An automated production line monitoring control management method and system

Abstract

The application discloses an automatic production line monitoring control management method and system, and particularly relates to the field of production line monitoring, which comprises calculating the flow distribution and the resource gap value of each edge node in a preset time period through the topology structure and historical data of the target production line, and then dynamically adjusting the resource allocation of each edge node and continuously optimizing the cycle. The automatic production line monitoring control management method and system predict the flow distribution change, realize the dynamic scheduling of resources from high-load nodes to low-load nodes, effectively alleviate the local response delay caused by uneven load, and reduce the fluctuation standard deviation of resource utilization; the double-layer planning algorithm is used to coordinate the resource gap calculation, which significantly improves the problem of excessive fluctuation of resource utilization; the global model parameters are dynamically corrected to ensure that the scheduling strategy evolves adaptively with the change of topology, and the stability in complex production scenarios is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of production line monitoring, and more particularly, to an automatic production line monitoring control management method and system. BACKGROUND

[0002] With the deep application of industrial automation technology, efficient monitoring and precise control of the production line are particularly important. Traditional automatic production line monitoring control systems mainly adopt a centralized architecture, which is composed of sensors distributed at each station, programmable logic controllers and a central monitoring server. The device state data is collected by distributed sensors, preliminarily processed by programmable logic controllers, and then uploaded to the central server for unified analysis and decision-making. However, this traditional architecture exposes technical bottlenecks such as high data processing delay, delayed response to abnormal working conditions, and insufficient multi-dimensional production data correlation analysis capability when facing multi-variety and small-batch production scenarios, making it difficult to meet the needs of dynamic adjustment and fine control in complex production processes.

[0003] To break through the performance limitations of the traditional architecture, a monitoring system based on distributed edge nodes is proposed. By deploying edge computing nodes at key stations of the production line, real-time sensor data is pre-processed and controlled locally, and only key feature data is uploaded to the central platform, while receiving global scheduling instructions from the central platform.

[0004] However, it still has some shortcomings in actual use, such as when the production line is reorganized, causing changes in data flow distribution, some edge nodes may have a steep increase in load, causing processing queue accumulation, while idle node resources cannot be effectively scheduled, ultimately causing local response delay. The resource scheduling lacks a mechanism to dynamically perceive the changes in production line topology, and there is a lack of fine-grained resource coordination capability between edge nodes, resulting in large fluctuations in overall resource utilization, which restricts the stability guarantee in complex production scenarios. SUMMARY

[0005] To overcome the above-mentioned defects of the prior art, the present application provides an automatic production line monitoring control management method and system, which solves the problems raised in the background art through the following solutions.

[0006] To achieve the above-mentioned purposes, the present application provides the following technical solutions:

[0007] An automatic production line monitoring control management method, comprising:

[0008] S1: Real-time capture of the physical connection state and position coordinate information of each device in the target production line, and generation of topological vector data representing the topological structure of the target production line;

[0009] S2: calculating a traffic distribution in a preset time period based on the topological vector data and historical data by using a prediction model;

[0010] S3: calculating a resource gap value of each edge node in the preset time period according to the traffic distribution and a target edge node load state, the resource gap value including a computing resource demand and a network bandwidth demand;

[0011] S4: dynamically adjusting a resource instance configuration allocated to each edge node according to the resource gap value;

[0012] S5: generating a task allocation scheme by using a decision model, and performing task scheduling based on the task allocation scheme;

[0013] S6: performing a continuous optimization cycle, the cycle at least including:

[0014] monitoring resource utilization of each edge node, reclaiming idle resources and using the idle resources to update calculation of the resource gap value;

[0015] updating the prediction model and the decision model based on model parameter synchronization between each edge node.

[0016] Preferably, the S1, the topological vector data includes:

[0017] real-time scanning of device space coordinates by a sensor pre-deployed on the production pipeline, and constructing a three-dimensional topological model based on a point cloud registration algorithm;

[0018] and calculating a device displacement amount by using a sliding window mechanism, and when a displacement variance of continuous N sampling periods exceeds a preset threshold, triggering a topological update event and generating an incremental topological change vector, wherein N is a positive integer dynamically adjusted according to a production beat, and the incremental topological change vector is used to update the topological vector data in real time.

[0019] Preferably, the S2, the prediction model is a spatio-temporal neural network;

[0020] the prediction model receives device connection relationship and spatial position information in the topological vector data, and outputs a traffic distribution with a confidence evaluation value;

[0021] the confidence evaluation value is used to dynamically adjust a weight of calculation of the resource gap value.

[0022] Preferably, the S3, the resource gap value of each edge node in the preset time period is calculated by using a bi-level programming algorithm, wherein an upper-level programming determines a computing resource demand ratio, and a lower-level programming solves a network bandwidth constraint condition, so as to obtain a multi-objective optimization solution of coordinating computing and network resources as the resource gap value.

[0023] Preferably, the S5, performing task scheduling, comprises:

[0024] The input information of the decision model is configured as the topology vector data updated in real time, the predicted traffic distribution, and the real-time load state and available resource information of each edge node;

[0025] Based on the input information, the task allocation scheme is generated by optimizing one or more objective functions, and the objective functions are used to jointly optimize at least two of the following: minimizing the end-to-end processing delay of tasks, maximizing the load balancing degree between edge nodes, and maximizing the overall resource utilization of the system;

[0026] The task allocation scheme is used to specify the allocation and migration path of tasks between edge nodes, so as to actively guide the computing load from high-load nodes to low-load nodes.

[0027] Preferably, the S5, performing task scheduling, further comprises:

[0028] Deploying a task processing queue with a multi-dimensional priority label at the target edge node, the multi-dimensional priority label being determined based on at least the task urgency and the migration efficiency;

[0029] And when it is detected that the task processing queue depth exceeds a threshold, preferentially migrating a specific task to other nodes according to the multi-dimensional priority label.

[0030] To achieve the above object, the present application provides the following technical scheme: an automatic production line monitoring control management system, comprising a system running database, a system central processing module and a user information terminal, implementing the above-mentioned automatic production line monitoring control management method, comprising:

[0031] A dynamic topology structure acquisition module: capturing the physical connection state and position coordinate information of each device in the target production line in real time, and generating topology vector data representing the topology structure of the target production line;

[0032] A traffic distribution prediction module: based on the topology vector data and historical data, a prediction model is used to calculate the traffic distribution in a preset time period;

[0033] A resource demand evaluation module: according to the traffic distribution and the load state of the target edge node, the resource gap value of each edge node in the preset time period is calculated, and the resource gap value includes the calculation resource demand and the network bandwidth demand;

[0034] A cooperative scheduling module: according to the resource gap value, the resource instance configuration allocated to each edge node is dynamically adjusted;

[0035] Task dynamic scheduling execution module: generating a task allocation scheme by using the decision model, and executing task scheduling based on the task allocation scheme;

[0036] Closed loop optimization module: performing a continuous optimization loop, the loop comprising at least:

[0037] Monitoring resource utilization of each edge node, reclaiming idle resources and using them to update the calculation of the resource gap value;

[0038] Updating the prediction model and the decision model based on the synchronization of model parameters between edge nodes;

[0039] The system operation database is all data texts of an automatic production line monitoring control management system, and real-time collection of information texts output by each module, the system central processing module is used for the information text instructions output by each module in the central control system, and the user information end is an information output device receiving an automatic production line monitoring control management system.

[0040] Preferably, the closed loop optimization module, based on the synchronization of model parameters between edge nodes, updates the prediction model and the decision model comprises:

[0041] Building a federated learning architecture between edge nodes, based on the scheduling decisions or local data training results of each edge node, using a model contribution-based parameter aggregation strategy to update the global model parameters.

[0042] The technical effects and advantages of the present application are:

[0043] 1. The present application generates an incremental topology change vector by capturing the physical connection state and displacement change of the equipment in real time, and accurately predicts the traffic distribution change by processing the equipment topology relationship, realizes the dynamic scheduling of resources from high-load nodes to low-load nodes, effectively alleviates the local response delay caused by uneven load, and reduces the fluctuation standard deviation of resource utilization rate;

[0044] 2. The present application coordinates multi-objective resource gap calculation through a double-layer planning algorithm, realizes fine-grained optimization of task migration path, and significantly improves the problem of too large resource utilization rate fluctuation;

[0045] 3. The present application dynamically corrects the global model parameters to ensure that the scheduling strategy evolves adaptively with the topology change, and improves the stability in complex production scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 An implementation flowchart of an automatic production line monitoring control management method according to an embodiment of the present application is provided.

[0047] Figure 2A system structure block diagram of an automatic production line monitoring control management system is provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort fall within the protection scope of the present application.

[0049] The terms used in the following embodiments of the present application are only for the purpose of describing the specific embodiments, and are not intended to be limiting on the present application. As used in the specification of the present application, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "and / or" used in the present application, means and includes any or all possible combinations of one or more listed items; in the description of the embodiments of the present application, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0050] As shown in the accompanying drawings Figure 1 An automatic production line monitoring control management method, first calculates the traffic distribution and resource gap value of each edge node in a preset time period through the topology structure and historical data of the target production line, and then dynamically adjusts the resource configuration allocated to each edge node and performs continuous optimization cycle. Specifically, the following steps are included:

[0051] S1: Real-time capture of the physical connection state and position coordinate information of each device in the target production line, generate topology vector data representing the topology structure of the target line;

[0052] S2: Based on the topology vector data and historical data, calculate the traffic distribution in a preset time period using a prediction model;

[0053] S3: According to the traffic distribution and target edge node load state, calculate the resource gap value of each edge node in the preset time period, the resource gap value includes computing resource demand and network bandwidth demand;

[0054] S4: According to the resource gap value, dynamically adjust the resource instance configuration allocated to each edge node;

[0055] S5: Generate a task allocation scheme using a decision model, and execute task scheduling based on the task allocation scheme;

[0056] S6: performing a continuous optimization loop, the loop comprising at least:

[0057] monitoring resource utilization of each edge node, reclaiming idle resources and using them to update the calculation of the resource gap value;

[0058] updating the prediction model and the decision model based on synchronization of model parameters between edge nodes.

[0059] Specifically, in S1, real-time scanning of device spatial coordinates is performed by sensors pre-deployed on the production line, a three-dimensional topology model is constructed based on a point cloud registration algorithm, and a sliding window mechanism is used to calculate device displacement. When the displacement variance of the last N sampling periods exceeds a preset threshold, a topology update event is triggered and an incremental topology change vector is generated, which is used to update the topology vector data in real time.

[0060] In this embodiment, the key positions include but are not limited to movable recombination robot end, fixed monitoring point, etc. Real-time scanning of the spatial coordinates of each key device on the production line is performed by pre-deploying millimeter wave radar arrays, UWB positioning tags, base stations or other high-precision sensors at key positions on the production line. The key devices include but are not limited to robots, conveyor controllers, vision detection units, etc. At the same time, through industrial standard communication protocols, connection is established with programmable logic controllers or host computers of key devices on the production line to collect structured data packets containing device unique identifiers, real-time three-dimensional coordinates, and network interface connection port states in real time.

[0061] In order to eliminate instantaneous jumps and high-frequency noise caused by electromagnetic interference or physical shielding in the industrial field of the multiple embodiments, a sliding average filtering algorithm is used to process the original coordinate data. In order to adapt to the change of device moving speed under different running states of the production line, the size of the filtering window can be adaptively adjusted according to the historical moving speed or current acceleration of the device. If it is detected that the device is in a high-speed moving state, the filtering window is appropriately reduced to ensure real-time performance. If the device is relatively stationary or moving slowly, the filtering window is increased to obtain smoother and more accurate coordinate estimates.

[0062] The construction process of the three-dimensional topological model can be specifically decomposed as follows: performing a voxel grid filtering operation on the device coordinate information after cleaning and filtering, setting the voxel size to regularize the data and reduce the computational complexity; using the RANSAC algorithm to remove background interference point clouds such as fixed or slowly moving conveyors; achieving preliminary spatial alignment of the RGB-D depth point cloud and the millimeter wave radar point cloud through the ICP algorithm; extracting key geometric features of the device surface, including detecting stable three-dimensional feature corners using the Harris 3D algorithm, and calculating a fast point feature histogram to describe the local surface normal vector information of each point; if the device displacement variance calculated within the preset sliding time window exceeds the preset threshold, triggering an incremental topological update process, and the updated result is encapsulated into a structured incremental topological change vector.

[0063] It should be noted that the embodiment also includes a dynamic adjustment mechanism, which dynamically calculates the size of the point cloud sampling and analysis window based on the production line beat cycle obtained in real time from the PLC and the like. The embodiment calculates the value of 200 milliseconds divided by the production line beat cycle by using the ceiling function to ensure that the window covers at least two production cycles to capture the complete dynamics.

[0064] Specifically, in S2, the prediction model is a spatio-temporal neural network; the prediction model receives the device connection relationship and spatial position information in the topological vector data, and outputs a traffic distribution with a confidence evaluation value; the confidence evaluation value is used to dynamically adjust the weight of the calculation of the resource gap value.

[0065] It should be noted that the spatio-temporal neural network constructs a graph convolution network to process the topological vector data between devices, and further takes each key device on the production line as a node of the graph, and the physical connection between devices as an edge, to establish an adjacency matrix, and the feature vector of the node of the adjacency matrix contains the topological vector data and the traffic mean value reflecting the historical load situation, while the weight w of the edge is determined by the actual bandwidth BW of the connection link and the physical distance between devices, and is specifically represented as: ij ij

[0066]

[0067] wherein, a represents a distance influence factor, p i and p j ​​The physical distance between devices i and j is represented; further, the physical connection relationship between devices is converted into a graph embedding vector of each device, and a global topology feature vector is generated through a graph pooling operation. For spatial position coordinate information, a three-dimensional sine position encoding scheme is adopted to map the (x, y, z) coordinates of each device into a high-dimensional feature vector, so as to calculate the confidence evaluation value through the structure of the double output head, wherein the main output head is responsible for predicting the future traffic distribution of each edge node, and the auxiliary output head calculates the confidence value in parallel. The calculation of the confidence evaluation value includes the cosine similarity between the current prediction result and the historical traffic pattern, the uncertainty degree of the internal feature representation of the model estimated by the Monte Carlo sampling method, and the logical check result of the integrity of the input data.

[0068] The weight coefficient is adjusted based on the calculated confidence. When the confidence is in the interval [0.9, 1.0], the weight coefficient is set to 1.2, indicating that the prediction result is highly trusted and can be directly used for aggressive resource pre-allocation. When the confidence is in the interval [0.7, 0.9), the weight is 1, and the resource is allocated according to the normal process while starting the standby resource monitoring. If the confidence decreases to the interval [0.5, 0.7), the weight coefficient is reduced to 0.8, the resource allocation amount is correspondingly reduced, and the manual review process is triggered. When the confidence is less than 0.5, the weight coefficient is further reduced to 0.5, and the resource allocation mode based on the historical average is switched to.

[0069] Specifically, in S3, the resource gap value of each edge node in the preset time period is calculated using a bi-level programming algorithm, wherein the upper-level programming determines the calculation resource demand ratio, and the lower-level programming solves the network bandwidth constraint condition to obtain a multi-objective optimization solution for coordinating calculation and network resources as the resource gap value.

[0070] In this embodiment, the upper-level programming target of the bi-level programming algorithm is to minimize the overall calculation resource gap, i.e., the sum of CPU demand and memory demand, and a constraint condition is imposed to specify that the allocation ratio of CPU and memory resources on a single edge node must be maintained within a preset range, which is set to 0.5≤CPU: memory≤2.0. In this embodiment, the Lagrange relaxation algorithm is used to relax the coupling constraints between nodes and update the Lagrange multiplier through iteration until the convergence condition is reached, thereby obtaining the optimal solution of the calculation resource demand. Further, the lower-level programming target is to optimize the network resources. In this embodiment, the target of the lower-level programming is set to maximize the bandwidth utilization of the entire network, while the constraint conditions include that the upper limit of the bandwidth that can be carried by each physical link in the network must not be broken, and the end-to-end transmission delay of key business data flow must be strictly controlled within a pre-set threshold.

[0071] It should be noted that the traffic distribution data generated in S2 is used for logical region division and mapping the divided regions to corresponding subnets in the network topology; the traffic peak of each subnet is used as the bandwidth demand constraint of the lower-level planning; at the same time, the historical load data of the edge node is used to initialize the resource allocation baseline of the upper-level planning.

[0072] Specifically, in S4, a resource pooling engine is constructed based on the resource gap value to allow high-load nodes to borrow the idle computing units of low-load nodes through a secure channel to dynamically adjust the resource configuration of the containerized instance.

[0073] Further, the containerization technology is used to abstract the computing resources and storage resources on each edge node into schedulable virtual units, and mark their load states as idle or occupied; based on the resource gap value of each edge node generated in S3, the type and scale of the resources to be allocated are calculated.

[0074] Further, a resource adjustment protocol process is designed and executed, when an edge node generates resource demand due to load, a resource request is sent to the resource pool through the message queue protocol, and the required resource type, bandwidth, quantity and tolerable delay are specified; at the same time, the low-load node reports the available idle resource information to the resource pooling engine, which combines the topology vector data generated in S1 to filter out candidate resource provider nodes that meet the delay constraint.

[0075] Specifically, in S5, the input information of the decision model is configured as the topology vector data updated in real time, the predicted traffic distribution, and the real-time load state and available resource information of each edge node; in this embodiment, based on the physical connection state and position coordinate information of each device in S1, combined with the network link bandwidth parameter, the delay value between each node is generated to form a global delay matrix for scheduling decision; the traffic prediction result output by S2 is used to obtain the task type distribution, data size and time distribution characteristics in a preset time period; the CPU utilization rate, memory occupancy rate, network bandwidth margin and container instance health state are collected in real time by the lightweight agent deployed on the edge node, and are packaged and reported in JSON format; the data is fused in real time through a unified data bus, the data update frequency is set to one second, and three partitions are divided, which correspond to topology data, traffic prediction and node state respectively, and are pushed to the decision model to ensure low delay and high consistency of the input information.

[0076] Furthermore, based on the input information, the task allocation scheme is generated by optimizing one or more objective functions. The objective functions are used to jointly optimize at least two of the following: minimizing the end-to-end processing latency of tasks, maximizing the load balancing between edge nodes, and maximizing the overall resource utilization of the system. The task allocation scheme is used to specify the allocation and migration paths of tasks among edge nodes, so as to actively redirect the computational load from high-load nodes to low-load nodes. In this embodiment, the initial task allocation scheme is solved by mixed integer programming, with constraints including the maximum number of hops between nodes ≤ 3 and the upper limit of resource capacity of a single node. The NSGA-II multi-objective genetic algorithm is used to iteratively optimize the coarse-grained scheme, using the task migration path as the gene encoding, generating new solutions through crossover and mutation operations, and screening the optimal solution based on the Pareto front, focusing on adjusting the task migration path of high-load nodes to reduce local hotspots.

[0077] Furthermore, a task processing queue with multi-dimensional priority labels is deployed at the target edge node. The multi-dimensional priority labels are determined at least based on task urgency and migration efficiency. When the depth of the task processing queue exceeds a threshold, specific tasks are preferentially migrated to other nodes according to the multi-dimensional priority labels. In this embodiment, each task is assigned a multi-dimensional priority label when entering the edge node, including urgency dynamically calculated based on the task deadline, migration efficiency inversely proportional to migration cost, and correlation measuring the contextual dependency strength between the task and the current node.

[0078] As attached Figure 2 The automated production line monitoring and control management system shown includes a system operation database, a system central processing module, and a user information terminal, and also includes:

[0079] Dynamic topology acquisition module: captures the physical connection status and position coordinate information of each device in the target production line in real time, and generates topology vector data that characterizes the topology of the target production line;

[0080] Traffic distribution prediction module: Based on the topology vector data and historical data, it uses a prediction model to calculate the traffic distribution within a preset time period;

[0081] Resource demand assessment module: Based on the traffic distribution and the load status of the target edge nodes, calculate the resource gap value of each edge node within the preset time period. The resource gap value includes the calculated resource demand and the network bandwidth demand.

[0082] Collaborative scheduling module: Based on the resource gap value, dynamically adjust the resource instance configuration allocated to each edge node;

[0083] A task dynamic scheduling execution module generates a task allocation scheme using the decision model and performs task scheduling based on the task allocation scheme;

[0084] A closed-loop optimization module performs a continuous optimization loop, which at least includes:

[0085] Monitoring the resource utilization of each edge node, recycling idle resources and updating the calculation of the resource gap value;

[0086] Based on the synchronization of model parameters between each edge node, the prediction model and the decision model are updated;

[0087] In this embodiment, the resource dynamic optimization loop includes monitoring the resource gap value and triggering the idle resource recycling mechanism, recycling resources feedback to the resource demand assessment module; a federated learning architecture is constructed between each edge node, and based on the scheduling decision or local data training result of each edge node, the global model parameters are updated using a model contribution-based parameter aggregation strategy.

[0088] It should be noted that the topology vector data of the dynamic topology structure acquisition module drives the traffic prediction of the traffic distribution prediction module, the resource gap evaluation of the resource demand assessment module depends on the prediction result of the traffic distribution prediction module, the resource allocation scheme of the collaborative scheduling module is controlled by the evaluation parameters of the resource demand assessment module, the scheduling decision of the task dynamic scheduling execution module synchronously receives the resource network state of the collaborative scheduling module and the prediction features of the traffic distribution prediction module, the optimization process of the closed-loop optimization module real-time corrects the model parameters of the traffic distribution prediction module and the task dynamic scheduling execution module, and each step forms a multi-layer closed-loop coupling architecture through data flow and control flow;

[0089] The system operation database is all data texts of an automatic production line monitoring control management system, and real-time collection of information texts output by each module, the system central processing module is used for information text instructions output by each module in the central control system, and the user information end is an information output device receiving an automatic production line monitoring control management system.

[0090] It should be noted that the system central processor is the core operation and control unit of the whole automatic production line monitoring control management system;Including one or more processing cores, through the use of various interfaces and lines, connecting each part in the whole system;By running or executing instructions, programs, code sets or instruction sets stored in the system running database, and being able to call the data stored therein, so as to execute various functions of an automatic production line monitoring control management system, including topology dynamic analysis algorithm, resource gap prediction model and low delay task migration instruction set and other operations, to ensure that the virtualization resource topology is dynamically reconstructed based on the edge node load state, the encryption tunnel establishment and removal of cross node task migration are implemented, and the real-time synchronization and reverse control of digital twin are triggered.

[0091] Among them, the communication bus 302 is used to realize the connection communication between components.

[0092] Among them, the system running database 303 is used to save a large amount of data related to an automatic production line monitoring control management, including mathematical model related data based on spatio-temporal neural network, deep reinforcement learning decision model and digital twin, etc. A large amount of historical running data is stored, including differential compression feature dictionary of task migration, mapping relationship of each priority label and resource allocation weight, and edge node log;The system central processor will frequently call these data from the system running database when executing various functions, so as to realize accurate control and efficient management of an automatic production line monitoring control management.

[0093] Among them, the user information end connects the display screen and the camera and other external devices through the standard wired interface or wireless interface to provide the interface for the user to interact with the system.

[0094] Secondly: the drawings of the disclosed embodiments only involve the structures involved in the disclosed embodiments, other structures can refer to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the present application can be combined with each other;

[0095] Finally: the above only describes the preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An automated production line monitoring control management method, characterized by, Comprise: S1: Real-time capture of the physical connection state and position coordinate information of each device in the target production pipeline, generate topological vector data representing the topological structure of the target pipeline; S2: Based on the topological vector data and historical data, use the prediction model to calculate the traffic distribution in the preset time period; S3: According to the traffic distribution and the load state of the target edge node, calculate the resource gap value of each edge node in the preset time period, the resource gap value includes the calculation resource demand and the network bandwidth demand; S4: According to the resource gap value, dynamically adjust the resource instance configuration allocated to each edge node; S5: Use the decision model to generate a task allocation scheme, and execute task scheduling based on the task allocation scheme; S6: Perform a continuous optimization cycle, the cycle at least includes: Monitoring the resource utilization of each edge node, recycling idle resources and used to update the calculation of the resource gap value; Based on the model parameter synchronization between each edge node, update the prediction model and the decision model.

2. The method of claim 1, wherein: The S1, the topological vector data includes: Real-time scanning of device space coordinates through sensors deployed in advance on the production pipeline, constructing a three-dimensional topological model based on point cloud registration algorithm; And using a sliding window mechanism to calculate the displacement amount, when the displacement variance of the continuous N sampling periods exceeds the preset threshold, triggering a topological update event and generating an incremental topological change vector, where N is a positive integer dynamically adjusted according to the production rhythm, the incremental topological change vector is used to update the topological vector data in real time.

3. The method of claim 1, wherein: The S2, the prediction model is a spatio-temporal neural network; The prediction model receives the device connection relationship and spatial position information in the topological vector data, and outputs the traffic distribution with confidence evaluation value; The confidence evaluation value is used to dynamically adjust the weight of the calculation of the resource gap value.

4. The method of claim 1, wherein: The S3, the calculation of the resource gap value of each edge node in the preset time period uses a double-layer planning algorithm, where the upper layer planning determines the calculation resource demand ratio, and the lower layer planning solves the network bandwidth constraint condition to obtain a multi-objective optimization solution that coordinates calculation and network resources as the resource gap value.

5. The method of claim 1, wherein: The S5, executing task scheduling includes: The input information of the decision model is configured as the topological vector data updated in real time, the predicted traffic distribution, and the real-time load state and available resource information of each edge node; Based on the input information, generate the task allocation scheme by optimizing one or more objective functions, which are used to jointly optimize at least two of the following: minimize the end-to-end processing delay of tasks, maximize the load balancing degree between edge nodes, and maximize the overall resource utilization of the system; The task allocation scheme is used to specify the allocation and migration path of tasks among edge nodes to actively direct the computing load from high-load nodes to low-load nodes.

6. The method of claim 5, wherein: The S5, executing task scheduling also includes: Deploy a task processing queue with a multi-dimensional priority label on the target edge node, the multi-dimensional priority label is determined based on at least task urgency and migration efficiency; and when detecting that the task processing queue depth exceeds a threshold, preferentially migrating a specific task to other nodes according to the multi-dimensional priority label.

7. An automatic production line monitoring control management system, comprising a system operation database, a system central processing module and a user information terminal, according to any one of the automatic production line monitoring control management method of claims 1-6, characterized in that, Also comprising: a dynamic topology acquisition module: capturing the physical connection state and position coordinate information of each device in the target production pipeline in real time, generating topology vector data representing the topology of the target pipeline; a flow distribution prediction module: based on the topology vector data and historical data, using a prediction model to calculate the flow distribution in a preset time period; a resource demand evaluation module: according to the flow distribution and the load state of the target edge node, calculating the resource gap value of each edge node in the preset time period, the resource gap value including the calculation resource demand and network bandwidth demand; a collaborative scheduling module: according to the resource gap value, dynamically adjusting the resource instance configuration allocated to each edge node; a task dynamic scheduling execution module: using a decision model to generate a task allocation scheme, and executing task scheduling based on the task allocation scheme; a closed-loop optimization module: performing a continuous optimization cycle, the cycle at least including: monitoring the resource utilization of each edge node, recycling idle resources and using them to update the calculation of the resource gap value; updating the prediction model and the decision model based on the model parameter synchronization between each edge node; The system operation database is all data texts of an automatic production pipeline monitoring control management system, and real-time collection of information texts output by each module, the system central processing module is used for controlling the information text instructions output by each module in the system, and the user information end is an information output device receiving an automatic production pipeline monitoring control management system.

8. The automated production line monitoring control management system of claim 7, wherein: The closed-loop optimization module, based on the model parameter synchronization between each edge node, updates the prediction model and the decision model, including: constructing a federated learning architecture between each edge node, based on the scheduling decision or local data training result of each edge node, using a model contribution-based parameter aggregation strategy to update the global model parameters.

Images