Rotary air preheater directional dust cleaning triggering method based on xRFM driving recovery boundary segmentation

Abstract

The application discloses a rotary air preheater directional soot cleaning trigger method based on xRFM driving recovery boundary segmentation, and belongs to the technical field of coal-fired boilers. The method comprises the following steps: acquiring pressure drop change records and soot cleaning time records of a cold end and a transition zone of a rotary air preheater, and obtaining a sector pressure drop sequence; extracting a pressure drop back-off amount according to the sector pressure drop sequence and the soot cleaning time records; calculating a recovery representation according to the pressure drop cumulative amount and the pressure drop back-off amount, and determining a current sector recovery state; inputting the recovery representation into xRFM, performing segmentation identification, and obtaining a boundary segmentation label; inputting the boundary segmentation label into a sector pressure drop recovery soot blowing criterion method, and obtaining a current segmentation recovery boundary; and judging the current sector recovery state and the current segmentation recovery boundary under the constraint of the current segmentation recovery boundary. The method realizes the classification selection of the recovery boundary through the boundary segmentation label, and improves the availability and consistency of the sector-level criterion under the conditions of working condition difference and soot blocking uncertainty.

Description

Technical Field

[0001] This invention relates to the field of coal-fired boiler technology, specifically to a method, system, equipment, and medium for directional ash removal triggering of a rotary air preheater based on xRFM-driven recovery boundary segmentation. Background Technology

[0002] Rotary air preheaters, as crucial equipment for utilizing waste heat from flue gas in coal-fired boilers, typically consist of a rotor and heat storage elements. During rotation, they facilitate heat exchange between flue gas and air. Because flue gas contains fly ash and sticky particles, the heat storage element channels are prone to ash accumulation and blockage during operation, especially in the cold end and transition zone, where condensation and ash deposition are more likely to occur, causing a gradual increase in channel resistance, manifested as a continuous increase in pressure drop. This increased pressure drop not only leads to increased induced draft fan load but also affects air distribution stability and furnace combustion organization, potentially causing further localized blockages and abnormal heating surfaces. To mitigate the impact of ash accumulation, soot blowing or cleaning devices are typically installed on-site, and the cleaning strategy is controlled based on changes in operating pressure drop. Since the air preheater is divided into multiple sectors along the circumference, the flue gas flow distribution, temperature conditions, and ash blockage morphology vary in different sectors. The pressure drop release amplitude and recovery speed after ash removal will also be different. Therefore, it is necessary to perform more granular identification on the relationship between pressure drop evolution and recovery after ash removal during continuous operation to support the triggering and allocation of directional ash removal, thereby ensuring flow capacity while avoiding unnecessary ash removal actions and disturbances.

[0003] In existing technologies, air preheater cleaning control often employs a timed and sequential approach, performing soot blowing on the cold end or specific areas according to a preset cycle, or using a threshold value for the overall unit pressure drop as a trigger. Some solutions utilize characteristics such as pressure drop slope and cumulative increment to assess soot blockage development, combining empirical criteria to determine the timing of cleaning. Some systems introduce data-driven models to predict pressure drop trends and provide cleaning suggestions based on the prediction results. For sector-level control scenarios, existing solutions typically use the current sector pressure drop or short-term changes as the basis for judgment, dividing sectors into those requiring cleaning and those temporarily not requiring cleaning according to fixed thresholds or fixed segmentation rules. The control system then issues cleaning commands to address localized soot blockage and control operational risks. Summary of the Invention

[0004] Existing technologies for air preheater cleaning control often rely on timed and sequential methods, or simple features like the current pressure drop of the entire unit or sector to determine cleaning timing. These methods struggle to provide fine-grained identification of the relationship between pressure drop evolution and post-cleaning recovery during continuous operation, failing to accurately support the triggering and allocation of directional cleaning. This can lead to unnecessary cleaning actions and disturbances, or ineffective handling of localized ash blockage. This invention provides a directional cleaning triggering method for rotary air preheaters based on xRFM-driven recovery boundary segmentation. By using boundary segmentation labels to categorize the recovery boundaries, sectors with different recovery capabilities can adapt to different boundary segmentation rules under the same operational scenario, improving the usability and consistency of sector-level criteria under varying operating conditions and ash blockage uncertainties.

[0005] To achieve the above objectives, the present invention provides the following technical solution.

[0006] In a first aspect, the present invention provides a method for triggering directional cleaning of a rotary air preheater based on xRFM-driven recovery boundary segmentation, comprising: Acquire the pressure drop change records and dust removal time records of each target sector in the cold end and transition zone of the rotary air preheater during continuous operation, and obtain the sector pressure drop sequence based on the pressure drop change records; Based on the sector voltage drop sequence and dust removal time records, extract the cumulative voltage drop before each dust removal time and the voltage drop decrease after each dust removal time; Based on the cumulative pressure drop and the pressure drop decrease, calculate the recovery characterization quantity that characterizes the relationship between pressure drop evolution and post-cleaning recovery, and determine the current sector recovery status based on the recovery characterization quantity; Input the recovery characterization quantity into xRFM, and segment the recovery characterization quantity according to the recovery boundary corresponding to different ash blockage recovery hidden states to obtain the boundary segment label; Input the boundary segment label into the sector pressure drop recovery soot blowing criterion method, and match the current sector recovery status to the recovery boundary category corresponding to the boundary segment label to obtain the current segment recovery boundary; Under the current segmented recovery boundary constraints, the current sector recovery status is compared with the current segmented recovery boundary. If the triggering conditions are met, a directional dust removal trigger command is generated; otherwise, a maintenance operation command is generated.

[0007] As a further improvement of the present invention, the step of acquiring the pressure drop change records and dust removal time records of each target sector in the cold end and transition zone of the rotary air preheater during continuous operation, and obtaining the sector pressure drop sequence based on the pressure drop change records, includes: The target sectors were determined according to the sector division of the cold end and transition zone of the rotary air preheater, and the pressure drop change records and dust removal time records of each target sector during continuous operation were collected. Using each dust removal moment in the dust removal moment record as a dividing point, the sector timing voltage drop data corresponding to each dust removal moment is obtained; The sector timing voltage drop data are arranged to obtain the sector voltage drop sequence of each target sector.

[0008] As a further improvement of the present invention, the step of extracting the cumulative voltage drop before each dust removal moment and the voltage drop decrease after each dust removal moment based on the sector voltage drop sequence and dust removal time record includes: Based on the dust removal time records, the continuous operation analysis range corresponding to the dust removal event is selected from the sector voltage drop sequence of each target sector. The pressure drop accumulation interval is determined based on the continuous operation analysis range, and the pressure drop increase between each adjacent record value in the pressure drop accumulation interval is extracted to obtain the rising change sequence; The pressure drop accumulation corresponding to the current dust removal moment is obtained by accumulating the increasing change sequence in the recording order; The sector voltage drop sequence from the current dust removal time to the moment when the first value not lower than the previous record appears in the sector voltage drop sequence is defined as the voltage drop fall interval. Adjacent record values ​​in the voltage drop fall interval are compared in the recording order. Based on the comparison results, the voltage drop decrease between each adjacent record value is extracted to obtain the voltage drop fall amount corresponding to the current dust removal time.

[0009] As a further improvement of the present invention, the step of calculating a recovery characterization quantity, which characterizes the relationship between pressure drop evolution and post-cleaning recovery, based on the cumulative pressure drop and the pressure drop decrease, and determining the current sector recovery status based on the recovery characterization quantity, includes: Based on the current cumulative pressure drop and the cumulative pressure drop corresponding to the previous cleaning time, calculate the change in the current cumulative pressure drop relative to the previous cleaning event. Based on the current pressure drop decrease and the pressure drop decrease corresponding to the previous cleaning time, calculate the change in the current pressure drop decrease relative to the previous cleaning event. Based on the current cumulative pressure drop and the current pressure drop decrease, calculate the difference between the current cumulative pressure drop and the current pressure drop decrease, and calculate the ratio between the current cumulative pressure drop and the current pressure drop decrease. The current cumulative pressure drop, the current pressure drop decrease, the change in the current cumulative pressure drop relative to the previous dust removal event, the change in the current pressure drop relative to the previous dust removal event, the difference between the current cumulative pressure drop and the current pressure drop decrease, and the ratio of the current cumulative pressure drop to the current pressure drop decrease are arranged in the input order of xRFM to form the recovery characterization quantity. Calculate the recovery boundary position corresponding to the current dust removal event based on the recovery characteristic quantity; Based on the location of the recovery boundary within the preset state segment, obtain the recovery state value and determine the recovery state value as the current sector recovery state.

[0010] As a further improvement of the present invention, the step of inputting the recovery characterization quantity into xRFM and segmenting the recovery characterization quantity according to the recovery boundary corresponding to different ash-blocking recovery hidden states to obtain boundary segmentation labels includes: The recovered representation is input into the input layer of xRFM, and the recovery mapping layer, composed of fully connected neurons, maps the recovered representation to obtain the mapping result. The mapping results are aggregated into recovery mapping features, and the recovery mapping features are input into the ash-blocking recovery hidden state layer composed of fully connected neurons. The recovery mapping features are compressed and represented to obtain the ash-blocking recovery hidden state vector. Based on the hidden state vector of ash blockage recovery, the current sector recovery state is assigned to the hidden state of high recovery ash blockage recovery, the hidden state of transitional recovery ash blockage recovery, and the hidden state of low recovery ash blockage recovery, and the assignment result of the hidden state of ash blockage recovery is obtained. The results of restoring the hidden state of the ash blockage are input into the criterion embedding layer composed of fully connected neurons for segment recognition, and the boundary segment labels are obtained.

[0011] As a further improvement of the present invention, the step of inputting the boundary segmentation label into the sector voltage drop recovery soot blowing criterion method and matching the current sector recovery state to the recovery boundary category corresponding to the boundary segmentation label to obtain the current segmented recovery boundary includes: Connect the sector voltage drop recovery soot blowing criterion method to the output of xRFM as the criterion header; The boundary segmentation label and the current sector recovery status are jointly input into the sector pressure drop recovery soot blowing criterion method; The recovery boundary category is determined based on the boundary segment label, and the boundary comparison interval corresponding to the recovery boundary category is called. In the sector voltage drop recovery soot blowing criterion method, a recovery boundary matching layer is set up to activate the matching neurons corresponding to the recovery boundary category and form the current boundary matching pathway. Input the current sector recovery status into the current boundary matching path, and obtain the boundary position matching result based on the relative position of the current sector recovery status with the start boundary, intermediate boundary and end boundary; determine the segment boundary position based on the boundary position matching result, and generate the current segment recovery boundary; The current segmented recovery boundary is passed to the triggering and determination layer of the sector pressure drop recovery soot blowing criterion method, which is used to determine the directional soot blowing trigger command and the maintenance operation command.

[0012] As a further improvement of the present invention, the step of determining the current sector recovery state and the current segment recovery boundary under the current segment recovery boundary constraints, generating a directional dust removal trigger command when the trigger condition is met, and generating a maintenance operation command when the trigger condition is not met includes: The current sector recovery status, the current segment recovery boundary, and the boundary segment label are input into the trigger determination layer of the sector voltage drop recovery soot blowing criterion method. Under the constraint of the current segment recovery boundary, the trigger interval corresponding to the current segment recovery boundary is determined according to the boundary segment label, and the trigger determination value is calculated according to the segment position and boundary direction of the current sector recovery status within the trigger interval. Determine the trigger value and the trigger range; When the trigger judgment value reaches the boundary condition corresponding to the trigger interval, the directional cleaning trigger judgment neuron is activated, and the judgment result of the directional cleaning trigger judgment neuron is determined as the directional cleaning trigger command. When the trigger judgment value does not meet the boundary conditions corresponding to the trigger interval, the maintain operation judgment neuron is activated, and the judgment result of the maintain operation judgment neuron is determined as the maintain operation instruction.

[0013] Secondly, the present invention provides a rotary air preheater directional cleaning triggering system based on xRFM-driven recovery boundary segmentation, comprising: Sector pressure drop sequence module: used to acquire the pressure drop change records and dust removal time records of each target sector in the cold end and transition zone of the rotary air preheater during continuous operation, and to obtain the sector pressure drop sequence based on the pressure drop change records; Pressure drop recovery module: used to extract the cumulative pressure drop before each cleaning moment and the pressure drop recovery after each cleaning moment based on the sector pressure drop sequence and cleaning time record; Sector recovery status module: Used to calculate the recovery characterization quantity, which characterizes the relationship between pressure drop evolution and post-cleaning recovery, based on the cumulative pressure drop and the pressure drop reduction, and to determine the current sector recovery status based on the recovery characterization quantity; Boundary segmentation label module: It is used to input the restored characterization quantity into xRFM, and to segment and identify the restored characterization quantity according to the restoration boundary corresponding to different ash-blocking restoration hidden states to obtain the boundary segmentation label; Segmented recovery boundary module: Used to input the boundary segment label into the sector voltage drop recovery soot blowing criterion method, and match the current sector recovery status to the recovery boundary category corresponding to the boundary segment label to obtain the current segmented recovery boundary; Directional dust removal trigger instruction module: Under the current segment recovery boundary constraints, it determines the current sector recovery status and the current segment recovery boundary. If the trigger conditions are met, it generates a directional dust removal trigger instruction; if the trigger conditions are not met, it generates a maintain operation instruction.

[0014] 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 executes the computer program to implement the aforementioned method for triggering directional cleaning of a rotary air preheater based on xRFM-driven recovery boundary segmentation.

[0015] 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 triggering directional cleaning of a rotary air preheater based on xRFM-driven recovery boundary segmentation.

[0016] Fifthly, the present invention provides a computer program product, including computer instructions, which, when executed by a processor, implement the aforementioned method for triggering directional cleaning of a rotary air preheater based on xRFM-driven recovery boundary segmentation.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention accurately acquires the pressure drop change records and soot blowing time records of each target sector in the cold end and transition zone of a rotary air preheater during continuous operation. Based on this, a sector pressure drop sequence is obtained, and the cumulative pressure drop and reduction before and after each soot blowing time are extracted. Recovery characteristics are calculated to determine the current sector recovery state. Furthermore, this invention introduces xRFM to segment and identify the recovery characteristics, obtaining boundary segment labels. Using a sector pressure drop recovery soot blowing criterion, the current sector recovery state is accurately matched to the corresponding recovery boundary category, determining the current segment recovery boundary. This achieves categorical selection of recovery boundaries, fully considering the differences in recovery capabilities among different sectors. In actual operation, different sectors may have different recovery capabilities due to factors such as location and operating environment. This invention's method enables sectors with different recovery capabilities to adapt to different boundary segmentation rules in the same business scenario, avoiding a one-size-fits-all control approach. This enhances the usability of sector-level criteria under complex operating conditions and uncertainties related to ash blockage, ensuring accurate determination of cleaning needs under various conditions. It also strengthens the consistency of criteria, reducing misjudgments caused by changes in operating conditions. By accurately determining the current sector recovery status and the current segment recovery boundary, it promptly generates directional cleaning trigger commands when trigger conditions are met, effectively addressing localized ash blockage issues. When trigger conditions are not met, it generates maintenance commands, avoiding unnecessary cleaning actions and disturbances. This ensures the stable and efficient operation of the rotary air preheater and reduces operating costs. Attached Figure Description

[0018] 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 1 This is a flowchart illustrating a method for triggering directional cleaning of a rotary air preheater based on xRFM-driven boundary segmentation according to the present invention. Figure 2 This is a schematic diagram of step S1 of the rotary air preheater directional cleaning triggering method based on xRFM driven recovery boundary segmentation according to the present invention; Figure 3 This is a schematic diagram of step S2 of the rotary air preheater directional cleaning triggering method based on xRFM driven recovery boundary segmentation according to the present invention; Figure 4 This is a schematic diagram of step S3 of the rotary air preheater directional cleaning triggering method based on xRFM driven recovery boundary segmentation according to the present invention; Figure 5 This is a schematic diagram of step S4 of the rotary air preheater directional cleaning triggering method based on xRFM driven recovery boundary segmentation according to the present invention; Figure 6 This is a schematic diagram of step S5 of the rotary air preheater directional cleaning triggering method based on xRFM driven recovery boundary segmentation according to the present invention; Figure 7 This is a schematic diagram of step S6 of the rotary air preheater directional cleaning triggering method based on xRFM driven recovery boundary segmentation according to the present invention; Figure 8 This is a schematic diagram of dust removal event alignment in an embodiment of the present invention; Figure 9 This is a schematic diagram illustrating the alignment, segmentation, and voltage drop characteristics of the target sector dust removal event in an embodiment of the present invention. Figure 10 This is a determination diagram of the pressure drop accumulation-fallback two-dimensional density field and directional triggering domain in an embodiment of the present invention. Detailed Implementation

[0019] 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.

[0020] 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.

[0021] Definitions: xRFM: refers to a recovery boundary segmentation identification network that takes recovery representation quantities as input. It consists of a recovery mapping layer, a ash-blocking recovery hidden state layer, and a criterion embedding layer. It is used to represent the hidden state relationship between pressure drop evolution and recovery after ash removal and output boundary segment labels to represent the recovery boundary category corresponding to the current sector recovery state, and to provide boundary selection parameters for the subsequent sector pressure drop recovery ash blowing criterion method.

[0022] Target sectors in the cold end and transition zone of the rotary air preheater: sectors that are determined according to sector division within the cold end and transition zone of the rotary air preheater, and for which pressure drop changes and cleaning times are recorded respectively.

[0023] Sector voltage drop sequence: The sequence formed by arranging the preceding and following operating voltage drops at each cleaning moment in the same target sector according to the order of cleaning moments.

[0024] Accumulated pressure drop: The amount of pressure drop increase between adjacent records in the sector pressure drop sequence from the previous cleaning time to the current cleaning time, accumulated in the order of recording.

[0025] Pressure drop reduction: The amount of pressure drop between adjacent records in the sector pressure drop sequence from the current dust removal time to the moment when the first record value not lower than the previous record value appears in the sector pressure drop sequence, calculated in the order of recording.

[0026] Recovery characteristics of the relationship between pressure drop evolution and post-cleaning recovery: current pressure drop accumulation, current pressure drop decrease, change of current pressure drop accumulation relative to the previous cleaning event, change of current pressure drop decrease relative to the previous cleaning event, difference between current pressure drop accumulation and current pressure drop decrease, and ratio of current pressure drop accumulation to current pressure drop decrease, arranged in the input order of xRFM.

[0027] Current sector recovery status: Calculated based on the recovery characteristic quantity and used to characterize the direction of change in the recovery capability of the target sector under the current dust removal event.

[0028] Boundary segmentation label: The label obtained by xRFM after segmenting and identifying the recovery characterization quantity according to the recovery boundary corresponding to different ash blockage recovery hidden states is used to characterize the recovery boundary category corresponding to the current sector recovery state.

[0029] Sector voltage drop recovery soot blowing criterion method: The method involves connecting to the output of xRFM and receiving the boundary segment tag and the current sector recovery status, and generating the current segment recovery boundary and judgment result based on the recovery boundary category.

[0030] Current segmented recovery boundary: The recovery boundary generated by the sector voltage drop recovery soot blowing criterion method based on the current sector recovery status under the recovery boundary category corresponding to the boundary segment label.

[0031] Triggering condition: Under the current segment recovery boundary constraints, the trigger judgment value corresponding to the current sector recovery state reaches the boundary condition corresponding to the trigger interval.

[0032] Existing air preheater cleaning control technologies often rely on timed and sequential methods, using simple features like the current pressure drop of the entire unit or sector to determine cleaning timing. This approach struggles to fine-grainedly identify the relationship between pressure drop evolution and post-cleaning recovery during continuous operation, failing to accurately support the triggering and allocation of directional cleaning. This can easily lead to unnecessary cleaning actions and disturbances, or ineffective handling of localized ash blockage. This invention provides a directional cleaning triggering method for rotary air preheaters based on xRFM-driven recovery boundary segmentation, such as... Figure 1 As shown, it includes: S100: Obtain the pressure drop change records and dust removal time records of each target sector in the cold end and transition zone of the rotary air preheater during continuous operation, and obtain the sector pressure drop sequence based on the pressure drop change records; S200: Based on the sector voltage drop sequence and dust removal time record, extract the cumulative voltage drop before each dust removal time and the voltage drop reduction after each dust removal time; S300: Based on the cumulative pressure drop and the pressure drop decrease, calculate the recovery characterization quantity that characterizes the relationship between pressure drop evolution and post-cleaning recovery, and determine the current sector recovery status based on the recovery characterization quantity; S400: Input the recovery characterization quantity into xRFM, and segment the recovery characterization quantity according to the recovery boundary corresponding to different ash blockage recovery hidden states to obtain the boundary segment label. The boundary segment label is used to characterize the recovery boundary category corresponding to the current sector recovery state, so that target sectors with different ash blockage forms and different recovery capabilities can be adapted to different recovery boundary segments. S500: Input the boundary segment label into the sector pressure drop recovery soot blowing criterion method, and match the current sector recovery status to the recovery boundary category corresponding to the boundary segment label to obtain the current segment recovery boundary; S600: Under the current segment recovery boundary constraints, the current sector recovery status is compared with the current segment recovery boundary. If the triggering conditions are met, a directional dust removal trigger command is generated; otherwise, a maintenance operation command is generated.

[0033] This method enables the categorized selection of recovery boundaries through boundary segmentation labels, allowing sectors with different recovery capabilities to adapt to different boundary segmentation rules in the same business scenario, thereby improving the availability and consistency of sector-level criteria under conditions of operating differences and dust blockage uncertainty.

[0034] The present invention will be further explained and described below with reference to the accompanying drawings.

[0035] A method for triggering directional dust removal in a rotary air preheater based on xRFM-driven boundary segmentation, comprising the following steps: like Figure 2 As shown, S1 is specifically: The target sectors were determined according to the sector division of the cold end and transition zone of the rotary air preheater, and the pressure drop change records and dust removal time records of each target sector during continuous operation were collected. Using each cleaning moment in the cleaning moment record as a dividing point, the voltage drop change record of each target sector is divided into the pre-running voltage drop before the cleaning moment and the post-running voltage drop after the cleaning moment, so as to obtain the sector timing voltage drop data corresponding to each cleaning moment. The sector timing voltage drop data are arranged in the order of each target sector and each dust removal time. The preceding and following operating voltage drops corresponding to each dust removal time are used to form the sector voltage drop sequence of each target sector. Based on the sequential relationship of each cleaning time within each target sector, the sector pressure drop sequence corresponding to the current cleaning time is sequentially associated with the sector pressure drop sequence corresponding to the previous cleaning time. This association is used to extract the cumulative pressure drop, the pressure drop decrease, and to calculate the recovery characteristic quantity.

[0036] like Figure 3 As shown, S2 is specifically: Based on the dust removal time record, the current dust removal time with the previous dust removal time is selected in the sector voltage drop sequence of each target sector in chronological order, and the sector voltage drop sequence between the previous dust removal time and the current dust removal time is determined as the continuous operation analysis range corresponding to this dust removal event. The continuous operation analysis range is used to limit the extraction boundaries of the cumulative voltage drop and the voltage drop fall. The sector pressure drop sequence within the continuous operation analysis range that is after the previous dust removal time and before the current dust removal time is determined as the pressure drop accumulation interval. Adjacent record values ​​in the pressure drop accumulation interval are compared in the recording order, and the pressure drop increase between each adjacent record value is extracted to obtain the rising change sequence. The increasing change sequence is accumulated in the order of recording to obtain the cumulative pressure drop corresponding to the current dust removal time. The cumulative pressure drop corresponds to the cumulative degree of flow obstruction caused by dust blockage in the target sector between two dust removals. A sequential correspondence is established between the cumulative pressure drop corresponding to the current dust removal time and the cumulative pressure drop corresponding to the previous dust removal time. The sector voltage drop sequence from the current dust removal time to the moment when the first value not lower than the previous record appears in the sector voltage drop sequence is defined as the voltage drop fall-off interval. The voltage drop fall-off interval is used to limit the range of voltage drop release under the influence of this dust removal action, and adjacent record values ​​in the voltage drop fall-off interval are compared in the recording order. Based on the comparison results, the pressure drop between each adjacent record value is extracted, and the pressure drop is accumulated in the order of the records to obtain the pressure drop recovery amount corresponding to the current cleaning time. The pressure drop recovery amount corresponds to the recovery range of the target sector after this cleaning action, and a sequential correspondence is established between the pressure drop recovery amount corresponding to the current cleaning time and the pressure drop recovery amount corresponding to the previous cleaning time. The cumulative pressure drop at the current dust removal moment and the pressure drop at the current dust removal moment are paired according to the same dust removal event to obtain the correspondence of the current dust removal event, so that the dust blockage development process and dust removal recovery process of the same target sector in this dust removal event enter the same calculation link; The cumulative pressure drop and the decrease in pressure drop corresponding to the previous cleaning time are associated with the current cleaning event in the order of cleaning time, forming the basis for comparing adjacent cleaning events called by S3 to calculate the recovery characterization quantity. This ensures that the recovery characterization quantity calculated by S3 simultaneously includes the cumulative pressure drop corresponding to the current cleaning time, the decrease in pressure drop corresponding to the current cleaning time, and the change relationship between adjacent cleaning events.

[0037] like Figure 4 As shown, S3 specifically refers to: The cumulative pressure drop at the current dust removal moment is defined as the current cumulative pressure drop, and the pressure drop at the current dust removal moment is defined as the current pressure drop. The current cumulative pressure drop is used to characterize the cumulative degree of flow obstruction before the current dust removal moment, and the current pressure drop is used to characterize the recovery range after the current dust removal moment. Based on the current cumulative pressure drop and the cumulative pressure drop corresponding to the previous ash removal time, the change in the current cumulative pressure drop relative to the previous ash removal event is calculated. Based on the current pressure drop decrease and the pressure drop decrease corresponding to the previous ash removal time, the change in the current pressure drop decrease relative to the previous ash removal event is calculated. The change in the current cumulative pressure drop relative to the previous ash removal event is used to characterize the development and change of ash blockage, and the change in the current pressure drop decrease relative to the previous ash removal event is used to characterize the change in recovery capacity. Based on the current cumulative pressure drop and the current pressure drop decrease, calculate the difference between the current cumulative pressure drop and the current pressure drop decrease, and calculate the ratio between the current cumulative pressure drop and the current pressure drop decrease. The difference between the current cumulative pressure drop and the current pressure drop decrease is used to characterize the deviation between the degree of ash blockage development and the recovery range after ash removal, and the ratio between the current cumulative pressure drop and the current pressure drop decrease is used to characterize the pressure drop release relationship. The current cumulative pressure drop, the current pressure drop decrease, the change in the current cumulative pressure drop relative to the previous dust removal event, the change in the current pressure drop relative to the previous dust removal event, the difference between the current cumulative pressure drop and the current pressure drop decrease, and the ratio of the current cumulative pressure drop to the current pressure drop decrease are arranged in the input order of xRFM to form the recovery characterization quantity. Based on the current cumulative pressure drop, current pressure drop decrease, the difference between the current cumulative pressure drop and the current pressure drop decrease, and the ratio between the current cumulative pressure drop and the current pressure drop decrease in the recovery characterization quantities, calculate the recovery boundary position corresponding to the current dust removal event. The recovery boundary position is used to characterize the interval between the target sector and the recovery boundary. Based on the location of the recovery boundary within the preset state segment, and combined with the change in the current cumulative pressure drop relative to the previous dust removal event and the change in the current pressure drop relative to the previous dust removal event, the recovery state value is calculated. The recovery state value is used to characterize the direction of change in the recovery capability of the target sector under the current dust removal event. The recovery status value is determined as the current sector recovery status, and the recovery characterization quantity is used as the input data of xRFM. The current sector recovery status is used to match the recovery boundary category by the sector voltage drop recovery soot blowing criterion method.

[0038] like Figure 5 As shown, S4 specifically refers to: The recovery characteristics are input into the xRFM input layer according to the current pressure drop accumulation, the current pressure drop reduction, the change of the current pressure drop accumulation relative to the previous dust removal event, the change of the current pressure drop reduction relative to the previous dust removal event, the difference between the current pressure drop accumulation and the current pressure drop reduction, and the ratio of the current pressure drop accumulation to the current pressure drop reduction. A recovery mapping layer composed of fully connected neurons is used to map the recovery representation quantity. The recovery mapping layer includes a pressure drop accumulation subgroup, a pressure drop fall subgroup, and a cumulative fall coupled subgroup. The pressure drop accumulation subgroup generates the pressure drop accumulation mapping result based on the current pressure drop accumulation amount and the change of the current pressure drop accumulation amount relative to the previous dust removal event. The pressure drop fall subgroup generates the pressure drop fall mapping result based on the current pressure drop fall amount and the change of the current pressure drop fall amount relative to the previous dust removal event. The cumulative fall coupled subgroup generates the cumulative fall coupled mapping result based on the difference between the current pressure drop accumulation amount and the current pressure drop fall amount and the ratio between the current pressure drop accumulation amount and the current pressure drop fall amount. The cumulative pressure drop mapping results, the pressure drop fall-back mapping results, and the cumulative fall-back coupled mapping results are aggregated into the recovery mapping features. The recovery mapping features are then input into the ash-blocking recovery hidden state layer composed of fully connected neurons. The recovery mapping features are compressed and represented to obtain the ash-blocking recovery hidden state vector. Based on the hidden state vector of ash blockage recovery, the current sector recovery state is assigned to the hidden state of high recovery ash blockage recovery, the hidden state of transitional recovery ash blockage recovery, and the hidden state of low recovery ash blockage recovery, and the assignment result of the hidden state of ash blockage recovery is obtained. The attribution result of the hidden state of the ash-blocking recovery is input into the criterion embedding layer composed of fully connected neurons. The correspondence between the attribution result of the hidden state of the ash-blocking recovery and the recovery boundary category is established in the criterion embedding layer. The criterion embedding layer then segments the recovery boundary according to the attribution result of the hidden state of the ash-blocking recovery, and obtains the boundary segmentation label. The recovery boundary category corresponding to the current sector recovery status is determined based on the boundary segment label, and the boundary segment label is used as the boundary selection parameter of the sector voltage drop recovery soot blowing criterion method.

[0039] When the hidden state attribution result of the ash-blocking recovery is input into the criterion embedding layer composed of fully connected neurons, the current sector recovery state is also input into the criterion embedding layer. The criterion embedding layer establishes a correspondence with the recovery boundary category based on the combination relationship between the hidden state attribution result of the ash-blocking recovery and the current sector recovery state, performs segmentation recognition of the recovery boundary, and outputs the boundary segmentation label.

[0040] like Figure 6 As shown, S5 specifically refers to: The sector voltage drop recovery soot blowing criterion method is connected to the output of xRFM as the criterion head, and the boundary segment label and the current sector recovery status are directly passed to the sector voltage drop recovery soot blowing criterion method as common inputs, so that the recovery boundary category represented by the boundary segment label and the current sector recovery status enter the same decision link. Based on the boundary segmentation label, the recovery boundary category is determined in the sector pressure drop recovery soot blowing criterion method, and the boundary comparison interval corresponding to the recovery boundary category is called. The boundary comparison interval is formed by the segmented boundary set preset according to the recovery boundary category. The segmented boundary set includes the starting boundary, intermediate boundary and ending boundary corresponding to the recovery boundary category. In the sector pressure drop recovery soot blowing criterion method, a recovery boundary matching layer is set up. The recovery boundary matching layer is composed of matching neurons corresponding to different recovery boundary categories. The matching neurons corresponding to the recovery boundary category are activated according to the recovery boundary category to form the current boundary matching pathway. Input the current sector recovery status into the current boundary matching path, and calculate the interval position of the current sector recovery status in the recovery boundary category based on the relative position of the current sector recovery status with the start boundary, intermediate boundary and end boundary, to obtain the boundary position matching result; Based on the boundary position matching results, the current sector recovery status is aligned with the recovery boundary category in segments, and the segment boundary position is determined by combining the boundary comparison interval and the segment boundary set. The segment boundary position is used to characterize the specific segment of the current sector recovery status within the recovery boundary category that is far from the recovery boundary. The current segment recovery boundary is generated based on the segment boundary location and recovery boundary category, and the current segment recovery boundary is determined as the judgment boundary under the recovery boundary category corresponding to the boundary segment label; The current segmented recovery boundary is passed to the triggering and determination layer of the sector pressure drop recovery soot blowing criterion method, and a correspondence is established between the current segmented recovery boundary and the current sector recovery state in the triggering and determination layer, which is used to determine the directional soot blowing trigger command and the maintenance operation command.

[0041] Based on the cumulative pressure drop and pressure drop amount corresponding to the current cleaning moment of the same target sector in multiple consecutive cleaning moments, which have the previous cleaning moment, these values ​​are compared sequentially with the cumulative pressure drop and pressure drop amount corresponding to the previous cleaning moment. The current sector recovery state where the current cumulative pressure drop amount corresponding to the cleaning moment is higher than the previous cumulative pressure drop amount corresponding to the cleaning moment and the current pressure drop amount corresponding to the cleaning moment is lower than the previous pressure drop amount corresponding to the cleaning moment is determined as the value basis for the end boundary. The current sector recovery state where the current cumulative pressure drop amount corresponding to the cleaning moment is lower than the previous cumulative pressure drop amount corresponding to the cleaning moment and the current pressure drop amount corresponding to the cleaning moment is higher than the previous pressure drop amount corresponding to the cleaning moment is determined as the value basis for the start boundary. The current sector recovery state that does not meet the above two comparison relationships is determined as the value basis for the intermediate boundary. Based on the value basis of the start boundary, the value basis of the intermediate boundary, and the value basis of the end boundary, a set of segmented boundaries corresponding to the recovery boundary category is formed.

[0042] When the current sector recovery status is input into the current boundary matching path, the recovery boundary position calculated based on the recovery characterization quantity is also input into the current boundary matching path. The segment boundary position is determined based on the boundary segment label, the current sector recovery status, and the correspondence between the recovery boundary position and the starting boundary, intermediate boundary, and ending boundary. Then, the current segment recovery boundary is generated based on the segment boundary position and the recovery boundary category.

[0043] like Figure 7 As shown, S6 specifically refers to: The current sector recovery status, the current segment recovery boundary, and the boundary segment label are input into the triggering decision layer of the sector voltage drop recovery soot blowing criterion method. The triggering decision layer includes directional soot blowing triggering decision neurons and maintenance operation decision neurons. Under the current segmented recovery boundary constraints, the trigger interval corresponding to the current segmented recovery boundary is determined according to the boundary segment label, and the trigger judgment value is calculated according to the segment position and boundary direction of the current sector recovery status within the trigger interval; The trigger determination value is compared with the trigger interval. When the trigger determination value reaches the boundary condition corresponding to the trigger interval, the directional cleaning trigger determination neuron is activated, and the determination result of the directional cleaning trigger determination neuron is determined as the directional cleaning trigger command. When the trigger judgment value does not meet the boundary conditions corresponding to the trigger interval, the maintain operation judgment neuron is activated, and the judgment result of the maintain operation judgment neuron is determined as the maintain operation instruction.

[0044] The present invention will be further explained and illustrated below with reference to specific embodiments.

[0045] like Figure 8 As shown, the continuous operation process of the same sector is represented by time strips, and the pressure drop record is divided into two segments before and after ash removal, using the ash removal point as the event point. Before ash removal, the pressure drop is represented by cumulative blocks stacked grid by grid, accumulating only the positive increase portion to obtain the cumulative pressure drop amount, which is used to characterize the degree of flow obstruction caused by ash blockage development between two ash removals. After ash removal, the cumulative pressure drop process is represented by release blocks released grid by grid, ending the statistics when the first stop point is encountered, obtaining the pressure drop reduction amount, which is used to characterize the recovery magnitude brought about by this ash removal. Through this event-aligned feature construction, the present invention incorporates ash blockage development and ash removal recovery into the same ash removal event for joint characterization, avoiding short-term fluctuation interference, improving the physical interpretability of features and the comparability between sectors, providing stable input for subsequent recovery boundary segmentation identification and directional ash removal triggering, thereby reducing false triggering and invalid ash removal, and improving operational stability.

[0046] In this embodiment, step S1 specifically includes: The sectors within the cold end and transition zone of the rotary air preheater, defined by sector division and with recorded pressure drop changes and dust removal times, are designated as target sectors of the cold end and transition zone of the rotary air preheater, and denoted as follows: .in, Indicates the total number of target sectors. Indicates the first Target sector.

[0047] For each target sector Record the pressure drop changes during continuous operation. Record of dust removal time .in, Indicates the target sector The number of records of pressure drop changes Indicates the target sector The number of dust removal times, Indicates the first Record the time corresponding to the change in pressure drop. Indicates the first Record the pressure drop value corresponding to each pressure drop change. Indicates the first A moment to clean up the dust.

[0048] Will and Sort them in ascending order of time, and... The start time of continuous operation will be determined. The end time of continuous operation is determined to ensure that the voltage drop change records and dust removal time records within the same target sector have a unified time sequence.

[0049] For any target sector Each dust removal moment ,by The pressure drop change records are segmented using these points as dividing points. The extraction rule for the current segment's operating pressure drop is: when... At that time, the time falls The pressure drop records within the system are arranged in chronological order to form the pressure drop data for the initial operating phase. ;when At that time, the time falls The pressure drop records within the system are arranged in chronological order to form the pressure drop data for the initial operating phase. .in, Indicates the first Number of records of pressure drop changes before each dust removal moment. Indicates the first stage of pressure drop during the initial operation. The recording time of each record Indicates the first stage of pressure drop during the initial operation. The voltage drop value of each record.

[0050] The extraction rule for the voltage drop in the downstream operation is: when At that time, the time falls The pressure drop records within the system are arranged in chronological order to form the pressure drop during subsequent operation. ;when At that time, the time falls The pressure drop records within the system are arranged in chronological order to form the pressure drop during subsequent operation. .in, Indicates the first Number of records of pressure drop changes after each dust removal moment. Indicates the pressure drop in the later stage of operation. The recording time of each record Indicates the pressure drop in the later stage of operation. The pressure drop value recorded. (This refers to the same dust removal time.) The corresponding front-end operating voltage drop and back-end operating voltage drop are indexed by the same index. Binding is performed to generate sector timing voltage drop data. Sector timing voltage drop data It contains two fields: the front-end operating voltage drop and the back-end operating voltage drop. The front-end operating voltage drop is a one-dimensional voltage drop time series, and the back-end operating voltage drop is a one-dimensional voltage drop time series. Both fields retain the recording time and recording order.

[0051] target sector All sector timing voltage drop data are arranged in chronological order of dust removal time to form a sector voltage drop sequence. The sector voltage drop sequence is a sequence formed by arranging the preceding and following operating voltage drops at each cleaning moment in the same target sector according to the order of the cleaning moments. For , index as Sector timing voltage drop data With index Sector timing voltage drop data In the same target sector Establish a sequential association within the index. The sector timing voltage drop data corresponds to the previous dust cleaning time, with the index being... The sector timing voltage drop data corresponds to the current dust removal time. This sequential correlation allows the accumulated voltage drop to be calculated from... Extracting data in the order of records allows the pressure drop to be calculated from... The data is extracted sequentially from the records, ensuring that the change in the current cumulative pressure drop relative to the previous dust removal event and the change in the current pressure drop relative to the previous dust removal event have clear comparison objects. The resulting raw input is not a uniform pressure drop value for the entire machine, but rather input based on the target sector number. And dust removal time index Variable-length bisegment sequences Each term in the variable-length bisegment sequence contains two fields, which correspond to the pressure drop in the first segment and the pressure drop in the second segment, respectively, and provide a direct source for the construction of the pressure drop accumulation, pressure drop fall, recovery characterization and six-dimensional recovery characterization.

[0052] In this embodiment, step S2 specifically includes: For any target sector Directly call the sector voltage drop sequence Record of dust removal time Will satisfy Dusting time Determine the current dust removal time as having the previous dust removal time, and... This is determined to be the time of the previous dust removal. and Define the time boundary of the current dust removal event, and in Establish continuous operation analysis scope Continuous operation analysis range In the middle, located After and located Previous records were used to extract the cumulative pressure drop, located at Subsequent records are used to extract the pressure drop. Continuous operation analysis range. The purpose of this setting is not to create a unified statistical window for the entire machine, but to extract the pre-cleaning operation process and post-cleaning recovery process corresponding to the same cleaning time within the same event boundary. This is to enable indexing... The current cumulative pressure drop relative to the previous cleaning event and the current decrease in pressure drop relative to the previous cleaning event have a defined comparison benchmark for the index. The same record extraction process is executed separately at the first dust removal moment, continuously running from the start time to... The preceding operating pressure drop records are used to form the cumulative base pressure drop according to the same rules. and will from The initial subsequent operating pressure drop records are used to form the base pressure drop reduction amount according to the same rules. .

[0053] Will Pressure drop in the middle and front sections Determined as the cumulative pressure drop range .for According to the order of the records and Compare. If Then extract the first Increase in pressure drop between adjacent recorded values ;like Then the first The increase in pressure drop corresponding to adjacent recorded values ​​is recorded as 0. All increases in pressure drop are recorded as follows. Accumulate the records in order to obtain the current dust removal time. Corresponding cumulative pressure drop Cumulative pressure drop This is the cumulative increase in voltage drop between adjacent records in the sector voltage drop sequence from the previous cleaning time to the current cleaning time, arranged in the order of recording. Accumulated voltage drop interval. When there is only one record, the cumulative pressure drop will be calculated. Set to 0. Accumulated voltage drop. Reflecting the target sector The cumulative degree of flow obstruction caused by ash blockage development between two ash removal events. By accumulating only the positive increase in adjacent records, the pressure drop development process before the ash removal event can be preserved as a physically meaningful cumulative amount, rather than directly equating local fluctuations with ash blockage development.

[0054] exist Pressure drop during mid-to-late stage operation Internal, self-index Search for the first record that satisfies the condition in the order of records. The record location is recorded, and the corresponding index is recorded as... .when If it exists, index the range The records within this range were determined to be the pressure drop range. ;when If it does not exist, all subsequent operating pressure drop records will be defined as the pressure drop reduction range. In the pressure drop range Within the system, adjacent record values ​​are compared in the order they are recorded. If the previous record value is higher than the next record value, the corresponding voltage drop is extracted. If the previous recorded value is not higher than the next recorded value, the pressure drop range will not be extended further. The pressure drop range will then be extended. The total pressure drop already extracted Accumulate the records in order to obtain the current dust removal time. Corresponding pressure drop Pressure drop This is the cumulative amount of voltage drop between adjacent records in the sector voltage drop sequence from the current dust removal time until the moment when the first value not lower than the previous record appears in the sector voltage drop sequence, calculated in the order of recording. Voltage drop recovery interval. When only one record is included, the pressure drop will be calculated. Set to 0. Pressure drop amount Reflecting the target sector The recovery range after the current dust removal action. Using "the first value not lower than the previous record value" as the stopping condition for the pressure drop range can separate the pressure drop release process directly caused by the dust removal action from the re-accumulation of dust or disturbance process after the dust removal action ends, thus ensuring that the pressure drop amount only represents the recovery range after dust removal.

[0055] Index at the same dust removal time Below, the cumulative pressure drop will be... With pressure drop Bind the processes to the same dust removal event, so that the dust blockage development process and the dust removal recovery process enter the same calculation link; then accumulate the pressure drop corresponding to the previous dust removal time. The pressure drop at the time of the previous dust removal Cumulative pressure drop corresponding to the current dust removal moment The pressure drop at the current dust removal moment Align according to the order of dust removal time. For each Event comparison vectors are formed according to a fixed field order. Event comparison vector This is a four-dimensional vector. The first dimension represents the cumulative pressure drop at the previous cleaning time, the second dimension represents the pressure drop reduction at the previous cleaning time, the third dimension represents the cumulative pressure drop at the current cleaning time, and the fourth dimension represents the pressure drop reduction at the current cleaning time. The recovery characteristic quantity calculation directly calls the event comparison vector. The four fields in the data are used to generate a six-dimensional recovery representation. Six-dimensional recovery representation quantity The six fields are, in order: current cumulative pressure drop, current pressure drop decrease, change in current cumulative pressure drop relative to the previous dust removal event, change in current pressure drop relative to the previous dust removal event, difference between current cumulative pressure drop and current pressure drop decrease, and ratio of current cumulative pressure drop to current pressure drop decrease. This is a six-dimensional recovery characterization. Each neuron in the xRFM input layer corresponds to one of the six neurons. xRFM uses the six-dimensional recovery representation corresponding to a single dust removal event as an input unit, and completes the recovery boundary segmentation recognition through the recovery mapping layer, the dust-blocking recovery hidden state layer, and the criterion embedding layer.

[0056] In this embodiment, step S3 specifically includes: For any target sector The current dust removal event directly calls the accumulated pressure drop corresponding to the previous dust removal time. The pressure drop at the time of the previous dust removal The cumulative pressure drop at the current dust removal moment The pressure drop corresponding to the current dust removal moment .Will This is determined as the current cumulative pressure drop. This is determined to be the current pressure drop. Current cumulative pressure drop. This characterizes the cumulative degree of flow obstruction prior to the current dust removal moment, and the current pressure drop. Characterizes the recovery magnitude after the current dust removal moment. Utilizes the current cumulative pressure drop. Subtract the cumulative pressure drop corresponding to the previous dust removal time. This method forms the change in the current cumulative pressure drop relative to the previous dust removal event. ; Using the current pressure drop amount Subtract the pressure drop at the time of the previous dust removal. This method forms the change in the current pressure drop relative to the previous dust removal event. . This indicates that the development of ash-blocking technology has intensified. This indicates that the development of ash-blocking technology has weakened. This indicates that the body's recovery ability is enhanced after cleaning. This indicates a weakened recovery capability after dust removal. The current cumulative pressure drop is used. Subtract the current pressure drop This method forms the difference between the current cumulative pressure drop and the current pressure drop decrease. ; Using the current cumulative pressure drop Divide by the current pressure drop This method forms the ratio of the current cumulative pressure drop to the current decrease in pressure. .when and At that time, Set to 1; when and At that time, Set to preset upper limit Preset upper limit Take 10.

[0057] Will , , , , and Arranged according to the input order of xRFM, the recovered characterization is formed. Recovery of characterization The xRFM is a six-dimensional vector. The first dimension represents the current cumulative pressure drop, the second dimension represents the current pressure drop decrease, the third dimension represents the change in the current cumulative pressure drop relative to the previous dust removal event, the fourth dimension represents the change in the current pressure drop decrease relative to the previous dust removal event, the fifth dimension represents the difference between the current cumulative pressure drop and the current pressure drop decrease, and the sixth dimension represents the ratio between the current cumulative pressure drop and the current pressure drop decrease. xRFM uses one recovery characteristic quantity corresponding to one dust removal event. As an input unit, the recovered characterization values ​​are received sequentially in ascending order of the dust removal time. The recursive state is not propagated between adjacent dust removal events. For the same target sector All valid dust removal events are sorted in ascending order by dust removal time to obtain the sequence of recovered characterization quantities. Recover the characterization sequence Each six-dimensional vector in the xRFM can be directly fed into the xRFM input layer.

[0058] To ensure that the recovery boundary position directly corresponds to the core field in the recovery representation quantity, before calculating the recovery boundary position corresponding to the current dust removal event, each dust removal event completed before the current dust removal time is analyzed. Formed according to the exact same rules as the current dust removal event. , , and Record the cumulative historical pressure drop. After sorting the values ​​from smallest to largest, the median value is taken to obtain the cumulative voltage drop reference threshold. Record the historical drop in pressure. Sort the values ​​from smallest to largest and take the median value to obtain the reference threshold for voltage drop reduction. Record the absolute value of the historical difference. After sorting the values ​​from smallest to largest, the median value is taken to obtain the difference reference threshold. Record historical ratio values Sort the values ​​in ascending order and take the median value to obtain the ratio reference threshold. When the number of historical records is even, the median is the arithmetic mean of the two middle numbers. , , or When any value in the record is 0, the corresponding value is replaced with the smallest positive value in the same historical record; if no smallest positive value exists in the same historical record, the corresponding value is set to 1. Based on the above reference threshold, the recovery boundary position corresponding to the current dust removal event is recorded as... And calculate using the following formula:

[0059] In the formula, Indicates the target sector The recovery boundary position under the current dust removal event; This indicates the current cumulative pressure drop; Indicates the target sector The cumulative reference threshold for voltage drop; This indicates the current pressure drop. Indicates the target sector The reference threshold for pressure drop reduction; This represents the difference between the current cumulative pressure drop and the current pressure drop decrease. Indicates the target sector The difference reference threshold; This represents the ratio of the current cumulative pressure drop to the current decrease in pressure. Indicates the target sector The proportional value reference threshold; Indicates the target sector index; Indicates the index of the current dust removal event; and This indicates the preset weighting coefficient.

[0060] Restore boundary position This is a one-dimensional continuous position code. Recover the boundary positions. At that time, the boundary position will be restored. Set to 0. Divide the preset state segment into the starting boundary segment. Intermediate boundary section and end boundary section Restore boundary position When it falls within the initial boundary segment, it indicates the target sector. The distance from the restoration boundary is relatively far; the location of the restoration boundary When it falls into the middle boundary segment, it indicates the target sector. At the boundary transition position; boundary restoration position When it falls into the end boundary segment, it indicates the target sector. It is nearing the end of the recovery boundary. The recovery state value is denoted as... When the boundary position is restored Falling into the initial boundary segment, and , At that time, Set to 1; when restoring the boundary position Falling into the end boundary segment, and , At that time, Set as In other cases Set to 0. Indicates the target sector Under the current dust removal event, recovery capabilities are shifting towards enhancement. Indicates the target sector Currently, the area is in a transitional recovery phase following the dust removal incident. Indicates the target sector Under the current dust removal event, the recovery ability is decreasing. [The recovery status value will be adjusted.] The current sector has been confirmed as being in recovery status.

[0061] In this implementation, xRFM employs a single-event forward inference structure. The input layer consists of six neurons, each receiving a recovered representation. The xRFM input consists of six fields; the recovery mapping layer has nine fully connected neurons, divided into three subgroups: pressure drop accumulation, pressure drop fallback, and accumulation fallback coupling, each with three neurons. The recovery mapping layer maps the six-dimensional recovery representation to a nine-dimensional recovery mapping feature; the gray-blocking recovery hidden state layer has six fully connected neurons, arranged in three pairs: high recovery, intermediate recovery, and low recovery. The gray-blocking recovery hidden state layer compresses the nine-dimensional recovery mapping feature into a six-dimensional gray-blocking recovery hidden state vector; the criterion embedding layer has three fully connected neurons, each corresponding to one of the three recovery boundary categories. The criterion embedding layer outputs three-dimensional boundary segment labels. xRFM has a 3-dimensional output, corresponding to three recovery boundary categories. Current sector recovery status. This is a one-dimensional state variable, used by the sector voltage drop recovery soot blowing criterion method. The resulting core input object is a six-dimensional recovery representation quantity. And the current sector recovery status in one dimension Six-dimensional recovery representation quantity Upon entering xRFM, xRFM performs forward calculations for each dust removal event, restoring the one-dimensional state of the current sector. Keep aligned with the same dust removal event.

[0062] In this embodiment, step S4 specifically includes: xRFM is a network that takes the recovered representation as input and sequentially passes it through a recovery mapping layer, a gray-blocking recovery hidden state layer, and a criterion embedding layer to segment and identify the recovered representation to obtain boundary segment labels. For any target sector The current dust removal event directly calls the recovery representation quantity. and the current sector recovery status Recovery of characterization It is a six-dimensional vector, with the first dimension being the current cumulative voltage drop. The second dimension represents the current pressure drop. The third dimension represents the change in the current cumulative pressure drop relative to the previous dust removal event. The fourth dimension represents the change in the current pressure drop relative to the previous dust removal event. The fifth dimension is the difference between the current cumulative pressure drop and the current pressure drop decrease. The sixth dimension is the ratio of the current cumulative pressure drop to the current decrease in pressure. Current sector recovery status. It is a one-dimensional state variable, taking the value of , or The xRFM input layer consists of six neurons, which receive data sequentially in a fixed field order. , , , , and xRFM performs independent forward reasoning for each cleaning event and does not propagate recursive state between adjacent cleaning events; therefore, the same target sector... Multiple dust removal events are called one by one in ascending order of dust removal time, without sharing the hidden state between events.

[0063] The input layer output enters the recovery mapping layer. The recovery mapping layer has nine fully connected neurons, divided into a voltage drop accumulation subgroup, a voltage drop fallback subgroup, and an accumulation-fallback coupling subgroup. Each subgroup contains three fully connected neurons. The voltage drop accumulation subgroup only receives the current accumulated voltage drop. The change in current cumulative pressure drop relative to the previous dust removal event And form a cumulative pressure drop mapping result. The pressure drop subgroup only receives the current pressure drop amount. The change in the current pressure drop relative to the previous dust removal event. And form a pressure drop mapping result. The cumulative drop-off coupling subgroup only receives the difference between the current cumulative voltage drop and the current voltage drop decrease. The ratio of the current cumulative pressure drop to the current pressure drop decrease. And form a cumulative fallback coupling mapping result. Among them, superscript Indicates the cumulative voltage drop subgroup, superscript Indicates pressure drop back to the drop group, superscript This represents the cumulative fallback coupling subgroup. Each fully connected neuron in each of the three subgroups performs a weighted sum of its corresponding input fields based on the connection and bias parameters fixed after training, and outputs a scalar. , and By concatenating the subgroups in the order of "accumulated pressure drop subgroup, pressure drop fallback subgroup, and cumulative fallback coupling subgroup", the recovered mapping features are obtained. Recover mapping features It is a nine-dimensional vector, with nine fields corresponding to the output values ​​of the three neurons in the three subgroups.

[0064] Restore mapping features Input the ash-blocking recovery hidden state layer. The ash-blocking recovery hidden state layer consists of six fully connected neurons, arranged in pairs according to the order of high-recovery, intermediate-recovery, and low-recovery hidden states, resulting in the ash-blocking recovery hidden state vector. Among them, superscript Indicates that the high-recovery ash plugging has returned to a hidden state, indicated by the superscript. Indicates transition recovery from ash plugging to hidden state, superscript This indicates that the low-recovery ash blockage has been restored to a hidden state. and Take the arithmetic mean to obtain the hidden state score for high-recovery ash plugging recovery; and Take the arithmetic mean to obtain the transition recovery and hidden state recovery score; and The arithmetic mean is taken to obtain the score for the low-recovery ash-blocking hidden state recovery. The three ash-blocking hidden state recovery scores are compared, and the score with the highest value corresponds to the ash-blocking hidden state assignment result. When two or three ash-blocking hidden state recovery scores are the same, the assignment result is determined in the order of low-recovery ash-blocking hidden state, transitional-recovery ash-blocking hidden state, and high-recovery ash-blocking hidden state. The ash-blocking hidden state assignment result is encoded into a three-dimensional ash-blocking hidden state indicator vector. High recovery ash plugging recovery hidden state corresponding to Transition recovery, dust plugging recovery, hidden state corresponding to Low recovery ash blockage recovery hidden state corresponding .

[0065] Restore the hidden state indicator vector of 3D ash filling Recovery status of the current sector The input vectors are fed into the criterion embedding layer. The criterion embedding layer uses three fully connected neurons, corresponding to the initial recovery boundary category, the intermediate recovery boundary category, and the final recovery boundary category, respectively. The input vector of the criterion embedding layer is denoted as... The input dimension is four-dimensional. To explicitly write the combination relationship between the hidden state attribution result of ash-blocking recovery and the current sector recovery state in the criterion embedding layer, the initial recovery boundary category matching value is first calculated. Intermediate recovery boundary category matching value Matching values ​​with end recovery boundary category Then, boundary segment labels are generated based on the three recovery boundary category matching values. The criterion embedding layer adopts the following segmentation recognition rules:

[0066] In the formula, Indicates the initial recovery boundary category matching value; This indicates the intermediate recovery boundary category matching value; This indicates the end of the recovery boundary category matching value; Represents the hidden state indicator vector for 3D ash plugging recovery The middle position corresponds to the indicator bit for restoring the hidden state of high-level ash blockage; Represents the hidden state indicator vector for 3D ash plugging recovery The corresponding indicator bit in the middle indicates the transition recovery and restoration of the hidden state after ash blockage. Represents the hidden state indicator vector for 3D ash plugging recovery The middle position corresponds to the indicator bit indicating the recovery of the hidden state from the low-level ash blockage. Indicates the current sector recovery status; Indicates the boundary segment label; This indicates the operation of encoding the category containing the maximum value as a one-hot label; This indicates selecting the category index with the largest matching value from a given set of categories; Indicates the recovery of the boundary category index; Indicates the target sector index; Indicates the index of the current dust removal event; superscript Indicates that the high-recovery ash plugging has returned to its hidden state; superscript Indicates transition recovery from ash plugging to hidden state; superscript Indicates that the low-recovery ash plugging has returned to a hidden state; superscript Indicates the initial recovery boundary category; superscript Indicates the intermediate recovery boundary category; superscript This indicates the end of the recovery boundary category.

[0067] In the formula , , For dimensionless indicator values ​​that take the value of 0 or 1, For the value to be , or The dimensionless state value, therefore the three recovery boundary category matching values , and All are dimensionless values ​​that can be directly compared. The hidden state indicator vector is recovered from the 3D ash-filling process. Unique heat properties and current sector recovery status The three-valued attributes jointly define that each current dust removal event corresponds to a unique primary matching direction. When When two maximum values ​​are found, the category index is determined in the order of end recovery boundary category, intermediate recovery boundary category, and start recovery boundary category. Boundary segment label. For a three-dimensional vector, it is written as When the initial recovery boundary category match value is the largest, the boundary segment label... Recorded as When the intermediate recovery boundary category matching value is the maximum, the boundary segment label... Recorded as When the boundary category matching value is at its maximum at the end of the recovery process, the boundary segment label is... Recorded as Boundary segmentation labels Used to characterize the current sector recovery status The corresponding recovery boundary category is used as the boundary selection parameter for the sector voltage drop recovery soot blowing criterion method.

[0068] For the same target sector All valid dust removal events are sorted in ascending order by dust removal time to form a six-dimensional recovery characterization sequence. xRFM for recovering six-dimensional characterization sequences Each recovery characterization in Performing a forward inference independently yields the following in sequence: 3D reconstruction mapping features, 6D ash-blocking reconstruction hidden state vector, 3D ash-blocking reconstruction hidden state indicator vector, and 3D boundary segmentation label. Therefore, the single-event input of xRFM is a six-dimensional recovery representation. The single-event output of xRFM is a 3D boundary segment label. The criteria for determining the one-dimensional current sector recovery state received by the embedded layer are as follows: Index of the same dust removal event Keep them aligned.

[0069] In this embodiment, step S5 specifically includes: The sector voltage drop recovery soot blowing criterion method is connected to the output of xRFM as the criterion header. For any target sector The current dust removal event directly calls the boundary segmentation label. Current sector recovery status , restore boundary position And historical event records of the same target sector that have been determined before the current dust removal event. .in, The first dimension represents the boundary segment label, and the first dimension corresponds to the initial recovery boundary category; The second dimension represents the boundary segment label, and the second dimension corresponds to the intermediate restored boundary category; The third dimension represents the boundary segment label, and the third dimension corresponds to the end of the boundary recovery category. This represents the index of historical soot blowing events. The input vector received by the sector voltage drop recovery soot blowing criterion method is denoted as... The input dimension is four-dimensional. Restore the boundary position. As an auxiliary matching quantity called internally within the recovery boundary matching layer, it does not change the four-dimensional interface structure between the sector voltage drop recovery soot blowing criterion and xRFM. The sector voltage drop recovery soot blowing criterion includes a recovery boundary matching layer and a trigger determination layer. The recovery boundary matching layer has three matching neurons, which correspond to the starting recovery boundary category, the intermediate recovery boundary category, and the ending recovery boundary category, respectively. The trigger determination layer has two determination neurons, which correspond to the directional soot blowing trigger command and the maintenance operation command, respectively.

[0070] To form a set of segmented boundaries corresponding to the recovery boundary categories, historical event records are first aggregated according to the recovery boundary categories. The recovery boundary category index is denoted as... ,in Corresponding to the initial recovery boundary category, Corresponding to the intermediate recovery boundary category, The corresponding end-of-recovery boundary category. For any category Filter out those that meet the requirements from historical event records and , Current sector recovery status The initial boundary value sequence is formed in ascending order according to the dust removal time; values ​​that meet the criteria are then selected. and , Current sector recovery status The final boundary value sequence is formed by sorting the dust removal times in ascending order; values ​​that meet the criteria are then selected. And the current sector recovery state does not fall into either of the aforementioned two comparison relationships. The intermediate boundary values ​​are formed in ascending order according to the dust removal time. The starting boundary, intermediate boundary, and ending boundary are denoted as follows: , and superscript Indicates the starting boundary, superscript Indicates the middle boundary, superscript This indicates the end boundary. When the number of records in the three sequences is odd, the middle value after sorting is taken as the corresponding boundary; when the number of records is even, the arithmetic mean of the two middle values ​​after sorting is taken as the corresponding boundary. The starting boundary value sequence is empty when... Set to 1; when the intermediate boundary value sequence is empty, Set to 0; when the end boundary value sequence is empty, Set as .when , and If the descending order is not satisfied, the three values ​​are reordered in descending order and assigned the starting boundary, middle boundary, and ending boundary values ​​sequentially. , and Forming recovery boundary categories Corresponding segment boundary set .

[0071] In forming a set of segmented boundaries Simultaneously, reference values ​​for restoring boundary positions are extracted from the same batch of historical event records. For any category Extract and restore boundary positions from historical events that satisfy the initial boundary comparison relationship. The starting position reference value is obtained by arranging the values ​​in ascending order and taking the median value. Extract and restore boundary positions from historical events that satisfy intermediate boundary comparison relationships. Arrange the values ​​in ascending order and take the median value to obtain the reference value at the middle position. Extract and restore boundary positions from historical events that satisfy the end boundary comparison relationship. After arranging the values ​​in ascending order, take the median value to obtain the reference value for the ending position. Starting position reference value Set to when missing Reference value at the middle position Set to when missing End position reference value Set to when missing .when , and If the ascending order is not satisfied, the three values ​​are reordered in ascending order and assigned the following reference values ​​in sequence: starting position reference value, middle position reference value, and ending position reference value. and Forming recovery boundary categories Corresponding boundary comparison interval .

[0072] In the current dust removal event determination process, the boundary segmentation labels are used first. Activate the matching neurons in the restored boundary matching layer that correspond to the restored boundary category. Boundary segment labels. At that time, the initiation recovery boundary category matching neuron is activated; boundary segmentation label At that time, the intermediate recovery boundary category matching neurons are activated; boundary segmentation labels are then applied. At this point, the activated boundary category matching neuron is activated. The activated matching neuron forms the current boundary matching pathway and reads the corresponding category. Piece boundary set Compare intervals with boundaries Then restore the current sector to its current state. and restore boundary position Input the current boundary matching path and calculate the interval position of the current sector recovery status within the recovery boundary category. Denote the segment boundary position as... , It is a one-dimensional location code. Indicates the starting segment. Indicates the middle section. Indicates the end of the paragraph. At that time, Set to 1; when At that time, Set to 3; when At that time, if and Then Set to 1; if and Then Set to 3; except for the two cases mentioned above, Set to 2. (By...) Achieve a unified expression of boundary position matching results and segment boundary positions.

[0073] Let the current segmented recovery boundary be denoted as , These are one-dimensional boundary values. The location of the segmented boundary. At that time, from the segment boundary set Read the starting boundary And assign to Segment boundary location At that time, from the segment boundary set Reading the intermediate boundary And assign to Segment boundary location At that time, from the segment boundary set Read the end boundary And assign to This yields the current segmented recovery boundary. Segmentation labels by boundary Select the recovery boundary category, and then restore the state from the current sector. and restore boundary position The specific boundary segments are jointly located, rather than using a unified fixed boundary for the entire machine.

[0074] Trigger the decision layer with decision vector The input dimension is two-dimensional. The trigger determination result is denoted as... ,in This indicates the output bit for the directional dust removal trigger command. Indicates the output bit of the execution instruction to be maintained. Segment boundary position. or At that time, if Then Set as ;like Then Set as Segment boundary location At that time, if Then Set as ;like Then Set as The resulting prediction process is a single-event forward decision process: xRFM outputs three-dimensional boundary segment labels for each dust removal event. The sector voltage drop recovery soot blowing criterion method reads the four-dimensional input vector for the same soot blowing event. And internally call to restore boundary position The boundary matching layer restores the one-dimensional piecewise boundary position. And one-dimensional current piecewise recovery boundary The decision layer is then triggered to output a two-dimensional decision result. For the same target sector For multiple dust removal events, the above judgment process is executed sequentially in ascending order of dust removal time. Each dust removal event completes a forward judgment independently, without propagating the hidden state between events. After the current dust removal event completes its judgment, the accumulated voltage drop, voltage drop reduction, boundary segment label, current sector recovery status, and recovery boundary position of the current dust removal event are added to the historical event record for use in updating the segment boundary set for the next dust removal event. Compare intervals with boundaries .

[0075] In this embodiment, step S6 specifically includes: The triggering layer of the sector voltage drop recovery soot blowing criterion method receives the current sector recovery status corresponding to the same current soot blowing event. Current segmented recovery boundary and boundary segmentation labels .in, This indicates the label bit corresponding to the initial recovery boundary category. This indicates the label bit corresponding to the intermediate recovery boundary category. This indicates the label bit corresponding to the end of the recovery boundary category. The internal input vector that triggers the decision layer is denoted as... Internal input vector The vector is a five-dimensional vector. The first dimension represents the current sector recovery state, the second dimension represents the current segment recovery boundary, and the third to fifth dimensions represent the three category bits of the boundary segment label. The trigger decision layer contains only two decision neurons: a directional dust removal trigger decision neuron and a maintenance operation decision neuron. The output vector of the trigger decision layer is denoted as... ,in, This indicates the output bit for the directional dust removal trigger command. This indicates that the output bit of the running instruction is maintained.

[0076] In the trigger determination layer, segmentation labels are first determined based on the boundaries. Determine the current segmented recovery boundary The corresponding trigger interval. Let the trigger interval be denoted as... Current sector recovery status. The state axis is fixed as , , ,in This indicates a change in recovery ability towards enhancement. Indicates a transitional recovery state. This indicates a shift in recovery capacity towards weakening, with the boundary direction fixed as from... point to .when At that time, the current dust removal event corresponds to the initial recovery boundary category and the triggering interval. The definition is all state segments on the state axis that satisfy the condition that "the current sector recovery state is lower than the current segment recovery boundary"; when At that time, the current dust removal event corresponds to the intermediate recovery boundary category, triggering interval. It is still determined to be all state segments on the state axis that satisfy the condition that "the current sector recovery state is lower than the current segment recovery boundary"; when At that time, the current dust removal event corresponds to the end and recovery boundary category, triggering the interval. The state segments on the state axis are defined as those that satisfy the condition that "the current sector recovery state is lower than or equal to the current segment recovery boundary".

[0077] In the trigger zone Once confirmed, restore the current sector status. Generate trigger judgment values ​​at the segment positions and boundary directions in the state axis. .when When this happens, the judgment value will be triggered. Set to 0, and determine the current sector recovery status as the high-side segment located at the current segment recovery boundary; when When this happens, the judgment value will be triggered. Set to 1, and determine the current sector recovery status as the boundary point segment located at the current segment recovery boundary; when When this happens, the judgment value will be triggered. Set to 2, and determine the current sector recovery status as the lower-side segment located at the current segment recovery boundary. Trigger determination value. The calculation only calls the current sector to restore the state. and the current segmented recovery boundary No additional intermediate variables will be introduced.

[0078] The initial recovery boundary category and the intermediate recovery boundary category use boundary conditions without boundary points, so the decision value is triggered only when the boundary condition is met. At that time, it is determined that the current sector recovery state has entered the trigger interval. The end of the recovery boundary category uses boundary conditions that include boundary points, therefore, when the judgment value is triggered... Or trigger judgment value At that time, it is determined that the current sector recovery state has entered the trigger interval. "Restore boundary in the current segment" Under constraints, trigger decision value Satisfy trigger interval The corresponding boundary condition is determined as the trigger condition. When the trigger condition is met, the directional cleaning trigger determination neuron is activated, and the output vector is... Set as If the triggering condition is not met, the neuron that determines the operation is activated, and the output vector is changed. Set as .

[0079] when When the current dust removal event is determined, the result is set as the directional dust removal trigger command; when At that time, the determination result of the current dust removal event is set as the maintain operation command. For the same target sector Multiple dust removal events are processed, and the internal input vectors are read one by one in ascending order of dust removal time. Generate trigger judgment values ​​one by one and output vector Each dust removal event completes a forward decision independently, without propagating the hidden state between dust removal events. By placing the boundary segment label, the current segment recovery boundary, and the current sector recovery state into the same trigger decision chain, the targeted dust removal trigger decision no longer relies on a unified threshold for the entire machine, nor on the black-box conclusion directly output by the model. This directly solves the core technical problem that existing technologies cannot accurately determine when to trigger targeted dust removal based on changes in recovery capability at the sector level.

[0080] In summary, this invention directly addresses the aforementioned technical problem through a closed-loop technical approach that combines pressure drop characterization based on dust removal events, hidden state segmentation identification, segment boundary constraint determination, and historical self-updating. In the sector operation scenarios of the cold end and transition zone of a rotary air preheater, dust blockage development and dust removal recovery are affected by ash load distribution, temperature condensation conditions, and airflow bias, resulting in sector differences and event uncertainties. If a uniform threshold for the entire unit or single-point pressure drop changes are still used as the basis, it can easily lead to premature or delayed directional dust removal triggering, causing continuous accumulation of flow resistance or ineffective dust removal disturbances. This invention, constrained by only the sector pressure drop change records and dust removal time records available on-site, constructs a sector pressure drop sequence with the dust removal time as the dividing point, extracts the cumulative pressure drop before dust removal and the pressure drop reduction after dust removal, and forms a six-dimensional recovery characterization quantity through the difference and ratio of the change quantity. At the same time, it calculates the recovery boundary position and obtains the current sector recovery state, and then establishes a dust blockage recovery hidden state mechanism by xRFM and outputs boundary segment labels, so that the key intermediate quantity, i.e. the current segment recovery boundary, can be reliably located. Subsequently, under the constraint of the segment recovery boundary, the current sector recovery state is triggered and a directional dust removal trigger command or a maintenance operation command is output. The judgment result is then written back to the historical event to update the segment boundary set, thereby realizing a closed-loop deliverable sector-level directional dust removal trigger.

[0081] Compared to common timed and sequential strategies that use fixed threshold criteria and directly output trigger conclusions, this invention makes targeted improvements to the algorithm structure to address sector differences and the uncertainty of dust removal recovery, thereby enhancing the aforementioned closed-loop effect. It accumulates only the positive increases of adjacent records before dust removal and limits the decline interval to the first moment after dust removal when the decline stops, ensuring that the amount of dust blockage development and the amount of dust removal release have clear physical directions and avoiding short-term fluctuations from influencing the recovery assessment. It uses xRFM to segment and identify the recovery characteristics and incorporates the combination relationship between hidden state affiliation and recovery state into the criterion embedding, making the recovery boundary no longer a globally fixed threshold but segmented and adapted according to the recovery capability category. It generates the current segmented recovery boundary and completes the trigger interval determination under the boundary segment label constraint through the criterion header, making the decision chain from pressure drop record to boundary location to trigger command a traceable intermediate link, thus improving the adaptability and consistency of directional dust removal triggering under different sector and dust blockage morphology conditions.

[0082] like Figure 9As shown, using the "dust removal moment" as the event anchor point, the pressure drop record of the same target sector is divided into a pre-dust removal sequence and a post-dust removal sequence. Before dust removal, the positive increment of pressure drop is accumulated to obtain the "cumulative pressure drop," which is used to characterize the intensity of dust blockage development. After dust removal, the drop range is accumulated to obtain the "pressure drop drop," which is used to characterize the dust removal recovery capability. This event-aligned feature construction decouples operational fluctuations from trend changes, making the dust blockage-recovery behavior under different sectors and operating conditions comparable and interpretable. This provides stable input for subsequent segmented recovery boundary identification and sector-level directional triggering, thereby improving the robustness of the criterion and reducing false triggering.

[0083] like Figure 10 As shown in the figure, the horizontal axis represents the cumulative pressure drop (clogging development), and the vertical axis represents the pressure drop reduction (cleaning recovery). Scattered points are grouped according to the recovery latent state (high / transitional / low recovery), and grayscale density and contour lines depict high-probability areas where events occur. Dashed boundaries and diagonally filled areas represent insufficient recovery triggering domains; that is, areas with smaller drop under the same cumulative level should be more likely to trigger directional cleaning. This visualization demonstrates that this invention does not use a uniform threshold for the entire machine, but rather achieves sector-level, state-adaptive triggering based on xRFM state recognition and segmented recovery boundary matching. This concentrates triggering on areas that truly require cleaning, reducing ineffective soot blowing and energy consumption / wear, and improving triggering accuracy and operational stability.

[0084] In summary, this method addresses the difficulty in accurately determining directional cleaning triggers due to differences in ash blockage morphology and recovery capabilities among different sectors in the cold end and transition zone. By acquiring pressure drop change records and cleaning time records during continuous operation of each target sector, a sector pressure drop sequence is constructed. The cumulative pressure drop before cleaning and the pressure drop reduction after cleaning are extracted, and a recovery characterization quantity is constructed, including the cumulative amount, the reduction amount, the difference in changes between events, and the ratio. The current sector recovery status is then calculated. The recovery characterization quantity is input into xRFM for recovery boundary segmentation identification to obtain boundary segment labels. These boundary segment labels, along with the current sector recovery status, are input into the sector pressure drop recovery soot blowing criterion to generate the current segmented recovery boundary. Under this boundary constraint, a directional cleaning trigger command or a maintenance operation command is determined and output. This invention achieves segmented adaptation of sector-level recovery boundaries, improving the applicability of directional cleaning triggers.

[0085] Secondly, by constructing a sector pressure drop sequence with the dust removal time as the dividing point, and extracting the cumulative pressure drop before dust removal and the pressure drop reduction after dust removal within each dust removal event, a six-dimensional recovery characterization quantity is formed by combining the changes in adjacent dust removal events and the cumulative drop difference and ratio. Simultaneously, the recovery boundary position is calculated to obtain the current sector recovery state. This technique, relying solely on the available on-site pressure drop change records and dust removal time records, incorporates the dust blockage development process and the dust removal recovery process into the same calculation chain, avoiding misjudgments caused by using only single-moment pressure drop or short-term slopes. This allows an interpretable correspondence between the recovery amplitude after dust removal and the degree of dust blockage accumulation before dust removal, thus providing a stable intermediate quantity basis for subsequent boundary determination. Furthermore, the six-dimensional recovery characterization quantity is input into the recovery mapping layer and the dust blockage recovery hidden state layer to form a hidden variable representation of the differences in different dust blockage forms and recovery capabilities. In the criterion embedding layer, the hidden state attribution result is correlated with the current sector recovery state, and boundary segment labels are output to characterize the recovery boundary category that the current sector should match. Compared to existing solutions that rely on fixed thresholds or uniform boundaries for judgment, and data-driven solutions that directly provide black-box trigger conclusions, this proposal achieves categorized selection of recovery boundaries through boundary segmentation labels. This allows sectors with different recovery capabilities to adapt to different boundary segmentation rules under the same business scenario, improving the usability and consistency of sector-level criteria under conditions of varying operating conditions and uncertainty related to dust blockage. Finally, by selecting the recovery boundary category based on the boundary segmentation label in the criterion header, and combining it with historical dust removal events to form a set of segmented boundaries under that category, the current sector recovery status is triggered and judged under the constraints of the current segmented recovery boundary, outputting a directional dust removal trigger command or a maintain operation command. This overall approach eliminates reliance on a uniform threshold for the entire system for critical decisions, and avoids completely delegating the triggering logic to the model's direct output. Instead, it uses the category results obtained from boundary segmentation identification to constrain the generation of criterion boundaries and the determination of trigger intervals, thereby achieving a closed-loop delivery at the sector scale from voltage drop recording to recovery status judgment to dust removal command output. This effectively addresses the on-site determination problem of when directional dust removal should be triggered under different sector recovery capabilities.

[0086] The second objective of this invention is to propose a rotary air preheater directional cleaning triggering system based on xRFM-driven recovery boundary segmentation, comprising: Sector pressure drop sequence module: used to acquire the pressure drop change records and dust removal time records of each target sector in the cold end and transition zone of the rotary air preheater during continuous operation, and to obtain the sector pressure drop sequence based on the pressure drop change records; Pressure drop recovery module: used to extract the cumulative pressure drop before each cleaning moment and the pressure drop recovery after each cleaning moment based on the sector pressure drop sequence and cleaning time record; Sector recovery status module: Used to calculate the recovery characterization quantity, which characterizes the relationship between pressure drop evolution and post-cleaning recovery, based on the cumulative pressure drop and the pressure drop reduction, and to determine the current sector recovery status based on the recovery characterization quantity; Boundary segmentation label module: It is used to input the restored characterization quantity into xRFM, and to segment and identify the restored characterization quantity according to the restoration boundary corresponding to different ash-blocking restoration hidden states to obtain the boundary segmentation label; Segmented recovery boundary module: Used to input the boundary segment label into the sector voltage drop recovery soot blowing criterion method, and match the current sector recovery status to the recovery boundary category corresponding to the boundary segment label to obtain the current segmented recovery boundary; Directional dust removal trigger instruction module: Under the current segment recovery boundary constraints, it determines the current sector recovery status and the current segment recovery boundary. If the trigger conditions are met, it generates a directional dust removal trigger instruction; if the trigger conditions are not met, it generates a maintain operation instruction.

[0087] 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.

[0088] 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.

[0089] 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.

[0090] 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.

[0091] 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.

[0092] 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 1 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.

[0093] 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.

[0094] 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 1 The various processes of the method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0095] 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.

[0096] 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 triggering directional dust removal in a rotary air preheater based on xRFM-driven boundary segmentation, characterized in that, include: Acquire the pressure drop change records and dust removal time records of each target sector in the cold end and transition zone of the rotary air preheater during continuous operation, and obtain the sector pressure drop sequence based on the pressure drop change records; Based on the sector voltage drop sequence and dust removal time records, extract the cumulative voltage drop before each dust removal time and the voltage drop decrease after each dust removal time; Based on the cumulative pressure drop and the pressure drop decrease, calculate the recovery characterization quantity that characterizes the relationship between pressure drop evolution and post-cleaning recovery, and determine the current sector recovery status based on the recovery characterization quantity; Input the recovery characterization quantity into xRFM, and segment the recovery characterization quantity according to the recovery boundary corresponding to different ash blockage recovery hidden states to obtain the boundary segment label; Input the boundary segment label into the sector pressure drop recovery soot blowing criterion method, and match the current sector recovery status to the recovery boundary category corresponding to the boundary segment label to obtain the current segment recovery boundary; Under the current segmented recovery boundary constraints, the current sector recovery status is compared with the current segmented recovery boundary. If the triggering conditions are met, a directional dust removal trigger command is generated; otherwise, a maintenance operation command is generated.

2. The method for triggering directional cleaning of a rotary air preheater based on xRFM-driven boundary segmentation for recovery, as described in claim 1, is characterized in that... The process involves acquiring records of pressure drop changes and dust removal times for each target sector in the cold end and transition zone of the rotary air preheater during continuous operation, and obtaining the sector pressure drop sequence based on the pressure drop change records, including: The target sectors were determined according to the sector division of the cold end and transition zone of the rotary air preheater, and the pressure drop change records and dust removal time records of each target sector during continuous operation were collected. Using each dust removal moment in the dust removal moment record as a dividing point, the sector timing voltage drop data corresponding to each dust removal moment is obtained; The sector timing voltage drop data are arranged to obtain the sector voltage drop sequence of each target sector.

3. The method for triggering directional cleaning of a rotary air preheater based on xRFM-driven boundary segmentation for recovery, as described in claim 1, is characterized in that... The step of extracting the cumulative pressure drop before each cleaning moment and the pressure drop decrease after each cleaning moment based on the sector pressure drop sequence and cleaning time records includes: Based on the dust removal time records, the continuous operation analysis range corresponding to the dust removal event is selected from the sector voltage drop sequence of each target sector. The pressure drop accumulation interval is determined based on the continuous operation analysis range, and the pressure drop increase between each adjacent record value in the pressure drop accumulation interval is extracted to obtain the rising change sequence; The pressure drop accumulation corresponding to the current dust removal moment is obtained by accumulating the increasing change sequence in the recording order; The sector voltage drop sequence from the current dust removal time to the moment when the first value not lower than the previous record appears in the sector voltage drop sequence is defined as the voltage drop fall interval. Adjacent record values ​​in the voltage drop fall interval are compared in the recording order. Based on the comparison results, the voltage drop decrease between each adjacent record value is extracted to obtain the voltage drop fall amount corresponding to the current dust removal time.

4. The method for triggering directional cleaning of a rotary air preheater based on xRFM-driven boundary segmentation for recovery, as described in claim 1, is characterized in that... The process of calculating a recovery characteristic quantity, which characterizes the relationship between pressure drop evolution and post-cleaning recovery, based on the cumulative pressure drop and the pressure drop decrease, and determining the current sector recovery status based on the recovery characteristic quantity, includes: Based on the current cumulative pressure drop and the cumulative pressure drop corresponding to the previous cleaning time, calculate the change in the current cumulative pressure drop relative to the previous cleaning event. Based on the current pressure drop decrease and the pressure drop decrease corresponding to the previous cleaning time, calculate the change in the current pressure drop decrease relative to the previous cleaning event. Based on the current cumulative pressure drop and the current pressure drop decrease, calculate the difference between the current cumulative pressure drop and the current pressure drop decrease, and calculate the ratio between the current cumulative pressure drop and the current pressure drop decrease. The current cumulative pressure drop, the current pressure drop decrease, the change in the current cumulative pressure drop relative to the previous dust removal event, the change in the current pressure drop relative to the previous dust removal event, the difference between the current cumulative pressure drop and the current pressure drop decrease, and the ratio of the current cumulative pressure drop to the current pressure drop decrease are arranged in the input order of xRFM to form the recovery characterization quantity. Calculate the recovery boundary position corresponding to the current dust removal event based on the recovery characteristic quantity; Based on the location of the recovery boundary within the preset state segment, obtain the recovery state value and determine the recovery state value as the current sector recovery state.

5. The method for triggering directional cleaning of a rotary air preheater based on xRFM-driven boundary segmentation for recovery, as described in claim 1, is characterized in that... The process involves inputting the restored characterization quantity into xRFM, segmenting and identifying the restored characterization quantity according to the restoration boundaries corresponding to different ash-blocking restoration hidden states, and obtaining boundary segment labels, including: The recovered representation is input into the input layer of xRFM, and the recovery mapping layer, composed of fully connected neurons, maps the recovered representation to obtain the mapping result. The mapping results are aggregated into recovery mapping features, and the recovery mapping features are input into the ash-blocking recovery hidden state layer composed of fully connected neurons. The recovery mapping features are compressed and represented to obtain the ash-blocking recovery hidden state vector. Based on the hidden state vector of ash blockage recovery, the current sector recovery state is assigned to the hidden state of high recovery ash blockage recovery, the hidden state of transitional recovery ash blockage recovery, and the hidden state of low recovery ash blockage recovery, and the assignment result of the hidden state of ash blockage recovery is obtained. The results of restoring the hidden state of the ash blockage are input into the criterion embedding layer composed of fully connected neurons for segment recognition, and the boundary segment labels are obtained.

6. The method for triggering directional dust removal in a rotary air preheater based on xRFM-driven boundary segmentation according to claim 1, characterized in that, The step of inputting boundary segment labels into the sector pressure drop recovery soot blowing criterion method and matching the current sector recovery status with the recovery boundary category corresponding to the boundary segment label to obtain the current segment recovery boundary includes: Connect the sector voltage drop recovery soot blowing criterion method to the output of xRFM as the criterion header; The boundary segmentation label and the current sector recovery status are jointly input into the sector pressure drop recovery soot blowing criterion method; The recovery boundary category is determined based on the boundary segment label, and the boundary comparison interval corresponding to the recovery boundary category is called. In the sector voltage drop recovery soot blowing criterion method, a recovery boundary matching layer is set up to activate the matching neurons corresponding to the recovery boundary category and form the current boundary matching pathway. Input the current sector recovery status into the current boundary matching path, and obtain the boundary position matching result based on the relative position of the current sector recovery status with the start boundary, intermediate boundary and end boundary; determine the segment boundary position based on the boundary position matching result, and generate the current segment recovery boundary; The current segmented recovery boundary is passed to the triggering and determination layer of the sector pressure drop recovery soot blowing criterion method, which is used to determine the directional soot blowing trigger command and the maintenance operation command.

7. The method for triggering directional cleaning of a rotary air preheater based on xRFM-driven boundary segmentation for recovery, as described in claim 1, is characterized in that... Under the current segmented recovery boundary constraints, the current sector recovery state is compared with the current segmented recovery boundary. If the triggering condition is met, a directional dust removal trigger command is generated; if the triggering condition is not met, a maintenance operation command is generated. This includes: The current sector recovery status, the current segment recovery boundary, and the boundary segment label are input into the trigger determination layer of the sector voltage drop recovery soot blowing criterion method. Under the constraint of the current segment recovery boundary, the trigger interval corresponding to the current segment recovery boundary is determined according to the boundary segment label, and the trigger determination value is calculated according to the segment position and boundary direction of the current sector recovery status within the trigger interval. Determine the trigger value and the trigger range; When the trigger judgment value reaches the boundary condition corresponding to the trigger interval, the directional cleaning trigger judgment neuron is activated, and the judgment result of the directional cleaning trigger judgment neuron is determined as the directional cleaning trigger command. When the trigger judgment value does not meet the boundary conditions corresponding to the trigger interval, the maintain operation judgment neuron is activated, and the judgment result of the maintain operation judgment neuron is determined as the maintain operation instruction.

8. A rotary air preheater directional cleaning triggering system based on xRFM-driven recovery boundary segmentation, characterized in that, include: Sector pressure drop sequence module: used to acquire the pressure drop change records and dust removal time records of each target sector in the cold end and transition zone of the rotary air preheater during continuous operation, and to obtain the sector pressure drop sequence based on the pressure drop change records; Pressure drop recovery module: used to extract the cumulative pressure drop before each cleaning moment and the pressure drop recovery after each cleaning moment based on the sector pressure drop sequence and cleaning time record; Sector recovery status module: Used to calculate the recovery characterization quantity, which characterizes the relationship between pressure drop evolution and post-cleaning recovery, based on the cumulative pressure drop and the pressure drop reduction, and to determine the current sector recovery status based on the recovery characterization quantity; Boundary segmentation label module: It is used to input the restored characterization quantity into xRFM, and to segment and identify the restored characterization quantity according to the restoration boundary corresponding to different ash-blocking restoration hidden states to obtain the boundary segmentation label; Segmented recovery boundary module: Used to input the boundary segment label into the sector voltage drop recovery soot blowing criterion method, and match the current sector recovery status to the recovery boundary category corresponding to the boundary segment label to obtain the current segmented recovery boundary; Directional dust removal trigger instruction module: Under the current segment recovery boundary constraints, it determines the current sector recovery status and the current segment recovery boundary. If the trigger conditions are met, it generates a directional dust removal trigger instruction; if the trigger conditions are not met, it generates a maintain operation instruction.

9. An electronic device, characterized in that, The method 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 rotary air preheater directional cleaning triggering method based on xRFM driven recovery boundary segmentation as described in any one of claims 1-7.

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 rotary air preheater directional cleaning triggering method based on xRFM-driven recovery boundary segmentation as described in any one of claims 1-7.

Images