SCADA intelligent alarm method and system based on AI driving and plug-in thermal management

By using AI-driven health scoring and plugin hot management, and dynamically scheduling high-precision diagnostic plugins, the problems of resource waste and delayed alarms in SCADA systems are solved, enabling proactive prevention and efficient fault alarms.

CN123115671APending Publication Date: 2026-11-17XIAN JIXU SOFTWARE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611318398.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing SCADA systems in industrial settings suffer from resource waste, outdated fixed threshold alarm mechanisms, and rigid systems that cannot be dynamically expanded, making it impossible to achieve early fault warnings and on-demand allocation of computing power.

Method used

A lightweight LSTM time-series inference model driven by AI is used to evaluate device health in real time. Combined with plug-in thermal management technology, high-precision diagnostic plug-ins are dynamically loaded and unloaded. Data is transmitted through the OPC UA protocol to achieve plug-in lifecycle management driven by health score.

Benefits of technology

It achieves a reduction of CPU utilization of more than 30%, moves the alarm timing forward to before the fault occurs, realizes proactive prevention, adapts to complex working conditions without restarting the system, and solves the problems of resource waste and delayed alarms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115671A_ABST
    Figure CN123115671A_ABST
Patent Text Reader

Abstract

The disclosure is an AI-driven and plug-in thermal management-based SCADA intelligent alarm method and system, which comprises a real-time data processing module, an alarm core module, an AI prediction and scheduling module, a thermal loading manager, a high-precision diagnostic plug-in and a plug-in warehouse. The "health score" output by the AI prediction and scheduling module is used as a trigger signal to control the loading and unloading of the high-precision diagnostic plug-in, and the average CPU occupancy rate is reduced by more than 30% through "on-demand loading", solving the problem of waste of computing power; two-stage wake-up mechanism active alarm: only loading without operation in the early warning stage, and only instantiation running in the accurate diagnosis stage, moving the alarm opportunity from "after the fault occurs" to "before the fault occurs", realizing the change from passive response to active prevention; the diagnosis function does not need to restart the system, and is automatically scheduled by the AI prediction and scheduling module, which is suitable for complex and variable industrial conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of fault alarm technology, and in particular to a SCADA intelligent alarm method and system based on AI-driven and plug-in thermal management. Background Technology

[0002] In existing SCADA (Supervisory Control and Data Acquisition) platforms, although most products adopt a plug-in extended architecture to improve the ability to iterate business functions, the underlying operation and scheduling mechanism has inherent defects that cannot be completely eliminated by conventional means such as parameter tuning and code optimization. The specific problems are as follows: 1. Full static loading leads to long-term consumption of computing power and memory idle resources. During the initialization phase of a traditional SCADA system, the plugin repository directory is traversed, and all fault diagnosis plugins (dynamic link libraries DLL / SO) are loaded into the process memory at once. An independent thread is then allocated to poll and execute the feature calculation logic. Industrial field equipment operates under rated normal conditions for more than 90% of the year, with no abnormal feature data input. The continuous empty loop calculation of high-precision diagnostic plugins consumes a large amount of CPU and memory, and the resource waste is linearly cumulative in multi-site cluster deployments.

[0003] 2. Fixed threshold triggering alarm mechanism only supports post-fault alarms and lacks early degradation prediction capability. Traditional alarm logic uses a static threshold discrimination method:

[0004] In the formula: The real-time data collected by the device's measuring points at time t. Alarm thresholds are set manually.

[0005] This logic only triggers an alarm after the equipment's operating parameters exceed a set threshold, which is a delayed response after a fault occurs. For slow time-varying deterioration faults such as progressive wear of motor bearings, slow aging of winding inter-turn insulation, corrosion and leakage in pipelines, and minor internal leakage in hydraulic systems, the parameter changes are consistently below the static threshold, making it impossible to detect early signs of faults. Furthermore, fluctuations in equipment load and changes in environmental temperature and humidity can lead to false alarms and missed alarms. Manually correcting the thresholds periodically is extremely costly, and there is no risk classification dynamic monitoring strategy based on time-series trends.

[0006] 3. The system operating environment is unaware of the limitations, diagnostic logic cannot be scaled up elastically, and the architecture is rigid. The existing plug-in scheduling module lacks an interface for perceiving device health status, making it impossible to dynamically add, switch, or disable diagnostic logic based on time-series data trends. When adding a new device fault diagnosis plug-in, the SCADA main service must be stopped and the process restarted to complete the loading. In continuous production scenarios such as rail transit, power grids, and chemical industries, shutdown and restart are not allowed, resulting in hard constraints on functional expansion. At the same time, it is impossible to allocate computing thread pool resources differently according to the fault risk level, with high-risk devices and normal devices occupying the same computing power, and resource scheduling lacking hierarchical priority.

[0007] In summary, critical industrial infrastructure urgently needs an intelligent SCADA alarm system that can dynamically assess health risks based on equipment time-series operating data, automatically manage and control the entire lifecycle of diagnostic plugins, achieve on-demand allocation of computing power, and provide early warning of faults.

[0008] It should be noted that this section is intended to provide background or context for the technical solutions of this disclosure as set forth in the claims. The description herein does not constitute an admission that it is prior art simply because it is included in this section. Summary of the Invention

[0009] The purpose of this invention is to provide a SCADA intelligent alarm method and system based on AI-driven and plug-in thermal management, thereby overcoming, to at least to some extent, one or more problems caused by the limitations and defects of related technologies.

[0010] This invention first provides a SCADA intelligent alarm method based on AI-driven and plug-in thermal management, the alarm method comprising: S1 collects multi-dimensional measurement point data from industrial equipment in real time and transmits the data via the OPC UA protocol; S2, the resident basic threshold alarm plugin compares the measurement point data at each sampling time with its corresponding preset alarm threshold. When the measurement point data reaches the preset alarm threshold, an internal original alarm flag is generated. The number of original alarm flags in the continuous sampling period is accumulated within the preset false alarm filtering time window. When the accumulated number reaches the preset effective counting threshold, a valid alarm signal is output. S3 receives the measurement point data in real time through the built-in lightweight LSTM temporal inference model, extracts the temporal features of the measurement points using the LSTM network, and obtains a standardized equipment health score through weighted calculation of the output layer and mapping by the Sigmoid activation function. S4. Based on the threshold range of the health score, perform dynamic lifecycle management on the high-precision diagnostic plugin. When the high-precision diagnostic plugin performs time-series fault feature analysis with a high-frequency sampling period, output the fault confidence in real time. S5. If the fault confidence is greater than or equal to the preset confidence threshold, a predictive early warning alarm is pushed and the fault type is marked. If the health score rises to the lower limit of the first threshold range for multiple consecutive sampling periods, the system automatically uninstalls the corresponding high-precision diagnostic plugin, destroys the computing thread and releases all memory and CPU resources, and the system falls back to the normal lightweight monitoring mode.

[0011] In this invention, in S1, the measurement point data are various process parameters collected in real time from industrial field equipment by the SCADA system to reflect the operating status of the equipment, including motor bearing temperature, winding current, vibration acceleration and velocity, pipeline pressure, flow rate, hydraulic oil level, and insulation resistance value.

[0012] In this invention, in step S2, the number of original alarm markers in a continuous sampling period is accumulated within a preset false alarm filtering time window. When the accumulated number reaches a preset effective counting threshold, the filtering calculation formula used in the process of outputting a valid alarm signal is as follows:

[0013] in, This represents the cumulative number of times the original alarm flags have been set. For the original alarm flag, This is a false alarm filtering window. The effective counting threshold is continuously triggered to filter false alarms caused by transient pulse interference.

[0014] In this invention, S3 is the standardized health score. The calculation formula is as follows: = Sigmoid(LSTM( ) * + ) In the formula: This is a multi-dimensional time series dataset of measurement points. Output the weight matrix for the model. As a bias term, the Sigmoid activation function maps the output to the [0,1] interval.

[0015] In this invention, S4 includes: When the health score is in the first threshold range, all high-precision diagnostic plugins are kept in disk hibernation mode and no computing resources are allocated to them. The system maintains normal lightweight monitoring. When the health score is in the second threshold range and the duration is not less than the preset duration, the plugin wake-up event is triggered. The system reads the plugin repository metadata, pre-allocates thread pool and memory resources for the high-precision diagnostic plugin that matches the current device degradation characteristics, completes loading and enters the ready waiting state but does not perform diagnostic calculations. When the health score is in the third threshold range, the system instantiates the corresponding high-precision diagnostic plugin, allocates an independent computing thread to perform time-series fault feature analysis with a high-frequency sampling period, and outputs the fault confidence in real time. The values ​​in the first threshold interval, the second threshold interval, and the third threshold interval decrease sequentially.

[0016] In this invention, the first threshold interval is [0.6, 1.0], the second threshold interval is [0.3, 0.6], and the third threshold interval is [0, 0.3].

[0017] The present invention further provides a SCADA intelligent alarm system based on AI-driven and plug-in thermal management, the alarm system comprising: The real-time data processing module is used to collect multi-dimensional measurement point data of industrial equipment in real time and transmit the data through the OPC UA protocol. The alarm core module is used by the resident basic threshold alarm plugin to compare the measurement point data at each sampling time with its corresponding preset alarm threshold. When the measurement point data reaches the preset alarm threshold, an internal original alarm flag is generated. Within a preset false alarm filtering time window, the number of original alarm flags in consecutive sampling periods is accumulated and counted. When the accumulated number reaches a preset effective count threshold, a valid alarm signal is output. Based on the confidence comparison result, a predictive pre-alarm is pushed to the front end and the fault type is marked. The AI ​​prediction and scheduling module receives the measurement point data in real time through a built-in lightweight LSTM temporal inference model, extracts the temporal features of the measurement points using the LSTM network, and obtains a standardized device health score through weighted calculation of the output layer and mapping using the Sigmoid activation function. Based on the threshold range of the health score, dynamic lifecycle management is performed on the high-precision diagnostic plugin, and full lifecycle management is performed on the high-precision diagnostic plugin. If the health score rises to a level not less than the lower limit of the first threshold range for several consecutive sampling periods, the hot-loading manager is controlled to automatically unload the corresponding high-precision diagnostic plugin, destroy the computing thread, and release all memory and CPU resources, and the system falls back to the normal lightweight monitoring mode. The hot-load manager, under the control of the AI ​​prediction and scheduling module, performs full lifecycle scheduling of high-precision diagnostic plugins. It is used to load high-precision diagnostic plugins from the plugin repository or unload them from the system, or enter a waiting-to-load-ready state without performing diagnostic operations. When the high-precision diagnostic plugin is loaded and running, it performs time-series fault feature analysis with a high-frequency sampling period and outputs the fault confidence level in real time. If the fault confidence level is greater than or equal to the preset confidence level threshold, the confidence level comparison result is pushed to the alarm core module. The plugin repository stores high-precision diagnostic plugins for various types of faults and is in hibernation by default.

[0018] The technical solution provided by this invention may include the following beneficial effects: This invention presents an AI-driven SCADA intelligent alarm method and system based on plug-in thermal management. It utilizes a "health score" output by an AI prediction and scheduling module as a trigger signal to control the loading and unloading of high-precision diagnostic plug-ins. Through "on-demand loading," the average CPU utilization is reduced by more than 30%, solving the problem of wasted computing power. A two-level wake-up mechanism provides proactive alarming: in the early warning stage, loading is performed without computation, while in the precise diagnosis stage, the plug-in is instantiated and run, shifting the alarm timing from "after the fault occurs" to "before the fault occurs," achieving a shift from passive response to proactive prevention. The diagnostic function does not require a system restart; it is automatically scheduled by the AI ​​prediction and scheduling module, adapting to complex and ever-changing industrial conditions. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0020] Figure 1 A flowchart illustrating an exemplary embodiment of the present disclosure of the SCADA intelligent alarm method based on AI-driven and plug-in thermal management is shown. Figure 2 This diagram illustrates the overall architecture of an AI-driven and plug-in thermal management-based SCADA intelligent alarm system in an exemplary embodiment of this disclosure. Figure 3 This illustration shows the interactive call flowchart between the real-time data processing module, the alarm core module, the AI ​​prediction and scheduling module, the plug-in repository, the hot loading manager, the high-precision diagnostic plug-in, and the HMI when an alarm event is triggered in an exemplary embodiment of this disclosure. Figure 4 This illustration shows the interactive call flowchart between the real-time data processing module, the alarm core module, the AI ​​prediction and scheduling module, the plug-in repository, the hot loading manager, the high-precision diagnostic plug-in, and the HMI when an alarm is recovered in an exemplary embodiment of this disclosure. Figure 5This diagram illustrates the interactive call flow between the real-time data processing module, alarm core module, AI prediction and scheduling module, plugin repository, hot loading manager, and high-precision diagnostic plugin when the system starts up in an exemplary embodiment of this disclosure. Detailed Implementation

[0021] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0022] Furthermore, the accompanying drawings are merely illustrative diagrams of embodiments of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0023] This example implementation first provides a SCADA intelligent alarm method based on AI-driven and plug-in thermal management. Please refer to [link / reference]. Figure 1 This method may include: S1-S5, as follows: S1 collects multi-dimensional measurement point data from industrial equipment in real time and transmits the data via the OPC UA protocol; S2, the resident basic threshold alarm plugin compares the measurement point data at each sampling time with its corresponding preset alarm threshold. When the measurement point data reaches the preset alarm threshold, an internal original alarm flag is generated. The number of original alarm flags in the continuous sampling period is accumulated within the preset false alarm filtering time window. When the accumulated number reaches the preset effective counting threshold, a valid alarm signal is output. S3 receives the measurement point data in real time through the built-in lightweight LSTM temporal inference model, extracts the temporal features of the measurement points using the LSTM network, and obtains a standardized equipment health score through weighted calculation of the output layer and mapping by the Sigmoid activation function. S4. Based on the threshold range of the health score, perform dynamic lifecycle management on the high-precision diagnostic plugin. When the high-precision diagnostic plugin performs time-series fault feature analysis with a high-frequency sampling period, output the fault confidence in real time. S5. If the fault confidence is greater than or equal to the preset confidence threshold, a predictive early warning alarm is pushed and the fault type is marked. If the health score rises to the lower limit of the first threshold range for multiple consecutive sampling periods, the system automatically uninstalls the corresponding high-precision diagnostic plugin, destroys the computing thread and releases all memory and CPU resources, and the system falls back to the normal lightweight monitoring mode.

[0024] This example implementation first provides an AI-driven SCADA intelligent alarm system based on plug-in thermal management. Please refer to [link / reference]. Figure 1 The alarm system includes: The real-time data processing module is used to collect multi-dimensional measurement point data of industrial equipment in real time and transmit the data through the OPC UA protocol. The alarm core module is used by the resident basic threshold alarm plugin to compare the measurement point data at each sampling time with its corresponding preset alarm threshold. When the measurement point data reaches the preset alarm threshold, an internal original alarm flag is generated. Within a preset false alarm filtering time window, the number of original alarm flags in consecutive sampling periods is accumulated and counted. When the accumulated number reaches a preset effective count threshold, a valid alarm signal is output. Based on the confidence comparison result, a predictive pre-alarm is pushed to the front end and the fault type is marked. The AI ​​prediction and scheduling module receives the measurement point data in real time through a built-in lightweight LSTM temporal inference model, extracts the temporal features of the measurement points using the LSTM network, and obtains a standardized device health score through weighted calculation of the output layer and mapping using the Sigmoid activation function. Based on the threshold range of the health score, dynamic lifecycle management is performed on the high-precision diagnostic plugin, and full lifecycle management is performed on the high-precision diagnostic plugin. If the health score rises to a level not less than the lower limit of the first threshold range for several consecutive sampling periods, the hot-loading manager is controlled to automatically unload the corresponding high-precision diagnostic plugin, destroy the computing thread, and release all memory and CPU resources, and the system falls back to the normal lightweight monitoring mode. The hot-load manager, under the control of the AI ​​prediction and scheduling module, performs full lifecycle scheduling of high-precision diagnostic plugins. It is used to load high-precision diagnostic plugins from the plugin repository or unload them from the system, or enter a waiting-to-load-ready state without performing diagnostic operations. When the high-precision diagnostic plugin is loaded and running, it performs time-series fault feature analysis with a high-frequency sampling period and outputs the fault confidence level in real time. If the fault confidence level is greater than or equal to the preset confidence level threshold, the confidence level comparison result is pushed to the alarm core module. The plugin repository stores high-precision diagnostic plugins for various types of faults and is in hibernation by default.

[0025] Please refer to Figures 3-5The implementation process of the technical solution of this application will be described below in conjunction with the alarm method and alarm system.

[0026] 1. System Architecture The alarm system of this application includes: a real-time data processing module, an alarm core module, an AI prediction and scheduling module, a hot-loading manager, a high-precision diagnostic plugin, and a plugin repository.

[0027] This application relies on a lightweight LSTM time-series AI model to output device health scores in real time. The health scores are used as the trigger conditions for plugin scheduling to achieve automated management of the entire lifecycle of high-precision diagnostic plugins, including hibernation, wake-up, and uninstallation. This enables on-demand allocation of computing resources and constructs a two-layer alarm architecture of "normal lightweight threshold monitoring + high-precision time-series diagnosis of fault precursors".

[0028] (1) Alarm core module The alarm core module is responsible for alarm lifecycle, status management, OPC UA communication, storage, and alarm filtering rule calculation.

[0029] The core filtering calculation formula is as follows:

[0030] in, This represents the cumulative number of times the original alarm flags have been set. For the original alarm flag, This is a false alarm filtering window. The effective counting threshold is continuously triggered to filter false alarms caused by transient pulse interference.

[0031] (2) AI Prediction and Scheduling Module The AI ​​prediction and scheduling module incorporates a lightweight LSTM fault prediction inference model, and receives real-time access to multi-dimensional measurement point time-series datasets (X = ... This is used to calculate the health score of devices in real time. Standardized calculation formula: = Sigmoid(LSTM( ) * + ) In the formula: Output the weight matrix for the model. As a bias term, the Sigmoid activation function maps the output to the [0,1] interval.

[0032] Health stratification judgment parameter threshold (configurable): Normal safe interval: Ht∈[0.6,1.0]; Warning transition interval: Ht∈[0.3,0.6); High-risk fault range: Ht∈[0,0.3).

[0033] (3) Plugin repository A high-precision diagnostic plugin for multiple types of faults with local persistent storage (DLL on Windows platform, SO dynamic library on Linux platform), which hibernates by default and does not consume runtime resources.

[0034] (4) Hot-load manager The plugin lifecycle scheduling and execution unit has a built-in dynamic link library loader, thread pool allocator, and resource recycling unit; it supports lifecycle management such as incremental hot loading of plugins, memory release, and thread destruction.

[0035] (5) High-precision diagnostic plugin Business diagnostics, custom fields, and high-precision analysis logic.

[0036] 2. Dynamic scheduling method for AI prediction and scheduling module (1) Normal stage ( [0.6,1.0]) The system only has a resident basic threshold alarm plugin (occupying a fixed low computing power), and the AI ​​prediction and scheduling module performs LSTM inference with a low-frequency sampling period to continuously monitor the time series data of the test points with low power consumption; all high-precision diagnostic plugins keep the disk in hibernation state, with zero memory usage and no additional thread operations.

[0037] (2) Early warning stage ( [0.3, 0.6), duration ≥ 3s for image stabilization determination) When the AI ​​prediction and scheduling module determines that the device health score is below a certain threshold (e.g., <0.6) and exceeds a set time (e.g., ≥3s), a plugin wake-up event is triggered. The hot loading manager reads the plugin repository metadata, matches the diagnostic plugin corresponding to the current device degradation characteristics, completes the pre-allocation of thread pool and the reservation of memory resources, and enters the waiting loading ready state without performing diagnostic operations.

[0038] (3) Precision diagnosis stage ( [0,0.3)) The hot-loading manager instantiates the plugin, allocates an independent computing thread, performs time-series fault feature analysis with a high-frequency sampling period, and outputs the fault confidence Conf in real time; it also synchronously counts the real-time computing power usage, and if it exceeds the CPU and memory limits, it automatically reduces the plugin's operation sampling frequency to achieve load balancing.

[0039] (4) Disposal phase If the plugin outputs a fault confidence level Conf ≥ (System default baseline value:) =0.75): Push predictive pre-alarms to the alarm core module and mark the fault type; If the AI ​​model outputs a health score that rises to a certain level for several consecutive sampling periods (e.g., 20), If the value is ≥ 0.6, the potential fault is determined to be eliminated; the hot-load manager automatically uninstalls the high-precision diagnostic plugin, destroys the computing threads, releases all memory and CPU resources, and the system falls back to the normal lightweight monitoring mode.

[0040] This application utilizes the "health score" output by the AI ​​prediction and scheduling module as a trigger signal to control the loading and unloading of high-precision diagnostic plugins. Through "on-demand loading," the average CPU utilization rate is reduced by more than 30%, solving the problem of wasted computing power. The two-level wake-up mechanism provides proactive alarms: in the early warning stage, only loading is performed without computation, while in the precise diagnosis stage, the plugin is instantiated and run, shifting the alarm timing from "after the fault occurs" to "before the fault occurs," realizing a shift from passive response to proactive prevention. The diagnostic function does not require a system restart and is automatically scheduled by the AI ​​prediction and scheduling module, adapting to complex and ever-changing industrial conditions.

[0041] It should be noted that although several modules of the system for executing actions are mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules described above can be embodied in one module. Conversely, the features and functions of one module described above can be further divided into multiple modules for embodiment. Components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0042] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0043] It should be noted that the installation of image acquisition and personal identification equipment in public places involved in this application is necessary for maintaining public safety, complies with relevant national regulations, and is accompanied by prominent warning signs. The collected personal images and identification information can only be used for the purpose of maintaining public safety and not for other purposes; or the images, personal identification data, etc. in this application are all legally and compliantly obtained or collected with the individual's separate consent.

Claims

1. A SCADA intelligent alarm method based on AI-driven and plug-in thermal management, characterized in that, The alarm method includes: S1 collects multi-dimensional measurement point data from industrial equipment in real time and transmits the data via the OPC UA protocol; S2, the resident basic threshold alarm plugin compares the measurement point data at each sampling time with its corresponding preset alarm threshold. When the measurement point data reaches the preset alarm threshold, an internal original alarm flag is generated. The number of original alarm flags in the continuous sampling period is accumulated within the preset false alarm filtering time window. When the accumulated number reaches the preset effective counting threshold, a valid alarm signal is output. S3 receives the measurement point data in real time through the built-in lightweight LSTM temporal inference model, extracts the temporal features of the measurement points using the LSTM network, and obtains a standardized equipment health score through weighted calculation of the output layer and mapping by the Sigmoid activation function. S4. Based on the threshold range of the health score, perform dynamic lifecycle management on the high-precision diagnostic plugin. When the high-precision diagnostic plugin performs time-series fault feature analysis with a high-frequency sampling period, output the fault confidence in real time. S5. If the fault confidence is greater than or equal to the preset confidence threshold, a predictive early warning alarm is pushed and the fault type is marked. If the health score rises to the lower limit of the first threshold range for multiple consecutive sampling periods, the system automatically uninstalls the corresponding high-precision diagnostic plugin, destroys the computing thread and releases all memory and CPU resources, and the system falls back to the normal lightweight monitoring mode.

2. The SCADA intelligent alarm method based on AI-driven and plug-in thermal management according to claim 1, characterized in that, In S1, the measurement point data are various process parameters collected in real time from industrial field equipment by the SCADA system to reflect the operating status of the equipment, including motor bearing temperature, winding current, vibration acceleration and velocity, pipeline pressure, flow rate, hydraulic oil level and insulation resistance value.

3. The SCADA intelligent alarm method based on AI-driven and plug-in thermal management according to claim 1, characterized in that, In S2, the number of original alarm markers in a continuous sampling period is accumulated within a preset false alarm filtering time window. When the accumulated number reaches a preset effective counting threshold, the filtering calculation formula used in the process of outputting a valid alarm signal is as follows: in, This represents the cumulative number of times the original alarm markers have been set. For the original alarm flag, This is a false alarm filtering window. The effective counting threshold is continuously triggered to filter false alarms caused by transient pulse interference.

4. The SCADA intelligent alarm method based on AI-driven and plug-in thermal management according to claim 1, characterized in that, In S3, the standardized health score The calculation formula is as follows: = Sigmoid(LSTM( ) * + ) In the formula: This is a multi-dimensional time series dataset of measurement points. Output the weight matrix for the model. As a bias term, the Sigmoid activation function maps the output to the [0,1] interval.

5. The SCADA intelligent alarm method based on AI-driven and plug-in thermal management according to claim 1, characterized in that, S4 include: When the health score is in the first threshold range, all high-precision diagnostic plugins are kept in disk hibernation mode and no computing resources are allocated to them. The system maintains normal lightweight monitoring. When the health score is in the second threshold range and the duration is not less than the preset duration, the plugin wake-up event is triggered. The system reads the plugin repository metadata, pre-allocates thread pool and memory resources for the high-precision diagnostic plugin that matches the current device degradation characteristics, completes loading and enters the ready waiting state but does not perform diagnostic calculations. When the health score is in the third threshold range, the system instantiates the corresponding high-precision diagnostic plugin, allocates an independent computing thread to perform time-series fault feature analysis with a high-frequency sampling period, and outputs the fault confidence in real time. The values ​​in the first threshold interval, the second threshold interval, and the third threshold interval decrease sequentially.

6. The SCADA intelligent alarm method based on AI-driven and plug-in thermal management according to claim 5, characterized in that, The first threshold interval is [0.6, 1.0], the second threshold interval is [0.3, 0.6], and the third threshold interval is [0, 0.3].

7. A SCADA intelligent alarm system based on AI-driven and plug-in thermal management, characterized in that, The alarm system includes: The real-time data processing module is used to collect multi-dimensional measurement point data of industrial equipment in real time and transmit the data through the OPC UA protocol. The alarm core module is used by the resident basic threshold alarm plugin to compare the measurement point data at each sampling time with its corresponding preset alarm threshold. When the measurement point data reaches the preset alarm threshold, an internal original alarm flag is generated. Within a preset false alarm filtering time window, the number of original alarm flags in consecutive sampling periods is accumulated and counted. When the accumulated number reaches a preset effective count threshold, a valid alarm signal is output. Based on the confidence comparison result, a predictive pre-alarm is pushed to the front end and the fault type is marked. The AI ​​prediction and scheduling module receives the measurement point data in real time through a built-in lightweight LSTM temporal inference model, extracts the temporal features of the measurement points using the LSTM network, and obtains a standardized device health score through weighted calculation of the output layer and mapping using the Sigmoid activation function. Based on the threshold range of the health score, dynamic lifecycle management is performed on the high-precision diagnostic plugin, and full lifecycle management is performed on the high-precision diagnostic plugin. If the health score rises to a level not less than the lower limit of the first threshold range for several consecutive sampling periods, the hot-loading manager is controlled to automatically unload the corresponding high-precision diagnostic plugin, destroy the computing thread, and release all memory and CPU resources, and the system falls back to the normal lightweight monitoring mode. The hot-load manager, under the control of the AI ​​prediction and scheduling module, performs full lifecycle scheduling of high-precision diagnostic plugins. It is used to load high-precision diagnostic plugins from the plugin repository or unload them from the system, or enter a waiting-to-load-ready state without performing diagnostic operations. When the high-precision diagnostic plugin is loaded and running, it performs time-series fault feature analysis with a high-frequency sampling period and outputs the fault confidence level in real time. If the fault confidence level is greater than or equal to the preset confidence level threshold, the confidence level comparison result is pushed to the alarm core module. The plugin repository stores high-precision diagnostic plugins for various types of faults and is in hibernation by default.