A multi-protocol air compressor station early warning system

By using a multi-protocol unified access module, an edge data preprocessing unit, a hierarchical intelligent early warning engine, and an early warning closed-loop management module, the problems of inconsistent protocols, delayed threshold early warnings, and insufficient rule iteration in the air compressor station equipment monitoring system have been solved, thus achieving the stability of equipment operation and the synchronous consistency of early warning information.

CN123072178APending Publication Date: 2026-11-10GUANGDONG XINZHUAN ENERGY SAVING TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202611425992.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-15
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

The existing air compressor station equipment monitoring system has several problems, including inconsistent protocols for heterogeneous equipment, making it difficult to centrally utilize data; single threshold warnings, leading to delayed hazard identification; different sources of warning events at internal and external terminals, resulting in inconsistent collaborative handling; and a lack of closed-loop iteration in warning rules, leading to insufficient adaptability.

Method used

It employs a multi-protocol unified access module, an edge data preprocessing unit, a hierarchical intelligent early warning engine, an internal and external dual-end early warning push module, and an early warning closed-loop management module to achieve dynamic parsing of heterogeneous protocols, data quality identification, multi-dimensional fusion judgment, and closed-loop optimization of early warning rules.

Benefits of technology

It enables plug-and-play functionality for heterogeneous devices, reduces false alarm and false alarm rates, ensures the continuity of device operation, and achieves consistent synchronization of early warning data between enterprise and client ends, thereby improving the iterative adaptability of early warning rules.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123072178A_ABST
    Figure CN123072178A_ABST
Patent Text Reader

Abstract

This invention discloses a multi-protocol-based early warning system for air compressor stations, relating to the fields of industrial IoT data processing and equipment operation and maintenance technology. It includes a multi-protocol unified access module, an edge data preprocessing unit, a hierarchical intelligent early warning engine, an internal and external dual-end early warning push module, and an early warning closed-loop management module. This invention achieves data acquisition from heterogeneous devices through multi-protocol unified access, ensures data quality through the edge preprocessing unit, realizes advanced early warning through triple-logic parallel judgment, ensures information consistency through internal and external dual-end push, and continuously optimizes early warning strategies through closed-loop management. It achieves dynamic access of heterogeneous protocol devices, early identification of equipment degradation trends, synchronization of early warning events between internal and external ends, and closed-loop iteration of early warning rules, reducing false alarm and false negative rates and ensuring the reliability and operational efficiency of air compressor stations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of industrial Internet of Things (IoT) data processing and equipment operation and maintenance technology, and particularly to a multi-protocol air compressor station early warning system. Background Technology

[0002] An air compressor station typically consists of multiple devices such as an air compressor, dryer, and electrical control cabinet. It plays a crucial role in providing a stable supply of compressed air to the production line, and its operational stability directly affects the continuity of the production line and product quality. With the development of Industrial Internet of Things (IIoT) technology, modern air compressor stations are increasingly deploying various sensors for data acquisition, enabling fault warnings through real-time monitoring of equipment operating parameters. For example, by deploying pressure, temperature, and vibration sensors in key components such as air compressors, air tanks, and pipelines, multi-source operating data is collected and uploaded to a monitoring platform for analysis and processing, achieving remote sensing of equipment status and anomaly identification.

[0003] Existing patent literature contains several technical solutions related to the monitoring and early warning of air compressor stations and related equipment. For example, Chinese patent application CN122024450A discloses an automatic early warning system for air compressor stations based on an air-bearing shaft. This system monitors the air supply pressure, bearing temperature, vibration amplitude, and rotational speed of the air-bearing shaft, calculates a comprehensive risk coefficient, and issues an alarm when a threshold is exceeded, thus achieving focused monitoring of the air-bearing shaft drive components. Chinese patent application CN121635227A discloses an air compressor station operation and maintenance method based on a GRU model. By collecting multi-source data and using the GRU model to predict the future operating status of the equipment, it achieves intelligent operation and maintenance of the air compressor station. Chinese patent application CN121720060A discloses a method for detecting pipeline leaks in air compressor stations based on dynamic pressure and multi-branch pipe pressure. By collecting pressure and flow data from the main pipe and each branch pipe, and combining a resistance model with machine learning algorithms, it achieves accurate detection and location of pipeline leaks.

[0004] However, the aforementioned technical solutions for monitoring air compressor stations and related equipment still have shortcomings in practical applications, specifically in four dimensions: protocol access, early warning mechanism, data synchronization, and rule optimization. In terms of protocol access, the communication protocols of devices from different manufacturers cover a variety of types, such as Modbus RTU / TCP, MQTT, OPC UA, and PLC private protocols. Traditional systems only support single protocol access or fixed protocol adaptation, which cannot flexibly adapt to the different communication needs of multiple types of devices in the whole station. It is difficult to achieve unified access and centralized collection of all devices in the station, resulting in the isolation of the operating data of each device and the formation of data silos.

[0005] In terms of fault early warning analysis, existing early warning systems rely solely on static fixed thresholds to determine faults and trigger alarms. They can only issue alarms when parameters reach critical values, making it difficult to identify slow degradation, gradual anomalies, and multi-device coupled anomalies in advance. Furthermore, using a unified judgment logic under different data quality conditions can easily lead to false alarms or missed alarms due to acquisition jitter, abnormal noise, or missing timing information.

[0006] Regarding data synchronization, existing technologies only allow early warning information to circulate within a closed loop on the enterprise's internal platform, without establishing a data synchronization mechanism. Customers cannot obtain completely consistent early warning events and equipment health data from the same source as the enterprise's internal data, making it impossible to monitor equipment malfunctions and potential failures in real time. Due to this information asynchrony, equipment users cannot take preventative measures in advance. When equipment malfunctions suddenly, there is a lack of buffer time to adjust production schedules, leading to unplanned shutdowns and reactive responses, impacting the efficiency and predictability of external collaboration.

[0007] In terms of rule optimization, the existing system makes low use of historical events and their outcomes, and lacks a mechanism to extract the basis for rule adjustment from historical false alarm samples and valid samples. This makes rule maintenance more reliant on human experience and difficult to iterate continuously. Summary of the Invention

[0008] The purpose of this invention is to provide a multi-protocol air compressor station early warning system to solve the problems in the existing technology, such as inconsistent protocols of heterogeneous equipment leading to difficulty in centralized data utilization, single threshold early warning leading to delayed hazard identification, different sources of early warning events at internal and external ends leading to inconsistent collaborative handling, and lack of closed-loop iteration of early warning rules leading to insufficient adaptability.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A multi-protocol-based air compressor station early warning system includes a multi-protocol unified access module, an edge data preprocessing unit, a hierarchical intelligent early warning engine, an internal and external dual-end early warning push module, and an early warning closed-loop management module.

[0010] The multi-protocol unified access module is configured with a protocol library, a driver manager, a protocol identification engine, and a protocol conversion engine, and is used to access air compressor station field devices that use different communication protocols; the driver manager adopts a plug-in architecture, supports dynamic loading of protocol drivers at runtime, and converts heterogeneous protocol data into a unified object model format. The edge data preprocessing unit is connected to the multi-protocol unified access module and is used to sequentially perform data denoising, outlier removal and breakpoint completion on the unified object model data, generate processed data carrying data quality identifiers and cache them locally. The hierarchical intelligent early warning engine is connected to the edge data preprocessing unit. It is used to perform fixed threshold judgment, trend judgment, and cross-device correlation judgment in parallel within a preset time window, output multiple candidate early warning events, and merge the candidate early warning events through preset fusion rules to determine the final early warning event and its corresponding early warning level. The fusion rules calculate a fusion score based on the level and number of hits of each judgment logic. When multiple judgment logics are hit at the same time, the fusion score is calculated comprehensively according to preset weights, and the final early warning level is improved when the improvement conditions are met. Among them, the data corresponding to different data quality labels have different confidence weights. The internal and external dual-end early warning push module is connected to the hierarchical intelligent early warning engine and is used to generate a unified event identifier for the merged early warning events, and push early warning information to the internal end of the enterprise and the client end simultaneously based on the identifier. The early warning closed-loop management module is connected to the hierarchical intelligent early warning engine. It is used to record early warning events and handling results, and to filter false alarm samples and valid samples from historical events to generate candidate rule update items. After manual review, the updated items are fed back to the hierarchical intelligent early warning engine.

[0011] As a preferred embodiment of this application, the specific workflow of the multi-protocol unified access module includes: When the driver manager starts, it scans the protocol plugin directory. When a new device is detected, it matches the corresponding driver from the protocol library based on the protocol identifier reported by the device. If the match is successful, the driver is loaded into memory using the class loading mechanism. If the match fails, the protocol recognition engine is triggered to perform message feature recognition. The protocol recognition engine adopts a three-level recognition strategy: the first level performs accurate matching based on frame header, length and function code; the second level extracts message statistical features and inputs them into a classifier for protocol prediction; the third level writes unrecognizable messages into a queue to be labeled. The protocol conversion engine maps different protocol fields to a unified object model according to the object model mapping rules, converts values ​​with different units and aligns timestamps uniformly, and outputs structured data packets.

[0012] As a preferred embodiment of this application, the specific processing logic of the edge data preprocessing unit includes: A sliding window filtering method is used to perform median smoothing on continuous sampled values ​​to eliminate spike noise caused by impulse interference; Using the statistical mean and standard deviation of the K most recent valid sampling points (excluding points marked as outliers) as a reference benchmark, outlier data points exceeding the preset outlier threshold are removed. When the time difference between adjacent sampling exceeds a preset threshold, a linear interpolation algorithm is triggered to complete the breakpoint. When the network is interrupted, the data is written to the local persistent cache. After the network is restored, the interrupted transmission is resumed in the order of timestamps, and the corresponding cache records are cleared after the data transmission is completed. The processed data is labeled with quality tags, and the tag categories include at least normal, noise-reduced, completed, and missing.

[0013] As a preferred embodiment of this application, the determination and fusion logic of the hierarchical intelligent early warning engine includes: The hierarchical intelligent early warning engine first parses the data quality identifier carried in the processed data, and then executes a differentiated data acceptance strategy based on the data quality identifier: for data with a normal quality code or that has undergone noise reduction, it participates in all judgment logic as normal; for data with a quality code that has been interpolated and completed, the confidence weight is reduced; for data with a quality code that is missing, the judgment at that time point is skipped. The fixed threshold determination is based on the following sources: equipment nameplate parameters, historical safe operating range statistics, and operation and maintenance experience settings. The trend determination is achieved by calculating the linear regression slope of the sampling points within a preset time window. If multiple consecutive windows satisfy the positive growth and the slope exceeds the preset rate of change, a trend warning is triggered. The association determination is based on the "if A and B, then C" form rule generated by historical event mining, which is triggered when multiple indicators meet the conditions simultaneously within the current time window; The result fusion rule is configured as follows: if only one judgment is hit, the corresponding level is output; if multiple judgments are hit at the same time, the fusion score is calculated according to the preset weight. When the fusion score reaches the improvement threshold and the number of hit judgments meets the preset conditions, the level is increased by one level based on the highest hit level; if an emergency alarm condition is involved, the emergency alarm level is directly output. Multiple candidate events triggered by the same device, the same time window, and the same anomaly source are merged into a single event ID to avoid duplicate alarms.

[0014] As a preferred embodiment of this application, the information processing mechanism of the internal and external dual-end early warning push module includes: Information pushed to clients is automatically filtered of internal processing remarks fields, retaining only the de-identified status information; The warning information received by the enterprise's internal end and the client end share the same event ID, ensuring that the two ends have a consistent understanding of the same fault event.

[0015] As a preferred embodiment of this application, the rule iteration mechanism of the early warning closed-loop management module includes: Distinguish between valid alarm samples and false alarm samples, and automatically generate candidate rules for threshold adjustment or correlation logic optimization based on the feature distribution of false alarm samples; All candidate rules must be manually reviewed and confirmed before being written into the rule base and then hot-updated and synchronized to the hierarchical intelligent early warning engine.

[0016] Compared with the prior art, the present invention has the following beneficial effects: By dynamically loading protocol plugins through the built-in plug-in driver manager of the multi-protocol unified access module, plug-and-play and dynamic parsing of heterogeneous protocol devices are realized, avoiding the overall system reconstruction caused by adding new device types and effectively reducing the cost of system access transformation.

[0017] Standardized data carrying data quality identifiers is generated by the edge data preprocessing unit. Combined with the multi-dimensional fusion judgment logic and dynamic weight fusion scoring mechanism of the hierarchical intelligent early warning engine, low-quality dirty data interference is effectively filtered out. It can identify equipment degradation trends in advance and accurately locate cross-equipment related faults, thereby significantly reducing false alarm rate and false alarm rate and ensuring the continuity of equipment operation.

[0018] By using a dual-end early warning push module to generate a unified event identifier and desensitization mechanism for early warning events, consistent synchronization of early warning data is achieved between the enterprise and client ends. Under the premise of ensuring data security, the problem of external response delay caused by information asymmetry is eliminated.

[0019] The monitoring-handling-feedback-optimization closed-loop mechanism established through the early warning closed-loop management module enables the system to automatically discover rule defects and generate optimization suggestions based on all historical data, breaking away from the limitations of traditional manual experience-based parameter tuning and continuously improving the pertinence of early warning rule iteration. Attached Figure Description

[0020] Figure 1 This is a modular architecture diagram of a multi-protocol air compressor station early warning system according to the present invention; Figure 2 This is a flowchart of the multi-protocol unified access module of a multi-protocol air compressor station early warning system according to the present invention; Figure 3 This is a flowchart of the edge data preprocessing unit of a multi-protocol air compressor station early warning system according to the present invention; Figure 4 This is a flowchart of the hierarchical intelligent early warning engine based on a multi-protocol air compressor station early warning system according to the present invention; Figure 5 This is a flowchart of the internal and external dual-end early warning push module of the multi-protocol air compressor station early warning system of the present invention; Figure 6 This is a flowchart of the early warning closed-loop management module of an early warning system based on a multi-protocol air compressor station according to the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the following embodiments are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention. All equivalent substitutions or modifications made based on the concept of the present invention should fall within the scope of protection of the present invention.

[0022] Example 1 like Figure 1 As shown, the present invention provides a multi-protocol air compressor station early warning system, including a multi-protocol unified access module, an edge data preprocessing unit, a hierarchical intelligent early warning engine, an internal and external dual-end early warning push module, and an early warning closed-loop management module.

[0023] The multi-protocol unified access module is configured with a protocol library, a driver manager, a protocol identification engine, and a protocol conversion engine, and is used to access air compressor station field devices that use different communication protocols. The driver manager adopts a plug-in architecture, supports dynamic loading of protocol drivers at runtime, and converts heterogeneous protocol data into a unified object model format. The edge data preprocessing unit is connected to the multi-protocol unified access module and is used to sequentially perform data denoising, outlier removal and breakpoint completion on the unified object model data, generate processed data carrying data quality identifiers and cache them locally. The hierarchical intelligent early warning engine is connected to the edge data preprocessing unit. First, it parses the data quality identifier carried by the processed data and executes a differentiated data acceptance strategy based on the data quality identifier. Then, it performs fixed threshold determination, trend determination and cross-device correlation determination in parallel within a preset time window, outputs multiple candidate early warning events, and completes the merging of candidate early warning events through preset fusion rules to determine the final early warning event and its corresponding early warning level. The internal and external dual-end early warning push module is connected to the hierarchical intelligent early warning engine and is used to generate a unified event identifier for the merged early warning events, and push early warning information to the internal end of the enterprise and the client end simultaneously based on the identifier. The early warning closed-loop management module is connected to the hierarchical intelligent early warning engine. It is used to record early warning events and their handling results, and automatically generate rule update items based on historical data to iteratively optimize the early warning strategy.

[0024] As a preferred embodiment of this application, the specific architecture and workflow of the multi-protocol unified access module are as follows: like Figure 2 As shown, the multi-protocol unified access module includes a protocol library, a driver manager, a protocol identification engine, and a protocol conversion engine.

[0025] The protocol library uses a relational database (such as MySQL) to store the metadata configuration information of the protocol drivers. Specifically, the data tables in the protocol library store the driver metadata for each protocol in a structured manner. Each driver metadata record includes at least: protocol name (such as Modbus TCP), version number, packet header characteristics (such as a specific hexadecimal frame header), verification method (such as CRC16), field parsing rules, and object model mapping rules. This protocol library serves as the underlying baseline configuration of the system, providing data support for subsequent automatic protocol identification and data parsing.

[0026] As an equivalent alternative, the protocol library can also use an in-memory database such as Redis or a local configuration file (such as XML or JSON format) to store the metadata configuration information, in order to adapt to deployment requirements under different computing power conditions.

[0027] The driver manager adopts an SPI (Service Provider Interface) plug-in architecture, decoupling the protocol parsing logic from the main program by defining standardized interface specifications, thereby achieving dynamic lifecycle management of the protocol driver. Each protocol plug-in is an independent software module that encapsulates the complete parsing logic of a single communication protocol, containing at least a protocol parser, encoder, field mapping rules, and metadata description file. The protocol plug-in exists as a compiled binary file (such as a .jar file in a Java environment or a .dll file in a .NET environment). Taking the Modbus RTU protocol as an example, its protocol plug-in includes: frame structure parsing logic (disassembly rules for address code, function code, register address, and CRC check); object model mapping rules (mapping register address 0x0001 to exhaust pressure attribute); and numerical conversion rules (multiplying the original register value by 0.01 to convert it to a physical quantity value).

[0028] The workflow of the driver manager includes: When the system starts up, the driver manager automatically scans the driver files (such as .jar or .dll files) in the local plugin directory. When a new device access request is detected, the protocol identifier reported by the device is read. The protocol identifier is an index key (e.g., a specific protocol type code) reported by the device during the handshake phase to uniquely characterize its communication features. The driver manager performs index matching in the protocol library; If a match is successful, the corresponding protocol driver is dynamically loaded into memory and instantiated using a class loading mechanism (such as Java ClassLoader), enabling it to parse the device data; if a match fails, the protocol recognition engine is triggered to perform automatic recognition.

[0029] As an equivalent alternative, the class loading mechanism can also be implemented using the OSGi modular framework or the .NET Assembly.LoadFrom method to achieve dynamic loading and isolation of drivers.

[0030] The protocol identification engine is used to solve the access problem of unknown or proprietary protocols, and it executes a three-level progressive identification strategy: The first level is exact matching, which extracts the frame header bytes, fixed-length fields and function codes of the data stream, and performs a hash comparison with the message header features in the protocol library. If the hash values ​​match, it is directly determined to be the corresponding protocol type. The second level is machine learning prediction. For fuzzy data streams without fixed features, the statistical feature vector of the message (including frame length variance, byte entropy distribution, and field change frequency) is extracted and input into a pre-trained classifier (such as a random forest classifier) ​​for probability prediction. If the prediction confidence is higher than the set threshold, it is determined to be the corresponding protocol type. As an equivalent alternative, the classifier can also be implemented using decision trees, support vector machines, or lightweight neural networks (such as MobileNet).

[0031] The third level is dynamic learning and inclusion. If the first two levels cannot identify the unknown protocol frame, it is written to the annotation queue (such as the Kafka message queue) along with the source device IP and timestamp for subsequent manual annotation and model iteration optimization.

[0032] After the driver is successfully loaded, the protocol conversion engine parses the original data frame according to the field parsing rules in the protocol library. The conversion process includes: first, stripping the frame header, frame tail and check redundancy bits, and extracting the original sensor values; then, mapping the private fields of different protocols (such as specific register addresses of Modbus) to unified standard object model attributes according to the object model mapping rules.

[0033] The object model mapping rule is a pre-configured field correspondence table used to define the conversion logic between private data fields of a specific protocol and attributes of a standard object model. Each mapping rule includes at least: source field identifier (i.e., field address or name in the original protocol), target attribute identifier (i.e., attribute name in the standard object model), data type conversion method (e.g., integer to floating-point), and unit conversion factor (e.g., multiplying the original value by 0.1 to convert to degrees Celsius).

[0034] For example, in the Modbus protocol, the mapping rule can be defined as follows: the 16-bit integer raw value read from register address 40001 is multiplied by a conversion factor of 0.1 and then mapped to the field named "exhaust temperature" in the standard object model, with the unit labeled in °C. In the MQTT protocol, the mapping rule can be defined as follows: the floating-point value with the key "pressure" in the JSON message is directly mapped to the field named "exhaust pressure" in the standard object model, with the unit labeled in MPa. Through this mapping rule, the protocol conversion engine can complete the standardized conversion of heterogeneous data in a configuration-driven manner, without the need to write hard-coded logic for each protocol.

[0035] The object model mapping rules adopt a hot-loading mechanism. When a new device type is added or the protocol version is updated, the operation and maintenance personnel can add or modify the mapping rules through the management interface, and the system can take effect without downtime, thereby enhancing the scalability of the system.

[0036] The standard object model is a unified data format used to eliminate device heterogeneity. Its field arrangement rules adopt a standardized key-value pair structure, including at least the following core fields: Device ID (used to globally and uniquely identify the device), Point ID (representing the specific sensor location), Unit, Timestamp (uniformly aligned to milliseconds), Value (a standardized value after unit conversion), and Quality Code (used to indicate whether the data has undergone noise reduction or interpolation). Finally, all parsed data is encapsulated into a data packet output in a preset format (such as Protobuf binary format), thus providing standardized data input for downstream edge preprocessing units.

[0037] The data packet encapsulation format is not limited to Protobuf. As equivalent alternatives, Avro, MessagePack, or JSON formats can also be used. The choice of different formats depends on the parsing capabilities and bandwidth constraints of the downstream system.

[0038] As a preferred embodiment of this application, the specific processing logic of the edge data preprocessing unit is as follows: like Figure 3 As shown, the edge data preprocessing unit receives unified object model data output by the multi-protocol unified access module, and performs the following preprocessing operations on the input data in sequence: 1. Data noise reduction processing: During operation, air compressor station equipment is prone to transient pulse interference, causing pulse-type noise with amplitudes that significantly deviate from the true values ​​in time-series data such as temperature, pressure, and vibration. If this noise is directly involved in subsequent early warning determination, it may lead to false triggering and affect system reliability. Therefore, this embodiment uses a sliding window midpoint filtering algorithm to smooth the continuously sampled values.

[0039] Specifically, the basic principle of the sliding window median filtering is as follows: a fixed-length data window is maintained on the time series data. The window slides as new data arrives, and the median of all data within the window is used as the output value at the current moment. The median statistic is insensitive to impulse noise: even if there are one or more extreme deviations within the window, the median can still stably reflect the central trend of the data. The system maintains a first-in-first-out (FIFO) sliding window of length N (N=5 in this embodiment). When a new data point enters, the median of the data within the window is calculated as the filtered output, and this median replaces the original value at the current moment.

[0040] The sliding window length N is a configurable parameter, and its value affects the balance between filtering strength and response sensitivity. A larger N value results in a stronger filtering effect but also a greater signal response delay, while a smaller N value results in a faster response but a weaker noise suppression capability. In this embodiment, the preferred range for N is 3 to 7. In high-frequency vibration sampling scenarios, N can be configured to 3; in low-frequency temperature sampling scenarios, N can be configured to 7.

[0041] For example, assume the original sampling sequence of the air compressor exhaust temperature sensor is [93.2, 93.5, 98.7, 93.6, 93.4], and the window length N=5. The system sorts the data within the window to [93.2, 93.4, 93.5, 93.6, 98.7], takes the median 93.5 to replace the original value 98.7 at the current moment, and outputs the filtered sequence as [93.2, 93.5, 93.5, 93.6, 93.4], where 98.7 is identified as impulse noise and discarded.

[0042] 2. Outlier removal: Even after noise reduction, outliers may still exist in the data due to transient sensor failures, momentary communication interference, or non-steady-state operation of equipment. If such outliers are included in subsequent statistical analysis, they will affect the accuracy of trend determination. Therefore, this embodiment employs a statistical method based on the 3σ criterion to identify and remove outliers.

[0043] The basic principle of the 3σ criterion is as follows: For a dataset that follows a normal or approximately normal distribution, the probability that its data values ​​fall within the mean μ ± 3σ range is approximately 99.73%. Values ​​outside this range can be considered low-probability events and identified as outliers. The system uses the most recent K valid sampling points (i.e., valid data points that do not include those marked as outliers) as the baseline sample set, calculates the mean μ and standard deviation σ of the data within the current time window, and marks values ​​outside the μ ± 3σ range as outliers.

[0044] For example, for the exhaust pressure data of an air compressor station within a certain time window, if μ = 0.75 MPa and σ = 0.02 MPa are calculated, then the confidence interval for μ ± 3σ is [0.69, 0.81] MPa. When the pressure value collected at a certain moment is 0.85 MPa, since 0.85 MPa > 0.81 MPa, it exceeds the confidence interval, and the system marks it as an outlier and removes it, not using it as input for subsequent early warning judgment.

[0045] 3. Breakpoint completion handling: In an Industrial Internet of Things (IIoT) environment, data loss at certain sampling points may occur due to factors such as instantaneous fluctuations in communication networks, temporary failures of sensor acquisition units, or connection timeouts between edge gateways and devices. Directly performing trend analysis with incomplete time-series data will affect the reliability of the calculation results. Therefore, this embodiment performs breakpoint completion processing on missing data points.

[0046] Specifically, the system monitors the timestamp difference Δt between adjacent valid sampling points in real time. If Δt exceeds a preset sampling period threshold, a data breakpoint is determined to have occurred. If the number of missing points is less than or equal to a preset threshold (the threshold is set to 3 in this embodiment), the system uses linear interpolation to complete the data.

[0047] The basic principle of linear interpolation is as follows: assuming that missing data exhibits a linear trend between its preceding and following valid data points, the missing value is estimated between the two points based on the time ratio and marked as imputed by interpolation. The formula for calculating the imputed value is:

[0048] in, For missing moments, and For the effective sampling times before and after, and This represents the corresponding valid value. If the number of missing consecutive points exceeds a preset threshold, it is marked as missing data.

[0049] 4. Data quality identification, local caching, and resume interrupted downloads: After preprocessing, the system adds a data quality identifier and a local persistent caching mechanism to each data point.

[0050] Specifically, the data quality identifier is a status marker byte automatically added by the system for each preprocessed data point, used to characterize the processing history experienced by that data point. The quality code includes: 0x00: Normal data, meaning the original data has not undergone any correction or processing, and has the highest reliability; 0x01: Noise reduction processing, i.e., the data points have been smoothed by median filtering; 0x02: Completed by interpolation, meaning the data points are estimated and generated using a linear interpolation algorithm; 0x03: Data missing, meaning that the number of consecutive missing points exceeds the preset threshold, and the system abandons the data completion.

[0051] The quality identifier (quality code) and standardized data are encapsulated and transmitted to the hierarchical intelligent early warning engine. When the engine receives data carrying different quality codes, it executes a differentiated data acceptance strategy based on the quality code: Quality code 0x00: Marked as high-confidence data, it will normally participate in all triple warning judgment logic, with a confidence weight of 100%; Quality code 0x01: Marked as filtered data, it will participate normally in all judgment logic, with a confidence weight of 100%; Quality code 0x02: Marked as estimated data, the confidence weight is reduced to 80% when participating in the judgment, and an alarm can only be triggered after confirmation by 3 consecutive sampling points in the fixed threshold judgment. Quality code 0x03: Marked as invalid data, skipping the judgment at this time point and not participating in any warning logic.

[0052] At the same time, data quality labeling also provides a basis for subsequent data traceability and problem investigation.

[0053] When network communication is interrupted, the edge gateway cannot upload preprocessed data to the central server in real time. In this situation, the system continuously writes the preprocessed standardized data to local persistent storage media (such as SSD storage or eMMC chips) to avoid data loss. The persistent storage employs a first-in, first-out (FIFO) circular queue management strategy. When cache space is insufficient, the earliest timestamp data record is overwritten first to ensure the continuity of new data storage. Specifically, when data is written, the system appends it in timestamp order; when the cached data volume reaches a preset capacity limit (e.g., 1GB), the system automatically cleans up historical records that have been re-uploaded to free up space; if the network interruption lasts too long and the cache is full, the earliest unuploaded record is overwritten in chronological order.

[0054] Once the system detects that network communication has returned to normal, it automatically triggers the retransmission process: the system first reads the record with the earliest timestamp in the local offline cache that has not yet been retransmitted, and uploads it to the central server in a first-in-first-out order. After each piece of data is successfully uploaded and a confirmation response is received from the server, the corresponding cache record is cleared. This process is repeated until all cached data has been processed.

[0055] As a preferred embodiment of this application, the determination and fusion logic of the hierarchical intelligent early warning engine is as follows: like Figure 4 As shown, the hierarchical intelligent early warning engine uses a preset time window as a unified analysis interval, and its processing flow includes three stages: data acceptance and quality response, triple parallel early warning judgment, candidate event merging and result fusion.

[0056] 1. Data Acceptance and Quality Response: When the hierarchical intelligent early warning engine receives data packets from the edge data preprocessing unit, it first parses the quality code field attached to each data point and executes a differentiated data acceptance strategy based on the quality code value. For data points with a quality code of 0x00 (normal), the engine marks them as high-confidence data and allows them to participate normally in all triple early warning judgment logic, with a confidence weight of 100%. For data points with a quality code of 0x01 (after noise reduction), the engine marks them as filtered data and allows them to participate normally in all judgment logic, with a confidence weight of 100%. However, the processing source of this data point is recorded separately (marked as source: filtered output) for subsequent source tracing analysis. For data points with a quality code of 0x02 (after interpolation completion), the engine marks them as estimated data, with a confidence weight when participating in the judgment logic. The weighting is reduced to 80%. Specifically, in trend analysis, the weight of this data point in slope calculation is halved; in fixed threshold determination, this data point does not trigger an alarm independently, but only when the parameter values ​​of three consecutive sampling points of the same device meet the threshold condition, in order to avoid false alarms caused by small deviations of a single interpolation point; for data points with a quality code of 0x03 (data missing), the engine marks them as invalid data, skips the determination of this time point, and does not participate in any alarm logic; this time point is recorded as a blank mark on the time series, and alarm triggering requires confirmation from the next valid data point.

[0057] Through the above mechanism, the hierarchical intelligent early warning engine can adopt differentiated judgment strategies for data of different quality, effectively reducing false alarms caused by data noise or completion errors while ensuring early warning sensitivity.

[0058] 2. Triple early warning and parallel judgment: Fixed threshold determination: The system has four built-in threshold levels for each operating parameter: alert threshold, warning threshold, alarm threshold, and emergency alarm threshold. The system compares the real-time collected parameter values ​​with these four threshold levels sequentially. When the measured value exceeds a certain threshold level but does not reach the next higher threshold level, an alarm of the corresponding level is triggered. The threshold sources include the equipment nameplate calibration parameters, historical safe operating ranges, and manually set values. When multiple sources exist for the same parameter, the system prioritizes the value according to the following order: equipment nameplate calibration parameters > historical safe operating range statistics > manually set values. The equipment nameplate calibration parameters have the highest priority, and manually set values ​​are only used as temporary configurations when no historical data is available.

[0059] Trend analysis and determination: Based on the K most recent sampling points within a preset time window (K=10 in this embodiment), a data sequence is constructed, and the least squares method is used to calculate the linear regression slope k. The calculation formula is as follows:

[0060] in, Sampling time, Here are the measured values ​​of the parameters at the corresponding time points, and n is the number of sampling points.

[0061] If the calculated slope k is positive and exceeds the preset upper limit of the normal rate of change k th (In this embodiment, the exhaust temperature k) th =0.3℃ / s, exhaust pressure k th If the slope of the sample is positive for M consecutive times (M=3 in this embodiment) and the slope is 0.02MPa / s, then it is determined to be an abnormal upward trend, triggering a trend warning.

[0062] When the trend determination hits independently (i.e., neither the fixed threshold determination nor the association rule determination hits), the system determines the warning level based on the multiple by which the slope exceeds the upper limit of the normal rate of change: if the multiple exceeds ≤1.5 times, it is determined as a warning level; if the multiple exceeds >1.5 times and ≤2.5 times, it is determined as an alarm level; and if the multiple exceeds >2.5 times, it is determined as an emergency alarm level. This ensures that the trend determination can independently output a clear warning level in a single hit scenario.

[0063] Through the above trend analysis, the system can identify abnormal trends in advance even before the parameters reach a fixed threshold, thus realizing the transformation from post-event alarms to pre-event warnings.

[0064] Association rule determination: Built-in association rule library, each rule is represented by the production rule of IF condition1 AND / OR condition2 THEN conclusion.

[0065] For example, the following rules are preset in this embodiment: Rule R001: If the exhaust pressure is <0.5MPa and the motor current is >120% of the rated current, then the system is considered to be leaking through a pipeline or malfunctioning a valve, triggering an alarm level warning. Rule R002: If housing vibration > 4.5 mm / s AND exhaust temperature > 85℃, then it is determined to be abnormal bearing wear, triggering an alarm level warning. Rule R003: If the dew point temperature is >0℃ and the dryer inlet air flow is normal, then the dryer regeneration is determined to be faulty, triggering a warning level alert. Rule R004: If voltage fluctuation > ±15% AND current fluctuation > ±20%, then a power supply abnormality is determined, triggering an emergency alarm level warning.

[0066] A corresponding alert will only be triggered when multiple indicators simultaneously meet the preconditions within the same preset time window. The rules in the association rule base are dynamically updated by the alert closed-loop management module based on historical fault cases and handling feedback results.

[0067] 3. Candidate event merging and result fusion: Furthermore, the hierarchical intelligent early warning engine performs candidate early warning event merging and result fusion: Multiple candidate events generated from the same device, within the same preset time window, and from the same root cause of an anomaly are merged into a single entry and assigned a unique event identifier. The criteria for determining the same root cause of an anomaly are: the triggering conditions of multiple candidate events point to the same device component or the same fault mechanism.

[0068] After the data is merged, the system determines the final warning level according to the following standardized result fusion rules: Rule 1: If only a single judgment logic is hit, directly output the original warning level corresponding to that judgment; Rule 2: If two or more judgment logics are triggered simultaneously, the system calculates a comprehensive risk score based on preset weighting coefficients and data quality indicators. The fusion scoring formula is as follows:

[0069] in, For the original risk score output by the i-th judgment logic, when the fixed threshold judgment hits, it is mapped according to the score corresponding to the four-level threshold (30 points for prompt, 50 points for warning, 75 points for alarm, and 100 points for emergency alarm); when the trend judgment hits, it is linearly mapped to the 0~100 score range according to the multiple of the slope exceeding the normal rate of change; when the correlation judgment hits, it is mapped to the 0~100 score range according to the normalized value of the ratio of the number of hit rules to the total number of preset rules. The basic weight coefficients corresponding to the i-th judgment logic are (the fixed threshold judgment weight is 0.4, the trend judgment weight is 0.35, and the correlation judgment weight is 0.25). This is the quality code correction coefficient corresponding to the i-th decision logic (0x00 is 1.00, 0x01 is 0.95, 0x02 is 0.80, and 0x03 is not included in the fusion). When the decision logic fails, the corresponding... The value is 0. If the fusion score... And the number of hits. 2. If any path determines that an emergency alarm condition has been met, the system will increase the current highest hit level by one level (for example, if the highest hit level is an alarm, it will be increased to an emergency alarm; if the highest hit level is a warning, it will be increased to an alarm). If the highest hit level is already an alarm level, it will be increased to an emergency alarm level. For example, when a certain early warning event simultaneously triggers both a fixed threshold determination (early warning level) and a trend determination, if the basic weighting coefficient of the fixed threshold determination... The basic weighting coefficient for trend determination ,corresponding And the quality code correction coefficients corresponding to the two judgments are both (Corresponding to quality code 0x00 or 0x01), then the fusion score is: .because and The system outputs a warning level.

[0070] Furthermore, when the fixed threshold judgment hits the alarm level (S1=75) and the trend judgment hits simultaneously (S2=80), and both judgments use data with a 0x00 quality code ( When ), the fusion score is: .because And the number of hits. 2. The system will upgrade the final warning level from the alarm level to the emergency alarm level.

[0071] Rule 3: If any judgment logic hits the emergency alarm condition, the emergency alarm level will be output directly, and the above-mentioned fusion scoring calculation will not be performed.

[0072] As a preferred embodiment of this application, the information processing mechanism of the internal and external dual-end early warning push module is as follows: like Figure 5As shown, the internal and external dual-end early warning push module receives early warning events generated by the hierarchical intelligent early warning engine. Each early warning event carries a unified event identifier (event ID). This unified event ID serves as the unique association key between the internal end and the client for that early warning event. Although the push messages from both ends differ in content anonymization, they share the same event ID, allowing the client's progress query and the internal update status to be associated based on the same event ID during subsequent processing. Based on this early warning event, the module derives two parallel push messages: The push module reads the complete basic fields of the early warning event and standardizes the push content into a unified data structure. The data structure includes at least the following fields: event ID (string type, globally unique), device identifier (string type), early warning level (enumeration type, values ​​are prompt / early warning / alarm / emergency alarm), trigger reason (string type, records the trigger logic type and specific conditions), occurrence time (timestamp type, accurate to milliseconds), and handling suggestion (string type, generated by matching the preset handling suggestion template according to the early warning level and fault type).

[0073] For internal enterprise users, the module matches the corresponding push method (including one or more of APP push, SMS, platform pop-up, and voice call) according to the warning level, and displays all raw data fields including internal maintenance notes and maintenance personnel information.

[0074] For the client, the module pushes targeted de-identified messages to the client's pre-configured receiver. The de-identification process employs a field-level filtering mechanism: the system iterates through each field in the complete data packet of the warning event based on the field whitelist in the client's push message template, retaining only the fields defined in the whitelist and discarding non-public fields such as internal handling notes and maintenance personnel information; the field whitelist at least includes the event ID, device parameters, fault cause, and general handling suggestions. For warning events with a warning level higher than a preset threshold, the module simultaneously triggers an SMS push, sending reminder information to the client's pre-configured contacts to ensure that high-level warnings reach key personnel on the client's side.

[0075] For example, when the hierarchical intelligent early warning engine generates an early warning event regarding an abnormal rise in the exhaust temperature of air compressor #1, the system assigns a globally unique event ID to the event, such as EVT20260617001. The enterprise's internal terminal receives a complete message containing this event ID, with the following content: EVT20260617001—Air compressor #1—Early warning level: Alarm—Trigger reason: Exhaust temperature rises from 82℃ to 96℃ within 10 seconds (slope 0.47℃ / s)—Occurrence time: 2026-06-17 14:30:25—Handling suggestion: Check if the cooler fins are blocked—Internal maintenance notes: Last maintenance time was 2026-06-01—Maintenance personnel: Zhang San, Li Si. At the same time, the client received the de-identified message: EVT20260617001—1# Air Compressor—Warning Level: Alarm—Fault Cause: Abnormal rise in exhaust temperature, it is recommended to check the cooling system—General Handling Recommendation: Contact maintenance personnel for on-site confirmation. Both ends share the event ID EVT20260617001, and the client can use this ID to query the system for subsequent handling progress.

[0076] As a preferred embodiment of this application, the rule iteration mechanism of the early warning closed-loop management module is as follows: like Figure 6 As shown, the early warning closed-loop management module relies on a unified event identifier (event ID) to connect early warning data across the entire chain. The specific workflow includes: Full-process recording: When the hierarchical intelligent early warning engine triggers the merging of early warning events and outputs a standard early warning event carrying a unified event identifier (event ID), the early warning closed-loop management module synchronously captures the event ID and automatically archives the complete event record using the event ID as the primary key. The basic fields of the early warning event, including event ID, trigger time, trigger rule, and device information, are already... Figure 5 The data structure of the internal and external dual-terminal early warning push module is defined as shown, and will not be repeated here. Based on this, this module supplements the following handling management fields: handling status (enumeration type, values ​​are pending order / processing / pending acceptance / completed), and handling conclusion (enumeration type, values ​​are real fault / false alarm / to be observed / cannot be reproduced). After the maintenance personnel complete the on-site handling through the enterprise internal terminal, they associate the corresponding event ID, fill in the handling information (including the person responsible for handling, action description, fault cause classification, and recovery time), and upload supporting materials.

[0077] Multi-dimensional data analysis: The system automatically performs full-volume historical event data analysis according to preset time periods (three time granularities in this embodiment: natural day, natural week, and natural month). The specific statistical dimensions are as follows: the frequency of early warnings is counted by device dimension to identify high-frequency faulty devices and guide the allocation of maintenance resources; the alarm contribution of fixed thresholds, trend analysis, and correlation rules is counted by triggering logic dimension to evaluate the effectiveness of each logic; the distribution ratio of real fault samples and false alarm samples is counted by handling conclusion dimension to evaluate the overall accuracy of the early warning strategy.

[0078] Automatic rule generation and iteration: Based on a fully archived dataset of historical events and their handling conclusions, candidate rule update items are automatically generated, specifically including the following two scenarios: For false alarm samples, the system extracts all feature data at the time the false alarm was triggered, analyzes the reasons for the false alarm, and automatically generates corresponding optimization suggestions based on the cause type. Specifically, the automatic analysis of false alarm samples follows this process: Feature extraction: For each sample marked as a false alarm, the system extracts the complete parameter sequence from the 10 sampling points before the trigger time to the 10 sampling points after the trigger time, generating the feature vector F of that sample. m ; Cause classification: The feature vector F m The input is fed into a pre-trained false positive classifier (using a decision tree model), which outputs the cause category to which the false positive sample belongs. The cause categories include overly sensitive threshold, excessively low slope threshold, overly broad association rule coverage, and others. Threshold matching and optimization suggestion generation: Based on the cause category, the system matches the corresponding optimization strategy template, fills in the specific parameter threshold, and outputs candidate rule update items. If the parameter value of the false alarm sample exceeds the threshold by less than or equal to 5% at the trigger time, and more than 95% of the statistical distribution of this parameter in the same historical period (e.g., the same period in the past 7 days) falls outside the original threshold, it is determined that the fixed threshold is too sensitive, and a threshold adjustment suggestion is generated, with an adjustment range of 5% to 10% of the current threshold. If the false alarm sample is triggered by trend analysis, but the rate of change at the trigger time is lower than the historical average rate of change of the same working condition under normal operation of the equipment, it is determined that the slope threshold is set too low, and a slope threshold adjustment suggestion is generated, with an adjustment range of 10% to 20% of the current slope. If the false alarm sample is triggered by association rules, but backtracking analysis reveals a large number of historical samples with similar trigger conditions but no actual failure, it is determined that the coverage of the association rule is too wide, and it is automatically recommended to add time window constraints (e.g., lasting more than 60 seconds) or auxiliary parameter verification conditions (e.g., and the equipment is in operation) to improve the accuracy of the rules.

[0079] To address missed cases and fault modes not covered by existing rules, association mining algorithms (such as Apriori or FP-Growth) are used for in-depth analysis of massive historical monitoring and handling data. Specifically, the new rule mining is performed according to the following process: Transaction construction: Extract all events marked as real faults within the past 3 months from the historical database, treat each fault event as a transaction, extract multi-parameter time-series data (including exhaust temperature, exhaust pressure, cooling water temperature, equipment current, vibration amplitude, etc.) within the 30-minute window before the fault, encode the changing trends of each parameter into discretized state items, and generate transaction dataset T; Frequent itemset generation: The Apriori algorithm is used to iteratively scan the transaction dataset T, setting a minimum support threshold of 1%, to generate all frequent itemsets that meet the support requirement. The support is the frequency of a particular itemset appearing in all transactions. Strong association rule extraction: For each frequent itemset, enumerate all possible non-empty subsets as antecedents, calculate the confidence score of the corresponding rule, and retain rules with a confidence score ≥70% as strong association rules. The confidence score is the conditional probability that the consequent also occurs given the occurrence of the antecedent. Candidate rule formatting: For the extracted strongly correlated rules, standardized candidate warning rules are generated according to a unified format (IF condition 1 AND / OR condition 2 THEN conclusion) and written into the queue to be reviewed.

[0080] For example, the system identifies a frequent itemset from historical data: {exhaust temperature rises ≥0.5℃ per minute, cooling water inlet temperature ≥32℃, exhaust pressure fluctuation ≥0.15MPa}, with a support of 1.2% and a confidence level of 82%. The system generates the following standardized candidate rule for this frequent itemset: IF Exhaust temperature rises ≥0.5℃ per minute AND Cooling water inlet temperature ≥32℃ AND Exhaust pressure fluctuation ≥0.15MPa THEN This is determined to be a decrease in cooling system heat dissipation efficiency, triggering an early warning level.

[0081] Review and Issuance: All automatically generated candidate rule updates are manually reviewed by operations personnel before being written into the rule base, and a version number is recorded (format: V{major version}.{minor version}.{revision version}, e.g., V2.1.3). The system has a built-in rule version rollback mechanism. If the number of false positives or false negatives increases abnormally after a new version rule is launched (e.g., an increase of more than 20% compared to the previous week), it supports rolling back to the previous stable version. Optimized rules that have completed review and version entry into the database are automatically synchronized and fed back to the hierarchical intelligent early warning engine, updating the engine's internal judgment logic.

[0082] It should be understood that the present invention has been described above by way of example with reference to the accompanying drawings, and is intended to enable those skilled in the art to clearly understand the technical solution of the present invention, and should not be construed as any limitation on the present invention. Those skilled in the art should understand that the above embodiments are merely illustrative, and in practical applications, any modifications, equivalent substitutions, improvements, changes or variations made within the spirit and principles of the present invention should be included within the scope of protection defined by the claims of the present invention.

Claims

1. A multi-protocol air compressor station early warning system, characterized in that, include: A multi-protocol unified access module is configured with a protocol library, a driver manager, a protocol identification engine, and a protocol conversion engine; the protocol library stores metadata configuration information for several communication protocols. The driver manager adopts a plug-in architecture, supports dynamic loading of protocol drivers at runtime, and converts heterogeneous protocol data into a unified object model format based on the metadata configuration information in the protocol library. The multi-protocol unified access module is also configured to automatically identify unmatched protocols through the protocol identification engine. The protocol identification engine executes a three-level progressive identification strategy: the first level is precise matching, the second level is machine learning prediction, and the third level is dynamic learning and inclusion. The identified protocol type is output to the driver manager to complete driver loading. When the driver manager starts up, it scans the protocol plugin directory. When a new device is detected, it matches the corresponding driver from the protocol library based on the protocol identifier reported by the device. If the match is successful, the driver is loaded into memory using the class loading mechanism. If the match fails, the protocol recognition engine is triggered to perform message feature recognition.

2. The system according to claim 1, characterized in that, The step of converting heterogeneous protocol data into a unified object model format based on the metadata configuration information in the protocol library includes: The protocol conversion engine maps different protocol fields to a unified object model according to the object model mapping rules, converts values ​​with different units and aligns timestamps in a unified manner, and outputs structured data packets. In the second-level strategy of the protocol identification engine, the machine learning prediction uses a random forest classifier; the message statistical features extracted by the random forest classifier include frame length variance, byte entropy distribution, and field change frequency. In the third-level strategy of the protocol recognition engine, unrecognizable messages are written into the Kafka annotation queue for subsequent manual annotation and model iteration training.

3. The system according to claim 1, characterized in that, It also includes an edge data preprocessing unit, which is communicatively connected to the multi-protocol unified access module. It is configured to sequentially perform data denoising, outlier removal, and breakpoint completion on the data in the unified object model format to generate processed data carrying a data quality identifier. The data quality identifier is used for downstream modules to execute differentiated admission strategies. It is also configured to cache the processed data locally when the network is interrupted and to perform breakpoint resumption after the network is restored.

4. The system according to claim 3, characterized in that, It also includes a hierarchical intelligent early warning engine, which is communicatively connected to the edge data preprocessing unit. It is configured to first parse the data quality identifier carried by the processed data and execute a differentiated data acceptance strategy based on the data quality identifier. It is also configured to perform fixed threshold determination, trend determination and cross-device correlation determination in parallel within a preset time window to generate multiple candidate early warning events, and to merge the candidate early warning events through preset fusion rules to determine the final early warning event and the corresponding early warning level.

5. The system according to claim 3, characterized in that, The specific processing logic of the edge data preprocessing unit includes: A sliding window filtering method is used to perform median smoothing on continuous sampled values ​​to eliminate spike noise caused by impulse interference; Using the statistical mean and standard deviation of the K most recent valid sampling points as a reference benchmark, outlier data points exceeding the preset outlier threshold are removed; When the time difference between adjacent sampling exceeds a preset threshold, a linear interpolation algorithm is triggered to complete the breakpoint. When the network is interrupted, the data is written to the local persistent cache, and the interrupted transmission is resumed in the order of timestamps after the network is restored. The processed data is labeled with quality tags, and the tag categories include at least normal, noise-reduced, completed, and missing.

6. The system according to claim 4, characterized in that, It also includes an internal and external dual-end early warning push module, which is connected to the hierarchical intelligent early warning engine and configured to generate a unified event identifier for the merged early warning events, and push early warning information to the internal enterprise end and the client end simultaneously according to the unified event identifier. It is also configured to perform field-level desensitization processing on the early warning information pushed to the client to hide the internal data fields.

7. The system according to claim 4, characterized in that, The judgment and fusion logic of the hierarchical intelligent early warning engine includes: The trend determination is achieved by calculating the linear regression slope of the sampling points within a preset time window. If multiple consecutive windows satisfy the positive growth and the slope exceeds the preset rate of change, a trend warning is triggered. The association determination is based on production rules generated from historical event mining, and is triggered when multiple indicators simultaneously meet the conditions within the current time window. The preset fusion rules are configured as follows: if only one judgment hits, the corresponding level is output; if multiple judgments hit simultaneously, the system calculates a fusion score based on the level and number of hits of each judgment logic. The fusion score comprehensively considers the preset weight coefficient of each judgment logic and the confidence correction coefficient corresponding to the data quality identifier. When the fusion score reaches the preset improvement threshold and the number of hit judgments meets the preset conditions, the highest hit level is increased by one level as the output; if an emergency alarm condition is involved, the emergency alarm level is directly output. Multiple candidate events triggered by the same device, the same time window, and the same anomaly source are merged into a single event ID.

8. The system according to claim 6, characterized in that, The information processing mechanism of the internal and external dual-end early warning push module includes: The push module reads the complete basic fields of the early warning event and standardizes the push content into a unified data structure. The data structure includes at least: event ID, device identifier, early warning level, triggering reason, occurrence time and handling suggestions. For internal enterprise use, the module matches the corresponding push method according to the warning level and displays all raw data fields, including internal maintenance notes and maintenance personnel information. For the client, the module pushes targeted de-identified messages to the client's pre-configured receiver, hiding internal enterprise handling notes and maintenance personnel information, and retaining only the event ID, device parameters, fault cause, and general handling suggestions. In the push notifications to clients, for warning events with a warning level higher than a preset threshold, the module simultaneously triggers an SMS push to send a reminder message to the client's pre-set contacts.

9. The system according to claim 4, characterized in that, It also includes an early warning closed-loop management module, which is connected to the hierarchical intelligent early warning engine and configured to record early warning events and handling results with the unified event identifier as the primary key, and automatically generate rule update items based on historical data; After review, the rule update item is fed back to the hierarchical intelligent early warning engine to iteratively update the judgment logic of the hierarchical intelligent early warning engine; The rule iteration mechanism of the early warning closed-loop management module includes: Distinguish between valid alarm samples and false alarm samples, and automatically generate candidate rules for threshold adjustment or correlation logic optimization based on the feature distribution of false alarm samples; All candidate rules are reviewed and confirmed before being written into the rule base and then hot-updated and synchronized to the hierarchical intelligent early warning engine. The system has a built-in rule version rollback mechanism. When the false positive rate or false negative rate of a new version rule increases by more than a preset percentage threshold compared to the previous week after the new version rule is launched, it will roll back to the previous version.

10. A multi-protocol air compressor station early warning method, applied to the system according to any one of claims 1 to 9, characterized in that, Includes the following steps: S1. Collect the operating data of multi-protocol devices at the air compressor station through the multi-protocol unified access module, and use the plug-in driver manager to dynamically load the protocol driver at runtime to convert the heterogeneous protocol data into a unified object model format. S2. The edge data preprocessing unit sequentially performs data denoising, outlier removal, and breakpoint completion on the unified object model data to generate processed data with data quality identifiers and cache it locally. S3. The data quality identifier carried by the processed data is analyzed by the hierarchical intelligent early warning engine, and a differentiated data acceptance strategy is executed based on the data quality identifier; Within a preset time window, fixed threshold determination, trend determination, and cross-device correlation determination are performed in parallel, and multiple candidate early warning events are output. The fusion score is calculated based on the level and number of hits of each judgment logic. Candidate warning events are merged through preset fusion rules to determine the final warning events and their corresponding warning levels. S4. Generate a unified event identifier for the merged early warning events through the internal and external dual-end early warning push module, and push early warning information to the internal end of the enterprise and the client end simultaneously based on the identifier; S5. The early warning closed-loop management module records early warning events and their handling results, and automatically generates rule update items based on historical data to iteratively optimize the early warning strategy.

Citation Information

Patent Citations

  • Air compression station house operation and maintenance method based on GRU model and robot

    CN121635227A

  • Air compression station pipeline leakage detection method based on dynamic pressure and multi-branch pipe pressure

    CN121720060A

  • Air compression station automatic early warning system based on air floating shaft

    CN122024450A