A method for predicting atmospheric methane concentration based on narrowband internet of things

Abstract

The application discloses a kind of based on narrowband internet of things atmospheric methane concentration prediction method, comprising the following steps: collecting methane and environmental data and regular formation multi-source monitoring sequence;Extract concentration change, disturbance coupling and sensing stability form local evolution microstate;Construct fact prediction branch and counterfactual missing report prediction branch to obtain counterfactual missing report influence;Determine prediction closure and closure inertia;Generate closure state and minimum recovery fragment candidate;When fracture triggers, upload closure fracture witness package and generate prediction result and closure update configuration;Write back closure update configuration and update each branch and closure state form prediction closed loop.The application uses the prediction closure method driven by counterfactual missing report influence, realizes atmospheric methane concentration low overhead prediction and recovery, with the advantages of low transmission burden, strong prediction continuity and high adaptability.

Description

Technical Field

[0001] This invention relates to the field of atmospheric environment monitoring technology, and in particular to a method for predicting atmospheric methane concentration based on narrowband Internet of Things. Background Technology

[0002] Current atmospheric methane concentration monitoring technologies mostly employ a fixed-period data acquisition, threshold-based alarm, and timed uploading via narrowband IoT. Methane sensors, in conjunction with temperature, humidity, air pressure, or wind speed and direction sensors, complete on-site monitoring, and then transmit the monitored values ​​to a cloud platform for storage, display, and basic predictive analysis. While some solutions incorporate historical data modeling, time-series forecasting, or edge-cloud collaborative processing, their technical focus remains primarily on monitoring device networking, sensor data uploading, and over-limit early warning, heavily relying on the completeness of the uploaded monitoring values ​​themselves.

[0003] However, the aforementioned existing technologies still have significant shortcomings in the low-bandwidth and low-power scenarios of narrowband IoT: On the one hand, existing solutions usually use whether the current monitoring value exceeds the threshold or is abnormal as the basis for uploading, lacking a judgment mechanism for "whether the current window's missing report will damage the subsequent prediction capability", making it difficult to balance the uploading burden and prediction continuity; on the other hand, existing solutions usually lack closed-loop modeling of the relationship between historical windows, current windows and future prediction windows, cannot generate minimum recovery fragment candidate results before the closure breaks, and cannot restore the cloud prediction state with minimum necessary evidence after the break is triggered, resulting in high communication overhead and insufficient prediction stability.

[0004] Therefore, how to provide a method for predicting atmospheric methane concentration based on narrowband Internet of Things is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an atmospheric methane concentration prediction method based on narrowband Internet of Things. This invention adopts a prediction closure method driven by the impact of counterfactual missing reports to achieve low-overhead prediction and recovery of atmospheric methane concentration, and has the advantages of low transmission burden, strong prediction continuity and high adaptability.

[0006] An atmospheric methane concentration prediction method based on narrowband Internet of Things according to an embodiment of the present invention includes the following steps: Collect methane concentration data and environmental auxiliary data, perform time alignment, quality verification and window normalization, and generate multi-source monitoring sequences; Multi-source monitoring sequences are input into the local evolution micro-state coding process to extract information on concentration changes, perturbation coupling, and sensor stability, thereby generating local evolution micro-state results; Based on multi-source monitoring sequences and local evolutionary micro-state results, a fact prediction branch and a counterfactual underreporting prediction branch are constructed to generate counterfactual underreporting impact results; Based on the counterfactual underreporting impact results and local evolution microstate results, the predicted closure results are determined and closure inertia results characterizing the closure maintenance capability are generated; Based on the predicted closure results and closure inertia results, closure state results are generated, and minimum recovery fragment candidate results are generated under closed tense state or inertial decay state. When the closure state result is in the rupture triggered state, a closure rupture witness packet is generated based on the minimum recovery fragment candidate result and uploaded via narrowband IoT. Combined with the historical cloud state, atmospheric methane concentration prediction result and closure update configuration result are generated. The process of writing the closure update configuration result back to the multi-source monitoring sequence updates the fact prediction branch, the counterfactual missing report prediction branch, and the closure status result, forming a prediction closed loop.

[0007] Optionally, the environmental auxiliary data specifically includes temperature data, relative humidity data, atmospheric pressure data, wind speed data, wind direction data, precipitation status data, and time-phase data.

[0008] Optionally, the generation of the local evolutionary microstate results specifically includes: Based on multi-source monitoring sequences, methane concentration data and environmental auxiliary data are extracted by window index to generate local evolution input fragments; Concentration change information is extracted from the methane concentration subsequence in the local evolution input fragment to generate concentration change results; Perturbation coupling information is extracted from environmental auxiliary data and concentration change results in the local evolution input segment, and perturbation coupling results are generated; Sensing stability information is extracted based on the quality check markers in the local evolution input segments and the sequence continuity within the window, and sensing stability results are generated. Local evolution microstate encoding is performed on the concentration change results, perturbation coupling results, and sensing stability results to generate local evolution microstate candidate results; Index binding and result encapsulation are performed on the candidate results of local evolution microstates to generate local evolution microstate results.

[0009] Optionally, the generation of the counterfactual underreporting impact result specifically includes: Extract the current window monitoring segment, reference window monitoring segment, and corresponding micro-state segment from multi-source monitoring sequences and local evolutionary micro-state results, and construct the input results of the fact prediction branch and the counterfactual missing report prediction branch; Perform fact extension calculation on the input results of the fact prediction branch to obtain the fact prediction branch results; Based on the current window missing report conditions, perform missing report extension calculation on the input results of the counterfactual missing report prediction branch to obtain the counterfactual missing report prediction branch results; The results of the fact prediction branch and the results of the counterfactual underreporting prediction branch are aligned and compared using the same prediction window index to form the branch difference results; Based on the branch difference results, the impact of missing reports is merged to form candidate results for counterfactual missing reports. Perform index binding and result writing on the candidate results of the counterfactual missing report impact, and output the counterfactual missing report impact results.

[0010] Optionally, the factual extension calculation reads the methane concentration value at the end of the reference window monitoring segment as the extension starting point, reads the window mean, adjacent difference, fluctuation intensity and directional persistence in the current window monitoring segment as the current window concentration evolution constraint, and reads the micro-state category in the local evolution micro-state result corresponding to the current window as the basis for extension mode selection.

[0011] Optionally, the generation of the predicted closure result and the closure inertia result specifically includes: Extract the current window closure determination elements and continuous window inertia determination elements from the counterfactual underreporting impact results and local evolution micro-state results to form the closure determination input results; Based on the closure determination input, perform closure boundary checks to obtain the closure boundary check results; The closure decision is executed based on the closure boundary check results and the local evolution microstate results corresponding to the current window to obtain the predicted closure result; Extract the direction continuity relationship, intensity change relationship, level change relationship, continuous window change relationship and micro-state continuity relationship from the continuous window inertia determination elements to form a closure inertia determination sequence; Perform an inertial decay check on the closure inertia determination sequence to obtain the inertial decay check result; The predicted closure result is correlated with the inertial decay check result and written to output the closure inertial result.

[0012] Optionally, the generation of the minimum recovery fragment candidate result specifically includes: State mapping is performed on the predicted closure result and the closure inertia result to form candidate closure state results; The candidate results of the closure state are checked in the order of execution according to the current window index and the indexes of consecutive adjacent windows to obtain the closure state result; When the closure state result is a closed tense state, extract the pre-recovery fragment to form the first candidate fragment result; When the closure state result is in an inertial decay state, an enhanced recovery fragment is constructed to form the second candidate fragment result; Minimum filtering is performed on the first and second candidate fragment results to obtain the minimum recovery fragment candidate results; The closure state result and the minimum recovery fragment candidate result are associated and written, and the fragment index result is output for the closure break witness package to call.

[0013] Optionally, the state mapping is determined by combining the predicted closure results (closure preservation, closure reduction, and closure breakage) with the predicted closure inertia results (high, medium, and low closure inertia). When the predicted closure result is a closure preservation result and the closure inertia result is a high closure inertia, the mapping is a closed steady state. When the predicted closure result is a closure preservation result and the closure inertia result is a medium closure inertia, the mapping is a closed tense state. When the predicted closure result is a closure reduction result and the closure inertia result is a medium or low closure inertia, the mapping is an inertia decay state. When the predicted closure result is a closure breakage result, or when the predicted closure result is a closure reduction result and the closure inertia result is a low closure inertia and the counterfactual missing report effect result corresponding to the current window is a severe effect, the mapping is a breakage triggered state.

[0014] Optionally, the generation of the atmospheric methane concentration prediction results and the closure update configuration results specifically includes: The closure state results are extracted as the current window corresponding fragment index result, counterfactual missing report impact result, and local evolution microstate result of the rupture trigger state, forming the rupture trigger input result; Perform witness encapsulation on the break trigger input result to obtain the closure break witness package; According to the narrowband IoT reporting requirements, the closure break witness packet is constructed and the link is written to form the narrowband IoT upload result; Read the closure break witness packet from the narrowband IoT upload results, and call the historical cloud state to perform cloud recovery alignment to form the recovered cloud state; Based on the restored cloud status, subsequent predictions are recursively performed, and the atmospheric methane concentration prediction results are output. The configuration writeback decision is executed based on the differences in state before and after recovery, resulting in a closure update configuration.

[0015] Optionally, the closure update configuration result associates and encapsulates the window regularization update item, the fact prediction branch update item, the counterfactual missing report prediction branch update item, the closure state update item, and the subsequent upload trigger update item.

[0016] The beneficial effects of this invention are: Compared to existing technologies, this invention does not merely perform conventional data collection, uploading, and threshold alarms for methane concentration. Instead, it constructs a continuous processing link around the question of whether a missing report in the current window will impair subsequent prediction capabilities. This link comprises a counterfactual missing report prediction branch, prediction closure results, closure inertia results, closure state results, and minimum recovery fragment candidate results. This transforms the uploading decision from simply relying on whether the current monitored value is abnormal to judging based on the actual impact of the missing report on subsequent prediction conclusions. This technical solution reduces redundant uploading during the closure maintenance phase, generates minimum recovery fragment candidate results in advance under closure tension or inertial decay states, and generates closure rupture witness packets for narrowband IoT uploading under the rupture trigger state. Thus, it maintains the continuity and recoverability of the cloud prediction link under limited bandwidth and low power consumption conditions.

[0017] Furthermore, this invention incorporates methane concentration data along with temperature, relative humidity, atmospheric pressure, wind speed, wind direction, precipitation status, and time-phase data into the local evolution micro-state coding process. This allows both the factual prediction branch and the counterfactual missing report prediction branch to be built on a more complete local state expression, enhancing the ability of atmospheric methane concentration prediction results to characterize local disturbances, fluctuation intensity, and evolution trends. Simultaneously, after closure breakdown, this invention does not simply retransmit the original data. Instead, it calls the closure breakdown witness packet through the minimum recovery fragment candidate result and combines it with historical cloud states to form the recovered cloud state and closure update configuration results. These are then written back to the multi-source monitoring sequence processing, thereby updating the factual prediction branch, the counterfactual missing report prediction branch, and the closure state results, forming a prediction closed loop. Therefore, this invention can improve the stability, continuity, and adaptive recovery capability for missing report scenarios in atmospheric methane concentration prediction while reducing the upload burden of narrowband IoT. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an atmospheric methane concentration prediction method based on narrowband Internet of Things proposed in this invention; Figure 2 This is a flowchart illustrating the generation of counterfactual underreporting impact results for an atmospheric methane concentration prediction method based on narrowband Internet of Things proposed in this invention. Figure 3 This is a flowchart illustrating the closure breaking witness upload and configuration write-back process for an atmospheric methane concentration prediction method based on narrowband Internet of Things proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-3 A method for predicting atmospheric methane concentration based on narrowband Internet of Things includes the following steps: Collect methane concentration data and environmental auxiliary data, perform time alignment, quality verification and window normalization, and generate multi-source monitoring sequences; Multi-source monitoring sequences are input into the local evolution micro-state coding process to extract information on concentration changes, perturbation coupling, and sensor stability, thereby generating local evolution micro-state results; Based on multi-source monitoring sequences and local evolutionary micro-state results, a fact prediction branch and a counterfactual underreporting prediction branch are constructed to generate counterfactual underreporting impact results; Based on the counterfactual underreporting impact results and local evolution microstate results, the predicted closure results are determined and closure inertia results characterizing the closure maintenance capability are generated; Based on the predicted closure results and closure inertia results, closure state results are generated, and minimum recovery fragment candidate results are generated under closed tense state or inertial decay state. When the closure state result is in the rupture triggered state, a closure rupture witness packet is generated based on the minimum recovery fragment candidate result and uploaded via narrowband IoT. Combined with the historical cloud state, atmospheric methane concentration prediction result and closure update configuration result are generated. The process of writing the closure update configuration result back to the multi-source monitoring sequence updates the fact prediction branch, the counterfactual missing report prediction branch, and the closure status result, forming a prediction closed loop.

[0021] In this embodiment, the environmental auxiliary data specifically includes temperature data, relative humidity data, atmospheric pressure data, wind speed data, wind direction data, precipitation status data, and time-phase data.

[0022] In this embodiment, the generation of local evolutionary microstate results specifically includes: Based on multi-source monitoring sequences, methane concentration data and environmental auxiliary data are extracted by window index to generate local evolution input fragments; The local evolution input segments include methane concentration subsequence, temperature subsequence, relative humidity subsequence, atmospheric pressure subsequence, wind speed subsequence, wind direction subsequence, precipitation state subsequence, and time phase subsequence aligned with the same window index, and retain the corresponding quality check mark and window start and end index; Concentration change information is extracted from the methane concentration subsequence in the local evolution input fragment to generate concentration change results; The concentration change information includes window mean, adjacent difference, fluctuation intensity and directional persistence. The adjacent difference is used to characterize the methane concentration change at adjacent sampling times, the fluctuation intensity is used to characterize the amplitude of methane concentration change within the window, and the directional persistence is used to characterize the degree of maintenance of continuous rise or continuous fall. The fluctuation intensity is the average of the absolute values ​​of the methane concentration difference at each adjacent sampling time within the window. Perturbation coupling information is extracted from environmental auxiliary data and concentration change results in the local evolution input segment, and perturbation coupling results are generated; The disturbance coupling information is obtained by calculating the change amplitude of environmental auxiliary data within the current window, matching it with the concentration change results in the same direction and coupling it with the amplitude, and obtaining the temperature, humidity and pressure coupling amount, wind field coupling amount, precipitation disturbance amount and time-period modulation amount. The disturbance coupling result is obtained by weighting and summing the preset non-negative coupling weights. Sensing stability information is extracted based on the quality check markers in the local evolution input segments and the sequence continuity within the window, and sensing stability results are generated. The sensing stability information includes the effective sampling ratio, the missing completion ratio, the anomaly removal ratio, and the continuity consistency. The sensing stability result is obtained by weighting the effective sampling ratio and the continuity consistency according to their respective stability weights, and then subtracting the result of weighting the missing completion ratio and the anomaly removal ratio according to their respective stability weights. Local evolution microstate encoding is performed on the concentration change results, perturbation coupling results, and sensing stability results to generate local evolution microstate candidate results; The local evolution microstate coding sequentially splices the concentration change results, perturbation coupling results, and sensing stability results according to a unified window index to form a local evolution feature group. Based on the preset microstate division rules, the local evolution feature group is classified and its intensity is graded to generate local evolution microstate candidate results. The pre-defined microstate classification rules are as follows: A step-by-step judgment is performed in the order of sensing stability results, concentration change results, and perturbation coupling results. First, the effective sampling ratio, missing data completion ratio, anomaly rejection ratio, and continuity consistency in the sensing stability results are read. When the effective sampling ratio is less than the sum of the missing data completion ratio and the anomaly rejection ratio, or when the continuity consistency is less than the missing data completion ratio, it is directly judged as a low-confidence microstate. In other cases, the adjacent differences and directional persistence in the concentration change results are read. When all adjacent differences are non-negative and at least one has a positive change, it is judged as a continuously rising state. When all adjacent differences are non-positive and at least one has a negative change, it is judged as a continuously falling state. When both positive and negative changes exist in adjacent differences and they alternate at least twice in time sequence... When the state is determined to be cyclical fluctuation, the rest are determined to be stable maintenance state. After the change state determination is completed, the perturbation coupling result is compared with the fluctuation intensity in the concentration change result. When the perturbation coupling result is greater than the fluctuation intensity, it is determined to be a perturbation-dominated state. When the perturbation coupling result is not greater than the fluctuation intensity and its direction of action is consistent with the direction of persistence, it is determined to be a cooperative maintenance state. When the perturbation coupling result is not greater than the fluctuation intensity and its direction of action is opposite to the direction of persistence, it is determined to be a cancellation and suppression state. Low confidence microstate, stable maintenance cooperative microstate, continuously rising cooperative microstate, continuously falling cooperative microstate, cyclical fluctuation cooperative microstate, continuously rising perturbation-dominated microstate, continuously falling perturbation-dominated microstate, cyclical fluctuation perturbation-dominated microstate, and stable maintenance cancellation microstate are regarded as closed microstate categories of candidate results for local evolution microstates. Index binding and result encapsulation are performed on the candidate results of local evolution microstates to generate local evolution microstate results.

[0023] In this embodiment, the generation of the counterfactual underreporting impact result specifically includes: Extract the current window monitoring segment, reference window monitoring segment, and corresponding micro-state segment from multi-source monitoring sequences and local evolutionary micro-state results, and construct the input results of the fact prediction branch and the counterfactual missing report prediction branch; The current window monitoring segment consists of methane concentration data and environmental auxiliary data corresponding to the current window index. The reference window monitoring segment consists of methane concentration data and environmental auxiliary data in the window that precedes and is adjacent to the current window. The input results of the fact prediction branch are formed by splicing the reference window monitoring segment, the current window monitoring segment, and the local evolution micro-state results corresponding to the current window in chronological order. The input results of the counterfactual missing report prediction branch are formed by combining the reference window monitoring segment, the current window index, and the local evolution micro-state results corresponding to the nearest window before the current window. Perform fact extension calculation on the input results of the fact prediction branch to obtain the fact prediction branch results; Based on the current window missing report conditions, perform missing report extension calculation on the input results of the counterfactual missing report prediction branch to obtain the counterfactual missing report prediction branch results; The missing report extension calculation keeps the current window index unchanged, reads the methane concentration value at the end of the monitoring segment of the reference window as the extension starting point, masks the methane concentration data and environmental auxiliary data in the monitoring segment of the current window, reads the micro-state category, window mean, adjacent difference, fluctuation intensity and directional persistence in the local evolution micro-state results corresponding to the nearest window before the current window as the basis for alternative concentration evolution constraints and extension mode selection, and outputs the counterfactual trend results, counterfactual peak results, counterfactual risk level results and counterfactual threshold crossing results in sequence along the same continuous prediction window as the factual extension calculation, and encapsulates them into counterfactual missing report prediction branch results; The results of the fact prediction branch and the results of the counterfactual underreporting prediction branch are aligned and compared using the same prediction window index to form the branch difference results; The branch difference results include trend offset results, peak offset results, risk level offset results, and threshold crossing offset results; Based on the branch difference results, the impact of missing reports is merged to form candidate results for counterfactual missing reports. The impact of missed reports is aggregated by sequentially scanning the trend offset, peak offset, risk level offset, and threshold crossing offset results using the same prediction window index. The prediction window where the first offset occurs is determined as the starting window of the missed report impact, and the prediction window where the last consecutive offset occurs is determined as the ending window of the missed report impact. The duration window of the missed report impact is determined by the number of consecutively covered prediction windows. The direction of the missed report impact is determined by the offset direction of the counterfactual missed report prediction branch result relative to the factual prediction branch result. When the counterfactual trend result is weaker than the factual trend result, the counterfactual peak result is lower than or later than the factual peak result, or the counterfactual risk level result is lower... When the result of factual risk level or counterfactual threshold crossing is later than the result of factual threshold crossing or changes from crossing to not crossing, it is determined to be an underestimation direction. When the result of counterfactual trend is stronger than the result of factual trend, the result of counterfactual peak is higher than or earlier than the result of factual peak, the result of counterfactual risk level is higher than the result of factual risk level, or the result of counterfactual threshold crossing is earlier than the result of factual threshold crossing or changes from not crossing to crossing, it is determined to be an overestimation direction. When the directions of the various offset results are inconsistent, the direction of the impact of the missing report is determined in the following order: threshold crossing offset result takes priority, risk level offset result takes second priority, peak offset result takes second priority, and trend offset result takes last priority. After the direction determination is completed, when the threshold crossing status changes or the risk level offset result crosses two or more risk levels, the impact intensity of the missing report is determined to be high intensity. When the threshold crossing status does not change but the threshold crossing time shifts, or the risk level offset result crosses one risk level, or the peak offset result includes both peak size offset and peak arrival window offset, the impact intensity of the missing report is determined to be medium intensity. The rest are determined to be low intensity. The impact level of a missed report is determined by combining the intensity of the missed report impact and the duration window of the missed report impact. When the intensity of the missed report impact is high and the duration window of the missed report impact covers two or more prediction windows, it is determined to be a severe impact. When the intensity of the missed report impact is medium, or when the intensity of the missed report impact is high but the duration window of the missed report impact only covers one prediction window, it is determined to be a moderate impact. The rest are determined to be mild impacts. The direction of the missed report impact, the intensity of the missed report impact, the level of the missed report impact, and the duration window of the missed report impact are encapsulated into a counterfactual missed report impact candidate result. Perform index binding and result writing on the candidate results of the counterfactual missing report impact, and output the counterfactual missing report impact results; The index binding and result writing process associates and encapsulates the candidate results of counterfactual missing reports with the current window index, the results of fact prediction branches, the results of counterfactual missing reports prediction branches, and the corresponding local evolution microstate results, to obtain the counterfactual missing report results that can be directly called for the determination of the prediction closure results.

[0024] In this embodiment, the factual extension calculation reads the methane concentration value at the end of the reference window monitoring segment as the extension starting point, reads the window mean, adjacent difference, fluctuation intensity and directional persistence in the current window monitoring segment as the current window concentration evolution constraint, and reads the micro-state category in the local evolution micro-state result corresponding to the current window as the basis for extension mode selection. When the microstate category is a continuously rising coordinated microstate or a continuously rising perturbation-dominated microstate, the future concentration sequence is formed by continuously recursively extrapolating in subsequent prediction windows according to the direction of the adjacent difference at the end of the current window, and superimposing the same-direction correction amount corresponding to the perturbation coupling result. When the microstate category is a continuously falling coordinated microstate or a continuously falling perturbation-dominated microstate, the future concentration sequence is formed by recursively extrapolating in the reverse direction of the adjacent difference at the end of the current window, and superimposing the reverse correction amount corresponding to the perturbation coupling result. When the microstate category is a reciprocating fluctuating coordinated microstate or a reciprocating fluctuating perturbation-dominated microstate, the future concentration sequence is formed by alternating extrapolation according to the most recent alternation order and fluctuation intensity of the adjacent difference within the current window. When the microstate category is a stable maintained coordinated microstate or a stable maintained offset microstate, the future concentration sequence is formed by slowly extrapolating with the mean of the current window as the center and according to the correction direction corresponding to the perturbation coupling result. When the microstate category is a low-confidence microstate, the future concentration sequence is formed by smoothly continuing the end concentration state of the monitoring segment of the reference window and the local evolution microstate result corresponding to the nearest window before the current window. After obtaining the future concentration sequence, continuously rising, continuously falling, stable, or fluctuating markers are extracted from the future concentration sequence as factual trend results. Extreme values ​​and their corresponding prediction windows in the future concentration sequence are extracted as factual peak results. Factual risk level results are generated according to the correspondence between the future concentration sequence and the methane concentration grading interval. Factual threshold crossing results are generated according to the comparison between the future concentration sequence and the methane concentration warning threshold. The factual trend results, factual peak results, factual risk level results, and factual threshold crossing results are encapsulated into factual prediction branch results.

[0025] In this embodiment, the generation of the predicted closure result and the closure inertia result specifically includes: Extract the current window closure determination elements and continuous window inertia determination elements from the counterfactual underreporting impact results and local evolution micro-state results to form the closure determination input results; The current window closure determination elements include the direction of the missing report impact, the intensity of the missing report impact, the level of the missing report impact, and the duration window of the missing report impact corresponding to the current window. The continuous window inertia determination elements include the direction of the missing report impact, the intensity of the missing report impact, the level of the missing report impact, the duration window of the missing report impact, and the micro-state category in the local evolution micro-state results corresponding to the consecutive adjacent windows before the current window. Based on the closure determination input, perform closure boundary checks to obtain the closure boundary check results; The closure boundary check first reads the impact level and duration window of the missing report in the current window. When the impact level is mild and the duration window of the missing report only covers one prediction window, it is determined that the current window is still inside the closure. When the impact level is mild and the duration window of the missing report covers two or more prediction windows, or when the impact level is moderate and the duration window of the missing report only covers one prediction window, it is determined that the current window is close to the closure boundary. When the impact level is moderate and the duration window of the missing report covers two or more prediction windows, or when the impact level is severe, it is determined that the current window exceeds the closure boundary. The closure decision is executed based on the closure boundary check results and the local evolution microstate results corresponding to the current window to obtain the predicted closure result; The closure decision outputs the closure preservation result when the current window is inside the closure and the local evolution microstate result corresponding to the current window is not a low-confidence microstate; the closure reduction result when the current window is close to the closure boundary, or when the current window is inside the closure but the local evolution microstate result corresponding to the current window is a low-confidence microstate; and the closure breaking result when the current window exceeds the closure boundary. Extract the direction continuity relationship, intensity change relationship, level change relationship, continuous window change relationship and micro-state continuity relationship from the continuous window inertia determination elements to form a closure inertia determination sequence; The directional continuity relationship indicates whether the direction of the missing report impact in the current window is consistent with that in the previous consecutive adjacent windows. The intensity change relationship indicates whether the intensity of the missing report impact in the current window increases, remains the same, or decreases. The level change relationship indicates whether the level of the missing report impact in the current window increases, remains the same, or decreases. The persistence window change relationship indicates whether the persistence window of the missing report impact in the current window expands, remains the same, or shortens. The micro-state continuity relationship indicates whether the micro-state category in the local evolution micro-state results continues to maintain the perturbation-dominated micro-state, cooperative micro-state, or low-confidence micro-state. Perform an inertial decay check on the closure inertia determination sequence to obtain the inertial decay check result; The inertial decay check is determined as follows: when the direction of the missing report impact is continuous and consistent, the intensity of the missing report impact does not decrease, the level of the missing report impact does not fall back, the duration window of the missing report impact does not shorten, and the micro-state continuity relationship continues to maintain a continuously rising disturbance-dominated micro-state, a continuously falling disturbance-dominated micro-state, a reciprocating fluctuation disturbance-dominated micro-state, or a low-confidence micro-state; when the direction of the missing report impact is continuous and consistent, and at least one of the three factors of the missing report impact intensity, missing report impact level, and missing report impact duration window does not continue to rise, or the micro-state continuity relationship changes from a disturbance-dominated micro-state to a cooperative micro-state; when the direction of the missing report impact is not continuous and consistent, or the intensity of the missing report impact decreases, or the level of the missing report impact falls back, or the duration window of the missing report impact shortens, it is determined as weak decay. The predicted closure result is correlated with the inertial decay check result and written to output the closure inertial result; The closure inertia result is denoted as high closure inertia when the predicted closure result is a closure-preserving result and the inertia decay check result is weak decay; medium closure inertia when the predicted closure result is a closure-preserving result and the inertia decay check result is medium decay, or when the predicted closure result is a closure-shrinking result and the inertia decay check result is weak decay; and low closure inertia when the predicted closure result is a closure-shrinking result and the inertia decay check result is medium or strong decay, or when the predicted closure result is a closure-breaking result.

[0026] In this embodiment, the generation of the minimum recovery fragment candidate result specifically includes: State mapping is performed on the predicted closure result and the closure inertia result to form candidate closure state results; The candidate results of the closure state are checked in the order of execution according to the current window index and the indexes of consecutive adjacent windows to obtain the closure state result; The sequential verification compares the candidate closure state of the current window with the candidate closure state of the consecutive adjacent windows before the current window, window by window. When the candidate closure state of the current window is higher than the candidate closure state of the consecutive adjacent windows before the current window, the candidate closure state of the current window remains unchanged. When the candidate closure state of the current window is lower than the candidate closure state of the consecutive adjacent windows before the current window and the impact level of the counterfactual misreporting of the current window has not decreased, the candidate closure state of the current window is upgraded to be consistent with the candidate closure state of the consecutive adjacent windows before the current window. When the candidate closure state of the current window is lower than the candidate closure state of the consecutive adjacent windows before the current window and the impact level of the counterfactual misreporting of the current window has decreased, the candidate closure state of the current window remains unchanged. When the closure state result is a closed tense state, extract the pre-recovery fragment to form the first candidate fragment result; The preparatory recovery segment takes the current window monitoring segment as the center, calls the most recent reference window monitoring segment forward, and calls the fact prediction branch result and counterfactual missing report prediction branch result corresponding to the prediction window that first appears offset in the prediction window covered by the window of continuous missing report impact. It extracts the window segments that are directly related to the trend offset result, peak offset result, risk level offset result and threshold crossing offset result, and retains the current window index, segment start index, segment end index, offset type and offset direction to form the first candidate segment result; When the closure state result is in an inertial decay state, an enhanced recovery fragment is constructed to form the second candidate fragment result; The enhanced recovery segment takes the current window monitoring segment as the center, calls the two most recent reference window monitoring segments forward, and calls the fact prediction branch results and counterfactual missing report prediction branch results corresponding to all prediction windows covered by the window of persistent missing report impact backward. It merges the window segments that are simultaneously associated with the peak offset result, risk level offset result and threshold crossing offset result, and supplements the local evolution microstate result, prediction closure result and closure inertia result corresponding to the current window to form the second candidate segment result. Minimum filtering is performed on the first and second candidate fragment results to obtain the minimum recovery fragment candidate results; The minimization screening is performed in the following order: first retaining threshold crossing offset results, then retaining risk level offset results, then retaining peak offset results, and finally retaining trend offset results. First, redundant window segments that do not change the direction of the impact of missing reports are deleted, then redundant window segments that do not change the intensity of the impact of missing reports are deleted, then redundant window segments that do not change the level of the impact of missing reports are deleted, and the shortest continuous window segment that can maintain the counterfactual missing report impact result of the current window unchanged after deletion is retained. The closure state result and the minimum recovery fragment candidate result are associated and written, and the fragment index result is output for the closure break witness package to call; The associated write operation only writes the closure state result and not the minimum recovery fragment candidate result when the closure state result is in a closed steady state. When the closure state result is in a closed tense state or an inertial decay state, the minimum recovery fragment candidate result is associated and encapsulated with the current window index, fragment start index, fragment end index, offset type, offset direction, predicted closure result, and closure inertial result.

[0027] In this embodiment, the state mapping is determined by combining the predicted closure results (closure preservation, closure reduction, and closure breakage) with the predicted closure inertia results (high, medium, and low closure inertia). When the predicted closure result is a closure preservation result and the closure inertia result is a high closure inertia, the mapping is a closed steady state. When the predicted closure result is a closure preservation result and the closure inertia result is a medium closure inertia, the mapping is a closed tense state. When the predicted closure result is a closure reduction result and the closure inertia result is a medium or low closure inertia, the mapping is an inertia decay state. When the predicted closure result is a closure breakage result, or when the predicted closure result is a closure reduction result and the closure inertia result is a low closure inertia and the counterfactual missing report effect result corresponding to the current window is a severe effect, the mapping is a breakage triggered state.

[0028] In this embodiment, the generation of atmospheric methane concentration prediction results and closure update configuration results specifically includes: The closure state results are extracted as the current window corresponding fragment index result, counterfactual missing report impact result, and local evolution microstate result of the rupture trigger state, forming the rupture trigger input result; The rupture trigger input results include the current window index, fragment start index, fragment end index, offset type, offset direction, predicted closure result, closure inertia result, missing report impact direction, missing report impact intensity, missing report impact level, missing report impact duration window, and micro-state category in the local evolution micro-state results. Only the window fragment corresponding to the smallest recovery fragment candidate result pointed to by the fragment index result is called. Perform witness encapsulation on the break trigger input result to obtain the closure break witness package; The witness encapsulation extracts the window segments corresponding to the smallest recovery fragment candidate results according to the fragment start index to the fragment end index. It retains the methane concentration data, environmental auxiliary data, local evolution microstate results, fact prediction branch results and counterfactual missing report prediction branch results corresponding to each window segment. It also associates and encapsulates the current window index, offset type, offset direction, prediction closure result, closure inertia result, missing report impact direction, missing report impact intensity, missing report impact level and missing report impact duration window with the extracted window segments. According to the narrowband IoT reporting requirements, the closure break witness packet is constructed and the link is written to form the narrowband IoT upload result; The message construction splits the closure break witness packet into an index message and a witness content message. The index message includes the current window index, fragment start index, fragment end index, offset type, offset direction, and missing report impact level. The witness content message includes methane concentration data, environmental auxiliary data, local evolution microstate results, fact prediction branch results, and counterfactual missing report prediction branch results in the window fragment corresponding to the minimum recovery fragment candidate results. The messages are written to the cloud receiving queue via narrowband IoT in the order of sending the index message first and the witness content message last. Read the closure break witness packet from the narrowband IoT upload results, and call the historical cloud state to perform cloud recovery alignment to form the recovered cloud state; The cloud recovery alignment writes the closure break witness package to the corresponding position in the historical cloud state according to the current window index and the fragment start index to the fragment end index. The minimum recovery fragment candidate result in the closure break witness package replaces the missing report continuation content in the same window range in the historical cloud state. The corresponding fact prediction branch result, counterfactual missing report prediction branch result, and counterfactual missing report impact result are updated synchronously within the window range. Based on the restored cloud status, subsequent predictions are recursively performed, and the atmospheric methane concentration prediction results are output. The subsequent prediction recursion uses the latest methane concentration data, environmental auxiliary data, and local evolution microstate results after the current window in the restored cloud state as the starting input for prediction. It recalculates the factual trend results, factual peak results, factual risk level results, and factual threshold crossing results along the subsequent continuous prediction windows. It associates and encapsulates the methane concentration prediction value, peak arrival window, risk level change, and threshold crossing status of the corresponding window into the atmospheric methane concentration prediction result. The configuration writeback decision is executed based on the differences in state before and after recovery, resulting in a closure update configuration.

[0029] In this embodiment, the closure update configuration result associates and encapsulates the window regularization update item, the fact prediction branch update item, the counterfactual missing report prediction branch update item, the closure status update item, and the subsequent upload trigger update item. The write-back decision first compares the difference between the restored cloud state and the historical cloud state in the coverage range of the minimum restoration fragment candidate result. When the number of windows covered by the restored fragment is greater than the number of windows covered by the original window, the window regularization update item is determined to be the new continuous regularization window range from the fragment start index to the fragment end index. When the number of windows covered by the restored fragment is equal to the number of windows covered by the original window, the window regularization update item is determined to maintain the original window coverage range. When the number of windows covered by the restored fragment is less than the number of windows covered by the original window and the restored fragment only covers the current window and its previous or next window, the window regularization update item is determined to be the new dual-window regularization range of the current window and its corresponding adjacent window. Then, read the methane concentration data, environmental auxiliary data, and local evolution microstate results in the window corresponding to the termination index of the recovery fragment. Determine the methane concentration value at the end of the window as the new fact extension starting point, and determine the window mean, adjacent difference, fluctuation intensity, and directional persistence of the window as the new current window concentration evolution constraints. Determine the microstate category in the local evolution microstate results corresponding to the window as the new extension mode selection basis, and use it as the fact prediction branch update item. Then, the window corresponding to the termination index of the recovered fragment is determined as the new most recently known window. The window mean, adjacent difference, fluctuation intensity, direction persistence and local evolution micro-state category of the window are read. The alternative concentration evolution constraint and the basis for selection of extension mode used for missing report extension calculation in the historical cloud status are replaced. The new most recently known window index, alternative concentration evolution constraint and the basis for selection of extension mode are written together as the counterfactual missing report prediction branch update item. Subsequently, the predicted closure result and closure inertia result recalculated from the cloud state after recovery are read. When the predicted closure result is a closure-maintained result and the closure inertia result is a high closure inertia, the closure state update term is determined as a closed steady state. When the predicted closure result is a closure-maintained result and the closure inertia result is a medium closure inertia, the closure state update term is determined as a closed tense state. When the predicted closure result is a closure-reduced result and the closure inertia result is a medium or low closure inertia, the closure state update term is determined as an inertia decay state. When the predicted closure result is a closure-breaking result, or when the predicted closure result is a closure-reduced result and the closure inertia result is a low closure inertia and the counterfactual missing report effect result after recovery is a severe effect, the closure state update term is determined as a breaking triggered state. Finally, the restored closure state update item and the restored counterfactual missing report impact level are combined for adjudication. When the closure state update item is in a broken state, the subsequent upload trigger update item is determined to be a direct upload. Once the subsequent window generates a minimum recovery fragment candidate result again, the generation of the closure broken witness packet is directly called. When the closure state update item is in an inertial decay state and the restored counterfactual missing report impact level is moderate or severe, the subsequent upload trigger update item is determined to be a tightened upload. When the subsequent window shows the same offset direction twice in a row, the minimum recovery fragment candidate result is directly called. When the closure state update item is in a closed tense state and the restored counterfactual missing report impact level is mild, the subsequent upload trigger update item is determined to be a preparatory upload. The subsequent window only generates a minimum recovery fragment candidate result and does not upload it immediately. When the closure state update item is in a closed steady state and the restored counterfactual missing report impact level is mild and the missing report impact duration window is shortened, the subsequent upload trigger update item is determined to be a regular upload, that is, it is restored to the upload method that is normally judged according to the closure state result.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a continuous atmospheric methane monitoring scenario on the periphery of a riverside chemical industrial park. Monitoring points were deployed at the downwind boundary of the park, around the wastewater regulation area, and at the intersection of roads outside the plant. Simultaneously, methane concentration data, temperature data, relative humidity data, atmospheric pressure data, wind speed data, wind direction data, precipitation status data, and time-phase data were collected on-site and transmitted back to the cloud via narrowband IoT. This scenario has long suffered from problems such as frequent wind direction changes, rapid changes in nighttime diffusion conditions, short-term missing reports from local sensors, and limited communication. Using a fixed-period upload method easily leads to numerous invalid uploads, abrupt changes in cloud trend judgments after local missing reports, and poor continuity of early warnings. This invention first forms a multi-source monitoring sequence at the monitoring end, then extracts information on concentration changes, disturbance coupling, and sensor stability to generate local evolution micro-state results. Subsequently, it constructs a fact prediction branch and a counterfactual missing report prediction branch to determine whether not reporting in the current window will affect subsequent predictions. Finally, based on the prediction closure result, closure inertia result, and closure state result, it determines whether to generate a minimum recovery fragment candidate result.

[0031] During continuous operation, on-site maintenance records, narrowband IoT upload logs, cloud trend trajectories, early warning trigger records, and handling receipts collectively demonstrate that this invention can maintain low-frequency uploads during stable periods, pre-reserve key segments under closed tension and inertial decay states, and upload only the closure rupture witness packet and complete cloud recovery alignment under rupture trigger states. Compared to the original fixed-period upload method, monitoring logs show a significant reduction in redundant backhauls, more timely remediation of missing reports during communication congestion periods, and better continuity of cloud prediction curves during weather changes and wind direction shifts. Handling logs show improved consistency between early warning information and on-site inspection conclusions, and a significant reduction in duplicate verifications and invalid alarms caused by short-term missing reports. Therefore, this invention can maintain the continuity and recoverability of atmospheric methane concentration prediction under controlled communication load conditions, making it suitable for atmospheric methane monitoring scenarios with strong disturbances, frequent missing reports, and limited reporting resources.

[0032] Table 1. Comparison of the overall performance of methane monitoring and forecasting based on the impact of counterfactual underreporting and the prediction closure mechanism.

[0033] From the perspective of communication burden and terminal overhead, the method of this invention shows significant advantages. The table shows that the average daily number of reported messages decreased from 1462 to 923, a reduction of 36.87%; the average data transmission volume per event decreased from 11.8KB to 4.9KB, a reduction of 58.47%; and the average daily power consumption on the communication side decreased from 63.4mAh to 41.7mAh, a reduction of 34.23%. This indicates that the method does not simply reduce sampling, but rather first forms local evolutionary micro-state results based on multi-source monitoring sequences, and then determines whether the current window truly affects subsequent predictions through factual prediction branches and counterfactual missing report prediction branches. This concentrates the uploading action on key segments corresponding to closed tense states, inertial decay states, and rupture trigger states, avoiding a large number of repeated backhauls during stable periods. According to the data in the table, all three communication and energy consumption indicators have significantly decreased.

[0034] In terms of prediction continuity and recovery capability, the method of this invention is more adaptable to scenarios with missing reports. The cloud recovery time after continuous missing reports was shortened from 48 minutes to 19 minutes, a reduction of 60.42%, and the number of continuous cloud prediction interruptions decreased from 21 times within 30 days to 8 times, a reduction of 61.90%. Simultaneously, the average absolute error of the two-hour rolling prediction decreased from 0.128 ppm to 0.084 ppm, a reduction of 34.38%, and the average error at peak arrival time decreased from 27 minutes to 15 minutes, a reduction of 44.44%. This result is due to the fact that this method does not wait for a break before remediation under closed tense or inertial decay states, but instead generates minimum recovery fragment candidate results in advance. When the closure state result transitions to a break-triggered state, a closure break witness packet is formed based on the minimum recovery fragment candidate results. After being transmitted back via narrowband IoT, cloud recovery alignment is performed, allowing the cloud to recover subsequent prediction links without relying on recalculating the entire historical data. According to the data in the table, the core indicators for both recovery and prediction improved simultaneously.

[0035] From the perspective of the consistency between early warning quality and on-site handling, the method of this invention can more accurately identify risk windows that truly have handling value. The accuracy rate of threshold crossing judgment improved from 86.9% to 94.1%, an increase of 7.2 percentage points; the proportion of invalid alarms decreased from 19.6% to 10.8%, a decrease of 8.8 percentage points; and the consistency rate of on-site verification improved from 83.7% to 92.4%, an increase of 8.7 percentage points. The reason for this result is that the impact of counterfactual underreporting is no longer limited to a single concentration threshold comparison, but rather a comprehensive analysis of trend offset, peak offset, risk level offset, and threshold crossing offset. Furthermore, the method uses prediction closure results, closure inertia results, and closure state results to perform layered adjudication of upload and recovery actions. Therefore, it can distinguish between short-term disturbances, continuous offsets, and true ruptures, making the uploaded content more focused on the minimum evidence necessary for recovery cloud prediction. According to the data in the table, the relevant indicators for early warning judgment and on-site verification have all shown synchronous improvement.

[0036] As can be seen from Table 1, this invention does not simply rely on compressing the upload frequency to reduce energy consumption. Instead, through the synergistic effects of local evolution micro-encoding, counterfactual missing report impact merging, prediction closure and closure inertia determination, minimum recovery fragment candidate result screening, closure break witness upload, and closure update configuration write-back, it simultaneously achieves reduced communication burden, faster missing report recovery, reduced prediction error, enhanced peak value prediction capability, reduced invalid alarms, and improved on-site verification consistency rate. This demonstrates that the method can maintain the continuity, stability, and adaptive recovery capability of atmospheric methane concentration prediction under narrowband IoT limited link conditions, and has high engineering application value.

[0037] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting atmospheric methane concentration based on narrowband Internet of Things, characterized in that, Includes the following steps: Collect methane concentration data and environmental auxiliary data, perform time alignment, quality verification and window normalization, and generate multi-source monitoring sequences; Multi-source monitoring sequences are input into the local evolution micro-state coding process to extract information on concentration changes, perturbation coupling, and sensor stability, thereby generating local evolution micro-state results; Based on multi-source monitoring sequences and local evolutionary micro-state results, a fact prediction branch and a counterfactual underreporting prediction branch are constructed to generate counterfactual underreporting impact results; Based on the counterfactual underreporting impact results and local evolution microstate results, the predicted closure results are determined and closure inertia results characterizing the closure maintenance capability are generated; Based on the predicted closure results and closure inertia results, closure state results are generated, and minimum recovery fragment candidate results are generated under closed tense state or inertial decay state. When the closure state result is in the rupture triggered state, a closure rupture witness packet is generated based on the minimum recovery fragment candidate result and uploaded via narrowband IoT. Combined with the historical cloud state, atmospheric methane concentration prediction result and closure update configuration result are generated. The process of writing the closure update configuration result back to the multi-source monitoring sequence updates the fact prediction branch, the counterfactual missing report prediction branch, and the closure status result, forming a prediction closed loop.

2. The method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 1, characterized in that, The environmental auxiliary data specifically includes temperature data, relative humidity data, atmospheric pressure data, wind speed data, wind direction data, precipitation status data, and time-phase data.

3. The method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 1, characterized in that, The generation of the local evolutionary microstate results specifically includes: Based on multi-source monitoring sequences, methane concentration data and environmental auxiliary data are extracted by window index to generate local evolution input fragments; Concentration change information is extracted from the methane concentration subsequence in the local evolution input fragment to generate concentration change results; Perturbation coupling information is extracted from environmental auxiliary data and concentration change results in the local evolution input segment, and perturbation coupling results are generated; Sensing stability information is extracted based on the quality check markers in the local evolution input segments and the sequence continuity within the window, and sensing stability results are generated. Local evolution microstate encoding is performed on the concentration change results, perturbation coupling results, and sensing stability results to generate local evolution microstate candidate results; Index binding and result encapsulation are performed on the candidate results of local evolution microstates to generate local evolution microstate results.

4. The method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 1, characterized in that, The generation of the counterfactual underreporting impact results specifically includes: Extract the current window monitoring segment, reference window monitoring segment, and corresponding micro-state segment from multi-source monitoring sequences and local evolutionary micro-state results, and construct the input results of the fact prediction branch and the counterfactual missing report prediction branch; Perform fact extension calculation on the input results of the fact prediction branch to obtain the fact prediction branch results; Based on the current window missing report conditions, perform missing report extension calculation on the input results of the counterfactual missing report prediction branch to obtain the counterfactual missing report prediction branch results; The results of the fact prediction branch and the results of the counterfactual underreporting prediction branch are aligned and compared using the same prediction window index to form the branch difference results; Based on the branch difference results, the impact of missing reports is merged to form candidate results for counterfactual missing reports. Perform index binding and result writing on the candidate results of the counterfactual missing report impact, and output the counterfactual missing report impact results.

5. The method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 4, characterized in that, The factual extension calculation reads the methane concentration value at the end of the reference window monitoring segment as the extension starting point, reads the window mean, adjacent difference, fluctuation intensity and directional persistence in the current window monitoring segment as the current window concentration evolution constraint, and reads the micro-state category in the local evolution micro-state result corresponding to the current window as the basis for extension mode selection.

6. The method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 1, characterized in that, The generation of the predicted closure result and the closure inertia result specifically includes: Extract the current window closure determination elements and continuous window inertia determination elements from the counterfactual underreporting impact results and local evolution micro-state results to form the closure determination input results; Based on the closure determination input, perform closure boundary checks to obtain the closure boundary check results; The closure decision is executed based on the closure boundary check results and the local evolution microstate results corresponding to the current window to obtain the predicted closure result; Extract the direction continuity relationship, intensity change relationship, level change relationship, continuous window change relationship and micro-state continuity relationship from the continuous window inertia determination elements to form a closure inertia determination sequence; Perform an inertial decay check on the closure inertia determination sequence to obtain the inertial decay check result; The predicted closure result is correlated with the inertial decay check result and written to output the closure inertial result.

7. The method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 1, characterized in that, The generation of the minimum recovery fragment candidate results specifically includes: State mapping is performed on the predicted closure result and the closure inertia result to form candidate closure state results; The candidate results of the closure state are checked in the order of execution according to the current window index and the indexes of consecutive adjacent windows to obtain the closure state result; When the closure state result is a closed tense state, extract the pre-recovery fragment to form the first candidate fragment result; When the closure state result is in an inertial decay state, an enhanced recovery fragment is constructed to form the second candidate fragment result; Minimum filtering is performed on the first and second candidate fragment results to obtain the minimum recovery fragment candidate results; The closure state result and the minimum recovery fragment candidate result are associated and written, and the fragment index result is output for the closure break witness package to call.

8. The method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 7, characterized in that, The state mapping is determined by combining the predicted closure results (closure preservation, closure reduction, and closure breakage) with the predicted closure inertia results (high, medium, and low closure inertia). When the predicted closure result is a closure preservation result and the closure inertia result is a high closure inertia, the mapping is a closed steady state. When the predicted closure result is a closure preservation result and the closure inertia result is a medium closure inertia, the mapping is a closed tense state. When the predicted closure result is a closure reduction result and the closure inertia result is a medium or low closure inertia, the mapping is an inertia decay state. When the predicted closure result is a closure breakage result, or when the predicted closure result is a closure reduction result and the closure inertia result is a low closure inertia and the counterfactual missing report effect result corresponding to the current window is a severe effect, the mapping is a breakage triggered state.

9. The method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 1, characterized in that, The generation of the atmospheric methane concentration prediction results and the closure update configuration results specifically includes: The closure state results are extracted as the current window corresponding fragment index result, counterfactual missing report impact result, and local evolution microstate result of the rupture trigger state, forming the rupture trigger input result; Perform witness encapsulation on the break trigger input result to obtain the closure break witness package; According to the narrowband IoT reporting requirements, the closure break witness packet is constructed and the link is written to form the narrowband IoT upload result; Read the closure break witness packet from the narrowband IoT upload results, and call the historical cloud state to perform cloud recovery alignment to form the recovered cloud state; Based on the restored cloud status, subsequent predictions are recursively performed, and the atmospheric methane concentration prediction results are output. The configuration writeback decision is executed based on the differences in state before and after recovery, resulting in a closure update configuration.

10. A method for predicting atmospheric methane concentration based on narrowband Internet of Things according to claim 9, characterized in that, The closure update configuration result encapsulates the window regularization update item, the fact prediction branch update item, the counterfactual missing report prediction branch update item, the closure state update item, and the subsequent upload triggered update item.

Images