Boiler industry data information transmission method based on big data

Abstract

The invention relates to the technical field of boilers, and provides a boiler industry data information transmission method based on big data, and the method comprises the steps: deploying an industrial-grade edge gateway, discovering and recording a network identifier, manufacturer information and a serial number of an access device, carrying out the verification, distributing a unique identifier for the device according to a preset naming rule, and generating a trusted device entry; timestamps and identifiers are stamped on the measuring point metadata, an original time sequence is preprocessed in an edge calculation unit, and statistical characteristics are calculated and compared with a historical baseline and a threshold value of equipment; deploying a pre-trained convolutional auto-encoder at an edge end, constructing and training a multivariable time sequence convolutional network at a cloud end, weighting edge scores, cloud end scores and the importance degrees of the measuring points to synthesize an abnormal priority, and performing mapping according to a preset threshold value; and mapping the exception priority with a strategy table, generating candidate scheduling instructions by a decision engine, verifying the candidate instructions in an edge sandbox, and collecting feedback data to evaluate a disposal effect.

Description

Technical Field

[0001] This invention relates to the field of boiler technology, specifically a data information transmission method for the boiler industry based on big data. Background Technology

[0002] Boiler data refers to various measurement, monitoring, and operating parameters collected and recorded during boiler operation, including real-time operating data, historical operating data, and equipment status data. These data are collected through smart instruments and monitoring systems and are used to monitor boiler operating status, optimize power generation efficiency, reduce fuel consumption, and support decision-making.

[0003] Chinese patent application number CN202210644580.2 discloses a method and system for transmitting boiler industry data information based on big data. The configuration method includes: a mobile terminal collecting boiler industry data information and receiving an access configuration sent by a boiler industry data information receiving center; the mobile terminal determining that it is accessing the boiler industry data information receiving center for the first time; the mobile terminal generating a connection request message A; after sending the connection request message A to the boiler industry data information receiving center, the mobile terminal receiving a response message A sent by the boiler industry data information receiving center, wherein the response message A includes synchronization information, resource allocation information, and a temporary identifier; after receiving the response message A, the mobile terminal sending boiler industry data information to the boiler industry data information receiving center, wherein the boiler industry data information is identified using a temporary identifier.

[0004] In the field of boiler technology, although there are technical solutions for collecting boiler industry data and receiving access configurations sent by boiler industry data receiving centers via mobile terminals, existing technologies have the following problems in scenarios involving remote monitoring and early warning of boilers with multi-system collaboration:

[0005] 1. Due to inconsistent interface and configuration specifications across different systems, there is a lack of effective data sharing and collaboration, resulting in information silos and difficulties in integration and maintenance when dealing with compatibility issues;

[0006] 2. To meet the needs of multi-system collaborative operation, there is some delay in data transmission during the boiler industry, which may lead to delayed early warning information, affect response speed, and consequently reduce the safety of boiler operation. Summary of the Invention

[0007] This invention provides a data information transmission method for the boiler industry based on big data, aiming to solve the problem of untimely sharing of early warning information when multiple systems are working together, which leads to data delays and thus affects security.

[0008] The technical solution adopted by this invention to solve the above-mentioned technical problems is as follows: A data information transmission method for the boiler industry based on big data is provided, comprising:

[0009] Unified access to sensing devices, generation of a trusted device list, deployment of industrial-grade edge gateways at the boiler site, discovery and recording of network identifiers, manufacturer information and serial numbers of access devices, uniqueness verification and address conflict detection in the registration database, allocation of unique identifiers to devices according to preset naming rules and registration of measurement point metadata, generation of trusted device entries;

[0010] Boiler data is collected in real time, preprocessed by the edge computing unit, and local timestamps and source device identifiers are added to the metadata of the measuring points to obtain the raw time series stream. The raw time series stream is then preprocessed in the edge computing unit to calculate statistical features and compare them with the equipment's historical baseline and threshold.

[0011] The detection model analyzes the data, classifies anomalies and issues warnings. A pre-trained convolutional autoencoder is deployed at the edge, and a multivariate temporal convolutional network is built and trained in the cloud. The edge score, cloud score and the importance of the measurement point are weighted to synthesize anomaly priority, and then mapped according to a preset threshold.

[0012] Automatic scheduling decisions are made, records are collected and coordinated in a closed loop, anomaly priorities are mapped to the policy table, candidate scheduling instructions are generated by the decision engine, candidate instructions are verified in the edge sandbox, and feedback data is collected during the execution period using short-window high-frequency sampling to evaluate the handling effect.

[0013] As a preferred implementation, the specific steps for generating a trusted device list through unified access to sensing devices are as follows: An industrial-grade edge gateway is selected as the edge computing unit at the boiler site. Various sensing devices are uniformly connected to the edge gateway. Basic device information is entered according to a pre-planned access form. By setting the edge gateway as a DHCP proxy, the MAC address, hostname, and assigned IP address of new devices are automatically recorded to obtain an identifiable device list. Devices are guided to perform initial registration and identity verification, and connectivity tests and data sampling time synchronization verification are performed to obtain a data source with a clear reporting source and consistent timestamps. A unique identifier is assigned to each device according to a unified naming rule and preset tags. The maintainer and responsible department are recorded. The information is summarized in the device ledger and a list containing status identifiers, security identifiers, and change records is generated, creating a trusted device list for data access, monitoring, and access control.

[0014] Based on the entered device serial number and recorded network identifier, uniqueness verification and address conflict detection are performed in the registration database. A certificate issuance and verification process is initiated to the device to obtain the device's trusted identity credentials. The network connectivity is tested using the ICMP protocol, and response latency and packet loss rate are recorded to obtain the device's reachability and protocol availability. Data is sampled from the device several times, and variance analysis and timestamp consistency verification are performed on the measured values ​​to evaluate the data reporting quality and time synchronization deviation. Devices that do not have a trusted identity or do not meet connectivity and data quality requirements are isolated and manually reviewed.

[0015] As a preferred implementation, the specific steps of the edge computing unit preprocessing for real-time acquisition of boiler data are as follows: the edge computing unit reads the original measurement values ​​in real time at a preset sampling frequency to obtain an original data stream with a local timestamp and source device identifier; the original data stream is subjected to bandpass filtering, noise reduction, and missing value interpolation to obtain a smooth and continuous time-series signal; the processed time-series signal is time-aligned and unit-normalized according to a sliding window, different units are converted into a unified unit and the sampling points are aligned; at the same time, the key statistical features of the data are calculated and compared in real time with the equipment historical baseline and a preset threshold.

[0016] The processed time series and extracted features are encapsulated into standardized data frames. The specific physical quantities measured within these data frames are then incrementally differentially encoded for adjacent time series data. The importance score of each measurement point is calculated, and safety, control importance, and historical fault correlation are fused according to preset weights. The formula for calculating the importance score is as follows:

[0017] ,

[0018] Where I represents the importance score of the measuring point. Represents security weights. Indicates control weights, Indicates the fault correlation weight, Indicates the safety of the measuring point. This indicates the control importance of the measuring point. Indicates the historical fault correlation of the measuring point;

[0019] By maintaining a baseline value and corresponding version number for each measurement point in the edge computing unit, the difference between the real-time sampled value and the reference baseline is calculated and compared with a preset difference threshold to determine whether a valid change has occurred. For measurement points determined to have a valid change, differential entries are constructed and differential frames are formed by compact encoding. At the same time, a complete snapshot is triggered periodically. The cloud returns an acknowledgment for successfully received and verified differential frames. The edge computing unit then updates the local baseline value and version number to obtain a baseline synchronization state that is consistent between the edge and the cloud.

[0020] As a preferred embodiment, the specific steps for deploying and calibrating the sensing device to collect and extract visual features are as follows:

[0021] A pre-trained anomaly detection model is applied to standardized data frames. A hybrid anomaly detection framework combining edge and cloud computing is adopted. At the edge computing unit, a convolutional autoencoder is used to reconstruct the original temporal window, simultaneously calculating statistics. The reconstruction error and statistics are then fused into an edge score according to preset weights. In the cloud, a multivariate temporal convolutional network is used to perform temporal modeling on the temporal window from multiple measurement points, obtaining a temporal anomaly score reflecting complex multivariate interactions. The edge score, cloud-based temporal anomaly score, and measurement point importance scores are then combined according to weights to form an anomaly priority, as shown in the formula:

[0022] ,

[0023] Where P represents the exception priority. Indicates marginal scores, Indicates the time-series anomaly score in the cloud. Indicates the importance score of the measuring point. Indicates edge weights, Indicates cloud weight, Indicates the weight of the measurement point;

[0024] The anomaly priority P is compared with a preset threshold. When P ≥ 0.8, it is determined to be the highest priority level; when 0.6 ≤ P < 0.8, it is determined to be the high priority level; when 0.4 ≤ P < 0.6, it is determined to be the medium priority level; and when P < 0.4, it is determined to be the low priority level. The system automatically generates warning messages containing measurement point identification, anomaly type, confidence level, time window, and relevant context, and sends them to the field edge execution unit according to priority, converting the prediction results into executable operation and maintenance instructions.

[0025] As a preferred implementation, the specific steps for deploying and calibrating the sensing device to collect and extract visual features are as follows: mapping the anomaly priority to a preset strategy table, calling the decision engine to generate a scheduling instruction to be executed, obtaining clear scheduling action candidates, performing security and constraint verification on the scheduling instruction in the edge sandbox, including permission verification and physical constraint checks, obtaining a set of executable actions that have been verified and are within the security constraints, issuing a notification to the relevant responsible person according to the execution level for verification, issuing a standardized scheduling request to the edge actuator, obtaining specific control actions that can be directly executed on site, and monitoring the relevant measuring points in real time with a shorter window during the execution process to obtain immediate feedback data for judging the control effect;

[0026] The execution results are written to the audit log. The data is automatically labeled and archived to the training sample queue in the cloud to obtain real samples for subsequent model retraining. A rollback strategy is triggered for execution failures. The hit rate, handling timeliness and manual intervention ratio of various scheduling decisions are summarized regularly, and the statistical results are included in the operation and maintenance indicator dashboard to obtain quantifiable performance evaluation and data basis for automatic fine-tuning of decision parameters.

[0027] The beneficial effects of this invention are as follows:

[0028] 1. This invention uses a unified access and trust mechanism to map various protocols into unified standardized data frames, thereby reducing system integration workload, lowering maintenance costs, and improving cross-system collaboration efficiency;

[0029] 2. This invention integrates edge scores and multivariate time series model scores according to weights to form anomaly priorities, thereby enabling the immediate identification of high-risk events and ensuring their reporting with the lowest latency and highest reliability, thus improving the reliability of handling and the system's adaptability. Attached Figure Description

[0030] Figure 1 This is a flowchart of a data information transmission method for the boiler industry based on big data.

[0031] Figure 2 This is a comparison chart showing the effects of a data information transmission method in the boiler industry based on big data. Detailed Implementation

[0032] To make the technical means, creative features, objectives, and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention.

[0033] Example 1, as Figure 1 This is a data information transmission method for the boiler industry based on big data, including the following steps:

[0034] Unify access to sensing devices and generate a list of trusted devices;

[0035] Real-time acquisition of boiler data, preprocessed by edge computing unit;

[0036] The detection model analyzes data, classifies anomalies, and issues warnings.

[0037] Automatic scheduling decisions are made, and records are collected to create a collaborative closed loop.

[0038] The following are the specific implementation steps: A data information transmission method for the boiler industry based on big data, wherein the specific steps for unified access to sensing devices and generation of a trusted device list are as follows:

[0039] An industrial-grade edge gateway is selected as the edge computing unit at the boiler site. Various sensing devices, such as temperature sensors, pressure sensors, and flow sensors, are uniformly connected to the edge gateway. Basic information of the devices, including manufacturer, model, serial number, installation location, measurement range, and unit, is entered according to a pre-planned access form. By setting the edge gateway as a DHCP proxy, the MAC address, hostname, and assigned IP address of new devices are automatically recorded to obtain an identifiable device list. The devices are guided to perform initial registration and identity verification, and connectivity tests and data sampling time synchronization verification are performed to obtain a data source with a clear reporting source and consistent timestamp. Each device is assigned a unique identifier according to a unified naming rule and preset tags, such as a naming rule of point-area-device type-function-serial number-version. At the same time, the maintenance person and responsible department are recorded. The information is summarized into the device ledger and a list containing status identifiers, security identifiers, and change records is generated to create a trusted device list for data access, monitoring, and access control.

[0040] Specifically, the guiding device performs initial registration and identity verification, and executes connectivity testing and data sampling time synchronization verification. Based on the entered device serial number and recorded network identifier, it performs uniqueness verification and address conflict detection in the registration database, initiates certificate issuance and verification process for the device, such as key exchange based on certificates, to obtain the device's trusted identity credentials, uses the ICMP protocol to test network connectivity, and records response latency and packet loss rate to obtain the device's reachability and protocol availability. It performs several data samplings on the device and performs variance analysis and timestamp consistency verification on the measured values ​​to evaluate the data reporting quality and time synchronization deviation. Devices that do not have trusted identities or do not meet connectivity and data quality requirements are isolated and manually reviewed.

[0041] The specific steps for real-time acquisition of boiler data and preprocessing by the edge computing unit are as follows:

[0042] The edge computing unit reads raw measurement values ​​such as temperature, pressure, and flow rate in real time according to a preset sampling frequency. For example, the temperature sensor frequency is 1Hz and the flow sensor frequency is 0.5Hz, resulting in a raw data stream with a local timestamp and source device identifier. The raw data stream is then processed by bandpass filtering, noise reduction, and missing value interpolation to obtain a smooth and continuous time-series signal. The processed time-series signal is then time-aligned and unit-normalized according to a sliding window. Specifically, different units are converted into a unified unit and the sampling points are aligned. At the same time, key statistical features of the data, such as mean, variance, and moving average, are calculated and compared in real time with the device's historical baseline and a preset threshold.

[0043] The processed time series and extracted features are encapsulated into standardized data frames. The specific physical quantities measured within these data frames, such as temperature, pipe pressure, and flow rate at a certain location, are abstracted as measurement points. Incremental differential encoding is applied to adjacent time series data, and the importance score of each measurement point is calculated. Safety, control importance, and historical fault correlation are then fused according to preset weights. Safety is assessed based on whether the measurement point is a protection trigger point; control importance is assessed based on whether the measurement point directly participates in the control algorithm; and historical fault correlation is assessed based on historical alarm logs. The formula for calculating the importance score is as follows:

[0044] ,

[0045] Where I represents the importance score of the measuring point. Represents security weights. Indicates control weights, Indicates the fault correlation weight, Indicates the safety of the measuring point. This indicates the control importance of the measuring point. Indicates the historical fault correlation of the measuring point;

[0046] Specifically, the incremental differential coding of adjacent time-series data involves maintaining a baseline value and a corresponding version number for each measurement point in the edge computing unit. The difference between the real-time sampled value and the reference baseline is calculated and compared with a preset difference threshold to determine whether a valid change has occurred. Differential entries are constructed only for measurement points that are determined to have a valid change, and differential frames are formed by compact coding, such as fixed-point differences. At the same time, a complete snapshot is triggered periodically. The cloud returns an acknowledgment for successfully received and verified differential frames. The edge computing unit updates the local baseline value and version number to obtain a baseline synchronization state consistent between the edge and the cloud.

[0047] The specific steps for the detection model to analyze data, classify anomalies, and issue warnings are as follows:

[0048] A pre-trained anomaly detection model is applied to standardized data frames. A hybrid anomaly detection framework combining edge and cloud computing is employed. At the edge computing unit, a convolutional autoencoder reconstructs the original temporal window, simultaneously calculating statistics such as mean and standard deviation. The reconstruction error and statistics are then fused together with preset weights to form an edge score. In the cloud, a multivariate temporal convolutional network models the temporal window from multiple measurement points, yielding a temporal anomaly score reflecting complex multivariate interactions. The edge score, cloud-based temporal anomaly score, and measurement point importance scores are weighted and combined to form an anomaly priority, as shown in the formula:

[0049] ,

[0050] Where P represents the exception priority. Indicates marginal scores, Indicates the time-series anomaly score in the cloud. Indicates the importance score of the measuring point. Indicates edge weights, Indicates cloud weight, Indicates the weight of the measurement point;

[0051] The anomaly priority P is compared with a preset threshold. When P ≥ 0.8, it is determined to be the highest priority level; when 0.6 ≤ P < 0.8, it is determined to be the high priority level; when 0.4 ≤ P < 0.6, it is determined to be the medium priority level; and when P < 0.4, it is determined to be the low priority level. The system automatically generates early warning messages containing measurement point identifiers, anomaly types, confidence levels, time windows, and relevant contexts, and sends them to the field edge execution units according to priority, transforming the prediction results into executable operation and maintenance instructions.

[0052] Specifically, a multivariate temporal convolutional network is used in the cloud to perform temporal modeling of temporal windows from multiple measurement points. A multivariate temporal convolutional network is constructed and trained in the cloud. A multivariate temporal convolutional network is a deep network that uses one-dimensional dilated convolution and residual connections in the time dimension to model signals from multiple measurement points simultaneously. The aligned time windows from multiple measurement points are used as the model input.

[0053] In the training phase of the multivariate temporal convolutional network, semi-supervised training is performed using labeled fault samples, with the goal of minimizing reconstruction error. In the inference phase, the output of the multivariate temporal convolutional network is normalized and calibrated according to the statistics of the training period. Each inference result is accompanied by a version number, input window identifier and output confidence, and written into the event log and reporting packet to obtain a traceable record of temporal anomaly judgment.

[0054] The specific steps for automatically making scheduling decisions and collecting and recording collaborative closed-loop processes are as follows:

[0055] The abnormal priority is mapped to the preset strategy table, and the decision engine is called to generate the scheduling instructions to be executed, so as to obtain clear scheduling action candidates, such as priority uplink, generating work order, and issuing control instructions. The scheduling instructions are verified for security and constraints in the edge sandbox, including permission verification and physical constraint check, so as to obtain a set of executable actions that have been verified and are within the security constraints. The relevant responsible persons are notified according to the execution level for verification. The standardized scheduling request is sent to the edge actuator to obtain the specific control actions that can be directly executed on site. During the execution process, the relevant measuring points are monitored in real time with a shorter window to obtain immediate feedback data for judging the control effect.

[0056] The execution results are written to the audit log. By automatically labeling the data pairs in the cloud and archiving them to the training sample queue, real samples that can be used for subsequent model retraining are obtained, which plays a role in building a continuous learning closed loop based on actual operation and maintenance results. In the event of execution failure, a rollback strategy is triggered, such as increasing the alarm level and generating a mandatory manual work order. The hit rate, handling timeliness and manual intervention ratio of various scheduling decisions are summarized regularly, and the statistical results are included in the operation and maintenance indicator dashboard to obtain quantifiable performance evaluation and data basis for automatically fine-tuning decision parameters.

[0057] like Figure 2 This is a comparison chart of the effects of a data information transmission method for the boiler industry based on big data. The horizontal axis lists key performance indicators, and the vertical axis represents exemplified performance scores, ranging from 0 to 100. Higher values ​​indicate better performance. The aim is to intuitively demonstrate the expected improvement of the present invention in key capabilities compared to typical existing technologies.

[0058] Example 2, based on Example 1 above, presents a data information transmission method for the boiler industry based on big data in a multi-system collaborative boiler remote monitoring and early warning scenario, specifically as follows:

[0059] Step 1: Deploy a trusted access and device fingerprint extraction module on the edge gateway. Actively probe and obtain the hardware identifier and protocol characteristics of the device through the Modbus protocol to form physical and protocol fingerprints for comparison with the registration database. Verify the device identity based on the TPM remote authentication mechanism and store the trusted key fingerprint. Perform read and write operations on the application layer interface, heartbeat monitoring, and variance and cross-correlation analysis of samples within the window to complete the verification of connectivity and time consistency. Then, quantitatively evaluate the device's reachability, protocol status, and data timing consistency. Record the fingerprint, verification results, latency statistics, and quality score in the device ledger. Automatically mark non-compliant devices as isolated and generate manual processing work orders to form an auditable and verified list of trusted devices.

[0060] Step 2: Multi-rate sampling is performed on different types of measurement points in the edge computing unit. Kalman filtering and wavelet denoising techniques are used to generate smooth time-series signals with confidence intervals. The time-series data is resampled on a unified time base and converted according to physical units to obtain aligned and comparable data sequences. Feature values ​​such as mean, variance, and root mean square value are calculated within a sliding window, and a health score is synthesized according to weights for threshold judgment and priority ranking. Standardized data frames are generated based on the change set indicated by the bitmap and fixed-point difference encoding. Data packets are placed in a preemptive priority queue according to the anomaly score and the importance of the measurement point, and a token bucket and multi-path redundancy strategy is used for transmission according to the link status.

[0061] Step 3: Perform short-term prediction on standardized data frames, calculate prediction residuals and several statistics, and combine the residuals and statistics into marginal anomaly scores according to preset weights. In the cloud, with multi-path aligned time windows as input, construct and train a multivariate temporal convolutional network with the goal of predicting future windows. Establish a probability density model for the error distribution output by the multivariate temporal convolutional network to generate normalized cloud anomaly scores. Combine the marginal scores, cloud scores and measurement point importance into anomaly priorities, map them to preset thresholds to create levels, and automatically generate alarm packets containing measurement point identifiers, confidence levels, time windows and necessary contexts to send to the field, obtain graded responses and drive manual handling.

[0062] Step four involves mapping the generated anomaly priorities to a pre-defined strategy table. The decision engine then outputs candidate scheduling instructions, resulting in a set of optional actions, including uplink priority, work order creation, and control commands. In the edge sandbox, permission verification, process constraints, and simulation playback are applied to the candidate instructions to obtain a set of validated executable instructions that meet safety boundaries. Based on the execution level, the corresponding responsible person is automatically notified, and standardized scheduling requests are issued to the edge actuators according to authorization. During execution, short-window high-frequency sampling is used for relevant measurement points, and the expected response is compared in real time. By writing instruction issuance records, execution logs, and effect data into the audit database, various cases are automatically labeled and archived as training samples, resulting in a real data pool for model retraining and strategy optimization.

[0063] The embodiments of the present invention described above are subject to modification and change of method by those skilled in the art without departing from the embodiments and broader aspects of the present invention. The appended claims are intended to include all such modifications and changes of method that do not depart from the present invention.

Claims

1. A big data-based boiler industry data information transmission method, characterized in that, Comprise: Uniformly access sensing devices, generate a list of trusted devices, deploy an industrial edge gateway in the boiler field, discover and record the network identity, manufacturer information and serial number of the access device, perform uniqueness verification and address conflict detection in the registration library, assign a unique identifier to the device according to the preset naming rules and register the measurement point metadata, generate a trusted device entry; Real-time acquisition of boiler data, edge computing unit preprocessing, local timestamp and source device identifier are added to the measurement point metadata, and the original time series flow is obtained, the original time series is preprocessed in the edge computing unit, the statistical characteristics are calculated and compared with the device historical baseline and threshold; Data analysis by detection model, classify abnormal situations and give early warning, deploy a pre-trained convolutional autoencoder on the edge, build and train a multivariate time series convolutional network on the cloud, weight the edge score, cloud score and measurement point importance to synthesize abnormal priority, and map according to the preset threshold; Automatic scheduling decision, collect and record the collaborative closed loop, map the abnormal priority and strategy table, generate candidate scheduling instructions by the decision engine, implement verification on the candidate instructions in the edge sandbox, and collect feedback data in the execution period with short window high frequency sampling to evaluate the treatment effect.

2. The big data-based boiler industry data information transmission method according to claim 1, characterized in that: The specific steps of uniformly accessing sensing devices and generating a list of trusted devices are: Select an industrial edge gateway as the edge computing unit of the boiler field, uniformly access various sensing devices on the edge gateway, enter device basic information according to the pre-planned access form, automatically record the MAC address, hostname and assigned IP address of new devices by setting the edge gateway as a DHCP proxy, and obtain an identifiable device list; Guide the device to perform first registration and identity verification, and perform connectivity test and data sampling time synchronization verification, obtain data sources with clear source and consistent timestamp, assign a unique identifier to each device according to the unified naming rules and preset labels, record the maintainer and responsible department, and generate a list containing status identifier, security identifier and change record, generate a trusted device list for data access, monitoring and permission management.

3. The big data-based boiler industry data information transmission method according to claim 2, characterized in that: The specific steps of uniformly accessing sensing devices and generating a list of trusted devices also include: According to the entered device serial number and recorded network identity, perform uniqueness verification and address conflict detection in the registration library, initiate certificate issuance and signature verification process to the device, obtain the trusted identity certificate of the device, test network connectivity using ICMP protocol, record response delay and packet loss rate, obtain device reachability and protocol availability, perform data sampling on the device several times and perform variance analysis and timestamp consistency verification on the measurement value, evaluate data reporting quality and time synchronization deviation, isolate devices without trusted identity, not meeting the connectivity and data quality requirements, and manually review.

4. The big data-based boiler industry data information transmission method according to claim 1, characterized in that: The specific steps of real-time acquisition of boiler data and edge computing unit preprocessing are: The edge computing unit reads original measurement values in real time at a preset sampling frequency, obtains an original data stream with a local timestamp and a source device identifier, performs band-pass filtering, denoising and missing value interpolation processing on the original data stream, obtains a smooth and continuous time series signal, performs time alignment and unit normalization on the processed time series signal according to a sliding window, converts different dimensions into a unified unit and aligns sampling points, simultaneously calculates key statistical features of the data, and compares the key statistical features with a historical baseline of the device and a preset threshold in real time.

5. The big data-based boiler industry data information transmission method according to claim 4, characterized in that: The specific steps of the edge computing unit preprocessing further include: The processed time series and the extracted features are packaged into a standardized data frame, the specific physical quantity measured in the data frame is subjected to incremental difference encoding, the important score of the measurement point is calculated, and the safety, control importance and historical fault correlation are fused according to a preset weight, and the calculation formula of the important score is: ， wherein I represents the importance score of the measurement point, represents the security weight, represents the control weight, represents the failure relevance weight, represents the security of the measurement point, represents the control importance of the measurement point, represents the historical failure relevance of the measurement point.

6. The big data-based boiler industry data information transmission method according to claim 4, characterized in that: The specific steps of the edge computing unit preprocessing further include: By maintaining a baseline value and a corresponding version number for each measurement point in the edge computing unit, calculating the difference between the real-time sampling value and the reference baseline, and comparing it with the preset difference threshold, a determination result of whether an effective change is generated is obtained, a difference entry is constructed for the measurement point determined as an effective change, and a difference frame is formed in compact encoding, and a complete snapshot is triggered periodically, and the edge computing unit updates the local baseline value and the version number according to the confirmation returned by the cloud for the successfully received and verified difference frame, so that the baseline synchronization state of the edge and the cloud is consistent.

7. The big data-based boiler industry data information transmission method according to claim 1, characterized in that: The specific steps of the detection model analyzing the data, classifying the abnormal conditions and giving an early warning are: The pre-trained abnormality detection model is applied to the standardized data frame, a hybrid abnormality detection framework of edge and cloud cooperation is adopted, a convolutional autoencoder is applied to the edge computing unit to reconstruct the original time series window, the statistical quantity is calculated, and the reconstruction error and the statistical quantity are fused into an edge score according to a preset weight, a multivariate time series convolution network is used in the cloud to model the time series window from multiple measurement points, a time series abnormality score reflecting complex multivariate interaction is obtained, and the edge score, the cloud time series abnormality score and the measurement point important score are combined into an abnormality priority according to the weight, and the formula is: ， where P represents an anomaly priority, represents an edge score, represents a cloud temporal anomaly score, represents a measurement point importance score, represents an edge weight, represents a cloud weight, represents a measurement point weight.

8. The big data-based boiler industry data information transmission method according to claim 7, characterized in that: The specific steps of the detection model analyzing the data, classifying the abnormal conditions and giving an early warning further include: The abnormality priority P is compared with a preset threshold, when P≥0.8, it is determined as the highest priority level, when 0.6≤P<0.8, it is determined as the high priority level, when 0.4≤P<0.6, it is determined as the medium priority level, and when P<0.4, it is determined as the low priority level; an early warning message containing the measurement point identifier, the abnormal type, the confidence, the time window and the related context is automatically generated, and is sent to the field edge execution unit according to the priority, and the prediction result is converted into an executable operation and maintenance instruction. 9.The big data-based boiler industry data information transmission method of claim 1, wherein: The specific steps of automatically taking a scheduling decision and collecting and recording a cooperative closed loop are: The abnormal priority is mapped with a preset policy table, a decision engine is called to generate a to-be-executed scheduling instruction, a clear scheduling action candidate is obtained, the scheduling instruction is subjected to safety and constraint checking in an edge sandbox, including permission checking and physical constraint checking, an executable action set verified and within safety constraints is obtained, a corresponding responsible person is notified according to an execution level for verification, a standardized scheduling request is issued to an edge executor, a specific control action directly executable in the field is obtained, related measuring points are monitored in a shorter window in real time during execution, and instant feedback data for judging control effect is obtained.

10. The big data-based boiler industry data information transmission method according to claim 9, characterized in that: The specific steps of automatically taking scheduling decisions and collecting records of a collaborative closed loop further include: The execution result is written into an audit log, the data is automatically labeled in the cloud and archived into a training sample queue, real samples for subsequent model retraining are obtained, a rollback strategy is triggered for execution failure, the hit rate, disposal timeliness and manual intervention rate of various scheduling decisions are regularly summarized, the statistical results are included in an operation and maintenance index board, and a quantifiable performance evaluation and data basis for automatically fine-tuning decision parameters are obtained.

Images