Self-adaptive scheduling industrial internet data acquisition control system and strategy

Abstract

The invention discloses a self-adaptive scheduling industrial internet data acquisition control system and strategy, and relates to the technical field of industrial internet data acquisition and self-adaptive scheduling control, and the system comprises a data sensing layer, a deployment multi-type intelligent sensor group, an edge gateway and a hybrid communication module. The sensor supports multi-protocol adaptive access and dynamically switches sampling frequency, the edge gateway preprocesses original data, and the communication module guarantees stable transmission through wired and wireless hybrid networking; the self-adaptive scheduling layer comprises a task grading unit, a load monitoring unit and a dynamic scheduling unit; the edge calculation layer is used for processing tasks with high real-time performance; the cloud collaboration layer stores data and mines, analyzes and constructs an equipment model; the security protection layer realizes identity verification, data encryption and intrusion detection; and the system management layer manages the full life cycle of the equipment and divides user permissions. According to the system, the problem of traditional industrial internet data acquisition scheduling pain points is solved, and the overall operation efficiency and safety of the system are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial internet data acquisition and adaptive scheduling control, and particularly relates to an adaptive scheduling industrial internet data acquisition control system and strategy. BACKGROUND

[0002] As the core carrier of industrial digital transformation, the industrial internet needs to realize real-time acquisition and accurate control of multiple types of data such as temperature, pressure, and vibration in the production field. However, the existing data acquisition system has obvious technical limitations. Traditional industrial data acquisition relies on fixed sampling frequency sensors, which cannot dynamically adjust according to the device running state. When the device is under high load, high-frequency data acquisition is needed to capture abnormal fluctuations, but the fixed sampling frequency leads to data lag. When the device is under low load, high-frequency sampling is still maintained, resulting in a large amount of redundant data occupying storage and bandwidth resources. At the same time, the sensor lacks self-calibration and multi-source fusion capabilities, is affected by environmental temperature and vibration interference, has large measurement deviation, and needs to be replaced manually after failure, resulting in data interruption. The communication network mostly uses single wired or wireless methods. Wired networks are difficult to lay in complex industrial sites, and wireless networks are easily affected by electromagnetic interference, leading to transmission interruption, making it difficult to balance the stability and flexibility of data transmission.

[0003] The lack of resource scheduling and edge-cloud collaboration capabilities is a key bottleneck restricting the efficiency of the industrial internet. Traditional systems use fixed task allocation strategies, uniformly allocating acquisition tasks to the cloud or edge nodes, without considering task priority and system real-time load. Emergency tasks (such as device fault warning data processing) may be delayed due to resource occupation by low-priority tasks, and edge nodes cannot migrate tasks to adjacent nodes when the load is too high, resulting in decreased computing efficiency. The cloud only stores data and performs simple analysis, without combining edge node local optimization results to generate a global scheduling scheme, making it difficult to achieve optimal resource allocation for the entire industrial internet. In addition, the task processing lacks a dynamic splitting mechanism, and complex tasks cannot be decomposed into sub-tasks for parallel processing according to data correlation and computational complexity, further reducing system response speed and failing to meet the high real-time requirements of industrial sites.

[0004] The real-time guarantee, security protection and model adaptation capability of the industrial internet also have short boards. In the process of data transmission from sensors to edge nodes and then to the cloud, the transmission delay accumulates with the increase of nodes, and the existing system lacks effective delay compensation mechanism, so that the control command cannot be issued in time due to delay, which may cause abnormal expansion of equipment. On the security protection level, single password verification is often used for identity authentication, which is easy to be cracked illegally; the security of data transmission and storage encryption algorithm is insufficient, and there is lack of real-time detection capability for network attacks, so the risk of industrial data leakage and illegal access to equipment is high. At the same time, the models such as equipment fault early warning are developed for specific production scenes, and a large amount of data needs to be collected for model training when the models are applied across scenes, so the adaptation period is long; the model update depends on manual triggering and cannot be automatically optimized based on new data, so it is difficult to adapt to the dynamic change of the operation environment in the industrial field, which restricts the long-term stable operation of the industrial internet. SUMMARY

[0005] The adaptive scheduling industrial internet data acquisition control system and strategy provided by the application solve the problems mentioned in the prior art.

[0006] In order to achieve the above purpose, the application adopts the following technical scheme: an adaptive scheduling industrial internet data acquisition control system, comprising:

[0007] A data perception layer is deployed in each production unit in the industrial field and includes multiple types of intelligent sensor groups, edge gateways and local communication modules. The intelligent sensor groups cover temperature, pressure, flow and vibration sensors. The edge gateway integrates a data preprocessing chip and can perform preliminary cleaning and format conversion on raw data. The local communication module uses a hybrid networking of industrial wireless and wired networking. An adaptive scheduling layer includes a task priority division unit, a resource load monitoring unit and a dynamic scheduling decision unit. The task priority division unit divides the acquisition tasks into four levels of emergency, high, medium and low according to relevant factors. The resource load monitoring unit collects relevant data of edge nodes and cloud servers in real time. The dynamic scheduling decision unit generates an optimal scheduling scheme based on the load state and task priority using an improved genetic algorithm.

[0008] An edge computing layer is composed of edge nodes distributed in the industrial field. Each node is equipped with an industrial-grade processor and a local storage unit to perform computing and processing on tasks with high real-time requirements scheduled to the local. The edge computing layer also has a data caching function to cache critical data when the network is interrupted.

[0009] A cloud collaboration layer is deployed in a cloud platform and includes a big data storage center, a deep analysis engine and a global optimization unit. The big data storage center uses a distributed file system. The deep analysis engine uses machine learning algorithms to mine historical data. The global optimization unit combines the local optimization results of each edge node.

[0010] The security protection layer includes an identity authentication unit, a data encryption unit, and an intrusion detection unit. The identity authentication unit adopts a multi-factor authentication mechanism to verify the relevant access identity. The data encryption unit encrypts the transmission and storage data. The intrusion detection unit monitors in real time based on relevant technologies, triggers an alarm and blocks the connection when an anomaly is found.

[0011] The system management layer includes a device management unit, a user permission unit, and a running monitoring unit. The device management unit implements the full life cycle management of the relevant devices. The user permission unit divides the management permissions according to the roles. The running monitoring unit displays the running state of each module of the system in real time.

[0012] Further, it also includes a dynamic resource allocation module, which includes a resource demand prediction unit and an allocation execution unit. The resource demand prediction unit analyzes the resource consumption law of historical tasks, combines the current task queue length, and outputs the resource demand curve in three time dimensions of 1 hour, 4 hours, and 12 hours in the future. The model training adopts the sliding window method. The allocation execution unit introduces the resource elastic expansion mechanism based on the prediction results and real-time load, and simultaneously adjusts the resource allocation rate of each task through Calculate the resource allocation rate of the i-th task at time t, where R i (t) is the resource allocation rate of the i-th task at time t, C i (s) is the resource demand of the i-th task at time s, P i (s) is the priority coefficient of the i-th task at time s, L(s) is the total load of the system at time s, t0 is the task start time, and t is the current time.

[0013] Further, it also includes a real-time compensation module, which includes a transmission delay monitoring unit and a compensation strategy generation unit. The transmission delay monitoring unit measures the transmission delay from the sensor to the edge node and from the edge node to the cloud in real time through timestamp comparison and network jitter analysis, and records the delay distribution law under different device types and network environments. The compensation strategy generation unit adopts a predictive compensation mechanism to fit the trend of historical delay data based on delay data, and simultaneously adjusts the transmission delay of each task through Calculate the delay compensation amount, where D comp is the total delay compensation amount, n is the number of transmission nodes, is the delay change rate of the k-th node, is the time compensation interval of the k-th node.

[0014] Further, the intelligent sensor group of the data perception layer also integrates a self-calibration unit and a multi-source fusion unit. The self-calibration unit automatically corrects measurement deviation according to environmental parameter changes. The temperature sensor uses a built-in temperature compensation curve. The vibration sensor uses an adaptive filtering algorithm to identify and filter environmental vibration interference in real time. The multi-source fusion unit uses a Kalman filter combined with a D-S evidence theory to fuse the data of multiple sensors at the same monitoring point. At the same time, the sensor group has a self-diagnosis function. By monitoring the power supply voltage, communication signal strength and measurement stability, and combining a fault identification model based on a BP neural network, it can determine whether there is a fault. When a fault is found, it actively sends an alarm message to the edge gateway and uploads the fault characteristic parameters. After receiving the alarm, the edge gateway automatically starts the backup sensor and records the historical data of the faulty sensor.

[0015] Further, the dynamic scheduling decision unit of the adaptive scheduling layer also introduces a reinforcement learning mechanism and a dynamic task splitting algorithm. The reinforcement learning mechanism takes system load, task completion time and resource utilization as reward function parameters. The decision model parameters are updated after completing each batch of tasks. Through continuous interaction with the environment, the scheduling strategy is optimized. The dynamic task splitting algorithm decomposes complex tasks into multiple subtasks according to data correlation and computational complexity, assigns them to different edge nodes using a parallel scheduling strategy, and uses a distributed message queue between subtasks. At the same time, the unit has a scheduling scheme simulation function.

[0016] Further, the edge nodes of the edge computing layer establish a collaborative communication mechanism and a load prediction sharing network. The collaborative communication mechanism uses a distributed hash table. When a certain edge node has a high load, it migrates part of the non-real-time tasks to a nearby node with a lower load, and only transmits the raw data and intermediate results that have not been processed. In the load prediction sharing network, each edge node broadcasts its load prediction results to neighboring nodes every 5 minutes. The receiving nodes adjust their local task queues based on their own prediction data. In addition, the edge nodes have heterogeneous computing capabilities. They automatically select CPU, GPU or FPGA for processing according to the task type. For image processing tasks, GPU acceleration is enabled, and for logical judgment tasks, FPGA acceleration is enabled, improving computing efficiency.

[0017] Further, the deep analysis engine of the cloud collaboration layer also includes a cross-scenario model migration unit and a federated learning framework. The cross-scenario model migration unit migrates a fault early warning model trained in a certain production scenario to a similar scenario by measuring the scenario difference degree using a maximum mean difference algorithm. The federated learning framework supports local model training by each edge node, and only uploads model parameter updates to the cloud. After aggregating the parameters, the cloud sends them to each node. The deep analysis engine supports model self-updating. After accumulating 5000 new data, it automatically re-trains the model. The global optimization unit is based on the local optimization results of each edge node.

[0018] Further, comprising the following steps:

[0019] Data sensing step, intelligent sensor group collects industrial field data such as temperature, pressure slowly varying parameters with low sampling frequency, for example, vibration, flow fast varying parameters with high sampling frequency, the collected data is preliminarily processed by edge gateway, and then transmitted to adaptive scheduling layer through hybrid communication network;

[0020] Task scheduling step, task priority division unit classifies collection tasks, resource load monitoring unit obtains system load state in real time, dynamic scheduling decision unit combines task priority and load state to generate task allocation scheme;

[0021] Edge processing step, after the edge node receives the allocated task, the data is feature extracted and abnormality identified, the abnormal data with high real-time requirement is directly generated control instruction and issued to field device, and the processing result and original data are cached;

[0022] Cloud analysis step, the cloud collaborative layer receives the data uploaded by the edge node, stores it to the big data storage center, the analysis engine mines the historical data, and the global optimization unit generates a global scheduling scheme and issues it to each edge node;

[0023] Security protection step, the identity authentication unit verifies the identity of all access devices, the data encryption unit encrypts the transmitted and stored data, and the intrusion detection unit monitors the network security state in real time, and triggers alarm and blocks connection when abnormality is found;

[0024] System management step, the device management unit manages all types of devices throughout the life cycle, the user permission unit controls the operation range of different roles, and the running monitoring unit displays the system running state in real time.

[0025] Further, the task scheduling step further includes load prediction sub-step and task rescheduling mechanism, the load prediction sub-step predicts the system load in future Δt time through Wherein L(t+Δt) is the system load in future Δt time, L(t) is the current system load, The load change rate is t, the current time, and Δt is the prediction time interval, and a sliding window mechanism is adopted; the task rescheduling mechanism is triggered when the deviation between the actual system load and the predicted load exceeds 15%, and the task priority and resource state are re-evaluated.

[0026] Further, the edge processing step further includes data compression sub-step and feature selection mechanism, the data compression sub-step adopts adaptive compression algorithm for non-critical data uploaded to the cloud through Calculating compression rate, wherein C is the data compression rate, S raw is the original data size, Scom To compress the data size after compression.

[0027] Compared with the prior art, the application has the beneficial effects that:

[0028] The application effectively solves the technical pain points of traditional industrial internet data collection and scheduling, significantly improving the accuracy and continuity of data collection. The intelligent sensor group of the data perception layer can dynamically adjust the sampling frequency according to the device load, avoiding the generation of redundant data and the lack of key data; the self-calibration unit combines environmental parameters to correct measurement deviation, the multi-source fusion unit integrates multi-sensor data to improve accuracy, and the self-diagnosis function automatically switches to standby equipment when the sensor fails, ensuring uninterrupted data collection; the hybrid communication network combines the advantages of wired and wireless, realizing stable data transmission and laying a high-quality foundation for subsequent data processing.

[0029] The system greatly improves the operation efficiency of the industrial internet through adaptive scheduling and resource optimization. The adaptive scheduling layer introduces a reinforcement learning mechanism and a dynamic task splitting algorithm, generates an optimal scheduling scheme based on task priority and system load, and the emergency task can get resources first, and the complex task is divided into sub-tasks for parallel processing; the dynamic resource allocation module realizes resource elasticity based on load prediction, avoiding resource waste and overload; the load prediction sub-step predicts system load changes in advance, and the task rescheduling mechanism adjusts the scheme smoothly when the load deviation exceeds the limit, reducing system fluctuations, ensuring efficient and stable scheduling process, and avoiding resource mismatch problems caused by traditional fixed scheduling.

[0030] The application also strengthens the edge-cloud collaboration capability, security protection and model adaptability, and comprehensively improves the practicality and security of the industrial internet. The edge computing layer can locally process tasks with high real-time requirements and quickly generate control instructions, the cloud collaboration layer integrates edge node results through a global optimization unit to optimize resources and production of the entire industrial internet; the cross-scene model migration unit shortens the model adaptation period, and the federated learning framework realizes model collaborative optimization while protecting data privacy; the security protection layer adopts multi-factor authentication, national encryption algorithm encryption and real-time intrusion detection to build a comprehensive security barrier; the system management layer realizes device lifecycle management and precise control of permissions to ensure system operation and maintenance specifications. Overall, the system realizes the whole process optimization of industrial internet data collection, scheduling, processing and security protection, improves the operation efficiency and security, and adapts to the dynamic needs of different industrial production scenes. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The schematic diagram of the adaptive scheduling industrial internet data collection control system proposed by the application is shown in the figure;

[0032] Figure 2A schematic block diagram of the strategy of the adaptive scheduling industrial internet data acquisition control system proposed in the present application;

[0033] Figure 3 A broken line graph of the relationship between task priority and resource allocation rate;

[0034] Figure 4 A scatter plot of system load and task completion delay;

[0035] Figure 5 A bar graph of the relationship between data importance and compression rate. DETAILED DESCRIPTION

[0036] 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 a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0037] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0038] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise explicitly and specifically limited. In addition, the terms "mounting", "connecting", "connection" should be broadly understood, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances, and the present application will be further described in detail with reference to the drawings.

[0039] Reference Figures 1 to 5 An adaptive scheduling industrial internet data acquisition control system, comprising:

[0040] The data sensing layer, deployed across various production units in the industrial site, comprises multiple types of intelligent sensor groups, edge gateways, and local communication modules. The intelligent sensor groups include temperature, pressure, flow, and vibration sensors, supporting multi-protocol adaptive access and automatically switching sampling frequencies based on equipment operating status, increasing to millisecond levels under high load and decreasing to second levels under low load. The edge gateway integrates a data preprocessing chip, capable of preliminary cleaning and format conversion of raw data. The local communication module employs a hybrid industrial wireless and wired network, with the wireless portion supporting LoRa and 5G dual-mode switching, and the wired portion based on industrial Ethernet for stable data transmission.

[0041] The adaptive scheduling layer includes a task priority division unit, a resource load monitoring unit, and a dynamic scheduling decision unit. The task priority division unit divides the data acquisition tasks into four levels: urgent, high, medium, and low, based on data real-time requirements, equipment importance, and production process relevance. The resource load monitoring unit collects real-time data on CPU utilization, memory usage, and network bandwidth consumption of edge nodes and cloud servers, generating a load status matrix. The dynamic scheduling decision unit uses an improved genetic algorithm to generate the optimal scheduling scheme based on the load status and task priority, achieving a reasonable allocation of tasks between the edge and the cloud.

[0042] The edge computing layer consists of edge nodes distributed across industrial sites. Each node is equipped with an industrial-grade processor and local storage unit. It can perform computational processing on tasks with high real-time requirements scheduled locally, including data feature extraction, outlier identification, and short-term trend prediction. The processing results can be directly used to generate control commands for field equipment. It also has a data caching function, which caches critical data when the network is interrupted and automatically synchronizes it to the cloud after the network is restored.

[0043] The cloud collaboration layer, deployed on a cloud platform, includes a big data storage center, a deep analysis engine, and a global optimization unit. The big data storage center uses a distributed file system to store historical data and processing results uploaded by edge nodes, supporting efficient retrieval of massive amounts of data. The deep analysis engine uses machine learning algorithms to mine historical data and build equipment fault early warning models and production optimization strategies. The global optimization unit combines the local optimization results of each edge node to generate a global scheduling scheme covering the entire industrial internet.

[0044] A security protection layer, which comprises an identity authentication unit, a data encryption unit and an intrusion detection unit, the identity authentication unit adopts a multi-factor authentication mechanism to verify the identity of the access of sensors, edge nodes and the cloud; the data encryption unit encrypts the data for transmission and storage, the transmission encryption adopts a national encryption algorithm, and the storage encryption adopts a partition key management; the intrusion detection unit monitors network attacks and illegal access in real time based on abnormal behavior analysis and feature matching, triggers a security alarm and blocks the connection when an anomaly is found;

[0045] A system management layer, which comprises a device management unit, a user permission unit and a running monitoring unit, the device management unit realizes the whole life cycle management of sensors, edge nodes and servers, including device registration, state monitoring, fault repair and firmware upgrade; the user permission unit divides the management permission according to roles, and different roles correspond to different operation ranges; the running monitoring unit displays the running state, data throughput and task completion rate of each module of the system in real time, and generates a system running report.

[0046] In the application, a dynamic resource allocation module is also included, which comprises a resource demand prediction unit and an allocation execution unit, the resource demand prediction unit constructs an LSTM time series prediction model by analyzing the resource consumption law of historical tasks and combining the current task queue length, outputs the resource demand curve in three time dimensions of 1 hour, 4 hours and 12 hours in the future, and the model training adopts a sliding window method, and the training data is updated once every 24 hours; the allocation execution unit introduces a resource elastic scaling mechanism based on the prediction result and real-time load, automatically starts a standby edge node when the predicted load exceeds a threshold, releases redundant resources when the load is below the threshold, and simultaneously adjusts the resource allocation ratio of each task through the integral formula The resource allocation ratio of the i-th task at time t is calculated, wherein R i (t) is the resource allocation ratio of the i-th task at time t, C i (s) is the resource demand of the i-th task at time s, P i (s) is the priority coefficient of the i-th task at time s (5 for an urgent task, 3 for a high priority, 2 for a medium priority, and 1 for a low priority), L(s) is the total load of the system at time s, t0 is the task start time, and t is the current time, the resource allocation ratio is dynamically adjusted through the integral formula, high-priority tasks obtain sufficient resources, containerization deployment technology is adopted to realize the rapid allocation and recovery of resources, the resource allocation response time is controlled within 500 milliseconds, and the overall operation efficiency of the system is improved.

[0047] In the application, a real-time compensation module is also included, which comprises a transmission delay monitoring unit and a compensation strategy generating unit, the transmission delay monitoring unit measures the transmission delay of real-time data from the sensor to the edge node and from the edge node to the cloud in real time through timestamp comparison and network jitter analysis, establishes a delay characteristic library, records the delay distribution law under different device types and network environments, and the delay measurement accuracy reaches 10 milliseconds; the compensation strategy generating unit adopts a predictive compensation mechanism based on the delay data, performs trend fitting on the historical delay data, predicts the possible delay fluctuation in advance, and simultaneously calculates the delay compensation amount through D is the total delay compensation amount, n is the number of transmission nodes, comp D is the total delay compensation amount, n is the number of transmission nodes, is the delay change rate of the kth node, is the time compensation interval of the kth node, the time compensation is performed on the data with high real-time requirement through the formula, the edge local priority response mechanism is adopted for the control instruction type data, the control instruction execution delay is compressed to within 200 milliseconds, the influence of the transmission delay on the control decision is reduced, and the timeliness of the control instruction is ensured.

[0048] In the application, the intelligent sensor group of the data perception layer also integrates a self-calibration unit and a multi-source fusion unit, the self-calibration unit can automatically correct the measurement deviation according to the environmental parameter change, the temperature sensor dynamically adjusts the measurement value within the range of-20 DEG C to 100 DEG C through the built-in temperature compensation curve, the environmental temperature is collected once every 30 minutes as a calibration reference, and the calibration error is controlled within ±0.5 DEG C; the vibration sensor adopts a self-adaptive filtering algorithm, can identify and filter out environmental vibration interference in real time, and distinguishes the inherent vibration of the equipment and the interference signal through spectrum analysis, and the signal-to-noise ratio is improved by more than 15 dB after filtering; the multi-source fusion unit adopts a method combining Kalman filtering and D-S evidence theory to perform fusion processing on the multi-sensor data of the same monitoring point, and the measurement accuracy is improved by 20%; meanwhile, the sensor group has a self-diagnosis function, can judge whether there is a fault by monitoring the self-supply voltage, the communication signal strength and the measurement stability, combining a fault identification model based on a BP neural network, the fault identification accuracy is more than 98%, when a fault is found, the sensor group actively sends an alarm information to the edge gateway and uploads the fault characteristic parameters, the edge gateway automatically starts a backup sensor after receiving the alarm, and simultaneously records the historical data of the fault sensor for post-analysis, the interruption time in data acquisition is controlled within 1 second, and the data acquisition continuity is ensured.

[0049] In the application, the dynamic scheduling decision unit of the adaptive scheduling layer also introduces a reinforcement learning mechanism and a dynamic task splitting algorithm. The reinforcement learning mechanism takes the system load, task completion time and resource utilization as reward function parameters, sets positive rewards (task completion in advance, improvement of resource utilization) and negative rewards (task delay, resource waste), the reward value quantization interval is -10 to 10, the decision model parameters are updated after completing each batch of tasks, the scheduling strategy is optimized through continuous interaction with the environment, and the scheduling efficiency is improved by 30%; the dynamic task splitting algorithm decomposes a complex task into multiple subtasks according to data correlation and computational complexity, establishes a subtask dependency graph, assigns different edge nodes using a parallel scheduling strategy, and realizes data interaction between subtasks through a distributed message queue, and the message transmission delay is less than 50 milliseconds; at the same time, the unit has a scheduling scheme simulation function, which can simulate the execution effect of different scheduling schemes before actual execution, including task completion rate, average delay and resource waste rate, an entropy weight-TOPSIS algorithm is introduced to comprehensively evaluate the simulation results, the evaluation index weight is objectively calculated by the entropy weight method, and the scheme with the optimal comprehensive performance is selected for execution, thereby reducing the risk of system performance decline caused by improper scheduling.

[0050] In the application, a cooperative communication mechanism and a load prediction sharing network are established between the edge nodes of the edge computing layer. The cooperative communication mechanism uses a distributed hash table (DHT) to realize node discovery and information sharing, the node discovery response time is less than 100 milliseconds, when the load of a certain edge node is too high, part of the non-real-time tasks can be migrated to a nearby node with lower load, the migration process uses an incremental data transmission method, only the raw data and intermediate results that have not been processed are transmitted, the data transmission amount is reduced by more than 60%, and the data consistency is ensured through a data consistency protocol before and after migration, and the data consistency is 100%; in the load prediction sharing network, each edge node broadcasts its load prediction result (CPU, memory occupation trend in the next 30 minutes) to the adjacent nodes every 5 minutes, the prediction error is controlled within 10%, and the receiving node adjusts the local task queue combined with its own prediction data; in addition, the edge node has heterogeneous computing capability, and can automatically select CPU, GPU or FPGA for processing according to the type of the task, GPU acceleration is enabled for image processing tasks, the processing speed is improved by 5 times, FPGA acceleration is enabled for logical judgment tasks, the processing speed is improved by 3 times, the computing efficiency is improved, when a fault occurs in the edge node, the task can be quickly switched to a backup node, the task switching time is less than 2 seconds, the computing task is not interrupted, and the fault tolerance of the system is improved.

[0051] In the application, the deep analysis engine of the cloud collaborative layer also includes a cross-scene model migration unit and a federated learning framework. The cross-scene model migration unit can migrate a fault early warning model trained in a certain production scene to a similar scene. The scene difference degree is measured by a maximum mean difference (MMD) algorithm. When the difference degree is lower than 0.2, the model is directly reused. When the difference degree is higher than 0.2, a small amount of on-site data is used to fine-tune the model parameters. The amount of fine-tuning data does not exceed 500, the model migration adaptation time is shortened to within 24 hours, and the model migration adaptation time is shortened to within 24 hours. The federated learning framework supports local training of models by each edge node, uploads only model parameter updates to the cloud, reduces parameter transmission by more than 90%, aggregates parameters in the cloud and distributes them to each node, realizes collaborative optimization of the model, and protects the privacy of industrial data. The deep analysis engine supports model self-updating. After accumulating 5000 new data, the model is automatically retrained. The sliding window mechanism is used to ensure that the model is always optimized based on the latest data, and the model prediction accuracy is maintained at more than 95%. The global optimization unit is based on the local optimization results of each edge node. An improved multi-objective genetic algorithm is used to construct an optimization function with the lowest energy consumption, the highest production efficiency and the smallest equipment loss as the target while meeting the local needs of each production unit. A global scheduling scheme is generated and distributed to each edge node to realize the collaborative optimization of the entire industrial internet, reduce the global energy consumption by 15%, and improve the production efficiency by 10%. The adaptive scheduling industrial internet data acquisition and control strategy,

[0052] The application comprises the following steps:

[0053] The data sensing step is that the intelligent sensor group collects industrial field data according to the dynamically adjusted sampling frequency. Slowly changing parameters such as temperature and pressure use a low sampling frequency, and fast changing parameters such as vibration and flow use a high sampling frequency. The collected data is preliminarily processed by the edge gateway and then transmitted to the adaptive scheduling layer through a hybrid communication network.

[0054] The task scheduling step is that the task priority division unit classifies the collection tasks, the resource load monitoring unit obtains the system load state in real time, and the dynamic scheduling decision unit generates a task allocation scheme by combining the task priority and the load state. The urgent and high priority tasks are allocated to the edge computing layer, and the medium and low priority tasks are allocated to the cloud collaborative layer.

[0055] The edge processing step is that after receiving the allocated tasks, the edge node extracts features and identifies abnormalities from the data. Control instructions are directly generated for real-time high priority data and issued to the field equipment. The processing results and the original data are cached and synchronized to the cloud at a set period.

[0056] The cloud-based analysis process involves the cloud collaboration layer receiving data uploaded from edge nodes and storing it in a big data storage center. The deep analysis engine mines historical data to build fault warning models and production optimization strategies. The global optimization unit generates a global scheduling plan and distributes it to each edge node.

[0057] The security protection steps include: the identity authentication unit verifies the identity of all access devices; the data encryption unit encrypts the transmitted and stored data; and the intrusion detection unit monitors the network security status in real time, triggers alarms and blocks the connection when an anomaly is detected.

[0058] The system management steps include: the equipment management unit manages all types of equipment throughout their lifecycle; the user permission unit controls the operational scope of different roles; and the operation monitoring unit displays the system's operating status in real time and generates operation reports.

[0059] In this invention, the task scheduling step further includes a load prediction sub-step and a task rescheduling mechanism. The load prediction sub-step... Predict the system load over a future time interval Δt, where L(t+Δt) is the system load over the future time interval Δt, and L(t) is the current system load. Let t be the load change rate, t be the current time, and Δt be the prediction time interval. A sliding window mechanism is used to update the prediction model every 10 minutes, with a window size of 2 hours. An attention mechanism is introduced to enhance the influence of recent data on the prediction results, achieving a prediction accuracy of over 85%. The task rescheduling mechanism is triggered when the deviation between the actual system load and the predicted load exceeds 15%. Task priorities and resource status are reassessed, and a new scheduling scheme is generated. A smooth transition strategy is used to gradually migrate tasks, with each migration not exceeding 20% ​​of the total task volume. This avoids system fluctuations caused by sudden scheduling changes. The task allocation scheme is adjusted in advance based on the prediction results, reducing the risk of performance degradation caused by sudden increases in system load and improving the foresight of scheduling decisions.

[0060] In this invention, the edge processing step also includes a data compression sub-step and a feature selection mechanism. The data compression sub-step employs an adaptive compression algorithm for non-critical data uploaded to the cloud. Calculate the compression ratio, where C is the data compression ratio and S is the compression ratio. raw S represents the original data size. comTo compress the size of the data, a mapping relationship between compression rate and data importance is established. The compression rate of important and real-time data is less than or equal to 30%, and the compression rate of non-important and low real-time data can be greater than or equal to 70%. At the same time, a compression quality verification algorithm is used to ensure that the error of the compressed data is within 5%. The feature selection mechanism is based on the Relief-F algorithm and mutual information entropy. The feature selection mechanism selects feature parameters with strong correlation with the device state from the collected data. The redundancy of feature selection is reduced by more than 40%, and redundant information is removed. The data usability is ensured while reducing the network transmission volume. The bandwidth occupation is reduced by 30%, and the efficiency of subsequent data analysis is improved.

[0061] The specific implementation of the system is further illustrated by two embodiments as follows:

[0062] Embodiment 1: Industrial Internet data acquisition and control application in automobile parts stamping workshop

[0063] This embodiment is applied to an automobile parts stamping workshop. The workshop includes 12 stamping machines, 8 conveying robots, and 6 conveyors. It needs to monitor 28 parameters such as temperature, pressure, vibration, and displacement. The equipment load reaches more than 85% during the peak production period and drops to less than 30% during the low production period. The system is required to realize real-time parameter acquisition, dynamic task scheduling, and rapid abnormal response.

[0064] I. System module deployment and parameter configuration

[0065] The data perception layer is deployed with multiple types of intelligent sensor groups, edge gateways, and local communication modules. The sensors include temperature, pressure, vibration, and displacement sensors, supporting multi-protocol adaptive access. The sampling frequency can be automatically switched according to the device operating state: 100ms / second for vibration and pressure sensors during high load of stamping machines, and 500ms / second during low load. The edge gateway integrates data preprocessing chips to perform preliminary cleaning and format conversion on raw data. The local communication module uses a hybrid network of industrial wireless and wired networks. The wireless part supports LoRa and 5G dual-mode switching, and the wired part is based on industrial Ethernet to ensure stable data transmission.

[0066] The adaptive scheduling layer includes task priority division, resource load monitoring, and dynamic scheduling decision units. The task priority division unit divides the acquisition tasks into four levels of emergency, high, medium, and low according to the real-time data requirements. The emergency task needs to be responded within 1 second. The resource load monitoring unit collects the load status of edge nodes and cloud servers in real time to generate a load state matrix. The dynamic scheduling decision unit generates an optimal scheduling scheme based on the load state and task priority using an improved genetic algorithm to realize reasonable allocation of tasks between edge and cloud. At the same time, a reinforcement learning mechanism is introduced to optimize the scheduling strategy.

[0067] Implementation of core functional modules

[0068] Dynamic resource allocation module: contains resource demand prediction and allocation execution unit, resource demand prediction unit constructs LSTM time series prediction model, outputs future three time dimension resource demand curve; allocation execution unit based on prediction results and real-time load, introduces resource elastic expansion mechanism, calculates task resource allocation rate through system designed formula, ensures that high priority task obtains sufficient resources, adopts containerized deployment technology to realize resource rapid allocation and recovery, response time control within 500 milliseconds.

[0069] Real-time compensation module: contains transmission delay monitoring and compensation strategy generation unit, transmission delay monitoring unit measures data transmission delay in real time, establishes delay feature library; compensation strategy generation unit based on delay data, adopts predictive compensation mechanism, calculates delay compensation amount through system designed formula, carries out time compensation for high real-time requirement data, controls instruction execution delay to be compressed to within 200 milliseconds.

[0070] Edge computing layer and cloud collaborative layer: after the edge node receives the high real-time requirement task allocated, data feature extraction, outlier identification are carried out, and on-site device control instruction is directly generated; at the same time, data is cached, and after network recovery, it is synchronized to the cloud. The big data storage center of the cloud collaborative layer stores historical data and processing results, the deep analysis engine constructs a device fault early warning model by using machine learning algorithm, and the global optimization unit combines the local optimization result of the edge node to generate a global scheduling scheme.

[0071] Control strategy execution data sensing step: sensors collect data at dynamically adjusted sampling frequency, which is transmitted to the adaptive scheduling layer through the hybrid communication network after being processed by the edge gateway; task scheduling step: hierarchical collection tasks, monitor system load, generate allocation scheme, allocate emergency and high priority tasks to edge computing layer; edge processing step: edge node processes data and generates control instruction, caches data and synchronizes to cloud; cloud analysis step: cloud stores data, mines and analyzes and generates global scheme; security protection step: verify access identity, encrypt data, monitor network security, alarm and block connection in abnormal situation; system management step: manage device whole life cycle, divide user rights, monitor system running state.

[0072] II. Effect verification data representation

[0073] Table 1: Performance index comparison before and after system deployment in automobile stamping workshop

[0074] Performance indicators Before deployment (traditional system) After deployment (this system) Promotion effect Average delay of emergency tasks (ms) 850 95 Significant reduction in delay Edge node resource utilization (%) 42 78 Significant improvement in resource utilization Data collection interruption time (s / month) 180 5 Significant reduction in collection interruption time Mold failure early warning accuracy (%) 72 96 Significant improvement in early warning accuracy

[0075] In Table 1: The data reflects the core advantages of the system. After deployment, the emergency task delay is greatly reduced, resulting from the dynamic task allocation and delay compensation mechanism, which ensures that emergency tasks have priority to access resources and that instructions are issued in a timely manner. The resource utilization rate of the edge node is significantly improved, thanks to the resource elastic expansion and load prediction, which avoids resource waste and overload. The interruption time in data acquisition is greatly shortened due to the sensor self-diagnosis and standby switching function. The mold failure warning accuracy is improved, reflecting the role of cloud and edge collaborative analysis. The overall data verifies the adaptability of the system to high-load industrial scenarios, solving the problems of traditional system scheduling rigidity, data interruption, and warning lag.

[0076] Example 2: Industrial Internet data acquisition and control application in a chemical ethylene production workshop

[0077] This example is applied to a large chemical ethylene production workshop, which includes 2 ethylene cracking furnaces, 4 propylene rectification towers, and 3 circulating water systems. It needs to monitor 35 parameters such as temperature, pressure, flow, and liquid level. The production process needs to strictly control temperature and pressure fluctuations, and the system is required to achieve stable data acquisition and safe scheduling in a high-risk environment.

[0078] I. System module deployment and parameter configuration

[0079] The intelligent sensors in the data perception layer are all explosion-proof, covering temperature, pressure, flow, and liquid level sensors. The sampling frequency is adjusted according to the process stage: the temperature sensor sampling frequency is 200 ms per time in the cracking furnace warming-up stage and 500 ms per time in the stable stage. The sensor integrates a self-calibration unit that can automatically correct measurement bias according to environmental parameter changes. The multi-source fusion unit uses Kalman filtering and D-S evidence theory to improve measurement accuracy. It has a self-diagnosis function with an accuracy of over 98% in fault identification, and automatically starts the standby sensor when a fault is detected. The edge gateway is explosion-proof, and the pre-processing algorithm adds hard clipping of outliers. The local communication module uses a combination of intrinsically safe industrial Ethernet and explosion-proof LoRa networking to avoid safety risks.

[0080] The adaptive scheduling layer and the edge computing layer The task priority division unit of the adaptive scheduling layer is classified according to safety impact and process correlation, and emergency tasks need to be responded within 0.5 seconds. The dynamic scheduling decision unit uses a dynamic task splitting algorithm to decompose complex tasks into sub-tasks for parallel processing, and the sub-tasks interact through a distributed message queue. The edge computing layer establishes a collaborative communication mechanism for the edge nodes, uses a distributed hash table to achieve node discovery and information sharing, and migrates part of the non-real-time tasks when the load is too high. The migration process uses incremental data transmission. It has heterogeneous computing capabilities and can automatically select processors for acceleration processing according to task type.

[0081] Implementation of core functional modules

[0082] Load prediction and task rescheduling: In the task scheduling step, the load prediction sub-step predicts future system load through the formula designed by the system, and updates the prediction model every 10 minutes using the sliding window mechanism to strengthen the influence of recent data on the prediction result. The task rescheduling mechanism is triggered when the deviation between the actual system load and the predicted load exceeds 15%, re-evaluates the task and resource status, generates a new scheduling scheme, and smoothly migrates tasks, with the amount of migrated tasks not exceeding 20% of the total task amount.

[0083] Data compression and feature selection: In the edge processing step, the data compression sub-step uses an adaptive compression algorithm for non-critical data, calculates the compression rate through the formula designed by the system, establishes a mapping relationship between the compression rate and the data importance, and ensures that the error of the compressed data is within 5%. The feature selection mechanism is based on the Relief-F algorithm and mutual information entropy, and selects feature parameters with strong correlation with device status and eliminates redundant information.

[0084] Cross-scenario model migration: The deep analysis engine of the cloud collaboration layer includes a cross-scenario model migration unit and a federated learning framework. The cross-scenario model migration unit can migrate a model trained in one production scenario to a similar scenario. The maximum mean difference algorithm is used to measure the scenario difference degree. When the difference degree is low, the model is directly reused, and when the difference degree is high, a small amount of on-site data is fine-tuned. The federated learning framework supports local training of models by each edge node, and only uploads model parameter updates to the cloud, realizing collaborative optimization of models.

[0085] Security protection and system management: The security protection layer includes identity authentication, data encryption, and intrusion detection units. The identity authentication unit uses a multi-factor authentication mechanism to verify the identity of device access. The data encryption unit encrypts the data transmitted and stored. Transmission encryption uses national encryption algorithms, and storage encryption uses partition key management. The intrusion detection unit is based on abnormal behavior analysis and feature matching, and monitors network attacks and illegal access in real time. When an anomaly is found, a security alarm is triggered and the connection is blocked. The system management layer includes device management, user permissions, and running monitoring units, realizes device lifecycle management, divides management permissions by role, and displays system running status and generates reports in real time.

[0086] II. Effect verification data representation

[0087] Table 2: Comparison of key indicators before and after deployment of the chemical ethylene plant system

[0088] Key indicators Before deployment (traditional system) After deployment (this system) Promotion effect Model cross-scene adaptation time (h) 72 22 Significant reduction in adaptation time Data transmission bandwidth occupancy (GB / day) 85 51 Significant reduction in bandwidth occupancy Security attack detection rate (%) 65 99 Significant improvement in attack detection rate Process parameter control accuracy (℃ / MPa) ±8 / ±0.15 ±3 / ±0.08 Significant improvement in control accuracy

[0089] In Table 2, the data reflects the value of the system adapting to the high-risk scene of chemical industry, the model adaptation time across scenes is greatly shortened, which is derived from the scene difference measurement and a small amount of data fine-tuning, avoiding the disadvantages of large data retraining; the bandwidth occupation is significantly reduced, which is attributed to data compression and feature selection, reducing non-critical data transmission; the safety attack detection rate is greatly improved, benefiting from multi-factor authentication and custom intrusion rules, blocking potential security risks; the process parameter control accuracy is significantly improved, reflecting the synergistic effect of real-time compensation and edge local control, ensuring that the parameters are stable within the safe range. The overall data verification system has advantages in explosion-proof and safe scheduling, meeting the high safety and stability requirements of chemical production.

[0090] Referring to Figure 3 : The figure directly verifies the core effect of the dynamic resource allocation module, and the numerical change fully conforms to the logic of the resource allocation formula - the higher the priority coefficient, the higher the resource allocation rate. For example, the resource allocation rate of the emergency task (priority coefficient 5) reaches 80%, which can quickly obtain CPU and memory resources, and complete the stamping die over-temperature data processing within 1 second, meeting the demand of the industrial field for rapid response to key parameter abnormalities; the allocation rate of the low-priority task (coefficient 1) is only 20%, which avoids occupying too many resources and affecting high-priority tasks. Compared with the traditional fixed resource allocation (each priority is 50%), the figure reflects the advantages of the system dynamically allocating resources according to priority, ensuring that key tasks are processed first, reducing the delay of emergency tasks caused by resource mismatch, significantly improving the overall operation efficiency of the system, and meeting the high requirements of the automobile stamping workshop for task response timeliness during the production peak period.

[0091] Referring to Figure 4 : The figure verifies the effectiveness of the load prediction and task rescheduling mechanism, and the data reflects two core effects: first, the higher the system load, the greater the task delay, which conforms to the objective law of industrial internet load and delay, such as the delay of 280ms without rescheduling when the load is 95%, which is far beyond the real-time response requirement of high-priority tasks (such as distillation tower liquid level abnormalities) in the chemical workshop; second, the delay is significantly reduced after rescheduling, from 200ms to 140ms when the load is 85%, and from 280ms to 180ms when the load is 95%, effectively avoiding the performance decline caused by sudden increase of system load. The fitting curve in the chart clearly shows the optimization amplitude of rescheduling on the delay, proving that the load prediction formula can predict the load change in advance, and through rescheduling, it adjusts the task allocation smoothly, ensures that the system still meets the real-time requirements under high load, improves the forward-looking of scheduling decisions, and meets the demand of continuous and stable operation of the chemical ethylene production workshop.

[0092] Referring to Figure 5The figure fully embodies the technical advantages of the data compression sub-step, and the values fully follow the design logic that "the higher the data importance, the lower the compression rate and the smaller the error". For example, the cracking furnace tube temperature data with extremely high importance (directly affecting production safety and process precision) has a compression rate of only 25%, an error of 2%, which reduces the data amount (from 100 MB to 25 MB), guarantees the parameter precision, and meets the requirement of the chemical plant for the control precision of the furnace tube temperature ± 3℃; the historical energy consumption statistical data with extremely low importance has a compression rate of 85%, an error of 8%, which greatly reduces the bandwidth occupation (from 85 GB / day to 51 GB / day), and does not affect the subsequent energy consumption analysis and report generation. Compared with the traditional fixed compression rate (all 50%), the figure proves that the compression strategy can achieve a precise balance between "data availability" and "bandwidth saving", effectively reduces the network transmission amount, improves the data analysis efficiency, and meets the actual needs of the real-time transmission and long-term storage of the massive data of the chemical ethylene plant.

[0093] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes to the technical range disclosed in the present application according to the technical solution and the inventive concept of the present application, which should be covered in the protection scope of the present application.

Claims

1. An adaptive scheduling industrial internet data acquisition and control system, characterized in that, include: The data sensing layer is deployed in various production units in the industrial site and includes multiple types of intelligent sensor groups, edge gateways and local communication modules. The intelligent sensor groups cover temperature, pressure, flow and vibration sensors. The edge gateway integrates a data preprocessing chip, which can perform preliminary cleaning and format conversion on the raw data; the local communication module adopts a hybrid industrial wireless and wired network. The adaptive scheduling layer includes a task priority division unit, a resource load monitoring unit, and a dynamic scheduling decision unit. The task priority division unit divides the collection tasks into four levels based on relevant factors; the resource load monitoring unit collects relevant data from edge nodes and cloud servers in real time; and the dynamic scheduling decision unit generates the optimal scheduling scheme based on load status and task priority using an improved genetic algorithm. The edge computing layer consists of edge nodes distributed across industrial sites. Each node is equipped with an industrial-grade processor and local storage unit to perform computational processing on tasks with high real-time requirements that are scheduled to the local machine. It also has a data caching function to cache critical data when the network is interrupted. The cloud collaboration layer, deployed on a cloud platform, includes a big data storage center, a deep analysis engine, and a global optimization unit. The big data storage center uses a distributed file system; the deep analysis engine uses machine learning algorithms to mine historical data; and the global optimization unit combines the local optimization results of each edge node. The security protection layer includes an identity authentication unit, a data encryption unit, and an intrusion detection unit. The identity authentication unit uses a multi-factor authentication mechanism to verify the relevant access identity; the data encryption unit encrypts the transmitted and stored data; and the intrusion detection unit monitors in real time based on relevant technologies, triggers alarms and blocks the connection when an anomaly is detected. The system management layer includes an equipment management unit, a user permission unit, and an operation monitoring unit. The equipment management unit manages the entire lifecycle of related equipment; the user permission unit divides management permissions according to roles; and the operation monitoring unit displays the real-time operating status of each module of the system.

2. The adaptive scheduling industrial internet data acquisition and control system according to claim 1, characterized in that, It also includes a dynamic resource allocation module, which comprises a resource demand prediction unit and an allocation execution unit. The resource demand prediction unit analyzes historical task resource consumption patterns and, combined with the current task queue length, outputs resource demand curves for the next 1 hour, 4 hours, and 12 hours. The model training uses a sliding window method. The allocation execution unit, based on the prediction results and real-time load, introduces a resource elastic scaling mechanism, and simultaneously... Calculate the resource allocation rate of the i-th task at time t, where R i (t) represents the resource allocation rate of the i-th task at time t, C i (s) represents the resource requirement of the i-th task at time s, P i L(s) is the priority coefficient of the i-th task at time s, L(s) is the total system load at time s, t0 is the task start time, and t is the current time.

3. The adaptive scheduling industrial internet data acquisition and control system according to claim 1, characterized in that, It also includes a real-time compensation module, which comprises a transmission delay monitoring unit and a compensation strategy generation unit. The transmission delay monitoring unit measures the transmission delay from the sensor to the edge node and from the edge node to the cloud in real time through timestamp comparison and network jitter analysis, recording the delay distribution patterns under different device types and network environments. The compensation strategy generation unit, based on the delay data, employs a predictive compensation mechanism to fit trends in historical delay data, while simultaneously... Calculate the delay compensation amount, where D comp The total delay compensation is n, where n is the number of transmission nodes. Let k be the rate of change of delay at the k-th node. This is the time compensation interval for the k-th node.

4. The adaptive scheduling industrial internet data acquisition and control system according to claim 1, characterized in that, The intelligent sensor group in the data perception layer also integrates a self-calibration unit and a multi-source fusion unit. The self-calibration unit automatically corrects measurement deviations based on changes in environmental parameters. The temperature sensor uses a built-in temperature compensation curve; the vibration sensor uses an adaptive filtering algorithm to identify and filter out environmental vibration interference in real time; the multi-source fusion unit uses a combination of Kalman filtering and DS evidence theory to fuse data from multiple sensors at the same monitoring point; at the same time, the sensor group has a self-diagnostic function, which monitors its own power supply voltage, communication signal strength, and measurement stability, and uses a fault identification model based on a BP neural network to determine whether a fault exists.

5. The adaptive scheduling industrial internet data acquisition and control system according to claim 1, characterized in that, The dynamic scheduling decision unit of the adaptive scheduling layer also introduces a reinforcement learning mechanism and a dynamic task splitting algorithm. The reinforcement learning mechanism uses system load, task completion time, and resource utilization as reward function parameters, and updates the decision model parameters after each batch of tasks is completed, optimizing the scheduling strategy through continuous interaction with the environment. The dynamic task splitting algorithm decomposes complex tasks into multiple sub-tasks according to data correlation and computational complexity, and assigns them to different edge nodes using a parallel scheduling strategy. The sub-tasks are communicated through a distributed message queue. At the same time, this unit has a scheduling scheme simulation function.

6. The adaptive scheduling industrial internet data acquisition and control system according to claim 1, characterized in that, A collaborative communication mechanism and a load prediction sharing network are established between edge nodes in the edge computing layer. The collaborative communication mechanism uses a distributed hash table. When an edge node is overloaded, some non-real-time tasks are migrated to nearby nodes with lower loads, and only unprocessed raw data and intermediate results are transmitted. In the load prediction sharing network, each edge node broadcasts its own load prediction results to neighboring nodes every 5 minutes. The receiving nodes adjust their local task queues based on their own prediction data. In addition, the edge nodes have heterogeneous computing capabilities, automatically selecting CPU, GPU, or FPGA for processing according to the task type. GPU acceleration is enabled for image processing tasks, and FPGA acceleration is enabled for logic judgment tasks to improve computing efficiency.

7. The adaptive scheduling industrial internet data acquisition and control system according to claim 1, characterized in that, The deep analysis engine in the cloud collaboration layer also includes a cross-scenario model transfer unit and a federated learning framework. The cross-scenario model transfer unit transfers a fault warning model trained in a production scenario to a similar scenario and measures the scenario difference through the maximum mean difference algorithm. The federated learning framework allows each edge node to train the model locally, only uploading model parameters to the cloud for updates. The cloud then aggregates the parameters and distributes them to each node. The deep analysis engine supports model self-updating, automatically retraining the model after accumulating 5,000 new data points. The global optimization unit is based on the local optimization results of each edge node.

8. A strategy for an industrial internet data acquisition and control system employing adaptive scheduling as described in any one of claims 1-7, characterized in that, Includes the following steps: In the data sensing step, the intelligent sensor group collects data from the industrial site at a dynamically adjusted sampling frequency. The low sampling frequency is used for slowly changing parameters such as temperature and pressure, while the high sampling frequency is used for rapidly changing parameters such as vibration and flow. After the collected data is initially processed by the edge gateway, it is transmitted to the adaptive scheduling layer through the hybrid communication network. The task scheduling process involves the task priority classification unit classifying the data collection tasks, the resource load monitoring unit acquiring the system load status in real time, and the dynamic scheduling decision unit generating a task allocation scheme by combining task priority and load status. In the edge processing steps, after receiving the assigned task, the edge node performs feature extraction and anomaly identification on the data. For abnormal data with high real-time requirements, it directly generates control commands and sends them to the field equipment. At the same time, it caches the processing results and the original data. The cloud-based analysis process involves the cloud collaboration layer receiving data uploaded from edge nodes and storing it in a big data storage center. The analysis engine mines historical data, and the global optimization unit generates a global scheduling plan and distributes it to each edge node. The security protection steps include: the identity authentication unit verifies the identity of all access devices; the data encryption unit encrypts the transmitted and stored data; and the intrusion detection unit monitors the network security status in real time, triggers alarms and blocks the connection when an anomaly is detected. The system management steps include: the equipment management unit manages all types of equipment throughout their lifecycle; the user permission unit controls the operational scope of different roles; and the operation monitoring unit displays the system's operating status in real time.

9. The strategy of the adaptive scheduling industrial internet data acquisition and control system according to claim 8, characterized in that, The task scheduling process also includes a load prediction sub-step and a task rescheduling mechanism. The load prediction sub-step... Predict the system load over a future time interval Δt, where L(t+Δt) is the system load over the future time interval Δt, and L(t) is the current system load. Δt represents the load change rate, t represents the current time, and Δt represents the predicted time interval. A sliding window mechanism is used. The task rescheduling mechanism is triggered when the deviation between the actual system load and the predicted load exceeds 15%, and the task priority and resource status are reassessed.

10. The strategy of the adaptive scheduling industrial internet data acquisition and control system according to claim 8, characterized in that, The edge processing steps also include a data compression sub-step and a feature selection mechanism. The data compression sub-step uses an adaptive compression algorithm on non-critical data uploaded to the cloud. Calculate the compression ratio, where C is the data compression ratio and S is the compression ratio. raw S represents the original data size. com This represents the size of the compressed data.

Images