Method and system for monitoring operation state of automatic control platform

Abstract

The invention relates to the technical field of data processing, and discloses an automatic control platform operation state monitoring method and system. The method comprises the steps that an acquisition terminal captures and processes platform multi-source data, a network transmission engine carries out encryption transmission, a decoding system analyzes the data and then constructs a digital twinborn model, a double-flow fusion network outputs health scores, the health scores are mapped to a fault map to predict risks, an early warning system generates an intervention strategy according to the risks and executes the intervention strategy, and a monitoring control closed loop is established. According to the invention, comprehensive, safe and accurate monitoring of the operation state of the automatic control platform is realized, and an effective prediction and intervention mechanism is provided before a fault occurs.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for monitoring the operating status of an automated control platform. Background Technology

[0002] Currently, automated control platforms are widely used in industrial production, energy management, transportation, and other fields. Their safe and stable operation is crucial for ensuring production continuity and equipment safety. Existing automated control platform operation status monitoring technologies mainly include simple monitoring methods based on threshold judgment, anomaly detection technologies based on statistical models, and fault diagnosis methods based on expert systems. Threshold-based monitoring methods set normal ranges for key parameters and trigger alarms when parameters exceed the thresholds. Anomaly detection technologies based on statistical models use historical data to construct parameter distribution models and identify anomalies that deviate from the normal distribution. Fault diagnosis methods based on expert systems rely on a pre-compiled rule base to infer possible causes of faults based on combinations of symptoms. In addition, some advanced systems also employ single-dimensional machine learning algorithms for status assessment, such as support vector machines, decision trees, and neural networks, to improve the accuracy of fault identification.

[0003] However, existing technologies suffer from several shortcomings, including limited data acquisition, insufficient transmission security, coarse feature extraction, one-sided status assessment, and inaccurate early warning. First, most systems collect only a single type of data (such as parameter data), lacking comprehensive analysis of video streams and control command streams, resulting in insufficient dimensions for status monitoring. Second, security measures during data transmission are simplistic and inadequate to address security threats in complex network environments. Third, feature extraction often employs static methods, failing to effectively capture the dynamic characteristics of the system's operation. Furthermore, status assessment is typically based on a single algorithm model, lacking fusion analysis of static attributes and dynamic behavior, reducing the comprehensiveness of the assessment. Finally, early warning systems are mostly triggered by fixed thresholds, unable to dynamically adjust response strategies based on the operating environment and task importance, leading to high false alarm rates and inaccurate intervention measures. These shortcomings severely limit the effectiveness of status monitoring for automated control platforms and fail to meet the stringent requirements for system reliability and predictive maintenance in today's highly complex industrial environments. Summary of the Invention

[0004] This application provides a method and system for monitoring the operating status of an automated control platform, which enables comprehensive, safe, and accurate monitoring of the operating status of the automated control platform and provides an effective prediction and intervention mechanism before a fault occurs.

[0005] Firstly, this application provides a method for monitoring the operational status of an automated control platform. The method includes: capturing a platform video stream, control command stream, and parameter stream via a data acquisition terminal; applying a lossless compression algorithm to the video stream; performing timing marking on the control command stream; and performing noise suppression on the parameter stream to generate multi-source heterogeneous data packets; inputting the multi-source heterogeneous data packets into a network transmission engine; the network transmission engine adjusting its transmission strategy based on network quality and applying an encryption protocol to form an encrypted data stream; and using a decoding system to parse the encrypted data stream and importing the parsing results into a feature extraction module. The module, namely the feature extraction module, constructs a digital twin feature model by combining a knowledge base and data mining techniques. A dual-stream fusion network is applied to the digital twin feature model. After processing the static and dynamic feature channels, the dual-stream fusion network integrates them through an attention mechanism to output a health score matrix. The health score matrix is ​​mapped to a fault knowledge graph, and an anomaly similarity is calculated using an inference engine. A long short-term memory network is then used to predict fault risks, generating a prediction report. Based on the prediction report, an early warning system is triggered. The early warning system generates intervention strategies according to the risk level and issues execution instructions through a secure channel, establishing a monitoring and control link.

[0006] Secondly, this application provides an automated control platform operation status monitoring system, the automated control platform operation status monitoring system comprising: The tagging module is used to capture platform video stream, control command stream and parameter stream through the acquisition terminal, apply lossless compression algorithm to the video stream, perform timing tagging on the control command stream, suppress noise in the parameter stream, and generate multi-source heterogeneous data packets. An encryption module is used to input the multi-source heterogeneous data packets into the network transmission engine, which adjusts the transmission strategy and applies an encryption protocol according to the network quality to form an encrypted data stream. The import module is used to parse the encrypted data stream using the decoding system and import the parsing results into the feature extraction module. The feature extraction module combines a knowledge base and data mining technology to construct a digital twin feature model. An integration module is used to apply a dual-stream fusion network to the digital twin feature model. The dual-stream fusion network processes the static and dynamic feature channels and then integrates them through an attention mechanism to output a health score matrix. The prediction module is used to map the health score matrix to the fault knowledge graph, calculate the anomaly similarity through the inference engine and combine it with the long short-term memory network to predict the fault risk and generate a prediction report. A module is established to trigger an early warning system based on the prediction report. The early warning system generates an intervention strategy according to the risk level and issues an execution command through a secure channel, thus establishing a monitoring and control link.

[0007] A third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the computer device to execute the above-described automated control platform operation status monitoring method.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method for monitoring the operating status of an automated control platform.

[0009] The technical solution provided in this application simultaneously captures platform video streams, control command streams, and parameter streams through a multi-port acquisition terminal, achieving multi-dimensional and comprehensive data acquisition. This provides a rich foundation of raw data for comprehensive monitoring of the platform's status. Furthermore, targeted processing algorithms are applied to different types of data, such as lossless compression for video streams, timing marking for control command streams, and noise suppression for parameter streams, ensuring data quality and transmission efficiency while significantly reducing data storage and transmission burden. The network transmission engine dynamically adjusts transmission strategies based on a network quality assessment matrix and applies multi-layer encryption protocols, effectively improving the security and reliability of data transmission in complex network environments and ensuring the integrity and confidentiality of monitoring data. The decoding system and feature extraction module, combined with a knowledge base and data mining techniques, construct a digital twin feature model that successfully integrates the platform's... Static structural information and dynamic behavioral characteristics provide accurate feature representations for a comprehensive assessment of the platform's status. The dual-stream fusion network integrates the parallel processing and attention mechanisms of static and dynamic feature channels, fully utilizing the stability of static features and the temporality of dynamic features. The generated health score matrix can reflect the platform's health status from multiple perspectives. Mapping the health score matrix to a fault knowledge graph and combining it with a long short-term memory network for anomaly similarity calculation and fault risk prediction significantly improves the accuracy and foresight of fault prediction, enabling the system to identify potential risks before a fault actually occurs. The early warning system, based on a tiered early warning and intervention strategy generation mechanism triggered by prediction reports, can adaptively adjust response measures according to the risk level and the current platform environment, realizing a shift from passive response to proactive intervention and significantly reducing the fault incidence and impact range. The artificial intelligence algorithms and models applied in this solution, such as inter-frame lossless compression algorithms, wavelet transform filtering technology, dual-stream fusion networks, attention mechanisms, long short-term memory networks, and fault knowledge graphs, not only fully consider the characteristics and needs of data in the industrial control field, but also specifically address the problems of low data processing efficiency, incomplete feature extraction, one-sided state assessment, and inaccurate early warning in traditional monitoring methods. It realizes intelligent operation of the entire process from data acquisition, transmission and processing, feature extraction, state assessment to early warning intervention, enabling the system to maintain high-precision monitoring and prediction capabilities in complex and ever-changing industrial environments, and significantly improving the reliability, security, and maintainability of the automation control platform. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1This is a schematic diagram of one embodiment of the automated control platform operation status monitoring method in this application. Figure 2 This is a schematic diagram of one embodiment of the automated control platform operation status monitoring system in this application. Figure 3 This is a schematic block diagram of the structure of the computer device in an embodiment of the present invention. Detailed Implementation

[0012] This application provides a method and system for monitoring the operational status of an automated control platform. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0013] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the automated control platform operation status monitoring method in this application includes: Step S101: Capture the platform video stream, control command stream, and parameter stream through the acquisition terminal; apply a lossless compression algorithm to the video stream; perform timing marking on the control command stream; and perform noise suppression on the parameter stream to generate multi-source heterogeneous data packets. Step S102: Input multi-source heterogeneous data packets into the network transmission engine. The network transmission engine adjusts the transmission strategy and applies an encryption protocol according to the network quality to form an encrypted data stream. Step S103: Use the decoding system to parse the encrypted data stream, and import the parsing result into the feature extraction module. The feature extraction module combines knowledge base and data mining technology to construct a digital twin feature model. Step S104: Apply a dual-stream fusion network to the digital twin feature model. After processing the static and dynamic feature channels, the dual-stream fusion network integrates them through an attention mechanism to output a health score matrix. Step S105: Map the health score matrix to the fault knowledge graph, calculate the anomaly similarity through the inference engine and combine it with the long short-term memory network to predict the fault risk and generate a prediction report; Step S106: Trigger the early warning system based on the prediction report. The early warning system generates an intervention strategy according to the risk level and issues an execution instruction through a secure channel to establish a monitoring and control link.

[0014] It is understood that the executing entity of this application can be an automated control platform operation status monitoring system, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0015] Specifically, the acquisition terminal first connects simultaneously to the video output interface, control interface, and communication interface of the automation control platform to capture the platform's video stream, control command stream, and parameter stream. Specifically, the multi-port acquisition terminal synchronously acquires raw video signals, control command signals, and operating parameter signals. For the raw video signal, a hardware-accelerated inter-frame lossless compression algorithm is used. This algorithm compares consecutive frames, identifies and retains key area information, and generates compressed video data. For example, when only a local area changes in the automation control platform's display interface, the compression algorithm only records the pixel information and position coordinates of the changed area, thus significantly reducing the data volume. For the control command signal, the acquisition terminal uses a high-precision clock source for microsecond-level timing marking, recording the difference between the transmission time and response time of each command, constructing a complete command timing chain, and forming timestamped control command data. This enables the system to accurately track the execution status and latency of control commands. Simultaneously, wavelet transform filtering technology is applied to the operating parameter signal to eliminate environmental interference and electromagnetic noise, extract key platform operating parameters, and generate noise-reduced parameter data. Finally, the compressed video data, timestamped control command data, and noise reduction parameter data are synchronized and calibrated according to a unified time base. Data format header information and checksums are added, and the data is encapsulated into multi-source heterogeneous data packets with internal correlation. After the multi-source heterogeneous data packets are input into the network transmission engine, they are first divided into three priority levels based on data importance: control data packets, monitoring data packets, and status data packets. The network status detector continuously collects the current network bandwidth utilization, communication latency, and packet loss rate to construct a network quality assessment matrix. The network transmission engine allocates transmission resources to data packets of different priority levels according to this matrix, ensuring that control data packets are transmitted first under network bandwidth constraints. Subsequently, a multi-layered protection structure is adopted for the multi-source heterogeneous data packets. First, a symmetric encryption algorithm (such as AES-256) is used to process the data content; then, a security protocol (such as TLS 1.3) is applied to protect the transmission channel; finally, a secure tunnel (such as IPSec) is established at the network layer to form an encrypted data stream. Digital signature information and timestamp data are also embedded in the encrypted data stream, and error detection and automatic data recovery during transmission are achieved through a segmented verification mechanism.

[0016] After receiving the encrypted data stream, the decoding system performs layered decryption using a multi-layer decryption processor, sequentially removing the network layer security tunnel, transport layer security protocol, and data encryption layer to restore the original data structure. For the decrypted video stream data, an image stream parsing engine is used for frame sequence reconstruction; for control command streams and parameter streams, a protocol identifier separates the data and restores the format, ultimately obtaining the original monitoring data. The feature extraction module extracts a standard set of operating parameters from the original monitoring data based on a platform type knowledge graph, including operating status identifiers, performance values, and environmental variables. An adaptive time window analyzer performs time-series slicing of the original monitoring data, calculates parameter fluctuation amplitude and rate of change, and identifies platform-specific operating characteristics. Finally, the standard set of operating parameters and platform-specific operating characteristics are weighted and merged to construct a digital twin feature model with spatiotemporal dual attributes, including static and dynamic feature channels. When processing the digital twin feature model, the dual-stream fusion network first uses a feature separator to split the model into static and dynamic feature data, which are then input into their respective feature channels. The spatial correlation extractor in the static feature channel performs multi-layer deep analysis on static feature data, uncovering potential relationships between features and generating spatial feature vectors. The temporal pattern recognizer in the dynamic feature channel performs continuous analysis on dynamic feature data, extracting parameter variation patterns and forming temporal feature vectors. The context-aware computing unit calculates weighting coefficients for the spatial and temporal feature vectors based on the platform's current operating environment. The feature mapping converter projects the weighted feature vectors onto the same computational space, performing data fusion through nonlinear transformation equations. From the fused data, the platform's overall health indicators, core component status scores, critical link stability values, and anomaly risk warning coefficients are extracted to form a multi-dimensional health scoring matrix.

[0017] The health score matrix is ​​projected onto the feature space of the fault knowledge graph through a feature mapper, establishing a coordinate correspondence between the current state and the fault mode. A graph traversaler performs bidirectional traversal along concept association paths in the fault knowledge graph, calculating the Euclidean distance between the current state feature points and various fault nodes to obtain an anomaly similarity ranking table. A multi-scale analyzer divides the historical health score data into time periods, extracting hourly fluctuation features, daily trend features, and weekly change features to form a multi-scale temporal feature set. This feature set is input into a Long Short-Term Memory (LSTM) network, which uses a triple structure of forget gate, input gate, and output gate to store historical state information and extrapolates key indicators temporally to generate future state prediction curves. Combining the anomaly similarity ranking table and the future state prediction curves, a Bayesian inference calculator determines the most likely fault type, fault probability value, and expected occurrence time, generating a structured fault prediction report. The early warning system extracts the fault type identifier, risk level value, and time urgency indicator from the prediction report, triggering a tiered early warning processing flow. The environmental perception unit collects parameters related to the current task importance of the platform, operating environment conditions, and historical intervention response records to construct a context information database. Based on this database, it dynamically adjusts the warning threshold parameters to create tiered warning thresholds that match the current scenario. It extracts the set of processing solutions with the highest historical effectiveness scores from the intervention strategy database and uses a gradient strategy optimization algorithm to finely calculate parameter adjustments, load control magnitudes, and system switching timings to generate targeted intervention strategies. These intervention strategies are converted into a standard control instruction set, with added security check codes and execution priority markers, and sent to the field control terminal via an encrypted transmission channel. The platform's response information during the intervention process is collected and analyzed in real time, recording intervention effectiveness evaluation data and system recovery indicators. This updates the intervention effectiveness database and threshold parameter table, completing the closed-loop construction of the monitoring and control chain.

[0018] Taking an automated control platform in a chemical plant as an example, the data acquisition terminal captures video streams from the control platform, control valve command streams, and temperature and pressure parameter streams. After applying an inter-frame lossless compression algorithm to the 1920×1080 resolution video stream, only data from changing areas on the interface is recorded, reducing the data volume per second from approximately 6MB to 0.8MB. Microsecond-level timing marking is performed on the control valve command stream, recording the time difference between command issuance and valve response. It was found that the response time of a key valve gradually increased from the standard 15ms to 28ms. After applying wavelet transform filtering technology to the temperature and pressure parameter stream, the ±0.5℃ temperature fluctuation noise was eliminated, accurately capturing the abnormal 0.8℃ temperature increase trend of the reactor over 30 minutes. Based on the current 80% bandwidth utilization and 15ms communication latency, the network transmission engine prioritizes the transmission of control data packets while reducing the sampling frequency of monitoring data packets. The health score matrix output by the dual-stream fusion network after processing the digital twin feature model showed that the stability value of a certain reactor temperature control loop dropped from the normal 92 to 78, and was mapped to the "temperature control system fault" node by the fault knowledge graph, with an anomaly similarity of 89%. The Long Short-Term Memory network predicted that this fault would develop into a serious fault within 8 hours. The early warning system immediately generated an intervention strategy, adjusted the parameters of the backup temperature control loop, and sent a switching command, successfully avoiding a potential production line downtime.

[0019] In this embodiment, a multi-port acquisition terminal simultaneously captures platform video streams, control command streams, and parameter streams, achieving multi-dimensional and comprehensive data acquisition. This provides a rich foundation of raw data for comprehensive monitoring of the platform's status. Targeted processing algorithms are applied to different data types, such as lossless compression for video streams, timing marking for control command streams, and noise suppression for parameter streams, ensuring data quality and transmission efficiency while significantly reducing data storage and transmission burden. The network transmission engine dynamically adjusts transmission strategies based on a network quality assessment matrix and applies multi-layer encryption protocols, effectively improving the security and reliability of data transmission in complex network environments and ensuring the integrity and confidentiality of monitoring data. The decoding system and feature extraction module, combined with a knowledge base and data mining techniques, construct a digital twin feature model that successfully integrates the platform's static data. The system integrates static structural information and dynamic behavioral characteristics, providing accurate feature representations for a comprehensive assessment of the platform's status. The dual-stream fusion network, through parallel processing and attention mechanisms of static and dynamic feature channels, fully leverages the stability of static features and the temporality of dynamic features, generating a health score matrix that reflects the platform's health status from multiple perspectives. Mapping the health score matrix to a fault knowledge graph and combining it with a long short-term memory network for anomaly similarity calculation and fault risk prediction significantly improves the accuracy and foresight of fault prediction, enabling the system to identify potential risks before a fault actually occurs. The early warning system, based on a tiered early warning and intervention strategy generation mechanism triggered by prediction reports, can adaptively adjust response measures according to the risk level and the current platform environment, achieving a shift from passive response to proactive intervention and significantly reducing the fault incidence and impact range. The artificial intelligence algorithms and models applied in this solution, such as inter-frame lossless compression algorithms, wavelet transform filtering technology, dual-stream fusion networks, attention mechanisms, long short-term memory networks, and fault knowledge graphs, not only fully consider the characteristics and needs of data in the industrial control field, but also specifically address the problems of low data processing efficiency, incomplete feature extraction, one-sided state assessment, and inaccurate early warning in traditional monitoring methods. It realizes intelligent operation of the entire process from data acquisition, transmission and processing, feature extraction, state assessment to early warning intervention, enabling the system to maintain high-precision monitoring and prediction capabilities in complex and ever-changing industrial environments, and significantly improving the reliability, security, and maintainability of the automation control platform.

[0020] In one specific embodiment, the process of performing step S101 may specifically include the following steps: (1) The video output interface, control interface and communication interface of the automation control platform are connected simultaneously through the multi-port acquisition terminal to acquire the original video signal, control command signal and operating parameter signal; (2) For the original video signal, the image of each frame is compared and processed by the hardware-accelerated inter-frame lossless compression algorithm to retain the key area information of the image and generate compressed video data. (3) For control command signals, use a high-precision clock source to perform microsecond-level timing marking, record the difference between command sending time and response time, construct a complete command timing chain, and form control command data with timestamps; (4) For the operating parameter signals, wavelet transform filtering technology is used to eliminate environmental interference and electromagnetic noise, extract key operating parameters of the platform, and generate noise reduction parameter data; (5) Synchronize and calibrate the compressed video data, time-stamped control command data and noise reduction parameter data according to a unified time reference, add data format header information and check code, and encapsulate them into multi-source heterogeneous data packets with internal correlation.

[0021] Specifically, the acquisition terminal connects to the platform's VGA, HDMI, or DVI video output interface via a video acquisition card to acquire raw video signals; it connects to the platform's RS-485, Modbus, or Profibus control interfaces via an industrial communication conversion module to capture control command signals; and it connects to the platform's Ethernet, serial port, or fieldbus communication interfaces via a data acquisition unit to acquire operating parameter signals. This multi-port parallel acquisition method ensures the integrity and synchronization of the data source, laying the foundation for subsequent comprehensive analysis. The acquired raw video signals are processed using a hardware-accelerated inter-frame lossless compression algorithm. This algorithm first performs pixel-level comparison of two consecutive frames to calculate the difference image matrix. Subsequently, it uses region segmentation technology to identify the boundaries of the difference regions and extract key regions with significant changes. For these key regions, the algorithm retains complete pixel information; while for unchanged regions, it only stores the region location index. For example, when monitoring an industrial control panel, if only a certain indicator light changes from green to red, the algorithm only needs to record the new pixel value and location information of that indicator light area, without needing to repeatedly store the data of the entire interface. This method reduces the amount of data to 10%-30% of the original video stream without compromising video quality, generating highly efficient compressed video data.

[0022] For processing control command signals, the acquisition terminal uses a built-in high-precision clock source for microsecond-level timing marking. When a control command is detected, its timestamp is recorded. When the corresponding response signal is captured, a timestamp is recorded. ; Calculate the time difference This process marks each instruction issued by the control platform, such as valve opening / closing instructions, motor start / stop instructions, and temperature setting change instructions, forming a complete instruction timing chain. Each instruction data includes fields such as instruction type identifier, instruction content, sending timestamp, response timestamp, and delay value. This time-stamped control instruction data can accurately reflect the response characteristics and instruction execution efficiency of the control system, providing a key basis for fault prediction. For operating parameter signals, the acquisition terminal uses wavelet transform filtering technology for processing. First, the original parameter signal is decomposed into wavelet components. By analyzing the energy distribution characteristics of each frequency domain, high-frequency components characterizing environmental interference and electromagnetic noise are identified. Then, an appropriate threshold function is designed to suppress or remove these high-frequency components. Finally, wavelet reconstruction is performed on each processed frequency component to obtain the noise-removed parameter signal. For example, data collected by a temperature sensor may contain high-frequency noise caused by power fluctuations; wavelet transform can effectively filter out these interferences and retain the true temperature change trend. This technology can extract key platform operating parameters, such as temperature, pressure, flow rate, voltage, and current, from noisy environments to generate high-quality noise reduction parameter data.

[0023] The acquisition terminal synchronizes and calibrates compressed video data, timestamped control command data, and noise reduction parameter data according to a unified time base, eliminating time deviations caused by different acquisition paths. Then, it adds a header with a unified format to each type of data, including fields such as data type identifier, source identifier, timestamp, and data length. Next, it calculates the checksum of the data content, such as CRC-32 or MD5 checksum, to ensure data integrity. Finally, these processed data and their metadata are organized and encapsulated according to a predefined data structure specification, forming multi-source heterogeneous data packets with internal correlation. This data packet structure design enables the subsequent parsing system to accurately identify and process different types of data while maintaining their time synchronization and logical connections. Taking the distillation tower control system of a petrochemical plant as an example, the multi-port acquisition terminal is simultaneously connected to the HDMI video output of the operator station, the Modbus control interface of the field control system, and the Ethernet interface of the temperature sensor network. The original video signal was a full-color image with a resolution of 1920×1080 and a frame rate of 30 frames per second. After processing with an inter-frame lossless compression algorithm, it was found that only the temperature display area and the alarm status indicator area in the operation interface changed frequently, while the rest was basically static. Therefore, the algorithm only retained the real-time image data of these key areas, reducing the video data volume from the original 178MB / minute to 27MB / minute. When timing the control command signals, it was recorded that the temperature regulating valve command usually takes 235 milliseconds from issuance to completion, but in one operation, this value suddenly increased to 412 milliseconds. The command timing chain clearly recorded this anomaly. After filtering the operating parameter signals through wavelet transform, the random fluctuations of ±0.8℃ and the periodic noise caused by 50Hz power supply interference were successfully eliminated, and the slow decreasing trend of the distillation column top temperature of 1.2℃ over 15 minutes was accurately captured. Ultimately, these three types of processed data are encapsulated into a unified multi-source heterogeneous data packet. Each data packet contains 5 seconds of synchronization data and an SHA-256 checksum to ensure the integrity and consistency of the data during transmission.

[0024] In one specific embodiment, the process of performing step S102 may specifically include the following steps: (1) Divide multi-source heterogeneous data packets into three priority levels according to the importance of the data: control data packets, monitoring data packets, and status data packets; (2) The network quality assessment matrix is ​​constructed by collecting the current network bandwidth utilization, communication latency and data packet loss rate in real time through the network status detector; (3) Based on the network quality assessment matrix, allocate transmission resources to data packets of different priority levels to ensure priority transmission of control data packets under network bandwidth constraints; (4) First, a symmetric encryption algorithm is used to process the data content of multi-source heterogeneous data packets. Then, a security protocol is applied to protect the transmission channel. Finally, a secure tunnel is established at the network layer to form a multi-layered protection structure for encrypted data streams. (5) Embed digital signature information and timestamp data in the encrypted data stream, and realize error detection and automatic data recovery during transmission through a segmented verification mechanism.

[0025] Specifically, after receiving multi-source heterogeneous data packets, the network transmission engine first prioritizes them according to their importance. Control data packets, containing instructions that directly control the automation platform, such as equipment start / stop instructions, parameter adjustment instructions, and emergency handling instructions, are given the highest priority. Monitoring data packets, containing key platform performance parameters and operational status information, such as real-time monitoring data of temperature, pressure, flow, and equipment load, are given medium priority. Status data packets, containing auxiliary information and environmental parameters, such as historical trend data, ambient humidity, and the status of non-critical components, are given the lowest priority. This three-tiered prioritization based on data importance ensures that core control functions are not affected when network resources are limited. The network status detector continuously monitors the current network environment quality, employing a distributed probe point strategy. Bandwidth utilization is obtained by statistically analyzing the percentage of available bandwidth used for traffic within a fixed time window, e.g., a current bandwidth utilization of 75%. Communication latency is calculated by sending probe packets and calculating the round-trip time, e.g., an average latency of 18 milliseconds. Packet loss rate is calculated by continuously sending test packets with consecutive sequence numbers and statistically analyzing the proportion of missing packets at the receiving end, e.g., a loss rate of 0.8%. The network status detector organizes these indicators into a network quality assessment matrix. The matrix elements include historical trends in the time dimension and the status of different network segments in the spatial dimension, forming a comprehensive assessment of the network environment.

[0026] Based on real-time data from the network quality assessment matrix, the network transmission engine dynamically adjusts its resource allocation strategy. When good network quality is detected, all three types of data packets are transmitted at their original sampling rate. When network bandwidth utilization exceeds 80% or communication latency increases by more than 30%, the transmission frequency of monitoring and status data packets is automatically reduced, such as decreasing the sampling rate of monitoring data packets from 10 times per second to 5 times per second and extending the update interval of status data packets from 1 second to 5 seconds. When the network is severely congested, status data packet transmission is suspended, and monitoring data packet transmission is reduced to the lowest frequency to ensure the real-time performance and reliability of control data packets. The network transmission engine also implements an adaptive congestion control algorithm, dynamically adjusting the retransmission timeout time and transmission window size according to the timeliness requirements of different priority data packets to ensure that the end-to-end latency of control data packets does not exceed a preset threshold. A multi-layered encryption architecture is used for data content security. First, the AES-256 symmetric encryption algorithm is used to process the data content itself, and the encryption key is updated periodically through a key management center. The encryption process segments the data blocks, with each block being 128 bits in size, and CBC (Cryptographic Block Chaining) mode is used to ensure encryption strength. Subsequently, the transmission channel is protected with the TLS 1.3 security protocol to establish an encrypted communication session, enabling authentication and key negotiation during data transmission. Finally, an IPSec secure tunnel is established at the network layer to encapsulate and encrypt the entire IP data packet, preventing eavesdropping and tampering attacks at the network layer. This triple protection structure, from data content and transmission channel to the network layer, ensures the confidentiality and integrity of data during transmission.

[0027] To enhance the reliability and tamper-proof capabilities of data transmission, additional security information is embedded in the encrypted data stream. Digital signature information is generated based on the asymmetric encryption algorithm RSA-2048, and the digest of the data content is signed with a private key. The receiving end can verify the authenticity and integrity of the data using the public key. Timestamp data is provided by a trusted time source, marking the precise time of data generation and transmission, preventing replay attacks and timing discrepancies. Furthermore, a segmented verification mechanism is implemented, dividing the data stream into fixed-size segments, each appended with a Reed-Solomon forward error correction code. This allows the receiving end to automatically recover partially lost or corrupted data when transmission errors are detected. When a serious error that cannot be recovered through error correction is detected, the system triggers a selective retransmission mechanism, re-requesting only the erroneous data segment, rather than the entire data packet, improving transmission efficiency and reliability. Taking the actual operation of a factory automation control system as an example, after receiving multi-source heterogeneous data packets containing valve control commands, the network transmission engine identifies them as control data packets and assigns them the highest priority. Meanwhile, the network status detector detected that the bandwidth utilization rate of the factory's intranet had reached 87%, the communication latency had increased from the normal 12 milliseconds to 25 milliseconds, and the packet loss rate was 1.2%. Based on this, the network quality assessment matrix showed that the network was in a state of moderate congestion. The network transmission engine immediately adjusted its resource allocation strategy, reducing the transmission frequency of status packets by 50% while maintaining the basic transmission rate of monitoring packets. At the same time, it allocated priority transmission channels for control packets to ensure that valve control commands could reach the execution terminal in a timely and reliable manner. In terms of data security processing, the system first encrypted the valve control commands using the AES-256 key valid for the day, then transmitted them through a TLS 1.3 secure channel, and established an IPSec tunnel between core network nodes. These three layers of encryption ensured that even if the data was intercepted, the original control command content could not be deciphered. At the same time, the system added a digital signature based on the factory's PKI system and a timestamp accurate to milliseconds to the encrypted data, and applied Reed-Solomon encoding to achieve 8-bit data recoverable forward error correction capability, successfully resisting 3-bit data errors that occurred during network jitter, ensuring the integrity and reliability of the control commands.

[0028] In one specific embodiment, the process of executing step S103 may specifically include the following steps: (1) The encrypted data stream is decrypted layer by layer by a multi-layer decryption processor to remove the encryption protection layer and restore the original data structure; (2) For the decrypted video stream data, the frame sequence is reconstructed using the image stream parsing engine. For the control command stream and parameter stream data, the data is separated and the format is restored using the protocol identifier to obtain the original monitoring data. (3) Extract a set of standard operating parameters from the original monitoring data based on the platform type knowledge graph, including operating status identifiers, performance values ​​and environmental variables; (4) The raw monitoring data is sliced ​​into time series by an adaptive time window analyzer to calculate the fluctuation amplitude and rate of change of parameters and identify the unique operating characteristics of the platform; (5) The standard operating parameter set and the platform-specific operating features are weighted and merged to construct a digital twin feature model with both spatiotemporal attributes. The digital twin feature model includes static feature channels and dynamic feature channels.

[0029] Specifically, the received encrypted data stream is decrypted layer by layer using a multi-layer decryption processor. The decryption process follows the reverse order of encryption. First, the IPSec security tunnel at the network layer is processed, and the tunnel encapsulation is decrypted after authentication using a pre-shared key or digital certificate. Then, the TLS 1.3 security protocol at the transport layer is processed, and the content of the transmission channel is decrypted using the session key. Finally, a symmetric key decryption algorithm (such as AES-256) is used to decrypt the data content itself. During the decryption process, the decryption processor also verifies the digital signature and timestamp information of the data to ensure the authenticity and timeliness of the data source. After multi-layer decryption, the encryption protection layer is completely removed, and the data is restored to the original structure of the multi-source heterogeneous data packets, retaining complete metadata such as data type identifiers, data format header information, and checksums. For different types of data streams after decryption, the system uses a dedicated parsing engine for processing. For video stream data, the image stream parsing engine first identifies the video frame boundaries in the data packets, and then decodes them according to the reverse process of the inter-frame lossless compression algorithm. Static areas are reconstructed using position indexes and reference frame data, while changed areas directly use new pixel data. The image stream parsing engine also performs frame sequence synchronization and buffer management, handling potential frame loss or out-of-order issues to ensure a smooth and coherent reconstructed video stream. For control command and parameter streams, the protocol identifier identifies different communication protocols, such as Modbus, PROFIBUS, and OPC UA, based on the data format header information, and then applies the corresponding protocol parsing rules for data separation and format restoration. The protocol identifier extracts information such as the type, content, execution status, and timestamp of each command, as well as attributes such as the name, value, unit, and quality marker of various parameters. Through these professional parsing processes, the system successfully acquires the raw monitoring data, laying the foundation for subsequent analysis.

[0030] The feature extraction module first extracts a set of standard operating parameters from the raw monitoring data based on a platform type knowledge graph. The platform type knowledge graph is a structured knowledge base containing information such as the definitions of standard operating parameters, normal value ranges, and relationships between parameters for different types of automated control platforms. Based on the model and configuration information of the current monitoring platform, the system queries the knowledge graph for matching parameter templates, and then extracts the corresponding operating status identifiers (such as operating mode, start / stop status, alarm level, etc.), performance values ​​(such as throughput, response time, processing capacity, etc.), and environmental variables (such as temperature, humidity, power supply voltage, etc.) from the raw monitoring data. These standard operating parameters constitute the core indicator set for evaluating the basic operating status of the platform, with clear physical meaning and evaluation standards. An adaptive time window analyzer performs time-series processing on the raw monitoring data to identify platform-specific operating characteristics. This analyzer can automatically adjust the analysis window size according to data characteristics, using a smaller time window (e.g., 100 milliseconds) for parameters with high-frequency changes and a larger time window (e.g., 10 minutes) for parameters with low-frequency changes. Within each time window, the statistical characteristics of the parameters are calculated, such as the mean, standard deviation, maximum, and minimum values. Simultaneously, the dynamic characteristics of the parameters are calculated, such as fluctuation amplitude (the difference between the maximum and minimum values), rate of change (first-order difference), acceleration (second-order difference), and periodicity (identified through Fast Fourier Transform). These time-series characteristics can reflect the changing patterns of the platform's operating state at different time scales, revealing dynamic behavioral patterns that cannot be directly expressed by standard parameter sets.

[0031] The system merges and weights standard operating parameters and platform-specific operating characteristics to construct a digital twin feature model. The weighting process is based on expert experience rules and historical data analysis, assigning importance coefficients to different types of features. Standard operating parameters typically receive higher base weights, while platform-specific operating characteristics receive dynamically adjusted weights based on their relevance to the platform's health status. The merged digital twin feature model possesses both spatiotemporal attributes: spatial attributes reflect the correlations and mutual influences between different parameters, forming a static feature channel; temporal attributes reflect the trends and patterns of parameter changes over time, forming a dynamic feature channel. The static feature channel focuses on describing the current state of each platform component and the relationships between components, while the dynamic feature channel focuses on describing the platform's behavioral patterns and state transition characteristics during operation. This dual-channel structure enables the feature model to comprehensively capture the platform's operational status information. Taking the lithography machine control system of a semiconductor manufacturing plant as an example, after receiving the encrypted data stream, the decoding system sequentially decrypts the IPSec tunnel protection, TLS session encryption, and AES-256 data encryption, while verifying the validity of the RSA-2048 digital signature, confirming that the data originates from an authorized acquisition terminal. The image stream parsing engine reconstructs the decrypted video stream, restoring the real-time screen of the lithography machine control interface, clearly displaying the exposure parameter settings and alignment status indicators. The protocol identifier parses the lithography machine's dedicated SECS / GEM protocol data, extracting equipment control commands and operating parameters. Based on the lithography equipment knowledge graph, the system extracts a standard set of operating parameters, including 25 core indicators such as wafer processing status, exposure energy value, alignment accuracy, and chamber pressure. The adaptive time window analyzer uses a 200-millisecond window for pressure parameters, detecting minute fluctuations of ±0.05Pa; uses a 5-minute window for temperature parameters, identifying a slow upward trend of 0.3°C / hour; and uses a 1-minute window for alignment accuracy, capturing periodic offsets synchronized with the vibration source. The system combines these standard parameters and timing features according to a preset weighting scheme to construct a digital twin feature model of the lithography machine. Its static feature channel describes the current working status and interrelationship of each component, while the dynamic feature channel records the changing patterns of key parameters and the equipment response characteristics, providing a comprehensive data foundation for subsequent health assessments.

[0032] In one specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Use a feature separator to split the digital twin feature model into static feature data and dynamic feature data, and input them into the static feature channel and dynamic feature channel respectively; (2) Perform multi-layer deep analysis on static feature data through the spatial correlation extractor in the static feature channel to mine the potential relationship between features and generate spatial feature vectors; (3) The time-series pattern recognizer in the dynamic feature channel performs continuous analysis on the dynamic feature data, extracts the parameter change law, and forms a time-dimensional feature vector; (4) The context-aware computing unit calculates the weight coefficients for the spatial dimension feature vector and the temporal dimension feature vector, and the weight coefficients change dynamically with the platform operating environment; (5) By using a feature mapping converter, the weighted spatial dimension feature vector and the time dimension feature vector are projected onto the same computational space, and data fusion is performed through nonlinear transformation equations; (6) Extract the overall health index of the platform, the status score of core components, the stability value of key links and the abnormal risk warning coefficient from the fused data to form a multi-dimensional health score matrix.

[0033] Specifically, the process of processing digital twin feature models in a dual-stream fusion network begins with a feature separator splitting the model. The feature separator uses a combination of principal component analysis (PCA) and convolutional neural networks (CNNs) to identify and separate static and dynamic features from the model. Static feature data mainly contains relatively stable information such as platform structural attributes, component relationships, and performance parameter baselines, including controller type, sensor configuration, communication interface parameters, and standard operating parameters. Dynamic feature data contains time-varying information such as parameter change trends, response patterns, and state transition characteristics, including temperature fluctuation curves, pressure change rates, and command response delay changes. The feature separator calculates the time sensitivity index of the features, classifying features with sensitivity below a threshold as static features and features with sensitivity above a threshold as dynamic features, thus completing the binary split of the feature model. The separated static and dynamic feature data are then input into static and dynamic feature channels for further processing, respectively. A spatial correlation extractor within the static feature channel performs multi-layer deep analysis on the input static feature data. The spatial correlation extractor consists of a hierarchical neural network, including a feature encoding layer, a correlation detection layer, and a feature fusion layer. The feature encoding layer uses an autoencoder to reduce the dimensionality of the original static features and remove redundant information. The association detection layer constructs an association graph between feature nodes through a graph convolutional network, where each node represents a feature and the edge weights between nodes represent the association strength between features. The association strength is determined by comprehensively calculating multiple indicators such as the correlation coefficient, mutual information, and conditional probability between features. The feature fusion layer aggregates information based on the association graph, and strongly correlated feature groups are identified as functional modules or subsystems. Through this multi-layered analysis process, the spatial association extractor uncovers the potential relationships between static features, such as control loop composition, system dependency chains, and functional groups, and encodes these relationships into fixed-dimensional spatial feature vectors.

[0034] Meanwhile, the time-series pattern recognizer within the dynamic feature channel performs continuous analysis on the dynamic feature data. The time-series pattern recognizer first applies a sliding window technique to the input time-series data to extract feature segments at different time scales. Then, it applies various time-series analysis methods, such as spectral analysis, wavelet transform, and recurrent neural networks, to these feature segments to identify time-domain features such as periodic patterns, trend changes, and abrupt changes. The time-series pattern recognizer also calculates derivative features such as the rate of change, fluctuation amplitude, and phase relationship of each parameter to capture the temporal dependencies and causal relationships between parameters. Finally, the extracted time-series features are integrated into a unified time-dimensional feature vector, which encodes the dynamic behavior patterns and state transition characteristics during platform operation.

[0035] The context-aware computation unit is responsible for calculating weight coefficients for the spatial and temporal feature vectors. This process can be represented by the following mathematical model:

[0036]

[0037] in, This represents the weight coefficient of the i-th spatial dimension feature. This represents the weight coefficient of the k-th time dimension feature. This represents the current environmental complexity index. Let represent the environmental complexity of the i-th spatial dimension feature. This represents the adaptive environmental complexity of the k-th time-dimensional feature. Let represent the correlation matrix between the i-th spatial feature and the j-th environmental factor. This represents the current value of the j-th environmental factor. This represents the sensitivity matrix between the k-th time-dimensional feature and the l-th dynamic condition. This represents the current value of the l-th dynamic condition. , , , , and To adjust parameters, this weighting method based on environmental complexity and conditional correlation enables the system to dynamically adjust the importance of features according to the platform's current operating environment. The feature map converter projects the weighted spatial and temporal feature vectors onto the same computational space and fuses the data through nonlinear transformation equations. The projection process uses tensor mapping technology to uniformly convert feature vectors of different dimensions and scales into tensor representations. The fusion process is achieved through a multi-head attention mechanism and residual connections. The multi-head attention mechanism allows the model to simultaneously focus on multiple feature combinations, capturing the complex interactions between spatial and temporal features. Residual connections preserve the original feature information, preventing information loss during the fusion process. Through this fusion method, the feature map converter generates a comprehensive representation that integrates static structural information and dynamic behavioral features.

[0038] Multiple health indicators are extracted from the fused data. The overall platform health indicator is calculated by applying a global scoring function to the fused data, reflecting the overall operational status of the platform. The core component status score targets critical components, such as the central processing unit, control loops, and communication units, and is calculated using a local scoring function, reflecting the health status of each critical component. The critical link stability value targets important functional links within the system, such as command transmission links, data processing links, and feedback control links, and is calculated through link response testing, reflecting the stability and reliability of the links. The anomaly risk warning coefficient is calculated based on pattern matching and trend extrapolation to predict potential system failure risks. These indicators together constitute a multi-dimensional health scoring matrix, comprehensively reflecting the platform's current and potential future health status.

[0039] Taking the blast furnace automation control platform of a steel plant as an example, after analyzing the digital twin feature model, the feature separator classifies 32 relatively stable features, such as controller configuration, sensor distribution, and tuyere parameters, into static feature data, and 28 time-varying features, such as temperature change trends, pressure fluctuation patterns, and flow response characteristics, into dynamic feature data. The spatial correlation extractor performs multi-layer analysis on the static feature data, discovering a strong correlation (correlation coefficient 0.87) between the furnace bottom temperature sensor parameters and the cooling water flow controller parameters, identifying a complete "temperature-flow" control loop structure, and simultaneously uncovering dependencies between multiple functional groups, such as the "airflow-fuel-temperature" ternary regulation system. These relationships are encoded into a 96-dimensional spatial feature vector. The temporal pattern recognizer analyzes the dynamic feature data, identifying the 8-hour periodic fluctuation pattern of furnace temperature under normal operating conditions, capturing the 15-minute time delay relationship between pressure changes and raw material addition, and discovering a trend where the flow regulation response time gradually increases from the standard 3 seconds to 4.5 seconds. These dynamic features are integrated into a 128-dimensional temporal feature vector. The context-aware computing unit detected that the blast furnace was currently in "high-production mode" and the ambient temperature was high (environmental complexity index of 0.78). Based on this, the average weight coefficient of the spatial dimension features was calculated to be 0.65, and the average weight coefficient of the temporal dimension features was 0.82, reflecting that dynamic features are more important in high-temperature and high-production environments. The feature mapping converter fused the weighted feature vectors to generate a comprehensive feature representation. The multi-dimensional health score matrix extracted from the fused data showed that the overall platform health index was 87 (out of 100), the status score of the cooling water control unit in the core component was 75, significantly lower than other components; the stability value of the "temperature-flow" critical link was 0.82, which is within the acceptable range; the abnormal risk warning coefficient was 0.35, indicating that the system has a moderate risk. These score data indicate that the cooling water control unit needs special attention.

[0040] In one specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Use a feature mapper to project the multidimensional score values ​​in the health score matrix onto the feature space of the fault knowledge graph to establish the coordinate correspondence between the current state and the fault mode; (2) By using a graph traversaler, bidirectional traversal is performed along the concept association path in the fault knowledge graph to calculate the Euclidean distance between the current state feature point and various fault nodes, and an anomaly similarity ranking table is obtained. (3) Based on the multi-scale analyzer, the historical health score data is divided into time periods, and hourly fluctuation features, daily trend features and weekly change features are extracted to form a multi-scale time series feature set; (4) Input the multi-scale temporal feature set into the memory unit network. The memory unit network saves historical state information through a triple structure of forget gate, input gate and output gate, performs temporal extrapolation on key indicators, and generates future state prediction curves. (5) Combining the anomaly similarity ranking table and the future state prediction curve, the most likely fault type, fault probability value and expected occurrence time are determined by the Bayesian inference calculator; (6) Organize the fault type description, risk level score, occurrence probability data and expected time information into a structured document to generate a fault prediction report.

[0041] Specifically, after receiving the multidimensional health score matrix, the feature mapper projects it onto the feature space of the fault knowledge graph. The health score matrix contains multidimensional score values ​​such as the overall platform health index, core component status scores, critical link stability values, and abnormal risk warning coefficients. These indicators constitute a state vector in a high-dimensional space. The feature mapper uses a dimensionality reduction mapping algorithm to convert the high-dimensional state vector into coordinate points in the feature space of the fault knowledge graph, preserving the relative relationships between indicators. The fault knowledge graph is a structured knowledge representation, where each fault mode is represented as a region or point in the feature space, and the similarity and causal relationships between faults are represented as connections between nodes. The feature mapping process establishes a spatial correspondence between the current system state and known fault modes, allowing complex multidimensional states to be directly compared with fault modes in a unified feature space. The graph traversaler performs bidirectional traversal analysis in the fault knowledge graph. Bidirectional traversal includes a forward traversal expanding outward from the current state point and a backward traversal converging inward from each fault node. The forward traversal calculates the Euclidean distance from the current state point to each fault node in the graph. The backward traversal starts from each fault node and calculates the matching degree between its possible previous states and the current state. The Euclidean distance calculation is based on the square root of the sum of squared weighted differences in each dimension of the feature space, with the weights dynamically set according to the importance of the features to fault identification. The graph traverser also considers the correlation strength between nodes, prioritizing traversal along highly correlated paths to ensure that the most relevant fault modes are found. After the traversal process is completed, the system sorts all possible fault modes according to the calculated Euclidean distance values, generating an anomaly similarity ranking table. The table records the type identifier, similarity score, and key feature differences for each possible fault.

[0042] The multi-scale analyzer divides historical health score data into time periods and extracts features. For short-term fluctuations, a sliding window (typically 1-4 hours) is used for hourly analysis to extract high-frequency features such as fluctuation frequency, amplitude, and abrupt changes. For medium-term trends, a smoothing process is performed using a daily window (typically 1-7 days) to extract features such as gradual trends, periodicity, and stability. For long-term changes, an aggregation analysis is performed using a weekly window (typically 1-4 weeks) to extract low-frequency features such as seasonal variations, aging characteristics, and long-period oscillations. Multi-scale analysis employs signal processing techniques such as wavelet transform and empirical mode decomposition to decompose the time series into different frequency components, separating noise and meaningful signals. Through this multi-scale time window analysis, the system can simultaneously capture short-term fluctuations and long-term evolution, forming a comprehensive multi-scale time series feature set. A memory unit network analyzes and predicts the multi-scale time series feature set. This network is based on a Long Short-Term Memory (LSTM) architecture and includes a forget gate, input gate, and output gate triple structure. The forget gate determines which historical information needs to be discarded, using the sigmoid function to calculate the forgetting coefficient and filter out information that is no longer relevant in the long term. The input gate controls the proportion of new information input at the current moment, selectively updating the memory unit state through the combined action of the sigmoid and tanh functions. The output gate determines the final output value based on the current memory unit state and the input information. After multiple rounds of training, the memory unit network can effectively capture the complex patterns of health score indicators changing over time, including long-term dependencies and short-term fluctuations. The system inputs the processed multi-scale temporal feature set into the memory unit network, and through repeated iterative calculations, extrapolates the future states of key health indicators, generating future state prediction curves. Each point on the curve contains the predicted health indicator value and its confidence interval.

[0043] The Bayesian inference calculator combines an anomaly similarity ranking table and future state prediction curves for fault prediction analysis. The calculator first establishes a prior distribution of fault probabilities, setting initial probabilities based on similarity scores in the anomaly similarity ranking table. Then, using the future state prediction curves as evidence, it calculates the posterior probabilities of various fault types. The Bayesian inference process considers the comprehensive matching degree of historical fault data, current state characteristics, and predicted trajectories, handling the competition and uncertainty of multiple potential faults. By calculating the probability distribution of each fault type at different time points, the Bayesian inference calculator determines the most likely fault type, the probability value of the fault occurrence, and the expected time point of occurrence. For fault prediction results with high probabilities, the system also analyzes the main influencing factors and possible evolution paths leading to the fault. The system organizes the key information of the fault prediction into a structured document, generating a fault prediction report. The report contains four main parts: fault type description (detailing the possible fault types, scope of impact, and typical symptoms), risk level score (a comprehensive risk score calculated based on fault probability, scope of impact, and severity), occurrence probability data (including the central probability value and probability distribution interval), and expected time information (the time point and time window when the fault may occur). The report is organized using a standardized XML structure or JSON format, facilitating system parsing and processing. It also includes visual charts such as risk trend graphs, failure probability distribution maps, and key parameter prediction curves, intuitively displaying the prediction results and providing a basis for subsequent early warning and intervention decisions.

[0044] Taking the automatic control system of a reactor in a chemical plant as an example, the feature mapper projects the real-time collected health score matrix (containing 92 indicators) onto the feature space of the fault knowledge graph to locate the coordinates of the current state point. The graph traversaler calculates that the Euclidean distance between this state point and the "temperature control loop fault" node is the smallest (0.28), followed by the "pressure sensor offset" node (0.43), generating an anomaly similarity ranking table containing 15 possible faults. The multi-scale analyzer analyzes the health score data of the past 3 months and finds that the temperature control deviation shows a periodic fluctuation of ±2.5°C in the 2-hour window, a trend of the mean gradually increasing by 0.8°C / day in the 3-day window, and captures the step characteristics related to changes in raw material batches in the 2-week window. After processing these multi-scale features, the memory unit network predicts that the temperature control deviation will exceed the safety threshold within 72 hours, and the control loop response time will increase by 40%. The Bayesian inference calculator integrates this information and determines that "temperature controller servo valve fault" is the most likely fault type (probability 0.82), which is expected to reach a critical state affecting production after 68 hours. The system generates a detailed prediction report, including a detailed description of the fault type, risk level (78 / 100), probability distribution curve, and expected time window (68±5 hours), providing maintenance personnel with sufficient preparation time and clear direction for intervention.

[0045] In one specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Use the risk analyzer to extract the fault type identifier, risk level value and time urgency indicator from the forecast report to trigger the graded early warning process; (2) Collect the current task importance parameters, operating environment condition data and historical intervention response records of the platform through the environmental perception unit to build a context information database; (3) Based on the context information database, the warning threshold is dynamically adjusted to form a hierarchical warning threshold value that matches the current scenario; (4) Extract the set of treatment schemes with the highest historical effect scores from the intervention strategy library, and use the gradient strategy optimization algorithm to perform fine calculations on parameter adjustment amount, load control range or system switching timing to generate targeted intervention strategies; (5) Convert the intervention strategy into a standard control instruction set, add a security check code and an execution priority flag, and send it to the field control terminal through an encrypted transmission channel; (6) Collect and analyze the platform response information in real time during the intervention process, record the intervention effect evaluation data and system recovery indicators, update the intervention effect library and threshold parameter table, and complete the closed-loop construction of the monitoring and control link.

[0046] Specifically, the risk analyzer, as the front-end processing unit of the early warning system, first extracts three key types of information from the fault prediction report: fault type identifier, risk level value, and time urgency indicator. The fault type identifier uses a hierarchical coding method, including equipment category, subsystem category, and specific fault mode. For example, "CT-03-0542" represents "control system - temperature regulation loop - sensor offset fault". The risk level value is based on a standardized scale of 0-100, comprehensively quantifying the fault probability value, impact range, and severity in the prediction report, and classifying them into three levels: low risk (0-30), medium risk (31-70), and high risk (71-100). The time urgency indicator reflects the expected time of fault occurrence, including remaining safe operating time and fault development rate, and is therefore divided into long-term warning (>72 hours), medium-term warning (24-72 hours), and emergency warning (<24 hours). Based on the extracted information, the risk analyzer automatically triggers a tiered early warning process matched to the risk level and urgency. For example, a high-risk emergency warning will initiate the highest-level processing procedure, including immediate notification of key personnel and preparation of emergency measures. The environmental perception unit collects contextual information from multiple data sources to build a comprehensive contextual information database. Task importance parameters are obtained from the production scheduling system, including factors such as the current task's priority, completion rate, planned capacity, and economic value. Operating environment condition data comes from the environmental monitoring system, including physical environmental factors such as temperature, humidity, power quality, and network status, as well as operational environment factors such as staffing and technical support availability. Historical intervention response records are extracted from the system log database, containing intervention measures taken in similar past situations and their effectiveness evaluation data. The environmental perception unit structures this data to generate standardized contextual information entries. Each entry includes a data item identifier, current value, reference range, and importance weight. These entries are indexed and categorized according to equipment type, fault category, and environmental characteristics, forming a contextual information database that can be quickly queried.

[0047] Based on a context information database, the early warning system dynamically adjusts the early warning threshold parameters. First, it extracts a set of baseline thresholds matching the current fault type from a standard threshold database, including default trigger conditions for different early warning levels. Then, it adjusts the sensitivity factor using the task importance parameter from the context information, employing a lower trigger threshold during important task execution to improve early warning sensitivity. It adjusts the fault tolerance factor using environmental condition data, appropriately relaxing the criteria for judging normal fluctuations under harsh environments. Finally, it adjusts the learning factor using historical intervention records, optimizing the current threshold based on the early warning accuracy in similar past scenarios. This multi-factor dynamic adjustment mechanism ensures a high degree of matching between the early warning threshold and the current scenario, avoiding false alarms due to oversensitivity and missed alarms due to undersensitivity. The final generated tiered early warning threshold includes trigger conditions for three levels: minor, general, and severe early warnings. Each level is further subdivided into multiple response strategies based on time urgency. When an early warning is triggered, the intervention decision module retrieves a matching processing plan from the intervention strategy database. The intervention strategy database is built based on historical cases, with each case including a fault characteristic description, the intervention measures taken, an evaluation of the intervention results, and a summary of expert experience. The retrieval process first narrows the search scope based on fault type identifiers. Then, it calculates the similarity between the current fault characteristics and historical cases, selecting the top N cases with the highest similarity as a reference set. The cases in the reference set are then sorted according to their historical effectiveness scores, and the set of treatment solutions with the highest scores is extracted. These solutions are subsequently refined using a gradient strategy optimization algorithm to calculate the optimal parameter adjustments, load control amplitude, or system switching timing for the current situation. The optimization algorithm employs a simulated annealing strategy, gradually approaching the optimal solution through multiple iterations, while considering the balance between intervention costs and expected effects, ultimately generating a customized intervention strategy for the current fault.

[0048] The process of converting intervention strategies into standard control instruction sets involves instruction format conversion, security verification, and priority marking. First, abstract intervention strategies are converted into specific control instructions; for example, parameter adjustment strategies are converted into instructions for setting specific parameters, and load control strategies are converted into instructions for device start / stop or power regulation. Then, a security checksum is added to each instruction, typically generated using the SHA-256 hash algorithm, to ensure the instructions are not tampered with during transmission and execution. Simultaneously, based on fault risk and time urgency, execution priority markers are added to the instructions; higher-priority instructions receive priority in resource allocation and execution. Finally, the instruction set is sent to the field control terminal via an encrypted transmission channel, typically using the TLS 1.3 protocol with a two-way authentication mechanism, to ensure the security and integrity of instruction transmission. Upon receiving the instructions, the field control terminal first verifies the security checksum, then executes the instructions according to the priority marker order, implementing the intervention strategy. Closed-loop monitoring and control are achieved by real-time acquisition of platform response information during the intervention process. This information includes data on the execution status of intervention instructions, actual values ​​after parameter adjustments, changes in system performance indicators, and the evolution of fault characteristics. Through high-frequency sampling and real-time analysis, intervention effectiveness evaluation indicators are calculated, such as fault characteristic mitigation rate, system stability recovery speed, and control accuracy. Simultaneously, the change curves of key indicators during the system recovery process are recorded, including the transition trajectory from abnormal to normal states and the stabilization time. This data is used to update two important knowledge bases: an intervention effectiveness database and a threshold parameter table. The intervention effectiveness database records the detailed process and result evaluation of each intervention, providing a reference for similar situations in the future; the threshold parameter table adjusts the warning thresholds and intervention trigger conditions based on the intervention results, optimizing future warning accuracy. Through this data-driven continuous self-optimization, a complete closed loop is formed between monitoring and control, and the system's fault prediction and intervention capabilities are continuously improved.

[0049] Taking a power system's distribution automation control platform as an example, the risk analyzer extracted the fault type "PC-07-0124" (distribution circuit - overload protection - trigger delay anomaly) from the prediction report, with a risk level of 68 (medium risk) and a time urgency of 36 hours (mid-term warning), triggering a level-two warning process. The environmental perception unit collected data showing that the current task being performed was "urban power supply guarantee" (importance score 92), the ambient temperature reached 34℃ (exceeding the normal operating reference range), and historical records showed that similar faults developed faster under high-temperature conditions. Based on this contextual information, the system dynamically adjusted the standard overload protection trigger delay warning threshold from 25 milliseconds to 18 milliseconds, forming a strict warning threshold for the current high-temperature, high-load scenario. The system retrieved the three highest-scoring historical treatment schemes from the intervention strategy library and calculated the optimal intervention parameters using a gradient strategy optimization algorithm: adjusting the load sharing ratio of the backup line to 40%, increasing the overload protection sensitivity of the main line by 15%, and arranging maintenance personnel to be on standby 2 hours in advance. These strategies were translated into 15 specific control commands, each with a SHA-256 security checksum and an "emergency-3" priority flag, and sent to the substation control terminal for execution via a TLS encrypted channel. After the intervention, real-time monitoring showed a successful reduction in line load, overload protection trigger latency recovering from an abnormal 28 milliseconds to a normal 14 milliseconds, and the fault risk score decreasing from 68 to 23. The complete process and effectiveness data of this successful intervention were recorded and used to update the intervention effectiveness database. Simultaneously, the warning threshold parameters of the overload protection system were optimized, providing valuable experience for handling similar situations in the future.

[0050] The above describes the method for monitoring the operating status of the automation control platform in the embodiments of this application. The following describes the system for monitoring the operating status of the automation control platform in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the automated control platform operation status monitoring system in this application includes: The tagging module is used to capture platform video stream, control command stream and parameter stream through the acquisition terminal, apply lossless compression algorithm to the video stream, perform timing tagging on the control command stream, suppress noise in the parameter stream, and generate multi-source heterogeneous data packets. An encryption module is used to input the multi-source heterogeneous data packets into the network transmission engine, which adjusts the transmission strategy and applies an encryption protocol according to the network quality to form an encrypted data stream. The import module is used to parse the encrypted data stream using the decoding system and import the parsing results into the feature extraction module. The feature extraction module combines a knowledge base and data mining technology to construct a digital twin feature model. An integration module is used to apply a dual-stream fusion network to the digital twin feature model. The dual-stream fusion network processes the static and dynamic feature channels and then integrates them through an attention mechanism to output a health score matrix. The prediction module is used to map the health score matrix to the fault knowledge graph, calculate the anomaly similarity through the inference engine and combine it with the long short-term memory network to predict the fault risk and generate a prediction report. A module is established to trigger an early warning system based on the prediction report. The early warning system generates an intervention strategy according to the risk level and issues an execution command through a secure channel, thus establishing a monitoring and control link.

[0051] Reference Figure 3 This invention also provides a computer device, which can be a server, and its internal structure can be as follows: Figure 3 As shown, the computer device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0052] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0053] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0054] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0055] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0056] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0057] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for monitoring the operational status of an automated control platform, characterized in that, The method for monitoring the operational status of the automated control platform includes: By capturing platform video stream, control command stream, and parameter stream through the acquisition terminal, a lossless compression algorithm is applied to the video stream, timing marking is performed on the control command stream, and noise suppression is performed on the parameter stream to generate multi-source heterogeneous data packets; The multi-source heterogeneous data packets are input into the network transmission engine, which adjusts the transmission strategy and applies an encryption protocol based on network quality to form an encrypted data stream. The encrypted data stream is parsed using a decoding system, and the parsing results are imported into a feature extraction module. The feature extraction module combines a knowledge base and data mining techniques to construct a digital twin feature model. A dual-stream fusion network is applied to the digital twin feature model. After processing the static and dynamic feature channels, the dual-stream fusion network integrates them through an attention mechanism to output a health score matrix. The health score matrix is ​​mapped to the fault knowledge graph, and the anomaly similarity is calculated by the inference engine and combined with the long short-term memory network to predict the fault risk and generate a prediction report. The prediction report triggers an early warning system, which generates an intervention strategy based on the risk level and issues execution instructions through a secure channel, thus establishing a monitoring and control link.

2. The method for monitoring the operating status of an automated control platform according to claim 1, characterized in that, The process involves capturing platform video streams, control command streams, and parameter streams via a data acquisition terminal; applying a lossless compression algorithm to the video streams; performing timing marking on the control command streams; and suppressing noise in the parameter streams to generate multi-source heterogeneous data packets. This includes: The multi-port acquisition terminal is connected to the video output interface, control interface and communication interface of the automation control platform at the same time to acquire raw video signals, control command signals and operating parameter signals. For the original video signal, a hardware-accelerated inter-frame lossless compression algorithm is used to compare each frame of the image, retain key image region information, and generate compressed video data. The control command signal is marked with microsecond-level timing using a high-precision clock source, the difference between the command sending time and the response time is recorded, a complete command timing chain is constructed, and control command data with timestamps is formed. For the aforementioned operating parameter signals, wavelet transform filtering technology is used to eliminate environmental interference and electromagnetic noise, extract key platform operating parameters, and generate noise-reduced parameter data. The compressed video data, timestamped control command data, and noise reduction parameter data are synchronized and calibrated according to a unified time base, and data format header information and check codes are added and encapsulated into multi-source heterogeneous data packets with internal correlation.

3. The method for monitoring the operating status of an automated control platform according to claim 1, characterized in that, The step of inputting the multi-source heterogeneous data packets into the network transmission engine, wherein the network transmission engine adjusts the transmission strategy according to network quality and applies an encryption protocol to form an encrypted data stream, includes: The multi-source heterogeneous data packets are divided into three priority levels according to the importance of the data: control data packets, monitoring data packets, and status data packets. A network quality assessment matrix is ​​constructed by collecting real-time data on current network bandwidth utilization, communication latency, and packet loss rate using a network status detector. Based on the network quality assessment matrix, transmission resources are allocated to data packets of different priority levels to ensure priority transmission of control data packets under network bandwidth constraints. The multi-source heterogeneous data packets are first processed using a symmetric encryption algorithm, then the transmission channel is protected by a security protocol, and finally a secure tunnel is established at the network layer, forming a multi-layered protection structure for the encrypted data stream. Digital signature information and timestamp data are embedded in the encrypted data stream, and error detection and automatic data recovery are achieved during transmission through a segmented verification mechanism.

4. The method for monitoring the operating status of an automated control platform according to claim 1, characterized in that, The encrypted data stream is parsed using a decoding system, and the parsing result is imported into a feature extraction module. This feature extraction module combines a knowledge base and data mining techniques to construct a digital twin feature model, including: The encrypted data stream is decrypted layer by layer by a multi-layer decryption processor to remove the encryption protection layer and restore the original data structure. For the decrypted video stream data, the image stream parsing engine is used to reconstruct the frame sequence. For the control command stream and parameter stream data, the protocol identifier is used to separate the data and restore the format to obtain the original monitoring data. From the original monitoring data, a standard set of operating parameters is extracted based on the platform type knowledge graph, including operating status identifiers, performance values ​​and environmental variables; The original monitoring data is sliced ​​into time series by an adaptive time window analyzer to calculate the parameter fluctuation amplitude and rate of change, and to identify the platform’s unique operating characteristics. The standard operating parameter set and platform-specific operating features are weighted and merged to construct a digital twin feature model with both spatiotemporal attributes. The digital twin feature model includes static feature channels and dynamic feature channels.

5. The method for monitoring the operating status of an automated control platform according to claim 1, characterized in that, The digital twin feature model is applied to a two-stream fusion network. After processing the static and dynamic feature channels, the two-stream fusion network integrates them through an attention mechanism to output a health score matrix, including: The digital twin feature model is split into static feature data and dynamic feature data using a feature separator, and then input into the static feature channel and the dynamic feature channel respectively. The static feature data is subjected to multi-layer deep analysis by a spatial correlation extractor within the static feature channel to uncover potential relationships between features and generate spatial dimensional feature vectors. The time-series pattern recognizer in the dynamic feature channel performs continuous analysis on the dynamic feature data, extracts the parameter change patterns, and forms a time-dimensional feature vector; The context-aware computing unit calculates weight coefficients for the spatial and temporal feature vectors, and the weight coefficients change dynamically with the platform's operating environment. By using a feature map-transformer, the weighted spatial eigenvectors and temporal eigenvectors are projected onto the same computational space, and data fusion is performed through nonlinear transformation equations. The platform's overall health indicators, core component status scores, critical link stability values, and abnormal risk warning coefficients are extracted from the fused data to form a multi-dimensional health scoring matrix.

6. The method for monitoring the operating status of an automated control platform according to claim 1, characterized in that, The process of mapping the health score matrix to a fault knowledge graph, calculating anomaly similarity through an inference engine, and predicting fault risk using a long short-term memory network to generate a prediction report includes: The multidimensional score values ​​in the health score matrix are projected onto the feature space of the fault knowledge graph using a feature mapper to establish the coordinate correspondence between the current state and the fault mode. The graph traversaler performs bidirectional traversal along the concept association path in the fault knowledge graph, calculates the Euclidean distance between the current state feature point and various fault nodes, and obtains the anomaly similarity ranking table. Based on the multi-scale analyzer, the historical health score data is divided into time periods, and hourly fluctuation features, daily trend features and weekly change features are extracted to form a multi-scale time series feature set. The multi-scale temporal feature set is input into the memory unit network. The memory unit network saves historical state information through a triple structure of forget gate, input gate and output gate, performs temporal extrapolation on key indicators, and generates future state prediction curves. By combining the aforementioned anomaly similarity ranking table with the future state prediction curve, the most likely fault type, fault probability value, and expected occurrence time are determined using a Bayesian inference calculator. The fault type description, risk level score, probability of occurrence data, and estimated time information are organized into a structured document to generate a fault prediction report.

7. The method for monitoring the operating status of an automated control platform according to claim 1, characterized in that, The system triggers an early warning system based on the prediction report. The early warning system generates an intervention strategy according to the risk level and issues execution instructions through a secure channel, establishing a monitoring and control link, including: The risk parser is used to extract the fault type identifier, risk level value and time urgency index from the prediction report to trigger a graded early warning process. The environmental perception unit collects current task importance parameters, operating environment condition data, and historical intervention response records to build a context information database. Based on the context information database, the warning threshold is dynamically adjusted to form a hierarchical warning threshold value that matches the current scenario; Extract the set of treatment schemes with the highest historical effect scores from the intervention strategy library, and use the gradient strategy optimization algorithm to perform fine calculations on parameter adjustment amount, load control range or system switching timing to generate targeted intervention strategies; The intervention strategy is converted into a standard control instruction set, a security check code and an execution priority flag are added, and the instruction is sent to the field control terminal through an encrypted transmission channel. The platform response information during the intervention process is collected and analyzed in real time, the intervention effect evaluation data and system recovery indicators are recorded, the intervention effect library and threshold parameter table are updated, and the closed-loop construction of the monitoring and control link is completed.

8. An automated control platform operation status monitoring system, used to implement the automated control platform operation status monitoring method as described in any one of claims 1-7, characterized in that, The automated control platform operation status monitoring system includes: The tagging module is used to capture platform video stream, control command stream and parameter stream through the acquisition terminal, apply lossless compression algorithm to the video stream, perform timing tagging on the control command stream, suppress noise in the parameter stream, and generate multi-source heterogeneous data packets. An encryption module is used to input the multi-source heterogeneous data packets into the network transmission engine, which adjusts the transmission strategy and applies an encryption protocol according to the network quality to form an encrypted data stream. The import module is used to parse the encrypted data stream using the decoding system and import the parsing results into the feature extraction module. The feature extraction module combines a knowledge base and data mining technology to construct a digital twin feature model. An integration module is used to apply a dual-stream fusion network to the digital twin feature model. The dual-stream fusion network processes the static and dynamic feature channels and then integrates them through an attention mechanism to output a health score matrix. The prediction module is used to map the health score matrix to the fault knowledge graph, calculate the anomaly similarity through the inference engine and combine it with the long short-term memory network to predict the fault risk and generate a prediction report. A module is established to trigger an early warning system based on the prediction report. The early warning system generates an intervention strategy according to the risk level and issues an execution command through a secure channel, thus establishing a monitoring and control link.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the automated control platform operation status monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, the computer program causing a processor, when executed by a processor, to perform the automated control platform operation status monitoring method as described in any one of claims 1 to 7.

Images