Substation personnel security management method

Abstract

The invention discloses a substation personnel security management method, which comprises the following steps of: acquiring video streams in a gate area by using an existing camera of a substation, and performing motion detection through a Gaussian mixture model (GMM) to trigger a Yolo + Sort multi-target detection and trajectory tracking algorithm so as to realize accurate identification of personnel access behaviors. By delimiting a virtual boundary and combining with Kalman filtering trajectory analysis, station entering or leaving characteristics of personnel are judged, real-time number data are synchronously compared with an ORM working system, and double auditing of operation information and personnel identities is achieved through an applet information registration and background cross verification mechanism. The pure vision-driven algorithm architecture can reduce hardware cost, through full-process management of multi-target detection, trajectory tracking and data collaboration, automation and safety of transformer substation personnel supervision are remarkably improved, and the method is particularly suitable for unattended transformer substation scenes and has high engineering practicability.

Description

Technical Field

[0001] This invention relates to the field of power system security protection technology, and in particular to a method for personnel security management in substations. Background Technology

[0002] In recent years, substations, as critical nodes in the power system, have faced extremely high security requirements for personnel access management. Traditional substation personnel security management primarily relies on manual registration, which has the following shortcomings:

[0003] Insufficient real-time capability: Manual verification is inefficient, especially after the work permit has been issued and there is no way to provide real-time warnings for unauthorized entry or unregistered personnel.

[0004] Inadequate security supervision: There are risks associated with unattended substations, such as lack of supervision by security guards, absenteeism, and unauthorized entry by unauthorized personnel.

[0005] Poor data collaboration: The personnel information registration and operation system (ORM operation system) lack real-time linkage, which is prone to information lag or omissions;

[0006] Low level of automation: It is impossible to automatically collect and analyze the flow of people and their entry and exit status, resulting in low management efficiency.

[0007] With the development of intelligent monitoring technology, some substations have deployed video surveillance systems. However, existing technologies mostly adopt single-target detection or static access control management, and have not formed a fully automated management system of "detection-tracking-verification-early warning", which is difficult to meet the intelligent needs of new substations for personnel security. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide a method for personnel security management in substations, which improves the safe operation of substations and has strong practicality.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: a method for personnel security management in substations, comprising the following steps:

[0010] Step 1: Use multi-target detection to detect personnel movement at the substation gate, and add an infrared sensing device. Use video stream motion monitoring to start the multi-target detection algorithm.

[0011] Step 2: Perform object detection using YOLOv5, whose objective function includes location loss, confidence loss, and class loss;

[0012] Step 3: Use the Sort algorithm to predict the target state through Kalman filtering and perform path tracking of the detected target;

[0013] Step 4: Determine whether the target has entry or exit characteristics;

[0014] Step 5: Establish a dynamic model of personnel numbers and set tolerance coefficients to provide early warnings of anomalies;

[0015] Step 6: Security personnel at each station register the work permit and the information of the personnel entering the station through the mini-program, and cross-compare the information with the ORM operation platform in the background. After the information is verified, the system will display "entry permitted" and the security personnel will allow entry.

[0016] In a preferred embodiment, step 1 employs video stream motion detection and algorithm triggering; a Gaussian mixture model (GMM) is used to model the background of the video stream in the substation gate area, and the pixel-level motion probability is calculated using the frame difference method.

[0017]

[0018] Where I(x,y,t) is the gray value of pixel (x,y) at time t, B(x,y,t) is the gray value of the background model, α is the threshold coefficient, and σ(x,y) is the variance of the background pixels; when the proportion of moving pixels... When β is the trigger threshold, the YOLO multi-target detection algorithm is automatically started.

[0019] In a preferred embodiment, in step 2, a person detection network is constructed based on the YOLO v5 model, and the objective function includes location loss, confidence loss, and category loss:

[0020] L=λ loc ·L loc +λ conf ·L conf +λ cls ·L cls

[0021] Where L is the total loss function, which is the weighted sum of location loss, confidence loss, and class loss; L loc It is the location loss, which measures the difference in location between the predicted bounding box and the ground truth bounding box; L conf It measures the difference between the confidence level that the predicted bounding box contains the target and the actual situation; Class loss L cls Measure the difference between the predicted category and the true category;

[0022] Position loss is calculated using CIoU:

[0023]

[0024] Among them, b, b gt ρ and c are the coordinates of the predicted bounding box and the ground truth bounding box, respectively; ρ is the Euclidean distance; and c is the length of the diagonal of the smallest rectangle enclosing the two boxes.

[0025]

[0026] Among them, α represents the weight coefficient of the aspect ratio loss; IoU is the ratio of the intersection area of the predicted bounding box and the ground truth bounding box to the union area, and its value range is [0, 1]. The larger the value, the better the match. w gt , hgt represents the width and height of the ground truth bounding box; w, h represent the width and height of the predicted bounding box.

[0027] Confidence loss L loc Adopts binary cross-entropy

[0028]

[0029] Among them, y is the true confidence, is the predicted confidence;

[0030] Class loss L cls Adopts cross-entropy: The detection result outputs the coordinates (x, y, w, h) of the personnel bounding box, the confidence S, and the class label.

[0031] In a preferred embodiment, in step 3, trajectory tracking and in-out behavior judgment:

[0032] Adopts the Sort algorithm to predict the target state through Kalman filtering:

[0033] State prediction:

[0034] Among them, the state vector includes position, scale, aspect ratio, and velocity. F is the state transition matrix, P is the state covariance matrix, which quantifies the uncertainty of the state estimate and is dynamically adjusted during the prediction and update processes to help the algorithm balance the weights of the model prediction and the observed data; Q is the process noise covariance;

[0035] Observation update:

[0036]

[0037] Among them, H is the observation matrix, z t is the observation value of the detection box, and R is the observation noise covariance.

[0038] In a preferred embodiment, in step 3, the inner and outer boundary lines of the gate L1: x = x1 and L2: x = x2 are defined, where x2 < x1, and the in-out behavior is judged by the trajectory point coordinates and the velocity direction:

[0039] Inbound feature: There exists a trajectory point p i (x i , yi ) Satisfy x i > x1 and x at the previous moment i-1 < x2, and the velocity component v x > 0;

[0040] Outbound feature: There exists a trajectory point p j (x j , y j ) Satisfy x j < x2 and x at the previous moment j-1 > x1, and the velocity component v x < 0.

[0041] In a preferred embodiment, in step 5, a dynamic model N(t) of the number of in-station personnel is established, and N(t) = N(t - 1)+ΔN in (t)-ΔN out (t) is updated based on the trajectory in-and-out events, where ΔN in (t) is the number of inbound personnel at time t, and ΔN out (t) is the number of outbound personnel; N(t) is transmitted to the ORM system through the MQTT protocol, and the comparison formula is: E(t)=|N(t)-N plan (t)|≤γ·N plan (t), where N plan (t) is the planned number of personnel, and γ is the tolerance coefficient. When E(t)>γ·N plan (t), an abnormal warning is triggered, analogous to the alarm sensitivity.

[0042] In a preferred embodiment, in step 6, the security personnel register the information D = {T, ID, Type, Area} through a mini-program. T, ID, Type, and Area respectively represent time, personnel ID, operation type, and area, and are compared with the ORM system data D orm ={T orm , ID orm , Type orm , Area orm} for cosine similarity calculation: When Sim≥δ is the matching threshold, the audit is passed, and permission to enter the station is pushed to the security and operation and maintenance personnel. Otherwise, the application is sent to the operation and maintenance personnel, and the operation and maintenance personnel decide whether to allow entry into the station.

[0043] Compared with existing technologies, this invention has the following advantages: This invention proposes a pure vision-triggered mechanism based on video stream motion detection, constructing a mathematical model of "motion detection - target recognition - trajectory tracking - data collaboration" to achieve high-precision monitoring of personnel entering and exiting substations, solving the problems of strong hardware dependence and poor algorithm robustness in traditional solutions. The pure vision-driven algorithm architecture of this invention can reduce hardware costs. Through full-process management of multi-target detection, trajectory tracking, and data collaboration, it significantly improves the automation and safety of substation personnel monitoring, making it particularly suitable for unmanned substation scenarios and possessing strong engineering practicality. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of a preferred embodiment of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0046] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0047] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application; as used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise; furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0048] A method to improve personnel security management in substations, referenced Figure 1 Multi-target detection is used to detect personnel at the substation gate, and infrared sensors are added to activate the multi-target detection algorithm. Video stream motion monitoring is used to activate the detection algorithm. YOLOv5 is used for target detection, and its objective function includes position loss, confidence loss, and category loss. The Sort algorithm is used to predict the target state through Kalman filtering and to track the path of the detected target. Two boundary lines are drawn to determine whether the target has entry or exit characteristics. A dynamic model of personnel quantity is established, and a tolerance coefficient is set to provide anomaly warnings. Security personnel at each station register work permits and personnel information through a mini-program, and cross-check the information with the ORM operation platform in the backend. After the work information is compared, entry is allowed, and security personnel grant passage.

[0049] The specific implementation process is as follows:

[0050] 1. First, video stream motion detection and algorithm triggering are used.

[0051] A Gaussian mixture model (GMM) is used to model the background of the video stream in the substation gate area, and the pixel-level motion probability is calculated using the frame difference method.

[0052]

[0053] Where I(x,y,t) is the gray value of pixel (x,y) at time t, B(x,y,t) is the gray value of the background model, α is the threshold coefficient, and σ(x,y) is the variance of the background pixels. When the proportion of moving pixels... When β is the trigger threshold, the YOLO multi-target detection algorithm is automatically started.

[0054] 2. Multi-target detection and feature extraction

[0055] A person detection network is built based on the YOLO v5 model. The objective function includes location loss, confidence loss, and class loss.

[0056] L=λ loc L loc +λ conf L conf +λ cls L cls

[0057] Position loss is calculated using CIoU (Complete Intersection over Union):

[0058]

[0059] Among them, b,b gt Let ρ be the coordinates of the predicted bounding box and the ground truth bounding box, ρ be the Euclidean distance, and c be the length of the diagonal of the smallest rectangle enclosing the two boxes.

[0060]

[0061] The confidence loss uses binary cross-entropy.

[0062]

[0063] Where y represents the true confidence level. To predict confidence levels;

[0064] The category loss uses cross-entropy: The detection results output the bounding box coordinates (x, y, w, h), confidence level S, and category label of the personnel.

[0065] 3. Trajectory tracking and entry / exit behavior determination

[0066] Adopt the Sort (Simple Online and Realtime Tracking) algorithm to predict the target state through Kalman filtering:

[0067] State prediction:

[0068] Among them, the state vector includes position, scale, aspect ratio and speed. F is the state transition matrix, and Q is the process noise covariance;

[0069] Observation update:

[0070]

[0071] Among them, H is the observation matrix, z t is the detection box observation value, and R is the observation noise covariance. Define the inner and outer boundary lines of the gate as L1: x = x1 and L2: x = x2 (x2 < x1), and judge the entry and exit behaviors by the trajectory point coordinates and the speed direction:

[0072] · Inbound characteristics:

[0073] There exists a trajectory point p i (x i , y i ) that satisfies x i > x1 and x at the previous moment i-1 < x2, and the velocity component v x > 0;

[0074] · Outbound characteristics:

[0075] There exists a trajectory point p j (x j ​​​​​​​​​​​​​​​​​​​​​​​​​​(t), where N plan (t) represents the planned number of people, and γ is the tolerance coefficient. When E(t) > γ·N plan An abnormal warning is triggered when (t), analogous to alarm sensitivity.

[0078] 5. Information registration and cross-verification

[0079] Security personnel register information D = {T, ID, Type, Area} (time, personnel ID, job type, area) via a mini-program, which is then compared with the data D in the ORM system. orm ={T orm ID orm Type orm Area orm Perform cosine similarity calculation:

[0080] When Sim≥δ is the matching threshold, the review is approved and a request for entry is sent to security and operations personnel. Otherwise, the request is sent to operations personnel, who will decide whether to allow entry.

Claims

1. A method for personnel security management in substations, characterized in that, It includes the following steps: Step 1: Use multi-object detection to detect the movement of personnel at the substation gate, and add an infrared induction device. The multi-object detection algorithm is started by video stream motion monitoring; Step 2: Use Yolo5 for object detection. Its objective function includes location loss, confidence loss, and class loss; Step 3: Adopt the Sort algorithm to predict the target state through Kalman filtering and track the path of the detected target; Step 4: Judge whether the target has the characteristics of entering the station or leaving the station; Step 5: Establish a dynamic model of the number of personnel and set a tolerance coefficient for abnormal warning; Step 6: The security personnel in each station register the work permit and the information of the entering personnel through a small program, and conduct cross-comparison with the ORM operation platform through the background. The operation information is compared. After passing, it shows that entry is allowed, and the security personnel release them.

2. The method for personnel security management in a substation according to claim 1, characterized in that, In the said Step 1, video stream motion detection and algorithm triggering are adopted; the Gaussian mixture model GMM is used to build the background model of the video stream in the substation gate area, and the pixel-level motion probability is calculated by the frame difference method Where I(x,y,t) is the gray value of pixel (x,y) at time t, B(x,y,t) is the gray value of the background model, α is the threshold coefficient, and σ(x,y) is the variance of the background pixels; when the proportion of moving pixels... When β is the trigger threshold, the YOLO multi-target detection algorithm is automatically started.

3. The method for personnel security management in a substation according to claim 1, characterized in that, In the said Step 2, a personnel detection network is constructed based on the Yolo v5 model, and the objective function includes location loss, confidence loss, and class loss: L=λ loc ·L loc +λ conf ·L conf +λ cls ·L cls Where L is the total loss function, which is the weighted sum of location loss, confidence loss, and class loss through λ. loc For position loss, L conf For confidence loss, L cls For category loss; Specifically: The location loss adopts CIoU, that is, the location loss can be obtained according to the following equation: Among them, b, b gt ρ and c are the coordinates of the predicted bounding box and the ground truth bounding box, respectively; ρ is the Euclidean distance; and c is the length of the diagonal of the smallest rectangle enclosing the two boxes. Where α is the weighting coefficient of the aspect ratio loss; IoU is the area of ​​the intersection of the predicted box and the ground truth box divided by the area of ​​their union; w gt h gt w and h are the width and height of the ground truth bounding box; v is the width and height of the predicted bounding box; and v is a metric for aspect ratio consistency. The confidence loss adopts binary cross-entropy Where y represents the true confidence level. To predict confidence levels; The category loss uses cross-entropy: The detection results output the bounding box coordinates (x, y, w, h), confidence level S, and category label of the personnel.

4. The method for personnel security management in a substation according to claim 1, characterized in that, In the said Step 3, trajectory tracking and in-out behavior judgment: Adopt the Sort algorithm to predict the target state through Kalman filtering: State prediction: Wherein, the state vector It includes location, scale, aspect ratio, and velocity. F is the state transition matrix, where P is the state covariance matrix, which quantifies the uncertainty of state estimation and is dynamically adjusted during prediction and updating to help the algorithm balance the weights of model prediction and observation data; Q is the process noise covariance. Observation update: Where H is the observation matrix, z t R represents the observed values ​​within the detection frame, and R represents the observation noise covariance.

5. The method for personnel security management in a substation according to claim 4, characterized in that, In the said Step 3, the inner and outer boundary lines of the gate are defined as L1: x = x1 and L2: x = x2, x2 < x1. The in-out behavior is judged by the trajectory point coordinates and the velocity direction: Inbound feature: There exists a trajectory point p i (x i ,y i ) satisfies x i > x1 and the previous moment x i-1 < x2, and the velocity component v x > 0; Outbound characteristics: There exists a trajectory point p j (x j , y j ) satisfies x j < x2 and at the previous moment x j-1 > x1, and the velocity component v x < 0.

6. The method for personnel security management in a substation according to claim 1, characterized in that, In step 5, a dynamic model N(t) of the number of people in the station is established, and N(t) is updated based on trajectory entry and exit events: N(t) = N(t-1) + ΔN in (t)-ΔN out (t), where ΔN in (t) represents the number of people entering the station at time t, ΔN out N(t) represents the number of passengers leaving the station; N(t) is transmitted to the ORM system via the MQTT protocol, and the comparison formula is: E(t) = |N(t) - N plan (t)|≤γ·N plan (t), where N plan (t) represents the planned number of people, and γ is the tolerance coefficient. When E(t) > γ·N plan An abnormal warning is triggered when (t), analogous to alarm sensitivity.

7. The method for personnel security management in a substation according to claim 1, characterized in that, In step 6, security personnel register information D = {T, ID, Type, Area} via a mini-program, where T, ID, Type, and Area represent time, personnel ID, job type, and area, respectively, and are compared with the ORM system data D. orm ={T orm ID orm Type orm Area orm Perform cosine similarity calculation: When Sim≥δ is the matching threshold, the review is approved and a request for entry is sent to security and operations personnel. Otherwise, the request is sent to operations personnel, who will decide whether to allow entry.

Images