Discrimination method for local loss compensation of radial seal plate of air preheater based on reflective agent

Abstract

The application discloses a kind of based on reflection intelligent agent's air preheater radial sealing sheet local loss compensation discrimination method, belong to air preheater technical field.Method includes obtaining air preheater according to the data of synchronous rotor circumferential phase, the repeating appearance degree of each circumferential phase in each thermal disturbance event is calculated, obtain fixed circumferential phase candidate set and input reflection intelligent agent, extract fixed circumferential phase loss compensation characteristic sequence and carry out persistent homology calculation, obtain persistent life, and the comparison result of persistent life and life threshold is written into reflection record, according to the iteration proof of updated reflection record, obtain loss compensation phase set, then calculate local effective compensation margin shortage grade, output local loss compensation discrimination result.This method can give the local effective compensation margin discrimination result corresponding to specific circumferential position.

Description

Technical Field

[0001] This invention relates to the field of air preheater technology, and specifically to a method, system, device, and medium for determining local loss of compensation of radial sealing sheet in an air preheater based on a reflective intelligent agent. Background Technology

[0002] As the core heat exchange equipment in the flue gas system of thermal power boilers, rotary air preheaters rely on a rotor carrying heat storage elements to recover heat between the flue gas side and the air side. Radial sealing plates, axial sealing plates, and circumferential sealing structures play a crucial role in limiting air leakage, maintaining heat exchange efficiency, and reducing fan energy consumption. The radial sealing plates are distributed circumferentially along the rotor and are continuously affected by multiple factors during operation, including thermal deformation, ash accumulation, soot blowing impact, temperature differences caused by load fluctuations, and long-term wear. Problems such as jamming, insufficient tracking compensation, and abnormally large gaps can easily occur in localized areas. Once localized uncompensation occurs, it often initially manifests as repeated deviations in operating signals such as temperature, pressure difference, oxygen content, and current at a fixed circumferential position during multiple thermal changes. This gradually evolves into increased air leakage, deteriorated heat exchange, increased flue gas temperature, and decreased equipment economy. Since it is usually difficult to directly obtain the real-time mechanical displacement and actual gap distribution of each sealing plate on-site, engineering practices rely more on existing operating measurement points for condition assessment. Existing technologies are mainly geared towards online monitoring of air preheater sealing status. By collecting signals such as temperature, pressure, flow rate, oxygen content, and load during air preheater operation, and combining them with air preheater operating mechanisms or empirical models, the sealing gap status, leakage level, or sealing effect can be estimated and monitored. Some solutions also combine historical data with current operating condition changes to perform trend analysis or inter-condition comparison to assist in maintenance planning and operational adjustments. The technical idea is essentially to use existing on-site measuring points to replace direct mechanical measurement, and to achieve online characterization of sealing performance through the mapping relationship between operating data and sealing status. However, existing technical solutions mainly rely on the overall air leakage level, average sealing status, or operational condition-related deviations to establish judgment paths. It is difficult to effectively separate evidence of local jamming with a fixed circumferential phase from load changes, soot blowing disturbances, and synchronous thermal fluctuations of the entire unit. Furthermore, they do not make sufficient use of the recurring relationship between multiple thermal disturbances and lack a continuous verification mechanism based on the same circumferential position, adjacent circumferential positions, and event sequence. Therefore, it is difficult to reliably identify the local, repetitive, and fixed-position abnormal mechanism of local loss of compensation of radial sealing sheet, and it is also difficult to further provide a judgment result of insufficient local effective compensation margin corresponding to a specific circumferential position. This has become a key bottleneck restricting the improvement of the refined operation and maintenance level of air preheaters. Summary of the Invention

[0003] To address the problem in existing technologies where load changes, soot blowing disturbances, and overall machine synchronization fluctuations are coupled under high-fouling conditions, leading to the masking of weak signs of localized circumferential phase jamming and difficulty in reliably determining local effective compensation margins, this invention provides a method for determining localized loss of compensation of the radial sealing sheet of an air preheater based on a reflective intelligent agent. Instead of directly receiving abnormal quantities or single event fragments from the overall machine at the reflection verification entry point, this method first filters a candidate set of fixed circumferential phases from the circumferential phase synchronization data. Then, it establishes a fixed circumferential phase jamming hypothesis around each candidate circumferential phase and simultaneously retrieves data fragments of the target circumferential phase and its adjacent circumferential phases in the same thermal disturbance event for comparison, thereby providing a determination result of the local effective compensation margin corresponding to the specific circumferential position.

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

[0005] In a first aspect, the present invention provides a method for determining local loss of compensation of radial sealing sheet of an air preheater based on a reflective intelligent agent, comprising: Data on the circumferential phase synchronization of the air preheater is obtained, and data segments corresponding to each circumferential phase are divided according to thermal disturbance events to obtain circumferential phase data segments. Thermal disturbance events are determined based on circumferential phase data fragments, and the repetition rate of each circumferential phase in each thermal disturbance event is calculated to obtain a fixed circumferential phase candidate set. Input the fixed circumferential phase candidate set into the reflective agent to extract the fixed circumferential phase loss compensation feature sequence; The persistent cohomology is calculated based on the fixed circumferential phase loss compensation characteristic sequence to obtain the persistent lifetime. The comparison result between the persistent lifetime and the lifetime threshold is written into the reflection record to obtain the updated reflection record. Based on the updated reflection record, the candidate set of fixed circumferential phases is iteratively verified, and candidate circumferential phases that satisfy the fixed circumferential phase stagnation hypothesis are selected to obtain the uncompensated phase set. Based on the uncompensated phase set and the fixed circumferential phase uncompensated feature sequence, the level of insufficient local effective compensation margin is calculated, and the local uncompensated discrimination result is output.

[0006] As a further improvement of the present invention, the step of acquiring data on rotor circumferential phase synchronization of the air preheater, dividing the data into data segments corresponding to each circumferential phase according to thermal disturbance events, and obtaining circumferential phase data segments includes: The system acquires temperature, differential pressure, oxygen content, current, load, and soot blowing event data of the air preheater in the rotor circumferential phase synchronization during continuous operation cycles. Based on the timestamp and rotor angular position information, the temperature, differential pressure, oxygen content, current, load, and soot blowing event data are mapped to each circumferential phase and arranged in circumferential phase order within the same rotor circumferential cycle. Based on the start and end times of the thermal disturbance event, continuous data containing the events before, during, and after the thermal disturbance event is extracted from the data corresponding to each circumferential phase. The captured continuous data is arranged in order of circumferential phase and thermal disturbance events to obtain circumferential phase data segments.

[0007] As a further improvement of the present invention, the step of determining thermal disturbance events based on circumferential phase data fragments and calculating the repetition rate of each circumferential phase in each thermal disturbance event to obtain a fixed circumferential phase candidate set includes: Based on the circumferential phase data segments, the load change slope, soot blowing event trigger marker, and continuous offset of temperature or pressure difference within a preset time window are extracted in the order of operation time to determine the start time, end time, and thermal disturbance event identifier, thus forming a thermal disturbance event sequence. Based on the thermal disturbance event sequence, an event reference segment corresponding to the thermal disturbance event identifier is formed; Based on the event comparison segments, the temperature shift, pressure difference shift, oxygen content shift, and current shift of each circumferential phase in each thermal disturbance event are calculated respectively. The temperature shift, pressure difference shift, oxygen content shift, and current shift are then combined according to the thermal disturbance event identifier and the circumferential phase sequence to form the circumferential phase anomaly trajectory corresponding to each circumferential phase. Based on the circumferential phase anomaly trajectory, the same circumferential phase is matched in different thermal disturbance events, and the recurrence score of each circumferential phase is calculated; based on the recurrence score, the degree of recurrence of each circumferential phase in each thermal disturbance event is determined, forming a fixed circumferential phase recurrence result; The results of repeated occurrence of fixed circumferential phases are compared with preset thresholds and thermal disturbance event coverage thresholds. Circumferential phases that meet the comparison conditions are selected and sorted from high to low repetition degree to obtain a candidate set of fixed circumferential phases.

[0008] As a further improvement of the present invention, the step of inputting a fixed circumferential phase candidate set into the reflective agent and extracting a fixed circumferential phase loss-compensation feature sequence includes: Input a fixed circumferential phase candidate set into the reflective agent, perform event matching and time alignment, and form an event sequence comparison result; Based on the results of the event sequence comparison, the changes in the target circumferential phase in terms of temperature, pressure difference, oxygen content and current are calculated respectively to form the phase residual; Based on the results of the event sequence comparison, the abnormal start time and abnormal recovery time of the target circumferential phase in the thermal disturbance event change segment are extracted to form a response hysteresis. Based on the results of the event sequence comparison, the differences in temperature, pressure difference, oxygen content and current of the target circumferential phase before and after the soot blowing event are calculated to form the soot blowing recovery difference. The differences in temperature, pressure difference, oxygen content and current between the target circumferential phase and the adjacent circumferential phases on both sides in the same thermal disturbance event and the same section are also calculated to form the adjacent circumferential phase difference. The phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase difference are combined with the target circumferential phase sequence according to the thermal disturbance event sequence to obtain a fixed circumferential phase loss compensation characteristic sequence.

[0009] As a further improvement of the present invention, the persistent cohomology calculation based on the fixed circumferential phase loss compensation feature sequence is performed to obtain the persistent lifetime, and the comparison result of the persistent lifetime and the lifetime threshold is written into the reflection record to obtain the updated reflection record, including: Based on the fixed circumferential phase loss compensation characteristic sequence, the order of appearance and disappearance of connected components corresponding to each level of connection relationship is obtained, forming the long-term lifetime; The longevity is compared with the lifetime threshold. If the comparison conditions are met, a continuous verification update result is generated. If the comparison conditions are not met, a verification intensity reduction update result is generated, resulting in a reflection record update result. Write the updated reflection record results into the corresponding candidate circumferential phase record position in the reflection record, and adjust the verification state of the corresponding candidate circumferential phase according to the updated reflection record results to obtain the updated reflection record.

[0010] As a further improvement of the present invention, the step of iteratively verifying the candidate set of fixed circumferential phases based on the updated reflection record, and screening candidate circumferential phases that satisfy the fixed circumferential phase lag assumption to obtain the uncompensated phase set includes: Based on the updated reflection record, extract the record value corresponding to each candidate circumferential phase in the fixed circumferential phase candidate set to form a candidate circumferential phase verification sequence; According to the candidate circumferential phase verification sequence, the verification results of the fixed circumferential phase stagnation hypothesis corresponding to each candidate circumferential phase in different thermal disturbance events are iteratively summarized to form the candidate circumferential phase verification results. The verification results of the candidate circumferential phases are compared with a preset threshold. If the comparison conditions are met, the corresponding candidate circumferential phases are determined to satisfy the fixed circumferential phase lag assumption. If the comparison conditions are not met, the corresponding candidate circumferential phases are eliminated to form a set of valid candidate circumferential phases. The set of valid candidate circumferential phases is then filtered from high to low according to the verification results of the candidate circumferential phases to obtain a set of uncompensated phases.

[0011] As a further improvement of the present invention, the step of calculating the local effective compensation margin deficiency level based on the uncompensated phase set and the fixed circumferential phase uncompensated feature sequence, and outputting the local uncompensated discrimination result, includes: Based on the uncompensated phase set, the extracted results are reorganized into a phase level calculation sequence that maintains the correspondence between events according to the uncompensated phase and the thermal disturbance event sequence; Based on the phase level calculation sequence, the phase residual and adjacent circumferential phase difference of the same uncompensated phase in each thermal disturbance event are combined and calculated, and the response hysteresis and soot blowing recovery difference are combined and calculated to form the margin calculation value corresponding to each uncompensated phase. Based on the margin calculation value, the margin calculation values ​​of each loss compensation phase in continuous thermal disturbance events are accumulated, and the maximum duration segment in the accumulated results is extracted to form the local effective compensation margin insufficient value. The local effective compensation margin deficiency value is compared with the preset level threshold to determine the local effective compensation margin deficiency level. The set of uncompensated phases is then associated with the local effective compensation margin deficiency level in a one-to-one correspondence according to the uncompensated phases to obtain the local uncompensated discrimination result.

[0012] Secondly, the present invention provides a system for determining the local loss of compensation of radial sealing sheet of an air preheater based on a reflective intelligent agent, comprising: Circumferential phase data segment module: used to acquire data of the air preheater synchronized with the rotor circumferential phase, divide the data segments corresponding to each circumferential phase according to the thermal disturbance event, and obtain the circumferential phase data segment; Fixed circumferential phase candidate set module: used to determine thermal disturbance events based on circumferential phase data segments, and calculate the repetition degree of each circumferential phase in each thermal disturbance event to obtain a fixed circumferential phase candidate set; Phase loss compensation feature sequence module: used to input a fixed circumferential phase candidate set into the reflective agent and extract a fixed circumferential phase loss compensation feature sequence; Update the reflection record module: It is used to perform persistent cohomology calculation based on the fixed circumferential phase loss compensation feature sequence, obtain the persistent lifetime, and write the comparison result of persistent lifetime and lifetime threshold into the reflection record to obtain the updated reflection record; Uncompensated phase set module: It is used to iteratively verify the fixed circumferential phase candidate set based on the updated reflection record, filter the candidate circumferential phases that satisfy the fixed circumferential phase stagnation assumption, and obtain the uncompensated phase set; Loss of compensation discrimination result module: It is used to calculate the level of insufficient local effective compensation margin based on the set of uncompensated phases and the fixed circumferential phase loss of compensation characteristic sequence, and output the local loss of compensation discrimination result.

[0013] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the aforementioned method for determining the local loss compensation of radial sealing sheet of an air preheater based on a reflective intelligent agent.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned method for determining local loss compensation of radial sealing sheet of an air preheater based on a reflective intelligent agent.

[0015] Fifthly, the present invention provides a computer program product, including computer instructions, which, when executed by a processor, implement the aforementioned method for determining the local loss compensation of radial sealing sheet of an air preheater based on a reflective intelligent agent.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This method fundamentally improves the signal-to-noise ratio and discrimination reliability of weak symptom extraction by establishing a mechanism of synchronous screening of the circumferential phase before the reflection and verification entry point and the assumption of fixed phase stagnation. First, the method synchronously divides the data based on the rotor circumferential phase, decoupling local abnormal signals originally mixed in with overall machine fluctuations according to the phase dimension. Then, by calculating the recurrence degree of each circumferential phase in thermal disturbance events, a fixed circumferential phase candidate set is constructed, effectively filtering out false correlations caused by single random disturbances. This ensures that the input entering the reflection agent has statistical stability, thereby avoiding misjudgments caused by noise-driven reflection agents. Secondly, the candidate set is input into the reflective agent to extract the uncompensated feature sequence, and persistent cohomology calculation is introduced to obtain the topological durability of the features. Topological invariants are used to replace empirical thresholds for preliminary discrimination, and the discrimination results are written into the reflection record to form an accumulative and traceable reasoning basis. Subsequently, the updated reflection record is used to iteratively verify the candidate set. Each iteration revolves around the fixed circumferential phase stagnation hypothesis, and data segments of the target phase and its two adjacent phases in the same thermal disturbance event are retrieved simultaneously for comparison, effectively distinguishing between true local uncompensation and global systemic fluctuations, and significantly suppressing false positives. Finally, this method calculates the level of insufficient local effective compensation margin based on the uncompensated phase set and the uncompensated feature sequence. It not only outputs a binary conclusion of whether there is uncompensation, but also provides a margin quantification level that can directly guide operation and maintenance decisions, realizing a leap from the ability to detect to the severity of the problem. Attached Figure Description

[0017] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. In the drawings: Figure 1This is a flowchart illustrating a method for determining local loss compensation of radial sealing sheet in an air preheater based on a reflective intelligent agent, according to the present invention. Figure 2 This is a schematic diagram illustrating the specific process of the present invention: a method for determining local loss compensation of radial sealing sheet in an air preheater based on a reflective intelligent agent. Figure 3 This is a flowchart illustrating step S1 in the present invention, a method for determining local loss compensation of radial sealing sheet of air preheater based on reflective intelligent agent; Figure 4 This is a flowchart illustrating step S2 in the method for determining local loss compensation of radial sealing sheet of air preheater based on reflective intelligent agent according to the present invention. Figure 5 This is a flowchart illustrating step S3 in the present invention, which is a method for determining the local loss of compensation of radial sealing sheet of air preheater based on reflective intelligent agent. Figure 6 This is a flowchart illustrating step S4 in the present invention, a method for determining local loss of compensation of radial sealing sheet of air preheater based on reflective intelligent agent; Figure 7 This is a flowchart illustrating step S5 in the present invention, a method for determining local loss compensation of radial sealing sheet of air preheater based on reflective intelligent agent; Figure 8 This is a flowchart illustrating step S6 in the present invention, which is a method for determining the local loss of compensation of radial sealing sheet of air preheater based on reflective intelligent agent. Figure 9 This is a schematic diagram of the network architecture of the algorithm of this invention; Figure 10 This is a schematic diagram of the second network architecture of the algorithm of the present invention; Figure 11 This is a schematic diagram of the network architecture of the algorithm of this invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0020] Definitions: Thermal disturbance events: refer to operating processes that are triggered by load changes or soot blowing events and result in continuous deviations in temperature or pressure differential; Circumferential phase data segment: refers to a data segment covering a continuous time range before, during, and after the thermal disturbance event, under the same thermal disturbance event; Event comparison segment: refers to the comparison segment formed by aligning the reference segment and response segment of each circumferential phase in the same thermal disturbance event according to the thermal disturbance event identifier and start time; Thermal disturbance event coverage threshold: refers to the minimum number of thermal disturbance events that must be met before the same circumferential phase can be included in the fixed circumferential phase candidate set; Fixed circumferential phase candidate set: refers to the set of circumferential phases whose repetition rate reaches a preset threshold and meets the thermal disturbance event coverage threshold; Reflective agent: refers to a processing model that establishes a fixed circumferential phase stagnation hypothesis around each candidate circumferential phase in a fixed circumferential phase candidate set, and generates a fixed circumferential phase loss compensation feature sequence based on the comparison results of the target circumferential phase and adjacent circumferential phases in each thermal disturbance event; Fixed circumferential phase stagnation hypothesis: refers to the hypothesis that the same candidate circumferential phase will repeatedly exhibit local abnormal shifts in multiple thermal disturbance events. Phase residual: refers to the offset obtained by comparing the changes in the target circumferential phase in temperature, pressure difference, oxygen content and current with the corresponding changes in the circumferential phase on both sides, in combination with the load change. Response lag: refers to the lag between the abnormal start and recovery times of the target circumferential phase and the start and end times of the thermal disturbance event, as well as the corresponding changes in the circumferential phases on both sides. Soot blowing recovery difference: refers to the difference in temperature, pressure difference, oxygen content and current of the target circumferential phase before and after the occurrence of a soot blowing event; Adjacent circumferential phase difference: refers to the temperature, pressure difference, oxygen content, and current difference between the target circumferential phase and the adjacent circumferential phases on both sides in the same thermal disturbance event and the same section; Fixed circumferential phase loss compensation characteristic sequence: refers to the sequence formed by continuously arranging the phase residual, response hysteresis, soot blowing recovery difference and adjacent circumferential phase difference according to the thermal disturbance event sequence and the target circumferential phase sequence; Rotor circumferential adjacency constraint: refers to the restriction condition that the connection relationship in the fixed circumferential phase loss compensation relationship diagram is established only within the local circumferential range corresponding to the candidate circumferential phase and the two adjacent circumferential phases on both sides; Thermal disturbance event consistency constraint: refers to the restriction that the node matching and connection relationship in the fixed circumferential phase loss compensation relationship diagram is established only between nodes of the same type of thermal disturbance event and whose thermal disturbance events correspond in sequence; Fixed circumferential phase loss compensation relationship graph: refers to a graph structure constructed based on the fixed circumferential phase loss compensation feature sequence, with the extracted results corresponding to each thermal disturbance event as nodes and the combined distance connection relationship as edges; Combined distance: refers to the distance between nodes determined by the combined differences of phase residual, response hysteresis, soot blowing recovery difference and adjacent circumferential phase difference; The order of appearance and disappearance of connected components refers to the order in which connected components begin to form and cease to exist independently when expanding the connection relationship in a fixed circumferential phase loss compensation relationship diagram in order of increasing combination distance. Durability: refers to the range in which the same connected component persists between the order in which it appears and the order in which it disappears. Reflection Log: This refers to the log entries used to write down the comparison results between the lifetime and lifetime threshold corresponding to the candidate circumferential phase and adjust the verification state accordingly. Iterative verification: refers to the process of summarizing the verification results of the fixed circumferential phase stagnation hypothesis in multiple thermal disturbance events in a round-by-round manner according to the candidate circumferential phase verification sequence, and continuously retaining or eliminating candidate circumferential phases based on the updated reflection record. Local effective compensation margin insufficiency value: refers to the degree of insufficiency corresponding to the largest sustained segment extracted from the cumulative result of the margin calculation of each uncompensated phase in the uncompensated phase set in continuous thermal disturbance events; Local effective compensation margin insufficient level: refers to the level result determined by comparing the local effective compensation margin insufficient value with the preset level threshold.

[0021] To address the problem in existing technologies where load changes, soot blowing disturbances, and overall machine synchronization fluctuations are coupled under high-fouling conditions, leading to the masking of weak signs of localized circumferential phase jamming and difficulty in reliably determining localized effective compensation margins, this invention provides a method for determining localized loss of compensation of the radial sealing plate of an air preheater based on a reflective intelligent agent. Figure 1 As shown, it includes: Data on the circumferential phase synchronization of the air preheater is obtained, and data segments corresponding to each circumferential phase are divided according to thermal disturbance events to obtain circumferential phase data segments. Thermal disturbance events are determined based on circumferential phase data fragments, and the repetition rate of each circumferential phase in each thermal disturbance event is calculated to obtain a fixed circumferential phase candidate set. Input the fixed circumferential phase candidate set into the reflective agent to extract the fixed circumferential phase loss compensation feature sequence; The persistent cohomology is calculated based on the fixed circumferential phase loss compensation characteristic sequence to obtain the persistent lifetime. The comparison result between the persistent lifetime and the lifetime threshold is written into the reflection record to obtain the updated reflection record. Based on the updated reflection record, the candidate set of fixed circumferential phases is iteratively verified, and candidate circumferential phases that satisfy the fixed circumferential phase stagnation hypothesis are selected to obtain the uncompensated phase set. Based on the uncompensated phase set and the fixed circumferential phase uncompensated feature sequence, the level of insufficient local effective compensation margin is calculated, and the local uncompensated discrimination result is output.

[0022] This method no longer directly receives abnormal quantities of the whole machine or single event fragments at the reflection and verification entry point. Instead, it first filters a fixed circumferential phase candidate set from the circumferential phase synchronization data, then establishes a fixed circumferential phase stagnation hypothesis around each candidate circumferential phase, and simultaneously retrieves data fragments of the target circumferential phase and the adjacent circumferential phases on both sides in the same thermal disturbance event for comparison.

[0023] The present invention will be further explained and described below with reference to the accompanying drawings.

[0024] like Figure 2 As shown, a method for determining local loss compensation of radial sealing sheet in an air preheater based on reflective agents includes the following steps: S1: As Figure 3 As shown, the temperature, pressure difference, oxygen content, current, load and soot blowing event data of the air preheater are obtained in circumferential phase synchronization of the rotor. The data segments corresponding to each circumferential phase are divided according to the thermal disturbance events to obtain the circumferential phase data segments. Specifically, the air preheater acquires temperature, differential pressure, oxygen content, current, load, and soot blowing event data in the rotor circumferential phase synchronization during continuous operation cycles, and maps the temperature, differential pressure, oxygen content, current, load, and soot blowing event data to each circumferential phase according to the timestamp and rotor angular position information, and arranges them in circumferential phase order within the same rotor circumferential cycle. Based on the load change slope, soot blowing event trigger marker, and continuous offset of temperature or pressure difference within a preset time window in the data corresponding to each circumferential phase, the start and end times of thermal disturbance events are determined, and the time period that meets the preset threshold condition is determined as a thermal disturbance event. According to the start and end times of the thermal disturbance event, continuous data containing the events before, during, and after the thermal disturbance event are extracted from the data corresponding to each circumferential phase, and the extraction range of each circumferential phase in the same thermal disturbance event is made consistent. The captured continuous data is arranged in order of circumferential phase and thermal disturbance events to obtain circumferential phase data segments.

[0025] The process involves inputting a fixed circumferential phase candidate set into a reflexive agent to obtain a fixed circumferential phase uncompensated feature sequence. The fixed circumferential phase candidate set is calculated from the repetition rate of circumferential phase data segments, indicating that certain circumferential phases exhibit anomalies multiple times in various thermal disturbance events. However, the fixed circumferential phase candidate set itself does not include the distinguishing condition between local jamming and overall machine disturbance. When the reflexive agent is directly applied to this scenario, it firstly suffers from a mismatch between the input representation and the scenario mechanism. Conventional reflexive agents receive entire anomaly descriptions or single event segments; the input object does not fix the difference relationship between the target circumferential phase and adjacent circumferential phases. Therefore, local offsets truly related to jamming under high-viscosity ash conditions will be mixed in with overall machine synchronous fluctuations caused by load changes, smoke temperature changes, and soot blowing disturbances in the same input object. The reflexive agent also suffers from a mismatch between its intermediate state update method and the scenario mechanism. Conventional reflection records are written to a single conclusion per round, primarily reflecting which type of anomaly is more prominent in that round. Radial seal jamming, however, corresponds to the persistent recurrence of a fixed circumferential phase across multiple thermal disturbance events. This persistent relationship is not organized into a stable intermediate object before writing. The problem stems from the input and writing objects themselves. Changing the threshold, memory length, or retrieval range while still using the same input representation and intermediate state update method will still result in weak signs of fixed circumferential phase jamming being scattered across multiple circumferential phase data segments. Based on this mechanism, this solution structurally modifies the reflection agent. The reflection agent no longer establishes broad hypotheses around system-wide anomalies, but instead establishes a fixed circumferential phase jamming hypothesis around each candidate circumferential phase in the fixed circumferential phase candidate set. After establishing the fixed circumferential phase jamming hypothesis, it does not directly read the target circumferential phase data, but simultaneously retrieves circumferential phase data segments of the target circumferential phase and adjacent circumferential phases in the same thermal disturbance event, and compares them according to the order of the thermal disturbance events. In this way, the generation of the fixed circumferential phase jamming hypothesis and the comparison with adjacent circumferential phases are organized into the same processing path, and the input representation is transformed into the form of "candidate circumferential phase - thermal disturbance event sequence - adjacent circumferential phase difference". Subsequently, phase residuals, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase differences are extracted from the comparison results to form a fixed circumferential phase uncompensated feature sequence. The phase residual corresponds to the degree of offset of the fixed circumferential phase, the response hysteresis corresponds to the hysteresis characteristics after the change of thermal gap, the soot blowing recovery difference corresponds to the correlation of ash accumulation and nodule formation, and the adjacent circumferential phase difference corresponds to the local boundary of the anomaly. The fixed circumferential phase uncompensated feature sequence thus organizes the local jamming evidence scattered in the circumferential phase data segments into an intermediate object that combines the fixed circumferential phase position, thermal disturbance event sequence, and adjacent comparison relationship. The nodes and connections of the fixed circumferential phase uncompensated relationship graph are all generated based on this intermediate object. Therefore, the fixed circumferential phase uncompensated feature sequence becomes a prerequisite for persistent coherence calculation to enter this scenario.

[0026] S2: As Figure 4 As shown, thermal disturbance events are determined based on circumferential phase data fragments, and the repetition rate of each circumferential phase in each thermal disturbance event is calculated to obtain a fixed circumferential phase candidate set. Specifically, the system receives circumferential phase data segments, extracts the load change slope, soot blowing event trigger marker, and continuous offset of temperature or pressure difference within a preset time window according to the running time sequence, and determines the start time, end time, and thermal disturbance event identifier of the thermal disturbance event when the load change condition or soot blowing event trigger condition is met and the continuous offset exceeds a preset threshold, thus forming a thermal disturbance event sequence. Based on the thermal disturbance event sequence, the circumferential phase data segments of each circumferential phase in the same thermal disturbance event are aligned according to the start time, and the reference segment before the thermal disturbance event and the response segment after the thermal disturbance event are extracted from the circumferential phase data segments corresponding to each circumferential phase to form an event comparison segment corresponding to the thermal disturbance event identifier. Based on the event comparison segments, the temperature shift, pressure difference shift, oxygen content shift, and current shift of each circumferential phase in each thermal disturbance event are calculated respectively. The temperature shift, pressure difference shift, oxygen content shift, and current shift are then combined according to the thermal disturbance event identifier and the circumferential phase sequence to form the circumferential phase anomaly trajectory corresponding to each circumferential phase. Based on the circumferential phase anomaly trajectory, the consistency of the anomaly direction, the order of anomaly onset, the duration of anomaly, and the recovery trend of the same circumferential phase in different thermal disturbance events are matched, and the recurrence score of each circumferential phase is calculated. Based on the recurrence score, the difference between the target circumferential phase and the adjacent circumferential phase in the same thermal disturbance event is compared, and the recurrence score is adjusted according to the difference comparison result to determine the degree of recurrence of each circumferential phase in each thermal disturbance event, thus forming a fixed circumferential phase recurrence result. The results of repeated occurrence of fixed circumferential phases are compared with preset thresholds and thermal disturbance event coverage thresholds. Circumferential phases that meet the comparison conditions are selected and sorted from high to low repetition degree to obtain a candidate set of fixed circumferential phases.

[0027] The recurrence degree of each circumferential phase in the results of fixed circumferential phase recurrence in thermal disturbance events formed under load change conditions is compared with a preset threshold and a thermal disturbance event coverage threshold. The recurrence degree of each circumferential phase in thermal disturbance events formed under soot blowing event trigger conditions is also compared with a preset threshold and a thermal disturbance event coverage threshold. If both comparisons meet the comparison conditions, the corresponding circumferential phase is retained as a circumferential phase in the fixed circumferential phase candidate set.

[0028] The process involves establishing a fixed circumferential phase loss compensation relationship graph based on the fixed circumferential phase loss compensation feature sequence, performing persistent cohomology calculations, and generating an updated reflection record. The fixed circumferential phase loss compensation feature sequence has extracted the phase residuals, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase differences related to local jamming. However, this sequence is still a continuous feature record unfolded according to thermal disturbance events, and the abnormal amplitude and rate of change under different thermal disturbance events will be affected by operating conditions. If the fixed circumferential phase loss compensation feature sequence is directly written into the reflection record, the reflection record will still be influenced by the strength of a single thermal disturbance event. In the next round of verification, the reflection agent will still update the fixed circumferential phase jamming hypothesis along the direction with the larger instantaneous amplitude. If persistent cohomology directly processes all feature points in a general way, the input object will not correspond to the scene mechanism. General persistent cohomology mainly relies on the geometric proximity relationship between samples and cannot directly express the rotor circumferential adjacency relationship and the correspondence between thermal disturbance events. The calculation result is closer to the global fluctuation structure than the fixed circumferential phase local jamming structure. Based on this, this scheme does not directly copy persistent cohomology, but extracts the core mechanism of persistent cohomology for "the continued existence of connected structures under changes in relational scale," and modifies the input objects and output action points of persistent cohomology. First, based on the fixed circumferential phase loss compensation feature sequence, a fixed circumferential phase loss compensation relationship graph is established under rotor circumferential adjacency constraints and thermal disturbance event consistency constraints. The rotor circumferential adjacency constraints limit the connection range to the target circumferential phase and adjacent circumferential phases, reflecting the local physical boundary of radial seal plate jamming; the thermal disturbance event consistency constraints limit the connection range to corresponding thermal change processes, reflecting the temporal comparability of response hysteresis and soot blowing recovery difference. The fixed circumferential phase loss compensation relationship graph established in this way is no longer an ordinary feature similarity graph, but a structured object with fixed circumferential phase position constraints and thermal disturbance event sequence constraints. Then, persistent cohomology calculation is performed on the fixed circumferential phase loss compensation relationship graph to obtain the persistent lifetime. Here, the persistence lifetime is not used as an independent diagnostic result, but rather as the basis for writing the reflection record: when the comparison result between the persistence lifetime and the lifetime threshold satisfies the condition for the continued existence of the fixed circumferential phase, the updated reflection record retains the candidate circumferential phase as a persistent verification object; when the comparison result does not satisfy the condition for the continued existence of the fixed circumferential phase, the updated reflection record lowers the verification intensity corresponding to the candidate circumferential phase. In this way, the output position of persistent cohomology is transformed into an intermediate state update stage of the reflection record. The reflection record no longer records the strength of a single anomaly, but rather the degree of persistence of the fixed circumferential phase uncompensated relationship graph in multi-scale relationship changes. The updated reflection record thus becomes the direct basis for screening the set of uncompensated phases, and whether the fixed circumferential phase stagnation hypothesis holds is also supported by the persistence lifetime corresponding to the fixed circumferential phase uncompensated relationship graph.

[0029] S3: As Figure 5 As shown, a fixed circumferential phase candidate set is input into a reflective agent. The reflective agent establishes a fixed circumferential phase lag hypothesis for each candidate circumferential phase, retrieves the corresponding circumferential phase data segment, compares the target circumferential phase with the circumferential phase data segments of adjacent circumferential phases in the order of thermal disturbance events, and extracts the phase residual, response hysteresis, soot blowing recovery difference and adjacent circumferential phase difference to obtain the fixed circumferential phase uncompensated feature sequence. Specifically, a fixed circumferential phase candidate set is input into the reflective agent. The reflective agent establishes a fixed circumferential phase stagnation hypothesis for each candidate circumferential phase based on the degree of repetition of each candidate circumferential phase in each thermal disturbance event, combined with the order of thermal disturbance events. The fixed circumferential phase stagnation hypothesis is limited to the abnormal hypothesis that the same candidate circumferential phase repetitively appears in multiple thermal disturbance events, thus forming a verification unit. According to the verification unit, each candidate circumferential phase is retrieved from the circumferential phase data segment as the circumferential phase data segment of the target circumferential phase in each thermal disturbance event, and the circumferential phase data segments of the two adjacent circumferential phases on both sides that are circumferentially adjacent to each target circumferential phase are retrieved simultaneously, so that the target circumferential phase and the two adjacent circumferential phases on both sides correspond to the same thermal disturbance event and the same time range, forming a control unit. The control unit is matched and time-aligned according to the order of thermal disturbance events. Within the same thermal disturbance event, the base segment before the thermal disturbance event, the thermal disturbance event change segment, and the thermal disturbance event recovery segment are divided in sequence. The data of the target circumferential phase and the adjacent circumferential phases on both sides in each segment are arranged in order to form the event sequence comparison result. Based on the results of the event sequence comparison, and with the reference section as a reference, the changes in temperature, pressure difference, oxygen content and current of the target circumferential phase are calculated respectively. The changes are then compared with the load changes in the same thermal disturbance event, and compared with the changes in the adjacent circumferential phases on both sides in the same section to form the phase residual. Based on the event sequence comparison results, the abnormal start time and abnormal recovery time of the target circumferential phase in the thermal disturbance event change segment are extracted, and the abnormal start time and abnormal recovery time are compared with the start time and end time of the thermal disturbance event, respectively. Then, combined with the change sequence relationship of the corresponding times of the adjacent circumferential phases on both sides, a response lag is formed. Based on the results of the event sequence comparison, the differences in temperature, pressure difference, oxygen content and current of the target circumferential phase before and after the soot blowing event are calculated to form the soot blowing recovery difference. The differences in temperature, pressure difference, oxygen content and current between the target circumferential phase and the adjacent circumferential phases on both sides in the same thermal disturbance event and the same section are also calculated to form the adjacent circumferential phase difference. The phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase difference are combined according to the thermal disturbance event sequence and the target circumferential phase sequence. The combination results corresponding to each thermal disturbance event are then continuously arranged under the same target circumferential phase to obtain a fixed circumferential phase loss compensation characteristic sequence.

[0030] When forming phase residuals based on the event sequence comparison results, the target circumferential phase is compared with the changes in temperature, pressure difference, oxygen content, and current in the same thermal disturbance event and the same segment on both sides. The offset of the comparison results relative to the reference segment is determined as the phase residual. When forming adjacent circumferential phase differences, the comparison results between the target circumferential phase and the adjacent circumferential phase on one side and the target circumferential phase and the adjacent circumferential phase on the other side are retained and arranged in circumferential order. When forming response hysteresis, the anomaly start time is determined based on the starting point of the continuous unidirectional changes in temperature, pressure difference, oxygen content, and current in the thermal disturbance event change segment and thermal disturbance event recovery segment. The anomaly recovery time is determined based on the ending point of the continuous unidirectional changes in temperature, pressure difference, oxygen content, and current.

[0031] S4: As Figure 6 As shown, based on the fixed circumferential phase loss compensation characteristic sequence, a fixed circumferential phase loss compensation relationship diagram is established under the rotor circumferential adjacency constraint and thermal disturbance event consistency constraint. The fixed circumferential phase loss compensation relationship diagram is subjected to persistent coherence calculation to obtain the persistent lifetime. The comparison result between the persistent lifetime and the lifetime threshold is written into the reflection record to obtain the updated reflection record. Specifically, based on the fixed circumferential phase loss compensation feature sequence, the phase residual, response hysteresis, soot blowing recovery difference and adjacent circumferential phase difference corresponding to each thermal disturbance event are extracted in the order of thermal disturbance events, and the extraction results corresponding to each thermal disturbance event are determined as the event node set; Based on the rotor circumferential adjacency constraint, the adjacent circumferential phase difference in the event node set is limited to the adjacent circumferential phases on both sides of the candidate circumferential phase, and the event node set that satisfies the rotor circumferential adjacency constraint is determined as the circumferential adjacency node set. Based on the set of adjacent nodes in the circumferential direction and the consistency constraint of thermal disturbance events, nodes belonging to the same type of thermal disturbance events and corresponding in the order of thermal disturbance events are matched, and the connection relationship is established based on the combination distance of nodes in phase residual, response hysteresis, soot blowing recovery difference and adjacent circumferential phase difference, so as to obtain a fixed circumferential phase loss compensation relationship diagram. The connection relationships of the fixed circumferential phase loss compensation relationship diagram are expanded step by step according to the combination distance from small to large, and the persistent cohomology calculation is performed on each level of connection relationship to obtain the appearance order and disappearance order of the connected components corresponding to each level of connection relationship, thus forming the persistent lifetime; The longevity is compared with the lifetime threshold. If the comparison conditions are met, a continuous verification update result is generated. If the comparison conditions are not met, a verification intensity reduction update result is generated, resulting in a reflection record update result. Write the updated reflection record results into the corresponding candidate circumferential phase record position in the reflection record, and adjust the verification state of the corresponding candidate circumferential phase according to the updated reflection record results to obtain the updated reflection record.

[0032] When establishing connections based on the set of circumferential adjacent nodes and the consistency constraints of thermal disturbance events, connections are established for nodes that belong to the same type of thermal disturbance events, have corresponding thermal disturbance event sequences, and whose phase residuals, response hysteresis, soot blowing recovery differences, and adjacent circumferential phase difference change directions are consistent. When writing the comparison results of the lifetime and lifetime threshold into the reflection record, the lifetime corresponding to the connected component with the most thermal disturbance events in the fixed circumferential phase loss compensation relationship graph is selected for comparison with the lifetime threshold, and the comparison results are written into the record position of the corresponding candidate circumferential phase in the reflection record.

[0033] S5: As Figure 7 As shown, the candidate set of fixed circumferential phases is iteratively verified based on the updated reflection record, and candidate circumferential phases that satisfy the fixed circumferential phase stagnation hypothesis are selected to obtain the uncompensated phase set. Specifically: Based on the updated reflection records, extract the record values ​​corresponding to each candidate circumferential phase in the fixed circumferential phase candidate set, and sort each candidate circumferential phase according to the record values ​​from high to low to form a candidate circumferential phase verification sequence; According to the candidate circumferential phase verification sequence, the verification results of the fixed circumferential phase stagnation hypothesis corresponding to each candidate circumferential phase in different thermal disturbance events are iteratively summarized, and the candidate circumferential phase verification results are formed based on the consistency between the iterative summary results and the updated reflection records. The verification results of the candidate circumferential phases are compared with the preset threshold. If the comparison conditions are met, the corresponding candidate circumferential phases are determined to satisfy the fixed circumferential phase stagnation assumption. If the comparison conditions are not met, the corresponding candidate circumferential phases are eliminated to form a set of effective candidate circumferential phases. The effective candidate circumferential phase set is filtered from high to low according to the verification results of the candidate circumferential phase to obtain the uncompensated phase set.

[0034] S6: As Figure 8 As shown, based on the uncompensated phase set and the fixed circumferential phase uncompensated feature sequence, the level of insufficient local effective compensation margin is calculated, and the local uncompensated discrimination result is output.

[0035] Specifically, based on the uncompensated phase set, the phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase difference of each uncompensated phase in each thermal disturbance event are extracted from the fixed circumferential phase uncompensated feature sequence. The extraction results are then reorganized into a phase level calculation sequence that maintains the correspondence between events according to the order of uncompensated phases and thermal disturbance events. Based on the phase level calculation sequence, the phase residual and adjacent circumferential phase difference of the same uncompensated phase in each thermal disturbance event are combined and calculated, and the response hysteresis and soot blowing recovery difference are combined and calculated to form the margin calculation value corresponding to each uncompensated phase. Based on the margin calculation value, the margin calculation values ​​of each loss compensation phase in continuous thermal disturbance events are accumulated, and the maximum duration segment in the accumulated results is extracted to form the local effective compensation margin insufficient value. The local effective compensation margin deficiency value is compared with the preset level threshold to determine the local effective compensation margin deficiency level. The set of uncompensated phases is then associated with the local effective compensation margin deficiency level in a one-to-one correspondence according to the uncompensated phases to obtain the local uncompensated discrimination result.

[0036] After the candidate set of fixed circumferential phases enters the reflective agent, it is first organized into a fixed circumferential phase uncompensated feature sequence. The criterion for local jamming is transformed from scattered circumferential phase data fragments into an intermediate object with fixed circumferential phase position, thermal disturbance event order, and adjacent differences. The fixed circumferential phase uncompensated feature sequence is further used to form a fixed circumferential phase uncompensated relationship graph under rotor circumferential adjacency constraints and thermal disturbance event consistency constraints. The persistent lifetime is obtained through persistent cohomology calculation and then written into the updated reflective record, so that the updating basis of the fixed circumferential phase jamming hypothesis is transformed into the degree of persistence of the fixed circumferential phase. This forms a continuous processing path of "fixed circumferential phase candidate set - fixed circumferential phase uncompensated feature sequence - fixed circumferential phase uncompensated relationship graph - updated reflective record - uncompensated phase set". The entire path revolves around the same scenario mechanism in which weak signs of local jamming under high viscous dust conditions are easily masked by the disturbance of the whole machine.

[0037] like Figures 9 to 11As shown, the model receives 16 candidate circumferential phases from a fixed circumferential phase candidate set at a time, along with the corresponding circumferential phase data segments. Each candidate circumferential phase is organized according to 8 thermal disturbance events, with 32 sampling points extracted from each thermal disturbance event. Each sampling point contains 6 channels of data, including temperature, pressure difference, oxygen content, current, load, and soot blowing event data. Therefore, the original input dimension of a single candidate circumferential phase is 8×32×6. The reflective agent first calculates a fixed circumferential phase loss-compensation feature sequence based on the circumferential phase data segments of the target circumferential phase and adjacent circumferential phases. Among them, the phase residual consists of 4 dimensions composed of the differences in temperature, pressure difference, oxygen content, and current relative to the load reference; the response hysteresis consists of 2 dimensions composed of the abnormal initiation hysteresis and abnormal recovery hysteresis; the soot blowing recovery difference consists of 4 dimensions composed of the differences in temperature, pressure difference, oxygen content, and current before and after soot blowing; and the adjacent circumferential phase difference consists of 4 dimensions composed of the differences in temperature, pressure difference, oxygen content, and current between the target circumferential phase and adjacent circumferential phases. Therefore, each thermal disturbance event obtains 14-dimensional features, and 8 thermal disturbance events constitute an 8×14 fixed circumferential phase loss-compensation feature sequence.

[0038] The reflective agent employs a single-backbone structure with shared parameters. The input encoding layer performs 1D convolution on an 8×14 fixed circumferential phase miscompensation feature sequence. The first convolutional layer has 32 kernels, a convolution length of 3, 14 input channels, and an output dimension of 8×32. The second convolutional layer has 64 kernels, a convolution length of 3, 32 input channels, and an output dimension of 8×64. Each convolutional neuron in both layers is connected to all input channels at the locations of three adjacent thermal perturbation events and includes weights, biases, and ReLU nonlinear units to compress the recurring local miscompensation patterns across thermal perturbation events into an event sequence representation. The 8×64 event sequence representation is averaged along the direction of the thermal perturbation events to obtain a 64-dimensional candidate circumferential phase representation. This 64-dimensional candidate circumferential phase representation is then fed into a fully connected layer with 32 neurons to obtain a 32-dimensional fixed circumferential phase hysteresis hypothesis representation.

[0039] The scenario-based transformation of the reflexive agent occurs in the input encoding and reflex update stages. The input encoding stage does not directly receive overall system anomalies; instead, it receives an 8×14 fixed circumferential phase uncompensated feature sequence constructed around a candidate set of fixed circumferential phases. This ensures the input object simultaneously contains three types of information: the target circumferential phase, adjacent circumferential phases, and the sequence of thermal disturbance events. The convolutional layer extracts the continuous shift relationship of the fixed circumferential phase through multiple thermal state changes along the thermal disturbance event sequence. The reflex update stage sets a 32-dimensional reflex record for each candidate circumferential phase. Each 32-dimensional reflex record corresponds one-to-one with a 32-dimensional fixed circumferential phase stagnation hypothesis representation. The 32 components record the cumulative state of the fixed circumferential phase stagnation hypothesis during repeated verification, transforming the updated object of the reflex record from a single anomaly description to a continuous uncompensated state of the fixed circumferential phase.

[0040] Persistent cohomology is embedded within the reflection update process. For each candidate circumferential phase, a fixed circumferential phase miscompensation relationship graph is first established based on an 8×14 fixed circumferential phase miscompensation feature sequence. This graph uses eight thermal disturbance event nodes as basic nodes, each with a feature dimension of 14. Under rotor circumferential adjacency constraints and thermal disturbance event consistency constraints, edge connections are generated based on the combined distance of phase residuals, response hysteresis, soot blowing recovery differences, and adjacent circumferential phase differences, resulting in an 8×8 adjacency matrix. Persistent cohomology performs connected component lifetime calculations on the 8×8 adjacency matrix, outputting a 1-dimensional persistent lifetime. This 1-dimensional persistent lifetime is concatenated with the 64-dimensional candidate circumferential phase representation and the 32-dimensional reflection record to form a 97-dimensional update input, which is then input into the reflection update layer. The reflection update layer employs a two-layer fully connected structure: the first layer has 64 neurons, and the second layer has 32 neurons. Each neuron is fully connected to all neurons in the previous layer and includes weight, bias, and ReLU nonlinear units, outputting the updated reflection record. In this way, the basis for updating the reflection record changes from the amplitude of a single thermal disturbance event to the degree of persistence of the fixed circumferential phase loss compensation relationship diagram in the relationship scale changes.

[0041] The output layer receives the updated reflection record and a feature summary of the fixed circumferential phase uncompensated feature sequence. The fixed circumferential phase uncompensated feature sequence is averaged along the direction of the thermal disturbance event to obtain a 14-dimensional feature summary. This 14-dimensional feature summary is concatenated with the 32-dimensional updated reflection record to form a 46-dimensional output-input. The 46-dimensional output-input is then passed sequentially through a fully connected layer of 16 neurons and a result layer of 4 neurons to form a 4-dimensional result vector for each candidate circumferential phase. In a feasible fixed example, the local effective compensation margin deficiency level is set to level 3. Therefore, the first dimension of the 4-dimensional result vector is used to form the uncompensated phase set, and the second to fourth dimensions are used to form the local effective compensation margin deficiency level. The 16 candidate circumferential phases correspond to a 16×4 output matrix. Candidate circumferential phases that meet the screening criteria constitute the uncompensated phase set, and the level with the largest value in the second to fourth dimensions corresponds to the local uncompensated discrimination result.

[0042] The fixed circumferential phase miscompensation feature sequence simultaneously drives the input encoding layer and the fixed circumferential phase miscompensation relation graph, forming a single backbone from the circumferential phase data fragment to the updated reflection record. Persistent cohomology does not directly process the whole-machine data, nor does it output local miscompensation discrimination results separately. The 1-permanent lifetime serves as a constraint within the reflection update layer, participating in the 32-dimensional reflection record update. This ensures that the miscompensated phase set and the level of insufficient local effective compensation margin formed by the output layer are always generated around the fixed circumferential phase stagnation assumption.

[0043] The present invention will be further explained and illustrated below with reference to specific embodiments.

[0044] In this embodiment, step S1 specifically includes: The air preheater generates a sequence of original synchronization records during continuous operation cycles. . Indicates a record index; Indicates the total number of records; ; Represents a timestamp; Indicates rotor angular position information; , , , , , These represent data for temperature, pressure difference, oxygen content, current, load, and soot blowing events, respectively. First, press... The original synchronous record sequence is sorted in ascending order, and then a rotor circumferential cycle is divided into... A fixed circumferential phase center ; Indicates the circumferential phase index; Indicates the total number of circumferential phases; Indicates the first The angular center of each circumferential phase. For each original synchronization record, the following is used: Circumferential phase mapping is completed, with the angle difference calculated based on the shortest circumferential difference within a rotor circumferential cycle. If multiple original synchronization records within the same rotor circumferential cycle are mapped to the same circumferential phase, the original synchronization record with the smallest angle difference is retained. If a circumferential phase is missing within the same rotor circumferential cycle, the corresponding rotor circumferential cycle is not included in the thermal disturbance event interception. After mapping, the records are rearranged according to chronological order and circumferential phase order to obtain... ; This indicates the synchronous sampling index that expands cyclically along the rotor circumference; Indicates the synchronous sampling index Position circumferential phase The corresponding six-channel record.

[0045] The thermal disturbance event refers to an operational process triggered by load changes or soot blowing events, resulting in continuous deviations in temperature or pressure differential. Thermal disturbance event identification is performed on the synchronous sampling index axis. The above will be performed. First, the load sequence will be analyzed. Perform adjacent differences, and within a length of The average value within the sliding window is used to obtain the load change slope sequence; This indicates the length of the calculation window for the load change slope. Then, the data from the soot blowing event... State transition detection is performed, and the position where the soot blowing event data changes from a non-triggered state to a triggered state is determined as the soot blowing event trigger marker. For each candidate trigger position... The length before the candidate trigger position is Using the reference window as a benchmark, the average temperature and average pressure difference within the reference window are calculated respectively; Indicates the length of the reference window. Then in The temperature and pressure difference sequences of all circumferential phases are checked within the preset time window; This indicates the length of the preset time window. If the load change slope reaches the preset load change slope threshold or the soot blowing event trigger flag is set, and at least one circumferential phase satisfies the condition that the temperature offset relative to the average temperature of the reference window continuously exceeds the preset temperature offset threshold, or the pressure offset relative to the average pressure offset of the reference window continuously exceeds the preset pressure offset threshold, and the offset direction remains consistent, then... The synchronous sampling index corresponding to the start time of the thermal disturbance event is determined. Searching backwards from the synchronous sampling index corresponding to the start time of the thermal disturbance event, when the temperature offset and pressure difference offset of all circumferential phases are continuous... When a synchronous sampling position returns to within the preset recovery threshold range, the position that first meets the condition is determined as the synchronous sampling index corresponding to the end time of the thermal disturbance event. All thermal disturbance events are sorted in ascending order of their start times to obtain... ; Indicates the index of thermal disturbance events; Indicates the first The start time of each thermal disturbance event corresponds to the synchronous sampling index; Indicates the first The end time of each thermal disturbance event corresponds to the synchronous sampling index.

[0046] For each thermal disturbance event Set the truncation length before the thermal disturbance event occurs. Length after thermal disturbance event recovery ; Indicates the length of the continuous data truncation prior to the occurrence of a thermal disturbance event; This indicates the truncation length of continuous data after recovery from the thermal disturbance event. For each circumferential phase... In the synchronous sampling index range Extract six consecutive channels of data. Circumferential phase. thermal disturbance events The corresponding truncation result is written as ; This indicates the discrete sampling index within the selected range. Synchronous sampling index interval. Maintaining consistency across all circumferential phases within the same thermal disturbance event ensures that continuous data before, during, and after the event share the same time boundaries across all circumferential phases. If ,or If the effective synchronization sampling index is exceeded, a thermal disturbance event will occur. It does not participate in the construction of circumferential phase data segments.

[0047] To ensure that the circumferential phase data segments corresponding to different thermal disturbance events have a uniform length, each truncation range is mapped to a unified discrete index axis. The continuous running cycle is selected to contain no less than [number missing] [units missing]. The operating range of a complete thermal disturbance event is determined, and the intercept range corresponding to each thermal disturbance event is uniformly selected between the start and end times of the interception. Each target sampling time. The resampling time step is denoted as... ; The fixed time interval between two adjacent target sampling times is represented by ; the allowed matching time window is denoted as . For discrete target locations Within the original truncation range, the time difference between the sampling time and the target sampling time is selected and denoted as . The original record; This represents the time difference between the original record timestamp and the target sampling time. If Then the original record with the smallest absolute time difference will be filled in. If discrete positions There is no satisfying The original record of the thermal disturbance event It does not participate in the construction of circumferential phase data segments. The data segments are selected in ascending order of their start times. A thermal perturbation event that meets the matching conditions, so the circumferential phase data segment corresponding to any circumferential phase is organized as follows: . for dimensional vector, The original input dimension is The circumferential phase data segment refers to a data segment covering a continuous time range before, during, and after the thermal disturbance event, under the same thermal disturbance event; the circumferential phase data segment is directly generated from... and express.

[0048] In this embodiment, step S2 specifically includes: Calling circumferential phase data fragments ,in Indicates the circumferential phase index. , Indicates the total number of circumferential phases; Indicates the thermal disturbance event index, in ; This represents the index of the sampling point within a segment of a thermal disturbance event. First, index the same thermal disturbance event. The load channels of all circumferential phases are averaged according to their circumferential phases to form a segmented load sequence; then the same thermal disturbance event is indexed. The soot blowing event data for all circumferential phases are taken as the maximum value according to the circumferential phase to form a segmented soot blowing sequence. The segmented load sequence is then divided into segments of length [missing information]. The load change slope is obtained by performing adjacent differential averaging within a sliding window; the position where the segment of the soot blowing sequence changes from a non-triggered state to a triggered state is determined as the soot blowing event trigger mark. For each thermal disturbance event segment, first take... The average temperature and average pressure difference are used as the reference values ​​for the segment, and then the length is... Within a preset time window, the continuous offset of temperature or pressure difference relative to the segment reference value in each circumferential phase is checked. The judgment rule for continuous offset is: on the same circumferential phase, the offset direction of all sampling points within the preset time window is consistent, and the absolute offset is not lower than the corresponding preset threshold. When the load change slope meets the load change condition, or the soot blowing event trigger mark meets the soot blowing event trigger condition, and at least one circumferential phase meets the continuous offset judgment rule, the position where the condition is first met is determined as the index corresponding to the start time of the thermal disturbance event. When temperature offset and pressure difference offset are continuous When a sampling point returns to the recovery threshold range, the position that first meets the condition is determined as the index corresponding to the end time of the thermal disturbance event. The thermal disturbance event is denoted as... ; This is used to distinguish between thermal disturbance events caused by load change conditions and thermal disturbance events caused by soot blowing event triggering conditions; when both types of conditions are simultaneously met for the first time at the same sampling point, the thermal disturbance event caused by the soot blowing event triggering conditions is written into the database. .

[0049] In determining , and Then, the start times of all circumferential phase data segments within the same thermal disturbance event are aligned. The aligned sampling point indices are denoted as... , The aligned discrete sequence is denoted as Alignment uses index remapping. This ensures that the start time of thermal disturbance events falls uniformly within the same timeframe. When the original index is mapped to Exceeding The range, or insufficient effective sampling points. At that time, thermal disturbance event index It is not included in the repetition rate calculation. After alignment, it will be... Determined as the benchmark section, The response segment is identified, and the index of the end time of the aligned thermal disturbance event is recorded as... , According to the thermal disturbance event identifier Alignment position at start time The reference segment and response segment are organized to obtain the event comparison segment. All circumferential phases in the event comparison segment share the same starting time position and the same segment boundary.

[0050] Within each event control segment, the baseline mean values ​​of temperature, pressure difference, oxygen content, and current are calculated for the baseline segment, and then anomaly vectors are constructed point-by-point in the response segment. . This indicates the deviation of the temperature from the average temperature reference. , , These represent the offsets of pressure difference, oxygen content, and current relative to their respective reference averages. The anomaly vector sequence corresponding to a single thermal disturbance event is formed by... Each sampling point and It consists of several channels, organized in the form of... ; same circumferential phase in The sequence of anomalous vectors on each thermal disturbance event is arranged in the order of the thermal disturbance events, forming a circumferential phase anomaly trajectory, organized as follows: Repeated occurrences are recorded as points. .calculate At that time, only in thermal disturbance event identifiers Matching identical thermal disturbance events. Indexing the target thermal disturbance events. Index of comparable thermal perturbation events First, compare the signs of the average offsets of the four channels in the response segment to determine the consistency of the anomaly direction; then record the sampling point indices where the four channels first exceed their respective offset thresholds, and form a starting order in chronological order to determine the consistency of the anomaly initiation sequence; then extract the length of the longest continuous interval where the absolute value of the four channel offsets is continuously higher than the corresponding offset threshold to determine the consistency of the anomaly duration segment; then from... Start by taking a length of The recovery window is used to check whether the four-channel offset continuously regresses towards zero, which is used to determine the consistency of the recovery trend. The four consistency judgment results are written as zero or one respectively, and then the average of the four judgment results is taken to form a thermal disturbance event index. Index of thermal disturbance events The matching values ​​are averaged across all comparable thermal disturbance events to form a circumferential phase index. Index of thermal disturbance events Repeated score .

[0051] In obtaining Then, the local difference between the target circumferential phase and the adjacent circumferential phases is introduced to adjust the scores for repeated occurrences. The index of the adjacent circumferential phase on the left is denoted as... The adjacent circumferential phase index on the right is denoted as The beginning and end circumferential phases are connected in a circular circumferential direction. The circumferential phase indices are then... Anomaly vector sequence and circumferential phase index and circumferential phase index Index of the same thermal disturbance event The anomaly vector sequence is compared point by point in the response segment. Calculate the average absolute values ​​of the temperature difference, pressure difference, oxygen content difference, and current difference, respectively. If the four channels are in circumferential phase index... If the average absolute value between the phase and at least one adjacent circumferential phase exceeds the corresponding adjacent difference threshold, then... Increase the preset adjustment amount If the four channels are in the circumferential phase index If the average absolute value between the phase and the adjacent circumferential phases on both sides does not exceed the corresponding adjacent difference threshold, then... Lower the preset adjustment amount In other cases, remain unchanged. Unchanged. The adjusted result is denoted as... and will Limited to between zero and one. Index the same circumferential phase. of Arranged in sequence according to thermal disturbance events, a fixed circumferential phase recurrence result is formed.

[0052] Let the set of thermal disturbance events that meet the load change conditions be denoted as . The set of thermal disturbance events indexed by the conditions for triggering the soot blowing event is denoted as . For each circumferential phase index Statistical analysis Not lower than the preset threshold Quantity and Not lower than the preset threshold Quantity, then calculate separately and Inside The average value. Only when The corresponding quantity meets the thermal disturbance event coverage threshold. The corresponding quantity meets the thermal disturbance event coverage threshold, and Inner mean and When the internal average values ​​all meet the preset threshold, the circumferential phase index Only then was it preserved. The entire circumferential phase will be preserved. Take the average to form the sorted values. Press again Sort from highest to lowest. Then sort the results by number of items at the top. Each circumferential phase index is denoted as , and each Corresponding raw circumferential phase data segment All are retained. Thus, the input to the reflective agent is organized as follows: Candidate circumferential phases in a fixed circumferential phase candidate set, and Each group corresponds one-to-one with a candidate circumferential phase data segment; the circumferential phase data segment corresponding to each candidate circumferential phase is maintained. The input organization format.

[0053] In this embodiment, step S3 specifically includes: One reception of a fixed circumferential phase candidate set There are 1 candidate circumferential phases, and the index of the candidate circumferential phase is denoted as . ,in Each candidate circumferential phase corresponds to Each thermal disturbance event and within each thermal disturbance event A circumferential phase data segment at each sampling point is denoted as... ,in , Circumferential phase The degree of repetition in each thermal disturbance event is denoted as . and will and Together written into the verification unit Proof unit The object of verification is used to determine the target circumferential phase, the sequence of thermal disturbance events, and the fixed circumferential phase stagnation hypothesis. The reflective agent revolves around the verification unit. Circumferential phase Establish a fixed circumferential phase lag assumption, and limit the fixed circumferential phase lag assumption to the circumferential phase. The hypothesis of repeatedly exhibiting local anomalous shifts in multiple thermal disturbance events needs to be verified.

[0054] Determine the circumferential phase according to the rotor circumferential sequence. The left adjacent circumferential phase index and the right adjacent circumferential phase index are denoted as follows: and The beginning and end circumferential phases are processed as a closed loop. Then, data is synchronously retrieved from the circumferential phase data segment. , and and will Recorded as a control unit The order of thermal disturbance events is determined using a thermal disturbance event index. The increasing order, corresponding to the unit Event matching and time alignment are performed under the same thermal disturbance event index and the same sampling point index. After alignment, Determined as the benchmark section, The region was identified as a thermal disturbance event change zone. The area was identified as the recovery zone for the thermal disturbance event. This indicates the alignment index corresponding to the end time of the thermal disturbance event. Thus, the target circumferential phase and the adjacent circumferential phases on both sides within the same time range of the same thermal disturbance event are unified onto the same index axis, and an event sequence comparison result is formed according to the baseline segment, the thermal disturbance event change segment, and the thermal disturbance event recovery segment.

[0055] In the formation of phase residuals, the change in the target circumferential phase relative to the reference phase in the thermal disturbance event variation segment is first calculated for each of the four channels: temperature, pressure difference, oxygen content, and current. This change is then correlated with the load change in the same thermal disturbance event and compared with the corresponding changes in the adjacent circumferential phases in the same segment. To ensure that the phase residuals simultaneously reflect the local shift of the target circumferential phase, the load change in the same thermal disturbance event, and the difference compared to the adjacent circumferential phases, the phase residual components of the four channels (temperature, pressure difference, oxygen content, and current) are calculated using the following formula:

[0056] In the formula, Indicates candidate circumferential phase Index of thermal disturbance events and channels The phase residual components below; Indicates the index of the candidate circumferential phase in the fixed circumferential phase candidate set; Indicates the index of thermal disturbance events; These indicate the channel identifiers, corresponding to temperature, pressure difference, oxygen content, and current, respectively. Indicates the target's circumferential phase index; Indicates the adjacent circumferential phase index on the left; Indicates the adjacent circumferential phase index on the right; Indicates in The adjacent circumferential phase index from which the value is taken; Indicates the sampling point index; This represents the alignment index corresponding to the end time of the thermal disturbance event; Indicates the target's circumferential phase in the channel The sampled value on, when Time to take ,when Time to take ,when Time to take ,when Time to take ; Indicates adjacent circumferential phases in the channel The sampled values ​​on; This represents the sampled value of the target circumferential phase in the load channel; This indicates a preset positive number, used to avoid the denominator being zero; Represents absolute value; This indicates that sampled values ​​within a given index range are accumulated; Corresponding to the benchmark section, Corresponding to the changing segment of the thermal disturbance event. According to... Arranged in order ,get 3D phase residual .

[0057] In the extraction of response lag, the determination length of continuous unidirectional change is denoted as... , This indicates the number of consecutive sampling points used to determine the onset and recovery times of the anomaly. First, the offsets of the four channels—temperature, pressure difference, oxygen content, and current—relative to their respective baseline averages are scanned point-by-point within the thermal disturbance event's variation range. When at least three of the four channels are in continuous... When all sampling points maintain the same offset direction and the absolute offset is not lower than their respective offset thresholds, the first sampling point that meets the conditions is determined as the index of the starting time of the target circumferential phase anomaly. Then, the offsets of the four channels are scanned point by point within the thermal disturbance event recovery section. When at least three of the four channels are in the same offset direction, the offset is determined. When the absolute offset continuously decreases as the sampling points regress towards zero, the last sampling point that meets the condition is determined as the target circumferential phase anomaly recovery time index. The corresponding indices of the left and right adjacent circumferential phases are extracted according to the same rules. When the target circumferential phase anomaly start time index is later than the left and right adjacent circumferential phase anomaly start time indices, the target circumferential phase anomaly start time index is relative to the thermal disturbance event start index. The offset is written as the abnormal start lag; when the target circumferential phase anomaly recovery time index is later than the left adjacent circumferential phase anomaly recovery time index, the right adjacent circumferential phase anomaly recovery time index, and later than At that time, the index of the target circumferential phase anomaly recovery time is relative to The offset is written as the anomaly recovery lag; the anomaly initiation lag and the anomaly recovery lag are combined into a single unit. Dimensional response lag .

[0058] In the extraction of soot blowing recovery difference, the target circumferential phase soot blowing event data is used. The index of the first sampling point that changes from a non-triggered state to a triggered state is denoted as . The length of the comparison window before and after the dust-blowing event is denoted as... , This indicates the number of consecutive sampling points used to calculate the difference in mean before and after the location where the dust blowing event was triggered. If it exists, take The length of the front is Comparison window and The length after is The comparison windows are used to calculate the mean differences of temperature, pressure difference, oxygen content, and current within the two comparison windows, respectively, to obtain... Weiss blowdown recovery difference ;when If it does not exist, Set to zero vector. Recover the difference by blowing away soot. The four components correspond to the changes in temperature, pressure difference, oxygen content, and current before and after the soot blowing event.

[0059] In extracting the phase difference between adjacent circumferential directions, the temperature, pressure, oxygen content, and current differences between the target circumferential phase and its adjacent left and right circumferential phases within the thermal disturbance event change zone are calculated first. Then, these differences are calculated again within the thermal disturbance event recovery zone. The left and right comparison results are retained in circumferential order. The comparison values ​​for the same channel within the thermal disturbance event change and recovery zones are averaged to form the left four-channel comparison values ​​and the right four-channel comparison values. Finally, the comparison value with the larger absolute value from the left and right four-channel comparison values ​​is selected by channel to form the... Dimensional adjacent circumferential phase difference Adjacent circumferential phase difference The four components correspond to the local difference boundaries between the target circumferential phase and the adjacent circumferential phases on both sides, respectively, for temperature, pressure difference, oxygen content, and current.

[0060] Will , , and Concatenate them sequentially to form an event feature vector Event feature vector The specific characteristic is composed of phase residuals Maintenance, response lag Difference in recovery between maintenance and soot blowing Maintaining adjacent circumferential phase difference Dimension, total dimension is .Will Candidate circumferential phases are obtained by sequentially arranging the thermal perturbation events. Fixed circumferential phase loss compensation characteristic sequence Fixed circumferential phase loss compensation characteristic sequence The organizational form is , of which Dimensional correspondence The first thermal disturbance event, the... Dimension corresponds to a single thermal perturbation event Each characteristic component. Proof unit. Responsible for determining the processing objects and the order of events, referring to the unit. Responsible for providing comparison data of the target circumferential phase and adjacent circumferential phases within the same thermal disturbance event, and fixing the circumferential phase loss-compensation characteristic sequence. It serves as the direct input to the input encoding layer of the reflective agent.

[0061] The reflexive agent employs a single-backbone structure with shared parameters. The input encoding layer uses a fixed circumferential phase-compensated feature sequence. As input, the first convolutional layer is set There are n convolutional kernels, and the convolution length is 1. The input channel is The output dimension is Second convolutional layer settings There are n convolutional kernels, and the convolution length is 1. The input channel is The output dimension is Each convolutional neuron in the first and second convolutional layers is connected to its neighboring neurons. All input channels at the location of each thermal perturbation event are connected, and weights, biases, and ReLU nonlinear units are included. The event sequence is represented by averaging along the direction of the thermal perturbation events, resulting in... Dimensional candidate circumferential phase representation Then Input contains A fully connected layer of neurons, containing weights, biases, and ReLU nonlinear units, yields... The fixed circumferential phase hysteresis hypothesis represents Reflecting on intelligent agents Each candidate circumferential phase undergoes the same processing procedure in parallel, forming a fixed circumferential phase loss-compensation feature sequence that corresponds one-to-one with each candidate circumferential phase. , Dimensional candidate circumferential phase representation and The fixed circumferential phase hysteresis hypothesis represents .

[0062] In this embodiment, step S4 specifically includes: For candidate circumferential phase Calling the fixed circumferential phase loss compensation feature sequence Candidate circumferential phase representation and reflection records . This represents the candidate circumferential phase index. Indicates candidate circumferential phase Index of thermal disturbance events The event feature vector below, From phase residual Response lag Soot blowing recovery difference Phase difference with adjacent circumferential directions Composition, the dimension of a single event feature vector is , A series of thermal disturbance events are arranged in sequence to form Fixed circumferential phase loss compensation characteristic sequence . This represents the output of the input encoding layer from the reflective agent. Dimensional candidate circumferential phase representation. Indicates candidate circumferential phase Before the current round enters the reflection and update phase Dimensional reflection record. Candidate circumferential phase. When entering the reflection and update phase for the first time, Initialize to Zero-dimensional vector; candidate circumferential phase When entering the reflection and update phase again, the data written in the previous round is directly called. Reflection Record as .

[0063] Extracted according to the order of thermal disturbance events The phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase difference corresponding to each thermal disturbance event are recorded, and the thermal disturbance events are identified. Index of the order of occurrence within similar thermal disturbance events Write node tags to obtain event nodes . This indicates a thermal disturbance event. This represents the sequential index of thermal disturbance events within the same category, ordered by their order of occurrence. Constitute candidate circumferential phase The set of event nodes. To satisfy the rotor circumferential adjacency constraint, the phase difference between adjacent circumferential directions is extracted. Simultaneous writing of adjacent circumferential phase source markers . Indicates candidate circumferential phase The left adjacent circumferential phase index, Indicates candidate circumferential phase The right-hand adjacent circumferential phase index. Only retain the adjacent circumferential phase source marker and... Consistent event nodes are represented as the set of circumferentially adjacent nodes. Candidate circumferential phases. of All thermal disturbance event nodes satisfy the rotor circumferential adjacency constraint, therefore the set of circumferential adjacency nodes includes Each node.

[0064] Within the circumferentially adjacent node set, nodes belonging to the same type of thermal disturbance event are matched. Matching is first performed based on the thermal disturbance event identifier. Grouping, and then maintaining the order index within the same group. The original order. The consistency constraint for thermal disturbance events is achieved by ensuring that the thermal disturbance event identifiers are the same and the difference in their occurrence order indices does not exceed a preset order difference threshold. For any two nodes that satisfy the thermal disturbance event consistency constraint... and First, calculate the average value of four sets of features: phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase difference. Then, determine the sign of the group average value based on whether it is greater than zero, less than zero, or equal to zero. When the signs of the group average values ​​of all four sets of features are consistent, it is determined that the node change direction is consistent, and the calculation of the combined distance begins. When any of the four sets of features has inconsistent directions, the node pair is directly... Mark as a pair of unconnected nodes.

[0065] The combined distance is denoted as The combined distance is based on four sets of numerical components: phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase difference. It also incorporates the occurrence sequence difference within similar thermal disturbance events, ensuring that the connectivity is simultaneously constrained by local uncompensated amplitude differences and the comparable order of thermal disturbance events. (Node) With nodes The combined distance between them is calculated using the following formula:

[0066] In the formula, Indicates candidate circumferential phase The The event node and the first The combined distance between event nodes; Indicates the candidate circumferential phase index; and Indicates the index of thermal disturbance events; , , , , These represent the preset weights corresponding to phase residual, response hysteresis, soot blowing recovery difference, adjacent circumferential phase difference, and occurrence sequence difference, respectively. The five preset weights are written using a preset weight table. This represents the characteristic component index, phase residual, soot removal recovery difference, and adjacent circumferential phase difference. Response lag ; Represents phase residual The One component; Indicates a delayed response. The One component; Indicates the difference in soot blowing recovery. The One component; Indicates the phase difference between adjacent circumferential directions The One component; and This indicates the order of occurrence of a thermal disturbance event within the same category of thermal disturbance events; Indicates the phase residual of the first Normalization thresholds for each component with the same dimension; Indicates the response lag. Normalization thresholds for each component with the same dimension; Indicates the difference between the recovery value and the soot blowing value. Normalization thresholds for each component with the same dimension; Indicates the phase difference with the adjacent circumferential direction. Normalization thresholds for each component with the same dimension; The order normalization threshold represents the difference in occurrence order with the same dimensions. Represents absolute value; This indicates that the components within the specified index range are accumulated. , , , and All values ​​are read from a preset normalized threshold table, and each normalized threshold maintains the same dimension as its corresponding component; therefore, the combined distance... It is a dimensionless numerical value.

[0067] Node pairs that satisfy the rotor circumferential adjacency constraint, thermal disturbance event consistency constraint, and change direction consistency condition will be written into the fixed circumferential phase loss compensation relationship diagram. Fixed circumferential phase loss compensation relationship diagram by Each thermal perturbation event node is used as a graph node to combine distances. As a border right, forming Weighted adjacency matrix Write zero values ​​to the diagonal positions, and write a "no connection" flag to node pairs that do not meet the connection conditions. Then... All finite combination distances are sorted in ascending order to obtain a progressively expanding threshold sequence. Each threshold in the progressively expanding threshold sequence corresponds to a connection level, and each connection level generates a... The binary adjacency matrix is ​​used. Persistent homology performs connected component tracking sequentially on all binary adjacency matrices, recording the order in which each connected component first forms as its occurrence order, and recording the order in which each connected component merges with an earlier connected component and no longer exists independently as its disappearance order. When a connected component remains independent at the last threshold, the level corresponding to the last threshold is determined as the disappearance order. The persistent homology is then subtracted from the threshold corresponding to the occurrence order to obtain the lifetime of the connected component. The binary adjacency matrix sequence is used to perform connected component lifetime calculations and outputs the phase for each candidate circumferential phase. Maintain long lifespan .

[0068] To maintain consistency with the reflexive logging rules, the number of thermal perturbation events contained in each connected component is counted across all connected components, and the connected component with the highest number of thermal perturbation events is selected as a candidate connected component. When multiple connected components have the same number of thermal perturbation events, the connected component with the longer durability is selected as the candidate connected component. The durability corresponding to the candidate connected component is denoted as . The lifetime threshold is denoted as .when At that time, the results are continuously verified and updated; when At that time, a verification intensity reduction update result is generated. The continuous verification update result and the verification intensity reduction update result together constitute the reflection record update result, and are written into the candidate circumferential phase. The position of the status marker.

[0069] Reflection and update layer is recorded as , This indicates a reflex update of the layer parameters. Dimensional candidate circumferential phase representation , Maintain long lifespan and Reflection Record Assembled in a fixed order The update layer updates the input. The reflection update layer uses a two-layer fully connected structure; the first layer sets... One neuron, second layer setup There are 10 neurons, and each neuron in the first and second layers is fully connected to all neurons in the previous layer, and includes weights, biases, and ReLU nonlinear units. Dimensional update input is fed into the reflection update layer After that, I received Reflections after the update . of The component represents the fixed circumferential phase hysteresis assumption. Each component corresponds one-to-one with the others and is used to record the candidate circumferential phase. The cumulative state in repeated verification. Candidate circumferential phases written into the reflection record The record position is determined, and the results of continuous verification or the results of verification intensity reduction are written into the candidate circumferential phase. The state position needs to be verified. Each candidate circumferential phase is processed in parallel using the same processing path with shared parameters, resulting in... A diagram showing the relationship between fixed circumferential phase loss compensation. , Durability and An updated reflection record .

[0070] In this embodiment, step S5 specifically includes: Call the candidate circumferential phase index from the fixed circumferential phase candidate set Candidate circumferential phase Corresponding updated reflection record and candidate circumferential phase The corresponding fixed circumferential phase loss compensation feature sequence . This represents the candidate circumferential phase index in the fixed circumferential phase candidate set; Indicates candidate circumferential phase Reflecting on the output of the update layer Reflections after maintenance and renewal; Indicates candidate circumferential phase Index of thermal disturbance events The event feature vector below; ; The organizational form is Let's start with each one. Extract record values Record value Adopt the updated reflection record of The arithmetic mean of the components. Then, according to the recorded values. Sort all candidate circumferential phases from high to low to obtain the candidate circumferential phase verification sequence. . In the candidate circumferential phase proof sequence, the first... Candidate circumferential phase indices corresponding to each position; This represents the position index in the candidate circumferential phase proof sequence.

[0071] Proof sequence along candidate circumferential phase Perform iterative verification one by one. For any candidate circumferential phase... , will the The cumulative summary of the stages obtained at the end of the round of iteration is denoted as ; ; Indicates candidate circumferential phase Before inclusion Following a thermal disturbance event Dimensional phase cumulative summary. During initialization, [the following will be done]: Take directly as .from To begin, first summarize the cumulative data from the previous stage. Each component is multiplied by the number of thermal disturbance events already included. Then, compare it with the event feature vector corresponding to the current thermal disturbance event. The corresponding components are added together, and then divided by the new number of thermal disturbance events. A new stage cumulative summary is obtained. .when At that time, Recorded as . Indicates candidate circumferential phase exist On a thermal disturbance event Dimensional feature summary. The former The round-by-round summary results of the phase residuals corresponding to each component, in the middle Each component corresponds to a round-by-round summary result with a response lag, and then... The results of summarizing the recovery differences of each component in each round are as follows: Each component corresponds to the round-by-round summation of the adjacent circumferential phase differences. The verification results of the fixed circumferential phase stagnation hypothesis in different thermal disturbance events are obtained through... The round-by-round update process is completed and summarized, therefore Directly characterize candidate circumferential phase The cumulative verification status in multiple thermal disturbance events.

[0072] The output layer is denoted as , Indicates the output layer parameters. Dimensional Feature Summary With candidate circumferential phase corresponding Reflections after the update Assembled in a fixed order to form Dimensional input / output . Indicates candidate circumferential phase Single inference input in the output layer; The former Each component comes from ,back Each component comes from . of The component represents the fixed circumferential phase hysteresis assumption. Each component corresponds one-to-one and is used to record the cumulative state of the fixed circumferential phase lag hypothesis during repeated verification. Output layer It adopts a two-layer forward structure, the first layer being... The first layer is a fully connected layer with 1 neuron, and the second layer is a fully connected layer with 1 neuron. The output layer consists of several neurons; each neuron in the first and second layers is fully connected to all neurons in the previous layer and includes weights, biases, and ReLU linear rectified nonlinear units. The output layer is calculated based on the candidate circumferential phase verification sequence. Forward inference is performed one by one, and hidden states are not passed between candidate circumferential phases. Output layer parameters In all The inference process of each candidate circumferential phase remains unchanged. Dimensional input / output Input / output layer After that, I received Dimensional result vector . This indicates the proof result of the candidate circumferential phase; , and This represents the three level components corresponding to the level of insufficient local effective compensation margin. Because the output layer reads simultaneously... and Candidate circumferential phase verification results That is, it is obtained by the joint forward discrimination of the iterative summary results and the updated reflection record.

[0073] Proof results of candidate circumferential phase With preset threshold Compare them. This represents the minimum threshold value required to satisfy the fixed circumferential phase lag assumption for a candidate circumferential phase. When At that time, the candidate circumferential phase Retained as a valid candidate circumferential phase; when At that time, the candidate circumferential phase Eliminate candidates from the fixed circumferential phase candidate set. Prove the candidate circumferential phases according to the verification sequence. Complete all After forward reasoning with each candidate circumferential phase, all valid candidate circumferential phases are obtained. The results are then verified based on these candidate circumferential phases. All valid candidate circumferential phases are filtered from high to low to obtain the uncompensated phase set. The model input for a single candidate circumferential phase in this step is... Dimensional input / output The model output is Dimensional result vector ;all Each candidate circumferential phase is verified according to the candidate circumferential phase verification sequence. The reasoning is completed one by one, and the corresponding candidate circumferential phase proof results are generated respectively. .

[0074] In this embodiment, step S6 specifically includes: Let the set of uncompensated phases be denoted as . Represents the uncompensated phase set, Indicates the first One phase of loss compensation, Indicates the uncompensated phase index. This represents the total number of uncompensated phases. For each uncompensated phase... Calling the fixed circumferential phase loss compensation feature sequence . Indicates uncompensated phase The corresponding fixed circumferential phase loss compensation feature sequence, Indicates uncompensated phase Index of thermal disturbance events The event feature vector below, Indicates the index of thermal disturbance events. Indicates uncompensated phase The corresponding number of thermal disturbance events, in This step directly calculates the levels of the values ​​in the fixed circumferential phase loss compensation feature sequence without introducing a new network model.

[0075] Will In The components are reorganized into phase level calculation units in a fixed order. , and by Forming a phase level calculation sequence . The organizational form is Single phase level calculation unit The Wei Zhi Di The phase residuals are temperature phase residual, pressure differential phase residual, oxygen content phase residual, and current phase residual, respectively; the third... Wei Zhi Di The dimensions are the lag at the onset of the anomaly and the lag at the recovery of the anomaly, respectively; the third... Wei Zhi Di The dimensions are the temperature-induced soot blowing recovery difference, pressure-induced soot blowing recovery difference, oxygen-content-induced soot blowing recovery difference, and current-induced soot blowing recovery difference, respectively; the third... Wei Zhi Di The dimensions are adjacent circumferential phase differences for temperature, pressure, oxygen content, and current, respectively. Through this recombination method, the phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase differences of a single uncompensated phase in each thermal disturbance event maintain the same event correspondence, and multiple thermal disturbance events are arranged consecutively in ascending index order.

[0076] For each phase level calculation unit Calculation margin value First, the preset normalized threshold table, which was already written during the combined distance calculation stage, is invoked. Then, the first... Wei Zhi Di Divide each dimension by the phase residual normalization threshold of the corresponding component, and then take the absolute value; then... Wei Zhi Di Divide each dimension by the normalized threshold of the adjacent circumferential phase difference of the corresponding component, and then take the absolute value; the resulting... Add up the normalized values ​​and divide by . This forms a local offset combination value. Then, the first... Wei Zhi Di Divide each dimension by the response hysteresis normalization threshold of the corresponding component, and then take the absolute value; then... Wei Zhi Di Divide each dimension by the normalized threshold of the soot blowing recovery difference for the corresponding component, and then take the absolute value; then obtain the... Add up the normalized values ​​and divide by . This forms the recovery hysteresis combination value. Then, the local offset combination value is added to the recovery hysteresis combination value and divided by... Obtain the thermal disturbance event index Calculated margin value When the difference in soot removal recovery is zero, the first... Wei Zhi Di The zero value is used in the calculation. Therefore, the phase loss is calculated. exist The margin calculation values ​​for each thermal disturbance event are sequentially arranged to form a margin sequence. .

[0077] In the margin sequence In the middle, it will satisfy The thermal disturbance events were identified as insufficient margin events. The thermal disturbance event indices were consecutive and all satisfied... A segment of thermal disturbance events is identified as a sustained period. The margin calculation value corresponding to a given thermal disturbance event is subtracted... The single-event deficit is obtained, and within the same duration segment, the single-event deficit is accumulated item by item in the order of thermal disturbance events to form the duration segment cumulative result. The duration segment cumulative results of all duration segments are compared, and the duration segment with the largest cumulative final value is selected as the maximum duration segment; when the cumulative final values ​​are the same, the duration segment containing more thermal disturbance events is selected; when the number of thermal disturbance events is still the same, the duration segment with the smaller initial thermal disturbance event index is selected. The cumulative final value corresponding to the maximum duration segment is recorded as the local effective compensation margin deficit value. . Indicates uncompensated phase The maximum duration of insufficiency during a series of thermal disturbance events.

[0078] Let the first-level threshold be denoted as The second-level threshold is denoted as and satisfy The local effective compensation margin is insufficient. and and Comparison. When At that time, determine the phase of loss compensation. The local effective compensation margin is insufficient at level one; when At that time, determine the phase of loss compensation. The level of insufficient local effective compensation margin is level two; when At that time, determine the phase of loss compensation. The local effective compensation margin is insufficient, classified as level three. Then, each uncompensated phase... With the corresponding local effective compensation margin insufficient value And the corresponding levels of insufficient local effective compensation margin are associated one by one according to the order of the uncompensated phase index, forming The discrimination record, Indicates uncompensated phase The level of insufficient local effective compensation margin.

[0079] In summary, the core technical problem this method aims to solve is how, under high ash viscosity conditions, using only available on-site data such as temperature, pressure difference, oxygen content, current, load, and soot blowing events, to reliably identify the loss of compensation caused by local jamming of the fixed circumferential phase radial sealing sheet from the synchronous thermal disturbance of the whole machine caused by load changes and soot blowing, and to form a judgment result of insufficient local effective compensation margin corresponding to the specific circumferential position. This method addresses the problem of weak signs of localized stagnation in fixed circumferential phases being masked and unreliable identification of local effective compensation margins under high-viscosity ash conditions, where load changes, soot blowing disturbances, and overall machine synchronous fluctuations are coupled. Under available field data on temperature, pressure difference, oxygen content, current, load, and soot blowing events, circumferential phase data segments are constructed based on thermal disturbance events. A candidate set of fixed circumferential phases is selected, and a stagnation hypothesis is established around these candidate phases. By comparing with adjacent circumferential phases, phase residuals, response hysteresis, soot blowing recovery differences, and adjacent circumferential phase differences are extracted to form a sequence of uncompensated characteristics characterizing the persistence of fixed circumferential phases. Furthermore, an uncompensated relationship diagram is established, persistent coherence calculations are performed, reflection records are updated, and the set of uncompensated phases and the level of insufficient local effective compensation margin are output through iterative verification. This invention improves the reliability of local uncompensation identification and is applicable to the monitoring and classification of local uncompensation of radial seals in air preheaters.

[0080] Secondly, this method no longer directly receives abnormal quantities of the entire unit or fragments of single events at the reflection and verification entry point. Instead, it first filters a candidate set of fixed circumferential phases from the circumferential phase synchronization data, then establishes a fixed circumferential phase lag hypothesis around each candidate circumferential phase, and simultaneously retrieves data fragments of the target circumferential phase and its two adjacent circumferential phases in the same thermal disturbance event for comparison. This invention further organizes the comparison results into four types of uncompensated features: phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase difference, and forms a fixed circumferential phase uncompensated feature sequence according to the thermal disturbance event sequence. Compared with existing whole-unit threshold methods, trend comparison methods, and direct learning methods, this invention incorporates local position, adjacent position differences, and event sequence into the input representation, enabling the differentiation between whole-unit synchronization fluctuations caused by load changes, flue gas temperature changes, and soot blowing disturbances and recurring local abnormal offsets in fixed circumferential positions. This improves the scenario adaptability and physical interpretability of the local uncompensated judgment of the air preheater radial seal. Furthermore, this method does not use persistent cohomology as an independent classification tool. Instead, it first establishes a fixed circumferential phase miscompensation relationship graph based on the fixed circumferential phase miscompensation feature sequence under rotor circumferential adjacency constraints and thermal disturbance event consistency constraints. Then, it characterizes the persistence of the fixed circumferential phase anomaly relationship in multi-scale connectivity changes through the persistence lifetime of connected components. Finally, the persistence lifetime is written into the reflection record as the basis for continuous verification or verification intensity reduction. Compared with existing methods that mainly rely on single anomaly amplitude, average offset, or ordinary similarity clustering, this invention changes the basis for updating the reflection record from instantaneous strength to persistent structure under multiple events and multiple scales. This can reduce the pulling effect of occasional strong disturbances, short-term soot blowing recovery, and operating condition switching on the verification direction, and help retain the local jamming signs that truly have fixed circumferential position stability, thereby enhancing the consistency between the candidate phase screening results and the actual local miscompensation mechanism. Finally, this method, utilizing only temperature, pressure difference, oxygen content, current, load, and soot blowing event data, completes thermal disturbance event segmentation, fixed circumferential phase candidate screening, reflective verification, uncompensated phase confirmation, and local effective compensation margin deficiency level calculation, forming a complete processing path from operation monitoring to maintenance location. This invention does not rely on direct displacement measurement or disassembly / inspection information of the sealing sheet, and can output an uncompensated phase set under continuous operating conditions, and combine the margin accumulation results in continuous thermal disturbance events to give the local effective compensation margin deficiency level. Compared to existing technologies that mostly focus on overall unit leakage judgment or overall sealing status alarms, this invention can further implement the diagnostic results to fixed circumferential positions and local compensation status, providing a more direct basis for air preheater seal maintenance, soot blowing strategy optimization, and operation and maintenance.

[0081] The second objective of this invention is to propose a local loss-compensation discrimination system for radial sealing plates of an air preheater based on a reflective intelligent agent, comprising: Circumferential phase data segment module: used to acquire data of the air preheater synchronized with the rotor circumferential phase, divide the data segments corresponding to each circumferential phase according to the thermal disturbance event, and obtain the circumferential phase data segment; Fixed circumferential phase candidate set module: used to determine thermal disturbance events based on circumferential phase data segments, and calculate the repetition degree of each circumferential phase in each thermal disturbance event to obtain a fixed circumferential phase candidate set; Phase loss compensation feature sequence module: used to input a fixed circumferential phase candidate set into the reflective agent and extract a fixed circumferential phase loss compensation feature sequence; Update the reflection record module: It is used to perform persistent cohomology calculation based on the fixed circumferential phase loss compensation feature sequence, obtain the persistent lifetime, and write the comparison result of persistent lifetime and lifetime threshold into the reflection record to obtain the updated reflection record; Uncompensated phase set module: It is used to iteratively verify the fixed circumferential phase candidate set based on the updated reflection record, filter the candidate circumferential phases that satisfy the fixed circumferential phase stagnation assumption, and obtain the uncompensated phase set; Loss of compensation discrimination result module: It is used to calculate the level of insufficient local effective compensation margin based on the set of uncompensated phases and the fixed circumferential phase loss of compensation characteristic sequence, and output the local loss of compensation discrimination result.

[0082] A third objective of this invention is to provide an electronic device comprising a processor, a memory, and a display screen. The memory and display screen are both connected to the processor, such as via a bus. Optionally, the electronic device may further include a transceiver. It should be noted that in practical applications, the transceiver is not limited to a single unit, and the structure of this electronic device does not constitute a limitation on the embodiments of this application.

[0083] The processor can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0084] A bus can include a pathway for transmitting information between the aforementioned components. The bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc.

[0085] The memory may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these.

[0086] The memory stores the application code that executes the solution of this application, and its execution is controlled by the processor. The processor executes the application code stored in the memory to implement the content shown in the foregoing method embodiments.

[0087] A fourth objective of this invention is to provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, performs the aforementioned functions. Figure 2 The illustrated method embodiments include various processes. For example, a memory may include instructions that can be executed by a processor of an electronic device to perform the described method.

[0088] A computer-readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device. A computer-readable storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, a computer-readable storage medium can be a portable computer disk, a hard disk, a USB flash drive, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), staging random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory stick, floppy disk, optical disk, magnetic disk, mechanical encoding device, or any combination thereof.

[0089] A fifth objective of this invention is to provide a computer program product comprising computer instructions that, when executed by a processor, implement the above-described... Figure 2 The various processes of the method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0090] Many embodiments and applications beyond the examples provided will be apparent to those skilled in the art upon reading the foregoing description. Therefore, the scope of this teaching should not be determined by reference to the foregoing description, but rather by reference to the foregoing claims and the full scope of their equivalents. For purposes of completeness, all articles and references, including patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein in the foregoing claims is not intended as a waiver of that subject matter, nor should it be construed as an indication that the applicant has not considered that subject matter as part of the disclosed inventive subject matter.

[0091] The above content provides a further detailed description of the present invention. It should not be construed that the specific embodiments of the present invention are limited to this. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention as defined by the submitted claims.

Claims

1. A method for determining local loss of compensation of radial sealing sheet in an air preheater based on reflective intelligent agents, characterized in that, include: Data on the circumferential phase synchronization of the air preheater is obtained, and data segments corresponding to each circumferential phase are divided according to thermal disturbance events to obtain circumferential phase data segments. Thermal disturbance events are determined based on circumferential phase data fragments, and the repetition rate of each circumferential phase in each thermal disturbance event is calculated to obtain a fixed circumferential phase candidate set. Input the fixed circumferential phase candidate set into the reflective agent to extract the fixed circumferential phase loss compensation feature sequence; The persistent cohomology is calculated based on the fixed circumferential phase loss compensation characteristic sequence to obtain the persistent lifetime. The comparison result between the persistent lifetime and the lifetime threshold is written into the reflection record to obtain the updated reflection record. Based on the updated reflection record, the candidate set of fixed circumferential phases is iteratively verified, and candidate circumferential phases that satisfy the fixed circumferential phase stagnation hypothesis are selected to obtain the uncompensated phase set. Based on the uncompensated phase set and the fixed circumferential phase uncompensated feature sequence, the level of insufficient local effective compensation margin is calculated, and the local uncompensated discrimination result is output.

2. The method for determining local loss of compensation of radial sealing sheet in an air preheater based on a reflective agent, as described in claim 1, is characterized in that... The process involves acquiring data on the rotor circumferential phase synchronization of the air preheater, dividing the data into segments corresponding to each circumferential phase according to thermal disturbance events, and obtaining circumferential phase data segments, including: The system acquires temperature, differential pressure, oxygen content, current, load, and soot blowing event data of the air preheater in the rotor circumferential phase synchronization during continuous operation cycles. Based on the timestamp and rotor angular position information, the temperature, differential pressure, oxygen content, current, load, and soot blowing event data are mapped to each circumferential phase and arranged in circumferential phase order within the same rotor circumferential cycle. Based on the start and end times of the thermal disturbance event, continuous data containing the events before, during, and after the thermal disturbance event is extracted from the data corresponding to each circumferential phase. The captured continuous data is arranged in order of circumferential phase and thermal disturbance events to obtain circumferential phase data segments.

3. The method for determining local loss of compensation of radial sealing sheet in an air preheater based on a reflective agent, as described in claim 1, is characterized in that... The process involves determining thermal disturbance events based on circumferential phase data fragments and calculating the repetition rate of each circumferential phase in each thermal disturbance event to obtain a fixed circumferential phase candidate set, including: Based on the circumferential phase data segments, the load change slope, soot blowing event trigger marker, and continuous offset of temperature or pressure difference within a preset time window are extracted in the order of operation time to determine the start time, end time, and thermal disturbance event identifier, thus forming a thermal disturbance event sequence. Based on the thermal disturbance event sequence, an event reference segment corresponding to the thermal disturbance event identifier is formed; Based on the event comparison segments, the temperature shift, pressure difference shift, oxygen content shift, and current shift of each circumferential phase in each thermal disturbance event are calculated respectively. The temperature shift, pressure difference shift, oxygen content shift, and current shift are then combined according to the thermal disturbance event identifier and the circumferential phase sequence to form the circumferential phase anomaly trajectory corresponding to each circumferential phase. Based on the circumferential phase anomaly trajectory, the same circumferential phase is matched in different thermal disturbance events, and the recurrence score of each circumferential phase is calculated; based on the recurrence score, the degree of recurrence of each circumferential phase in each thermal disturbance event is determined, forming a fixed circumferential phase recurrence result; The results of repeated occurrence of fixed circumferential phases are compared with preset thresholds and thermal disturbance event coverage thresholds. Circumferential phases that meet the comparison conditions are selected and sorted from high to low repetition degree to obtain a candidate set of fixed circumferential phases.

4. The method for determining local loss of compensation of radial sealing sheet of air preheater based on reflective intelligent agent according to claim 1, characterized in that, The step of inputting a fixed circumferential phase candidate set into the reflective agent and extracting a fixed circumferential phase loss-compensation feature sequence includes: Input a fixed circumferential phase candidate set into the reflective agent, perform event matching and time alignment, and form an event sequence comparison result; Based on the results of the event sequence comparison, the changes in the target circumferential phase in terms of temperature, pressure difference, oxygen content and current are calculated respectively to form the phase residual; Based on the results of the event sequence comparison, the abnormal start time and abnormal recovery time of the target circumferential phase in the thermal disturbance event change segment are extracted to form a response hysteresis. Based on the results of the event sequence comparison, the differences in temperature, pressure difference, oxygen content and current of the target circumferential phase before and after the soot blowing event are calculated to form the soot blowing recovery difference. The differences in temperature, pressure difference, oxygen content and current between the target circumferential phase and the adjacent circumferential phases on both sides in the same thermal disturbance event and the same section are also calculated to form the adjacent circumferential phase difference. The phase residual, response hysteresis, soot blowing recovery difference, and adjacent circumferential phase difference are combined with the target circumferential phase sequence according to the thermal disturbance event sequence to obtain a fixed circumferential phase loss compensation characteristic sequence.

5. The method for determining local loss of compensation of radial sealing sheet in an air preheater based on a reflective agent, as described in claim 1, is characterized in that... The process involves calculating persistent cohomology based on a fixed circumferential phase loss compensation feature sequence to obtain the persistent lifetime, and then comparing the persistent lifetime with the lifetime threshold in the reflection record to obtain an updated reflection record, including: Based on the fixed circumferential phase loss compensation characteristic sequence, the order of appearance and disappearance of connected components corresponding to each level of connection relationship is obtained, forming the long-term lifetime; The longevity is compared with the lifetime threshold. If the comparison conditions are met, a continuous verification update result is generated. If the comparison conditions are not met, a verification intensity reduction update result is generated, resulting in a reflection record update result. Write the updated reflection record results into the corresponding candidate circumferential phase record position in the reflection record, and adjust the verification state of the corresponding candidate circumferential phase according to the updated reflection record results to obtain the updated reflection record.

6. The method for determining local loss of compensation of radial sealing sheet in an air preheater based on a reflective agent, as described in claim 1, is characterized in that... The process involves iteratively verifying the candidate set of fixed circumferential phases based on the updated reflection records, filtering out candidate circumferential phases that satisfy the fixed circumferential phase lag assumption, and obtaining a set of uncompensated phases, including: Based on the updated reflection record, extract the record value corresponding to each candidate circumferential phase in the fixed circumferential phase candidate set to form a candidate circumferential phase verification sequence; According to the candidate circumferential phase verification sequence, the verification results of the fixed circumferential phase stagnation hypothesis corresponding to each candidate circumferential phase in different thermal disturbance events are iteratively summarized to form the candidate circumferential phase verification results. The verification results of the candidate circumferential phases are compared with a preset threshold. If the comparison conditions are met, the corresponding candidate circumferential phases are determined to satisfy the fixed circumferential phase lag assumption. If the comparison conditions are not met, the corresponding candidate circumferential phases are eliminated to form a set of valid candidate circumferential phases. The set of valid candidate circumferential phases is then filtered from high to low according to the verification results of the candidate circumferential phases to obtain a set of uncompensated phases.

7. The method for determining local loss of compensation of radial sealing sheet of air preheater based on reflective intelligent agent according to claim 1, characterized in that, The step involves calculating the level of insufficient local effective compensation margin based on the uncompensated phase set and the fixed circumferential phase uncompensated feature sequence, and outputting the local uncompensated discrimination result, including: Based on the uncompensated phase set, the extracted results are reorganized into a phase level calculation sequence that maintains the correspondence between events according to the uncompensated phase and the thermal disturbance event sequence; Based on the phase level calculation sequence, the phase residual and adjacent circumferential phase difference of the same uncompensated phase in each thermal disturbance event are combined and calculated, and the response hysteresis and soot blowing recovery difference are combined and calculated to form the margin calculation value corresponding to each uncompensated phase. Based on the margin calculation value, the margin calculation values ​​of each loss compensation phase in continuous thermal disturbance events are accumulated, and the maximum duration segment in the accumulated results is extracted to form the local effective compensation margin insufficient value. The local effective compensation margin deficiency value is compared with the preset level threshold to determine the local effective compensation margin deficiency level. The set of uncompensated phases is then associated with the local effective compensation margin deficiency level in a one-to-one correspondence according to the uncompensated phases to obtain the local uncompensated discrimination result.

8. A system for determining local uncompensated radial sealing strips of an air preheater based on a reflective agent, wherein the method for determining local uncompensated radial sealing strips of an air preheater based on a reflective agent as described in any one of claims 1-7 is characterized in that, include: Circumferential phase data segment module: used to acquire data of the air preheater synchronized with the rotor circumferential phase, divide the data segments corresponding to each circumferential phase according to the thermal disturbance event, and obtain the circumferential phase data segment; Fixed circumferential phase candidate set module: used to determine thermal disturbance events based on circumferential phase data segments, and calculate the repetition degree of each circumferential phase in each thermal disturbance event to obtain a fixed circumferential phase candidate set; Phase loss compensation feature sequence module: used to input a fixed circumferential phase candidate set into the reflective agent and extract a fixed circumferential phase loss compensation feature sequence; Update the reflection record module: It is used to perform persistent cohomology calculation based on the fixed circumferential phase loss compensation feature sequence, obtain the persistent lifetime, and write the comparison result of persistent lifetime and lifetime threshold into the reflection record to obtain the updated reflection record; Uncompensated phase set module: It is used to iteratively verify the fixed circumferential phase candidate set based on the updated reflection record, filter the candidate circumferential phases that satisfy the fixed circumferential phase stagnation assumption, and obtain the uncompensated phase set; Loss of compensation discrimination result module: It is used to calculate the level of insufficient local effective compensation margin based on the set of uncompensated phases and the fixed circumferential phase loss of compensation characteristic sequence, and output the local loss of compensation discrimination result.

9. An electronic device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for determining the local loss of compensation of radial sealing sheet of an air preheater based on any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for determining local loss compensation of radial sealing sheet of an air preheater based on a reflective agent, as described in any one of claims 1-8.

Images