Industrial internet of things oriented edge node data real-time processing method
Patent Information
- Application Number
- CN202611251070.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-10-27
Smart Images
Figure CN122994341A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial Internet of Things (IIoT) and edge computing technology, specifically to a method for real-time data processing of edge nodes for industrial IoT. Background Technology
[0002] With the rapid development of the Industrial Internet of Things (IIoT) and smart manufacturing, a large number of sensors, smart meters, controllers, and actuators are deployed in industrial settings, continuously generating massive amounts of heterogeneous data from multiple sources. Edge computing, by pushing computing, storage, and network resources to edge nodes closer to the data source, can effectively reduce data transmission latency, lower cloud load, and meet the low latency and high reliability requirements of industrial real-time control, predictive maintenance, and production optimization scenarios.
[0003] However, existing edge node data processing methods for the Industrial Internet of Things still have the following problems:
[0004] First, multi-source data fusion and standardization are difficult. Industrial field equipment is diverse, communication protocols and data formats are not uniform, sampling frequencies of various data sources differ significantly, and timestamps are difficult to align precisely, directly affecting the accuracy of subsequent anomaly detection, feature extraction, and real-time analysis.
[0005] Second, the adaptability of data preprocessing is insufficient. Traditional anomaly detection methods often use fixed thresholds or simple statistical rules, which are difficult to adapt to the dynamic changes in industrial operating conditions and are prone to false alarms or missed alarms; noise reduction processing often uses filtering methods with fixed parameters, which cannot be adaptively adjusted according to data characteristics.
[0006] Third, the decision between local processing and data offloading lacks flexibility. Existing methods typically pre-set fixed strategies for local processing or uploading to the cloud, without comprehensively considering the node's current resource consumption, network bandwidth, and the real-time requirements of data processing tasks, making it difficult to dynamically optimize between local computing and data offloading.
[0007] Fourth, edge analytics models are outdated. Traditional edge nodes typically deploy static analytics models, which cannot utilize real-time field data for online learning and updates, making it difficult to adapt to changes in equipment status and operational conditions; frequent uploads to the cloud for training also result in significant communication overhead and latency.
[0008] Fifth, the control command generation and cloud-edge collaboration mechanism are not perfect. The conversion from processing results to field control commands relies heavily on fixed rules and lacks effective integration with fault knowledge and semantic information, making it difficult to achieve intelligent decision-making in complex industrial scenarios; there is also a lack of efficient data synchronization and model aggregation mechanisms between edge nodes and the cloud.
[0009] In summary, existing technologies cannot simultaneously meet the requirements of real-time performance, accuracy, adaptability, and collaboration in edge node data processing within an Industrial Internet of Things (IIoT) environment. Summary of the Invention
[0010] (I) Technical problems to be solved
[0011] The technical problem to be solved by this invention is to overcome the above difficulties and provide a real-time data processing method for edge nodes in the Industrial Internet of Things that can improve the consistency of multi-source data, adaptive anomaly detection and noise reduction capabilities, realize dynamic decision-making for local processing and offloading, support online updates of edge models and cloud-edge collaboration, and meet the real-time and reliability requirements of industrial field data processing.
[0012] (II) Technical Solution
[0013] To address the aforementioned technical problems, the present invention provides a method for real-time data processing of edge nodes in the Industrial Internet of Things, comprising the following steps:
[0014] Step S1: The edge node collects multi-source data from industrial field equipment in real time. The multi-source data includes sensor data, equipment operating status data, and control command data.
[0015] Step S2: Perform timestamp alignment and format normalization on the multi-source data to obtain a standardized data stream;
[0016] Step S3: Perform anomaly detection and noise reduction on the standardized data stream using a sliding time window to obtain preprocessed data;
[0017] Step S4: Based on the current computing resource status, network status, and latency requirements of the data processing task of the edge node, determine whether to perform real-time processing locally;
[0018] Step S5: If it is determined that real-time processing is performed locally, the edge node uses the locally deployed real-time analysis model to perform real-time analysis on the preprocessed data to obtain the processing result; if it is determined that real-time processing is not performed locally, the preprocessed data or intermediate processing result is unloaded to an adjacent edge node or the cloud, and the returned processing result is received.
[0019] Step S6: Generate control commands based on the processing results and send the control commands to the industrial field execution unit.
[0020] As an improvement, in step S1, after the edge node collects multi-source data from industrial field equipment in real time, the method further includes: validating the multi-source data and removing data that exceeds the range, is repetitive, has incorrect timestamps, or does not conform to the preset format.
[0021] As an improvement, step S2 performs timestamp alignment and format normalization on the multi-source data, including: adding a unified timestamp to the data from each data source, and using interpolation or resampling to align data with different sampling frequencies to the same time axis; converting data with different protocols and formats into a unified data structure, the unified data structure including data source identifier, timestamp, data type, value, and unit.
[0022] As an improvement, step S3 utilizes a sliding time window to perform anomaly detection and noise reduction on the standardized data stream, including: determining an adaptive threshold based on the statistical characteristics of the data within the current window; detecting anomalies in the data within the window based on the adaptive threshold; marking or removing the detected anomalies; and smoothing the data within the window using median filtering, mean filtering, or Kalman filtering.
[0023] As an improvement, step S4 determines whether to perform real-time processing locally based on the current computing resource status, network status, and latency requirements of the data processing task of the edge node. This includes: obtaining the CPU utilization rate, memory utilization rate, task queue length, and uplink network bandwidth of the edge node; estimating the estimated completion time of local processing based on the type, data volume, and latency requirements of the data processing task; determining that real-time processing is performed locally when the estimated completion time meets the latency requirements and both the CPU utilization rate and memory utilization rate are lower than the corresponding preset thresholds; otherwise, determining that real-time processing is not performed locally.
[0024] As an improvement, step S5 unloads the preprocessed data or intermediate processing results to adjacent edge nodes or the cloud, including: determining the priority of the data to be unloaded; selecting an unloading target based on the priority of the data to be unloaded, task latency requirements, and available resources of adjacent edge nodes; compressing high-priority data and transmitting it first, caching low-priority data locally, and transmitting it when the network is idle.
[0025] As an improvement, in step S5, the edge node uses a locally deployed real-time analysis model to perform real-time analysis on the preprocessed data to obtain processing results, including: using a lightweight neural network model or an incremental learning model to extract and identify features from the preprocessed data to obtain prediction results of equipment operating status, fault type, remaining lifespan, or product quality.
[0026] As an improvement, the lightweight neural network model or incremental learning model is updated in the following way: the cloud uses aggregated historical data to train an initial model, compresses the initial model, and distributes it to the edge nodes; the edge nodes use newly collected real-time data to perform online incremental updates on the distributed model, and periodically upload the updated model parameters to the cloud for aggregation.
[0027] As an improvement, step S6 generates control instructions based on the processing results, including: matching the processing results with a preset rule base or fault knowledge graph to generate equipment start / stop control instructions, parameter adjustment instructions, or production scheduling suggestions.
[0028] As an improvement, the method further includes: synchronizing the processing results, preprocessed data, and model update parameters of the edge nodes to the cloud according to a preset period or event-triggered method, so that the cloud can perform global model training or production optimization based on the data aggregated from multiple edge nodes.
[0029] (III) Beneficial Effects
[0030] The advantages of this invention compared to the prior art are:
[0031] 1. This invention unifies multi-source heterogeneous data into a standardized data stream through timestamp alignment and format normalization, thereby improving the accuracy and consistency of subsequent processing;
[0032] 2. This invention employs adaptive threshold anomaly detection and filtering noise reduction based on a sliding time window, which can adapt to dynamic changes in industrial operating conditions, reduce false alarms and missed alarms, and improve data quality.
[0033] 3. This invention dynamically decides whether to process locally or offload tasks based on the real-time resource status, network status, and task latency requirements of edge nodes, thereby optimizing resource utilization and meeting real-time requirements.
[0034] 4. The present invention deploys lightweight neural network models or incremental learning models at edge nodes, supports online updates and cloud-edge collaboration, and can adapt to changes in equipment status and operating conditions, thereby improving analysis accuracy;
[0035] 5. This invention generates control commands by matching the processing results with a rule base or fault knowledge graph, and realizes cloud-edge data synchronization and model aggregation, thereby improving the level of intelligent decision-making and global optimization capabilities in industrial sites. Attached Figure Description
[0036] Figure 1 This is a flowchart illustrating the real-time data processing method for edge nodes in the Industrial Internet of Things (IIoT) according to the present invention.
[0037] Figure 2 This is a schematic diagram of the edge node and cloud collaborative architecture of the edge node data real-time processing method for industrial IoT of the present invention.
[0038] Figure 3 This is a schematic diagram of the sliding time window anomaly detection and noise reduction process in the real-time data processing method for edge nodes of the Industrial Internet of Things according to the present invention.
[0039] Figure 4This is a flowchart illustrating the local processing and offloading decision-making process of the real-time edge node data processing method for the Industrial Internet of Things according to the present invention.
[0040] Figure 5 This is a schematic diagram of the edge node real-time analysis model update process of the edge node data real-time processing method for the Industrial Internet of Things according to the present invention. Detailed Implementation
[0041] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.
[0042] Combined with appendix Figure 1-5 A real-time data processing method for edge nodes in the Industrial Internet of Things (IIoT) is proposed. This method can be applied to edge computing nodes in the IIoT, which can be industrial gateways, edge servers, embedded controllers, or other devices with computing and communication capabilities. The method includes the following steps:
[0043] S1: Edge nodes collect multi-source data from industrial field equipment in real time.
[0044] Edge nodes connect to industrial field devices via wired or wireless communication interfaces to collect multi-source data in real time. This multi-source data includes sensor data, equipment operating status data, and control command data. Sensor data includes parameters such as temperature, pressure, vibration, current, and flow rate; equipment operating status data includes parameters such as equipment on / off status, operating mode, speed, and load rate; and control command data includes start / stop commands and parameter settings from a host computer or controller.
[0045] Optionally, after collecting multi-source data, the validity of the multi-source data can be verified, and data that exceeds the measurement range, is duplicated, has incorrect timestamps, or does not conform to the preset format can be removed to ensure the data quality entering subsequent processing stages. For example, if the data of a temperature sensor exceeds its measurement range of -40℃ to 150℃, the data will be removed or marked as invalid; if multiple duplicate data appear at the same timestamp from the same data source, only one will be retained; if the data timestamp deviates significantly from the normal time series or has an incorrect format, it will be removed.
[0046] S2. Perform timestamp alignment and format normalization on the multi-source data to obtain a standardized data stream.
[0047] Since different data sources may use different sampling frequencies, time bases and data formats, it is necessary to perform timestamp alignment and format normalization on multi-source data.
[0048] Specifically, a unified timestamp is added to the data from each data source, and interpolation or resampling is used to align data with different sampling frequencies to the same timeline. For example, for data sources with lower sampling frequencies, linear interpolation or spline interpolation methods can be used to fill in missing time points; for data sources with higher sampling frequencies, downsampling or aggregation methods can be used to align them with the unified timeline. The unified timestamp can be obtained based on the local clock of the edge node or synchronization using the Network Time Protocol (NTP).
[0049] In terms of format normalization, data from different protocols and formats is converted into a unified data structure. This unified data structure includes the data source identifier, timestamp, data type, value, and unit. For example, a standardized data entry can be represented as: {source_id:"sensor_01",timestamp:1630000000,data_type:"temperature",value:85.3,unit:"℃"}. Through format normalization, subsequent processing modules do not need to concern themselves with underlying protocol differences and can directly operate on the standardized data stream.
[0050] S3. Use a sliding time window to perform anomaly detection and noise reduction on the standardized data stream to obtain preprocessed data.
[0051] Figure 3 The flowchart illustrates the anomaly detection and noise reduction process using a sliding time window, specifically including:
[0052] (1) Determine an adaptive threshold based on the statistical characteristics of the data within the current window, and perform outlier detection on the data within the window according to the adaptive threshold. The sliding time window size can be set according to the data characteristics and task requirements, for example, the window length is 100 data points and the sliding step size is 10 data points. Statistical characteristics include mean, standard deviation, median, interquartile range, skewness, kurtosis, etc. For example, using the 3σ criterion, when a data point in the window deviates from the window mean by more than 3 times the standard deviation, the point is determined to be an outlier; or a robust method based on the median and interquartile range is used to determine data points that exceed the range of [Q1-1.5×IQR,Q3+1.5×IQR] as outliers. Since the threshold is dynamically adjusted with the window data, it can adapt to changes in working conditions.
[0053] (2) Detected outliers are marked or removed, and the data within the window is smoothed using median filtering, mean filtering, or Kalman filtering. Data points confirmed as outliers can be directly removed and filled in using interpolation methods, or simply marked for subsequent analysis. Median filtering is suitable for removing impulse noise, mean filtering is suitable for removing random noise, and Kalman filtering is suitable for data with dynamic models. Optionally, if the data contains periodic fluctuations, an adaptive filtering method can be used.
[0054] S4. Based on the current computing resource status, network status, and latency requirements of the data processing task at the edge node, determine whether to perform real-time processing locally.
[0055] Figure 4 A flowchart illustrating the local processing and unloading decision-making process is shown, specifically including:
[0056] (1) Obtain the CPU utilization, memory utilization, task queue length, and uplink network bandwidth of the edge nodes. These metrics can be obtained in real time through the resource monitoring module of the edge nodes. For example, CPU utilization represents the current computing load, memory utilization represents the storage load, task queue length represents the number of tasks to be processed, and uplink network bandwidth represents the current network transmission capacity.
[0057] (2) Estimate the expected completion time of local processing based on the type of data processing task, data volume, and latency requirements. The expected completion time of local processing can be estimated based on task complexity, data volume, and currently available computing resources. For example, if the task is real-time fault diagnosis, the data volume is N standardized data points, the model inference time is t_infer, and the data preprocessing time is t_pre, then the estimated completion time of local processing can be estimated as t_local = t_pre + t_infer + t_queue, where t_queue is the task queuing time.
[0058] (3) When the estimated completion time meets the latency requirements and both CPU utilization and memory utilization are below the corresponding preset thresholds, it is determined that real-time processing will be performed locally; otherwise, it is determined that real-time processing will not be performed locally. The preset thresholds can be set according to the actual scenario, for example, the CPU utilization threshold is 80% and the memory utilization threshold is 85%. If the CPU utilization or memory utilization exceeds the threshold, or the estimated completion time exceeds the maximum allowable latency of the task, the task will be offloaded to an adjacent edge node or the cloud. This can avoid excessive processing latency due to local overload and ensure real-time performance.
[0059] Figure 2 This diagram illustrates a collaborative architecture between edge nodes and the cloud in an embodiment of the present invention. Multiple edge nodes are deployed in an industrial field, each connected to sensors, devices, and execution units in its respective area. Adjacent edge nodes can be interconnected via a local area network (LAN). Each edge node is connected to the cloud via a wide area network (WAN). The cloud is responsible for global model training, data aggregation, and production optimization.
[0060] S5. If it is determined that real-time processing is to be performed locally, the edge node will use the locally deployed real-time analysis model to perform real-time analysis on the preprocessed data and obtain the processing result; if it is determined that real-time processing is not to be performed locally, the preprocessed data or intermediate processing result will be unloaded to the adjacent edge node or the cloud and the returned processing result will be received.
[0061] For local processing, edge nodes utilize locally deployed real-time analytics models to analyze preprocessed data in real time. These real-time analytics models can be lightweight neural network models or incremental learning models. Lightweight neural network models, such as compressed deep learning models like MobileNet, ShuffleNet, and TinyBERT, are suitable for resource-constrained edge devices. Incremental learning models can update model parameters online as new data arrives, eliminating the need for retraining from scratch. The model extracts and identifies features from the preprocessed data to obtain predictions of equipment operating status, fault type, remaining lifespan, or product quality. For example, for vibration sensor data, the model can output the bearing fault type; for equipment operating data, the model can predict the remaining lifespan.
[0062] In the case of unloading, preprocessed data or intermediate processing results are unloaded to adjacent edge nodes or the cloud, and the returned processing results are received. Specifically, this includes:
[0063] (1) Determine the priority of the data to be unloaded. Priority can be determined based on factors such as data importance, urgency, and task type. For example, data involving safety interlocks has the highest priority, followed by process monitoring data, and historical statistical data has a lower priority.
[0064] (2) Select the unloading target based on the priority of the data to be unloaded, the task latency requirements, and the available resources of adjacent edge nodes. Adjacent edge nodes can be other edge nodes within the same local area network or edge servers at a higher level. When selecting, the computing power, current load, network latency, and reliability of the target node can be comprehensively considered.
[0065] (3) High-priority data is compressed and transmitted first, while low-priority data is cached locally and transmitted when the network is idle. Compression can use lossless or lossy compression algorithms to reduce the amount of data transmitted. High-priority data is sent immediately through a reliable transmission protocol; low-priority data is stored in a local cache queue and uploaded when network bandwidth is idle or the task is not urgent. This ensures that urgent tasks are processed in a timely manner and makes effective use of network resources.
[0066] S6. Generate control commands based on the processing results and send the control commands to the industrial field execution units.
[0067] The processing results can include equipment operating status, fault type, remaining lifespan, or product quality predictions. These results are then matched against a preset rule base or fault knowledge graph to generate equipment start / stop control commands, parameter adjustment commands, or production scheduling suggestions. For example, when equipment temperature exceeds a safety threshold and the model outputs an overheating fault, a shutdown command or load reduction command is generated based on the rule base; when the predicted remaining lifespan of the equipment is below a preset threshold, a maintenance reminder or production scheduling suggestion is generated. Control commands are sent to industrial field execution units, such as PLCs, frequency converters, and robot controllers, via industrial communication protocols (such as Modbus, OPCUA, Profinet, etc.).
[0068] Furthermore, the method in this embodiment of the invention may also include model update and cloud-edge collaboration steps:
[0069] Regarding model updates, lightweight neural network models or incremental learning models are updated in the following ways: the cloud uses aggregated historical data to train an initial model, compresses the initial model, and distributes it to edge nodes; the edge nodes use newly collected real-time data to perform online incremental updates on the distributed model, and periodically upload the updated model parameters to the cloud for aggregation. Figure 5 The diagram illustrates the model update process. For example, the cloud uses historical data from multiple industrial sites to train a general fault diagnosis model. A lightweight model is then obtained through compression techniques such as model pruning, quantization, and knowledge distillation, and distributed to various edge nodes. During actual operation, the edge nodes incrementally learn the model using newly collected real-time data, adapting the model to local equipment characteristics. At regular intervals or when model parameters change beyond a threshold, the updated model parameters are uploaded to the cloud. The cloud performs federated averaging or aggregation of the model parameters from multiple edge nodes to obtain a new global model, which is then redistributed, achieving continuous optimization through cloud-edge collaboration.
[0070] Regarding cloud-edge synchronization, the processing results, preprocessed data, and model update parameters of edge nodes can be synchronized to the cloud according to a preset cycle or event-triggered method. This allows the cloud to perform global model training or production optimization based on the data aggregated from multiple edge nodes. For example, processing results and preprocessed data can be synchronized every hour; synchronization can be triggered immediately when a critical fault is detected; and model update parameters can be synchronized at a fixed time each day. After the cloud aggregates data from multiple edge nodes, it can perform global model training, equipment health status assessment, and production process optimization, further improving the overall level of intelligence.
[0071] The method of this invention achieves efficient real-time processing of edge node data in an industrial IoT environment through multi-source data acquisition, timestamp alignment and format normalization, adaptive anomaly detection and noise reduction, local / offload dynamic decision-making, real-time edge analysis and online model updates, control command generation and cloud-edge collaboration. It can meet the requirements of industrial sites for low latency, high reliability and adaptability.
[0072] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, alterations, deletions of some features, additions of features, or recombinations of features to the above embodiments within the scope of the present invention without departing from the principles and spirit of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the innovative principles of the present invention shall still fall within the scope of the technical solutions of the present invention.
Claims
1. A method for real-time data processing of edge nodes for the Industrial Internet of Things, characterized by: Includes the following steps: Edge nodes collect multi-source data from industrial field equipment in real time, including sensor data, equipment operating status data, and control command data. The multi-source data is timestamped and formatted to obtain a standardized data stream; Anomaly detection and noise reduction are performed on the standardized data stream using a sliding time window to obtain preprocessed data. Based on the current computing resource status, network status, and latency requirements of the data processing task of the edge node, determine whether to perform real-time processing locally. If it is determined that real-time processing should be performed locally, the edge node will use the locally deployed real-time analysis model to perform real-time analysis on the preprocessed data to obtain the processing result. If it is determined that the preprocessed data or intermediate processing results are not to be processed locally in real time, the preprocessed data or intermediate processing results are unloaded to an adjacent edge node or the cloud, and the returned processing results are received. Control commands are generated based on the processing results and then sent to the industrial field execution units.
2. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 1, characterized in that: The multi-source data is aligned with timestamps and normalized in format, including: Add a uniform timestamp to the data from each data source, and use interpolation or resampling to align data with different sampling frequencies to the same timeline; Data from different protocols and formats is converted into a unified data structure, which includes a data source identifier, timestamp, data type, value, and unit.
3. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 1, characterized in that: Anomaly detection and noise reduction are performed on the standardized data stream using a sliding time window, including: An adaptive threshold is determined based on the statistical characteristics of the data within the current window, and outlier detection is performed on the data within the window based on the adaptive threshold. Detected outliers are marked or removed, and the data within the window is smoothed using median filtering, mean filtering, or Kalman filtering.
4. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 1, characterized in that: Based on the current computing resource status, network status, and latency requirements of the data processing task at the edge node, determine whether to perform real-time processing locally, including: Obtain the CPU utilization, memory utilization, task queue length, and uplink network bandwidth of the edge nodes; Estimate the expected completion time of local processing based on the type of data processing task, the amount of data, and latency requirements; If the estimated completion time meets the latency requirements and both CPU utilization and memory utilization are lower than the corresponding preset thresholds, it is determined that real-time processing will be performed locally; otherwise, it is determined that real-time processing will not be performed locally.
5. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 1, characterized in that: Unloading the preprocessed data or intermediate processing results to adjacent edge nodes or the cloud includes: Determine the priority of the data to be uninstalled; The uninstallation target is selected based on the priority of the data to be uninstalled, the task latency requirements, and the available resources of adjacent edge nodes; High-priority data is compressed and transmitted first, while low-priority data is cached locally and transmitted when the network is idle.
6. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 1, characterized in that: The edge nodes utilize locally deployed real-time analysis models to perform real-time analysis on the preprocessed data, obtaining processing results, including: The preprocessed data is used to extract and identify features using a lightweight neural network model or an incremental learning model to obtain prediction results for equipment operating status, fault type, remaining lifespan, or product quality.
7. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 6, characterized in that: The lightweight neural network model or incremental learning model is updated in the following ways: The cloud uses aggregated historical data to train an initial model, and then compresses the initial model and sends it to the edge nodes; The edge nodes use newly collected real-time data to perform online incremental updates to the distributed model, and periodically upload the updated model parameters to the cloud for aggregation.
8. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 1, characterized in that: Based on the processing result, control instructions are generated, including: The processing results are matched with a preset rule base or fault knowledge graph to generate equipment start / stop control commands, parameter adjustment commands, or production scheduling suggestions.
9. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 1, characterized in that: It also includes synchronizing the processing results, preprocessed data, and model update parameters of the edge nodes to the cloud according to a preset cycle or event-triggered method, so that the cloud can perform global model training or production optimization based on the data aggregated from multiple edge nodes.
10. The real-time data processing method for edge nodes in the Industrial Internet of Things according to claim 1, characterized in that: After the edge node collects multi-source data from industrial field equipment in real time, it also includes: The validity of the multi-source data is verified, and data that exceeds the range, is duplicated, has incorrect timestamps, or does not conform to the preset format is removed.