A multi-branch fluid element adaptive timing scheduling control method and system

Abstract

The application discloses a kind of multi-branch fluid element adaptive timing scheduling control method and system, belong to automatic control technical field, this method first acquisition each branch fluid controlled element real-time differential pressure, medium moisture content and flue gas temperature, extract four-dimensional characteristic parameters;Based on four-dimensional characteristic parameters distinguish three kinds of blockades, such as conventional gradual progressive dust, soft agglomeration impact, hard agglomeration compaction, construct double threshold value decision logic and utilize temperature and humidity dynamic correction;By fitting differential pressure timing curve, the critical point of compaction solidification and blowback failure is predicted, and the switching prediction signal is triggered;Combined with agglomeration impact frequency and local blockage pressure gradient, construct risk index, sort priority and output branch switching scheduling instruction, complete fluid controlled element switching;The application can accurately identify the type of blockage, predict switching time, and improve the running stability and service life of fluid controlled element.

Description

Technical Field

[0001] This invention belongs to the field of automatic control technology, specifically a method and system for adaptive timing scheduling control of multi-branch fluid elements. Background Technology

[0002] In industrial flue gas purification and other operating conditions, fluid-controlled components are frequently impacted by agglomerated pollutants under high humidity and temperature environments, easily leading to soft agglomerated impact blockage and hard agglomerated compaction blockage. Unlike conventional gradual ash accumulation blockage, this type of blockage is characterized by its sudden onset, significant damage, and susceptibility to failure during backflushing. Existing fluid-controlled component control technologies mostly rely on a single differential pressure threshold for blockage determination, only capable of identifying conventional ash accumulation and unable to distinguish between agglomerated impact and compaction blockage. Furthermore, they do not consider the influence of temperature and humidity on blockage characteristics, and fixed thresholds are prone to misjudgment. Simultaneously, traditional methods lack accurate prediction of agglomerated agglomerate compaction and solidification, as well as the critical failure point during backflushing, making early warning difficult and easily leading to fluid-controlled component failure, decreased purification efficiency, and even equipment damage. In addition, fluid-controlled component switching often uses timed or fixed differential pressure triggering modes, failing to consider the actual blockage risk priority of multiple branch fluid-controlled components, resulting in blind switching strategies, wasted fluid-controlled component resources, and increased operating costs. Therefore, there is an urgent need for a fluid-controlled component switching control method that can accurately identify the type of agglomerated pollutant impact, dynamically adjust the determination threshold, predict the failure critical point, and optimize the switching strategy. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes an adaptive time-series scheduling control method and system for multi-branch fluid components, aiming to solve problems such as difficulty in distinguishing blockage types, inaccurate prediction of switching timing, and unreasonable switching strategies caused by agglomerated pollutant impacts in fluid-controlled components. The method first collects real-time differential pressure, medium moisture content, and flue gas temperature of each branch fluid-controlled component, extracting four-dimensional feature parameters. Based on these four-dimensional feature parameters, it distinguishes three types of blockage: conventional progressive ash accumulation, soft agglomeration impact, and hard agglomeration compaction, constructing a dual-threshold judgment logic and using dynamic temperature and humidity correction. By fitting the differential pressure time-series curve, it predicts the critical point of compaction solidification and backflushing failure, triggering a switching prediction signal. Combining the frequency of agglomeration impacts with the local blockage pressure gradient, it constructs a risk index, prioritizes components, and outputs branch switching scheduling instructions to complete the switching of fluid-controlled components. This invention can accurately identify blockage types, predict switching timing, and improve the operational stability and service life of fluid-controlled components.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] An adaptive timing scheduling control method for multi-branch fluid elements includes:

[0006] The real-time differential pressure, medium moisture content, and flue gas temperature of the fluid-controlled components in each branch are obtained, and four-dimensional characteristic parameters are extracted.

[0007] Based on the four-dimensional feature parameters, conventional progressive ash accumulation blockage, soft agglomeration impact blockage and hard agglomeration compaction blockage are distinguished to obtain the preliminary classification result of the blockage condition. Then, the preliminary classification result of the blockage condition is re-verified based on the dual threshold judgment logic. At the same time, the dual threshold judgment logic is dynamically corrected by the medium moisture content and flue gas temperature, and the corrected dual threshold judgment logic is output.

[0008] Based on the modified dual-threshold judgment logic, the pressure difference time-series curve generated by the real-time pressure difference of each fluid controlled element is fitted to predict the critical point of agglomeration, compaction, solidification, and backflushing failure, trigger the element switching prediction signal, and output the identifier of the fluid element to be scheduled.

[0009] The system collects the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element identifier to be scheduled, sorts the agglomeration and blockage risk priority of each branch fluid controlled element, outputs the branch switching scheduling command, and completes the switching of fluid controlled elements according to the branch switching scheduling command. After the switching, the system collects the four-dimensional characteristic parameters of the newly put fluid controlled element in real time.

[0010] Specifically, four-dimensional feature parameters are extracted, including:

[0011] The real-time pressure difference is segmented into time domain segments based on the moisture content of the medium and the temperature of the flue gas to obtain segmented pressure difference time series data.

[0012] Based on the segmented differential pressure time series data, the differential pressure rise slope is obtained;

[0013] Based on the segmented differential pressure time series data, the maximum fluctuation difference between the real-time differential pressure and the preset pulsating reference differential pressure is calculated to obtain the differential pressure pulsation amplitude.

[0014] Based on the segmented pressure difference time series data, the sudden pressure difference corresponding to the instantaneous impact of aggregated pollutants is extracted, and the sudden pressure difference is used as a weighting coefficient and substituted into the pre-constructed short-time pressure difference distortion coefficient calculation model to output the short-time pressure difference distortion coefficient.

[0015] Multiple pressure acquisition units are arranged along the axial direction of the folds of the fluid-controlled element to acquire axial segmented pressure data of the folds, divide multiple pressure gradient detection intervals, calculate the differential pressure gradient value of each interval based on the axial segmented pressure data, and then calculate the differential pressure gradient difference between two adjacent pressure gradient detection intervals to obtain the pressure gradient difference of the fold interval.

[0016] The pressure difference rise slope, pressure difference pulsation amplitude, short-time pressure difference distortion coefficient, and pressure gradient difference in the folded region are all output as four-dimensional feature parameters.

[0017] Specifically, the construction process of the short-time pressure difference distortion coefficient calculation model includes:

[0018] Acquire segmented differential pressure time series data under historical operating conditions and label sudden differential pressure events caused by instantaneous impacts of aggregated pollutants;

[0019] Extract the impact amplitude and impact duration for each sudden pressure difference event, and define the ratio of impact amplitude to impact duration as the instantaneous impact intensity;

[0020] Using instantaneous impact intensity as the independent variable and short-term pressure difference distortion coefficient as the target variable, an exponentially weighted moving average method is used to fit and construct a calculation model for the short-term pressure difference distortion coefficient. The expression for the calculation model of the short-term pressure difference distortion coefficient is as follows: Where D is the current short-term pressure difference distortion coefficient, and I is the instantaneous impact intensity. As a smoothing factor, The output is the short-time differential pressure distortion coefficient of the previous moment, and the completed short-time differential pressure distortion coefficient calculation model.

[0021] Specifically, based on the aforementioned four-dimensional characteristic parameters, conventional progressive ash accumulation blockage, soft agglomeration impact blockage, and hard agglomeration compaction blockage are distinguished, including:

[0022] The four-dimensional feature parameters are preset to match the characteristic response modes of conventional gradual ash accumulation blockage, soft agglomeration instantaneous change threshold range, and hard agglomeration compaction cumulative threshold range, respectively.

[0023] The pressure difference rise slope, pressure difference pulsation amplitude, short-time pressure difference distortion coefficient, and pressure gradient difference in the folded area are compared one by one with the corresponding steady-state threshold range of conventional ash accumulation, the instantaneous change threshold range of soft agglomeration, and the cumulative threshold range of hard agglomeration compaction, and the transient response component and steady-state cumulative component of each characteristic parameter are extracted.

[0024] Based on the extracted transient response components and steady-state cumulative components, a feature coupling judgment matrix is ​​constructed. The transient abrupt change features are characterized by the differential pressure pulsation amplitude and the short-term differential pressure distortion coefficient, which are used to match soft agglomeration impact blockage. The steady-state cumulative features are characterized by the differential pressure rise slope and the pressure gradient difference in the folded interval, which are used to match hard agglomeration compaction blockage. When the four-dimensional feature parameters show a steady-state change as a whole, it is matched with conventional progressive ash accumulation blockage, thus obtaining the preliminary classification results of the blockage conditions.

[0025] Specifically, the verification of the preliminary classification results of siltation conditions based on the dual-threshold judgment logic includes:

[0026] Four-dimensional characteristic parameters including differential pressure rise slope, differential pressure pulsation amplitude, short-term differential pressure distortion coefficient, and pressure gradient difference in folded regions are obtained. The weighted sum of differential pressure pulsation amplitude and short-term differential pressure distortion coefficient is defined as the instantaneous change characteristic value, and the weighted sum of differential pressure rise slope and pressure gradient difference in folded regions is defined as the steady-state cumulative characteristic value.

[0027] Based on historical calibration data, instantaneous change thresholds and steady-state accumulation thresholds are preset, and a first timing window and a second timing window are set; the length of the first timing window is less than the length of the second timing window.

[0028] Within the first timing window, the difference between the peak and trough values ​​of the real-time differential pressure is calculated as the instantaneous fluctuation amplitude. When the instantaneous fluctuation amplitude is greater than the instantaneous sudden change threshold, an instantaneous warning flag is triggered.

[0029] Within the second timing window, the cumulative offset of the real-time differential pressure is calculated using an exponentially weighted moving average method. When the cumulative offset exceeds the steady-state cumulative threshold, a steady-state warning flag is triggered.

[0030] When the instantaneous warning sign is triggered, the working condition is confirmed as soft agglomeration impact blockage; when the steady-state warning sign is triggered, the working condition is confirmed as hard agglomeration compaction blockage; when neither sign is triggered, the working condition is confirmed as conventional progressive ash accumulation blockage; the blockage condition judgment result after secondary verification is output.

[0031] Specifically, the logic for dynamically correcting the dual threshold determination using the moisture content of the medium and the flue gas temperature includes:

[0032] Obtain the moisture content of the medium and the flue gas temperature at the current sampling time;

[0033] Based on the moisture content of the medium, query the preset moisture content-instantaneous correction coefficient mapping table and the preset moisture content-steady-state correction coefficient mapping table, and output the first instantaneous correction coefficient and the first steady-state correction coefficient;

[0034] Based on the flue gas temperature at the current sampling time, query the preset temperature-instantaneous correction coefficient mapping table and the preset temperature-steady-state correction coefficient mapping table, and output the second instantaneous correction coefficient and the second steady-state correction coefficient;

[0035] Multiply the first instantaneous correction coefficient by the second instantaneous correction coefficient to obtain the instantaneous comprehensive correction factor, and multiply the first steady-state correction coefficient by the second steady-state correction coefficient to obtain the steady-state comprehensive correction factor.

[0036] Obtain the preset instantaneous mutation threshold and steady-state accumulation threshold, multiply the instantaneous mutation threshold by the instantaneous comprehensive correction factor to obtain the corrected instantaneous mutation threshold, and multiply the steady-state accumulation threshold by the steady-state comprehensive correction factor to obtain the corrected steady-state accumulation threshold.

[0037] The modified instantaneous mutation threshold is used to replace the instantaneous mutation threshold, and the modified steady-state cumulative threshold is used to replace the steady-state cumulative threshold. The updated dual threshold judgment logic is then output.

[0038] Specifically, the step of fitting the differential pressure time-series curves generated by the real-time differential pressure for each fluid-controlled element based on the modified dual-threshold judgment logic includes:

[0039] Extract the corrected instantaneous change threshold and the corrected steady-state cumulative threshold from the updated dual threshold judgment logic. At the same time, obtain the pressure difference time series curve generated by the real-time pressure difference of each fluid controlled element. Use the corrected steady-state cumulative threshold as an adaptive adjustment factor to adjust the sliding window step size in reverse and output the adjusted sliding window step size. The steady-state cumulative threshold and the window step size are inversely proportional.

[0040] According to the output sliding window step size, take three consecutive points on the pressure difference time series curve window by window, calculate the ratio of the central angle to the arc length based on the three-point common circle method, and use it as the curvature value of the center point of each window, and output the initial curvature sequence.

[0041] The modified instantaneous mutation threshold is used as an outlier suppression factor to filter the initial curvature sequence. The original pressure difference corresponding to each curvature value is traversed to three consecutive points. The vertical distance between the middle point and the line connecting the two points is calculated. If the vertical distance is greater than the outlier suppression factor, the mean of the curvature of the two points is used to replace the current curvature value. The smooth curvature sequence after outlier suppression is output.

[0042] Specifically, the critical point of predicting the compaction, solidification, and backflushing failure of agglomerated clumps triggers a component switching prediction signal, outputting an identifier of the fluid component to be scheduled, including:

[0043] Obtain a smooth curvature sequence, input the smooth curvature sequence into a time-series prediction network based on gated recurrent units in chronological order, and output a curvature time-series input vector;

[0044] The gated recurrent unit in the time-series prediction network updates the smooth curvature values ​​at multiple consecutive sampling times sequentially and outputs the curvature prediction values ​​at multiple sampling times.

[0045] The output curvature prediction value is compared with the preset curvature runaway threshold. When the curvature prediction value at two consecutive sampling times exceeds the curvature runaway threshold, it is determined that the critical point of agglomeration, compaction, solidification, and backflushing failure has been reached.

[0046] When the critical point is reached, a component switching prediction signal is triggered, the identifier of the currently controlled fluid component is obtained as the identifier of the fluid component to be scheduled, and the identifier of the fluid component to be scheduled is output.

[0047] Specifically, the process involves collecting the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element to be scheduled, prioritizing the agglomeration and blockage risk of each branch fluid controlled element, and outputting branch switching scheduling instructions, including:

[0048] Collect the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element identifier to be scheduled, and output the agglomeration impact frequency and local blockage pressure gradient of the current fluid controlled element;

[0049] The frequency of agglomeration impacts and the local blockage pressure gradient are weighted and fused to construct the agglomeration and blockage risk index for each fluid-controlled element, and the agglomeration and blockage risk index of each fluid-controlled element is output.

[0050] All online operating branch fluid-controlled components are sorted in descending order of their agglomeration and blockage risk index to obtain a risk priority queue;

[0051] Extract the top-ranked fluid controlled component identifier from the risk priority queue as the fluid controlled component identifier to be removed in this switchover; query the idle fluid controlled component identifier with the highest health score in the standby fluid controlled component pool as the fluid controlled component identifier to be deployed; output the fluid controlled component identifier to be removed and the fluid controlled component identifier to be deployed.

[0052] The identifiers of the fluid-controlled components to be removed, the identifiers of the fluid-controlled components to be put into operation, and the switching order list generated based on the risk priority queue are encapsulated together into a branch switching scheduling instruction, which is then output to the fluid-controlled component switching actuator.

[0053] Specifically, the steps for constructing the agglomeration and clogging risk index for each fluid-controlled element include:

[0054] The agglomeration impact frequency and local blockage pressure gradient were obtained and standardized to obtain normalized impact frequency and normalized pressure gradient.

[0055] The normalized impact frequency is multiplied by a preset first weighting coefficient to obtain the weighted impact component, and the normalized pressure gradient is multiplied by a preset second weighting coefficient to obtain the weighted pressure component.

[0056] The weighted impact component and the weighted pressure component are added together to obtain the agglomeration and clogging risk index of the fluid-controlled component.

[0057] An adaptive timing scheduling control system for multi-branch fluid components includes:

[0058] The feature extraction module is used to obtain the real-time differential pressure, medium moisture content and flue gas temperature of the fluid controlled components in each branch, and to extract the segmented differential pressure time series data from the segmented time domain of the real-time differential pressure, and extract four-dimensional feature parameters from it.

[0059] The blockage type identification module is used to distinguish between conventional progressive ash accumulation blockage, soft agglomeration impact blockage and hard agglomeration compaction blockage based on four-dimensional feature parameters. It establishes a dual threshold judgment logic based on instantaneous change threshold and steady-state cumulative threshold, and uses the medium moisture content and flue gas temperature to query the corresponding mapping table to obtain correction coefficients and dynamically correct the dual threshold judgment logic.

[0060] The critical point prediction module is used to obtain a smooth curvature sequence by fitting the differential pressure time curve based on the modified dual threshold judgment logic, predict the critical point of compaction solidification and backflushing failure of agglomerates, trigger the element switching prediction signal and output the identifier of the fluid element to be scheduled.

[0061] The switching execution module is used to collect the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element to be switched, construct the agglomeration and blockage risk index, determine the identifiers of the fluid controlled elements to be removed and to be put into operation, and generate branch switching scheduling instructions. After the instructions are issued, the switching of the fluid controlled element is completed, and at the same time, the four-dimensional characteristic parameters of the newly put into fluid controlled element are collected in real time.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] 1. This invention proposes an adaptive time-series scheduling control method for multi-branch fluid components. By extracting four-dimensional characteristic parameters of differential pressure, it can accurately distinguish three types of blockage: conventional progressive ash accumulation, soft agglomeration impact, and hard agglomeration compaction. Combined with dynamic correction of dual threshold judgment logic based on medium moisture content and flue gas temperature, it effectively avoids the problem of misjudgment and missed judgment caused by fixed thresholds being easily affected by operating conditions, thus improving the accuracy and reliability of blockage identification. At the same time, by fitting the differential pressure time-series curve, it predicts the critical point of agglomeration compaction solidification and backflushing failure, realizing early warning of the switching timing of fluid controlled components, avoiding excessive blockage or sudden failure of fluid controlled components, and ensuring the stable operation of the purification system.

[0064] 2. This invention proposes an adaptive time-sequential scheduling control method for multi-branch fluid components. Based on the frequency of agglomeration impact and the local blockage pressure gradient, a blockage risk index is constructed. The risk priority of multi-branch fluid controlled components is sorted to achieve precise switching on demand. This method abandons the blind switching mode triggered by traditional timed or fixed pressure difference, significantly reducing the waste of fluid controlled component resources and lowering operating costs. After switching, the characteristic parameters of the new fluid controlled component are updated in real time to form a closed-loop control, continuously optimizing the judgment and switching strategy, further extending the overall service life of the fluid controlled components and improving the system operating efficiency. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of an adaptive timing scheduling control method for multi-branch fluid elements according to the present invention;

[0066] Figure 2This is a flowchart illustrating the principle of an adaptive timing scheduling control method for multi-branch fluid elements according to the present invention.

[0067] Figure 3 This is a diagram of the architecture of an adaptive timing scheduling control system for multi-branch fluid elements according to the present invention. Detailed Implementation

[0068] Example 1

[0069] Please see Figures 1-2 The present invention provides an embodiment of an adaptive timing scheduling control method for multi-branch fluid elements, the method comprising S1 to S4, including the following steps:

[0070] S1: Obtain the real-time differential pressure, medium moisture content, and flue gas temperature of the fluid-controlled components in each branch, and extract four-dimensional characteristic parameters;

[0071] Furthermore, in each filtration branch of the industrial flue gas purification system, high-precision differential pressure sensors at both ends of the fluid-controlled element, moisture content sensors on the medium conveying pipeline, and temperature sensors in the flue are all connected to the real-time data acquisition unit. The data acquisition unit performs preliminary filtering on the data collected by each sensor to remove invalid noise data caused by equipment vibration, instantaneous electromagnetic interference, and minor airflow pulsations, retaining only the valid data that truly reflects the operating conditions of the fluid-controlled element. Subsequently, the data acquisition unit synchronizes and aligns the processed real-time differential pressure data, medium moisture content data, and flue gas temperature data according to a unified timestamp to ensure that the three types of data are completely matched in the time dimension. After completing the data synchronization and alignment, four-dimensional feature parameter extraction is performed based on the real-time differential pressure data.

[0072] Furthermore, this embodiment takes the bag filter fluid controlled element filtration branch in an industrial boiler flue gas purification system as an example. The system includes six parallel filtration branches, each equipped with a set of bag filter fluid controlled elements of the same specifications. High-precision differential pressure sensors are installed at the inlet and outlet of each branch fluid controlled element. The sensor measurement range covers the entire operating range of the fluid controlled element under normal operation, gradual ash accumulation, agglomeration and impact, and compaction and blockage. A capacitive moisture content sensor is installed near the inlet of the fluid controlled element in the medium conveying pipeline to monitor the moisture content of the medium entering the fluid controlled element in real time. A platinum resistance temperature sensor is installed near the inlet of the filter unit in the main flue to collect the flue gas temperature entering the filtration system in real time. All sensors are connected to the system's data acquisition terminal via industrial Ethernet. The terminal collects various data at a sampling frequency of 10 times per second, and at the same time, the built-in digital filtering unit automatically filters out high-frequency interference signals.

[0073] S2: Based on the four-dimensional feature parameters, distinguish between conventional progressive ash accumulation blockage, soft agglomeration impact blockage and hard agglomeration compaction blockage to obtain the preliminary classification result of the blockage condition. Then, based on the dual threshold judgment logic, re-verify the preliminary classification result of the blockage condition. At the same time, use the medium moisture content and flue gas temperature to dynamically correct the dual threshold judgment logic and output the corrected dual threshold judgment logic.

[0074] Furthermore, in this embodiment, considering the actual operating characteristics of the fluid-controlled components in the boiler flue gas purification system, conventional progressive ash accumulation blockage is characterized by a slow and steady increase in differential pressure of the fluid-controlled components over time, with all four-dimensional characteristic parameters showing a gentle and continuous trend without significant abrupt changes. Soft agglomeration impact blockage typically occurs when fine dust particles in the flue gas adhere to moisture and form soft agglomerates. When these agglomerates momentarily impact the surface of the fluid-controlled components, they manifest as rapid jumps and violent fluctuations in differential pressure within a short period, with a sharp increase in the short-term differential pressure distortion coefficient and differential pressure pulsation amplitude. Hard agglomeration compaction blockage occurs when soft agglomerates continuously accumulate and compress on the surface of the fluid-controlled components, gradually hardening and compacting, resulting in a continuous and slow increase in differential pressure, uneven pressure gradient distribution, and a steadily increasing differential pressure rise slope. By comprehensively comparing the four-dimensional characteristic parameters, the system can accurately identify the three types of blockage, avoiding misjudgment based on a single feature. At the same time, the dual-threshold judgment logic takes into account both instantaneous impact and long-term accumulation blockage risks, and combined with dynamic temperature and humidity correction thresholds, effectively improving the accuracy and adaptability of blockage judgment for fluid-controlled components.

[0075] S3: Fit the pressure difference time-series curve generated by the real-time pressure difference of each fluid controlled element according to the modified dual threshold judgment logic, predict the critical point of agglomeration, compaction, solidification and backflushing failure, trigger the element switching prediction signal, and output the identifier of the fluid element to be scheduled.

[0076] Furthermore, in this embodiment, dust in boiler flue gas tends to form soft agglomerates when humidity is high. Initially, these can be cleaned by pulse backflushing. However, as operating time increases, the agglomerates will continue to accumulate, harden, and gradually block the pores of the fluid-controlled elements. At this point, the backflushing cleaning effect will decrease significantly or even fail completely. If the fluid-controlled elements are not switched in time, it will lead to excessive system pressure difference, excessive flue gas emissions, and a surge in equipment energy consumption. By fitting the pressure difference time-series curve, the inflection point where the pressure difference curve changes from a gradual rise to a sharp rise can be accurately captured. This is the critical point where the agglomerates compact and solidify, and backflushing fails. This avoids wasting fluid-controlled elements due to premature switching or causing system failure due to delayed switching. The prediction accuracy can effectively meet the needs of stable operation of industrial systems.

[0077] S4: Collect the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element identifier to be scheduled, sort the agglomeration and blockage risk priority of each branch fluid controlled element, output the branch switching scheduling command, and complete the switching of fluid controlled elements according to the branch switching scheduling command. After the switching, collect the four-dimensional characteristic parameters of the newly put fluid controlled element in real time.

[0078] Furthermore, in this embodiment, the industrial flue gas purification system typically adopts a multi-branch parallel operation mode. The fluid control components in each branch are affected by differences in flue gas distribution, dust concentration, and humidity, resulting in varying degrees of blockage and risks. If only a single faulty fluid control component is switched, it can easily lead to excessive performance differences among the fluid control components in each branch, causing system imbalance. By prioritizing the blockage risk of all fluid control components, precise and orderly switching of fluid control components can be achieved. The fluid control components with the highest risk and worst performance are replaced first, while the optimal backup fluid control components are matched to ensure balanced performance of fluid control components in each branch and stable system operation. After the switch is completed, feature extraction and status monitoring of the new fluid control components are performed immediately, enabling real-time monitoring of the operating status of the new fluid control components and timely detection of potential anomalies.

[0079] Extract four-dimensional feature parameters, including:

[0080] S1.1: The real-time pressure difference is segmented and extracted in the time domain based on the moisture content of the medium and the flue gas temperature to obtain segmented pressure difference time series data;

[0081] Furthermore, the specific process of segmenting the real-time differential pressure based on the moisture content of the medium and the flue gas temperature to obtain segmented differential pressure time-series data includes: firstly, retrieving the currently collected real-time moisture content and flue gas temperature values; dividing the moisture content into three intervals (low, medium, and high humidity) and the flue gas temperature into three intervals (low, medium, and high temperature); and combining these intervals to form multiple different operating condition combinations; retrieving the pre-constructed operating condition-time domain segmentation rule mapping table; and, based on the current operating condition combination, matching the table to obtain the corresponding time domain segmentation rule. The time domain segmentation rule clearly defines the key parameters used for differential pressure feature extraction under the current operating condition, including the single-segment segmentation time... The data collection process involves defining the length of the data, the starting interval between adjacent segments, the overlap ratio of data between consecutive segments, and the threshold for abnormal fluctuations. Subsequently, the continuously acquired real-time differential pressure time-series data is segmented according to defined truncation rules, dividing the complete long-series differential pressure data into multiple fixed-duration, time-continuous, and non-interfering segments. Each segment corresponds to a specified operating condition interval, accurately reflecting the pressure difference variation characteristics of the fluid-controlled components under that condition, avoiding distortion in feature extraction caused by mixing data from different operating conditions. After segmentation, the validity of each segment of differential pressure time-series data is verified, eliminating invalid segments with missing data or excessive abnormal fluctuations, retaining only complete, stable, and valid segmented differential pressure time-series data.

[0082] Furthermore, the operating condition-time domain truncation rule mapping table is pre-constructed using 24 months of full historical operating data from the system, after offline calibration. Operating condition intervals are first precisely defined, with medium moisture content categorized as: low moisture (less than 10%), medium moisture (10%~25%), and high moisture (greater than 25%); flue gas temperature is categorized as: less than 120℃ (low temperature), 120℃~180℃ (medium temperature), and greater than 180℃ (high temperature). These two intervals are then combined to generate low-moisture low-temperature, low-moisture medium-temperature, low-moisture high-temperature, medium-moisture low-temperature, and medium-moisture... Nine independent operating condition combinations were identified: medium temperature, medium humidity and high temperature, high humidity and low temperature, high humidity and medium temperature, and high humidity and high temperature. For each operating condition combination, four core extraction parameters were calibrated based on the corresponding pressure difference variation patterns: single-segment extraction duration, starting interval between adjacent segments, data overlap ratio between consecutive segments, and abnormal fluctuation threshold. The calibration followed these principles: for high humidity and low temperature conditions, where soft agglomeration and impact are frequent and pressure difference changes are severe, shorter segment durations, lower data overlap ratios, and lower abnormal fluctuation thresholds were configured; for low humidity and high temperature conditions, where dust dispersion is good... For pressure differentials that change slowly and steadily, a long segment duration, high data overlap ratio, and high abnormal fluctuation threshold are configured. For other intermediate operating conditions, linear interpolation is used to match the corresponding parameters. Here, the single segment interception duration refers to the continuous collection duration of single segment pressure differential data, the adjacent segment start interval refers to the difference in start time between two segments, the overlap ratio of preceding and following segments refers to the percentage of shared data between adjacent segments in the total duration of a single segment, and the abnormal fluctuation judgment threshold refers to the maximum deviation of the pressure differential within a segment from the benchmark pressure differential limit, such as 150 Pa. If this threshold is exceeded, the segment data is judged to be abnormal. The operating condition-time domain interception rule mapping table is stored in the system control logic storage unit using a two-dimensional key-value pair structure with moisture content range and temperature range combination identifiers as keys and four interception parameters as values. At the same time, an automatic iterative update mechanism is set. Every 30 days of system operation, new operating condition running data is automatically collected, and the interception parameters corresponding to each operating condition combination are refitted and corrected to complete the dynamic iterative optimization of the mapping table, ultimately forming a standard operating condition-time domain interception rule mapping table that adapts to all operating conditions.

[0083] Furthermore, in this embodiment, the moisture content and temperature of the boiler flue gas directly affect the formation rate, viscosity, and adhesion strength of agglomerated pollutants on the surface of the fluid-controlled element, thus leading to significant differences in the pressure difference variation of the fluid-controlled element. For example, under high humidity and low temperature conditions, dust easily forms highly viscous soft agglomerates, causing drastic fluctuations in the pressure difference of the fluid-controlled element; under low humidity and high temperature conditions, dust has good dispersion, and the pressure difference of the fluid-controlled element rises steadily. By segmenting the pressure difference data under different temperature and humidity conditions, the pressure difference variation characteristics under different conditions can be separated, making the extracted feature parameters more targeted and representative, and improving the ability of feature parameters to characterize the blockage state of the fluid-controlled element.

[0084] S1.2: Based on the segmented differential pressure time series data, the differential pressure rise slope is obtained;

[0085] Furthermore, the specific process of obtaining the pressure difference rise slope based on segmented pressure difference time series data includes: for each valid segmented pressure difference time series data, firstly, the segmented pressure difference time series data is smoothed preprocessed to eliminate small random fluctuations in the segmented pressure difference time series data and retain the overall upward trend of the pressure difference; then, in each segmented pressure difference time series data, the pressure difference value at the start time is selected as the baseline value, and the pressure difference value at the end time is selected as the final value, while the time span corresponding to this segment is recorded; based on the baseline pressure difference, the final pressure difference, and the time span, the average rate of change of the pressure difference with time during this period is calculated, which is the pressure difference rise slope corresponding to this segment; for segments with obvious steps... The segmented data with phased changes can be further subdivided into multiple sub-time periods. The rising slope of each sub-time period is calculated, and then the average is taken to obtain the comprehensive rising slope of the entire data segment. This ensures that the slope can accurately reflect the overall rising trend of the pressure difference of the fluid-controlled component within that time period. After calculating the slope of all segmented pressure difference time series data in sequence, a complete pressure difference rising slope sequence is obtained. This sequence can intuitively reflect the rate of increase of the pressure difference of the fluid-controlled component with changes in operating conditions. It is a core characteristic indicator for judging the progressive ash accumulation blockage and hard agglomeration compaction blockage of the fluid-controlled component. The calculation formulas for the slope and the mean are existing technologies in this field and are not the inventive solutions of this application. They will not be elaborated here.

[0086] S1.3: Based on the segmented differential pressure time series data, calculate the maximum fluctuation difference between the real-time differential pressure and the preset pulsating reference differential pressure, and obtain the differential pressure pulsation amplitude;

[0087] Furthermore, the specific process of calculating the maximum fluctuation difference of the real-time differential pressure deviating from the preset pulsating reference differential pressure based on segmented differential pressure time series data, and obtaining the differential pressure pulsation amplitude, includes: firstly, retrieving the preset pulsating reference differential pressure value. The pulsating reference differential pressure is set based on the historical average differential pressure of the fluid controlled element under clean conditions, normal operation, and no agglomeration impact conditions. It is a benchmark reference value for measuring differential pressure fluctuations. In the industrial boiler flue gas purification bag fluid controlled element system applied in this embodiment, it is uniformly set to 150Pa; for each segmented differential pressure time series data, the real-time differential pressure at each sampling time within that segment is traversed. For each real-time differential pressure value, the difference between the real-time differential pressure value and the pulsating reference differential pressure is calculated to form a differential pressure deviation difference sequence. The absolute value of the differential pressure deviation difference sequence is processed to eliminate the influence of positive and negative deviation directions, focusing only on the magnitude of the deviation. Then, the difference with the largest value is selected from the differential pressure deviation difference difference sequence after absolute value processing. This difference is the differential pressure pulsation amplitude corresponding to the current segment of differential pressure time series data. The pulsation amplitude calculation of all segment data is completed in sequence, and the results are summarized to form a differential pressure pulsation amplitude sequence, which can accurately reflect the intensity of the fluctuation of the differential pressure of the fluid controlled element caused by the impact of agglomerated pollutants.

[0088] Furthermore, in this embodiment, when soft agglomerated pollutants instantaneously impact the fluid-controlled element, the pressure difference will fluctuate violently within a short period of time, and the deviation from the normal reference pressure difference will increase. During the conventional gradual ash accumulation and hard agglomerated compaction process, the pressure difference fluctuation amplitude is small and close to the reference pressure difference. The amplitude of the pressure difference pulsation can quantitatively measure the severity of the pressure difference fluctuation. The larger the amplitude, the stronger the agglomerated impact and the higher the risk of soft agglomerated blockage. This feature can be used to quickly distinguish instantaneous impact blockage from other types of blockage.

[0089] S1.4: Based on the segmented pressure difference time series data, extract the sudden pressure difference corresponding to the instantaneous impact of aggregated pollutants, substitute the sudden pressure difference as a weighting coefficient into the pre-constructed short-time pressure difference distortion coefficient calculation model, and output the short-time pressure difference distortion coefficient.

[0090] Furthermore, the specific process of extracting the abrupt pressure difference corresponding to the instantaneous impact of agglomerated pollutants based on segmented pressure difference time-series data, and substituting the abrupt pressure difference as a weighting coefficient into the pre-constructed short-time pressure difference distortion coefficient calculation model to output the short-time pressure difference distortion coefficient includes: firstly, detecting abrupt points in each segmented pressure difference time-series data; by comparing the pressure difference change amplitude at adjacent sampling times, identifying the moments when the pressure difference changes abruptly within a short period of time. These abrupt changes are the direct manifestation of the instantaneous impact of agglomerated pollutants on the fluid-controlled components, and the corresponding pressure difference change is the abrupt pressure difference; extracting the abrupt pressure difference value corresponding to each abrupt event to ensure that the extraction result accurately reflects the impact intensity; and then retrieving the pre-constructed short-time pressure difference distortion coefficient. The coefficient calculation model for short-term differential pressure distortion has been trained and optimized using a large amount of historical operating data, and can accurately correlate sudden pressure drops with the degree of short-term differential pressure distortion. The extracted sudden pressure drops are used as the weight input parameters of the short-term differential pressure distortion calculation model and substituted into the model for calculation. The short-term differential pressure distortion calculation model combines information such as sudden pressure drop weights, impact duration, and historical distortion data to comprehensively calculate the short-term differential pressure distortion coefficient at the current moment. The short-term differential pressure distortion coefficient can quantitatively characterize the degree of distortion of the pressure time series curve caused by the instantaneous impact of agglomerated pollutants. The larger the distortion coefficient, the more severe the pressure distortion caused by the impact and the more severe the local blockage of the fluid-controlled components.

[0091] Furthermore, in this embodiment, the instantaneous impact of agglomerated pollutants not only causes pressure differential fluctuations but also causes the pressure differential time-series curve to deviate from its normal smooth upward trend, resulting in significant distortion. The short-term pressure differential distortion coefficient comprehensively considers factors such as the magnitude of the sudden pressure differential change and the duration of the impact. Compared with the single pressure differential pulsation amplitude, it can more comprehensively reflect the degree of damage to the pressure differential change law of the fluid-controlled element caused by the instantaneous impact. The higher the distortion coefficient, the more concentrated the impact of the soft agglomerates, the more severe the local blockage, and the higher the risk of compaction and solidification.

[0092] S1.5: Arrange multiple pressure acquisition units along the axial direction of the folds of the fluid-controlled element to acquire the axial segmented pressure data of the folds, divide multiple pressure gradient detection intervals, calculate the differential pressure gradient value of each interval based on the axial segmented pressure data, and then obtain the differential pressure gradient difference between two adjacent pressure gradient detection intervals to obtain the pressure gradient difference of the fold interval.

[0093] Furthermore, the specific process of dividing the folds of the fluid-controlled element into multiple pressure gradient detection intervals along their axial direction, calculating the pressure gradient value for each interval, and then obtaining the pressure gradient difference between two adjacent pressure gradient detection intervals to obtain the pressure gradient difference between the fold intervals includes: first, determining the total length of the folds along their axial direction. In this embodiment, the total axial length of the folds is uniformly set to 800mm. The folds are then uniformly divided according to a preset interval division precision of 10mm, meaning the entire 800mm axial length is uniformly divided into 80 continuous, equal-length, non-overlapping pressure gradient detection intervals. The test intervals are each 10mm long and correspond to a continuous section along the axial direction of the folds in the fluid-controlled element, covering the entire folded area from the inlet to the outlet. Next, real-time differential pressure data corresponding to each pressure gradient detection interval is retrieved. This data is collected by miniature pressure sensors pre-positioned at different locations along the axial direction of the folds in the fluid-controlled element. These miniature pressure sensors are evenly spaced at 10mm intervals, with a total of 80 sensors. Each sensor corresponds to one pressure gradient detection interval. The unit of the collected real-time differential pressure data is uniformly set to Pa, enabling precise measurement. The system accurately reflects the real-time pressure status within the corresponding interval. Next, the real-time differential pressure data within each pressure gradient detection interval is preprocessed using a moving average filtering algorithm. The moving average window length is set to 5 sampling points, and the step size is set to 1 sampling point. Instantaneous random fluctuations are filtered out using the moving average filtering algorithm, outputting smoothed differential pressure data for each interval. Then, a first-order difference algorithm is used to calculate the differential pressure gradient value based on the smoothed differential pressure data of each interval. The parameters of the first-order difference algorithm are set to an axial distance step size of 10 mm. The formula for calculating the differential pressure gradient value is the current interval's smoothed differential pressure data minus the smoothed differential pressure data of the adjacent previous interval. Divide by the axial distance step size to obtain the pressure gradient value corresponding to each interval; then arrange all the calculated pressure gradient values ​​in order of the axial direction of the fluid controlled element folds to form a pressure gradient sequence; finally, calculate the difference between each pair of adjacent pressure gradient values ​​in the pressure gradient sequence in turn. The difference between adjacent values ​​is calculated by direct subtraction. The result is the pressure gradient difference of the fold interval. All pressure gradient differences of the fold intervals are arranged in order to form a pressure gradient difference sequence of the fold intervals. The specific calculation formulas of the moving average filtering algorithm and the first-order difference algorithm are existing technologies in this field and are not the inventive solution of this application. They will not be elaborated here.

[0094] Furthermore, in this embodiment, hard agglomerate compaction blockage is usually manifested as a large amount of agglomerate clumps accumulating and hardening in a local area of ​​the fluid-controlled element, resulting in uneven pressure distribution in the axial direction of the fluid-controlled element and a sharp increase in the local pressure gradient; in conventional progressive ash accumulation blockage, the fluid-controlled element is uniformly blocked and the pressure gradient difference is small; the pressure gradient difference in the folded area can accurately capture the characteristics of local compaction blockage of the fluid-controlled element and effectively distinguish between uniform ash accumulation blockage and local compaction blockage.

[0095] S1.6: The pressure difference rise slope, pressure difference pulsation amplitude, short-time pressure difference distortion coefficient, and pressure gradient difference in the folded area are output as four-dimensional feature parameters.

[0096] Furthermore, in this embodiment, the four-dimensional feature parameters each have their own emphasis and complement each other: the slope of the pressure difference rise reflects the rate of blockage accumulation, the amplitude of the pressure difference pulsation reflects the instantaneous impact intensity, the short-term pressure difference distortion coefficient reflects the degree of impact distortion, and the pressure gradient difference in the folded area reflects the uniformity of blockage distribution; after the four parameters are combined, they can comprehensively cover the characteristics of the three types of blockage: conventional ash accumulation, soft agglomeration impact, and hard agglomeration compaction, avoiding the limitations of a single feature and improving the comprehensiveness and accuracy of the blockage status identification of fluid-controlled components.

[0097] The construction process of the short-time pressure difference distortion coefficient calculation model includes:

[0098] S1.4.1: Obtain segmented pressure difference time series data under historical operating conditions and label sudden pressure difference events caused by instantaneous impacts of agglomerated pollutants;

[0099] Furthermore, the acquisition of segmented differential pressure time-series data under historical operating conditions includes: retrieving the complete historical operating database of the industrial flue gas purification system over the past few months or even years. The database stores full data such as real-time differential pressure, medium moisture content, flue gas temperature, and equipment operating parameters of each branch fluid-controlled component under different operating conditions and different operating stages; the system segments the real-time differential pressure data in the historical database according to segmentation rules to generate historical segmented differential pressure time-series data, covering various typical operating conditions such as low humidity and low temperature, low humidity and high temperature, medium humidity and medium temperature, high humidity and low temperature, and high humidity and high temperature, to ensure the comprehensiveness and representativeness of the historical data.

[0100] Furthermore, the specific process of labeling sudden pressure difference events caused by instantaneous impacts of agglomerated pollutants includes: the system organizes professional technicians to manually label each segment of historical pressure difference time series data, combining historical operating condition records, equipment operation logs, manual inspection reports, and other auxiliary information; during the labeling process, technicians focus on identifying moments when the pressure difference changes abruptly and deviates from the normal trend within a short period of time, and confirm whether these sudden changes are caused by instantaneous impacts of agglomerated pollutants, in conjunction with operating condition information such as flue gas dust concentration, humidity, and temperature at that time; for confirmed instantaneous impact events, technicians accurately label key information such as the start time, end time, sudden pressure difference value, and impact duration of the event, forming a sudden pressure difference event labeling library containing a large number of labeled samples.

[0101] Furthermore, in this embodiment, the historical operating condition data covers various actual operating scenarios such as system startup, normal operation, load fluctuation, sudden changes in dust concentration, and abnormal increase in humidity. The labeled sudden pressure difference events include soft agglomeration impact cases with different intensities, durations, and operating conditions. The sample quantity is sufficient, the types are rich, and the labeling is accurate, which can comprehensively reflect the characteristics and laws of instantaneous impact of agglomerated pollutants in actual operating conditions.

[0102] S1.4.2: Extract the impact amplitude and impact duration of each sudden pressure difference event, and define the ratio of impact amplitude to impact duration as the instantaneous impact intensity;

[0103] Furthermore, the specific process of extracting the impact amplitude and impact duration of each sudden pressure difference event, and defining the ratio of impact amplitude to impact duration as the instantaneous impact intensity, includes: the system retrieving the constructed sudden pressure difference event annotation library, traversing each annotated instantaneous impact event of aggregated pollutants in the library; for each annotated instantaneous impact event of aggregated pollutants, the system extracts two core key parameters from the annotation information: one is the impact amplitude, that is, the maximum change in pressure difference from the start time of the sudden change to the peak time during the impact, which directly reflects the severity of the impact; the other is the impact... Duration, or the total time from the onset of an impact to its return to normal, reflects the sustained impact range. The system ensures the accuracy of impact amplitude and duration extraction, down to the minimum sampling period, to avoid extraction errors. Subsequently, for each impact event, the ratio of impact amplitude to impact duration is calculated, and this ratio is defined as the instantaneous impact intensity. The instantaneous impact intensity integrates both impact amplitude and duration, enabling a more comprehensive and accurate quantification of the overall intensity of the instantaneous impact of agglomerated pollutants. Compared to a single impact amplitude or duration, it better reflects the degree of damage to fluid-controlled components caused by the impact.

[0104] Furthermore, in this embodiment, different instantaneous impact events of aggregated pollutants have significant differences in impact amplitude and duration: some impacts have large amplitudes and short durations, belonging to high-intensity instantaneous impacts; some impacts have small amplitudes and long durations, belonging to low-intensity sustained impacts. Combining the two instantaneous impact intensities can uniformly quantify the comprehensive intensity of different types of impacts, avoiding the one-sidedness of evaluation by a single parameter, enabling the short-time pressure difference distortion coefficient calculation model to accurately identify impact events of different intensities, and improving the adaptability of the short-time pressure difference distortion coefficient calculation model to various impact conditions.

[0105] S1.4.3: Using instantaneous impact intensity as the independent variable and short-term pressure difference distortion coefficient as the target variable, an exponentially weighted moving average method is used to fit and construct a calculation model for the short-term pressure difference distortion coefficient. The expression for the calculation model of the short-term pressure difference distortion coefficient is as follows: Where D is the current short-term pressure difference distortion coefficient, and I is the instantaneous impact intensity. As a smoothing factor, The output is the short-time differential pressure distortion coefficient of the previous moment, and the completed short-time differential pressure distortion coefficient calculation model.

[0106] Furthermore, the specific process of constructing a short-term pressure difference distortion coefficient calculation model using the exponential weighted moving average method with instantaneous impact intensity as the independent variable and short-term pressure difference distortion coefficient as the target variable, and outputting the completed short-term pressure difference distortion coefficient calculation model, includes: retrieving the sudden pressure difference event annotation library, extracting the instantaneous impact intensity data corresponding to each sudden pressure difference event from the library, the instantaneous impact intensity being calculated from the impact amplitude and impact duration of the sudden pressure difference event, with the impact amplitude unit uniformly in Pa and the impact duration unit uniformly in seconds; retrieving the quantitative analysis results of the distortion degree of the pressure difference time series curve, which are output from the pressure difference time series curve calculation process; and combining the instantaneous impact intensity data with the short-term pressure difference distortion coefficient calculation model. Based on the instantaneous impact intensity data and quantitative analysis results, the target value of the short-term pressure difference distortion coefficient corresponding to each sudden pressure difference event is determined. The target value of the short-term pressure difference distortion coefficient is uniformly set to a range of 0 to 1. All instantaneous impact intensity data and corresponding target values ​​of the short-term pressure difference distortion coefficient are summarized to form a training dataset. The training dataset consists of the independent variable instantaneous impact intensity and the target variable short-term pressure difference distortion coefficient. Preprocessing operations are performed on the training dataset, including two sub-steps: outlier removal and data normalization. Outlier removal adopts the 3σ criterion, with the standard deviation factor parameter of the 3σ criterion uniformly set to 3. Data normalization adopts a linear normalization algorithm, with the mapping interval parameter of the linear normalization algorithm uniformly set to 0 to... 1. An exponentially weighted moving average method is used to construct a short-term pressure difference distortion coefficient calculation model. This model comprises three layers: an input layer, a weighted calculation layer, and an output layer. The input layer receives the instantaneous impact intensity independent variable, the weighted calculation layer performs the exponentially weighted moving average calculation, and the output layer outputs the short-term pressure difference distortion coefficient target variable. Training parameters for the exponentially weighted moving average method are set, including the initial value of the smoothing factor, the range of the smoothing factor, the number of training iterations, and the error convergence threshold. Specifically, the initial value of the smoothing factor is uniformly set to 0.3, the range of the smoothing factor is uniformly set to 0.1 to 0.9, the number of training iterations is uniformly set to 500, and the error convergence threshold is uniformly set... The value is 0.001; the eighth step is to start iterative training of the model. In each iteration, the weighted calculation layer calculates the weighted value of the historical short-term pressure difference distortion coefficient based on the current smoothing factor. When calculating the weighted value, the impact data of the last 10 sampling times are assigned a weight of 70%, and the impact data before the last 10 sampling times are assigned a weight of 30%. The model output value is generated by fitting the weighted value with the current instantaneous impact intensity. The mean square error between the model output value and the target value of the short-term pressure difference distortion coefficient in the training dataset is calculated. The mean square error is calculated using the mean square error algorithm. When the mean square error is greater than the error convergence threshold, the smoothing factor value is updated in reverse using the gradient descent method. The learning rate parameter of the gradient descent method is uniformly set to 0.01; Repeat the iterative training and smoothing factor update process until the number of iterations reaches the preset number or the mean square error is less than or equal to the error convergence threshold to complete the model parameter optimization; Perform model fitting effect verification. The verification process selects test datasets under low humidity and low temperature, low humidity and high temperature, medium humidity and medium temperature, high humidity and low temperature, and high humidity and high temperature conditions. The proportion of the test dataset to the total data volume is uniformly set to 20%. The verification index is the mean absolute error. When the mean absolute error is less than or equal to 0.02, the fitting effect is considered qualified; Perform model generalization ability test. The test process selects instantaneous impact event data of aggregated pollutants that have not participated in the training. The impact amplitude range of the test data is uniformly set to 20%. The impact pressure is set to a range of 50 Pa to 500 Pa, and the impact duration is uniformly set to a range of 0.5 seconds to 5 seconds. A pass rate of over 95% indicates acceptable generalization ability. Once the model fitting accuracy meets the preset requirements and the generalization ability passes the test, a short-time pressure difference distortion coefficient calculation model is output, with optimized parameters, ready for direct real-time operation. The model input is the instantaneous impact intensity, and the model output is the short-time pressure difference distortion coefficient. The calculation formulas for the linear normalization algorithm, exponentially weighted moving average calculation, squared error mean algorithm, and gradient descent method are existing technologies in this field and are not inventive solutions for this application; therefore, they will not be elaborated upon here.

[0107] Based on the aforementioned four-dimensional characteristic parameters, conventional progressive ash accumulation blockage, soft agglomeration impact blockage, and hard agglomeration compaction blockage can be distinguished, including:

[0108] S2.1: Preset the steady-state threshold range of conventional ash accumulation, the instantaneous change threshold range of soft agglomeration, and the cumulative threshold range of hard agglomeration compaction corresponding to the four-dimensional feature parameters, and match the feature response modes of conventional progressive ash accumulation blockage, soft agglomeration impact blockage, and hard agglomeration compaction blockage respectively;

[0109] Furthermore, the specific process of matching the characteristic response modes of conventional progressive ash accumulation blockage, soft agglomeration impact blockage, and hard agglomeration compaction blockage corresponding to the preset four-dimensional characteristic parameters to the conventional ash accumulation steady-state threshold range, soft agglomeration instantaneous change threshold range, and hard agglomeration compaction accumulation threshold range respectively includes: retrieving the historical operating condition four-dimensional characteristic parameter statistical database of the industrial flue gas purification system. The storage time span of the historical operating condition four-dimensional characteristic parameter statistical database is uniformly set to 24 months, containing full data of four-dimensional characteristic parameters corresponding to the three blockage types: conventional progressive ash accumulation blockage, soft agglomeration impact blockage, and hard agglomeration compaction blockage. The four-dimensional characteristic parameters include the pressure difference rise slope, pressure difference pulsation amplitude, short-time pressure difference distortion coefficient, and pressure gradient difference in the folded area; from the historical operating condition four-dimensional characteristic parameters... Statistical data on the pressure differential rise slope corresponding to conventional progressive ash accumulation blockage, soft agglomeration impact blockage, and hard agglomeration compaction blockage were extracted from the characteristic parameter statistical database. The unit of the pressure differential rise slope was uniformly set to Pa / s. A numerical distribution statistical algorithm was used to analyze the numerical distribution range, fluctuation pattern, and trend of the pressure differential rise slope under the three blockage types. The statistical interval step size parameter of the numerical distribution statistical algorithm was uniformly set to 0.1 Pa / s. Based on the analysis results of the numerical distribution statistical algorithm, three threshold intervals corresponding to the pressure differential rise slope were set: the steady-state threshold interval for conventional ash accumulation was uniformly set to 0.1 Pa / s to 0.5 Pa / s; the instantaneous change threshold interval for soft agglomeration was uniformly set to 1.2 Pa / s to 3.0 Pa / s; and the threshold interval for hard agglomeration compaction was set to... The cumulative threshold range was uniformly set to 0.6 Pa / s to 1.1 Pa / s. Differential pressure pulsation amplitude statistics were extracted from the historical four-dimensional characteristic parameter statistical database for three different blockage types, with the unit for differential pressure pulsation amplitude uniformly set to Pa. A numerical distribution statistical algorithm was used to analyze the numerical distribution range, fluctuation patterns, and trends of differential pressure pulsation amplitude under the three blockage types. The statistical interval step size parameter of the numerical distribution statistical algorithm was uniformly set to 50 Pa. Based on the analysis results of the numerical distribution statistical algorithm, three threshold ranges corresponding to the differential pressure pulsation amplitude were set: the steady-state threshold range for conventional ash accumulation was uniformly set to 100 Pa to 300 Pa; the instantaneous change threshold range for soft agglomeration was uniformly set to 800 Pa to 2000 Pa; and the threshold range for hard agglomeration... The cumulative threshold range for agglomeration compaction was uniformly set to 400 Pa to 750 Pa. Statistical data on short-term differential pressure distortion coefficients corresponding to three blockage types were extracted from the historical four-dimensional characteristic parameter statistical database. The value range of the short-term differential pressure distortion coefficients was uniformly set to 0 to 1. A numerical distribution statistical algorithm was used to analyze the numerical distribution range, fluctuation patterns, and trends of the short-term differential pressure distortion coefficients under the three blockage types. The statistical interval step size parameter of the numerical distribution statistical algorithm was uniformly set to 0.05. Based on the analysis results of the numerical distribution statistical algorithm, three threshold ranges corresponding to the short-term differential pressure distortion coefficients were set: the steady-state threshold range for conventional ash accumulation was uniformly set to 0.05 to 0.20, and the instantaneous abrupt change threshold range for soft agglomeration was uniformly set to 0.50 to 0.90. The cumulative threshold range for hard agglomeration compaction is uniformly set to 0.25 to 0.45. Statistical data on the pressure gradient difference in folded sections corresponding to three blockage types are extracted from the historical four-dimensional characteristic parameter statistical database. The unit for the pressure gradient difference in folded sections is uniformly set to Pa / mm. A numerical distribution statistical algorithm is used to analyze the numerical distribution range, fluctuation pattern, and trend of the pressure gradient difference in folded sections under the three blockage types. The step size parameter of the statistical interval in the numerical distribution statistical algorithm is uniformly set to 2 Pa / mm. Based on the analysis results of the numerical distribution statistical algorithm, three threshold ranges corresponding to the pressure gradient difference in folded sections are set: the steady-state threshold range for conventional ash accumulation is uniformly set to 2 Pa / mm to 8 Pa / mm; the instantaneous change threshold range for soft agglomeration is uniformly set to 18 Pa / mm to 40 Pa / mm; and the cumulative threshold range for hard agglomeration compaction is uniformly set to 9 Pa / mm to 17 Pa / mm. The corresponding four-dimensional characteristic parameters are verified. The boundaries of the three threshold intervals are designed to ensure that the boundaries of the steady-state threshold interval for conventional ash accumulation, the instantaneous change threshold interval for soft agglomeration, and the cumulative threshold interval for hard agglomeration compaction are clear and non-overlapping. A transition interval is reserved between adjacent threshold intervals, with the width of the transition interval uniformly set to 1.5 times the step size of the corresponding feature parameter statistical interval. The steady-state threshold intervals for conventional ash accumulation, corresponding to the pressure difference rise slope, pressure difference pulsation amplitude, short-term pressure difference distortion coefficient, and pressure gradient difference in the folded interval, are matched with the characteristic response mode of conventional gradual ash accumulation blockage. The instantaneous change threshold interval for soft agglomeration is matched with the characteristic response mode of soft agglomeration impact blockage, and the cumulative threshold interval for hard agglomeration compaction is matched with the characteristic response mode of hard agglomeration compaction blockage. The matching relationships of the four-dimensional feature parameters, the three threshold intervals, and the characteristic response modes of the three blockage types are summarized to construct a feature-threshold-blockage type mapping table. The numerical distribution statistical algorithm is existing technology in this field and is not an inventive solution of this application; therefore, it will not be elaborated upon here.

[0110] Furthermore, in this embodiment, the four-dimensional characteristic parameters of the three types of blockage show significant differences: in the case of conventional progressive ash accumulation blockage, the pressure difference rise slope is small, the pulsation amplitude is low, the distortion coefficient is close to zero, and the pressure gradient difference is uniform; in the case of soft agglomeration impact blockage, the pressure difference pulsation amplitude and distortion coefficient far exceed the conventional range, and instantly jump to the abrupt change threshold range; in the case of hard agglomeration compaction blockage, the pressure difference rise slope and pressure gradient difference continue to increase, entering the compaction accumulation threshold range; by presetting a precise threshold range, the real-time four-dimensional characteristic parameters can be quickly and accurately matched with the three types of blockage, realizing the initial differentiation of blockage types and laying the foundation for the subsequent establishment of dual threshold judgment logic.

[0111] S2.2: Compare the differential pressure rise slope, differential pressure pulsation amplitude, short-term differential pressure distortion coefficient, and pressure gradient difference in the folded area with the corresponding steady-state threshold range of conventional ash accumulation, the instantaneous change threshold range of soft agglomeration, and the cumulative threshold range of hard agglomeration compaction one by one, and extract the transient response component and steady-state cumulative component of each characteristic parameter.

[0112] Furthermore, the specific process of comparing the pressure differential rise slope, pressure differential pulsation amplitude, short-time pressure differential distortion coefficient, and pressure gradient difference in the folded area with the corresponding steady-state threshold ranges for conventional ash accumulation, instantaneous change threshold ranges for soft agglomeration, and cumulative compaction threshold ranges for hard agglomeration, and extracting the transient response components and steady-state cumulative components of each feature parameter includes: the system sequentially retrieves the four-dimensional feature parameters under the current real-time operating condition, as well as the preset three types of threshold ranges; each feature parameter, including the pressure differential rise slope, pressure differential pulsation amplitude, short-time pressure differential distortion coefficient, and pressure gradient difference in the folded area, is compared with its corresponding steady-state threshold range for conventional ash accumulation, instantaneous change threshold range for soft agglomeration, and cumulative compaction threshold range for hard agglomeration, in sequence; during the comparison process, it is determined which threshold range the current feature parameter value falls into: if it falls into the steady-state threshold range for conventional ash accumulation, it indicates that the current feature parameter value falls into the corresponding threshold range. The feature parameters are in a stable and normal state. If they fall into the threshold range of instantaneous mutation in soft agglomeration, it indicates that the feature parameter has an instantaneous mutation response. If they fall into the threshold range of cumulative compaction in hard agglomeration, it indicates that the feature parameter exhibits steady-state cumulative change. Based on the comparison results, the corresponding transient response component and steady-state cumulative component are extracted for each feature parameter: the transient response component characterizes the degree of mutation of the feature parameter under the influence of instantaneous impact, and is only effective when the feature parameter falls into the threshold range of instantaneous mutation in soft agglomeration; the steady-state cumulative component characterizes the degree of cumulative change of the feature parameter over time, and is effective when the feature parameter falls into the steady-state threshold range of conventional ash accumulation or the threshold range of cumulative compaction in hard agglomeration. The system sequentially completes the comparison and component extraction of all four-dimensional feature parameters to ensure that the transient response and steady-state cumulative features of each feature parameter can be accurately captured, providing basic component data for the construction of the subsequent feature coupling judgment matrix.

[0113] Furthermore, in this embodiment, the transient response component mainly reflects the short-term and drastic changes in characteristic parameters caused by the instantaneous impact of soft agglomerate pollutants, which is the key to identifying soft agglomerate impact blockage; the steady-state cumulative component mainly reflects the long-term and stable cumulative changes in characteristic parameters caused by the ash accumulation and compaction process of fluid-controlled components, which is the core of identifying conventional gradual ash accumulation blockage and hard agglomerate compaction blockage; by extracting the components, the four-dimensional characteristic parameters are decomposed into two types of core information, namely transient and steady-state, which can more accurately focus on the core characteristics of different blockage types, reduce interference from irrelevant information, and improve the accuracy of blockage type determination.

[0114] S2.3: Based on the extracted transient response components and steady-state cumulative components, a feature coupling judgment matrix is ​​constructed. The transient change characteristics are represented by the differential pressure pulsation amplitude and the short-time differential pressure distortion coefficient to match soft agglomeration impact blockage. The steady-state cumulative characteristics are represented by the differential pressure rise slope and the pressure gradient difference in the folded interval to match hard agglomeration compaction blockage. When the four-dimensional feature parameters show a steady-state change, it matches conventional progressive ash accumulation blockage, thus obtaining the preliminary classification results of the blockage conditions.

[0115] Further, the specific process of S2.3 includes: retrieving the extracted transient response components and steady-state cumulative components. The transient response components consist of transient response components of pressure difference pulsation amplitude and transient response components of short-time pressure difference distortion coefficient. The steady-state cumulative components consist of steady-state cumulative components of pressure difference rise slope and steady-state cumulative components of pressure gradient difference in the folded area. Numerical normalization processing is performed on the transient response components and steady-state cumulative components. The normalization processing uses a linear normalization algorithm, with the mapping interval parameter of the linear normalization algorithm uniformly set to 0 to 1. The normalized transient response components and normalized steady-state cumulative components are output. The normalized pressure difference pulsation... The transient response components of the amplitude value and the transient response components of the short-time pressure difference distortion coefficient are weighted and fused. The weight parameters of the weighting fusion algorithm are set to 0.5 and 0.5 respectively, and the comprehensive value of the transient change feature group is output. The steady-state cumulative components of the normalized pressure difference rise slope and the steady-state cumulative components of the pressure gradient difference in the folded area are weighted and fused. The weight parameters of the weighting fusion algorithm are set to 0.5 and 0.5 respectively, and the comprehensive value of the steady-state cumulative feature group is output. A two-dimensional feature coupling judgment matrix is ​​constructed, with the matrix dimension uniformly set to 3×3. The rows correspond to the strength level of the transient change feature, and the columns correspond to the high and low levels of the steady-state cumulative feature. The matrix level is set. Thresholds are uniformly set at 0.3 and 0.7 for classifying the strength of transient mutation features and steady-state cumulative features, respectively. Transient mutation features are divided into three levels: weak, medium, and strong; steady-state cumulative features are divided into three levels: low, medium, and high. Pre-defined matching relationships are applied to each cell of the two-dimensional feature coupling judgment matrix: cells with weak transient mutation features are matched with cells with low steady-state cumulative features for conventional progressive dust accumulation and blockage; cells with strong transient mutation features are matched with cells with any steady-state cumulative features for soft agglomeration and impact blockage; cells with any transient mutation features are matched with cells with high steady-state cumulative features for hard agglomeration and compaction blockage; and the remaining cells are matched with other features. Cell matching is performed for regular progressive dust accumulation blockage; the comprehensive value of the instantaneous mutation feature group is compared with the threshold for classifying the strength level of the instantaneous mutation feature to determine the level of the instantaneous mutation feature; the comprehensive value of the steady-state cumulative feature group is compared with the threshold for classifying the high and low levels of the steady-state cumulative feature to determine the level of the steady-state cumulative feature; based on the instantaneous mutation feature level and the steady-state cumulative feature level, the corresponding cell of the two-dimensional feature coupling judgment matrix is ​​located, the preset matching relationship of the cell is retrieved, and the blockage type corresponding to the current fluid controlled element is output; the blockage condition judgment results of all fluid controlled elements after secondary verification are summarized to generate a blockage condition judgment result table.

[0116] Furthermore, in this embodiment, the feature coupling judgment matrix organically combines instantaneous mutation features with steady-state accumulation features, making full use of the complementarity of four-dimensional feature parameters to effectively solve the problem of inaccurate judgment of a single feature under complex working conditions. For example, under some working conditions, there may be both light soft agglomeration impact and slow dust accumulation at the same time, and a single feature is easy to confuse them. However, the coupling matrix can accurately distinguish the dominant blockage type by comparing the strength of the two types of features. Through the comprehensive judgment of this matrix, the accuracy of the classification of the three blockage types can be greatly improved, providing a reliable type basis for the establishment of dual threshold judgment logic and threshold correction.

[0117] The verification of the preliminary classification results of siltation conditions based on the dual-threshold judgment logic includes:

[0118] S2.4: Obtain four-dimensional characteristic parameters including differential pressure rise slope, differential pressure pulsation amplitude, short-term differential pressure distortion coefficient, and pressure gradient difference in the folded area. Define the weighted sum of differential pressure pulsation amplitude and short-term differential pressure distortion coefficient as the instantaneous change characteristic value, and define the weighted sum of differential pressure rise slope and pressure gradient difference in the folded area as the steady-state cumulative characteristic value.

[0119] Further, the specific process of S2.4 includes: retrieving the four-dimensional characteristic parameters under the current real-time operating conditions. These four-dimensional characteristic parameters include the differential pressure rise slope, differential pressure pulsation amplitude, short-time differential pressure distortion coefficient, and pressure gradient difference in the folded area. The differential pressure rise slope is in Pa / s, the differential pressure pulsation amplitude is in Pa, the short-time differential pressure distortion coefficient ranges from 0 to 1, and the pressure gradient difference in the folded area is in Pa / mm. The completeness and validity of the four-dimensional characteristic parameters are verified using a data rationality judgment algorithm. The algorithm parameters include the reasonable range of the differential pressure rise slope. The parameters are defined as follows: pressure pulsation amplitude within a range of 0.1 Pa / s to 3.0 Pa / s; reasonable range for differential pressure pulsation amplitude of 100 Pa to 2000 Pa; reasonable range for short-term differential pressure distortion coefficient of 0 to 1; and reasonable range for pressure gradient difference in folded regions of 2 Pa / mm to 40 Pa / mm. Abnormal parameters exceeding these reasonable ranges are removed, and valid four-dimensional feature parameters are output. Pre-set weighting coefficients for transient change features are retrieved. These coefficients include a weighting coefficient for differential pressure pulsation amplitude and a weighting coefficient for short-term differential pressure distortion coefficient. These weighting coefficients are obtained through training and optimization based on historical operating data. The amplitude shift weighting coefficient is uniformly set to 0.6, and the short-time differential pressure distortion coefficient weighting coefficient is uniformly set to 0.4. The differential pressure pulsation amplitude in the effective four-dimensional feature parameters is multiplied by the differential pressure pulsation amplitude weighting coefficient to output the weighted pulsation component. The short-time differential pressure distortion coefficient in the effective four-dimensional feature parameters is multiplied by the short-time differential pressure distortion coefficient weighting coefficient to output the weighted distortion component. The weighted pulsation component is added to the weighted distortion component to output the instantaneous change feature value. The preset steady-state cumulative feature weighting coefficient is retrieved, which includes the differential pressure rise slope weighting coefficient. The weighting coefficients for the pressure gradient difference in the folded region are obtained by training and optimization based on historical working condition data. The weighting coefficients for the pressure gradient rise slope and the pressure gradient difference in the folded region are both uniformly set to 0.5. The pressure gradient rise slope in the effective four-dimensional feature parameters is multiplied by the pressure gradient rise slope weighting coefficient to output the weighted slope component. The pressure gradient difference in the folded region in the effective four-dimensional feature parameters is multiplied by the pressure gradient difference weighting coefficient to output the weighted gradient component. The weighted slope component and the weighted gradient component are added together to output the steady-state cumulative feature value.

[0120] Furthermore, in this embodiment, the differential pressure pulsation amplitude and the short-term differential pressure distortion coefficient are both instantaneous impact-sensitive features. The instantaneous abrupt change feature value obtained after weighted fusion can more comprehensively quantify the impact intensity and avoid the one-sidedness of a single feature. The differential pressure rise slope reflects the blockage accumulation rate, and the pressure gradient difference in the folded interval reflects the uniformity of blockage distribution. The steady-state accumulation feature value after weighted fusion can comprehensively assess the overall severity of blockage accumulation. The two core feature values ​​focus on the two core risks of instantaneous impact and steady-state accumulation, respectively, providing a simple and effective judgment index for the dual-threshold judgment logic, and improving the efficiency and accuracy of the judgment.

[0121] S2.5: Based on historical calibration data, preset instantaneous change threshold and steady-state accumulation threshold, and simultaneously set a first timing window and a second timing window; the length of the first timing window is less than the length of the second timing window;

[0122] Further, the specific process of S2.5 includes: First, retrieving the historical calibration database of the industrial flue gas purification system. This database has a uniform storage time span of 24 months and contains calibration data for instantaneous mutation characteristic values ​​and steady-state cumulative characteristic values ​​corresponding to soft agglomeration impact blockage and hard agglomeration compaction blockage. The calibration data is integrated from multiple sources, including manual on-site calibration, equipment fault records, and fluid-controlled component failure analysis reports, covering various blockage cases under different operating conditions, loads, and dust characteristics. Second, extracting the instantaneous mutation characteristic value calibration data corresponding to soft agglomeration impact blockage from the historical calibration database. The instantaneous mutation characteristic value is obtained by adding the weighted component of the differential pressure pulsation amplitude and the weighted component of the short-time differential pressure distortion coefficient. Third, performing critical value statistical analysis on the instantaneous mutation characteristic value calibration data. The confidence parameter for the critical value statistics is uniformly set to 95%, and the critical value corresponding to the instantaneous mutation characteristic value when soft agglomeration impact blockage occurs is set to 0.75. This critical value is preset as the instantaneous mutation threshold. The steady-state cumulative characteristic value calibration data corresponding to hard agglomeration compaction blockage is extracted from the historical calibration database. The steady-state cumulative characteristic value is obtained by adding the weighted component of the pressure difference rise slope and the weighted component of the pressure gradient difference in the folded area. Critical value statistical analysis is performed on the steady-state cumulative characteristic value calibration data, and the critical value corresponding to the steady-state cumulative characteristic value when hard agglomeration compaction blockage occurs is set to 0.70. This critical value is preset as the steady-state cumulative threshold. A first timing window is set to monitor instantaneous change characteristics. The window length parameter is uniformly set to 5 seconds and the sampling frequency parameter is uniformly set to 10Hz. A second timing window is set to monitor steady-state cumulative characteristics. The window length parameter is uniformly set to 60 seconds and the sampling frequency parameter is uniformly set to 1Hz. The length relationship between the first timing window and the second timing window is verified to confirm that the length of the first timing window is less than the length of the second timing window. The preset instantaneous change threshold, steady-state cumulative threshold, first timing window, and second timing window parameters are output.

[0123] Furthermore, in this embodiment, both the instantaneous mutation threshold and the steady-state accumulation threshold are calibrated based on a large number of real historical operating conditions, which fits the actual operating scenario and avoids the disconnect between theoretical thresholds and actual operating conditions. The short-duration first timing window can capture the instantaneous impact of soft agglomeration at the millisecond or second level in a timely manner, avoiding missed impact detection. The long-duration second timing window can filter out the small fluctuations in steady-state accumulation characteristic values ​​caused by short-term operating condition fluctuations, and only identify continuous and real blockage accumulation trends, avoiding false alarms. The combination of dual thresholds and dual timing windows forms a precise monitoring system that adapts to different blockage characteristics.

[0124] S2.6: Within the first timing window, the difference between the peak and valley values ​​of the real-time differential pressure is calculated as the instantaneous fluctuation amplitude. When the instantaneous fluctuation amplitude is greater than the instantaneous sudden change threshold, an instantaneous warning flag is triggered.

[0125] Further, the specific process of S2.6 includes: starting a first timing window with a length parameter of 5 seconds and a sampling frequency parameter of 10Hz; continuously collecting real-time differential pressure data during the timing of the first timing window, with the unit of real-time differential pressure data being Pascals, forming a real-time differential pressure sequence within the window, with a total of 50 data points in the real-time differential pressure sequence; traversing the real-time differential pressure sequence within the window, filtering out the largest value in the sequence, defining it as the differential pressure peak, and filtering out the smallest value in the real-time differential pressure sequence, defining it as the differential pressure trough; calculating the difference between the differential pressure peak and the differential pressure trough. The first step is to obtain the instantaneous fluctuation amplitude, with the unit being Pascals. The second step is to retrieve the preset instantaneous change threshold, which has a value of 0.75. The instantaneous fluctuation amplitude is then compared with the instantaneous change threshold. If the instantaneous fluctuation amplitude is less than or equal to the instantaneous change threshold, the instantaneous warning flag is not triggered. If the instantaneous fluctuation amplitude is greater than the instantaneous change threshold, the instantaneous warning flag is triggered, the warning flag status is updated to valid, and the warning trigger time, instantaneous fluctuation amplitude, and real-time differential pressure sequence data within the window are recorded to generate instantaneous warning record information.

[0126] Furthermore, in this embodiment, the short duration of the first timing window ensures that the instantaneous fluctuation amplitude can quickly reflect the latest impact situation, avoiding warning delays; the instantaneous fluctuation amplitude is directly calculated based on real-time differential pressure, the data source is direct and reliable, and it can truly reflect the actual differential pressure fluctuation state of the fluid-controlled component; the warning is triggered by comparing with the instantaneous change threshold, the logic is simple and direct, the response speed is fast, and it can issue a warning at the first moment when soft agglomeration impact occurs, effectively preventing the risk of damage to the fluid-controlled component caused by soft agglomeration impact blockage.

[0127] S2.7: Within the second timing window, the cumulative offset of the real-time differential pressure is calculated using an exponentially weighted moving average method. When the cumulative offset exceeds the steady-state cumulative threshold, a steady-state warning flag is triggered.

[0128] Further, the specific process of S2.7 includes: starting a second timing window with a length parameter of 60 seconds and a sampling frequency parameter of 1Hz; continuously collecting real-time differential pressure data during the timing of the second timing window, with the unit of real-time differential pressure data being Pascals, forming a real-time differential pressure sequence within the window, with a total of 60 data points in the sequence; retrieving the parameters of the exponentially weighted moving average method, uniformly setting the smoothing factor parameter to 0.3 and the initial reference differential pressure to 150Pa; weighting the real-time differential pressure sequence within the window, assigning a weight of 30% to the data in the first 30 seconds of the window and a weight of 70% to the data in the last 30 seconds of the window, and outputting the result. The weighted average differential pressure sequence is calculated. The difference between the last data point in the weighted average differential pressure sequence and the initial reference differential pressure is calculated to obtain the cumulative offset, which is in Pa. The preset steady-state cumulative threshold is retrieved, with a value of 0.70. The cumulative offset is compared with the steady-state cumulative threshold. If the cumulative offset is less than or equal to the steady-state cumulative threshold, the steady-state warning flag is not triggered. If the cumulative offset is greater than the steady-state cumulative threshold, the steady-state warning flag is triggered, the warning flag status is updated to valid, and the warning trigger time, cumulative offset, real-time differential pressure sequence data within the window, smoothing factor parameters, and initial reference differential pressure parameters are recorded to generate steady-state warning record information.

[0129] Furthermore, in this embodiment, the application of the long-duration characteristic of the second timing window and the exponentially weighted moving average method can effectively filter out short-term pressure fluctuations caused by factors such as flue gas flow fluctuations, load adjustments, and short-term humidity changes, focusing only on the long-term, continuous upward trend of pressure difference caused by ash accumulation and compaction of the fluid-controlled components; the cumulative offset can quantitatively characterize the cumulative increase in pressure difference, with stable values ​​and obvious trends, avoiding the randomness of pressure difference values ​​at a single moment; by comparing with the steady-state cumulative threshold to trigger an early warning, the early trend of hard agglomeration compaction and blockage can be accurately identified, providing early warning.

[0130] S2.8: When the instantaneous warning flag is triggered, the working condition is confirmed as soft agglomeration impact blockage; when the steady-state warning flag is triggered, the working condition is confirmed as hard agglomeration compaction blockage; when neither flag is triggered, the working condition is confirmed as conventional progressive ash accumulation blockage; output the blockage condition judgment result after secondary verification.

[0131] Further, the specific process of S2.8 includes: real-time reading of the status of instantaneous warning flags and steady-state warning flags, where the status of instantaneous warning flags is either valid or invalid, and the status of steady-state warning flags is either valid or invalid; pre-setting blockage type determination logic rules, which include four trigger combinations: the first combination is instantaneous warning flag valid and steady-state warning flag invalid; the second combination is instantaneous warning flag invalid and steady-state warning flag valid; the third combination is instantaneous warning flag valid and steady-state warning flag valid; and the fourth combination is instantaneous warning flag invalid and steady-state warning flag invalid; matching the current flag status with the pre-set determination logic rules, and performing the corresponding blockage type determination; when the first combination matches successfully, determining the current fluid-controlled element. The blockage type is soft agglomeration impact blockage; when the second combination match is successful, the current fluid controlled element blockage type is determined to be hard agglomeration compaction blockage; when the third combination match is successful, the current fluid controlled element blockage type is preferentially determined to be hard agglomeration compaction blockage, and the risk of soft agglomeration impact superposition is marked simultaneously; when the fourth combination match is successful, the current fluid controlled element blockage type is determined to be conventional progressive ash accumulation blockage; read the current time, fluid controlled element identifier, blockage type, instantaneous warning flag status, steady-state warning flag status, instantaneous fluctuation amplitude, cumulative offset, instantaneous mutation threshold, and steady-state cumulative threshold information; integrate the above information to generate a blockage condition judgment result that has been verified twice, and output the blockage condition judgment result that has been verified twice.

[0132] The logic for dynamically correcting the dual threshold determination based on the moisture content of the medium and the flue gas temperature includes:

[0133] S2.9: Obtain the moisture content of the medium and the flue gas temperature at the current sampling time;

[0134] Furthermore, the specific process of obtaining the moisture content of the medium and the flue gas temperature at the current sampling moment includes: During each data sampling cycle, the system collects real-time moisture content data of the medium entering the fluid-controlled element using a high-precision moisture content sensor installed on the medium delivery pipeline. The sensor outputs an analog signal in real-time, which is converted into a digital signal by a signal conversion module and then transmitted to the system data processing unit. Simultaneously, the system collects real-time flue gas temperature data entering the filtration system using a high-precision temperature sensor installed in the main flue. The temperature signal output by the temperature sensor is filtered, amplified, and then synchronously transmitted to the system data processing unit. The system processes the received moisture content data and flue gas temperature data... The system performs validity verification, checking whether the data is within the normal measurement range of the sensor, whether there are any sudden anomalies, and whether there is any missing data. If the data is normal and valid, the system records the medium moisture content and flue gas temperature values ​​corresponding to the current sampling time to ensure that the data accurately corresponds to the sampling time. If the data is abnormal, the system automatically removes the abnormal data, retrieves the valid data from the previous sampling time as a temporary replacement, and generates an abnormal data record for subsequent sensor status monitoring and fault diagnosis. After completing data verification and recording, the system outputs the medium moisture content and flue gas temperature data at the current sampling time for use in the subsequent threshold correction step, ensuring that the threshold correction is based on real-time and accurate operating condition data.

[0135] Furthermore, in this embodiment, the moisture content of the medium and the flue gas temperature are the core operating parameters affecting the formation, viscosity, adhesion strength, and pressure difference variation of agglomerated pollutants. The higher the moisture content, the easier it is for dust to form sticky soft agglomerates, and the more intense the impact. The lower the flue gas temperature, the stronger the stickiness of the agglomerates, and the easier they are to adhere and compact. Real-time and accurate acquisition of these two operating condition data is a prerequisite for dynamically correcting the dual thresholds and ensuring that the thresholds are adapted to the actual operating conditions. The high precision of the sensors and the data verification mechanism ensure the reliability of the operating condition data and avoid threshold correction deviations due to data errors.

[0136] S2.10: Based on the moisture content of the medium, query the preset moisture content-instantaneous correction coefficient mapping table and the preset moisture content-steady-state correction coefficient mapping table, and output the first instantaneous correction coefficient and the first steady-state correction coefficient;

[0137] Furthermore, the specific process of S2.10 includes: constructing a moisture content-instantaneous correction coefficient mapping table and a moisture content-steady-state correction coefficient mapping table. Both mapping tables are constructed based on historical operating condition test data and field operation experience of the industrial flue gas purification system. The time span of the historical operating condition test data is uniformly set to 24 months, covering the entire operating condition range of extremely low humidity, low humidity, medium humidity, high humidity, and extremely high humidity. The field operation experience comes from the threshold correction records of fluid-controlled components under different medium moisture content conditions; dividing the medium moisture content range, with the interval division accuracy uniformly set to 5%, and successively setting the medium moisture content range to less than 10%, The moisture content ranges are 10% to 15%, 15% to 20%, 20% to 25%, 25% to 30%, and greater than 30%. A unique instantaneous correction factor is assigned to each moisture content range, with a uniform value of 1.0 to 1.5. The instantaneous correction factor is 1.0 for the range less than 10%, 1.1 for 10% to 15%, 1.2 for 15% to 20%, 1.3 for 20% to 25%, 1.4 for 25% to 30%, and [missing value] for greater than 30%. 1.5 Generate a moisture content-instantaneous correction coefficient mapping table; assign a unique steady-state correction coefficient to each medium moisture content range. The steady-state correction coefficient is uniformly set to a range of 1.0 to 1.4: 1.0 for moisture content less than 10%, 1.05 for 10% to 15%, 1.1 for 15% to 20%, 1.2 for 20% to 25%, 1.3 for 25% to 30%, and 1.4 for moisture content greater than 30%. The table is generated; the current medium moisture content value is retrieved, and the medium moisture content value is in percentage; the current medium moisture content value is matched with the medium moisture content range in the moisture content-instantaneous correction coefficient mapping table to determine the range to which the current medium moisture content belongs, and the instantaneous correction coefficient corresponding to the range is extracted and defined as the first instantaneous correction coefficient; the current medium moisture content value is matched with the medium moisture content range in the moisture content-steady-state correction coefficient mapping table to determine the range to which the current medium moisture content belongs, and the steady-state correction coefficient corresponding to the range is extracted and defined as the first steady-state correction coefficient; the first instantaneous correction coefficient and the first steady-state correction coefficient are output.

[0138] Furthermore, in this embodiment, the moisture content of the medium is a key factor affecting the impact intensity of soft agglomerates and the compaction rate of hard agglomerates: as the moisture content increases, the soft agglomerate clumps become more viscous and the impact becomes more intense, requiring an appropriate increase in the instantaneous change threshold to avoid frequent false alarms; at the same time, agglomerates are easier to compact in high-humidity environments, requiring an appropriate increase in the steady-state cumulative threshold to adapt to changes in compaction rate; the moisture content-correction coefficient mapping table is constructed based on actual working conditions and can accurately quantify the impact of moisture content on the dual thresholds. The correction coefficient can be quickly obtained by querying the mapping table, which is efficient and accurate, providing reliable moisture content correction parameters for dynamic threshold correction.

[0139] S2.11: Based on the flue gas temperature at the current sampling time, query the preset temperature-instantaneous correction coefficient mapping table and the preset temperature-steady-state correction coefficient mapping table, and output the second instantaneous correction coefficient and the second steady-state correction coefficient;

[0140] Furthermore, the specific process of S2.11 includes: constructing a temperature-instantaneous correction coefficient mapping table and a temperature-steady-state correction coefficient mapping table. Both mapping tables are constructed based on historical operating condition test data and field operation experience of the industrial flue gas purification system. The time span of the historical operating condition test data is uniformly set to 24 months, covering the entire operating condition range of low temperature, medium-low temperature, medium temperature, medium-high temperature, and high temperature. The field operation experience comes from the threshold correction records of fluid controlled components under different flue gas temperature conditions; dividing the flue gas temperature range, with the interval division accuracy uniformly set to 20℃, and sequentially setting the flue gas temperature range. The temperature ranges are defined as follows: below 120℃, 120℃ to 140℃, 140℃ to 160℃, 160℃ to 180℃, and above 180℃. A unique instantaneous correction coefficient is assigned to each flue gas temperature range, with a uniform value range of 1.0 to 1.5. Specifically, the instantaneous correction coefficient is 1.5 for the range below 120℃, 1.3 for 120℃ to 140℃, 1.2 for 140℃ to 160℃, 1.1 for 160℃ to 180℃, and 1.1 for the range above 180℃. A positive coefficient of 1.0 is used to generate a temperature-instantaneous correction coefficient mapping table. A unique steady-state correction coefficient is assigned to each flue gas temperature range, with values ​​ranging from 1.0 to 1.4. The steady-state correction coefficient is 1.4 for temperatures below 120℃, 1.25 for 120℃ to 140℃, 1.15 for 140℃ to 160℃, 1.05 for 160℃ to 180℃, and 1.0 for temperatures above 180℃. This generates a temperature-steady-state correction coefficient mapping table. The table retrieves the current flue gas temperature value, in °C; it matches the current flue gas temperature value with the flue gas temperature range in the temperature-instantaneous correction coefficient mapping table to determine the range to which the current flue gas temperature belongs, extracts the instantaneous correction coefficient corresponding to that range, and defines it as the second instantaneous correction coefficient; it matches the current flue gas temperature value with the flue gas temperature range in the temperature-steady-state correction coefficient mapping table to determine the range to which the current flue gas temperature belongs, extracts the steady-state correction coefficient corresponding to that range, and defines it as the second steady-state correction coefficient; finally, it outputs the second instantaneous correction coefficient and the second steady-state correction coefficient.

[0141] Furthermore, in this embodiment, flue gas temperature directly affects the physical properties of agglomerated pollutants: as temperature decreases, the viscosity of soft agglomerates increases significantly, resulting in greater impact and destructive force. Simultaneously, the agglomerates adhere more firmly to the surface of the fluid-controlled element, accelerating the compaction speed. Conversely, as temperature increases, the viscosity of agglomerates decreases, the impact effect diminishes, and the compaction difficulty increases. The temperature-correction coefficient mapping table accurately quantifies the impact of temperature changes on the dual thresholds. At low temperatures, the dual thresholds are appropriately increased to avoid false alarms caused by low temperatures. At high temperatures, the dual thresholds are appropriately decreased to promptly capture the risks of weak impacts and slow compaction. The temperature correction coefficient is quickly obtained by querying the mapping table, complementing the moisture content correction coefficient and comprehensively covering the impact of core operating parameters on the thresholds.

[0142] S2.12: Multiply the first instantaneous correction coefficient by the second instantaneous correction coefficient to obtain the instantaneous comprehensive correction factor; multiply the first steady-state correction coefficient by the second steady-state correction coefficient to obtain the steady-state comprehensive correction factor.

[0143] S2.13: Obtain the preset instantaneous mutation threshold and steady-state accumulation threshold, multiply the instantaneous mutation threshold by the instantaneous comprehensive correction factor to obtain the corrected instantaneous mutation threshold, and multiply the steady-state accumulation threshold by the steady-state comprehensive correction factor to obtain the corrected steady-state accumulation threshold.

[0144] Furthermore, in this embodiment, the initial dual thresholds are only applicable to standard operating conditions. In actual operation, operating conditions fluctuate frequently, and fixed thresholds are prone to misjudgment and missed judgment. By dynamically correcting the dual thresholds through real-time operating parameters, the thresholds can be adjusted synchronously with changes in humidity and temperature, always adapting to the current actual operating conditions. The corrected instantaneous change threshold can accurately match the severity of soft agglomeration impact under the current operating conditions, avoiding frequent false alarms caused by strong impact and low threshold under high humidity and low temperature, or missed alarms caused by weak impact and high threshold under low humidity and high temperature. The corrected steady-state cumulative threshold can accurately match the rate of hard agglomeration compaction under the current operating conditions, avoiding early alarms caused by fast compaction and low threshold, or late alarms caused by slow compaction and high threshold. The dynamically corrected dual thresholds significantly improve the accuracy and reliability of blockage determination under different operating conditions.

[0145] S2.14: Replace the instantaneous mutation threshold with the corrected instantaneous mutation threshold, replace the steady-state cumulative threshold with the corrected steady-state cumulative threshold, and output the updated dual threshold judgment logic.

[0146] Furthermore, in this embodiment, the updated judgment logic fully considers the combined effects of humidity and temperature on agglomerated pollutants and blockage of fluid-controlled components, and can adapt to various complex operating conditions from mild to extreme. Based on the subsequent curvature fitting, critical point prediction, risk ranking and other operations based on the updated judgment logic, it can more accurately capture the state changes of fluid-controlled components, make the optimal fluid-controlled component switching decision, effectively improve the operational stability, safety and economy of the industrial flue gas purification system, and reduce fluid-controlled component losses and operating energy consumption.

[0147] The step of fitting the differential pressure time-series curves generated by the real-time differential pressure for each fluid-controlled element based on the modified dual-threshold judgment logic includes:

[0148] S3.1: Extract the corrected instantaneous change threshold and the corrected steady-state cumulative threshold from the updated dual threshold judgment logic. At the same time, obtain the pressure difference time series curve generated by the real-time pressure difference of each fluid controlled element. Use the corrected steady-state cumulative threshold as an adaptive adjustment factor to adjust the sliding window step size in reverse and output the adjusted sliding window step size. The steady-state cumulative threshold and the window step size are inversely proportional.

[0149] Further, the specific steps of S3.1 include: retrieving the dynamically updated dual-threshold judgment logic, which has undergone dual correction for medium moisture content and flue gas temperature, and internally stores two types of core threshold parameters: the corrected instantaneous change threshold and the corrected steady-state accumulation threshold. The corrected instantaneous change threshold is used to identify the instantaneous impact of soft agglomerates, and the corrected steady-state accumulation threshold is used to measure the degree of compaction and accumulation of hard agglomerates. Both thresholds are dimensionless values, uniformly set to a range of 0.5 to 1.5; extracting the corrected instantaneous change threshold and the corrected steady-state accumulation threshold from the updated dual-threshold judgment logic. The extraction process uses parameter... The data reading command is executed, with a uniform reading precision set to four decimal places to ensure accurate value reading. After extraction, the specific values ​​of the two thresholds are recorded, with the corrected steady-state cumulative threshold used as an adaptive adjustment factor. The real-time differential pressure data acquisition unit for each branch fluid-controlled element is activated. This unit continuously acquires the inlet and outlet differential pressure data of each fluid-controlled element at a preset sampling frequency of 10Hz, covering the current complete operating cycle of the fluid-controlled element to ensure data continuity. The acquired real-time differential pressure data is sorted and organized chronologically to generate a differential pressure time-series curve for each fluid-controlled element. The differential pressure time-series curve, with time on the horizontal axis and real-time differential pressure on the vertical axis, comprehensively reflects the continuous process of differential pressure change over time in the fluid-controlled component. The number of data points on the curve is consistent with the sampling duration and sampling frequency. A preset baseline parameter and adjustment range for the sliding window step size are established. The sliding window step size represents the data span selected for a single curvature fitting. The baseline step size is uniformly set to 5 seconds, and the step size adjustment range is uniformly set to 1 to 10 seconds to ensure a reasonable and effective adjustment range. An inverse mapping relationship is established between the steady-state cumulative threshold and the sliding window step size. This inverse relationship follows a linear inverse proportional algorithm, the core logic of which is that the larger the steady-state cumulative threshold, the smaller the sliding window step size, and the smaller the steady-state cumulative threshold. The smaller the sliding window step size, the larger the calculation precision is set to four decimal places to ensure the accuracy of the mapping calculation. The extracted and corrected steady-state cumulative threshold is substituted into the linear inverse proportional algorithm for calculation. First, the algorithm benchmark correspondence is set. When the steady-state cumulative threshold is 1.0, the sliding window step size is equal to the benchmark step size of 5 seconds. Linear inverse proportional conversion is performed based on this benchmark. The sliding window step size is adjusted in reverse according to the conversion rules. When the corrected steady-state cumulative threshold is greater than 1.0, it indicates that the risk of compaction accumulation of the fluid controlled element is high, and the window step size needs to be reduced to accurately capture pressure difference changes. After adjustment, the step size is less than five seconds. When the corrected steady-state cumulative threshold is less than 1...At a value of 0, the risk of compaction accumulation in the fluid-controlled element is low, allowing for an increase in the window step size to improve computational efficiency. The adjusted step size is greater than 5 seconds. Boundary checks are performed on the adjusted sliding window step size to ensure it is no less than the minimum step size of 1 second and no greater than the maximum step size of 10 seconds. If the step size exceeds the boundary, it is automatically locked to the corresponding boundary value to avoid invalid step sizes. The adaptively adjusted sliding window step size is output, precisely matching the current risk of compaction accumulation in the fluid-controlled element, and can be directly used for pressure differential time-series curve fitting calculations.

[0150] S3.2: According to the output sliding window step size, take three consecutive points on the pressure difference time series curve window by window, calculate the ratio of the central angle to the arc length based on the three-point common circle method, and use it as the curvature value of the center point of each window, and output the initial curvature sequence.

[0151] Furthermore, the specific steps of S3.2 include: retrieving the output adaptive sliding window step size; retrieving the generated and stored differential pressure time series curve of the fluid controlled element; determining the starting position for window-by-window truncation, defaulting to the first valid data point of the differential pressure time series curve, with subsequent windows shifting sequentially according to the sliding window step size, ensuring no overlap or gap between windows during the shifting process to guarantee continuous and complete data points; selecting all continuous data points within the current window on the differential pressure time series curve according to the time span corresponding to the sliding window step size, with the number of data points included in each window depending on the step size and... The sampling interval is determined; for example, with a step size of 5 seconds, each window contains 50 consecutive differential pressure data points, and all windows maintain the same number of data points. Within each window, three consecutive differential pressure data points are selected at the beginning, middle, and end positions of the window, strictly following chronological order. The three data points are arranged sequentially, with a time interval of 0.1 seconds between adjacent data points to ensure temporal continuity and uniform distribution. The basic parameters of the three-point concircle method are determined; this method is used to calculate the local curvature of the curve, and the basic parameters include a uniform time interval of 0.1 seconds between data points. Three consecutive differential pressure data points are used to determine the geometric relationship of the plane containing these three points. If the three points are not on a straight line, a circle is uniquely determined; if they are on a straight line, the central angle is zero by default, and the arc length is the straight-line distance between the three points. The central angle corresponding to the circle determined by the three points is calculated. The central angle is the included angle between the centers of the arcs, calculated using a geometric angle calculation method. The arc length between the three points is calculated using the ratio of the circumference to the central angle. The calculated central angle value is divided by the arc length value to obtain the curvature value corresponding to the center point of the current window. The curvature value is a dimensionless value. It directly reflects the curvature of the pressure difference curve at the current window position; the curvature value of the current window center point is arranged according to the order of the windows on the time series curve, and the curvature value corresponding to all windows is calculated in sequence to form a continuous and ordered curvature data set; the generated curvature data set is time-series organized to ensure that each curvature value corresponds one-to-one with the timestamp of the corresponding window center point, with a timestamp accuracy of 0.1 seconds, and finally outputs a complete and ordered initial curvature sequence. The formulas for calculating geometric angles and arc lengths are existing technologies in this field and are not the inventive solutions of this application, and will not be elaborated here.

[0152] S3.3: The corrected instantaneous mutation threshold is used as the outlier suppression factor to filter the initial curvature sequence. Specifically, the original pressure difference corresponding to each curvature value is traversed to three consecutive points, and the vertical distance between the middle point and the line connecting the two points is calculated. If the vertical distance is greater than the outlier suppression factor, the mean of the curvature of the two points is used to replace the current curvature value, and the smooth curvature sequence after outlier suppression is output.

[0153] Further, the specific steps of S3.3 include: retrieving the dynamically corrected transient mutation threshold, and directly defining the corrected transient mutation threshold as an outlier suppression factor to identify abnormal outliers in the initial curvature sequence; retrieving the generated initial curvature sequence, which is arranged in chronological order, with each curvature value corresponding to a window center point on the differential pressure time series curve; retrieving the original differential pressure time series data corresponding to the generation of the initial curvature sequence; establishing the correspondence between the initial curvature sequence and the original differential pressure data, where each curvature value in the initial curvature sequence corresponds to a set of three consecutive data points on the original differential pressure time series curve, with the three data points being the previous one of the current window center points. After establishing the correlation between the pressure difference data at different times, the current pressure difference data, and the pressure difference data at the next time, it ensures that each curvature value can be accurately traced back to its corresponding three original pressure difference data points. Starting from the first curvature value in the initial curvature sequence, each curvature value in the initial curvature sequence is traversed sequentially. For the currently traversed curvature value, its corresponding three original continuous pressure difference data points are retrieved, defined as the previous point, midpoint, and next point, respectively, with a time interval of 0.1 seconds between the three data points. Based on the pressure difference values ​​and time positions of the previous and next points, a line segment is determined between the two points. The perpendicular distance from the midpoint to this line segment is calculated using a perpendicular distance algorithm, and the algorithm parameters include... A time interval of 0.1 seconds and a pressure differential accuracy to one decimal place are used to ensure accurate and reliable calculation results. The calculated vertical distance value is compared with the outlier suppression factor, which is the corrected instantaneous mutation threshold, ranging from 0.5 to 1.5. The comparison process strictly follows the numerical relationship. When the vertical distance is greater than the outlier suppression factor, the current curvature value is determined to be an outlier and needs correction. In this case, the previous and next curvature values ​​are retrieved, and the arithmetic mean of the two curvature values ​​is calculated. This average is used to replace the current outlier curvature value. When the vertical distance is less than or equal to the outlier suppression factor, the current curvature value is determined to be a normal and valid value. If correction is needed, the current curvature value is retained unchanged. The process involves sequentially traversing, comparing, and correcting all curvature values ​​in the initial curvature sequence, maintaining the temporal order of the sequence during traversal to ensure the continuity of the filtered sequence. The curvature data after all corrections are rearranged according to the original temporal order to form a new curvature sequence, which is the smoothed curvature sequence after outlier suppression. The smoothed curvature sequence is then validated for integrity, checking if the sequence length matches the initial curvature sequence to ensure no data loss or corruption. Once the validation passes, the smoothed curvature sequence is output. The calculation formula for the vertical distance algorithm is existing technology in this field and is not an inventive solution of this application; therefore, it will not be elaborated upon here.

[0154] The critical point of predicted agglomeration, compaction, solidification, and backflushing failure triggers a component switching prediction signal, outputting the identifier of the fluid component to be scheduled, including:

[0155] S3.4: Obtain the smooth curvature sequence, input the smooth curvature sequence into the time series prediction network based on gated recurrent units in time order, and output the curvature time series input vector;

[0156] Furthermore, a time-series prediction network based on gated recurrent units is used to process the temporal characteristics of curvature sequences, adapting to the dynamic changes in the pressure differential curvature of fluid-controlled components in industrial flue gas purification systems. The network consists of three core modules: an input layer, a gated recurrent unit hidden layer, and an output layer. These modules are fully connected to ensure smooth and lossless data transmission. The input layer receives the smooth curvature sequence and performs data preprocessing. It has one neuron, matching the single data dimension of the smooth curvature sequence, and receives continuous temporal data. The hidden layer captures the temporal correlation of the smooth curvature sequence. It has two gated recurrent unit neurons, each with 64 state dimensions. The activation function is the hyperbolic tangent function, with default parameters that require no adjustment. The forget gate threshold is set to 0.5 to control the degree of forgetting historical curvature data. When the forget gate output value is greater than 0.5, historical data features are retained; when it is less than or equal to 0.5, historical data features are retained. When the threshold is equal to 0.5, some historical data is forgotten to avoid interference from redundant information; the input gate threshold is set to 0.6 to control the input weight of the current curvature data. When it is greater than 0.6, the influence of the current data is enhanced, and when it is less than or equal to 0.6, the influence of the current data is weakened; the output gate threshold is set to 0.5 to control the output ratio of the hidden layer state to ensure the rationality of the output data; the gradient clipping threshold of the hidden layer is set to 1.0 to prevent gradient explosion during network training; the core function of the output layer is to output the curvature temporal input vector. The number of neurons in the output layer is set to 64, which is consistent with the state dimension of the neurons in the hidden layer. The output layer uses a linear activation function with no additional parameters. The output vector has 64 dimensions, each corresponding to a temporal feature parameter. The output rate is consistent with the input layer, which is ten data points per second to ensure temporal synchronization between input and output. The hyperbolic tangent function and the linear activation function are existing technologies in this field and are not the inventive solutions of this application, so they will not be described in detail here.

[0157] Furthermore, the training data for the time-series prediction network based on gated recurrent units uses historical smooth curvature sequence data from industrial flue gas purification systems. The data time span is set to 24 months, covering curvature changes under different operating conditions and blockage types. The total training data is set to 100,000 sets, with 80% used as the training set and 20% as the test set. The training batch size is set to 32 sets, the number of training iterations is set to 500, the learning rate is set to 0.01, and the learning rate decay coefficient is set to 0.95, decaying once every 50 iterations. The loss function is the mean squared error function. During training, training is stopped when the loss value of the test set is less than 0.001 to ensure that the network training accuracy meets the standard. After training, the network parameters are saved in the standard model parameter format.

[0158] Furthermore, in this embodiment, the processed smooth curvature sequence is input point by point into the input layer of the temporal prediction network in chronological order. During the input process, the data input rate is kept stable, consistent with the rate of 10 data points per second set for the input layer. The input layer normalizes each input curvature value and then transmits the data to the hidden layer of the gated recurrent unit. The hidden layer of the gated recurrent unit receives the normalized curvature data transmitted from the input layer and processes it according to preset gating rules. The forget gate, input gate, and output gate work together to filter and fuse historical and current curvature data, capturing the temporal correlation features of the curvature sequence. Each neuron of the gated recurrent unit performs computational processing on the input data and outputs the hidden layer state value. The computational process of the hidden layer strictly follows the core logic of the gated recurrent unit, ensuring the temporal correlation features are accurately reflected. The accuracy of feature capture is ensured. The hidden layer transmits the processed temporal feature data to the output layer. The output layer transforms the hidden layer state values ​​through a linear activation function, generating a 64-dimensional curvature temporal input vector. The value of each dimension corresponds to a temporal feature of the smooth curvature sequence. The generation process of the output vector ensures that the temporal order is consistent with the input smooth curvature sequence. The completeness and accuracy of the output curvature temporal input vector are verified. The verification includes whether the vector dimension is 64, whether the numerical precision is four decimal places, and whether the temporal order is consistent with the input smooth curvature sequence, ensuring that the output curvature temporal input vector has no deviation. After the verification is passed, a complete and accurate curvature temporal input vector is output, which contains all the temporal features of the smooth curvature sequence and can be directly used to predict the critical point of backflush failure of fluid controlled components.

[0159] S3.5: Through the gated recurrent unit in the time-series prediction network, the smooth curvature values ​​at multiple consecutive sampling times are updated sequentially, and the curvature prediction values ​​at multiple sampling times are output.

[0160] Furthermore, the specific steps of S3.5 include: retrieving the trained and saved temporal prediction network and smooth curvature sequence; setting the initial state parameters of the gated recurrent unit, with the initial hidden state value uniformly set to 0, ensuring that the gated recurrent unit starts state updates from the initial zero state; configuring the core operation parameters of the gated recurrent unit, with the forget gate threshold set to 0.5, the input gate threshold set to 0.6, the output gate threshold set to 0.5, the activation function using the hyperbolic tangent function, and the gradient clipping threshold set to 1.0 to prevent gradient explosion during operation; and prioritizing time-ordered operations. Subsequently, the smooth curvature values ​​of 50 consecutive sampling times are sequentially input into two neurons of the hidden layer of the gated recurrent unit, with the input rate maintained at 10 data points per second. For the first input smooth curvature value, the first state update of the gated recurrent unit is initiated. First, the input smooth curvature value and the initial hidden state are processed through the reset gate. The reset gate receives the input smooth curvature value and the initial hidden state, performs calculations according to a preset threshold of 0.5, and controls the retention ratio of historical hidden state information. Since the initial hidden state is 0, the output value of the reset gate after the calculation is 0.5, which just reaches the threshold and is retained. The base hidden state remains unchanged. An update gate processes the current input smooth curvature value. The update gate receives the input smooth curvature value and the initial hidden state, performs calculations with a threshold of 0.6, and outputs a weight value to control the influence of the current smooth curvature value on the new candidate hidden state. If the updated gate output value is greater than 0.6, it strengthens the weight of the current smooth curvature value. After activation by the hyperbolic tangent function, the current smooth curvature value is fused with the historical hidden state filtered by the reset gate to obtain the candidate hidden state component to be updated. The update gate weight is then added to the historical hidden state and the candidate hidden state component to complete the hidden state update. The first update of the hidden state is performed. Simultaneously, the updated hidden state is processed through an output mapping gate. The output mapping gate receives the input smooth curvature value and the updated hidden state, performs calculations according to a threshold of 0.5, and outputs a weight value. The updated hidden state is activated by the hyperbolic tangent function and multiplied by the weight value output by the output mapping gate to obtain the first updated hidden state value. Using the first updated hidden state as the initial state, the smooth curvature value at the second sampling time is input, and the process is repeated iteratively. The reset gate is based on the current input smooth curvature value and the previously updated hidden state, according to a threshold of 0.A threshold of 5 is used to filter valid information from historical hidden states, update the weights of the current smooth curvature value of the gate control, and output the output ratio of the hidden state mapped by the gate control, thus completing the second update of the hidden state. The smooth curvature values ​​of 50 consecutive sampling times are processed sequentially. Each input of a smooth curvature value at a sampling time completes one state update of the gated recurrent unit. The updated hidden state is used as the initial state for the next sampling time operation, ensuring the continuity and temporal correlation of state updates. The operation strictly follows preset parameters and operational logic to ensure the accuracy of each state update. After completing the state update at each sampling time, the gated recurrent unit transmits the updated hidden state value to the output layer of the temporal prediction network. The output layer uses a linear activation function to... The hidden state values ​​are transformed to generate curvature prediction values ​​for the corresponding sampling moments. State updates and curvature prediction value generation are performed sequentially for 50 consecutive sampling moments. The curvature prediction values ​​for each sampling moment are arranged in chronological order to ensure that the temporal order of the prediction values ​​is completely consistent with the temporal order of the input smooth curvature values, avoiding temporal discrepancies. The integrity of the generated curvature prediction values ​​for multiple sampling moments is verified. This checks whether the number of prediction values ​​matches the number of consecutive sampling moments in the input (50 in total) and whether the precision of each prediction value is four decimal places, ensuring no missing data or insufficient precision. After successful verification, the curvature prediction values ​​for multiple consecutive sampling moments are output. These curvature prediction values ​​reflect the changing trend of the subsequent smooth curvature sequence.

[0161] S3.6: Compare the output curvature prediction value with the preset curvature runaway threshold. When the curvature prediction value at two consecutive sampling times exceeds the curvature runaway threshold, it is determined that the critical point of agglomerated agglomerate compaction solidification and backflushing failure has been reached.

[0162] Furthermore, the specific steps of S3.6 include: retrieving the curvature prediction value corresponding to the previously generated smooth curvature sequence, and simultaneously retrieving the preset curvature runaway threshold, wherein the curvature runaway threshold is set to 0.3; extracting the curvature prediction value at each sampling moment in chronological order, and comparing it one by one with the curvature runaway threshold, the comparison process adopts a point-by-point comparison method, only requiring direct comparison of the curvature prediction value at a single sampling moment with the runaway threshold of 0.3; during the comparison process, strictly recording the curvature prediction value and comparison result at each sampling moment, and simultaneously recording the time node of each sampling moment to ensure that each curvature prediction value corresponds to a specific sampling moment. To avoid timing discrepancies, the time interval between each sampling moment is 0.1 seconds. The focus is on monitoring the curvature prediction values ​​at two consecutive sampling moments. It is determined whether both of these predicted curvature values ​​exceed a preset curvature runaway threshold of 0.3. First, the curvature prediction value at the first sampling moment is compared. If this value is greater than 0.3, the curvature prediction value at the next adjacent sampling moment is compared to confirm if it is also greater than 0.3. If the curvature prediction value at the first sampling moment is greater than 0.3, while the curvature prediction value at the second sampling moment is less than or equal to 0.3, the condition of exceeding the threshold for two consecutive sampling moments is not met, and the critical point is determined not to have been reached. The process continues for the next sampling moment. The curvature prediction values ​​at each sampling time are compared until two consecutive sampling times exceeding the threshold are found. If the curvature prediction values ​​at two consecutive adjacent sampling times both exceed 0.3, and these two sampling times are consecutive and without interval, with a time interval of 0.1 seconds, then the critical point of compaction, solidification, and backflushing failure of agglomerated clumps is determined. During the comparison process, if the curvature prediction value at a single sampling time exceeds the threshold, but the curvature prediction values ​​at adjacent sampling times do not exceed the threshold, only the anomaly at that sampling time is recorded, and it is not determined as a critical point. If the curvature prediction values ​​at two non-consecutive sampling times exceed the threshold, with one or more sampling times in between, the difference is considered significant. If the threshold is not exceeded, it is not considered a critical point. Critical point determination is only performed when two consecutive, uninterrupted sampling times both exceed the threshold. After determining that a critical point has been reached, the curvature prediction values, corresponding sampling times, and comparison results of the two consecutive sampling times are recorded in detail with a precision of four decimal places. At the same time, the current comparison process is stopped, and the determination of the critical point is completed. If no two consecutive sampling times are found where the curvature prediction values ​​exceed the runaway threshold, the curvature prediction values ​​of the sampling times are compared one by one in chronological order until two consecutive sampling times that meet the conditions are found, or the comparison of all sampling times is completed, to ensure that no possible critical point is missed.

[0163] S3.7: When the critical point is determined to be reached, trigger the element switching prediction signal, obtain the identifier of the currently controlled fluid element as the identifier of the fluid element to be scheduled, and output the identifier of the fluid element to be scheduled.

[0164] Furthermore, the specific steps of S3.7 include: real-time monitoring of the critical point determination results; when the curvature prediction values ​​at two consecutive sampling times both exceed the preset curvature runaway threshold, immediately triggering a component switching prediction signal. This signal is a pure logic trigger instruction, used only to initiate the acquisition and output of fluid controlled component identifiers; initiating the fluid controlled component identifier acquisition program, which automatically identifies all currently operating fluid controlled components. Each fluid controlled component is assigned a unique identifier, which uses a digital code with a length of 6 bits, ranging from 000001 to 999999. Each code corresponds to a unique fluid controlled component, ensuring the uniqueness and traceability of the identifier and avoiding confusion between different fluid controlled component identifiers; and using the system's built-in fluid controlled component location algorithm, which requires no additional parameter settings and matches the corresponding fluid controlled component based solely on the sampling time and data source of the currently monitored curvature data, thus clarifying the specific location of the fluid controlled component that has reached the critical point, ensuring the accuracy of the fluid controlled component to be switched, and preventing misjudgments of fluids. The controlled element status is as follows: A unique 6-digit identifier corresponding to the fluid controlled element to be switched is extracted. The extraction process only requires comparing the location of the fluid controlled element corresponding to the critical point with the fluid controlled element identifier database stored in the system to ensure that the extracted identifier completely corresponds to the fluid controlled element to be switched, without any identification errors or confusion. The extracted fluid element identifier to be scheduled is verified. The verification includes whether the identifier's code length is 6 digits and whether the code is within the range of 000001 to 999999. After verification, the identifier is confirmed as the fluid element identifier to be scheduled. The verification process does not require additional parameter settings; only the code format and range are checked to ensure they meet the requirements. The confirmed fluid element identifiers to be scheduled are organized and output clearly according to the system's preset output format. The output process ensures the integrity of the identifier information and simultaneously records the time of triggering the critical point and the corresponding curvature data. The location positioning algorithm is a one-to-one traceability positioning algorithm based on sampled data stream label matching, relying on the hardware port identifier and timestamp label inherent in the monitoring data to complete the accurate positioning of the fluid controlled element.

[0165] The process involves collecting the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element to be scheduled, prioritizing the agglomeration and blockage risk of each branch fluid controlled element, and outputting branch switching scheduling instructions, including:

[0166] S4.1: Collect the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element identifier to be scheduled, and output the agglomeration impact frequency and local blockage pressure gradient of the current fluid controlled element;

[0167] Furthermore, the specific steps of S4.1 include: retrieving the unique identification information of the fluid-controlled element to be switched, combining it with the basic data of the fluid-controlled elements stored in the system, identifying the target fluid-controlled element for which data is to be collected, ensuring that the collected agglomeration impact frequency and local blockage pressure gradient correspond to the fluid-controlled element to be switched, avoiding confusion with data from other fluid-controlled elements, and specifying the data collection time range, uniformly set to the most recent 1 hour, to ensure that the collected data can accurately reflect the current operating status of the fluid-controlled element, with the collection interval set to 10 seconds, i.e., data is collected once every 10 seconds, ensuring data continuity and timeliness; starting the data acquisition program, targeting the fluid to be switched... For the controlled element, all its operating data within the past hour is collected, with a focus on parameters related to agglomeration impacts and pressure data related to local blockage. The frequency of agglomeration impacts is collected using a fixed time interval statistical method, with a statistical period of 1 minute. This means that the number of agglomeration impacts experienced by the fluid-controlled element is counted every minute to ensure accurate recording of each impact event without omissions or misrecording. A clear statistical standard for agglomeration impact frequency is defined: the criterion for a single agglomeration impact is a momentary increase in local pressure difference of 5 Pa or more. If this condition is met, it is considered a single agglomeration impact and included in the agglomeration impact frequency statistics. During the statistical process, if multiple instances of agglomeration impact occur within one minute... All impacts meeting the specified conditions are counted separately, without repetition or omission, to ensure the accuracy of the agglomeration impact frequency statistics. Local blockage pressure gradient data of the fluid-controlled element to be switched are collected, strictly adhering to a preset sampling frequency of once per second. Each collected data point represents the current actual pressure value of the fluid-controlled element, with a precision of one decimal place to ensure the accuracy of the pressure data. Simultaneously, a standard for calculating the pressure gradient is established: the pressure difference between two adjacent sampling times is divided by the time interval to obtain the pressure change rate per unit time, i.e., the local blockage pressure gradient. The collected agglomeration impact frequency data are then processed, and the agglomeration impact frequency per minute over the past hour is statistically analyzed. The number of impacts is counted, and then the average agglomeration impact frequency within that hour is calculated as the final result of the agglomeration impact frequency for the fluid-controlled element to be switched. The collected pressure data is processed to calculate the pressure difference between two adjacent sampling times. Combined with the sampling time interval, the local blockage pressure gradient is calculated. The pressure gradient at each sampling time is calculated separately. The calculated agglomeration impact frequency and local blockage pressure gradient data are verified. The verification includes whether the data is complete and accurate. The agglomeration impact frequency must be an integer, and the local blockage pressure gradient must be retained to one decimal place. If there is missing data or the accuracy is not up to standard, the data at the corresponding time time must be re-collected until the data meets the requirements.The verified agglomeration impact frequency and local blockage pressure gradient data are organized and arranged sequentially according to the acquisition time to ensure a one-to-one correspondence between the data and the acquisition time, without any temporal disorder. The final output is the agglomeration impact frequency and local blockage pressure gradient corresponding to the fluid control element to be switched.

[0168] S4.2: The frequency of agglomeration impact is weighted and fused with the local blockage pressure gradient to construct the agglomeration and blockage risk index of each fluid-controlled element, and the agglomeration and blockage risk index of each fluid-controlled element is output.

[0169] S4.3: Arrange all online branch fluid-controlled components in descending order of their agglomeration and blockage risk index to obtain a risk priority queue;

[0170] S4.4: Extract the top-ranked fluid controlled component identifier from the risk priority queue as the fluid controlled component identifier to be removed in this switch; query the idle fluid controlled component identifier with the highest health score in the standby fluid controlled component pool as the fluid controlled component identifier to be deployed; output the fluid controlled component identifier to be removed and the fluid controlled component identifier to be deployed.

[0171] Furthermore, the specific steps of S4.4 include: retrieving the risk priority queue, where each entry corresponds to an identifier of a fluid-controlled component and its corresponding risk score. The risk score ranges from 0 to 1, with a higher score indicating a higher risk of blockage and requiring priority handling. The queue is strictly ordered from highest to lowest risk score to avoid disorder and ensure that the highest-risk fluid-controlled component is placed at the top of the queue; defining the reading rules for the risk priority queue, reading the fluid-controlled component identifiers and corresponding risk scores sequentially from front to back, with a reading precision of two decimal places to ensure accurate identification of the fluid-controlled component ranked first. The identifier is a unique numerical code used to uniquely distinguish each fluid controlled component, avoiding confusion with other fluid controlled components. If queue data is missing during the reading process, the process immediately returns to retrieve the queue again, ensuring that the identifier of the first-ranked fluid controlled component is accurate. The identifier of the first-ranked fluid controlled component in the risk priority queue is extracted and identified as the fluid controlled component to be switched and removed. Simultaneously, relevant basic information about this fluid controlled component is recorded, including its runtime, degree of blockage, and previous detection data, ensuring that the identifier of the fluid controlled component to be removed accurately corresponds to the specific fluid controlled component, preventing identifier errors. In cases of confusion or misidentification, the backup fluid controlled component pool data query program is initiated. This pool is a database that pre-stores information on all idle fluid controlled components. Each idle component has a unique identifier and a corresponding health score, ranging from 0 to 1. A higher score indicates a better condition and higher priority for deployment. The health score is calculated based on a comprehensive assessment of the component's runtime, past blockages, maintenance records, and other data. A health score screening criterion is established, clearly defining the idle fluid controlled component with the highest health score in the backup pool as the one to be deployed. The screening process strictly adheres to the health score criteria. The scores are sorted from highest to lowest using a step-by-step comparison method, first comparing the integer part and then the decimal part to ensure accurate sorting results and prevent situations where scores are the same but the order is disordered. The sorting precision is two decimal places. All idle fluid controlled components in the standby fluid controlled component pool are compared for health scores, and the selection is carried out sequentially starting from the highest score. If multiple fluid controlled components have the same health score, the idle time of the fluid controlled components is used as an auxiliary criterion. Fluid controlled components with longer idle times are given priority as fluid controlled components to be put into use, ensuring that the fluid controlled components to be put into use not only meet the health standards but can also be put into use quickly, avoiding situations where they cannot operate normally after being put into use.The system identifies the idle fluid control element with the highest health score. This identifier is also a unique numerical code, consistent with the coding rules of the fluid control element to be removed, facilitating system identification and switching operations. Simultaneously, it records basic information such as the specific health score and idle duration of the fluid control element to be deployed, ensuring that the fluid control element meets the standards for deployment and is free from faults and damage. The identified fluid control element identifiers and the fluid control element to be deployed undergo dual verification, including the uniqueness of the identifiers and the standardization of the coding format, ensuring that the risk level corresponding to the fluid control element to be removed is prioritized. The first fluid controlled element in the priority queue is identified as the idle fluid controlled element with the highest health status in the backup fluid controlled element pool, with no errors or confusion in identification. The relevant information of the fluid controlled element identifications to be removed and those to be added is compiled, clearly distinguishing their purposes: the identification for fluid controlled elements to be removed is used for fluid controlled element removal operations, while the identification for fluid controlled elements to be added is used for fluid controlled element switching and replacement. This ensures that the two identifications are not confused or omitted. After compilation, the final identifications of the fluid controlled elements to be removed and those to be added are output.

[0172] S4.5: Encapsulate the identifier of the fluid controlled component to be removed, the identifier of the fluid controlled component to be put into operation, and the switching order list generated based on the risk priority queue into a branch switching scheduling instruction, and output it to the fluid controlled component switching actuator.

[0173] The specific steps for constructing the agglomeration and clogging risk index for each fluid-controlled element include:

[0174] S4.2.1: Obtain the agglomeration impact frequency and local blockage pressure gradient and perform standardization processing to obtain normalized impact frequency and normalized pressure gradient;

[0175] Further, the specific steps in S4.2.1 include: retrieving the acquired agglomeration impact frequency data and local blockage pressure gradient data, where the agglomeration impact frequency is the number of agglomeration impacts experienced by the current fluid-controlled element per unit time, and the local blockage pressure gradient is the pressure change gradient formed during the blockage process of the current fluid-controlled element; both types of data are arranged in chronological order; setting the core parameters for standardization processing, where the standardization processing adopts a linear normalization algorithm; and, based on previous experimental data and actual operating conditions, setting the range of agglomeration impact frequency to 0 to 10 times per second. Zero cycles per second (CPS) represents a no-impact state, while 10 CPS represents a critical impact state. This range covers all possible impact frequencies. Considering the actual operation of the fluid-controlled components, the CPS is set to Pa / s, specifically ranging from 50 Pa / s to 200 Pa / s. This range encompasses pressure gradient changes under different degrees of blockage, including low pressure gradients for slight blockage and high pressure gradients for severe blockage. For agglomeration impact frequencies, with 0 CPS as the minimum and 10 CPS as the maximum, all agglomeration impact frequency data are linearly mapped to the 0-1 range. During the calculation process, it is ensured that each... The data points are accurately mapped. For example, a cluster impact frequency of 5 times per second is mapped to 0.5 after calculation, and a cluster impact frequency of 2 times per second is mapped to 0.2, ensuring that all data are evenly distributed within the target range. For localized blockage pressure gradients, the same linear normalization algorithm is used, with 50 Pa / s as the minimum and 200 Pa / s as the maximum, to map all pressure gradient data to the range of 0 to 1. For example, a pressure gradient of 200 Pa / s is mapped to 0.5, and a pressure gradient of 50 Pa / s is mapped to 0.25, ensuring consistency with the normalization standard for cluster impact frequencies. The normalization... The data on agglomeration impact frequency and local blockage pressure gradient were verified to check whether each data point was within the range of 0 to 1 and whether there were any abnormal data points outside the range. If abnormal data was found, normalization calculations were performed again to ensure data accuracy. At the same time, the corresponding relationship of each data point was recorded to avoid data corruption. The normalized agglomeration impact frequency was defined as the normalized impact frequency, and the normalized local blockage pressure gradient was defined as the normalized pressure gradient. The normalized impact frequency and normalized pressure gradient were organized and arranged in chronological order, and the time point corresponding to each data point was recorded.

[0176] S4.2.2: Multiply the normalized impact frequency by the preset first weighting coefficient to obtain the weighted impact component, and multiply the normalized pressure gradient by the preset second weighting coefficient to obtain the weighted pressure component;

[0177] Further, the specific steps of S4.2.2 include: retrieving the normalized impact frequency and normalized pressure gradient data after the previous standardization process, and simultaneously retrieving the preset first weighting coefficient and second weighting coefficient, where the first weighting coefficient is set to 0.6 and the second weighting coefficient is set to 0.4; multiplying the obtained normalized impact frequency data one by one by the preset first weighting coefficient 0.6, the calculation is performed according to the linear weighting algorithm, each normalized impact frequency data corresponds to one multiplication operation, and four decimal places are retained during the operation, the value obtained after the operation is the weighted impact component, the unit of the weighted impact component is consistent with the normalized impact frequency, no additional conversion is required; the obtained weighted impact components are sorted and arranged in chronological order to ensure that each weighted component is consistent with the normalized impact frequency. Each impact component corresponds to a unique time point, and the specific value of each weighted impact component is recorded. The obtained normalized pressure gradient data is multiplied one by one by a preset second weighting coefficient of 0.4. The accuracy of each value is checked to ensure that it meets the requirement of four decimal places and matches the numerical range of the normalized pressure gradient. If a calculation deviation occurs, it is recalculated in time to ensure the accuracy of the weighted pressure components and avoid affecting the judgment due to calculation errors. The calculation results of the weighted impact components and weighted pressure components are sorted according to the time sequence to ensure that each time point corresponds to a unique weighted impact component and weighted pressure component, forming a complete correspondence with the previous normalized data and original feature data, providing accurate weighted feature data for blockage type determination and status assessment.

[0178] S4.2.3: Add the weighted impact component and the weighted pressure component to obtain the agglomeration and clogging risk index of the fluid-controlled element.

[0179] Example 2

[0180] Please see Figure 3 Another embodiment of the present invention provides: an adaptive timing scheduling control system for multi-branch fluid elements, comprising:

[0181] Feature extraction module 10 is used to obtain the real-time differential pressure, medium moisture content and flue gas temperature of each branch fluid controlled element, segment the real-time differential pressure into time domain segments to obtain segmented differential pressure time series data, and extract four-dimensional feature parameters such as differential pressure rise slope, differential pressure pulsation amplitude, short-time differential pressure distortion coefficient and pressure gradient difference in folded intervals from the differential pressure.

[0182] The blockage type identification module 20 is used to distinguish between conventional progressive ash accumulation blockage, soft agglomeration impact blockage and hard agglomeration compaction blockage based on four-dimensional feature parameters. It establishes a dual threshold judgment logic based on instantaneous change threshold and steady-state cumulative threshold, and uses the medium moisture content and flue gas temperature to query the corresponding mapping table to obtain correction coefficients, dynamically correct the dual threshold judgment logic, and improve the accuracy of blockage judgment.

[0183] The critical point prediction module 30 is used to adjust the sliding window step size with the steady-state cumulative threshold and suppress outliers with the instantaneous mutation threshold based on the modified dual threshold judgment logic, and fit the pressure difference time series curve to obtain a smooth curvature sequence; input the sequence into the gated cyclic unit time series prediction network to predict the critical points of agglomeration, compaction, solidification and backflushing failure, trigger the element switching prediction signal and output the identifier of the fluid element to be scheduled;

[0184] The switching execution module 40 is used to collect the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element to be switched, construct an agglomeration and blockage risk index by weighted fusion, sort the risk priority of each branch fluid controlled element, determine the identifier of the fluid controlled element to be removed and the fluid controlled element to be put into operation, and generate a branch switching scheduling command. After the command is issued, the fluid controlled element switching is completed. At the same time, the four-dimensional characteristic parameters of the newly put fluid controlled element are collected in real time to form a closed loop control.

[0185] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. An adaptive timing scheduling control method for multi-branch fluid elements, characterized in that, include: The real-time differential pressure, medium moisture content, and flue gas temperature of the fluid-controlled components in each branch are obtained, and four-dimensional characteristic parameters are extracted. Based on the four-dimensional feature parameters, conventional progressive ash accumulation blockage, soft agglomeration impact blockage and hard agglomeration compaction blockage are distinguished to obtain the preliminary classification result of the blockage condition. Then, the preliminary classification result of the blockage condition is re-verified based on the dual threshold judgment logic. At the same time, the dual threshold judgment logic is dynamically corrected by the medium moisture content and flue gas temperature, and the corrected dual threshold judgment logic is output. Based on the modified dual-threshold judgment logic, the pressure difference time-series curve generated by the real-time pressure difference of each fluid controlled element is fitted to predict the critical point of agglomeration, compaction, solidification, and backflushing failure, trigger the element switching prediction signal, and output the identifier of the fluid element to be scheduled. The system collects the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element identifier to be scheduled, sorts the agglomeration and blockage risk priority of each branch fluid controlled element, outputs the branch switching scheduling command, and completes the switching of fluid controlled elements according to the branch switching scheduling command. After the switching, the system collects the four-dimensional characteristic parameters of the newly put fluid controlled element in real time.

2. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 1, characterized in that, Extract four-dimensional feature parameters, including: The real-time pressure difference is segmented into time domain segments based on the moisture content of the medium and the temperature of the flue gas to obtain segmented pressure difference time series data. Based on the segmented differential pressure time series data, the differential pressure rise slope is obtained; Based on the segmented differential pressure time series data, the maximum fluctuation difference between the real-time differential pressure and the preset pulsating reference differential pressure is calculated to obtain the differential pressure pulsation amplitude. Based on the segmented pressure difference time series data, the sudden pressure difference corresponding to the instantaneous impact of aggregated pollutants is extracted, and the sudden pressure difference is used as a weighting coefficient and substituted into the pre-constructed short-time pressure difference distortion coefficient calculation model to output the short-time pressure difference distortion coefficient. Multiple pressure acquisition units are arranged along the axial direction of the folds of the fluid-controlled element to acquire axial segmented pressure data of the folds, divide multiple pressure gradient detection intervals, calculate the differential pressure gradient value of each interval based on the axial segmented pressure data, and then calculate the differential pressure gradient difference between two adjacent pressure gradient detection intervals to obtain the pressure gradient difference of the fold interval. The pressure difference rise slope, pressure difference pulsation amplitude, short-time pressure difference distortion coefficient, and pressure gradient difference in the folded region are all output as four-dimensional feature parameters.

3. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 2, characterized in that, The construction process of the short-time pressure difference distortion coefficient calculation model includes: Acquire segmented differential pressure time series data under historical operating conditions and label sudden differential pressure events caused by instantaneous impacts of aggregated pollutants; Extract the impact amplitude and impact duration for each sudden pressure difference event, and define the ratio of impact amplitude to impact duration as the instantaneous impact intensity; Using instantaneous impact intensity as the independent variable and short-term pressure difference distortion coefficient as the target variable, an exponentially weighted moving average method is used to fit and construct a calculation model for the short-term pressure difference distortion coefficient. The model expression is as follows: Where D is the current short-term pressure difference distortion coefficient, and I is the instantaneous impact intensity. As a smoothing factor, The output is the short-time differential pressure distortion coefficient of the previous moment, and the completed short-time differential pressure distortion coefficient calculation model.

4. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 3, characterized in that, Based on the aforementioned four-dimensional characteristic parameters, conventional progressive ash accumulation blockage, soft agglomeration impact blockage, and hard agglomeration compaction blockage can be distinguished, including: The four-dimensional feature parameters are preset to match the characteristic response modes of conventional gradual ash accumulation blockage, soft agglomeration instantaneous change threshold range, and hard agglomeration compaction cumulative threshold range, respectively. The pressure difference rise slope, pressure difference pulsation amplitude, short-time pressure difference distortion coefficient, and pressure gradient difference in the folded area are compared one by one with the corresponding steady-state threshold range of conventional ash accumulation, the instantaneous change threshold range of soft agglomeration, and the cumulative threshold range of hard agglomeration compaction, and the transient response component and steady-state cumulative component of each characteristic parameter are extracted. Based on the extracted transient response components and steady-state cumulative components, a feature coupling judgment matrix is ​​constructed. The transient abrupt change features are characterized by the differential pressure pulsation amplitude and the short-term differential pressure distortion coefficient, which are used to match soft agglomeration impact blockage. The steady-state cumulative features are characterized by the differential pressure rise slope and the pressure gradient difference in the folded interval, which are used to match hard agglomeration compaction blockage. When the four-dimensional feature parameters show a steady-state change as a whole, it is matched with conventional progressive ash accumulation blockage, thus obtaining the preliminary classification results of the blockage conditions.

5. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 4, characterized in that, The verification of the preliminary classification results of siltation conditions based on the dual-threshold judgment logic includes: Four-dimensional characteristic parameters including differential pressure rise slope, differential pressure pulsation amplitude, short-term differential pressure distortion coefficient, and pressure gradient difference in folded regions are obtained. The weighted sum of differential pressure pulsation amplitude and short-term differential pressure distortion coefficient is defined as the instantaneous change characteristic value, and the weighted sum of differential pressure rise slope and pressure gradient difference in folded regions is defined as the steady-state cumulative characteristic value. Based on historical calibration data, instantaneous change thresholds and steady-state accumulation thresholds are preset, and a first timing window and a second timing window are set; the length of the first timing window is less than the length of the second timing window. Within the first timing window, the difference between the peak and trough values ​​of the real-time differential pressure is calculated as the instantaneous fluctuation amplitude. When the instantaneous fluctuation amplitude is greater than the instantaneous sudden change threshold, an instantaneous warning flag is triggered. Within the second timing window, the cumulative offset of the real-time differential pressure is calculated using an exponentially weighted moving average method. When the cumulative offset exceeds the steady-state cumulative threshold, a steady-state warning flag is triggered. When the instantaneous warning sign is triggered, the working condition is confirmed as soft agglomeration impact blockage; when the steady-state warning sign is triggered, the working condition is confirmed as hard agglomeration compaction blockage; when neither sign is triggered, the working condition is confirmed as conventional progressive ash accumulation blockage; the blockage condition judgment result after secondary verification is output.

6. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 5, characterized in that, The logic for dynamically correcting the dual threshold determination based on the moisture content of the medium and the flue gas temperature includes: Obtain the moisture content of the medium and the flue gas temperature at the current sampling time; Based on the moisture content of the medium, query the preset moisture content-instantaneous correction coefficient mapping table and the preset moisture content-steady-state correction coefficient mapping table, and output the first instantaneous correction coefficient and the first steady-state correction coefficient; Based on the flue gas temperature at the current sampling time, query the preset temperature-instantaneous correction coefficient mapping table and the preset temperature-steady-state correction coefficient mapping table, and output the second instantaneous correction coefficient and the second steady-state correction coefficient; Multiply the first instantaneous correction coefficient by the second instantaneous correction coefficient to obtain the instantaneous comprehensive correction factor, and multiply the first steady-state correction coefficient by the second steady-state correction coefficient to obtain the steady-state comprehensive correction factor. Obtain the preset instantaneous mutation threshold and steady-state accumulation threshold, multiply the instantaneous mutation threshold by the instantaneous comprehensive correction factor to obtain the corrected instantaneous mutation threshold, and multiply the steady-state accumulation threshold by the steady-state comprehensive correction factor to obtain the corrected steady-state accumulation threshold. The modified instantaneous mutation threshold is used to replace the instantaneous mutation threshold, and the modified steady-state cumulative threshold is used to replace the steady-state cumulative threshold. The updated dual threshold judgment logic is then output.

7. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 6, characterized in that, The step of fitting the differential pressure time-series curves generated by the real-time differential pressure for each fluid-controlled element based on the modified dual-threshold judgment logic includes: Extract the corrected instantaneous change threshold and the corrected steady-state cumulative threshold from the updated dual threshold judgment logic. At the same time, obtain the pressure difference time series curve generated by the real-time pressure difference of each fluid controlled element. Use the corrected steady-state cumulative threshold as an adaptive adjustment factor to adjust the sliding window step size in reverse and output the adjusted sliding window step size. The steady-state cumulative threshold and the window step size are inversely proportional. According to the output sliding window step size, take three consecutive points on the pressure difference time series curve window by window, calculate the ratio of the central angle to the arc length based on the three-point common circle method, and use it as the curvature value of the center point of each window, and output the initial curvature sequence. The modified instantaneous mutation threshold is used as an outlier suppression factor to filter the initial curvature sequence. The original pressure difference corresponding to each curvature value is traversed to three consecutive points. The vertical distance between the middle point and the line connecting the two points is calculated. If the vertical distance is greater than the outlier suppression factor, the mean of the curvature of the two points is used to replace the current curvature value. The smooth curvature sequence after outlier suppression is output.

8. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 7, characterized in that, The critical point of predicted agglomeration, compaction, solidification, and backflushing failure triggers a component switching prediction signal, outputting the identifier of the fluid component to be scheduled, including: Obtain a smooth curvature sequence, input the smooth curvature sequence into a time-series prediction network based on gated recurrent units in chronological order, and output a curvature time-series input vector; The gated recurrent unit in the time-series prediction network updates the smooth curvature values ​​at multiple consecutive sampling times sequentially and outputs the curvature prediction values ​​at multiple sampling times. The output curvature prediction value is compared with the preset curvature runaway threshold. When the curvature prediction value at two consecutive sampling times exceeds the curvature runaway threshold, it is determined that the critical point of agglomeration, compaction, solidification, and backflushing failure has been reached. When the critical point is reached, a component switching prediction signal is triggered, the identifier of the currently controlled fluid component is obtained as the identifier of the fluid component to be scheduled, and the identifier of the fluid component to be scheduled is output.

9. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 8, characterized in that, The process involves collecting the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element to be scheduled, prioritizing the agglomeration and blockage risk of each branch fluid controlled element, and outputting branch switching scheduling instructions, including: Collect the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element corresponding to the fluid element identifier to be scheduled, and output the agglomeration impact frequency and local blockage pressure gradient of the current fluid controlled element; The frequency of agglomeration impacts and the local blockage pressure gradient are weighted and fused to construct the agglomeration and blockage risk index for each fluid-controlled element, and the agglomeration and blockage risk index of each fluid-controlled element is output. All online operating branch fluid-controlled components are sorted in descending order of their agglomeration and blockage risk index to obtain a risk priority queue; Extract the top-ranked fluid controlled component identifier from the risk priority queue as the fluid controlled component identifier to be removed in this switchover; query the idle fluid controlled component identifier with the highest health score in the standby fluid controlled component pool as the fluid controlled component identifier to be deployed; output the fluid controlled component identifier to be removed and the fluid controlled component identifier to be deployed. The identifiers of the fluid-controlled components to be removed, the identifiers of the fluid-controlled components to be put into operation, and the switching order list generated based on the risk priority queue are encapsulated together into a branch switching scheduling instruction, which is then output to the fluid-controlled component switching actuator.

10. The adaptive timing scheduling control method for multi-branch fluid elements as described in claim 9, characterized in that, The specific steps for constructing the agglomeration and clogging risk index for each fluid-controlled element include: The agglomeration impact frequency and local blockage pressure gradient were obtained and standardized to obtain normalized impact frequency and normalized pressure gradient. The normalized impact frequency is multiplied by a preset first weighting coefficient to obtain the weighted impact component, and the normalized pressure gradient is multiplied by a preset second weighting coefficient to obtain the weighted pressure component. The weighted impact component and the weighted pressure component are added together to obtain the agglomeration and clogging risk index of the fluid-controlled component.

11. A multi-branch fluid element adaptive timing scheduling control system, used to implement the multi-branch fluid element adaptive timing scheduling control method according to any one of claims 1-10, characterized in that, include: The feature extraction module is used to obtain the real-time differential pressure, medium moisture content and flue gas temperature of the fluid controlled components in each branch, and to extract the segmented differential pressure time series data from the segmented time domain of the real-time differential pressure, and extract four-dimensional feature parameters from it. The blockage type identification module is used to distinguish between conventional progressive ash accumulation blockage, soft agglomeration impact blockage and hard agglomeration compaction blockage based on four-dimensional feature parameters. It establishes a dual threshold judgment logic based on instantaneous change threshold and steady-state cumulative threshold, and uses the medium moisture content and flue gas temperature to query the corresponding mapping table to obtain correction coefficients and dynamically correct the dual threshold judgment logic. The critical point prediction module is used to obtain a smooth curvature sequence by fitting the differential pressure time curve based on the modified dual threshold judgment logic, predict the critical point of compaction solidification and backflushing failure of agglomerates, trigger the element switching prediction signal and output the identifier of the fluid element to be scheduled. The switching execution module is used to collect the agglomeration impact frequency and local blockage pressure gradient of the fluid controlled element to be switched, construct the agglomeration and blockage risk index, determine the identifiers of the fluid controlled elements to be removed and to be put into operation, and generate branch switching scheduling instructions. After the instructions are issued, the switching of the fluid controlled element is completed, and at the same time, the four-dimensional characteristic parameters of the newly put into fluid controlled element are collected in real time.

Images