Edge node-based distributed IO data preprocessing and feedback system

Abstract

The application relates to the technical field of industrial automation control, and discloses a distributed IO data preprocessing and feedback system based on an edge node, which comprises an initial sampling module, a data processing module, a priority scheduling module, a cooperative control module and a learning optimization module.The initial sampling module collects sampling data based on an adaptive sampling strategy using a distributed IO; the data processing module obtains filtered data and calculates a data quality score based on the filtered data and a trust score; the priority scheduling module calculates a comprehensive priority and outputs a scheduling result based on the comprehensive priority; the cooperative control module executes hierarchical control decisions in combination with neighbor node cooperative information and outputs control actions; and the learning optimization module updates system parameters.Through intelligent data processing, real-time scheduling optimization, distributed cooperation and comprehensive fault processing, the application significantly improves the data quality, real-time performance, reliability and adaptive capacity of a distributed IO system, and provides a perfect technical solution for industrial automation and underground coal mine control.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and more specifically, to a distributed IO data preprocessing and feedback system based on edge nodes. Background Technology

[0002] In the field of industrial automation control, distributed I / O systems, as core components connecting field devices and the control center, are widely used in complex scenarios such as coal mines, intelligent manufacturing, and smart buildings. In underground coal mine control, distributed I / O is often used to collect various types of signals, including digital and analog signals, such as gas concentration, temperature and humidity, and equipment operating status, and to achieve remote communication and control via protocols such as CANopen and EtherCAT. In such scenarios, the number of I / O modules often approaches the expansion limit, and challenges such as strong electromagnetic interference and high real-time data requirements are faced.

[0003] A Chinese patent with authorization announcement number CN116719758B discloses a PROFINET distributed I / O synchronization method, which includes configuring scan lock timers Ttim for the interface module and each I / O module respectively. All modules' Ttim have the same target reload cycle time and target synchronization lock offset time. Each time the synchronization pulse edge arrives, the reload cycle count value of each module's Ttim is dynamically adjusted according to the deviation between the actual synchronization lock offset time and the target synchronization lock offset time, so as to achieve synchronization between the modules' Ttim. Each module calculates and determines the synchronization period in the Ttim reload event and the synchronization pulse edge event, and realizes the synchronization of input and output.

[0004] However, in complex industrial environments such as underground coal mines, when the number of IO modules approaches the expansion limit, the system needs to transmit multiple types of data simultaneously. Existing synchronization methods process data according to a fixed cycle, which cannot distinguish priorities. This may cause urgent data to exceed the safety response threshold due to queuing delays, thus failing to guarantee the reliability and security of control in complex industrial environments. Summary of the Invention

[0005] The purpose of this invention is to provide a distributed IO data preprocessing and feedback system based on edge nodes in order to solve the above-mentioned problems.

[0006] This invention provides a distributed I / O data preprocessing and feedback system based on edge nodes, comprising:

[0007] The initial sampling module sets a trust score for each distributed I / O and uses distributed I / O to collect sampling data based on an adaptive sampling strategy.

[0008] The data processing module filters the sampled data to obtain filtered data, and calculates a data quality score based on the filtered data and the trust score.

[0009] The priority scheduling module calculates a comprehensive priority based on filtered data and data quality scores, and outputs the scheduling result based on the comprehensive priority.

[0010] The collaborative control module, based on the scheduling results and combined with the collaborative information of neighboring nodes, performs hierarchical control decisions and outputs control actions;

[0011] The learning optimization module updates system parameters based on the execution results of control actions and data quality scores.

[0012] Furthermore, the steps for calculating the synthesis priority include:

[0013] The scheduling event intensity of the filtered data is calculated using an exponentially weighted moving average method. The scheduling event intensity at the current moment is the sum of the preset event forgetting factor and the scheduling event intensity at the previous moment and the event indicator. The event indicator is the difference between 1 and the preset event forgetting factor multiplied by the event indicator value at the current moment. When the filtered data exceeds the preset alarm threshold of the IO data channel of the distributed IO, the event indicator value is 1; otherwise, the event indicator value is 0.

[0014] Calculate the overall priority of each IO data channel. The overall priority is the sum of the current scheduling event intensity multiplied by the preset scheduling event intensity weight coefficient, the difference between 1 and the data quality score multiplied by the preset quality score weight coefficient, the current preset control urgency multiplied by the preset urgency weight coefficient, and the business criticality multiplied by the preset criticality weight coefficient.

[0015] Furthermore, the scheduling results include emergency queues and regular queues:

[0016] When the overall priority of the IO data channel is greater than or equal to the preset emergency queue admission threshold, it will be assigned to the emergency queue for processing. The emergency queue adopts the earliest deadline priority scheduling algorithm, and the IO data channel task with the earliest deadline is executed first. The deadline is the current time plus the delay time. The delay time is the preset base delay time divided by the control urgency adjustment factor. The control urgency adjustment factor is the current time preset control urgency multiplied by the preset urgency sensitivity coefficient plus the delay constant term.

[0017] When the overall priority of an IO data channel is less than the preset emergency queue admission threshold, it will be assigned to the regular queue for processing. The regular queue adopts a probabilistic selection mechanism. The probability of each IO data channel obtaining bandwidth is calculated by dividing the exponential function value of the overall priority of each IO data channel by the sum of the exponential function values ​​of the overall priorities of multiple IO data channels participating in the scheduling. The base of the exponential function is the natural constant, and the exponent is the sharpness adjustment parameter multiplied by the overall priority. The sharpness adjustment parameter is the preset base sharpness parameter multiplied by the sharpness adjustment factor. The sharpness adjustment factor is the preset priority difference sensitivity coefficient multiplied by the standard deviation of the overall priorities of multiple IO data channels participating in the scheduling at the current time plus the sharpness constant term.

[0018] Furthermore, the initialization settings for the intensity of scheduled events include:

[0019] When the system starts, a preset initial value for the intensity of scheduling events is allocated according to the type of IO data channel;

[0020] After the system enters the running state, if the abnormal frequency within the preset preheating period is greater than the preset abnormal frequency threshold, the scheduling event intensity benchmark value is increased based on the preset upward adjustment step size; if the abnormal frequency is less than the preset stability threshold, the scheduling event intensity benchmark value is decreased based on the preset downward adjustment step size. The abnormal frequency is the frequency of occurrence of abnormal events, and the abnormal event is the filtered data that exceeds the preset alarm threshold of the IO data channel.

[0021] After the warm-up period ends, a dynamic confidence interval for the scheduling event intensity is established for each IO data channel. The dynamic confidence interval lies between the product of the scheduling event intensity benchmark value and the preset lower boundary coefficient and the product of the scheduling event intensity benchmark value and the preset upper boundary coefficient.

[0022] Furthermore, the hierarchical control decision-making includes a local rapid decision-making layer, a neighbor node collaborative verification layer, and a conservative monitoring and observation layer;

[0023] If the scheduling result is an emergency queue and the data quality score is greater than the preset high threshold, the local fast decision layer directly executes the predefined local control action;

[0024] If the scheduling result is an emergency queue and the data quality score is less than the preset high threshold, or if the scheduling result is a regular queue and the control event intensity is greater than the preset collaborative verification threshold, the neighbor node collaborative verification layer combines the neighbor node collaborative information to execute the control action; whereby, the control event intensity is the absolute difference between the mean of the current filtered data and the mean of the historical filtered data, divided by the standard deviation of the historical filtered data, and then normalized.

[0025] If the scheduling result is a regular queue and the intensity of the control event is less than the preset collaborative verification threshold, the conservative monitoring and observation layer will not execute any control actions.

[0026] Furthermore, the combination of neighbor node collaboration information includes:

[0027] Generate an event summary and exchange it with neighboring nodes. The event summary includes the control event strength and confidence level. The confidence level is the data quality score multiplied by a preset confidence level adjustment factor. Neighboring nodes are edge computing nodes that are in the same network segment, have established stable communication connections, and have overlapping functions or overlapping monitoring areas.

[0028] The number of neighboring nodes whose control event strength is greater than a preset collaborative verification threshold is counted. When the number of neighboring nodes exceeding the preset collaborative verification threshold is greater than or equal to a preset consistency threshold, the multiple neighboring nodes execute coordinated control actions; otherwise, no control actions are executed.

[0029] Furthermore, the step of filtering the sampled data includes:

[0030] Adaptive median filtering and adaptive Kalman filtering are performed on the analog data in the sampled data;

[0031] The adaptive median filter adjusts the window radius based on the signal rate of change and the noise standard deviation. The window radius is a preset base window radius multiplied by a window adjustment factor. The window adjustment factor is a preset signal rate of change weight multiplied by the negative of the normalized signal rate of change, plus a noise standard deviation weight multiplied by the normalized noise standard deviation, plus a preset constant term. The signal rate of change is the absolute value of the difference between adjacent sampling points in the sampled data. The noise standard deviation is the product of the median of the absolute differences between multiple data points within the preset sliding window and the window median, and a preset normal distribution correction coefficient. The window median is the median of all data points within the sliding window.

[0032] Adaptive Kalman filtering is based on observation residuals and updates the observation noise covariance using an exponentially weighted moving average method. The new observation noise covariance is the sum of a preset forgetting factor multiplied by the observation noise covariance of the previous time step plus a residual term. The residual term is the difference between 1 and the preset forgetting factor multiplied by the square of the current observation residual, where the observation residual is the absolute difference between the actual observation value and the predicted observation value.

[0033] Multi-level debouncing processing is performed on the digital data in the sampled data, including pulse width constraint filtering based on the minimum effective pulse width threshold and a multi-window voting mechanism. The multi-window voting mechanism determines the final digital state output according to the majority principle. When the number of occurrences of the high-level state and the low-level state of the digital data are equal, the output state of the previous moment is maintained.

[0034] Furthermore, the steps for calculating the data quality score include:

[0035] The data quality score is 1 divided by the quality combination item, and the quality combination item is 1 plus the weighted power of the negative quality index of the natural constant.

[0036] For analog data, the weighted combination of quality indicators is: the preset observation residual weight multiplied by the negative value of the observation residual, the preset signal-to-noise ratio weight multiplied by the signal-to-noise ratio at the current time, and the preset first trust score weight multiplied by the sum of the trust scores; where the signal-to-noise ratio is the logarithm of the absolute ratio of signal power to noise power to the base 10, multiplied by the preset signal-to-noise constant, the signal power is the mean square value of the signal within the sliding window, and the noise power is the square of the noise standard deviation.

[0037] For digital data, the weighted combination of quality indicators is the sum of the preset glitch probability weight multiplied by the glitch probability, the preset second trust score weight multiplied by the trust score, and the preset consistency weight multiplied by the voting consistency index; where the glitch probability is the number of detected glitch divided by the total number of sampling points in the sliding window, and the voting consistency index is the number of sampling points with consistent states in the preset voting window divided by the total number of sampling points in the preset voting window.

[0038] Furthermore, the system parameters include alarm thresholds and trust scores, and the steps for adaptively updating system parameters include:

[0039] Update the alarm thresholds for each IO data channel, including:

[0040] Preset the initial alarm threshold values ​​for each IO data channel;

[0041] When the false alarm rate in the preset statistics window exceeds the preset false alarm rate threshold, or the false alarm rate exceeds the preset false alarm rate threshold, the threshold base update is triggered.

[0042] For analog data, the alarm thresholds include an upper alarm threshold and a lower alarm threshold. The new upper alarm threshold is the product of a preset dynamic threshold adjustment coefficient and the standard deviation of the sliding statistical window data, plus the mean of the sliding statistical window data. The new lower alarm threshold is the mean of the sliding statistical window data minus the dynamic threshold term, where the dynamic threshold term is the preset dynamic threshold adjustment coefficient multiplied by the standard deviation of the sliding statistical window data. For digital data, the alarm thresholds include an abnormal state change frequency threshold. The new abnormal state change frequency threshold is the product of a preset dynamic frequency abnormality coefficient and the standard deviation of the state change frequency, plus the mean of the state change frequency. The standard deviation of the sliding statistical window data, the mean of the sliding statistical window data, the standard deviation of the state change frequency, and the mean of the state change frequency are calculated based on historical filtered data within the preset sliding statistical window.

[0043] Update the trust score. The new trust score is the preset update rate multiplied by the current trust score, plus 1 minus the preset update rate, multiplied by the current data quality indicator function value. When the data quality score exceeds the preset acceptance threshold, the data quality indicator function value is 1; otherwise, the data quality indicator function value is 0.

[0044] Furthermore, the adaptive sampling strategy includes:

[0045] An initial priority score is established for each distributed IO IO data channel. The initial priority score is the data quality score multiplied by the preset data quality weight, plus the preset control urgency multiplied by the preset control urgency weight, plus the preset business criticality multiplied by the preset business criticality weight.

[0046] Calculate the sampling period, which is the preset basic sampling period divided by the sampling adjustment term. The sampling adjustment term is the product of the preset sensitivity adjustment parameter and the comprehensive priority at the previous moment plus 1.

[0047] The beneficial effects of this invention are as follows: This invention achieves efficient and accurate acquisition of distributed I / O data through multi-dimensional priority evaluation and differentiated sampling mechanisms. Event-driven sampling is used for digital signals, and a dual adaptive strategy based on signal activity is employed for analog signals, significantly reducing system load and network traffic.

[0048] Through a two-stage cascaded processing of adaptive median filtering and adaptive Kalman filtering, impulse noise, Gaussian noise, and various interferences are effectively suppressed, improving data quality and signal-to-noise ratio. A multi-level debouncing mechanism is implemented for digital data to eliminate spurious state changes and ensure signal authenticity. Differential scheduling strategies are established for emergency queues and regular queues. The emergency queue uses the EDF algorithm to ensure hard real-time constraints for critical data, while the regular queue uses Softmax probability allocation to achieve fair and efficient resource allocation, ensuring priority transmission of critical data while avoiding starvation of low-priority data.

[0049] By establishing a dynamic model of scheduling event intensity using the EWMA method, and combining multiple factors such as data quality score and control urgency to calculate comprehensive priority, intelligent resource allocation and scheduling decisions can be achieved.

[0050] By establishing a neighbor node collaborative information exchange mechanism, and through event digest generation, confidence calculation, and statistical fingerprint verification, multi-node verification is achieved to reduce the risk of single-point failures and improve the overall decision-making reliability of the distributed system. A control strategy that prioritizes local decision-making while supplementing it with collaborative verification ensures both response speed and decision reliability. A comprehensive reset logic and control action recording mechanism are established. System parameters are adaptively optimized using historical feedback data, including intelligent updates to alarm thresholds and dynamic maintenance of trust scores, enabling system self-improvement and continuous performance enhancement.

[0051] A hierarchical health monitoring system is established, encompassing comprehensive assessments of hardware health, communication health, data health, and functional health. This system ensures reliable operation by combining preventative maintenance, proactive fault handling, and reactive fault recovery. Data integrity is guaranteed through digital signatures and HMAC algorithms, and a hierarchical key management system is established. Signature verification and timeliness checks prevent data tampering and replay attacks. When equipment fails, control tasks are automatically reallocated and system topology is adjusted to ensure no control blind spots and maintain basic operational safety even in failure situations.

[0052] This invention significantly improves the data quality, real-time performance, reliability, and adaptability of industrial distributed I / O systems through intelligent data processing, real-time scheduling optimization, distributed collaboration, self-learning mechanisms, and comprehensive fault handling, providing advanced technical solutions for industrial automation and underground coal mine control. Attached Figure Description

[0053] Figure 1 This is a module example diagram of the distributed IO data preprocessing and feedback system based on edge nodes according to the present invention;

[0054] Figure 2 This is an example diagram of the computational data quality scoring of the edge node-based distributed IO data preprocessing and feedback system of the present invention;

[0055] Figure 3 This is an example diagram of the output scheduling results of the distributed IO data preprocessing and feedback system based on edge nodes of the present invention;

[0056] Figure 4 This is an example diagram illustrating the control actions of the distributed IO data preprocessing and feedback system based on edge nodes according to the present invention. Detailed Implementation

[0057] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0058] Example 1:

[0059] Distributed I / O data preprocessing and feedback system based on edge nodes, such as Figure 1 As shown, it includes:

[0060] The initial sampling module 101 sets a trust score for each distributed IO and uses distributed IO to collect sampling data based on an adaptive sampling strategy.

[0061] The initial sampling module, as the starting point of the system, is responsible for establishing the basic conditions for system operation. Its output sampled data and system parameters are directly passed to the data processing module for subsequent processing. Through a systematic initialization process and intelligent sampling strategy, this module lays a solid operational foundation for the entire distributed IO data preprocessing system.

[0062] Distributed I / O modules are intelligent data acquisition devices deployed at various locations in industrial sites. Each distributed I / O module integrates multiple I / O data channels, capable of independently completing the acquisition, preliminary processing, and network transmission of field signals. I / O data channels refer to the data acquisition and transmission channels between various I / O devices such as sensors and actuators and the distributed I / O module, including analog input channels, digital input channels, analog output channels, and digital output channels. Each distributed I / O module typically contains 8-32 I / O data channels, communicating with edge computing nodes via fieldbus or Ethernet to form a hierarchical distributed data acquisition network architecture.

[0063] When the system starts up, the initial sampling module first performs a comprehensive initialization process, establishing reliable basic parameters and operating environment for the subsequent data processing pipeline through a systematic detection, evaluation, and calibration process:

[0064] The system automatically discovers all connected distributed I / O modules through network scanning and device enumeration mechanisms, obtaining basic parameters such as module device identification, model information, and channel configuration. Subsequently, a comprehensive self-test is performed on each I / O data channel, including channel connectivity testing, signal integrity detection, and basic function verification. Channel connectivity testing verifies the communication link by sending test signals; signal integrity detection checks electrical characteristics such as signal amplitude and frequency response; and basic function verification verifies core functions such as data acquisition, conversion, and transmission. The self-test process generates a detailed I / O data channel status table, recording the availability status, signal type label, measurement range, sampling accuracy, response time, and other basic characteristic parameters of each channel. The signal type label includes analog and digital signals. The I / O data channel status table provides a basis for subsequent sampling strategy formulation.

[0065] Based on self-test results and historical operational data, the system establishes an initial trust score for each distributed I / O module. The trust score is represented by a continuous value between 0 and 1, where 1 represents completely trustworthy and 0 represents completely untrustworthy. The initial value is uniformly set to a medium trust level of 0.5. The trust score is calculated using a four-dimensional comprehensive evaluation system: data quality stability (based on data comparison and output consistency with neighboring modules); communication reliability (based on network packet loss rate and latency jitter); fault recovery capability (based on self-test pass rate and anomaly handling capability); and historical performance consistency (based on historical failure rate and performance stability). This score not only reflects the reliability of data quality in each distributed I / O module, but more importantly, it establishes a trust mechanism for mutual verification and dynamic adjustment between modules. This provides crucial prior information for data quality score calculation in the data processing module and allows the learning optimization module to continuously and dynamically adjust based on the module's actual operational performance.

[0066] Considering the geographically dispersed deployment of distributed I / O modules, the system employs a hierarchical timing synchronization mechanism to ensure time consistency among modules. First, edge computing nodes serve as local time bases, acquiring standard time via high-precision time synchronization protocols such as GPS, NTP, or IEEE 1588. Then, each distributed I / O module periodically calibrates its clock with the edge computing nodes through time synchronization messages, calculating and compensating for network transmission latency. The estimated timing deviation is obtained by subtracting the module's local time from the system reference time. This deviation estimate is not only used for data timestamp correction but, more importantly, provides a unified timing basis for multi-node collaborative decision-making in subsequent collaborative control modules, ensuring that all nodes in the distributed system can coordinate their work based on a consistent time reference.

[0067] After completing system initialization, the initial sampling module adopts an intelligent adaptive sampling strategy. This strategy takes the IO data channel status table and trust score as input, and achieves efficient and accurate collection of distributed IO data through multi-dimensional priority evaluation and differentiated sampling mechanism.

[0068] The system establishes a comprehensive initial priority score for each IO data channel, which is based on a weighted fusion calculation of three core dimensions. The initial priority score is calculated as follows: the initial priority score is the data quality score multiplied by a preset data quality weight, plus the preset control urgency multiplied by a preset control urgency weight, plus the preset business criticality multiplied by a preset business criticality weight.

[0069] The control urgency is determined based on the I / O data channel's position and response time within the control loop to determine its real-time requirements. The calculation method is as follows: control urgency equals the normalized loop position importance multiplied by a preset loop position weighting coefficient, plus the normalized response time urgency multiplied by a preset response time weighting coefficient. The normalized loop position importance is based on control loop topology analysis, categorizing I / O data channels into five levels according to their role in the control loop: primary control variable importance is 1.0, key feedback variable importance is 0.8, auxiliary control variable importance is 0.6, monitoring variable importance is 0.4, and status indicator variable importance is 0.2. The normalized response time urgency is determined based on the ratio of the required response time of the I / O data channel to the system's standard response time; a shorter response time requirement indicates higher urgency. The default value for the preset loop position weighting coefficient is 0.6, and the default value for the preset response time weighting coefficient is 0.4; the sum of these weighting coefficients equals 1.

[0070] Business criticality is assessed based on the importance and safety impact of the IO data channel in the entire production process. The calculation method is as follows: Business criticality equals normalized production importance multiplied by a preset production importance weighting coefficient, plus normalized safety impact multiplied by a preset safety impact weighting coefficient. Normalized production importance is determined based on the impact of the IO data channel on production continuity and product quality, and is divided into five levels: 1.0 for core production variables, 0.8 for critical production variables, 0.6 for important production variables, 0.4 for general production variables, and 0.2 for auxiliary production variables. Normalized safety impact is determined based on the potential impact of IO data channel failures or anomalies on personnel safety, equipment safety, and environmental safety, and is also divided into five levels: 1.0 for severe safety impact, 0.8 for major safety impact, 0.6 for moderate safety impact, 0.4 for minor safety impact, and 0.2 for no safety impact. The default value for both the preset production importance weighting coefficient and the preset safety impact weighting coefficient is 0.5, and the sum of these weighting coefficients equals 1.

[0071] The weight parameters for each dimension are configured according to specific application scenarios and security requirements. The default values ​​for data quality weight, control urgency weight, and business criticality weight are 0.4 and 0.2, respectively. The sum of the weights for the three dimensions equals 1, ensuring that the initial priority score is within the range of 0 to 1. The resulting initial priority score not only guides the current sampling strategy but also provides a basic priority reference for the priority scheduling module. Furthermore, it is dynamically refined and updated during system operation through a learning optimization module.

[0072] The system implements a dynamic sampling period adjustment mechanism to achieve intelligent allocation of sampling resources. The adjustment algorithm first determines the boundary constraints of the sampling period, namely the minimum and maximum sampling period limits. Within these constraints, the sampling period is calculated using an adaptive formula, where the sampling period is the preset base sampling period divided by a sampling adjustment term, which is the product of a preset sensitivity adjustment parameter and the overall priority at the previous moment, plus 1. This formula ensures that high-priority I / O data channels receive a higher sampling frequency, while low-priority channels use a lower sampling frequency to save system resources and network bandwidth. The sensitivity adjustment parameter controls the degree of influence of priority on the sampling frequency, enabling continuous optimization of the sampling strategy; the default value is 1.5.

[0073] Considering the different characteristics of digital and analog signals, the system employs differentiated sampling strategies to improve sampling efficiency and data quality. For digital signals, the system uses an event-driven sampling mechanism, detecting signal edge changes (including rising and falling edges) through hardware interrupts or software polling, triggering sampling and data transmission only when the state changes. This mechanism effectively avoids invalid sampling of stable digital signals, significantly reducing system load and network traffic. When a change in the state of a digital signal is detected, the system immediately captures the change event and records a precise timestamp. The sampling result, along with timing information, is transmitted to the data processing module for debouncing and validity verification. For analog signals, the system employs a dual adaptive sampling strategy based on signal activity and overall priority. The signal change rate is obtained by calculating the difference between adjacent sampling points; a larger change rate indicates a more active signal, requiring a higher sampling frequency to accurately capture the signal's dynamic characteristics. The system weights and fuses the signal change rate with overall priority, allocating more sampling resources to important and active analog channels to ensure high-fidelity acquisition of critical signals. The sampled analog data, carrying complete timestamps and quality identifiers, is transmitted to the data processing module for multi-level filtering and quality assessment, providing a high-quality data foundation for subsequent priority scheduling and coordinated control.

[0074] Data processing module 102 filters the sampled data to obtain filtered data, and calculates a data quality score based on the filtered data and the trust score, specifically as follows: Figure 2 As shown;

[0075] The data processing module, as the core data processing component of the system, receives the raw sampled data and system parameters output by the initial sampling module. Through unified data adaptation, intelligent multi-level filtering, and a comprehensive quality assessment mechanism, it outputs high-quality filtered data and data quality scores, providing reliable data input for the priority scheduling module. This module employs a categorized processing strategy, implementing differentiated preprocessing procedures based on the different characteristics of analog and digital data to ensure optimal processing results for all types of I / O data channels.

[0076] The data processing module first receives the sampling data transmitted by the initial sampling module. This sampling data includes a complete timestamp, IO data channel identifier, signal type marker, sampled value, and a trust score provided by the initial sampling module. Based on the signal type information recorded in the IO data channel status table, the system classifies the received data stream into analog and digital data, preparing for subsequent differentiated processing.

[0077] Based on data classification, the system performs an initial quality assessment of the raw data using a windowed statistical analysis method, establishing the parameter basis for filtering. This assessment process employs a sliding window mechanism, with the window length dynamically adjusted based on the sampling frequency and signal characteristics of the IO data channel to ensure the effectiveness and real-time performance of the statistical analysis. The sliding window length is calculated by multiplying the base window length by a sampling frequency adjustment factor. The base window length is set as the minimum time span containing sufficient statistical samples, typically 10-50 sampling points. The sampling frequency adjustment factor is determined based on the ratio of the actual sampling frequency of the current IO data channel to the standard sampling frequency, ensuring that IO data channels with different sampling frequencies can obtain appropriate statistical windows.

[0078] For analog data, the system uses the median deviation (MAD) method to estimate the noise standard deviation. This method has excellent robustness and can effectively resist the influence of outliers. The calculation process of the noise standard deviation is as follows: first, calculate the absolute deviation of all data points within the sliding window from the median of the window; then, take the median of these deviations; finally, multiply by the normal distribution correction factor of 1.4826 to obtain the estimated value of the noise standard deviation, where the median of the window is the median of all data points within the sliding window. Based on the noise standard deviation, the system further calculates the signal-to-noise ratio (SNR). Specifically, take the common logarithm of the absolute ratio of signal power to noise power, and then multiply it by a preset SNR constant to obtain the SNR in decibels. The default value of the preset SNR constant is 20. In this embodiment, the common logarithm is the logarithm with base 10. The signal power is obtained by calculating the mean square value of the signal within the sliding window, while the noise power is calculated based on the square of the noise standard deviation. A higher SNR indicates better signal quality and less noise interference. This indicator is not only used to judge the reliability of the data, but also provides an important basis for the adaptive adjustment of subsequent filtering parameters.

[0079] For digital data, the system assesses signal stability through a glitch detection mechanism. A glitch refers to a very short-lived, spurious signal change, typically caused by electromagnetic interference, poor contact, or transient disturbances during signal transmission. Glitch detection is based on the minimum pulse width constraint principle. The system sets a minimum effective pulse width threshold, determined by the response time and historical interference characteristics recorded in the IO data channel status table. The minimum effective pulse width threshold is calculated by multiplying a preset base pulse width threshold by a preset interference sensitivity adjustment factor. The base pulse width threshold is set based on the physical characteristics and control requirements of the IO data channel, while the interference sensitivity adjustment factor is dynamically adjusted based on historical glitch statistics and environmental interference levels. The glitch probability is equal to the number of detected glitches divided by the total number of sampling points within the sliding window. This probability provides a quantitative basis for subsequent debouncing parameter settings and digital filtering processing.

[0080] Based on the initial quality assessment results, the system performs two-stage cascaded filtering on the analog data. By leveraging the complementary advantages of different filtering algorithms, it achieves comprehensive suppression of various noises and interferences.

[0081] Median filtering, as the first-stage filtering process, is primarily used to suppress impulse noise and sudden interference. The system employs an adaptive window-length median filtering algorithm, dynamically adjusting the filter window size based on the signal's local characteristics to minimize signal delay while maintaining filtering effectiveness. The filter window is symmetrically arranged, extending forward and backward by several sampling points from the current moment. The adaptive adjustment of the filter window radius is based on a comprehensive evaluation of the signal's rate of change and noise standard deviation. When the signal's rate of change is high, a smaller window radius is used to maintain the signal's dynamic characteristics; when the noise standard deviation is high, a larger window radius is used to enhance the filtering effect.

[0082] The window radius is calculated as follows: the window radius is the preset base window radius multiplied by the window adjustment factor. The window adjustment factor is the preset signal rate of change weight multiplied by the negative value of the normalized signal rate of change, plus the noise standard deviation weight multiplied by the normalized noise standard deviation, plus a preset constant term. The default value for the base window radius is 3 sampling points, the default value for the signal rate of change weight is 0.5, the default value for the noise standard deviation weight is 0.5, and the default value for the constant term is 1. The signal rate of change is obtained by calculating the absolute value of the difference between adjacent sampling points. Normalization ensures the comparability of signals with different ranges. This adaptive mechanism ensures that the median filter can be optimized and adjusted according to the real-time characteristics of the signal, providing a preprocessed data basis for subsequent Kalman filtering.

[0083] Kalman filtering, as the second-stage filtering process, uses a state-space model to perform optimal signal estimation, effectively suppressing Gaussian noise and providing signal predictability. This embodiment employs an adaptive Kalman filtering algorithm, which can dynamically adjust filtering parameters according to the statistical characteristics of the signal, adapting to time-varying noise environments and signal characteristics.

[0084] The state-space model of Kalman filtering is based on the physical characteristics and dynamic behavior of the signal. For general industrial signals, the system adopts a second-order state model, where the state vector contains two components: signal value and signal rate of change. The construction of the state transition matrix considers the time evolution characteristics of the signal, where the diagonal elements are set to 1 to represent the continuity of the state, and the off-diagonal elements are determined according to the sampling interval and the dynamic characteristics of the signal. The specific form of the state transition matrix is: the first row is [1, sampling interval], and the second row is [0, 1]. This matrix describes the relationship between the signal value and the signal rate of change. The observation matrix is ​​set to [1, 0], representing the directly observed signal value.

[0085] The initialization of the process noise covariance matrix and the observation noise covariance matrix is ​​based on the results of the initial quality assessment. The process noise covariance matrix reflects the uncertainty of the system model, and its diagonal elements are determined according to the expected amplitude of signal variation and model accuracy. The initial value of the observation noise covariance matrix is ​​set to the square of the noise standard deviation, directly reflecting the measurement uncertainty.

[0086] The recursive processing of Kalman filtering includes two core steps: prediction and update. The prediction step predicts the state at the next time step based on the system's dynamic model. State prediction is achieved by multiplying the state transition matrix by the optimal state estimate from the previous time step. Covariance prediction propagates the covariance from the previous time step through the state transition matrix and adds the influence of process noise covariance. The update step corrects the prediction results using the observation data at the current time step. First, the Kalman gain is calculated, representing the degree of confidence assigned between the predicted and observed values. The Kalman gain is calculated using the prediction covariance, the observation matrix, and the observation noise covariance. Then, the observation residuals are weighted and fused into the state prediction using the Kalman gain to obtain the optimal state estimate, where the observation residuals are the absolute differences between the actual and predicted observations. Finally, the covariance matrix of the state estimate is updated.

[0087] To adapt to time-varying noise characteristics and signal dynamics, the system employs an adaptive method to update the noise covariance parameter. Statistical analysis of the observation residuals provides the basis for adaptive updates. When the observation residuals are consistently large, it indicates increased observation noise or model mismatch, requiring an increase in the corresponding noise covariance. Conversely, when the observation residuals are consistently small, it indicates a decrease in noise levels, allowing for a reduction in the noise covariance to improve filtering accuracy. The adaptive update uses an exponentially weighted moving average method. The update formula for the observation noise covariance is: the new observation noise covariance equals the preset forgetting factor multiplied by the observation noise covariance at the previous time step, plus a residual term. The residual term is the difference between 1 and the preset forgetting factor multiplied by the square of the current observation residual. The preset forgetting factor ranges from 0.9 to 0.99, determined based on statistical analysis balancing filtering stability and adaptive response speed. A larger value indicates a stronger dependence on historical data and a smoother change in noise covariance; a smaller value indicates a faster response to current data and a quicker adaptation to changes in noise characteristics.

[0088] The focus of digital data preprocessing is to eliminate spurious state changes caused by various interferences, ensuring that the output digital signal accurately reflects the actual state of the monitored object. The system employs a multi-level debouncing mechanism, combined with time-domain filtering and logical decision-making, to achieve reliable processing of digital signals.

[0089] Based on the minimum effective pulse width threshold determined in the initial quality assessment, the system performs pulse width-constrained filtering on the digital signal. This filtering mechanism detects the duration of a state change; only state changes with a duration greater than the minimum effective pulse width threshold are considered valid signals. Pulse width detection employs a state counter method. When a state change is detected, a state counter is activated; the counter continuously increments during the state hold period; when the state changes again or the counter reaches the minimum effective pulse width threshold, the validity of the previous state change is determined. If the counter value is greater than or equal to the minimum effective pulse width threshold, the state change is considered valid; otherwise, the state change is identified as a glitch and filtered out.

[0090] The dynamic adjustment of the minimum effective pulse width threshold is based on statistical analysis of the glitch probability. When the glitch probability exceeds the preset upper limit, the system automatically increases the minimum effective pulse width threshold to enhance anti-interference capability; when the glitch probability is below the preset lower limit and the system response speed requirement is high, the system appropriately decreases the minimum effective pulse width threshold to improve response sensitivity. The step size of the threshold adjustment is determined according to the degree of deviation of the glitch probability to ensure the stability and effectiveness of the adjustment process.

[0091] Building upon pulse-width constrained filtering, the system further employs a multi-window voting mechanism to enhance the reliability of digital signal decision-making. This mechanism statistically votes on the digital signal state across multiple time windows, determining the final digital state output based on the majority principle. The preset voting window settings consider the signal's dynamic characteristics and interference patterns; the number of windows is typically set to an odd number to ensure the uniqueness of the voting results. The window length is determined based on the minimum effective state duration of the signal and system response requirements.

[0092] The voting algorithm is implemented as follows: at each voting moment, the system counts the digital signal states in each preset voting window and calculates the number of occurrences of the high-level state and the low-level state; the voting result is determined according to the majority principle, that is, the state that occurs more often is taken as the final output; when the number of occurrences of the high-level state and the low-level state is equal, the system maintains the output state of the previous moment to ensure the continuity and stability of the output.

[0093] The multi-window voting mechanism significantly improves the reliability of digital signal processing through temporal redundancy. Even if a strong disturbance occurs at one moment, it can be corrected by the correct state at other moments. The optimization of the preset voting window parameters is based on historical interference statistics and system performance requirements, and typically includes 10-20 sampling points.

[0094] After differential filtering, the system performs a unified quality score calculation on the filtered data from all IO data channels, establishing a comprehensive data quality assessment system. The data quality score is represented by a continuous value from 0 to 1, where 1 represents the highest quality and 0 represents the lowest quality. This score will serve as an important input parameter for the comprehensive priority calculation in the priority scheduling module.

[0095] Data quality scores are calculated using the Sigmoid function, which possesses favorable mathematical properties, mapping any real number to the interval between 0 and 1, thus achieving a smooth normalization of the quality score. The Sigmoid function is calculated as follows: the data quality score equals 1 divided by the quality combination term, which is 1 plus the natural constant raised to the power of a negative quality index weighted combination. The quality index weighted combination is a linear combination of multiple quality factors, each assigned a weight parameter based on its impact on data quality.

[0096] For analog data, the weighted combination of quality indicators is calculated as follows: Preset observation residual weight multiplied by the negative value of the observation residual, plus a preset signal-to-noise ratio (SNR) weight multiplied by the current SNR, plus a preset first trust score weight multiplied by the trust score. The default value for the preset observation residual weight is 0.3, obtained through quantitative analysis of the impact of observation residuals on data quality. The default value for the preset SNR weight is 0.4, determined based on historical data analysis of the correlation between SNR and system performance. The default value for the preset first trust score weight is 0.3, set through statistical analysis of module reliability assessment. The observation residual reflects the accuracy of the filtering algorithm in predicting the signal; a smaller residual indicates a better filtering effect and higher signal quality.

[0097] For digital data, the weighted combination of quality indicators is calculated as follows: a preset glitch probability weight multiplied by the glitch probability, plus a preset second trust score weight multiplied by the trust score, plus a preset consistency weight multiplied by the voting consistency index. The preset glitch probability weight is negative, with a default value of -0.4, determined based on statistical analysis of the impact of glitch on digital signal reliability; a higher glitch probability indicates lower quality. The preset second trust score weight is positive, with a default value of 0.45, determined based on quantitative analysis of the impact of module reliability on data quality. The preset consistency weight is positive, with a default value of 0.15, reflecting the consistency of multi-window voting, obtained based on historical data analysis of consistency verification results. The voting consistency index is calculated as follows: the voting consistency index equals the number of consistent sampling points within a preset voting window divided by the total number of sampling points within the preset voting window. The window length is dynamically adjusted according to the sampling frequency, typically containing 10-20 sampling points. The glitch probability directly reflects the interference level of the digital signal; a lower probability indicates a more stable and reliable signal.

[0098] The data processing module outputs two parts: filtered data and data quality score, both transmitted to the priority scheduling module in a unified data format. The output data format includes complete information such as the IO data channel identifier, filtered data value, data quality score, timestamp, signal type marker, and processing status identifier. The processing status identifier records the specific processing flow of the data sample, providing a reference for subsequent module processing decisions.

[0099] Filtered data preserves the physical meaning and dimensions of the original data, ensuring that subsequent modules can correctly understand and use it. Data quality scoring provides an important basis for the comprehensive priority calculation of subsequent priority scheduling modules; high-quality data will receive higher processing priority, while low-quality data may trigger additional verification and processing mechanisms.

[0100] Through this unified and comprehensive data preprocessing mechanism, the data processing module provides a high-quality and highly reliable data foundation for the entire distributed IO data preprocessing system, ensuring that subsequent priority scheduling and collaborative control can make decisions based on reliable data, thereby improving the performance and reliability of the entire system.

[0101] Priority scheduling module 103 calculates a comprehensive priority based on filtered data and data quality scores, and outputs the scheduling result based on the comprehensive priority, as detailed below. Figure 3 As shown;

[0102] The priority scheduling module, as the core scheduling decision-making link of the system, undertakes the critical task of intelligently scheduling and allocating high-quality data output from the data processing module according to business needs and real-time requirements. This module receives filtered data and data quality scores from the data processing module, combines them with the initial priority score provided by the initial sampling module, constructs a complete priority evaluation system, and implements differentiated data transmission strategies based on a dual-queue scheduling mechanism. Finally, the scheduling results are transmitted to the collaborative control module for corresponding control decisions. The entire scheduling process forms a complete data pipeline from data input, priority calculation, queue allocation to scheduling output, ensuring that critical data can be transmitted with optimal timing and bandwidth resources.

[0103] The system establishes a complete alarm threshold preset mechanism in the priority scheduling module, pre-setting scientifically reasonable alarm thresholds for each IO data channel as the basis for anomaly detection and scheduling decisions. The alarm threshold system adopts a bidirectional boundary detection mechanism, including two core components: an upper alarm threshold and a lower alarm threshold, forming a complete anomaly detection boundary range. The upper alarm threshold is used to detect anomalies where IO data channel values ​​exceed the upper limit of the normal operating range. When the monitored data exceeds the upper alarm threshold, it indicates that the monitored object may be in an overload, overvoltage, overtemperature, or other excessive abnormal state. The lower alarm threshold is used to detect anomalies where IO data channel values ​​are below the lower limit of the normal operating range. When the monitored data is below the lower alarm threshold, it indicates that the monitored object may be in an underload, undervoltage, undertemperature, or other insufficient abnormal state. The upper and lower alarm thresholds together constitute a symmetrical or asymmetrical anomaly detection interval, providing the system with comprehensive anomaly monitoring capabilities and ensuring the timely detection of various types of operational anomalies.

[0104] The preset alarm threshold is based on a comprehensive analysis at three levels. The first level is the equipment specification benchmark, which determines the basic threshold range based on the equipment's technical specifications, range, and accuracy level recorded in the IO data channel status table, ensuring that the threshold setting conforms to the equipment's physical characteristics and technical limitations. The second level is historical data statistics, which calculates the statistical characteristic parameters of the data distribution, including mean, standard deviation, and quantiles, by analyzing the historical operating data of this type of IO data channel, and establishes an initial threshold based on the statistical distribution. The third level is production safety constraints, which, in combination with specific production process requirements and safe operating procedures, determines the threshold boundaries that meet the requirements of production safety and production continuity.

[0105] For analog I / O data channels, the preset alarm thresholds adopt a modified 3σ principle. The preset upper alarm threshold equals the historical data mean plus the preset initial threshold coefficient multiplied by the historical data standard deviation, and the preset lower alarm threshold equals the historical data mean minus the preset initial threshold coefficient multiplied by the historical data standard deviation. The preset initial threshold coefficient is determined based on the importance of the I / O data channel and production requirements. For critical control variables, the coefficient is set to 2.0, determined based on statistical analysis of strict anomaly detection requirements to ensure timely detection of anomalies. For general monitoring variables, the coefficient is set to 2.5, obtained through quantitative evaluation balancing detection accuracy and false alarm rate, balancing detection sensitivity and system stability. For auxiliary indicator variables, the coefficient is set to 3.0, set through historical data analysis of relaxed detection strategies and optimized resource allocation to reduce unnecessary alarm interference.

[0106] State change frequency is a core parameter for anomaly detection in digital I / O data channels. It is defined as the number of state changes of a digital signal per unit time, typically expressed in times per minute. A state change refers to the transition between high and low levels of a digital signal, including the rising edge transition from low to high and the falling edge transition from high to low. The statistical analysis of state change frequency is based on the count of state change events within a specified time window. The average state change frequency is obtained by dividing the total number of state changes by the length of the time window. This parameter effectively reflects the dynamic characteristics and time-domain behavior patterns of digital signals and is a crucial quantitative indicator for distinguishing normal signal changes from abnormal signal jitter.

[0107] The calculation of state change frequency employs a sliding window statistical method. The window length is determined based on the control cycle and signal characteristics of the digital I / O data channel, typically set to include a time span sufficient for statistical samples to ensure the accuracy and representativeness of the frequency statistics. The system identifies state change events of the digital signal using an edge detection algorithm, recording the timestamp and change type of each state change to establish a time sequence of state change events. Based on this time sequence, the system calculates the number of state changes within a specified time window and, combined with the window length, calculates the state change frequency. The statistics of state change frequency are not only used for anomaly detection at the current moment, but more importantly, they establish a benchmark for the normal state change pattern of the digital I / O data channel through the accumulation of historical data, providing a statistical basis for setting the threshold for abnormal state change frequency.

[0108] For digital I / O data channels, the preset alarm thresholds are mainly used for detecting anomalies in state change patterns. The system statistically analyzes historical state change frequencies to establish a baseline for normal state change patterns. The range of normal state change frequencies is determined based on the 95% confidence interval of historical data. The threshold for abnormal state change frequencies is set to 1.5 times the upper limit of the normal frequency, determined based on statistical analysis of abnormal digital signal patterns. The threshold for abnormal state duration is set as the boundary value of the normal duration range, determined according to the timing requirements of the digital control logic.

[0109] The abnormal state change frequency threshold is a core component of the digital I / O data channel alarm threshold system, specifically designed to detect abnormalities in the state change patterns of digital signals. This threshold is defined as the abnormal detection boundary for the number of digital signal state changes per unit time. When the actual state change frequency exceeds this threshold, the system determines that the digital I / O data channel has an abnormal state change pattern, triggering corresponding alarm and scheduling response mechanisms. The abnormal state change frequency threshold differs fundamentally from traditional analog alarm thresholds in its detection targets and judgment mechanisms. Analog alarm thresholds primarily detect signal amplitude exceeding limits, while the abnormal state change frequency threshold focuses on abnormal changes in the time-domain characteristics of digital signals, including frequent jitter, abnormal oscillations, and excessively frequent state switching—all timing anomalies.

[0110] The calculation of the abnormal state change frequency threshold is based on statistical analysis of historical data and the physical characteristics of digital signals. The system first statistically analyzes the historical state change frequencies of digital I / O data channels within a specified time window, establishing a probability distribution model of the state change frequency. The normal state change frequency range is determined using the 95% confidence interval of historical data, which reflects the statistical regularity of digital signal state changes under normal operating conditions. The abnormal state change frequency threshold is set to 1.5 times the upper limit of the normal frequency. This multiplier is determined based on statistical analysis of abnormal patterns in digital signals, effectively identifying genuine abnormal state changes while avoiding false alarms caused by fluctuations in normal operating conditions. The selection of the multiplier comprehensively considers the noise characteristics of the digital signal, the response requirements of the control system, and the sensitivity requirements of anomaly detection. Verification and analysis using a large amount of industrial field data ensures the accuracy and reliability of anomaly detection.

[0111] During system operation, the abnormal state change frequency threshold, as an important component of the alarm threshold system, works in conjunction with the analog alarm threshold to form a complete IO data channel anomaly detection mechanism. When the actual state change frequency of a digital IO data channel exceeds the abnormal state change frequency threshold, the system records this event as an abnormal state change event and sets the event indicator value to 1 in the scheduling event intensity calculation, directly affecting the comprehensive priority evaluation and resource allocation strategy of that IO data channel. The triggering of the abnormal state change frequency threshold is not only used for alarm indication, but more importantly, it provides the priority scheduling module with a quantitative indicator of the abnormal activity of digital signals, enabling the system to promptly identify and respond to abnormal state change patterns of digital IO data channels, ensuring reliable monitoring and timely processing of critical digital signals.

[0112] The preset alarm thresholds are transmitted to each module of the system through a unified parameter management mechanism, providing a basic standard for the quality assessment of the data processing module, the anomaly detection of the collaborative control module, and the parameter update of the learning optimization module.

[0113] Based on preset alarm thresholds and input filtered data, the system uses the Exponential Weighted Moving Average (EWMA) method to dynamically model the intensity of scheduling events for each IO data channel. This modeling process provides real-time abnormal activity indicators for comprehensive priority calculation.

[0114] EWMA is a time series smoothing technique that assigns decreasing weights to historical data, giving more influence to recent data while preserving historical trends. The strength of scheduled events is calculated recursively. The strength of a scheduled event at the current moment is the sum of a preset event forgetting factor and the event strength at the previous moment, plus the event indicator value. The event indicator value is the difference between 1 and the preset event forgetting factor multiplied by the event indicator value at the current moment.

[0115] The event indication value is determined by an indication function. When the filtered data output by the data processing module exceeds the preset alarm threshold of the corresponding IO data channel, the event indication value is set to 1, indicating that an abnormal event has been detected; otherwise, the event indication value is set to 0, indicating that the data is within the normal range. The event forgetting factor takes a value between 0 and 1, controlling the degree of influence of historical data. The closer the value is to 1, the stronger the dependence on historical data and the smoother the change in the intensity of the scheduled event, which is suitable for environments with high noise. The closer the value is to 0, the faster the response to current data and the more sensitive the change in the intensity of the scheduled event, which is suitable for rapidly changing operating conditions.

[0116] During system startup, the intensity of scheduling events needs to be appropriately initialized to ensure the system can quickly enter a stable scheduling state. The initialization process employs a three-stage progressive strategy, using systematic parameter configuration and adaptive adjustment mechanisms to achieve a smooth transition from cold start to steady-state operation, ensuring that the intensity of scheduling events on each IO data channel accurately reflects its actual abnormal activity level.

[0117] Upon system startup, due to the lack of historical operational data, differentiated initial value settings are required based on the type characteristics and business importance of the IO data channels. The system assigns corresponding initial values ​​for scheduling event strength to different types of IO data channels based on the channel classification information recorded in the IO data channel status table established by the initial sampling module, combined with the initial priority evaluation results.

[0118] For safety-critical I / O data channels, such as emergency shutdown systems, safety interlock circuits, and fire alarms, the initial value for scheduling event intensity is set to a preset safety-critical initial value, with a default value of 0.2. This value is determined based on statistical analysis of the safety system's high sensitivity to abnormal events, and is obtained through quantitative evaluation of historical safety event response times and false alarm rates, ensuring that safety-related anomalies can be responded to in a timely manner.

[0119] For critical I / O data channels, such as control loops for major production parameters and control of key equipment, the initial value of the scheduling event intensity is set to the preset critical control initial value, with a default value of 0.15. This value is obtained based on a balance analysis of the control system's requirements for appropriate response to process anomalies, and is determined through comprehensive optimization calculations of control loop stability and response speed, ensuring control performance while avoiding oversensitivity.

[0120] For monitoring-type IO data channels, such as trend monitoring, status indication, and performance monitoring, the initial value of the scheduling event intensity is set to the preset monitoring-type initial value, with a default value of 0.1. This value is set through statistical analysis of the monitoring system's basic sensitivity to data changes, and is obtained based on the frequency and importance of changes in historical monitoring data, ensuring that important status changes can be detected in a timely manner.

[0121] For auxiliary I / O data channels, such as environmental parameters, device status indicators, and maintenance information, the initial value of the scheduling event intensity is set to the preset auxiliary initial value, with a default value of 0.05. This value is determined based on the assessment and analysis of the impact of auxiliary information on system operation, and is obtained through quantitative analysis of the correlation between auxiliary parameters and main control variables, thus avoiding resource waste while maintaining necessary monitoring.

[0122] Meanwhile, the initial value of the preset event forgetting factor is uniformly set to 0.8. This value is obtained based on the balance analysis of the dependence on historical information during the startup phase and is determined through simulation optimization of system startup stability and data convergence speed. A larger forgetting factor value ensures that the system has better stability during the startup phase when historical data is scarce.

[0123] After the system enters the operational state, a preset warm-up period is established. The default warm-up period is 50 sampling periods, which is determined based on a balance between system parameter convergence analysis and startup time requirements. By collecting actual data changes from each IO data channel, the baseline value of the scheduling event intensity for each channel is dynamically adjusted. The warm-up learning period is a critical stage for the system to transition from theoretical configuration to actual operation.

[0124] During the warm-up period, the system establishes a real-time statistical profile for each IO data channel, recording the following: frequency of abnormal events, statistics of the number of times exceeding the preset alarm threshold; data fluctuation amplitude, quantified by standard deviation and coefficient of variation; data quality score distribution, statistics of the proportion of data at different quality levels; and signal activity index, reflecting the frequency of signal changes.

[0125] Based on the statistical analysis results, the system uses an adaptive baseline value adjustment algorithm to correct the initial value of the scheduling event intensity.

[0126] When the actual abnormal frequency of the IO data channel is higher than the preset abnormal frequency threshold, the default value of the abnormal frequency threshold is 0.1 times / minute, which is determined based on the statistical analysis of abnormal events in the industrial field. The abnormal frequency is the frequency of occurrence of abnormal events, indicating that the channel is more active than expected. The system adjusts the step size according to the preset step size, which is 0.02 by default. This is obtained through the balance analysis of parameter stability and response speed, and the baseline value of its scheduling event intensity is appropriately increased to enhance the sensitivity to abnormalities of the channel.

[0127] When the IO data channel performs stably and the frequency of anomalies is lower than the preset stability threshold (default value is 0.02 times / minute, determined based on historical data analysis of stable system operation), it indicates that the channel is more stable than expected. The system maintains the current baseline value or adjusts it downward according to the preset step size (default value is 0.01) to ensure the conservatism and stability of system parameter adjustments, appropriately reduce the baseline value of scheduling event intensity, and avoid over-responding to normal fluctuations.

[0128] Simultaneously, the preset event forgetting factor is gradually adjusted to a preset target value of 0.9 according to a preset adjustment strategy. This target value is set through statistical analysis that balances stability and response sensitivity. The adjustment process uses a linear interpolation method, with an adjustment step size of 0.002 for each sampling period, determined based on parameter smoothness transition and system stability requirements, to ensure the smoothness of parameter changes and system stability.

[0129] After the warm-up period, the system establishes the normal operating range and long-term dynamic adjustment mechanism for the intensity of scheduling events on each IO data channel based on accumulated statistical information. This stage marks the system's transition from learning mode to normal operating mode.

[0130] The system establishes a dynamic confidence interval for the intensity of scheduling events for each IO data channel. The preset normal range boundary coefficient is [0.5, 2.0], meaning the normal operating range is set as [baseline value × preset lower boundary coefficient, baseline value × preset upper boundary coefficient], where the preset lower boundary coefficient is 0.5 and the preset upper boundary coefficient is 2.0. The boundary coefficients are determined based on the statistical distribution analysis of the preheating cycle operating data. The lower boundary coefficient of 0.5 corresponds to the lower limit of the 95% confidence interval, and the upper boundary coefficient of 2.0 corresponds to the upper limit of the 95% confidence interval, ensuring that the normal range can cover the vast majority of normal operating states.

[0131] When the intensity of actual scheduled events exceeds the normal operating range, the system triggers an anomaly analysis and parameter correction mechanism. The correction mechanism adopts a tiered anomaly handling strategy: for minor exceedances, a parameter fine-tuning mechanism is activated to recalculate the baseline value based on the most recent statistical data; for severe exceedances, a parameter rollback mechanism is activated to roll back the baseline value of the scheduled event intensity to a stable value determined during the warm-up period; for persistent anomalies, a neighbor node verification mechanism is activated to cross-verify data with neighbor nodes through the collaborative control module; and for extreme anomalies, a manual intervention alarm is triggered to notify operations and maintenance personnel for manual analysis and handling.

[0132] The scheduling event intensity obtained by EWMA modeling can smoothly track the trend of abnormal events in each IO data channel, avoid overreaction of the system due to occasional anomalies, and maintain sensitivity to continuous anomalies, providing a stable and reliable metric for anomaly activity for subsequent priority calculation.

[0133] Based on the scheduling event intensity modeling, the system performs comprehensive priority calculation on each IO data channel. This calculation process integrates the initial priority score of the initial sampling module with the current real-time status information to form a dynamically updated priority evaluation system.

[0134] The overall priority is calculated through a weighted combination of multiple factors. The calculation formula is the sum of each factor multiplied by its corresponding weight parameter. Specifically, it is the sum of the following: the intensity of the current scheduling event multiplied by a preset scheduling event intensity weight parameter; the difference between 1 and the data quality score multiplied by a preset quality score weight parameter; the urgency of the current control multiplied by a preset urgency weight parameter; and the business criticality multiplied by a preset criticality weight parameter. Here, the intensity of the current scheduling event, derived from the scheduling event intensity modeling, reflects the real-time anomaly level of the IO data channel; a higher intensity indicates a higher priority for processing. The difference between 1 and the data quality score comes from the data quality score of the data processing module; the difference is used because lower quality data requires higher attention and processing priority. The urgency of the current control reflects the real-time requirements of the IO data channel for the control system's response speed; a higher urgency indicates a lower latency tolerance. The IO data channel business criticality constant reflects the importance and security impact of the channel within the entire business system.

[0135] The weight parameters for each factor are all non-negative, and their settings reflect the system's understanding of the importance of different factors and its inclination towards business needs. For example, a safety-critical system may assign higher values ​​to the scheduling event intensity weight parameter and the business criticality weight parameter to ensure that abnormal events and critical business data receive the highest priority; while a performance optimization system may place more emphasis on the allocation of the quality score weight parameter and the urgency weight parameter.

[0136] Based on the results of comprehensive priority calculation, the system establishes a dual-queue scheduling mechanism to allocate data transmission tasks with different real-time requirements to the corresponding scheduling queues, thereby obtaining scheduling results and implementing differentiated scheduling strategies and bandwidth allocation to ensure that all types of data can obtain transmission services suitable for their characteristics. The scheduling results include an emergency queue and a regular queue.

[0137] The emergency queue is specifically designed to handle high-priority data with strict real-time requirements, employing the Earliest Deadline First (EDF) scheduling algorithm to ensure that hard real-time constraints are met. When the overall priority score is greater than or equal to the emergency queue admission threshold, the corresponding IO data channel data will be allocated to the emergency queue for processing.

[0138] The EDF algorithm is a dynamic priority scheduling strategy that assigns priorities based on the absolute deadlines of tasks. The task with the earliest deadline receives the highest priority and is executed first. The deadline is determined based on the urgency of the I / O data channel control and the business criticality. The deadline equals the current time plus a delay, where the delay is inversely proportional to the urgency of control; the higher the urgency, the smaller the allowable delay.

[0139] The delay time is calculated as follows: the delay time equals the preset base delay time divided by the control urgency adjustment factor. The control urgency adjustment factor equals the current control urgency multiplied by the preset urgency sensitivity coefficient plus a delay constant term, with a default value of 1. The preset base delay time is determined based on the system's standard response time requirements, typically set to 2 to 5 times the system's average processing cycle, with a default value of 100 milliseconds. The preset urgency sensitivity coefficient controls the degree to which control urgency affects the delay time, with a default value of 2.0. A larger coefficient indicates a more sensitive system to changes in control urgency and a lower tolerance for delays in high-urgency tasks. The current control urgency is calculated from the multi-dimensional initial priority assessment in the initial sampling module, ranging from 0 to 1. Using this calculation method, the delay time for the highest urgency task (control urgency 1.0) is approximately one-third of the base delay time, while the delay time for a low-urgency task (control urgency 0.2) is approximately 0.83 times the base delay time, achieving differentiated delay management based on urgency.

[0140] System schedulability is determined by processor utilization. The system is schedulable and can meet the deadlines of all hard real-time tasks when the sum of processor utilization for all tasks is less than 1. Processor utilization equals the sum of the ratios of each task's execution time to its cycle time. Execution time is estimated based on packet size and processing complexity, while the cycle time is determined by the sampling frequency of the I / O data channel. The emergency queue will directly trigger local control actions or neighbor verification mechanisms in the collaborative control module to ensure the fastest possible response to critical events.

[0141] The regular queue processes data whose overall priority score is less than the admission threshold of the emergency queue. It adopts a probabilistic selection mechanism based on the overall priority score and allocates bandwidth through a normalized exponential function (Softmax function) to achieve a fair and efficient resource allocation strategy.

[0142] The Softmax function is a probability distribution function that transforms any real-valued vector into a probability distribution, ensuring that the sum of all probabilities equals 1. The bandwidth allocation probability is calculated as follows: the probability of each I / O data channel receiving bandwidth is equal to the exponential function value of its overall priority score, divided by the sum of the exponential function values ​​of the overall priority scores of all I / O data channels participating in the scheduling. The base of the exponential function is the natural constant e, and the exponent is the sharpness adjustment parameter multiplied by the overall priority score.

[0143] The sharpness adjustment parameter controls the concentration of the probability distribution and is a key parameter for the flexibility of the scheduling strategy. The larger the sharpness adjustment parameter value, the more concentrated the probability of high-priority IO data channels obtaining bandwidth, and the more the scheduling will favor high-priority channels. The smaller the sharpness adjustment parameter value, the more uniform the probability of each IO data channel obtaining bandwidth, and the fairer the scheduling.

[0144] The sharpness adjustment parameter is calculated as follows: it equals the preset base sharpness parameter multiplied by the sharpness adjustment factor. The sharpness adjustment factor equals the preset priority difference sensitivity coefficient multiplied by the current time-based comprehensive priority standard deviation plus a sharpness constant term, with a default value of 1. The preset base sharpness parameter is determined based on the system's scheduling strategy. For safety-critical systems, a larger value is typically set to highlight high-priority data, while for load-balanced systems, a smaller value is typically set to maintain fairness; the default value is 3.0. The preset priority difference sensitivity coefficient controls the degree of influence of priority distribution on the sharpness adjustment parameter; the default value is 2.0. A larger priority difference sensitivity coefficient indicates a more sensitive system to priority differences. The current time-based comprehensive priority standard deviation is obtained by calculating the standard deviation of the comprehensive priority scores of all participating IO data channels, reflecting the dispersion of the priority distribution. When the comprehensive priority standard deviation is large, indicating significant priority differences, the sharpness adjustment parameter is increased, making scheduling more biased towards high-priority channels. When the comprehensive priority standard deviation is small, indicating insignificant priority differences, the sharpness adjustment parameter is decreased, making scheduling more equitable.

[0145] This probabilistic scheduling mechanism ensures the priority transmission of high-priority data while preventing low-priority data from being starved, thus achieving efficient utilization of system resources. The conventional queue provides an ordered data flow for the collaborative control module, supporting comprehensive control decisions and system optimization.

[0146] Through the collaborative operation of the dual-queue scheduling mechanism, the priority scheduling module achieves precise scheduling of data of different types and importance, ensuring that the system can meet the hard real-time requirements of critical data while maintaining overall scheduling efficiency and fairness, thus providing an optimized data transmission foundation for subsequent collaborative control.

[0147] The collaborative control module 104, based on the scheduling result and combined with the neighbor node collaborative information, performs hierarchical control decisions and outputs control actions, specifically as follows: Figure 4 As shown;

[0148] The collaborative control module, as the core of the system's decision-making and execution, receives scheduling results from the priority scheduling module and filtered data from the data processing module. Through a hierarchical control decision-making architecture, combined with collaborative information from neighboring nodes, it achieves fast and reliable control action output. This module adopts a control strategy that prioritizes local decision-making and supplements it with collaborative verification, improving decision reliability while ensuring response speed.

[0149] To improve the overall decision-making reliability of the distributed system, the collaborative control module exchanges collaborative information with neighboring nodes and reduces the risk of single points of failure and misjudgment through a multi-node verification mechanism.

[0150] Each node generates an event digest containing key state information and exchanges it with its neighboring nodes. The event digest includes control event strength, confidence level, and statistical fingerprint. The confidence level is calculated as the data quality score multiplied by a preset trust adjustment factor, where the trust adjustment factor is determined based on the trust score of the corresponding distributed I / O module; the higher the trust score, the closer the adjustment factor is to 1. The statistical fingerprint is generated by digesting key data using the MD5 hash algorithm, ensuring integrity and consistency verification during data transmission.

[0151] The system automatically discovers neighboring nodes and establishes reliable communication links through network broadcasting and response mechanisms. Neighboring nodes are defined as edge computing nodes within the same network segment that have established stable communication connections and whose monitoring ranges overlap functionally or geographically. The communication protocol uses UDP broadcasting for node discovery and TCP connections for reliable event digest exchange. The system maintains a list of neighboring nodes, recording each node's network address, communication status, response time, and reliability score, providing a basis for node selection in collaborative decision-making.

[0152] A k-consistency threshold mechanism is employed for the collaborative judgment of global anomalies. This mechanism counts the number of neighboring nodes whose control event intensity exceeds a preset collaborative verification threshold. When the number of neighboring nodes exceeding the preset collaborative verification threshold reaches or exceeds a preset consistency threshold k, the system considers a global anomaly to have occurred and requires coordinated control actions. The preset consistency threshold k needs to balance the system's sensitivity and robustness. The preset k value is set within the range of 60%-80% of the total number of participating nodes, obtained based on a quantitative evaluation of the global anomaly detection effect and system robustness, ensuring effective detection of global anomalies while avoiding overreaction caused by false alarms from a few nodes.

[0153] Based on the input data fusion results and neighbor node collaboration information, the collaborative control module adopts a hierarchical control decision-making mechanism. According to different combinations of control event intensity and data quality scores, it selects the corresponding control actions in the following sequence to achieve a fast and reliable control response:

[0154] If the scheduling result is an emergency queue and the data quality score exceeds a preset high threshold (typically set to 0.9), obtained through a quantitative assessment of data reliability and control decision reliability, it indicates the detection of a highly reliable emergency anomaly. In this case, the local rapid decision-making layer directly executes predefined local control actions without waiting for confirmation from neighboring nodes, achieving the fastest possible response speed. These predefined local control actions are pre-configured based on the control logic and safety requirements of the IO data channel, including emergency protection measures such as alarm output, equipment shutdown, and valve closure. The execution results of the control actions are monitored through an observation and feedback mechanism, including action execution status, response time, and control effect evaluation.

[0155] When the scheduling result is an emergency queue and the data quality score is less than a preset high threshold, or when the scheduling result is a regular queue and the control event strength is greater than a preset collaborative verification threshold, the system initiates the neighbor node collaborative verification layer. The default value of the collaborative verification threshold is 0.6. The neighbor node collaborative verification layer first sends a verification request to neighbor nodes, containing a local event summary and verification query information. After receiving the verification request, the neighbor node performs anomaly detection based on its own monitoring data and returns the verification result. The verification result includes the neighbor node's control event strength, confidence level, and suggested control action. The local node collects the verification results from all neighbor nodes and uses a k-consistency threshold mechanism to determine whether to execute a control action. If more than k neighbor nodes confirm an anomaly, the corresponding control action is executed; otherwise, a conservative monitoring strategy is adopted, and the system state changes are observed further.

[0156] When the scheduling result is a regular queue, the control event strength is below the co-validation threshold, and the data quality score is normal, the system adopts a conservative monitoring and observation layer. This layer does not perform active control actions but instead increases the monitoring frequency of relevant IO data channels and records system state change trends. The increased monitoring frequency is achieved by sending a sampling frequency adjustment request back to the initial sampling module, ensuring timely detection of potential anomalies. Simultaneously, the system continuously calculates the control event strength and data quality score; once a higher level of control conditions is met, it immediately switches to the corresponding control strategy.

[0157] To ensure the stability and reliability of the control system, the collaborative control module employs multiple protection mechanisms when executing control actions to prevent control oscillations and malfunctions:

[0158] A protection interval is set for each control action to prevent frequent control switching caused by minor fluctuations near the threshold. The protection interval employs a hysteresis design, with a certain difference between the preset uplink and downlink thresholds, forming a control dead zone. The width of the preset protection interval is determined based on the noise level of the IO data channel and the control accuracy requirements, typically set as a multiple of the preset signal standard deviation. The default value for the signal standard deviation multiple is 2-3 times, determined based on statistical analysis balancing noise suppression and control accuracy. A control action is triggered when the intensity of a control event rises from below the preset downlink threshold to above the preset uplink threshold; a reset action is triggered when the intensity of a control event falls from above the preset uplink threshold to below the preset downlink threshold.

[0159] A robust reset logic mechanism is established to ensure the control system can recover from an abnormal state to a normal state. Reset conditions include: the intensity of a control event remaining below a preset reset threshold for a preset time, neighboring nodes confirming the anomaly has been eliminated, and manual reset commands. The preset reset threshold is typically set to 70%-80% of the control trigger threshold, obtained through quantitative assessment of system state recovery confirmation and reset reliability, ensuring that the reset is only executed after the system state has truly returned to normal. The preset reset time delay is determined based on the characteristics of the controlled object. The preset fast response system time range is set to several seconds to tens of seconds, determined through historical data analysis of the fast response system's characteristics and reset safety. The preset slow response system time range is set to several minutes to tens of minutes, determined based on an assessment of the slow response system's stability and reset timeliness.

[0160] The execution process and results of all control actions are recorded in detail, forming a complete control history data set. The records include: trigger time, control event intensity, data quality score, selected control strategy, neighbor node verification results, control action type, execution status, response time, and control effect evaluation. This recorded data serves as observation results fed back to the learning and optimization module for adaptive adjustment of system parameters and continuous optimization of the control strategy. Control effect evaluation is quantified by comparing changes in system state before and after control, including anomaly elimination effectiveness, response time performance, and system stability indicators.

[0161] The learning optimization module 105 updates system parameters based on the execution results of control actions and data quality scores.

[0162] The learning optimization module, as the intelligent core of the system, undertakes the crucial task of adaptively optimizing system parameters based on historical feedback data. This module establishes a complete closed-loop learning mechanism, receiving control actions and observation results from the collaborative control module. Combining this with historical operating data, it continuously optimizes and adjusts the system's key parameters, feeding back the optimized parameters to the initial sampling module, data processing module, priority scheduling module, and collaborative control module. This enables the entire distributed IO data preprocessing system to achieve self-improvement and performance enhancement.

[0163] Based on the established closed-loop learning data system, the learning optimization module adopts a unified parameter optimization framework to make coordinated and consistent adaptive adjustments to the system's key parameters. Guided by the unified optimization framework, the learning optimization module performs categorized optimization of the system's key parameters, ensuring that the adjustment of each parameter not only meets the local optimization objectives but also serves to improve the overall performance of the system.

[0164] As a core parameter for anomaly detection, the alarm threshold directly impacts the system's false alarm and false negative rates. The learning and optimization module, based on the control action execution results fed back from the collaborative control module, employs a multi-level adaptive update strategy to intelligently and dynamically adjust the alarm thresholds for each IO data channel, thereby continuously optimizing the accuracy of anomaly detection.

[0165] The learning optimization module receives the alarm thresholds preset by the priority scheduling module as the basic standard, and performs incremental updates to the preset alarm thresholds based on historical feedback data. The basic update mechanism follows the principle of stability first, ensuring that the threshold adjustment process does not cause drastic fluctuations in system operation, while maintaining adaptability to long-term trends such as production changes and equipment aging.

[0166] The basic update process adopts a conservative adjustment strategy. For analog I / O data channels, the alarm threshold update range is limited to ±5% of the preset alarm threshold, determined based on a comprehensive consideration of system stability requirements and production safety constraints. The update frequency is controlled to no more than once every 24 hours, determined according to the time constant of the production process and the responsiveness of operators, ensuring the stability and predictability of threshold changes. For digital I / O data channels, the update of the abnormal state change frequency threshold also adopts a gradual adjustment method, with a single update range not exceeding 3% of the current threshold, determined based on the stability requirements of the digital control logic, to avoid false alarms or missed alarms caused by sudden threshold changes.

[0167] The learning optimization module establishes a complete evaluation system for control action execution results, providing accurate feedback information for the basic updates of alarm thresholds. The evaluation of control action execution results includes four core dimensions:

[0168] Control response performance is evaluated by measuring the response time and execution accuracy of control actions. Response time metrics include segmented time measurements such as the delay time from the control command to the actuator action, the transmission time from the actuator action to the controlled object's response, and the feedback time from the controlled object's response to sensor detection. Execution accuracy metrics are evaluated by comparing the deviation between the actual control effect and the expected control target, including steady-state accuracy and dynamic accuracy. Steady-state accuracy is the deviation after control stabilizes, while dynamic accuracy is the maximum deviation during the control process.

[0169] Effective control is assessed by analyzing the actual impact of control actions on the system state. Effective control is defined as follows: after the control action is executed, the controlled variable converges to the target value within a preset time window, and the reduction in deviation exceeds a preset effective control threshold (default value is 50% of the initial deviation), determined based on statistical analysis of the control effect's significance. Ineffective control includes three scenarios: no significant change in the controlled variable after the control action is executed; the controlled variable changes in the wrong direction; and the oscillation of the controlled variable intensifies, leading to system instability.

[0170] System stability recovery assessment is achieved by monitoring the impact of control actions on the overall system stability. Stability indicators include dynamic characteristic parameters such as system oscillation amplitude, convergence time, and overshoot. A good control action should enable the system to reach a new steady state within a preset settling time, with a smooth transition without oscillations; the settling time is typically 3-5 time constants.

[0171] Resource consumption assessment is achieved by statistically analyzing the execution cost and resource usage of control actions. Resource consumption includes the number of actuator actions, energy consumption levels, communication bandwidth usage, and computational resource consumption. An efficient control strategy should minimize resource consumption while achieving the control objective.

[0172] Based on a comprehensive evaluation of the results of control actions, the system establishes a multi-dimensional threshold-based update trigger mechanism. The evaluation of trigger conditions adopts a sliding window statistical method, with the statistical window length set to either the most recent 100 control actions or the control history of the most recent 24 hours, selecting the one with the larger sample size to ensure statistical validity.

[0173] The system establishes a complete false alarm identification and statistical system to achieve dynamic monitoring of alarm threshold sensitivity. False alarm identification is based on a multi-dimensional evaluation of the control action execution results, specifically including three levels of judgment criteria:

[0174] The first level is the control effectiveness assessment. When the control action executed after an alarm is evaluated as ineffective, i.e., the controlled variable shows no significant improvement or changes in the wrong direction after the control action is executed, the alarm is marked as a suspected false alarm. The second level is the system stability impact assessment. When the control action leads to a deterioration in system stability indicators, such as increased oscillation amplitude, prolonged convergence time, or increased overshoot, the alarm is further confirmed as a false alarm. The third level is the resource consumption rationality analysis. When the resource consumption of the control action does not match the actual control effect, i.e., high resource consumption but weak control effect, the alarm is ultimately determined to be a false alarm.

[0175] The false alarm rate is calculated by dividing the number of false alarms within the statistical window by the total number of alarm triggers within the statistical window, and then multiplying by 100% to obtain the percentage false alarm rate. When the false alarm rate within the statistical window exceeds the preset false alarm rate threshold, it indicates that the current alarm threshold is too sensitive, and the alarm threshold is relaxed and updated. The default value of the false alarm rate threshold is 5%, which is determined based on statistical analysis that balances anomaly detection accuracy and system response efficiency.

[0176] The system establishes a missed detection mechanism based on multi-source information fusion, identifying anomalies that the system failed to detect in a timely manner through post-event analysis and cross-validation. Missed detection relies on four main information sources:

[0177] Anomalies discovered through manual inspections include equipment malfunctions, deviations in process parameters, and product quality issues found by operators during on-site inspections. These anomalies often fail to trigger corresponding alarms in the system. Missed events are identified through time correlation analysis between manual inspection records and system alarm records. Deviations detected during periodic calibration and testing, including deviations found during regular instrument calibration, equipment testing, and process parameter verification, are analyzed retrospectively when calibration results show significant differences between the actual state and the system test results, exceeding the normal error range, to determine the time point and duration of the missed alarm.

[0178] Inconsistencies in cross-validation between neighboring nodes are identified through correlation analysis with adjacent IO data channels or related process parameters to pinpoint data inconsistencies. When relevant parameters show abnormal trends but the target channel does not trigger an alarm, missed events are confirmed through process mechanism analysis. Historical data backtracking analysis, through in-depth mining and pattern recognition of historical data, reveals abnormal patterns that were not detected at the time but were later proven to exist. These abnormal patterns are confirmed as missed events through statistical analysis and expert knowledge verification.

[0179] The false negative rate is calculated as follows: the total number of false negative events identified through various means within the statistical window is divided by the total number of actual abnormal events occurring within the statistical window, including both detected and missed abnormal events, and then multiplied by 100% to obtain the percentage false negative rate. The total number of actual abnormal events is obtained by statistically analyzing system alarm records, manual inspection records, periodic verification results, and neighbor node verification information. When the false negative rate within the statistical window exceeds the preset false negative rate threshold, it indicates that the current alarm threshold is too lenient, triggering a threshold tightening update. The default value of the false negative rate threshold is 2%, obtained based on a quantitative assessment of the completeness of anomaly detection and system security.

[0180] When the invalid control rate in the statistics window exceeds the preset invalid control rate threshold, it indicates that the alarm threshold setting is improper, leading to frequent invalid control interventions, and triggers a threshold optimization update. The default value for the preset invalid control rate threshold is 30%. It is set through historical data analysis of control effectiveness and resource utilization efficiency.

[0181] When the frequency of control actions causing system stability indicators to deteriorate exceeds a preset stability deterioration threshold, threshold adjustment is triggered to improve control performance. Deterioration of system stability indicators includes increased oscillation amplitude, prolonged convergence time, and increased overshoot. The preset stability deterioration threshold has a default value of 15%, determined based on an analysis balancing system stability requirements and control sensitivity.

[0182] After triggering a basic threshold update, the system employs a sliding window statistical analysis method to adjust the alarm threshold based on the historical data distribution characteristics of the IO data channel. The sliding window statistical analysis method presets a sliding statistical window, the length of which is determined according to the temporal correlation and statistical stability requirements of the data. It is typically set to a time span containing sufficient statistical samples, such as the most recent 1000 valid data points or data from the most recent 24 hours, taking the larger value to ensure statistical validity. Within the sliding window, the system calculates statistical characteristic parameters of the historical filtered data, including mean, standard deviation, skewness, and kurtosis, to describe the central tendency, dispersion, and distribution pattern of the data distribution.

[0183] For analog I / O data channels, the dynamic adjustment of alarm thresholds is based on the normal distribution assumption and anomaly detection theory, combined with statistical feedback on false alarm rate and missed alarm rate to achieve precise adjustment. The formula for updating the upper alarm threshold is: the new upper alarm threshold equals the product of the preset dynamic threshold adjustment coefficient and the standard deviation of the sliding statistical window data, plus the mean of the sliding statistical window data; the formula for updating the lower alarm threshold is: the new lower alarm threshold equals the mean of the sliding statistical window data minus the dynamic threshold term, where the dynamic threshold term is the preset dynamic threshold adjustment coefficient multiplied by the standard deviation of the sliding statistical window data.

[0184] The adaptive update mechanism of the dynamic threshold adjustment coefficient fully considers the statistical feedback of false alarm rate and false negative rate to achieve intelligent threshold adjustment. When the false alarm rate exceeds the preset false alarm rate threshold, it indicates that the current threshold is too sensitive. The coefficient is increased by a preset increment to loosen the threshold and reduce false alarms caused by an overly tight threshold. The default value of the preset increment is 0.1. When the false negative rate exceeds the preset false negative rate threshold, it indicates that the current threshold is too lenient. The coefficient is decreased by a preset decrement step to tighten the threshold and improve the completeness of anomaly detection. The default value of the preset decrement step is 0.05. When the ineffective control rate or the frequency of system stability deterioration exceeds the corresponding threshold, the coefficient is finely adjusted based on the specific control effect evaluation results to ensure that the threshold adjustment can both improve detection performance and maintain system stability.

[0185] The coefficient adjustment employs an improved gradient descent method, with the adjustment step size determined based on the deviation between the false alarm rate and the missed alarm rate, as well as the degree of improvement in control effectiveness. The formula for calculating the adjustment step size is: the current adjustment step size equals the base adjustment step size multiplied by the deviation adjustment factor. This deviation adjustment factor is determined based on the ratio of the actual false alarm rate (or missed alarm rate) to the target false alarm rate (or target missed alarm rate), ensuring the stability and convergence of the adjustment process. To prevent over-adjustment of the coefficients, the system sets the dynamic threshold adjustment coefficient to a range of 1.5 to 4.0, determined based on statistical analysis balancing industrial noise levels and anomaly detection accuracy.

[0186] For digital I / O data channels, the dynamic adjustment of alarm thresholds is mainly reflected in the optimization of anomaly detection parameters such as the frequency and duration of state changes. Intelligent adjustment is also achieved based on statistical feedback of false alarm rate and missed alarm rate. The system statistically analyzes the frequency of state changes within a sliding statistical window, calculates the mean and standard deviation of the state change frequency, and establishes a normal range for the state change frequency. The updated calculation method for the abnormal state change frequency threshold is as follows: the new abnormal state change frequency threshold equals the mean of the state change frequency plus the dynamic frequency anomaly coefficient multiplied by the standard deviation of the state change frequency.

[0187] The adaptive adjustment mechanism of the dynamic frequency anomaly coefficient is similar to that of the analog channel, making full use of statistical information on false alarm rate and missed alarm rate. When the false alarm rate of the digital alarm exceeds the preset false alarm rate threshold, it indicates that the frequency anomaly detection is too sensitive. The coefficient is increased by a preset step size to reduce false alarms and avoid triggering unnecessary alarms due to normal state fluctuations. When the missed alarm rate of the digital alarm exceeds the preset missed alarm rate threshold, it indicates that the frequency anomaly detection is too lenient. The coefficient is decreased by a preset step size to improve detection sensitivity and ensure that true digital anomaly patterns can be detected in a timely manner. The variation range of the dynamic frequency anomaly coefficient is set between 1.5 and 3.0, determined based on statistical analysis of digital signal anomaly patterns and control response requirements, ensuring the stability and effectiveness of the adjustment process.

[0188] Considering the diverse operating conditions in industrial settings and the potentially significant differences in normal data distribution under different conditions, the system establishes a multi-condition basic threshold management mechanism, maintaining independent statistical systems for false alarm and false negative rates for each condition. Condition identification is based on cluster analysis of key production parameters. The system automatically identifies and classifies different operating conditions, establishing an independent alarm threshold system for each. Condition classification employs an improved K-means clustering algorithm, with clustering features including the mean, standard deviation, trend, and correlation of key production parameters. The number of clusters is determined through the elbow rule and silhouette coefficient analysis, typically set to 3-8 condition categories. This is determined based on statistical analysis balancing the complexity of industrial operating modes and management efficiency, ensuring the rationality and effectiveness of condition classification.

[0189] Alarm thresholds for each operating condition are calculated and maintained independently. The system establishes a dedicated threshold parameter file for each operating condition, recording the data distribution characteristics, abnormal modes, historical statistics of control action execution results, and false alarm and missed alarm statistics specific to that operating condition for each IO data channel. False alarm identification standards and missed alarm detection methods remain consistent across different operating conditions, but statistical benchmarks and threshold parameters are set independently according to the characteristics of each operating condition, ensuring the targeted and effective adjustment of thresholds. During operating condition switching, the system automatically loads the alarm threshold parameters and false alarm / missed alarm statistical benchmarks for the corresponding operating condition, enabling rapid threshold adaptation. Operating condition switching detection is based on the Mahalanobis distance calculation between real-time production parameters and the centers of each operating condition. When the Mahalanobis distance between the current production parameters and a certain operating condition center is minimized and less than the preset operating condition identification threshold, the system switches to the corresponding operating condition mode. The preset operating condition identification threshold is determined based on the separation between operating conditions and switching stability requirements, typically set to 1.8 times the standard deviation of the data distribution within the operating condition, and is set through historical data analysis balancing operating condition identification accuracy and switching stability.

[0190] During the switching of operating conditions, the system employs a gradual threshold transition mechanism to avoid abrupt threshold changes impacting system stability. The alarm threshold during the transition period is calculated using a weighted average method. The transition threshold equals the previous operating condition threshold multiplied by its weight, plus the new operating condition threshold multiplied by its weight. The weights are dynamically adjusted based on the confidence level and time progression of the switching operation. The transition time is typically set to 5-15 minutes, determined through a quantitative assessment balancing stability and response timeliness during the switching operation, ensuring the system has sufficient time to adapt to the new operating condition characteristics.

[0191] To prevent frequent changes in alarm thresholds from affecting system stability, the learning optimization module establishes a robust mechanism to ensure the stability of threshold updates. This mechanism, through multi-level constraints and verification, ensures that threshold updates both respond to system performance requirements and maintain system operational stability.

[0192] Threshold updates employ a gradual adjustment strategy, with each update strictly limited to the preset maximum adjustment ratio of the current threshold. For analog I / O data channels, a single threshold adjustment is limited to 10% of the current threshold, determined based on an analysis balancing threshold stability and adaptive response speed. For digital I / O data channels, a single frequency threshold adjustment is limited to 15% of the current threshold, obtained through quantitative evaluation of digital signal characteristics and control response requirements. Large threshold adjustments require multiple gradual updates, with each update interval no less than the preset minimum update interval (default 30 minutes). This interval is set through statistical analysis balancing system adaptation time and response timeliness, ensuring the system has sufficient time to adapt to the new threshold setting and evaluate its effectiveness.

[0193] Threshold updates must undergo a rigorous two-stage validity verification process before officially taking effect. The first stage is historical data backtesting verification. The system uses historical data from the last 72 hours to verify the detection effect of the new threshold. This is combined with manual inspection records from the historical period, periodic verification results, and cross-validation information from neighboring nodes. The expected false positive rate, false negative rate, and control effectiveness indicators are recalculated according to the aforementioned false positive and false negative rate statistical methods. The passing criteria for backtesting verification are: the expected false positive rate does not exceed 120% of the target false positive rate, the expected false negative rate does not exceed 110% of the target false negative rate, and the expected control effectiveness is not lower than 95% of the current level. The second stage is real-time performance evaluation verification. After passing the historical data backtesting verification, the new threshold enters a 2-hour trial operation period. During this period, the system runs both the old and new thresholds in parallel, comparing their actual detection effects and control action execution results. Simultaneously, a real-time false positive and false negative rate statistical mechanism is activated for parallel monitoring. During the trial operation, the actual performance of the new threshold must meet the following conditions to pass the real-time performance evaluation: an actual false positive rate improvement of no less than 5%, an actual false negative rate improvement of no less than 3%, or a control effectiveness improvement of no less than 2%.

[0194] Only the new threshold that passes the two-stage verification can officially replace the current threshold. If either stage of verification fails, the system retains the original threshold setting and records the detailed reasons for the update failure, including the specific indicators of failure, the degree of deviation, and possible directions for improvement, providing a reference for subsequent optimization. After three consecutive update failures, the system automatically starts the threshold parameter diagnostic program to analyze possible systemic problems and generate operation and maintenance suggestions.

[0195] The updated alarm thresholds are synchronized to all relevant modules of the system through a unified parameter feedback mechanism, ensuring that the entire distributed IO data preprocessing system coordinates its work based on consistent anomaly detection standards. Parameter feedback adopts a publish-subscribe model, with the learning and optimization module acting as the parameter publisher and other modules as parameter subscribers, achieving real-time synchronization of parameter updates.

[0196] In the priority scheduling module, the updated alarm threshold is used to recalculate the intensity of scheduling events, directly affecting the priority assessment and resource allocation strategy of each IO data channel. When the alarm threshold is tightened, the anomaly detection sensitivity of the corresponding IO data channel increases, its scheduling event intensity increases accordingly, and it obtains a higher processing priority; when the alarm threshold is relaxed, the scheduling event intensity decreases accordingly, making room for other more urgent IO data channels to process resources.

[0197] In the collaborative control module, the updated alarm thresholds serve as a crucial basis for anomaly event judgment, directly impacting the triggering conditions and control strategy selection for control decisions. The fundamental updates to the thresholds enable the collaborative control module to formulate control decisions based on more accurate anomaly detection results, reducing unnecessary control interventions caused by false alarms and improving the timeliness of anomaly response to missed alarms.

[0198] In the data processing module, updating the alarm threshold provides a more accurate reference standard for calculating the data quality score, enabling the quality score to more accurately reflect the actual status of the IO data channel and provide a more reliable basis for subsequent data processing and decision-making.

[0199] This unified and coordinated threshold management mechanism not only improves the local performance of each module, but more importantly, it achieves global optimization of the entire system, ensuring that the distributed IO data preprocessing system can make data processing and control decisions based on consistent anomaly detection standards, thus significantly improving the system's coordination, reliability and overall performance.

[0200] Trust scores, as a crucial indicator of the reliability of distributed I / O modules, directly impact the reliability of data quality assessment and system decision-making. The learning optimization module employs an exponentially weighted moving average method to dynamically maintain trust scores, continuously adjusting the trust level based on the device's historical performance and current data quality.

[0201] The trust score is updated according to the following rule: the new trust score equals the preset update rate multiplied by the current trust score, plus 1 minus the preset update rate, multiplied by the current data quality indicator function value. The data quality indicator function is defined as follows: when the data quality score exceeds a preset acceptance threshold, the data quality indicator function value is 1, indicating good device performance; otherwise, the data quality indicator function value is 0, indicating poor device performance. The acceptance threshold is determined based on the system's quality requirements and is typically set as the median or 70th percentile of the data quality score.

[0202] The choice of update rate parameter needs to balance the stability and responsiveness of the trust score. A larger update rate (close to 1) makes the trust score change slowly, which can resist occasional data quality fluctuations, but it is slower to respond to continuous changes in device performance; a smaller update rate (close to 0) makes the trust score more sensitive to current performance and can quickly reflect changes in device status, but it may cause inappropriate fluctuations in the trust score due to occasional anomalies. The system dynamically adjusts the update rate based on the historical stability of the device and the current operating environment, using a larger update rate for devices with stable historical performance and a smaller update rate for newly deployed devices or devices with unstable historical performance.

[0203] The updated trust score is transmitted to the initial sampling module and the data processing module through a parameter feedback mechanism. In the initial sampling module, it is used to establish and maintain the trust score system, which affects the formulation of the sampling strategy. In the data processing module, it is used to calculate the data quality score, which serves as an important input indicator for quality assessment, ensuring that the entire system can perform data processing and decision-making based on accurate equipment reliability assessment.

[0204] Through this unified and systematic parameter optimization mechanism, the learning optimization module achieves coordinated optimization of key parameters of the entire distributed IO data preprocessing system. This not only improves the local performance of each module, but more importantly, it enables continuous improvement of the overall system performance, ensuring that the system can adapt to the ever-changing industrial environment and business needs and maintain optimal operating status.

[0205] The fault handling module 106 serves as the core of security for the distributed I / O data preprocessing system, undertaking the critical tasks of maintaining stable system operation, ensuring secure data transmission, and handling various anomalies. This module establishes a complete fault detection, diagnosis, handling, and recovery mechanism to ensure the distributed system maintains coordination, consistency, and reliability in the face of various internal and external interferences and faults. The module's design follows the principles of layered protection, proactive monitoring, rapid response, and automatic recovery, providing comprehensive security for the entire system.

[0206] The fault handling module first establishes a comprehensive system health status monitoring system. Through multi-dimensional and multi-level monitoring mechanisms, it monitors the system's operating status in real time, providing an accurate status information basis for subsequent fault diagnosis and handling.

[0207] The system adopts a hierarchical health monitoring architecture to perform real-time health status assessments on each distributed I / O module and edge computing node. The health status assessment is based on a comprehensive analysis of four core dimensions: hardware health is assessed by monitoring hardware indicators such as CPU utilization, memory usage, storage space utilization, and temperature. The preset normal ranges are as follows: preset CPU utilization threshold below 80%, determined based on statistical analysis balancing system performance and stability; preset memory usage threshold below 85%, obtained through quantitative assessment of memory management efficiency and system response speed; preset storage space utilization threshold below 90%, set through historical data analysis of the impact of storage capacity on system operation; and preset operating temperature range between -10°C and 60°C, determined according to equipment specifications and environmental adaptability technical specifications. Communication health is assessed by statistically analyzing network packet loss rate, communication latency, connection stability, and data transmission error rate. The preset packet loss rate threshold should be below 0.1%, determined based on network performance analysis of communication reliability requirements. The preset communication latency threshold should be below 100 milliseconds, obtained through quantitative evaluation of the real-time control system's response time requirements. The preset connection interruption threshold should be less than 3 times per hour, set based on analysis of the impact of connection stability on system continuity. Data health is based on statistical analysis of data quality scores, including a comprehensive evaluation of data integrity, consistency, and timeliness. Functional health is assessed by evaluating the response time, processing capacity, and error rate of each module.

[0208] The health score for each dimension is calculated using a weighted scoring method. The health score is the weighted average of the indicators for each dimension. The preset weights are as follows: hardware health score is 0.3, determined based on statistical analysis of the impact of hardware failures on system operation; communication health score is 0.3, obtained through quantitative assessment of the impact of communication failures on data transmission reliability; data health score is 0.2, set through historical data analysis of the impact of data quality on control accuracy; and functional health score is 0.2, determined based on assessment of the impact of functional anomalies on system performance. The overall health score is represented by a continuous value between 0 and 1, where 1 represents complete health and 0 represents complete failure. An early warning mechanism is triggered when the health score falls below a preset warning threshold (default value 0.7, determined based on a balance analysis between system warning response time and fault prevention effectiveness). A fault handling process is initiated when the score falls below a preset fault handling threshold (default value 0.5, obtained through quantitative assessment of fault handling timeliness and system availability). Emergency isolation measures are implemented when the score falls below a preset emergency isolation threshold (default value 0.3, set through statistical analysis of system security and fault propagation risk).

[0209] Based on the distributed timing synchronization mechanism established by the initial sampling module, the fault handling module continuously monitors the timing synchronization status between each node. Timing deviation monitoring is achieved by periodically comparing the local time of each distributed I / O module with the system reference time. The estimated timing deviation is calculated as follows: timing deviation equals system reference time minus module local time. Under normal circumstances, the timing deviation should be controlled within a preset normal deviation threshold of ±10 milliseconds, determined based on the technical specifications for the timing accuracy requirements of the real-time control system. When the timing deviation exceeds a preset abnormal alarm threshold, a timing abnormal alarm is triggered. The default value of the abnormal alarm threshold is ±50 milliseconds, obtained based on a quantitative assessment of the impact of timing deviation on system coordination. When the timing deviation exceeds a preset calibration start threshold, the timing calibration process is initiated. The default value of the calibration start threshold is ±100 milliseconds, set through statistical analysis of the impact of timing deviation on data validity.

[0210] Timing synchronization quality assessment is based on the statistical characteristics of timing deviations, including indicators such as the mean deviation, standard deviation, and maximum deviation. The timing synchronization quality score is calculated as 1 minus the normalized standard deviation of the timing deviation, where normalization maps the standard deviation to a range of 0 to 1. A timing synchronization quality score higher than a preset "good synchronization" threshold indicates good timing synchronization, with a default value of 0.8, determined based on statistical analysis of the impact of timing synchronization quality on system performance. A score between a preset "synchronization concern" threshold indicates that synchronization quality generally requires attention, with a default value between 0.5 and 0.8, obtained through quantitative assessment of the impact of synchronization quality on data reliability. A score lower than a preset "synchronization anomaly" threshold indicates synchronization anomalies requiring immediate attention, with a default value of 0.5, set through historical data analysis of the impact of synchronization anomalies on system stability.

[0211] The system establishes a comprehensive communication link health monitoring mechanism to evaluate the transmission quality and stability of each communication link in real time. Link health monitoring includes two levels: physical link monitoring and logical link monitoring. Physical link monitoring assesses the basic transmission capability of the link by detecting physical layer parameters such as signal strength, bit error rate, and connection status. Logical link monitoring assesses the actual transmission performance of the link by analyzing network layer parameters such as packet transmission success rate, round-trip time, and throughput.

[0212] The quantitative assessment of link health employs a comprehensive scoring method. The scoring formula is: Link health equals a preset transmission success rate weight multiplied by the packet transmission success rate, plus a preset latency performance weight multiplied by the normalized latency performance index, plus a preset stability weight multiplied by the connection stability index. Specifically, the preset transmission success rate weight is set to 0.4, determined based on the analysis of the core impact of transmission success rate on link reliability; the preset latency performance weight is set to 0.3, obtained from the quantitative assessment of the impact of latency on real-time communication performance; and the preset stability weight is set to 0.3, determined through statistical analysis of the impact of connection stability on system continuity. The normalized latency performance index is obtained by inversely mapping the ratio of actual latency to target latency, and the connection stability index is calculated based on statistical analysis of connection interruption frequency and duration.

[0213] Based on the trust scoring system established by the initial sampling module, the fault handling module is responsible for the continuous dynamic maintenance and management of the trust scores to ensure that the trust scores can accurately reflect the actual reliability level of each distributed IO module.

[0214] The system establishes a hierarchical trust management mechanism, classifying and managing distributed I / O modules according to different trust scores. The preset trust score grading standards are as follows: High trust level threshold range (0.8, 1.0], determined based on statistical analysis of equipment reliability and critical task capability, indicating stable and reliable equipment operation, excellent data quality, and suitability for critical control tasks; Medium trust level threshold range (0.5, 0.8], obtained based on quantitative assessment of basic equipment performance and general task adaptability, indicating basically normal equipment operation, acceptable data quality, and suitability for general monitoring tasks; Low trust level threshold range (0.2, 0.5], set through historical data analysis of equipment instability and non-critical task risks, indicating unstable equipment operation, poor data quality, and suitability only for non-critical monitoring tasks; Untrustworthy level threshold range [0.0, 0.2], determined based on assessment and analysis of serious equipment problems and system security risks, indicating serious equipment problems, unreliable data, and the need for isolation.

[0215] Devices with different trust levels enjoy different permissions and assume different responsibilities in the system. Data from high-trust-level devices receives higher weight in the data processing module, higher priority in the priority scheduling module, and greater decision-making influence in the collaborative control module. Data from low-trust-level devices requires additional verification and filtering, and their control recommendations need to be confirmed by other high-trust devices before they can be implemented. Untrusted devices are automatically isolated, their data does not participate in system decisions, and maintenance alarms are generated to prompt manual intervention.

[0216] To ensure the security of data transmission and storage in the distributed system, the fault handling module establishes a comprehensive data integrity protection mechanism, using cryptographic methods to guarantee the authenticity, integrity, and confidentiality of the data.

[0217] The system generates digital signature digests for critical data packets and control actions, ensuring the integrity and authenticity of data transmission and storage. The digital signature digests are generated using a hash-based Message Authentication Code (HMAC) algorithm, which combines hash functions and key encryption techniques to provide robust data protection.

[0218] The signature generation process employs a two-tiered protection mechanism. The first tier is data integrity protection, which uses a SHA-256 hash operation on the data packet content to obtain a fixed-length digest value. This digest value exhibits one-way hashing and avalanche effect, meaning any small data change will lead to a significant alteration of the digest value. The second tier is identity authentication protection, which uses a system security key to perform an HMAC operation on the digest value to generate the final digital signature. This dual protection mechanism can both detect data tampering through hash verification and verify the authenticity of the data source through key verification.

[0219] The system employs a hierarchical key management system. The master key is generated and distributed centrally by the system administrator and used to generate working keys for each node. Working keys are held independently by each edge computing node and used for daily data signing and verification. Key updates utilize a periodic rotation mechanism: the master key is updated annually, and working keys are updated monthly to ensure key security. Key distribution is conducted through a secure channel, employing the RSA asymmetric encryption algorithm to protect the security of key transmission.

[0220] The receiving end performs integrity verification on received data packets, which includes two steps: signature verification and timeliness check. Signature verification is achieved by recalculating the HMAC value of the data packet and comparing it with the received signature. If they match, it indicates that the data has not been tampered with and the source is trustworthy. Timeliness check is achieved by comparing the timestamp of the data packet with the current time. If the time difference exceeds the preset validity period, the data packet is rejected to prevent replay attacks. The preset validity period is usually 5 minutes.

[0221] For data packets that fail to be verified, the system adopts a tiered processing strategy. Occasional verification failures may be caused by network transmission errors, and the system records alarm information and requests retransmission. Continuous verification failures may indicate a security threat or equipment failure, and the system immediately isolates the relevant equipment and generates a security alarm. Batch verification failures may indicate key leakage or system attack, and the system activates emergency security mode and notifies the administrator for manual intervention.

[0222] Based on the aforementioned health monitoring and safety protection mechanisms, the fault handling module establishes a multi-level fault handling system, which ensures the reliable operation of the system under various abnormal conditions by combining preventive maintenance, proactive fault handling and passive fault recovery.

[0223] When link health monitoring detects communication link anomalies, the system initiates a tiered response mechanism. Minor link anomalies, such as occasional packet loss or latency fluctuations, are handled through adaptive retransmission and path optimization. The system automatically increases the number of retransmissions, adjusts data packet size, selects alternative communication paths, and reduces the transmission frequency of non-critical data to alleviate link load.

[0224] Severe link anomalies, such as network interruption, excessively high packet loss rate, or communication delay timeout, trigger a communication degradation mode. The system automatically switches to a local conservative control strategy, ceasing data exchange with neighboring nodes and making independent control decisions based on local data and preset security parameters. The parameter settings of the conservative control strategy follow the principle of safety first, setting the control threshold to 80% of the normal value, extending the response time by 50%, and limiting the control action amplitude to within 60% of the normal range, ensuring that the system can maintain basic safe operation even in the event of communication interruption.

[0225] The communication recovery mechanism includes two modes: active probing and passive waiting. Active probing detects link recovery status by periodically sending heartbeat packets, with the probing interval starting at 1 second and gradually increasing to a maximum of 60 seconds. Passive waiting detects link recovery by monitoring network activity, and immediately attempts to connect when a stable network signal is detected. After link recovery, the system performs state synchronization and data verification to ensure that the recovered system state is consistent with that of other nodes.

[0226] When timing anomalies are detected by timing synchronization monitoring, the system initiates a multi-level timing calibration mechanism. Minor timing deviations are handled through software clock compensation. Based on statistical analysis of historical timing data, the system establishes a clock drift model to predict and compensate for clock deviations. Minor timing deviations are defined as deviations within ±50 milliseconds. The clock drift model employs a linear regression method, analyzing deviation data from the most recent 100 timing synchronization cycles to calculate the clock drift rate and deviation trend, thereby enabling the prediction and proactive compensation of future timing deviations.

[0227] Severe timing deviations exceeding ±50 milliseconds trigger mandatory timing calibration. The system suspends data processing and performs multiple rounds of timing synchronization with the time reference node, recalibrating the local clock via Network Time Protocol (NTP) or Precision Time Protocol (PTP). The calibration process employs an averaging method based on multiple measurements to eliminate the impact of network latency jitter and ensure calibration accuracy. After calibration, the system retrospectively corrects the timestamps of historical data to guarantee data timing consistency.

[0228] In the event of extreme timing anomalies, with deviations exceeding ±100 milliseconds or complete timing synchronization failure, a local clock compensation mechanism is activated. The system switches to independent timing mode, maintaining a relatively accurate time base based on the local crystal oscillator and historical timing statistics. Local clock compensation is achieved by establishing a clock error accumulation model. This model predicts the accumulation trend of clock errors based on the device's historical clock performance data and makes corresponding compensation adjustments.

[0229] For distributed I / O modules whose trust scores consistently fall below the threshold or whose health deteriorates significantly, the system executes a device isolation and system refactoring process. Device isolation employs a progressive strategy: first, the device is marked as untrusted, and its data is no longer used for critical control decisions but can still be used for monitoring and reference; if the device's condition continues to deteriorate, the device is completely isolated, ceasing all data exchange and control interactions; finally, detailed maintenance prompts are generated, including a description of the fault symptoms, possible cause analysis, and suggested remedial measures, guiding maintenance personnel in device inspection and maintenance.

[0230] The system reconfiguration mechanism addresses equipment failures by reallocating control tasks and adjusting the system topology. When critical equipment is isolated, the system automatically assigns its control tasks to other healthy equipment. When multiple devices in a certain area fail simultaneously, the system adjusts its control strategy to expand the monitoring range of healthy equipment and ensure that no control blind spots occur. When the number of failed devices exceeds the system's redundancy capacity, the system activates an emergency safety mode and adopts the most conservative control strategy to maintain basic safe operation.

[0231] To address system-level failures such as software deadlocks, memory leaks, and processor anomalies, the fault handling module integrates a hardware watchdog mechanism and a software monitoring mechanism. The hardware watchdog monitors the system's operational status through independent hardware circuitry. The system needs to periodically send a preset heartbeat signal to the watchdog, typically once per second, determined based on an analysis balancing system response time and monitoring accuracy. If the watchdog does not receive a heartbeat signal within a preset timeout period (typically 5 seconds, determined through statistical analysis balancing fault detection timeliness and false alarm rate), a serious system failure is considered, and the corresponding processing module or the entire system is automatically restarted.

[0232] The software monitoring mechanism detects software-level anomalies through multi-threaded monitoring and mutual checks. The main monitoring thread is responsible for checking the response time and processing capacity of each functional module. When a module is found to have a timeout or a decrease in processing capacity, it attempts to restart the module. The auxiliary monitoring thread is responsible for checking system resource usage, including CPU utilization, memory usage, and disk space. When resource usage exceeds a preset safety threshold, typically 80%, it is set based on historical data analysis balancing system stability and resource utilization efficiency, and performs resource cleanup and optimization operations. The mutual check mechanism verifies the module's operating status through preset periodic communication frequencies between modules. The frequency is determined based on an assessment and analysis balancing monitoring coverage and system overhead, ensuring the reliability of the monitoring system itself.

[0233] The fault recovery process employs a tiered restart strategy. Module-level restarts only restart the faulty functional module, maintaining the normal operation of other modules, suitable for localized software failures. Process-level restarts require restarting the entire application process and clearing all memory states, suitable for memory leaks or state chaos. System-level restarts require restarting the entire edge computing node, suitable for hardware failures or operating system anomalies. After each restart, the system performs a complete self-check and state recovery to ensure that the restarted system can function normally and maintain coordination with other nodes.

[0234] Through this systematic and multi-layered fault handling mechanism, the fault handling module provides comprehensive security for the entire distributed IO data preprocessing system, ensuring that the system can maintain a stable and reliable operating state when faced with various internal and external interferences and faults, and providing solid technical support for safe production in industrial sites.

[0235] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. An edge node based distributed IO data preprocessing and feedback system, characterized in that, The method comprises the following steps: An initial sampling module sets a trust score for each distributed IO, and uses the distributed IO to collect sampling data based on an adaptive sampling strategy; The adaptive sampling strategy comprises the following steps: An initial priority score is established for each IO data channel of the distributed IO, the initial priority score being a data quality score multiplied by a preset data quality weight, plus a preset control urgency multiplied by a preset control urgency weight, plus a preset business criticality multiplied by a preset business criticality weight; A sampling period is calculated, the sampling period being a preset basic sampling period divided by a sampling adjustment term, the sampling adjustment term being a preset sensitivity adjustment parameter multiplied by a comprehensive priority at a previous time point plus 1; A data processing module performs filtering processing on the sampling data to obtain filtered data, and calculates a data quality score based on the filtered data and the trust score; A priority scheduling module calculates a comprehensive priority based on the filtered data and the data quality score, and outputs a scheduling result based on the comprehensive priority; the step of calculating the comprehensive priority comprises the following steps: An exponential weighted moving average method is used to calculate a scheduling event intensity of the filtered data; a scheduling event intensity at a current time point is a preset event forgetting factor multiplied by a scheduling event intensity at a previous time point plus an event indication term; the event indication term is a difference between 1 and the preset event forgetting factor multiplied by an event indication value at the current time point; when the filtered data exceeds a preset alarm threshold of the IO data channel of the distributed IO, the event indication value is 1, otherwise the event indication value is 0; A comprehensive priority of each IO data channel is calculated, the comprehensive priority being a sum of the scheduling event intensity at the current time point multiplied by a preset scheduling event intensity weight coefficient, a difference between 1 minus the data quality score multiplied by a preset quality score weight coefficient, the preset control urgency at the current time point multiplied by a preset urgency weight coefficient, and the business criticality multiplied by a preset criticality weight coefficient; A collaborative control module executes a hierarchical control decision based on the scheduling result in combination with neighbor node collaborative information, and outputs a control action; the hierarchical control decision comprises a local fast decision layer, a neighbor node collaborative verification layer, and a conservative monitoring and observation layer; If the scheduling result is an emergency queue and the data quality score is greater than a preset high threshold, the local fast decision layer directly executes a predefined local control action; If the scheduling result is an emergency queue and the data quality score is less than the preset high threshold, or the scheduling result is a regular queue and a control event intensity is greater than a preset collaborative verification threshold, the neighbor node collaborative verification layer executes a control action in combination with neighbor node collaborative information; the control event intensity is an absolute difference between a current filtered data and a historical filtered data mean, divided by a standard deviation of the historical filtered data, and subjected to normalization processing; If the scheduling result is a regular queue and the control event intensity is less than the preset collaborative verification threshold, the conservative monitoring and observation layer does not execute a control action; A learning optimization module updates system parameters based on an execution result of the control action and the data quality score.

2. The edge node based distributed IO data preprocessing and feedback system according to claim 1, wherein, The scheduling result comprises an emergency queue and a regular queue: When a comprehensive priority of an IO data channel is greater than or equal to a preset emergency queue access threshold, the IO data channel is assigned to the emergency queue for processing. The emergency queue adopts an earliest deadline first scheduling algorithm, and an IO data channel task with the earliest deadline is preferentially executed, the deadline being the current time plus a delay time, the delay time being a preset basic delay time divided by a control urgency adjustment factor, the control urgency adjustment factor being a preset control urgency at the current time multiplied by a preset urgency sensitivity coefficient plus a delay constant term; When the comprehensive priority of the IO data channel is less than a preset emergency queue admission threshold, the IO data channel is assigned to a regular queue for processing; The regular queue adopts a probabilistic selection mechanism, and the probability of each IO data channel obtaining bandwidth is an exponential function value of the comprehensive priority of each IO data channel divided by a sum of exponential function values of comprehensive priorities of a plurality of IO data channels participating in scheduling, the base of the exponential function being a natural constant, and the exponent being a sharpness adjustment factor multiplied by the comprehensive priority; the sharpness adjustment factor being a preset basic sharpness parameter multiplied by a sharpness adjustment factor, the sharpness adjustment factor being a preset priority difference sensitivity coefficient multiplied by a standard deviation of the comprehensive priorities of the plurality of IO data channels participating in scheduling at the current time plus a sharpness constant term.

3. The edge node based distributed IO data pre-processing and feedback system of claim 1, wherein, The initialization setting of the scheduling event intensity includes: When the system starts, a preset scheduling event intensity initial value is assigned according to the type of the IO data channel; After the system enters a running state, if the frequency of abnormal events in a preset warm-up period is greater than a preset abnormal frequency threshold, the scheduling event intensity reference value is increased based on a preset up-regulation step, and if the frequency of abnormal events is less than a preset stability threshold, the scheduling event intensity reference value is decreased based on a preset down-regulation step; the frequency of abnormal events is the frequency of abnormal events, and the abnormal event is filtered data that exceeds a preset alarm threshold of the IO data channel; After the warm-up period ends, a dynamic confidence interval of the scheduling event intensity is established for each IO data channel, and the dynamic confidence interval is between the product of the scheduling event intensity reference value and a preset lower boundary coefficient and the product of the scheduling event intensity reference value and a preset upper boundary coefficient.

4. The edge node based distributed IO data pre-processing and feedback system of claim 1, wherein, The combination of the neighbor node cooperative information includes: Generating an event summary and exchanging with a neighbor node; the event summary includes a control event intensity and a confidence, and the confidence is a data quality score multiplied by a preset trust adjustment factor; the neighbor node is an edge computing node that is in the same network segment, has established a stable communication connection, and has a functional overlap or a monitoring area overlap in the monitoring range; The number of neighbor nodes with a control event intensity greater than a preset cooperative verification threshold is counted, and when the number of neighbor nodes exceeding the preset cooperative verification threshold is greater than or equal to a preset consistency threshold, the plurality of neighbor nodes perform a coordinated control action, otherwise no control action is performed.

5. The edge node based distributed IO data pre-processing and feedback system of claim 1, wherein, The step of filtering the sampling data includes: Performing adaptive median filtering and adaptive Kalman filtering on analog data in the sampling data; The adaptive median filter adjusts the window radius based on the signal change rate and the noise standard deviation, the window radius is a preset basic window radius multiplied by a window adjustment factor, the window adjustment factor is a preset signal change rate weight multiplied by the negative value of the normalized signal change rate, plus a noise standard deviation weight multiplied by the normalized noise standard deviation, plus a preset constant term; the signal change rate is the absolute value of the difference between adjacent sampling points in the sampling data, and the noise standard deviation is the product of the median of the absolute difference between a plurality of data points in a preset sliding window and a median in the window and a preset normal distribution correction coefficient, and the median in the window is the median of all data points in the sliding window; The adaptive Kalman filter updates the observation noise covariance through an exponential weighted moving average method based on the observation residual, the new observation noise covariance is a preset forgetting factor multiplied by the observation noise covariance at the previous time plus a residual term, the residual term is the difference between 1 and the preset forgetting factor multiplied by the square of the current observation residual, wherein the observation residual is the absolute difference between the actual observation value and the predicted observation value; The multi-level debouncing processing is performed on the digital data in the sampling data, including pulse width constraint filtering based on a minimum effective pulse width threshold, and a multi-window voting mechanism; the multi-window voting mechanism determines the final digital state output according to the majority principle, and when the number of occurrences of the high level state and the low level state of the digital data is equal, the output state of the previous time is maintained.

6. The edge node based distributed IO data preprocessing and feedback system according to claim 5, wherein, The steps of calculating the data quality score include: The data quality score is 1 divided by the quality combination term, and the quality combination term is 1 plus the negative quality index weighted combination power of the natural constant; For analog data, the quality index weighted combination is the sum of the preset observation residual weight multiplied by the negative value of the observation residual, the preset signal-to-noise ratio weight multiplied by the signal-to-noise ratio at the current time, and the preset first trust score weight multiplied by the trust score; wherein the signal-to-noise ratio is the logarithm to the base 10 of the absolute ratio of the signal power to the noise power, multiplied by a preset signal-to-noise constant to obtain, the signal power is the mean square value of the signal in the sliding window, and the noise power is the square of the noise standard deviation; For digital data, the quality index weighted combination is the sum of the preset glitch probability weight multiplied by the glitch probability, the preset second trust score weight multiplied by the trust score, and the preset consistency weight multiplied by the voting consistency index; wherein the glitch probability is the number of detected glitches divided by the total number of sampling points in the sliding window, and the voting consistency index is the number of state consistent sampling points in the preset voting window divided by the total number of sampling points in the preset voting window.

7. The edge node based distributed IO data pre-processing and feedback system of claim 1, wherein, The system parameters include the alarm threshold and the trust score, and the steps of updating the system parameters include: Updating the alarm threshold of each IO data channel includes: Presetting the initial value of the alarm threshold of each IO data channel; Triggering the threshold basic update when the false alarm rate in the preset statistical window exceeds the preset false alarm rate threshold, or the miss alarm rate exceeds the preset miss alarm rate threshold; For analog data, the alarm threshold includes an upper alarm threshold and a lower alarm threshold, the new upper alarm threshold is the product of a preset dynamic threshold adjustment coefficient and the standard deviation of the sliding statistical window data plus the mean value of the sliding statistical window data, and the new lower alarm threshold is the mean value of the sliding statistical window data minus a dynamic threshold term, the dynamic threshold term is the product of the preset dynamic threshold adjustment coefficient and the standard deviation of the sliding statistical window data; for digital data, the alarm threshold includes an abnormal state change frequency threshold, the new abnormal state change frequency threshold is the product of a preset dynamic frequency abnormality coefficient and the standard deviation of the state change frequency plus the mean value of the state change frequency; wherein the standard deviation of the sliding statistical window data, the mean value of the sliding statistical window data, the standard deviation of the state change frequency, and the mean value of the state change frequency are calculated based on the historical filtered data in the preset sliding statistical window; The trust score is updated, the new trust score is the preset update rate multiplied by the current trust score, plus the difference of 1 minus the preset update rate multiplied by the value of the current data quality indication function; the data quality indication function value is 1 when the data quality score exceeds the preset acceptance threshold, otherwise the data quality indication function value is 0.

Images