Substation safety risk management and control platform integrated with artificial intelligence algorithm

Abstract

The invention relates to a substation safety risk management and control platform integrated with an artificial intelligence algorithm. The substation safety risk management and control platform integrated with the artificial intelligence algorithm comprises the following management and control methods: obtaining three-dimensional substation field data; obtaining operation area data based on the three-dimensional substation field data; obtaining three-dimensional position data of the person; judging the risk level of the personnel based on the three-dimensional position data of the personnel and the regional data of the personnel; and obtaining corresponding alarm information based on the risk level. According to the transformer substation safety risk management and control platform integrated with the artificial intelligence algorithm, real-time three-dimensional reconstruction is performed on a transformer substation site through the mobile video acquisition device in combination with the laser radar, intelligent, active and real-time supervision functions of site operation behaviors are realized, an operation area is identified through a fence detection algorithm, personnel positions are tracked in real time through a target identification algorithm, and the safety risk management and control accuracy is improved. Analysis and recognition algorithms of various behaviors are used for monitoring and preventing dangerous behaviors, and the accuracy of safety risk management and control of the transformer substation is improved.

Description

Technical Field

[0001] The present invention relates to the field of substation safety, and more specifically, to a substation safety risk management and control platform integrating artificial intelligence algorithms. Background Art

[0002] Today's society pays more and more attention to the safety and reliability of power supply. As an important part of the power grid, the reliability of substations is closely related to the safe and stable operation of the power grid. The level of substation safety management and control directly affects the probability of safety accidents. Once they occur, they will cause unimaginable consequences and losses. The existing safety management and control system includes various safety operation rules and regulations, safety risk management and control platforms, video surveillance systems, etc.

[0003] However, safety operation rules and regulations, manual video monitoring systems and other methods heavily rely on the safety awareness of the staff themselves and the supervision of the supervisors, and have shortcomings such as low efficiency, high labor costs, and strong subjectivity. In addition, traditional manual supervision methods often have problems such as low efficiency or insufficient supervision.

[0004] The popularity of intelligent inspection robots and drones in video surveillance systems has provided a large amount of video image data. Intelligent patrol systems are used in substations, and intelligent cameras, infrared (ultraviolet) cameras, etc. are installed around the stations to monitor substations. This has greatly promoted the research and application of algorithms for identifying safety violations in substations. Video surveillance combined with artificial intelligence can effectively solve the problems existing in traditional manual video surveillance.

[0005] However, the intelligent inspection robots and drones in video surveillance systems often have the problem of low accuracy. They are unable to accurately measure and record the specific details of the substation, and are unable to accurately locate the equipment. This accuracy problem will affect the effectiveness and accuracy of the overall safety risk management. Summary of the invention

[0006] The purpose of the present invention is to provide a substation safety risk management and control platform with a simple structure, reasonable design and integrated artificial intelligence algorithm in order to solve the above problems.

[0007] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0008] A substation safety risk management and control platform integrating artificial intelligence algorithms, which includes the following management and control methods:

[0009] Obtain 3D substation field data;

[0010] Obtaining operation area data based on 3D substation field data;

[0011] Obtaining three-dimensional position data of personnel;

[0012] Determine the risk level of the person based on the three-dimensional location data of the person and the area data where the person is located;

[0013] Based on the risk level, corresponding alarm information is obtained.

[0014] Based on mobile video acquisition devices and LiDAR equipment, obtain 3D substation field data;

[0015] Specifically, based on the mobile video acquisition device and the laser radar device, the point cloud data and the position data of the device are obtained, the rough alignment is performed based on the position data of the device, and the point cloud line and surface information is extracted for fine alignment;

[0016] When aligning multi-view point clouds: the coordinate transformation includes rotation transformation and translation transformation. The reference point cloud is denoted as P, and the three-dimensional coordinates of the point in the cloud P are p = (x p ,y p ,z p ) T , the three-dimensional coordinates of the point in cloud Q are q = (x q ,y q ,z q ) T , the transformation formula is as follows:

[0017]

[0018] Right now

[0019]

[0020] In the formula, R = R x R y R z is a 3*3 rotation matrix, t=(T x ,T y ,T z ) T is a 3*1 translation matrix, T pq It is a 4*4 rotation and translation transformation matrix;

[0021] Suppose the point clouds Q0, Q1...Q m-1 , there is an overlapping area between two adjacent perspectives, and the stitching process is as follows:

[0022] 4) Take the point cloud Q0 as the reference coordinate system P;

[0023] 5) Taking Q0 as the source point cloud and Q1 as the moving point cloud, the point cloud registration calculates the transformation matrix T from Q1 to Q0 01 ;

[0024] 6) Taking Q1 as the source point cloud and Q2 as the moving point cloud, the point cloud registration calculates the transformation matrix T from Q2 to Q1 12 , then the transformation matrix T from Q2 to Q0 02 =T 01 T 12 ;

[0025] Similarly, calculate Q i (i=2,...,m-1) is transformed to the transformation matrix T of Q0 0i ;

[0026] Based on the transformation matrix T 0i Obtain 3D substation field data.

[0027] The fence support column and fence cloth strip data are obtained based on the target detection algorithm, and M fence support columns {p0,...p m} and N fence strips {f0,...f n}two parts;

[0028] Based on the fence support column and fence strip data, splice to determine the working area. The splicing process is as follows:

[0029] a. Randomly select a support column p i ;

[0030] b. Support column p i Find the distance threshold on both sides less than ζ cloth strip f l 、f r ;

[0031] c.f l 、f r Find the support column p whose distance threshold is less than ζ lj 、p rj ,p lj ≠p i ,p rj ≠p i ;

[0032] d. From the support column p lj 、p rj Find the distance threshold less than ζ and with f l 、f r Inconsistent fabric strips lj 、f rj ;

[0033] e. From f lj 、f rj Find the distance threshold less than ζ and with p lj 、p rj Inconsistent support columns lk 、p rk;

[0034] f. Return to process d and e until no connected support columns or cloth strips can be found, and the operation area is determined by a series of connected support columns and cloth strips;

[0035] g. From the detected M fence support columns {p0,...p m} and N fence strips {f0,...f n}Delete a series of connected support columns and strips of cloth;

[0036] h. Return to process a to g until no connected support columns or strips of cloth can be found.

[0037] Based on the three-dimensional position data of the person and the data of the area where he / she is located, the risk level of the person is judged. Specifically, a target detection algorithm is used to judge whether the person is a staff member, and based on the three-dimensional position data of the person and the data of the area where he / she is located, whether he / she is in the working area is judged. If the person is not a working person and appears in the fenced working area, a dangerous behavior alarm is issued;

[0038] If the person is a worker and appears within the fenced work area, a behavioral analysis and recognition algorithm will be used to detect and prevent dangerous behavior of the person.

[0039] The target recognition algorithm and trajectory tracking algorithm are used to track the three-dimensional position of personnel in real time. The process is as follows:

[0040] 4) Use target detection algorithm to detect the two-dimensional position of people in multi-view visual images;

[0041] 5) Determine the matching relationship between people in multi-view visual images;

[0042] c. Initially, it can be assumed that the value of the person's Z axis in the world coordinates is 0, and the 3D position of the person can be estimated. The person with the closest 3D position in multiple images can be matched;

[0043] The camera imaging model is:

[0044]

[0045] Among them, z c is the distance depth coefficient, u d 、v d is the pixel coordinate point, f cx 、f cy is the focal length of the camera, d x ,d y is the physical length of a pixel, u0 and v0 are the intersection points of the camera optical axis and the image plane, and x w ,y w 、z w is the world coordinate point, R 3*3 、T3*1 is the camera external parameter, M is the projection matrix;

[0046] Assuming that the operator is on the ground, the Z-axis coordinate of the operator in the world coordinate is z w = 0, according to the pixel position u of the ground point of the person in the picture d 、v d , substitute into the camera imaging model, and solve the X and Y axis coordinates x of the operator in the world coordinates w ,y w ;

[0047] d. Use trajectory tracking algorithms to determine the similarity of personnel trajectories and determine the matching relationship of personnel in multi-view visual images;

[0048] 6) Calculate the three-dimensional position of the person in the multi-view visual picture based on the camera imaging principle and the matching relationship between the person in the multi-view visual picture;

[0049] Assume that the two-dimensional positions of the operators in V images are (u v ,v v ),v∈[1,...,V], whose world coordinates are (x w ,y w ,z w ), which can be obtained from the camera imaging model

[0050]

[0051] Among them, M v is the camera projection matrix of the vth frame, and solving the equation yields (x w ,y w ,z w ).

[0052] The three-dimensional electronic perimeter algorithm is used to achieve safety protection of key areas and equipment. Specifically, based on the three-dimensional models of various types of equipment and the three-dimensional positions of personnel obtained by the target recognition algorithm, the closest spatial distance d between the operator and various types of equipment is calculated. If the closest spatial distance d is less than or equal to the dangerous area intrusion distance threshold ξ, then the dangerous area intrusion warning of the person is issued:

[0053] Assume the 3D position of the device is x b ,y b ,z b , the three-dimensional position of the person is x w ,y w ,z w ,but

[0054]

[0055] Set risk levels, specifically including the first risk level, the second risk level, the third risk level, the fourth risk level and the fifth risk level.

[0056] Monitoring and predicting dangerous behaviors of people:

[0057] Based on multiple consecutive pictures, the dangerous behaviors in the pictures are obtained and the risk level is obtained, including:

[0058] If the behavior of not wearing a helmet correctly is shown in multiple consecutive pictures, the second risk level is generated, triggering a dangerous behavior reminder;

[0059] If multiple consecutive pictures show that the user is not wearing a safety belt when working at height, the third risk level is generated, triggering a dangerous behavior reminder;

[0060] If one of the pictures shows a fence-crossing behavior, the fourth risk level is generated, triggering a danger alarm and a dangerous behavior reminder;

[0061] If one of the pictures shows a person not wearing a helmet correctly and crossing a fence, the fifth risk level is generated, triggering a high-level danger alarm and a dangerous behavior reminder;

[0062] Based on multiple consecutive pictures, personnel movement path data is generated, and subsequent movement path data is predicted based on the movement path data. Based on the predicted data and fence position data, dangerous behaviors of personnel are predicted. When the predicted data coincides with the fence position data, the first risk level is generated, triggering a dangerous behavior warning.

[0063] The beneficial effects of the present invention are as follows: the present invention uses a mobile video acquisition device combined with a laser radar to perform real-time three-dimensional reconstruction of the substation site and construct a three-dimensional model of the power station, which can provide more effective distance information, and at the same time perform intelligent analysis of the real-time video of the temporary operation site to achieve intelligent, active and real-time supervision of on-site operation behaviors: identify the operation area through a fence detection algorithm, track the location of personnel in real time through a target recognition algorithm, monitor various behavior analysis and recognition algorithms to prevent the occurrence of dangerous behaviors, and achieve safety protection of key areas and equipment through a three-dimensional electronic perimeter algorithm, thereby improving the accuracy of substation safety risk management and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a flow chart of the control platform method of the present invention;

[0065] Figure 2 It is a flow chart of the control platform of the present invention. DETAILED DESCRIPTION

[0066] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0067] like Figure 1 , Figure 2 As shown, a substation safety risk management and control platform integrating artificial intelligence algorithms includes the following management and control methods:

[0068] First, based on the mobile video acquisition device and LiDAR equipment, the 3D substation field data is obtained;

[0069] Specifically, based on the mobile video acquisition device and the laser radar device, point cloud data and equipment location data are obtained. Point cloud data is a set of three-dimensional data generated by the laser radar. Each point has X, Y, and Z coordinates, representing a point in space. These data are used to build a detailed three-dimensional model of the substation. The equipment location data is the coordinate information of the equipment location and its surrounding environment obtained from the laser radar and the mobile video acquisition device. These data help to understand the specific location and layout of the equipment in the three-dimensional space. Rough registration is performed based on the equipment location data, that is, preliminary data alignment is performed based on the equipment location data to ensure that the laser radar point cloud data and the image data collected by the mobile video roughly match. On the basis of rough registration, the lines and surface information in the point cloud are further extracted to identify and analyze the geometric shape and structural features of the equipment. The point cloud data is accurately aligned through advanced algorithms to ensure that the point cloud data and the video image data are completely matched in space, reduce errors, and make subsequent analysis more accurate.

[0070] When aligning multi-view point clouds: the coordinate transformation includes rotation transformation and translation transformation. The reference point cloud is denoted as P, and the three-dimensional coordinates of the point in the cloud P are p = (x p ,y p ,z p ) T , the three-dimensional coordinates of the point in cloud Q are q = (x q ,y q ,z q ) T , the transformation formula is as follows:

[0071]

[0072] Right now

[0073]

[0074] In the formula, R = R x R y R z is a 3*3 rotation matrix, t=(T x ,T y ,T z ) T is a 3*1 translation matrix, T pq It is a 4*4 rotation and translation transformation matrix;

[0075] Suppose the point clouds Q0, Q1...Q m-1 , there is an overlapping area between two adjacent perspectives, and the stitching process is as follows:

[0076] 7) Take the point cloud Q0 as the reference coordinate system P;

[0077] 8) Taking Q0 as the source point cloud and Q1 as the moving point cloud, the point cloud registration calculates the transformation matrix T from Q1 to Q0 01 ;

[0078] 9) Taking Q1 as the source point cloud and Q2 as the moving point cloud, the point cloud registration calculates the transformation matrix T from Q2 to Q1 12 , then the transformation matrix T from Q2 to Q0 02 =T 01 T 12 ;

[0079] Similarly, calculate Q i (i=2,...,m-1) is transformed to the transformation matrix T of Q0 0i ;

[0080] Based on the transformation matrix T 0i Obtain 3D substation field data.

[0081] The operation area data is obtained based on the three-dimensional substation field data; specifically, the fence support column and fence cloth strip data are obtained based on the target detection algorithm, and M fence support columns {p0,...p m} and N fence strips {f0,...f n}two parts;

[0082] Based on the fence support column and fence strip data, splice to determine the working area. The splicing process is as follows:

[0083] a. Randomly select a support column p i ;

[0084] b. Support column p i Find the distance threshold on both sides less than ζ cloth strip f l 、f r ;

[0085] c.f l 、f r Find the support column p whose distance threshold is less than ζ lj 、p rj ,p lj ≠p i ,p rj ≠p i ;

[0086] d. From the support column p lj 、p rj Find the distance threshold less than ζ and with f l 、f r Inconsistent fabric strips lj 、f rj ;

[0087] e. From f lj 、f rj Find the distance threshold less than ζ and with p lj 、p rj Inconsistent support columns lk 、p rk ;

[0088] f. Return to process d and e until no connected support columns or cloth strips can be found, and the operation area is determined by a series of connected support columns and cloth strips;

[0089] g. From the detected M fence support columns {p0,...p m} and N fence strips {f0,...f n}Delete a series of connected support columns and strips of cloth;

[0090] h. Return to process a to g until no connected support columns or strips of cloth can be found.

[0091] Furthermore, the three-dimensional position data of the personnel is obtained; specifically, the three-dimensional position of the personnel is tracked in real time through the target recognition algorithm and the trajectory tracking algorithm, and the process is as follows:

[0092] 1) Use target detection algorithm to detect the two-dimensional position of people in multi-view visual images;

[0093] 2) Determine the matching relationship between people in multi-view visual images;

[0094] a. Initially, it can be assumed that the value of the person's Z axis in the world coordinates is 0, and the 3D position of the person can be estimated. The person with the closest 3D position in multiple images can be matched;

[0095] The camera imaging model is:

[0096]

[0097] Among them, z c is the distance depth coefficient, u d 、v d is the pixel coordinate point, f cx 、f cy is the focal length of the camera, d x ,d y is the physical length of a pixel, u0 and v0 are the intersection points of the camera optical axis and the image plane, and x w ,y w 、z w is the world coordinate point, R 3*3 、T 3*1 is the camera external parameter, M is the projection matrix;

[0098] Assuming that the operator is on the ground, the Z-axis coordinate of the operator in the world coordinate is Z w = 0, according to the pixel position u of the ground point of the person in the picture d 、v d , substitute into the camera imaging model, and solve the X and Y axis coordinates x of the operator in the world coordinates w ,y w ;

[0099] b. Determine the matching relationship of people in multi-view visual images by judging the similarity of people's trajectories through trajectory tracking algorithm;

[0100] 3) Calculate the three-dimensional position of the person in the multi-view visual picture based on the camera imaging principle and the matching relationship between the person in the multi-view visual picture;

[0101] Assume that the two-dimensional positions of the operators in V images are (u v ,v v ),v∈[1,...,V], whose world coordinates are (x w ,y w ,z w ), which can be obtained from the camera imaging model

[0102]

[0103] Among them, M v is the camera projection matrix of the vth frame, and solving the equation yields (x w ,y w ,z w ).

[0104] The risk levels are set, specifically including a first risk level, a second risk level, a third risk level, a fourth risk level and a fifth risk level, and corresponding alarm information is obtained based on the risk levels.

[0105] Based on the three-dimensional position data of the person and the data of the area where he / she is located, the risk level of the person is determined; specifically, a target detection algorithm is used to determine whether the person is a staff member, and based on the three-dimensional position data of the person and the data of the area where he / she is located, whether he / she is in the working area is determined. If the person is not a working person and appears in the fenced working area, a dangerous behavior alarm is issued;

[0106] If the person is a worker and appears in the fenced work area, a behavior analysis and recognition algorithm will be used to detect and prevent dangerous behavior of the person;

[0107] Furthermore, monitor and predict dangerous behaviors of people: Based on multiple consecutive pictures, obtain the dangerous behaviors in the pictures and obtain the risk level, including:

[0108] If the behavior of not wearing a helmet correctly is shown in multiple consecutive pictures, the second risk level is generated, triggering a dangerous behavior reminder;

[0109] If multiple consecutive pictures show the behavior of working at height without wearing a safety belt, the third risk level is generated, triggering a dangerous behavior reminder;

[0110] If one of the pictures shows a fence-crossing behavior, the fourth risk level is generated, triggering a danger alarm and a dangerous behavior reminder;

[0111] If one of the pictures shows a person not wearing a helmet correctly and crossing a fence, the fifth risk level is generated, triggering a high-level danger alarm and a dangerous behavior reminder;

[0112] Generate personnel movement path data based on multiple consecutive images, predict subsequent movement path data based on the movement path data, predict personnel dangerous behavior based on the predicted data and fence position data, and generate the first risk level and trigger dangerous behavior warning when the predicted data coincides with the fence position data;

[0113] Dangerous behavior reminders can be alarm sounds or voice warnings to remind on-site personnel of safety operation issues;

[0114] Hazard alarms can be flashing lights or indicator lights, providing visual warning signals;

[0115] Advanced hazard alerts can be continuous and high-volume alarm sounds as well as flashing warning lights to further enhance the warning effect.

[0116] In addition, the safety protection of key areas and equipment is achieved through the three-dimensional electronic perimeter algorithm. Specifically, based on the three-dimensional models of various types of equipment and the three-dimensional positions of personnel obtained by the target recognition algorithm, the closest spatial distance d between the operator and various types of equipment is calculated. If the closest spatial distance d is less than or equal to the dangerous area intrusion distance threshold ξ, then the dangerous area intrusion warning of the person is issued:

[0117] Assume the 3D position of the device is x b ,y b ,z b , the three-dimensional position of the person is x w ,y w ,z w ,but

[0118]

[0119] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A substation safety risk management and control platform integrating artificial intelligence algorithms, characterized in that: The platform includes the following control methods: Obtain 3D substation field data; Obtaining operation area data based on 3D substation field data; Obtaining three-dimensional position data of personnel; Determine the risk level of the person based on the three-dimensional location data of the person and the area data where the person is located; Based on the risk level, corresponding alarm information is obtained.

2. According to the substation safety risk management and control platform integrating artificial intelligence algorithm according to claim 1, it is characterized in that: Based on mobile video acquisition devices and LiDAR equipment, obtain 3D substation field data; Specifically, based on the mobile video acquisition device and the laser radar device, the point cloud data and the position data of the device are obtained, the rough alignment is performed based on the position data of the device, and the point cloud line and surface information is extracted for fine alignment; When aligning multi-view point clouds: the coordinate transformation includes rotation transformation and translation transformation. The reference point cloud is denoted as P, and the three-dimensional coordinates of the point in the cloud P are p = (x p ,y p ,z p ) T , the three-dimensional coordinates of the point in cloud Q are q = (x q ,y q ,z q ) T , the transformation formula is as follows: Right now In the formula, R = R x R y R z is a 3*3 rotation matrix, t=(T x ,T y ,T z ) T is a 3*1 translation matrix, T pq It is a 4*4 rotation and translation transformation matrix; Suppose the point clouds Q0, Q1...q under M perspectives to be stitched m-1 , there is an overlapping area between two adjacent perspectives, and the stitching process is as follows: 1) Take the point cloud Q0 as the reference coordinate system P; 2) Taking Q0 as the source point cloud and Q1 as the moving point cloud, the point cloud registration calculates the transformation matrix T from Q1 to Q0 01 ; 3) Taking Q1 as the source point cloud and Q2 as the moving point cloud, the point cloud registration calculates the transformation matrix T from Q2 to Q1 12 , then the transformation matrix T from Q2 to Q0 02 =T 01 T 12 ; Similarly, calculate Q i (i=2,...,m-1) is transformed to the transformation matrix T of Q0 0i ; Based on the transformation matrix T 0i Obtain 3D substation field data.

3. According to the substation safety risk management and control platform integrating artificial intelligence algorithm according to claim 2, it is characterized in that: The fence support column and fence cloth strip data are obtained based on the target detection algorithm, and M fence support columns {p0,...p m } and N fence strips {f0,...f n }two parts; Based on the fence support column and fence strip data, splice to determine the working area. The splicing process is as follows: a. Randomly select a support column p i ; b. Support column p i Find the distance threshold on both sides less than ζ cloth strip f l 、f r ; c.f l 、f r Find the support column p whose distance threshold is less than ζ lj 、p rj ,p lj ≠p i ,p rj ≠p i ; d. From the support column p lj 、p rj Find the distance threshold less than ζ and with f l 、f r Inconsistent fabric strips lj 、f rj ; e. From f lj 、f rj Find the distance threshold less than ζ and with p lj 、p rj Inconsistent support columns lk 、p rk ; f. Return to process d and e until no connected support columns or / and cloth strips can be found, and the operation area is determined by a series of connected support columns and cloth strips; g. From the detected M fence support columns {p0,...p m } and N fence strips {f0,...f n }Delete a series of connected support columns and strips of cloth; h. Return to process a to g until no connected support columns and / or cloth strips can be found.

4. According to the substation safety risk management and control platform integrating artificial intelligence algorithm according to claim 3, it is characterized in that: Based on the three-dimensional position data of the person and the data of the area where he / she is located, the risk level of the person is judged. Specifically, a target detection algorithm is used to judge whether the person is a staff member, and based on the three-dimensional position data of the person and the data of the area where he / she is located, whether he / she is in the working area is judged. If the person is not a working person and appears in the fenced working area, a dangerous behavior alarm is issued; If the person is a worker and appears within the fenced work area, a behavioral analysis and recognition algorithm will be used to detect and prevent dangerous behavior of the person.

5. A substation safety risk management and control platform integrating artificial intelligence algorithm according to claim 4, characterized in that: The target recognition algorithm and trajectory tracking algorithm are used to track the three-dimensional position of personnel in real time. The process is as follows: 1) Use target detection algorithm to detect the two-dimensional position of people in multi-view visual images; 2) Determine the matching relationship between people in multi-view visual images; a. Initially, the value of the person's Z axis in the world coordinates is set to 0, and the 3D position of the person is estimated. The person with the closest 3D position in multiple images can be matched; The camera imaging model is: Among them, z c is the distance depth coefficient, u d 、v d is the pixel coordinate point, f cx 、f cy is the focal length of the camera, d x ,d y is the physical length of a pixel, u0 and v0 are the intersection points of the camera optical axis and the image plane, and x w ,y w 、z w is the world coordinate point, R 3*3 、T 3*1 is the camera external parameter, M is the projection matrix; Assume that the operator is on the ground, and the Z-axis coordinate of the operator in the world coordinate is z w = 0, according to the pixel position u of the ground point of the person in the picture d 、v d , substitute into the camera imaging model, and solve the X and Y axis coordinates x of the operator in the world coordinates w ,y w ; b. Determine the matching relationship of people in multi-view visual images by judging the similarity of people's trajectories through trajectory tracking algorithm; 3) Calculate the three-dimensional position of the person in the multi-view visual picture based on the camera imaging principle and the matching relationship between the person in the multi-view visual picture; Assume that the two-dimensional positions of the operators in V images are (u v ,v v ),v∈[1,...,V], whose world coordinates are (x w ,y w ,z w ), which can be obtained from the camera imaging model Among them, M v is the camera projection matrix of the vth frame, and solving the equation yields (x w ,y w ,z w ).

6. A substation safety risk management and control platform integrating artificial intelligence algorithm according to claim 5, characterized in that: The three-dimensional electronic perimeter algorithm is used to achieve safety protection of key areas and equipment. Specifically, based on the three-dimensional models of various types of equipment and the three-dimensional positions of personnel obtained by the target recognition algorithm, the closest spatial distance d between the operator and various types of equipment is calculated. If the closest spatial distance d is less than or equal to the dangerous area intrusion distance threshold ξ, then the dangerous area intrusion warning of the person is issued: Let the three-dimensional position of the device be x b ,y b ,z b , the three-dimensional position of the person is x w ,y w ,z w ,but 7. A substation safety risk management and control platform integrating artificial intelligence algorithm according to claim 6, characterized in that: Set risk levels, specifically including the first risk level, the second risk level, the third risk level, the fourth risk level and the fifth risk level.

8. A substation safety risk management and control platform integrating artificial intelligence algorithm according to claim 7, characterized in that: Monitor and predict dangerous behaviors of personnel; Based on multiple consecutive pictures, the dangerous behaviors in the pictures are obtained and the risk level is obtained, including: If the behavior of not wearing a helmet correctly is shown in multiple consecutive pictures, the second risk level is generated, triggering a dangerous behavior reminder; If multiple consecutive pictures show the behavior of working at height without wearing a safety belt, the third risk level is generated, triggering a dangerous behavior reminder; If one of the pictures shows a fence-crossing behavior, the fourth risk level is generated, triggering a danger alarm and a dangerous behavior reminder; If one of the pictures shows a person not wearing a helmet correctly and crossing a fence, the fifth risk level is generated, triggering a high-level danger alarm and a dangerous behavior reminder; Based on multiple consecutive pictures, personnel movement path data is generated, and subsequent movement path data is predicted based on the movement path data. Based on the predicted data and fence position data, dangerous behaviors of personnel are predicted. When the predicted data coincides with the fence position data, the first risk level is generated, triggering a dangerous behavior warning.

Images