, the operation data of the substation is detected, which specifically includes the following steps:

[0012] Step Q1: Data preprocessing: data cleaning, missing value processing and outlier processing are performed on the historical operation data to obtain preprocessed historical operation data;

[0013] Step Q2: Data conversion, performing Fourier transform on the pre-processed historical operation data to obtain spectrum data;

[0014] Step Q3: Feature extraction, use the CNN model to extract features from the spectrum data to obtain historical operation features;

[0015] Step Q4: Operation detection, set the feature threshold according to the historical operation characteristics, repeat steps Q1-Q3, extract the real-time operation characteristics of the real-time operation data, when the real-time operation characteristics are not within the feature threshold, the real-time operation data is abnormal, and an operation alarm message is issued.

[0016] Furthermore, in the intelligent analysis module, an adaptive attention-based dislocation spatiotemporal feature extraction algorithm is used to detect intrusion behavior, which specifically includes the following steps:

[0017] Step S1: Video segmentation: dividing the point cloud video into N point cloud video segments of equal length according to the step length d, and dividing the safe area in the point cloud;

[0018] Step S2: Anchor point sampling: select the central frame of each point cloud video clip as the anchor frame, use the farthest point sampling method to sample the anchor points of each anchor frame, and propagate the sampling results to the adjacent frames of the point cloud video clip;

[0019] Step S3: constructing a spatiotemporal local region, using the KNN algorithm to select K nearest neighboring points of the anchor point, and defining the region surrounded by the K nearest neighboring points as the spatiotemporal local region;

[0020] Step S4: local area feature acquisition, using 4D convolution to extract features of all point clouds in the local spatiotemporal area to obtain point features in the local area, where the 4D convolution includes three-dimensional position convolution plus one-dimensional time convolution;

[0021] Step S5: Extract global static features. Use the adaptive attention-enhanced hierarchical spatial algorithm to extract global static features from the anchor box:

[0022] r j t =[ max i=1,...,km { MLP([( l j,i t - o j t ); e j,i t ; f j,i t ]⨀A) } ; o j t ] ;

[0023] In the formula, represents the coordinates of the i-th point in the j-th spatiotemporal local region in the point cloud, represents the coordinates of the centroid of the jth local space-time region, Represents the point feature in the local area of ​​the i-th point, represent and The Euclidean distance between Represents the attention score calculated by the multi-head attention model, [( l j,i t - o j t ); e j,i t ; f j,i t ] represents multi-head attention processing, where multi-head attention includes query Q, key K and value V, and the inputs are , and , represents the dot product operation, represents multi-layer perceptron processing, represents the number of inliers in a local region of spacetime, Represents the global static characteristics of the jth spatiotemporal local region;

[0024] Step S6: Extraction of dislocation spatiotemporal dynamic features, specifically including the following steps:

[0025] Step S61: Calculation of spatial dislocation time and space position:

[0026] x t+k = h t+k ⨁[( t o t l -1 )-0.5] ;

[0027] In the formula, Represents the point cloud video clip Time point, Represents the length of the point cloud video clip, represents the three-dimensional position of a local region in space and time, represents the addition operation of the vectors corresponding to the features, The spatially displaced space-time position representing a local region of space-time;

[0028] Step S62: motion activation function design:

[0029] ;

[0030] In the formula, represents the basis function, represents the spline function, and represents the preset trainable weights, represents the motion activation function;

[0031] Step S63: Calculation of dislocation spatiotemporal dynamic characteristics:

[0032] ;

[0033] Where N represents the total number of point cloud video clips, represents the motion activation function, represents the feature extraction operation, Represents the dynamic characteristics of dislocation space-time, The feature that represents the superposition of the global static features of all local regions in space and time, and The counting group is responsible for the cumulative sum counting in the formula;

[0034] Step S7: Intrusion detection, activate and predict the dislocation spatiotemporal dynamic features to obtain a predicted point cloud image. When the predicted point cloud image exceeds the preset point cloud safety area, an intrusion alarm message is issued.

[0035] The beneficial results achieved by the present invention using the above scheme are as follows:

[0036] (1) In view of the problem that the existing technology can only perform simple monitoring when monitoring the substation and cannot identify and detect intrusion behavior based on the monitored images, the present invention adopts a method based on IoT perception to construct a point cloud video of the substation with laser, and realizes the identification and detection of intrusion behavior by capturing motion on the point cloud video;

[0037] (2) To address the problem that the existing computer vision technology is slow in recognizing dynamic geometric information in monitoring and is difficult to quickly issue danger alarms when used in substation monitoring, the present invention adopts a dislocated spatiotemporal feature extraction algorithm based on adaptive attention. By aggregating the local area features and global static features of the point cloud, the dislocated spatiotemporal features are calculated. The dislocated spatiotemporal features coordinate the integrity of the micro-transformation and macro-structural transformation in the point cloud. At the same time, trainable basis functions and spline functions are used to effectively capture the complex multivariate relationships of intrusion behaviors in the point cloud, reduce calculation parameters, and improve the recognition efficiency and accuracy of intrusion behaviors. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A module diagram of a substation intelligent monitoring system based on IoT perception provided by the present invention;

[0039] Figure 2 Flowchart of the adaptive attention-based misaligned spatiotemporal feature extraction algorithm.

[0040] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0042] Example 1, see Figure 1 , an intelligent monitoring system for substations based on IoT perception, including sensors, interfaces, equipment operation detection modules, intelligent analysis modules, alarm modules and remote monitoring and control modules;

[0043] The sensors include temperature sensors, humidity sensors, current sensors and voltage sensors, which respectively collect data on the temperature and humidity of the substation and the current and voltage of the transformer, capacitor, reactor, lightning arrester and battery pack to obtain operation data, including real-time operation data and historical operation data. The point cloud video of the substation is collected and constructed by laser to obtain the point cloud video, and the operation data and the point cloud video are respectively sent to the equipment operation detection module and the intelligent analysis module through the interface;

[0044] The interface connects the equipment operation detection module and the intelligent analysis module to the sensor;

[0045] The equipment operation detection module detects the operation data of the substation, and when there is an abnormality in the real-time operation data, an operation alarm message is issued and transmitted to the alarm module;

[0046] The intelligent analysis module uses an adaptive attention-based staggered spatiotemporal feature extraction algorithm to detect intrusion behavior on the point cloud video. If there is intrusion behavior, an intrusion alarm message is issued and transmitted to the alarm module;

[0047] The alarm module receives operation alarm information and intrusion alarm information and issues an alarm, while saving all alarm information;

[0048] The remote monitoring and control module displays the substation's operating data, point cloud video and all alarm information in real time, allowing users to remotely control substation equipment.

[0049] By performing the above operations, in order to solve the problem that the existing technology can only perform simple monitoring of the substation and cannot identify and detect intrusion behavior based on the monitored images, the present invention adopts a method based on Internet of Things perception, uses laser to construct a point cloud video of the substation, and realizes the identification and detection of intrusion behavior by capturing motion on the point cloud video.

[0050] Example 2, see Figure 2 Based on the above embodiment, this embodiment performs intrusion behavior detection in the intelligent analysis module based on the dislocation spatiotemporal feature extraction algorithm of adaptive attention, which specifically includes the following steps:

[0051] Step S1: Video segmentation: divide the point cloud video into 10 point cloud video segments of equal length according to a step length of 10 seconds, and divide the safe area in the point cloud;

[0052] Step S2: Anchor point sampling: select the central frame of each point cloud video clip as the anchor frame, use the farthest point sampling method to sample the anchor points of each anchor frame, and propagate the sampling results to the adjacent frames of the point cloud video clip;

[0053] Step S3: constructing the spatiotemporal local area, using the KNN algorithm to select the 8 nearest neighboring points of the anchor point, and defining the area surrounded by the 8 nearest neighboring points as the spatiotemporal local area;

[0054] Step S4: local area feature acquisition, using 4D convolution to extract features of all point clouds in the local spatiotemporal area to obtain point features in the local area, where the 4D convolution includes three-dimensional position convolution plus one-dimensional time convolution;

[0055] Step S5: Extract global static features. Use the adaptive attention-enhanced hierarchical spatial algorithm to extract global static features from the anchor box:

[0056] r j t =[ max i=1,...,km { MLP([( l j,i t - o j t ); e j,i t ; f j,i t ]⨀A) } ; o j t ] ;

[0057] In the formula, represents the coordinates of the i-th point in the j-th spatiotemporal local region in the point cloud, represents the coordinates of the centroid of the jth local space-time region, Represents the point feature in the local area of ​​the i-th point, represent and The Euclidean distance between Represents the attention score calculated by the multi-head attention model, [( l j,i t - o j t ); e j,i t ; f j,i t ] represents multi-head attention processing, where multi-head attention includes query Q, key K and value V, and the inputs are , and , represents the dot product operation, represents multi-layer perceptron processing, represents the number of inliers in a local region of spacetime, Represents the global static characteristics of the jth spatiotemporal local region;

[0058] Step S6: Extraction of dislocation spatiotemporal dynamic features;

[0059] Step S7: Intrusion detection, activate and predict the dislocation spatiotemporal dynamic features to obtain a predicted point cloud image. When the predicted point cloud image exceeds the preset point cloud safety area, an intrusion alarm message is issued.

[0060] By performing the above operations, in order to solve the problem that the existing computer vision technology is slow in recognizing the dynamic geometric information in monitoring and difficult to issue danger alarms quickly when applied to substation monitoring, the present invention adopts a dislocated spatiotemporal feature extraction algorithm based on adaptive attention, aggregates the local area features and global static features of the point cloud, and calculates the dislocated spatiotemporal features. The dislocated spatiotemporal features coordinate the integrity of the micro-transformation and the macro-structural transformation in the point cloud, and simultaneously adopts trainable basis functions and spline functions to effectively capture the complex multivariate relationship of the intrusion behavior in the point cloud, reduce the calculation parameters, and improve the recognition efficiency and accuracy of the intrusion behavior.

[0061] Example 3, see Figure 2 This embodiment is based on the above embodiment, and step S6 specifically includes the following steps:

[0062] Step S61: Calculation of spatial dislocation time and space position:

[0063] x t+k = h t+k ⨁[( t o t l -1 )-0.5] ;

[0064] In the formula, Represents the point cloud video clip Time point, Represents the length of the point cloud video clip, represents the three-dimensional position of a local region in space and time, represents the addition operation of the vectors corresponding to the features, The spatially displaced space-time position representing a local region of space-time;

[0065] Step S62: motion activation function design:

[0066] ;

[0067] In the formula, represents the basis function, represents the spline function, and represents the preset trainable weights, represents the motion activation function;

[0068] Step S63: Calculation of dislocation spatiotemporal dynamic characteristics:

[0069] ;

[0070] Where N represents the total number of point cloud video clips, represents the motion activation function, represents the feature extraction operation, Represents the dynamic characteristics of dislocation space-time, The feature that represents the superposition of the global static features of all local regions in space and time, and It is the counting group, responsible for the cumulative sum of the counts in the formula.

[0071] Embodiment 4: Based on the above embodiment, this embodiment detects the operation data of the substation in the equipment operation detection module, and specifically includes the following steps:

[0072] Step Q1: Data preprocessing: data cleaning, missing value processing and outlier processing are performed on the historical operation data to obtain preprocessed historical operation data;

[0073] Step Q2: Data conversion, performing Fourier transform on the pre-processed historical operation data to obtain spectrum data;

[0074] Step Q3: Feature extraction, use the CNN model to extract features from the spectrum data to obtain historical operation features;

[0075] Step Q4: Operation detection, set the feature threshold according to the historical operation characteristics, repeat steps Q1-Q3, extract the real-time operation characteristics of the real-time operation data, when the real-time operation characteristics are not within the feature threshold, the real-time operation data is abnormal, and an operation alarm message is issued.

[0076] Embodiment 5: This embodiment is based on the above embodiment, and step Q2 specifically includes the following steps:

[0077] Prepare time series data: Prepare preprocessed historical running data to ensure that the data sampling frequency meets the Nyquist sampling theorem to avoid aliasing;

[0078] Perform Fourier transform: Apply continuous Fourier transform to the preprocessed historical operation data to convert it from the time domain to the frequency domain to obtain the result of continuous Fourier transform;

[0079] Calculate spectrum data: Extract the amplitude spectrum (amplitude information of frequency components) and phase spectrum (phase information of frequency components) from the results of continuous Fourier transform to construct spectrum data, where the amplitude spectrum represents the intensity of the frequency components and the phase spectrum represents the phase of the frequency components.

[0080] Embodiment 6: This embodiment is based on the above embodiment. In step S2, the farthest point sampling method specifically includes the following steps:

[0081] Preparing original point cloud data: preparing original point cloud data, that is, the point cloud of the central frame;

[0082] Select initial point: randomly select an initial point from the original point cloud data as the first sampling point;

[0083] Calculate the distance: For each point, calculate the Euclidean distance between it and all points in the selected sampling point set;

[0084] Select the farthest point: select the point farthest from the initial point set as the next sampling point, that is, select the point farthest from the current point to the selected sampling point set;

[0085] Repeat steps: Repeat steps 3 and 4 until a preset number of sampling points are selected;

[0086] Sampling point set formation: After the above steps, a representative sampling point set is finally formed, and the centroid of the sampling point set is the anchor point.

[0087] Embodiment 7: This embodiment is based on the above embodiment, and point cloud construction specifically includes the following steps:

[0088] Data acquisition: Use laser scanner to obtain point information in three-dimensional space;

[0089] Data preprocessing: preprocess the raw data, including denoising, filtering, registration and other operations to improve data quality and accuracy;

[0090] Point cloud generation: convert the processed data into a point cloud data structure, which is usually represented as a set of three-dimensional points, each of which contains three-dimensional position information;

[0091] Point cloud registration: Align point clouds from different data sources into the same coordinate system to obtain a complete and consistent point cloud.

[0092] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0093] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0094] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent monitoring system for substations based on IoT perception, characterized in that: It includes sensors, interfaces, equipment operation detection modules, intelligent analysis modules, alarm modules and remote monitoring and control modules; The sensors include a temperature sensor, a humidity sensor, a current sensor and a voltage sensor, which respectively collect data on the temperature, humidity, current and voltage of the substation to obtain operation data, including real-time operation data and historical operation data, collect and construct point cloud video of the substation through laser to obtain point cloud video, and send the operation data and point cloud video to the equipment operation detection module and the intelligent analysis module through the interface respectively; The interface connects the equipment operation detection module and the intelligent analysis module to the sensor; The equipment operation detection module uses a convolutional network method to detect the operation data of the substation. When there is an abnormality in the real-time operation data, an operation alarm message is issued and transmitted to the alarm module; The intelligent analysis module uses an adaptive attention-based staggered spatiotemporal feature extraction algorithm to detect intrusion behavior on the point cloud video. If there is intrusion behavior, an intrusion alarm message is issued and transmitted to the alarm module; The intrusion behavior detection is performed using the dislocated spatiotemporal feature extraction algorithm based on adaptive attention, which specifically includes the following steps: Step S1: Video segmentation: dividing the point cloud video into N point cloud video segments of equal length according to the step length d, and dividing the safe area in the point cloud; Step S2: Anchor point sampling: select the central frame of each point cloud video clip as the anchor frame, use the farthest point sampling method to sample the anchor points of each anchor frame, and propagate the sampling results to the adjacent frames of the point cloud video clip; Step S3: constructing a spatiotemporal local region, using the KNN algorithm to select K nearest neighboring points of the anchor point, and defining the region surrounded by the K nearest neighboring points as the spatiotemporal local region; Step S4: local area feature acquisition, using 4D convolution to extract features of all point clouds in the local spatiotemporal area to obtain point features in the local area, where the 4D convolution includes three-dimensional position convolution plus one-dimensional time convolution; Step S5: Extract global static features. Use the adaptive attention-enhanced hierarchical spatial algorithm to extract global static features from the anchor box: ； In the formula, represents the coordinates of the i-th point in the j-th spatiotemporal local region in the point cloud, represents the coordinates of the centroid of the jth local space-time region, Represents the point feature in the local area of ​​the i-th point, represent and The Euclidean distance between Represents the attention score calculated by the multi-head attention model, represents multi-head attention processing, where multi-head attention includes query Q, key K and value V, and the inputs are , and , represents the dot product operation, represents multi-layer perceptron processing, represents the number of inliers in a local region of spacetime, Represents the global static characteristics of the jth spatiotemporal local region; Step S6: Extraction of dislocation spatiotemporal dynamic features, specifically including the following steps: Step S61: Calculation of spatial dislocation time and space position: ； In the formula, Represents the point cloud video clip Time point, Represents the length of the point cloud video clip, represents the three-dimensional position of a local region in space and time, represents the addition operation of the vectors corresponding to the features, The spatially displaced space-time position representing a local region of space-time; Step S62: motion activation function design: ； In the formula, represents the basis function, represents the spline function, and represents the preset trainable weights, represents the motion activation function; Step S63: Calculation of dislocation spatiotemporal dynamic characteristics: ； Where N represents the total number of point cloud video clips, represents the motion activation function, represents the feature extraction operation, Represents the dynamic characteristics of dislocation space-time, The feature that represents the superposition of the global static features of all local regions in space and time, and The counting group is responsible for the cumulative sum counting in the formula; Step S7: Intrusion detection, activating and predicting the dislocation spatiotemporal dynamic features to obtain a predicted point cloud image. When the predicted point cloud image exceeds the preset point cloud safety area, an intrusion alarm message is issued; The alarm module receives and stores the operation alarm information and the intrusion alarm information, and issues an alarm at the same time; The remote monitoring and control module displays the substation's operating data, point cloud video and all alarm information in real time, allowing users to remotely control substation equipment.

2. According to claim 1, a substation intelligent monitoring system based on IoT perception is characterized in that: The equipment operation detection module uses a convolutional network method to detect the operation data of the substation, which specifically includes the following steps: Step Q1: Data preprocessing: data cleaning, missing value processing and outlier processing are performed on the historical operation data to obtain preprocessed historical operation data; Step Q2: Data conversion, performing Fourier transform on the pre-processed historical operation data to obtain spectrum data; Step Q3: Feature extraction, use the CNN model to extract features from the spectrum data to obtain historical operation features; Step Q4: Operation detection, set the feature threshold according to the historical operation characteristics, repeat steps Q1-Q3, extract the real-time operation characteristics of the real-time operation data, when the real-time operation characteristics are not within the feature threshold, the real-time operation data is abnormal, and an operation alarm message is issued.

Images