Distribution line construction supervision system based on edge calculation and swarm intelligence

Abstract

Aiming at the problems of insufficient real-time performance, low data processing efficiency, poor collaboration and the like in the existing distribution line construction supervision, the invention provides a construction supervision system, which adopts a distributed architecture and comprises an edge computing node and a cloud center, and data communication exists between the cloud center and the edge computing node. The edge computing nodes are in data communication with the existing systems of one or more construction sites; the cloud center is configured to deploy an image detection model and execute image detection model training; and / or configuring and deploying a task scheduling algorithm, and performing task scheduling on the plurality of construction tasks; and the edge computing node is configured and deployed with a trained image detection model and performs behavior recognition on the monitoring video of the construction task. Cooperative work is achieved through a task scheduling algorithm, localization rapid processing of construction site data is achieved through the edge computing node, complex supervision tasks are completed jointly, and the real-time performance and reliability of construction supervision are improved through the cloud center and the edge computing node.

Description

Technical Field

[0001] This case involves a distribution line construction supervision system, and particularly involves a distribution line construction supervision system based on edge computing and swarm intelligence. Background Art

[0002] The existing systems at the distribution line construction site include a monitoring system, an operation and maintenance system, a dispatching system, and other systems. In addition, there are video devices and other devices. Each system and device works independently, collects data independently, and then processes and configures it manually. Therefore, there are problems such as insufficient real-time performance, low data processing efficiency, and poor coordination. Summary of the Invention

[0003] To solve or at least partially solve the problems of insufficient real-time performance, low data processing efficiency, and poor coordination existing in traditional distribution line construction supervision, this case aims to use edge computing technology to achieve rapid local processing of on-site construction data, and integrate the collective decision-making ability of distributed devices through swarm intelligence algorithms, so as to build an efficient, intelligent, and reliable construction supervision network. The technical solutions are as follows.

[0004] A construction supervision system, the construction supervision system adopts a distributed architecture, including a cloud center and multiple edge computing nodes. There is data communication between the cloud center and the edge computing nodes, and there is data communication between the edge computing nodes and one or more existing systems at the construction site. The existing systems include a monitoring system and a dispatching system; the cloud center is configured to deploy an image detection model and perform image detection model training; and / or is configured to deploy a task scheduling algorithm to schedule multiple construction tasks; the edge computing node is configured to deploy the trained image detection model to perform behavior recognition on the monitoring video of the construction task.

[0005] In an implementation manner of the above technical solution: The cloud center sets up a cloud server, and the cloud server is configured to store the monitoring video of the construction task or other videos.

[0006] In an implementation manner of the above technical solution: The monitoring video or other videos are stored in the Object Storage Service (OSS) manner.

[0007] In an implementation manner of the above technical solution: The cloud center and the edge computing nodes communicate using the MQTT protocol.

[0008] In an implementation manner of the above technical solution: The image detection model is an improved YOLO algorithm model, and the improved YOLO algorithm model introduces a Spatial Pyramid Pooling module into the original YOLO algorithm model and uses GIoU as the loss function.

[0009] In one implementation of the above technical solution: The goal of the task scheduling algorithm is to minimize the function where:

[0010]

[0011] In the formula: The subscript i is the task identifier, the subscript min represents the minimum value, the subscript max represents the maximum value, TCC is the total task completion cost, LB is the resource load balance degree, and TCT is the total task completion time.

[0012] In one implementation of the above technical solution: The task scheduling algorithm adopts an improved beluga whale algorithm. The improvements of the improved beluga whale algorithm compared with the original beluga whale algorithm include introducing a simulated annealing algorithm for local position update and adopting a quasi-oppositional learning strategy for global position update; The local position update includes:

[0013] (1) When the optimized objective function value of the potential new position is lower than the current position, let the current beluga whale go to the potential new position in the next step to complete the position update;

[0014] (2) When the optimized objective function value of the potential new position is not lower than the current position, calculate the acceptance probability p of the worse position to guide the next step of the current beluga whale:

[0015]

[0016] In the formula: Obj T is the objective function value of the current position of the beluga whale, Obj T+1 is the objective function value of the new position, T is the current iteration number, C is the cooling rate of the simulated annealing parameter, and Temperature is the initial temperature of the simulated annealing;

[0017] The global position update includes: Denote the number of beluga whale populations as N, the search space dimension as d, and the position of the i-th beluga whale in the d-th dimensional space is expressed as respectively represent the lower bound and upper bound of, define the corresponding reverse solution is Then the reverse solution under the quasi-oppositional learning strategy is:

[0018]

[0019] In the formula: rand(0, 1) is a random number between (0, 1), i = 1, 2,..., N, j = 1, 2,..., d.

[0020] In one implementation of the above technical solution: The edge computing node collects data from the existing system through the OPC UA protocol.

[0021] In one implementation of the above technical solution: The edge computing node obtains data from the device through the device's own protocol, binds the device's data information source to the device object information model, and updates the data of the device object information model when the device's data information source changes.

[0022] In one implementation of the above technical solution: The edge computing node is configured to send a video analysis request to the cloud center when the computing power does not meet the set threshold, and obtain the behavior recognition result of monitoring the construction task from the cloud center; The cloud center is configured to respond to the video analysis request of the edge computing node, send and store the behavior recognition result of monitoring the construction task video to the edge computing node.

[0023] The beneficial technical effects of this case: Through this distributed architecture using edge computing nodes and cloud centers, the construction supervision system can achieve efficient, intelligent, and reliable all-round supervision. It not only greatly improves the real-time performance and accuracy of supervision, but also realizes collaborative work on multiple construction sites through a task scheduling algorithm. The task scheduling algorithm uses an optimized beluga algorithm, and the optimized beluga algorithm has strong robustness and high adaptability, thus improving the practicality of the construction supervision system. This innovative supervision method effectively improves the construction quality and safety level, and at the same time provides important technical support and experience accumulation for the construction and operation and maintenance of future smart grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only exemplary, and for those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0025] Figure 1 、 one Schematic diagram of the system structure in one implementation.

[0026] Figure 2 、 one Schematic diagram of the device object information modeling process in one implementation.

[0027] Figure 3 、 one Schematic diagram of the data collection process in one implementation.

[0028] Figure 4 、 one MQTT communication architecture in one implementation.

[0029] Figure 5 、 oneSchematic diagram of the video data management process in a certain implementation manner.

[0030] Figure 6 、 one Schematic diagram of the SPP structure in a certain implementation manner.

[0031] Figure 7 、 one Schematic diagram of GIoU in a certain implementation manner.

[0032] Figure 8 、 one Schematic diagram of the swarm intelligence algorithm process in a certain implementation manner.

[0033] Figure 9 、 one Schematic diagram of the process of the improved beluga whale optimization algorithm in a certain implementation manner. Specific implementation manner

[0034] In view of the problems existing in the traditional construction supervision of distribution lines, such as insufficient real-time performance, low data processing efficiency, and poor coordination, the technical solution proposed in this disclosure improves the real-time performance of construction supervision by deploying multiple edge computing nodes at the construction site and using the nodes to independently collect and preliminarily analyze data; and by setting up a cloud center, the task scheduling algorithm of the cloud center is used to achieve collaborative work of multiple construction sites and jointly complete complex supervision tasks. At the same time, the supervision system integrates resources in the cloud, realizing collaborative computing between the edge and the cloud, which not only ensures real-time performance and reliability, but also provides powerful data analysis and decision-making support capabilities.

[0035] For the explanation of the relevant terms of the technical solution of this disclosure, see Table 1 below.

[0036] Table 1

[0037]

[0038]

[0039]

[0040] The following clearly and completely describes how to implement the technical solution of this case. Obviously, the described implementation manners are only a part of the implementation manners of this case, rather than all the implementation manners. Based on the implementation manners in this case, all other implementation manners obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0041] (I) Overall system structure

[0042] See Figure 1 , the existing system at the construction site includes a monitoring system, an operation and maintenance system, a dispatching system, and other systems.

[0043] The supervision system adopts a distributed architecture and is built based on the existing systems and other devices at the construction site without affecting the operation of the existing systems. The supervision system is divided into two layers hierarchically: the cloud center and the edge computing nodes at the construction site. There is data communication between the cloud center and the edge computing nodes, and there is data communication between the edge computing nodes and one or more existing systems at the construction site. In Figure 1 it, although it is shown that one edge computing node corresponds to one construction site, according to the computing power of the edge computing node, it can also correspond to multiple construction sites.

[0044] Among them, the cloud center includes a cloud storage center that stores the video data of the construction site. The cloud center also includes an algorithm training platform for training algorithm models, and the training data is the data published by the edge computing nodes to the cloud center. The cloud center publishes the trained algorithm models to each edge computing node to ensure the update of the algorithms.

[0045] Among them, the built-in GPU in the edge computing node provides computing power for the behavior recognition algorithm calculation in the operation process. The data collected by the edge computing node from the existing systems at the construction site is backed up in the local storage after participating in the information data flow. The edge computing node can publish the collected data to the cloud center for the algorithm training platform in the cloud center to train algorithm models. When the cloud center publishes deep learning inference or judgment algorithms, etc., the edge computing node will store the newly published deep learning inference or judgment algorithms.

[0046] Part of the data for the operation of the supervision system comes from the original systems and devices at the construction site and new data sources can be added according to requirements. In order not to affect the existing systems, only the required data is collected from the existing systems according to requirements, and the data is collected from the existing systems to the edge computing nodes through the OPC UA protocol. For the data that is not available at present, the device data can also be transmitted to the edge computing nodes through other protocols, such as the device's own protocol.

[0047] (2) Data Transmission and Storage of the Existing Systems, Node Layer and Center Layer

[0048] The communication feature between the existing systems at the construction site and the edge computing nodes is that when the edge computing node collects data from the existing systems using the OPC UA protocol, based on the information specification and for the actual requirements of the supervision of the construction operation process, the device object information is modeled, and the construction process is as Figure 2As shown below: For a device object, obtain its device attributes. The device attributes include device information such as parameter descriptions, initial values, and numerical types. In addition, consider the device object as a node and set the corresponding node identifier Node ID for the current device. Based on this, an information model is obtained, and the information model is classified according to the device parameter information. An XML description file is generated according to the information model, and the data information source of the device is bound to the information model using the Node ID, so that when the device information changes, the corresponding data information is updated.

[0049] Taking a distance sensor as an example, establish a Dnc Machine information model object, classify the information model according to the device parameter information, and set information such as the Node ID, parameter description, initial value, and numerical type of the corresponding node. Data acquisition is completed in the Open62541 open-source SDK. An OPC UA server is established, and then the information model file of the device object is loaded into the server. The data information source of the device is bound to the information model through the Node ID. When the device information changes, the corresponding device data is updated through the SDK. Finally, by accessing the OPC UA server (i.e., the edge computing node) through the client, various data information of the device can be read.

[0050] The data acquisition process is as Figure 3 shown. In Figure 3 it, the edge computing node reads the device operation parameters, configures the read operation parameters to be shared in memory, and initializes the UA data set. The OPC UA server in the edge computing node is started, and a listening task is started. If there is a client data reading request, the device parameter information is read from the shared memory and returned. At the same time as configuring the operation parameters to be shared in memory, it also includes accessing the device driver, and when the data source is updated, the shared memory information is updated. In one implementation, the edge computing node uses the relational database MySQL for real-time data storage. The advantage of using MySQL is that SQL Bridge supports creating transaction groups (Transaction Groups), which synchronize data between the PLC and the database. Through the transaction group, it is possible to log in to the database from the PLC, move data back from the database to the PLC, or even keep the two synchronized. The transaction group reads the real-time data values from the OPC address in the simplest form and stores them in the SQL database. SQL Bridge can automatically create and manage database tables for each transaction group, enabling data recording and querying without the need for the technical foundation of writing SQL queries or creating database tables.

[0051] In one implementation, the edge computing node communicates with the cloud center to transfer real-time data through the MQTT protocol. The cloud center located in the cloud subscribes to corresponding topics as needed, and part of the real-time data is also stored in the cloud for unified data mining, data analysis, and data backup and recovery when necessary. In this implementation, three MQTT modules are used, namely the MQTT TransmitterModule, the MQTT Engine Module, and the MQTT DistributorModule to implement MQTT data communication.

[0052] Specifically, refer to Figure 4 the schematic diagram of the MQTT communication architecture. In Figure 4 it, the edge computing node acts as an MQTT publisher and sends data to the broker located in the cloud center. The broker synchronizes relevant information to the central station according to the content subscribed by the central station. The broker can be set either at the edge computing node or in the cloud, and a third-party broker can also be used.

[0053] The MQTT Transmitter module can generate an MQTT Sparkplug message after the tag value is updated and then send the data to the broker, enabling the subscriber to apply the data. The MQTT Transmitter supports the transmission and sending of OPC tags and also supports the generation and publication of customizations. Therefore, setting the MQTT Transmitter at the edge computing node makes the edge computing node act as an MQTT publisher. The tags are the object categories in the image in object detection and the position and range of the object in the image. For example, in an actual scenario, there are insulators and transformers. Correspondingly, the tags are insulators, transformers, and their positions and ranges in the image.

[0054] The MQTT Distributor module can be placed in the cloud center as a communication broker or a third-party server can be used as the MQTT communication broker.

[0055] The MQTT Engine module can act as an MQTT protocol subscriber, installed in the cloud center to receive messages published by the broker.

[0056] Refer to Figure 4 the schematic diagram of the MQTT communication architecture. In Figure 4 it, the edge computing node acts as an MQTT publisher and sends data to the broker located in the cloud center. The broker synchronizes relevant information to the central station according to the content subscribed by the central station. The broker can be set either at the edge computing node or in the cloud, and a third-party broker can also be used.

[0057] In one implementation, the management of video data includes the processes of uploading, storing, processing, and invoking video data. The video data is uploaded to OSS for storage and then transferred to the cloud server ECS for processing. After that, the results are transferred to the edge computing node or the client, and at the same time, the results are stored in OSS. The client can also call historical videos for viewing at any time.

[0058] In one implementation, the edge computing node can choose to download the video data stored in the cloud server ECS for behavior recognition, or it can choose to directly perform behavior recognition on the cloud server ECS and only obtain one behavior recognition result.

[0059] In one implementation, behavior recognition is preferentially performed on the monitoring videos of construction tasks at the edge computing node. When the computing power of the edge computing node is insufficient, a video analysis request is sent to the cloud center, and the behavior recognition result of the monitoring video of the construction task is obtained from the cloud center; the cloud center is configured to respond to the video analysis request of the edge computing node, send and store the behavior recognition result of the monitoring video of the construction task to the edge computing node. Insufficient computing power means that when performing a computing task, the required computing resources or capabilities are insufficient to meet the task requirements, resulting in low computing efficiency or inability to complete the task, such as setting hardware resource metric indicators for judgment.

[0060] See Figure 5 , the construction videos collected by the network camera are managed and video-analyzed on the ESC cloud server. The ESC cloud server stores the videos using the OSS object storage method and establishes index storage. Based on the stored videos, after video transcoding and media storage, the videos can be shared with the CDN content delivery network, and then the content is distributed to the client through the CDN content delivery network. The client can also directly perform video playback from the OSS storage object storage or manage and video-analyze the videos in the ECS cloud server. Here, the client can be the user side or the edge computing node.

[0061] (III) Behavior Cognition

[0062] In one implementation, for the behavior recognition of various operations during the construction process, the present disclosure uses a deep learning algorithm based on YOLO for image detection, optimizes the loss function and structure in the algorithm, and integrates the results into the job process supervision. While enhancing the system's supervision ability for the job process, it also enhances the system's intelligence level and perception ability. Using supervision systems such as the operation ticket and work ticket management functions, monitoring the video information provided by the equipment, the system analysis capabilities provided by the edge computing node, and the operations and supervision of the operators, the collaborative supervision of humans, machines, and objects in the construction operation process is realized to a certain extent.

[0063] Specifically, an SPP module is introduced to solve the problem of non-fixed image size in convolutional neural networks, so that image data of any size can be input. After adding the SPP module, the network can crop and scale pictures without distortion, reducing the loss of picture information. Secondly, it can avoid the repeated extraction of picture feature values to a certain extent, thus accelerating the calculation time of the network and saving calculation costs.

[0064] Comparing the SPP structure diagram of the present disclosure ( Figure 6 (b) in it) with the general SPP structure diagram of the prior art ( Figure 6 (a) in it), it can be seen that in the technical solution of the present disclosure, the SPP module is a four-parallel-branch structure composed of 3 max-pooling layers plus a skip connection. Adding the SPP module to the algorithm is not for distortion-free cropping. Drawing on the idea of the spatial pyramid in this module, using this module to find the global and local features of pictures in the model and combining the local and global features of image data can enable the feature tensor to better represent the features of the image, further improving the detection accuracy.

[0065] Specifically, the loss function uses GIoU (Generalized Intersection over Union) to replace the method of using mean squared error in the original YOLO algorithm. GIoU is an improved IoU (Intersection over Union) metric used to evaluate the overlap between the predicted bounding box and the ground truth bounding box in object detection.

[0066] The calculation formula of GIoU is as follows.

[0067]

[0068] Where: A and B are any two boxes in the plane, and C is the smallest box that can enclose A and B. See Figure 7 The illustration of GIoU.

[0069] At this time, the loss function is:

[0070] Loss GIoU = 1 - GIoU

[0071] When using GIoU as the distance, the loss function of the model has non-negativity and uncertainty. At the same time, as a ratio operation, GIoU also has the property of scale invariance. At the same time, GIoU repeatedly considers the overlapping method between the graphics and can fully reflect the characteristics of the overlapping method between the bounding box and the ground truth bounding box during bounding box regression. By introducing GIoU as the distance, the position and size relationship between the bounding box and the ground truth bounding box in the loss function can be reflected.

[0072] The specific implementation method is as follows:

[0073] First, assume that the predicted bounding box B p and the ground truth bounding box B t have the following coordinates:

[0074] And there is

[0075] In the formula, is the upper left coordinate of B p , is the lower right coordinate of B p ; is the upper left coordinate of B t , is the lower right coordinate of B t .

[0076] Then the area of the predicted bounding box and the ground truth bounding box B p and B t is:

[0077]

[0078] In the formula, S p is the area of the predicted bounding box, and S t is the area of the ground truth bounding box.

[0079] The area of the overlapping part between the ground truth bounding box B t and the predicted bounding box B p is:

[0080]

[0081] In the formula, is the upper left coordinate of the overlapping part, is the lower right coordinate of the overlapping part.

[0082] The coordinates and area of the smallest bounding box C that contains the predicted bounding box and the ground truth bounding box are:

[0083]

[0084] In the formula, is the upper left coordinate of the smallest bounding box C, is the lower right coordinate of the smallest bounding box C.

[0085] The following formula is the value of IoU:

[0086]

[0087] The following formula is the value of GIoU:

[0088]

[0089] The loss function regarding GIoU is as follows:

[0090] Loss GIoU = 1 - GIoU

[0091] (4) Task Scheduling

[0092] In one implementation, the task scheduling in the edge computing node adopts the group intelligence algorithm based on the cloud platform - the beluga whale algorithm. To address the problem that it is prone to falling into local optimal values, the simulated annealing algorithm is introduced. The Metropolis criterion is used to calculate the probability and accept the sub-optimal solution with this probability, enabling the algorithm to jump out of the local optimum. At the same time, a quasi-opposite learning strategy is added after the whale fall step of the beluga whale optimization algorithm to avoid one-way search, expand the solution range, and improve the global search ability of the algorithm. The schematic diagram of the group intelligence algorithm process is shown in Figure 8 .

[0093] (4.1) Establish a cloud computing task scheduling model.

[0094] Specifically, the mapping between the tasks (construction objectives) submitted by users and virtual machines (representing the set of workers in reality) can be described as follows: The user task set is T = {t1, t2,..., t n}, the virtual machine set is V = {v1, v2,..., v m}, and n >> m. It is stipulated that a task is only allowed to run on one virtual machine. Then the relationship between the task and the virtual machine can be represented by the matrix TV map as follows:

[0095]

[0096] T i V j means that task t i is assigned to virtual machine v j to run. When t i is assigned to v j , T i V j equals 1, otherwise it equals 0.

[0097] After completing the pre-sorting of user tasks and virtual machines, the formulas for calculating the priorities of tasks and virtual machines are as follows:

[0098] t i = taskLength i + taskSize i + taskOutput i

[0099] v j = cpj ×0.1 + bw j ×0.1 + size j ×0.05 + ram j ×0.1

[0100] Where: taskLength i represents the computational complexity of user task t i , taskSize i represents the data size of user task t i , taskOutput i represents the size of the user task result output; cp j represents the computing power of virtual machine v j , bw j is the CPU bandwidth of virtual machine vj, size j represents the external memory size of the virtual machine; ram j represents the memory size of the virtual machine.

[0101] Among them, cp j can be calculated by the following formula:

[0102] cp j = numCPU j × mipsCPU j

[0103] mips is used to describe the computing speed of a computer, mipsCPU j is the computing speed of a single processor of virtual machine vj, numCPU j is the number of processors of virtual machine vj, and the product of the two is used to measure the computing power of this virtual machine.

[0104] The user tasks and virtual machines are pre-sorted according to the above rules to form a new task-virtual machine mapping relationship, so that lightweight tasks are assigned to low-performance virtual machines and large-load tasks are assigned to high-performance virtual machines.

[0105] The Time matrix represents the predicted running time of the task set on the virtual machine:

[0106]

[0107] In this matrix, Timeij represents task t i on virtual machine v j 's predicted running time. The running time consists of two parts: data transfer time and computing time, and is defined as:

[0108]

[0109] (4.2) Construct the fitness function.

[0110] First, create the task completion time function.

[0111] The virtual machines process tasks in parallel and without interruption at full load. The total completion time is the maximum of the predicted execution times among all virtual machines. The total task completion time TCT (Total Completion Time) can be calculated from the Time matrix:

[0112]

[0113] In the formula: represents the sum of the predicted running times of all tasks allocated to the virtual machines. Taking the maximum value of this value among all virtual machines gives the total task completion time.

[0114] Secondly, create the resource load balance function.

[0115] During task scheduling, the system load balance has a significant impact on system performance and is closely related to QoS and resource utilization. To address this issue, a resource load balance function is introduced and defined as:

[0116]

[0117] In the formula: m represents the total number of virtual machines in the system, represents the total number of tasks allocated to virtual machine v j , represents the average number of tasks allocated to each virtual machine. This formula is used to measure the overall performance stability of virtual machine v j during the process of processing tasks with the number of .

[0118] Then, create the task cost function.

[0119] The total cost TCC (Total Completion Cost) for the system to complete tasks can be obtained by multiplying the running time of the virtual machine by its corresponding cost per unit time:

[0120]

[0121] In the formula: Cost j is the cost per unit time of virtual machine v j , and the calculation formula is:

[0122] Cost j = mispsCPU j × 3 + bw j × 0.1 + size j × 0.05 + ram j × 0.1

[0123] Subsequently, a fitness function is constructed.

[0124] Specifically, the selected algorithm objective is to minimize the total user task completion time, system load balance, and total task completion cost, which is a multi-objective optimization problem.

[0125] Multi-objective optimization needs to satisfy several objective functions simultaneously. Generally, there are interactions between objective functions, and the optimization of one objective function will sacrifice the interests of other objective functions. The cloud computing task scheduling strategy in this patent comprehensively evaluates from three aspects: the total time span of user tasks TCT (Total Completion Time), system load balance LB (Load Balance), and total task completion cost TCC (Total Completion Cost). Its mathematical model is:

[0126] min(TCT), min(LB), min(TCC)

[0127] Among them, min(TCT) represents the objective function of minimizing the total task completion time, min(LB) represents the objective function of minimizing the system load balance, and min(TCC) represents the objective function of minimizing the total task completion cost.

[0128] This disclosure adopts the geometric mean method commonly used to solve multi-objective problems, transforming the multi-objective into a single-objective optimization problem. The optimized objective function can be expressed as:

[0129]

[0130] Where: The subscript i is the task identifier, the subscript min represents the minimum value, and the subscript max represents the maximum value.

[0131] (4.3) Optimization of the beluga whale algorithm

[0132] The optimization objective of the improved beluga whale optimization algorithm is the overall best of task completion time, system load balance, and task completion cost, that is, downward optimization.

[0133] Introduce the simulated annealing algorithm, and design the rule for updating the position of the beluga whale as:

[0134] (1) When the value of the optimized objective function at the potential new position is lower than the current position, let the current beluga whale move to the potential new position in the next step to complete the position update:

[0135] (2) When the optimized objective function value of the potential new position is not lower than the current position, the Metropolis criterion is introduced to propose the calculation formula p of the acceptance probability of the worse position to guide the next move of the current beluga whale:

[0136]

[0137] In the formula: Obj T is the objective function value of the current position of the beluga whale, Obj T+1 is the objective function value of the new position, T is the current iteration number, p C is the cooling rate of the simulated annealing parameter, and Temperature is the initial temperature of the simulated annealing.

[0138] As the number of iterations increases, the acceptance probability p of the worse position will gradually decrease. This mechanism helps the beluga whale optimization algorithm to jump out of the local optimum in the early stage of optimization.

[0139] Introduce the quasi-opposite learning strategy. Assume that the number of beluga whales in the population is N, the search space is d-dimensional, and the position of the i-th beluga whale in the d-th dimensional space is represented as:

[0140]

[0141] In the formula: i = 1, 2,..., N. j = 1, 2,..., d. respectively represent the lower bound and the upper bound. Define the corresponding opposite solution as:

[0142]

[0143] Based on the requirements of this disclosure, the calculation method of the quasi-opposite solution is designed as follows:

[0144]

[0145] In the formula: rand(0, 1) is a random number between (0, 1).

[0146] The flow chart of the improved beluga whale optimization algorithm is as Figure 9 shown.

[0147] The general process is as follows: First, initialize the beluga whale population, calculate the objective function values of all beluga whales, and select the minimum value as the current optimal position. Then calculate the balance factor B f , when B f > 0.5, the algorithm enters the exploration stage and calculates the next position of the beluga whale; when B f ≤ 0.5, the algorithm enters the exploitation stage and calculates the next position of the beluga whale.

[0148] If the objective function value of the next position of the beluga whale is less than that of the current position of the beluga whale, accept this new position and transfer the beluga whale to the new position.

[0149] If the objective function value of the next position of the beluga whale is greater than that of the current position of the beluga whale, the existing beluga whale optimization algorithm will directly abandon this position, making the algorithm unable to jump out of the local optimum.

[0150] The improved beluga whale optimization algorithm introduces the Metropolis criterion. According to this criterion, calculate the probability p of accepting a worse position, generate a random number between (0, 1), compare the size of p and this random number. If the random number is greater than the probability p, abandon this new position. If the random number is less than the probability p, accept this new position and transfer the beluga whale to the new position; then calculate the objective function values of all beluga whales and select the minimum value as the current optimal position to complete the update of the current optimal position.

[0151] Calculate the whale fall probability W f , compare B f and W f When B f <W f , the algorithm enters the whale fall stage until the number of algorithm iterations reaches the preset upper limit, and the algorithm ends and outputs the optimal solution.

[0152] In the description of the present disclosure, the orientation or positional relationship indicated by terms such as "upper left", "lower right", "upper", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present application.

[0153] (V) Summary

[0154] In the present disclosure, by deploying edge computing nodes at the construction site, the local and rapid processing of data is realized. This distributed architecture significantly reduces data transmission latency and improves the system response speed. For example, in video data processing, the edge node can perform preliminary analysis and only transmit key information to the cloud, greatly reducing the network bandwidth requirement and improving the supervision efficiency. By introducing the improved YOLO algorithm for image recognition, and by adding the SPP module and the GIoU loss function, the recognition accuracy of various operations at the construction site is significantly improved. This not only enhances the perception ability of the supervision system but also improves the detection efficiency of illegal operations, thus better ensuring construction safety and quality. By adopting the improved beluga whale optimization algorithm, the collaborative work of multiple construction sites is realized. This collaborative mechanism not only improves the overall decision-making ability of the system but also enhances the adaptability to complex construction environments. For example, in task scheduling, the improved beluga whale algorithm can comprehensively consider the completion time, system load, and cost to make a better resource allocation decision.

[0155] The supervision system of the present disclosure adopts a distributed architecture, enabling flexible addition or adjustment of edge computing nodes according to actual needs. This design facilitates future system upgrades and function expansions and can adapt to distribution line construction projects of different scales and types. Through the combination of edge computing and cloud storage, the supervision system not only ensures local processing of sensitive data but also realizes cloud backup of important data. This design improves data processing efficiency while enhancing data security and reliability.

[0156] The present disclosure collects data from existing systems through the OPC UA protocol and uses the MQTT protocol to implement communication between edge nodes and the cloud center. This multi-protocol combination ensures the compatibility of the system with existing devices and provides efficient real-time data transmission capabilities.

[0157] Through the description of the above embodiments, those skilled in the art can clearly understand that the supervision system of the present disclosure can be implemented by means of software plus necessary general-purpose hardware, and of course, it can also be implemented by dedicated hardware including application-specific integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. Generally, functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structures for implementing the same function can also be diverse, such as analog circuits, digital circuits, or dedicated circuits. However, in more cases for the present disclosure, software program implementation is a better implementation method.

[0158] Although the embodiments of the present disclosure have been described above in conjunction with the accompanying drawings, the present disclosure is not limited to the above specific embodiments and application fields. The above specific embodiments are merely illustrative and guiding, rather than restrictive. Those of ordinary skill in the art can also make many forms under the inspiration of this specification and without departing from the scope protected by the claims of the present disclosure, and all of these fall within the scope of protection of the present disclosure.

Claims

1. A construction supervision system, characterized in that: The construction supervision system adopts a distributed architecture including a cloud center and multiple edge computing nodes. There is data communication between the cloud center and the edge computing nodes. One edge computing node has data communication with one or more existing systems at construction sites. The existing systems include a monitoring system and a scheduling system; The cloud center is configured to deploy an image detection model and perform image detection model training; and / or is configured to deploy a task scheduling algorithm to schedule multiple construction tasks; The edge computing node is configured to deploy the trained image detection model to perform behavior recognition on the monitoring video of the construction task.

2. The construction supervision system according to claim 1, characterized in that: The cloud center is provided with a cloud server, and the cloud server is configured to store the monitoring video of the construction task or other videos.

3. The construction supervision system according to claim 2, characterized in that: The monitoring video or other videos are stored in the Object Storage Service (OSS) manner.

4. The construction supervision system according to claim 1, characterized in that: The cloud center and the edge computing nodes communicate using the MQTT protocol.

5. The construction supervision system according to claim 1, wherein: The image detection model is an improved YOLO algorithm model. The improved YOLO algorithm model introduces a Spatial Pyramid Pooling module into the original YOLO algorithm model and uses GIoU as the loss function.

6. The construction supervision system according to claim 1, wherein The goal of the task scheduling algorithm is to minimize the function where: In the formula: the subscript i is the task identifier, the subscript min represents the minimum value, the subscript max represents the maximum value, TCC is the total task completion cost, LB is the resource load balance degree, and TCT is the total task completion time.

7. The construction supervision system according to claim 1, characterized in that, The task scheduling algorithm adopts an improved beluga algorithm. The improvements of the improved beluga algorithm compared to the original beluga algorithm include introducing a simulated annealing algorithm for local position update and adopting a quasi-opposite learning strategy for global position update; The local position update includes: (1) When the optimized objective function value of the potential new position is lower than the current position, let the current beluga move to the potential new position in the next step to complete the position update; (2) When the optimized objective function value of the potential new position is not lower than the current position, calculate the acceptance probability p of the worse position to guide the next move of the current beluga: Where: Obj T is the objective function value at the current position of the beluga whale, Obj T+1 is the objective function value at the new position, T is the current iteration number, C is the cooling rate of the simulated annealing parameter, and Temperature is the initial temperature of the simulated annealing; The global position update includes: denoting the number of beluga whales as N, the dimensionality of the search space as d, and the position of the i-th beluga whale in the d-th dimensional space as respectively represent the lower and upper bounds of, define the corresponding reverse solution is Then the reverse solution under the quasi-reverse learning strategy is: In the formula: rand(0, 1) is a random number between (0, 1), i = 1, 2,..., N, j = 1, 2,..., d.

8. The construction supervision system according to claim 1, characterized in that, The edge computing node collects data from the existing system through the OPC UA protocol.

9. The construction supervision system according to claim 1, wherein The edge computing node obtains data from the device through the device's own protocol and binds the device's data information source to the device object information model. When the device's data information source changes, the device object information model is updated with data.

10. The construction supervision system according to claim 1, characterized in that, The edge computing node is configured to send a video analysis request to the cloud center when the computing power does not meet the set threshold and obtain the behavior recognition result of the monitoring video of the construction task from the cloud center; The cloud center is configured to respond to the video analysis request of the edge computing node, send and store the behavior recognition result of the monitoring video of the construction task to the edge computing node.

Images