An atmospheric pollutant automatic sampling and analyzing method and device

Abstract

The application discloses an atmospheric pollutant automatic sampling and analyzing method and device, extracts periodic opening rules of an emission source from a sampling flow time sequence zero fluctuation deviation period, implements concentrated sample enrichment with a differential pressure abnormal jump as a trigger condition in a discharge node time window; the enriched sample is subjected to multi-wavelength differential optical detection to generate an original absorption intensity, a pure reference absorption value is extracted by using a single-component dominant signal in a multi-component covariance degree sudden drop period, an adaptive updated cross-sensitivity correction matrix is established to realize accurate quantification; the quantitative result is compared with a meteorological diffusion prediction value in each period, near-end emission contribution is identified according to the consistency of the downwind matching strength, an emission rule index is generated from the periodic rhythm of the concentration residual after correction; the instrument baseline drift is diagnosed in the index low period, reliable data sets are screened, a traceability list is constructed, a detection and analysis report is generated, and the emission peak sample capture rate, multi-component quantitative accuracy and long-term credibility of the traceability conclusion are improved.

Description

Technical Field

[0001] This invention relates to the field of atmospheric environmental monitoring technology, and in particular to an automatic sampling and analysis method and device for atmospheric pollutants. Background Technology

[0002] Under long-term unattended operation, continuous atmospheric pollutant monitoring stations experience intermittent interruptions in gas flow due to external interference or equipment fluctuations. Existing data processing workflows lack an automatic mechanism to identify the validity of these interruption periods, leading to the systematic interference of abnormal data with subsequent analysis results. Industrial emissions typically follow fixed temporal patterns driven by production shifts or process nodes. Existing sampling schemes lack the ability to proactively identify these emission temporal patterns, resulting in a uniform distribution of sample collection times rather than concentration during critical emission periods, and a low capture rate for initial high-concentration peak samples.

[0003] In multi-component optical detection, target components interfere with each other in their characteristic wavelength bands. Existing methods rely on fixed parameters set at the factory for component separation, which cannot adaptively adjust to seasonal changes in the actual coexisting atmospheric components. As the operating cycle lengthens, quantitative accuracy continues to decline. The measured concentration includes contributions from direct emissions near the monitoring site as well as regional background emissions from long-distance transport, making it difficult to quantitatively separate the two. After long-term operation, the slow decay of optical path transmission efficiency causes a trend of baseline drift in the measurement. Existing diagnostic procedures rely on periodic manual correction, and the drift deviation between calibration cycles continues to accumulate, affecting the long-term reliability of monitoring data. Summary of the Invention

[0004] This invention discloses an automatic sampling and analysis method and device for air pollutants. It aims to identify the periodic time sequence of emission sources by the abnormal deviation pattern of flow and drive differential pressure enrichment to concentrate in key periods. It extracts adaptive cross-interference correction parameters based on multi-component covariant fracture window to achieve accurate quantification. It decouples the near-end emission contribution by combining meteorological diffusion model and automatically locates the baseline drift diagnosis window of the instrument by relying on emission pattern index. It screens reliable datasets to construct a source tracing list and outputs detection and analysis reports.

[0005] The first aspect of this invention provides an automatic sampling and analysis method for air pollutants, comprising the following steps: Ambient air samples and meteorological monitoring data are collected. Based on the time periods of abnormal sampling flow deviation in the ambient air samples, flow deviation records are obtained. Based on the meteorological monitoring data, the atmospheric diffusion level is determined by identifying the characteristics of calm wind residence periods. Based on the periodic pattern of sampling interruption times identified in the flow deviation records, key sampling batches are formed. These key sampling batches are then used to trigger concentrated enrichment based on abnormal differential pressure spikes to obtain concentrated contaminated samples. The concentrated pollutant sample is subjected to multi-wavelength differential optical detection to generate the original absorption intensity. The multi-component covariance break time period is identified based on the original absorption intensity to determine the interference correction amount. The equivalent concentration deviation of each pollutant component is calculated based on the interference correction amount to establish a quantitative detection result. Based on the quantitative detection results and the atmospheric diffusion level, a meteorological decoupling record is established to analyze the meteorological mismatch characteristics of pollutant concentration. The meteorological decoupling record is used to quantify the local contribution ratio of near-end emission intensity to obtain a corrected concentration value. The meteorological correction residual time-series periodic law is analyzed from the corrected concentration value to output the emission law index. Based on the emission pattern index, baseline drift characteristics during low emission periods are identified, and an instrument deviation index is output. Periodic offset exceeding the threshold is screened for the instrument deviation index to obtain a reliable dataset. The reliable dataset is used to construct a multi-pollutant emission source tracing list and generate a detection and analysis report.

[0006] A second aspect of the present invention provides an automatic sampling and analysis device for air pollutants, comprising: The data acquisition module is used to collect ambient air samples and meteorological monitoring data, collect flow deviation records based on the abnormal deviation period of the sampling flow from the ambient air samples, and determine the atmospheric diffusion level based on the characteristics of the calm wind residence period identified by the meteorological monitoring data. The batch screening module is used to identify the periodic pattern of sampling interruption times based on the flow deviation record to form key sampling batches, and to use the key sampling batches to trigger concentrated enrichment based on differential pressure abnormal jumps to obtain concentrated contaminated samples; The spectral detection module is used to perform multi-wavelength differential optical detection on the concentrated pollutant sample to generate the original absorption intensity, identify the multi-component covariance break time period based on the original absorption intensity to determine the interference correction amount, and calculate the equivalent concentration deviation of each pollutant component based on the interference correction amount to establish a quantitative detection result. The meteorological correction module is used to analyze the meteorological mismatch characteristics of pollutant concentration based on the quantitative detection results and the atmospheric diffusion level, establish a meteorological decoupling record, use the meteorological decoupling record to quantify the local contribution ratio of near-end emission intensity to obtain the corrected concentration value, and analyze the time series periodic law of meteorological correction residual from the corrected concentration value to output the emission law index. The report generation module is used to identify baseline drift characteristics during low-emission periods based on the emission pattern index, output an instrument deviation index, screen for periodic offset exceeding the threshold period based on the instrument deviation index to obtain a reliable dataset, and construct a multi-pollutant emission source tracing list from the reliable dataset to generate a detection and analysis report.

[0007] The beneficial effects of this invention are reflected in the following points: 1. The zero-fluctuation interval in the sampling pump flow time series carries the time series information of interruption events. The fixed start-stop behavior of the emission source leaves a periodic imprint in the flow time series through pipeline pressure disturbance. The frequency peaks of the interval distribution of adjacent interruption times are transformed into emission rhythm consistency through inverse misalignment correlation analysis. The time coordinates of each emission node automatically emerge from the flow anomaly record. The enrichment enhancement mode triggered by the differential pressure anomaly jump concentrates the sampling resources to the initial high concentration stage of emission. The capture rate of emission peak samples continues to increase with the accumulation of emission rhythm confidence. 2. The covariance of the multi-component absorbance time series drops sharply at the moment of emission source switching. The covariance break window exposes the pure reference signal dominated by single-component absorption. The cross-sensitivity correction coefficient is continuously extracted from the measured signal to replace the fixed factory calibration parameters. The correction matrix is ​​layered according to atmospheric diffusion level and fused with new batch data by time-reverse attenuation weighting. The cross-interference correction accuracy does not degrade with the extension of operating time. The impact of seasonal atmospheric composition changes on quantitative accuracy is actively compensated. 3. The downwind consistency between the measured concentration and the meteorological diffusion prediction concentration quantitatively separates the local near-end emission contribution from the regional transport background. The periodic rhythm of the corrected concentration residual is quantified by the emission regularity index to weigh the emission intensity of each time period. The emission contribution in the low phase interval of the emission regularity index approaches zero. Any non-zero systematic reading comes from the instrument itself. This interval is used in reverse as an interference-free diagnostic window for baseline drift. The trend drift rate and fixed bias component are accurately analyzed from historical batch accumulation. After the correction amount is deducted, the confidence of the reliable dataset and the traceability list automatically increases with the operation cycle. Attached Figure Description

[0008] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0009] Figure 1 This is a flowchart illustrating an automatic sampling and analysis method for air pollutants according to the present invention.

[0010] Figure 2 This is a schematic diagram of the multi-component covariant fracture window identification of the present invention.

[0011] Figure 3 This is a structural block diagram of an automatic sampling and analysis device for air pollutants according to the present invention.

[0012] Wherein: 1-Component covariance coefficient AB sequence; 2-Component covariance coefficient AC sequence; 3-Covariance fracture window; 4- Fracture window period; 5-Single component dominant period; 6-Pure reference absorbance extraction point; 7-False fracture removal section; 8-Covariance coefficient fracture threshold line. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0015] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0016] The technical solutions of the embodiments of this application will be described below.

[0017] like Figure 1 As shown, this embodiment of the invention provides an automatic sampling and analysis method for air pollutants, including the following steps S101-S105: Step S101: Collect ambient air samples and meteorological monitoring data. Based on the abnormal deviation period of the ambient air sample, obtain the flow deviation record. Based on the meteorological monitoring data, identify the characteristics of the calm wind residence period and determine the atmospheric diffusion level.

[0018] Specifically, ambient air samples and meteorological monitoring data are collected. Ambient air samples are continuously drawn by an automatic sampling pump installed at the monitoring point at a set flow rate. After particulate matter is removed by a filter membrane, the gas sample enters a multi-stage absorption tube. Each absorption tube sequentially enriches gaseous pollutant components of different polarities. The sampling pump records the air flow rate in real time and writes it into the sampling log at a fixed frequency. The validity of the ambient air sample is determined by the continuity and range of flow rate readings in the sampling log. When the monitoring point is deployed downwind of the industrial park's emission outlet, the intermittent operation of the emission source will leave a temporal trace in the enrichment amount of the ambient air sample. Sample batches with consistently high enrichment amounts at the same time period for several consecutive days mostly correspond to concentrated emissions from the emission source at fixed times. The batch time records of ambient air samples thus become the temporal basis for subsequently identifying periodic emission patterns. Meteorological monitoring data is collected by a combination of ultrasonic anemometers, barometric pressure and humidity sensors, rain gauges, and total radiation meters. Wind direction, wind speed, air temperature, relative humidity, atmospheric pressure, and solar radiation intensity are written into the meteorological log at a time resolution synchronized with the sampling frequency of ambient air samples. Continuous periods in the meteorological log where wind speed is consistently below the calm wind threshold (e.g., 0.5 m / s) directly reflect the degree to which the near-surface pollutant diffusion and dilution capacity is limited, serving as the core temporal input for determining the atmospheric diffusion level. Ambient air samples and meteorological monitoring data are bidirectionally linked by the sampling time, ensuring that each batch of ambient air samples precisely corresponds to the meteorological monitoring data at that time, providing a time-aligned data foundation for subsequent batch-by-batch quantitative analysis of meteorological diffusion contributions.

[0019] In some embodiments, obtaining flow deviation records based on the abnormal deviation periods of the ambient air sample collection includes: generating a sampling flow sequence using the instantaneous flow change nodes of the sampling pump operating conditions detected by the ambient air sample; calculating the coupling slope of air path resistance and flow attenuation on the sampling flow sequence to obtain a flow deviation distribution; identifying anomalous ultra-stable zero-fluctuation intervals based on the flow deviation distribution to form a set of deviation periods; and outputting a flow deviation record by aggregating the maximum deviation duration within the set of deviation periods.

[0020] A sampling flow sequence is generated by detecting sudden changes in the instantaneous flow rate of the sampling pump under ambient air sample conditions. The sampling log of the ambient air sample records instantaneous flow rate readings for each sampling period. After being arranged in chronological order, the difference between flow rate readings of adjacent periods constitutes a beat-by-beat difference component. The moment when the absolute value of the difference component exceeds the sudden change threshold (e.g., 0.2 liters per minute) is marked as a sudden change node. All sudden change nodes and their corresponding flow rate readings together constitute the basic data layer of the sampling flow sequence. Under normal operating conditions, the sampling pump flow rate decreases slowly and monotonically due to the increased filter membrane load. The absolute value of the difference component in the sampling flow sequence is generally small and there are no obvious sudden changes during the normal operating period. When the sampling pipeline is briefly blocked by an external object and then clears itself, the sampling flow sequence shows a sharp peak shape after a sudden drop in flow rate at the time of blockage, which differs in timing from the pipeline pressure fluctuations caused by intermittent emissions from the emission source. The former drops sharply and then quickly recovers to the pre-blockage level, while the latter maintains the flow rate at a new equilibrium point for a period of time after the sudden change before slowly changing. Two types of sudden changes are distinguished based on their recovery rate after the sudden change: sudden change nodes with a recovery time shorter than several sampling periods are recorded as instantaneous blockage nodes. Instantaneous blockage nodes are not included in the valid candidates for subsequent flow interruption pattern recognition to prevent pipeline interference events from being mistakenly identified as emission behavior signals. During ambient air sample collection, short-term fluctuations in grid voltage will cause changes in the sampling pump motor speed. The corresponding sampling flow sequence will show a brief disturbance with the same frequency as the voltage fluctuation. The voltage monitoring module records a voltage anomaly flag in the ambient air sample collection log. Sudden change nodes in the sampling flow sequence within the validity period of the flag are marked with a power disturbance label. These nodes do not participate in the main sequence fitting of the gas path resistance coupling slope during the flow deviation distribution calculation stage.

[0021] The flow deviation distribution is obtained by calculating the coupling slope of gas path resistance and flow rate decay for the sampled flow rate sequence. The stable segments between abrupt change nodes in the sampled flow rate sequence correspond to the continuous operating range of the sampling pump under stable load. The rate of flow rate decay with increasing filter membrane load within each stable segment is quantified by the linear regression slope. A steeper slope indicates a larger increase in gas path resistance, which is mostly caused by the rapid accumulation of high-concentration particulate matter on the filter membrane surface. After determining the decay slope of each stable segment, the theoretical flow rate at subsequent times of that segment is extrapolated using the slope. The difference between the measured flow rate and the theoretical extrapolated value constitutes the flow deviation at that time. The flow deviations throughout the entire process are arranged by time to form the flow deviation distribution. During periods of heavy pollution, PM2.5 concentrations surge, causing the attenuation slope of the stable segment in the sampling flow sequence to increase several times more sharply than usual. Flow deviations are densely distributed with negative deviations during these periods, and measured flow rates remain consistently below the normal attenuation baseline. Sampling efficiency is significantly reduced due to rapid membrane blockage. Conversely, short-term spikes in gas pressure caused by intermittent emissions from emission sources exhibit alternating positive and negative pulse patterns in the flow deviation distribution. These patterns differ significantly from the continuous negative deviations caused by membrane blockage, making them easily distinguishable. Differentiating these two types of characteristics in the flow deviation distribution pattern is a crucial prerequisite for accurate identification of deviation time periods. When calculating the slope of the stable segment near the power disturbance marker node in the sampling flow sequence, the slopes of the disturbance-free segments are replaced by excluding the four sampling cycles before and after the marker node. This ensures that brief flow fluctuations caused by voltage volatility do not affect the accuracy of estimating the gas path resistance coupling slope.

[0022] Based on the flow deviation distribution, abnormally stable zero-fluctuation intervals are identified to form a set of deviation time periods. Under normal sampling conditions, the flow deviation distribution oscillates randomly around zero at each moment, and the oscillation amplitude reflects the normal random fluctuation of the gas path resistance within the sampling period. When the sampling pump stops due to power failure or abnormal shutdown of the control module, the pump body stops, the gas path flow is completely interrupted, and the flow meter reading locks near a certain abnormal value and no longer changes. The flow deviation distribution exhibits an extremely stable form with extremely low oscillation amplitude in the corresponding time period, which is completely different from the random oscillation under normal operating conditions. In the flow deviation distribution, the interval where the variance of the deviation sequence frame by frame is lower than the normal oscillation amplitude benchmark for 15 consecutive sampling periods is identified as the zero-fluctuation interval. The zero-fluctuation interval must be separated from the simple low-load stable segment. The low-load stable segment has a smaller oscillation amplitude but still has discernible random fluctuations. The deviation value of the true zero-fluctuation interval is almost locked at a fixed level and has no random component. The distinction between the two is based on the threshold that the variance of the deviation sequence frame by frame is lower than the lower limit of the sensor quantization noise. Continuous segments below this threshold enter the candidate set of deviation time periods. The set of deviation time periods only includes such completely locked intervals. During the midday low-load period, the sampling pump operates steadily at a low flow rate, with the deviation frame variance being approximately two to three times the noise lower limit, still exhibiting discernible random fluctuations. However, after the sampling pump completely shuts down due to a power outage, the deviation frame variance drops to less than one-tenth of the noise lower limit, with the variance magnitudes of the two scenarios differing by more than an order of magnitude. When a sudden drop in temperature causes condensation and water accumulation in the sampling pipeline, the gas path resistance increases abruptly at the water accumulation location, resulting in a sustained and significant negative deviation in the flow deviation distribution, although the oscillation amplitude does not approach zero. This does not meet the zero-fluctuation criteria and is therefore excluded from the deviation period set. Thus, the two types of anomalies—water blockage and sampling interruption—can be effectively distinguished based on their morphological characteristics. Each candidate interval in the deviation period set, along with the original deviation sequence, is retained for verification of interval boundaries during the aggregation phase.

[0023] The traffic deviation record is output by summing the duration of the maximum deviation within the deviation time period set. The start and end times of each candidate zero-fluctuation interval in the deviation time period set are accurate to the sampling period granularity. The maximum number of consecutive frames in which the deviation value deviates from the normal oscillation baseline within the interval is multiplied by the sampling period to convert it into the maximum deviation duration. The start and end times of each candidate interval and the maximum deviation duration together constitute the basic entries of the traffic deviation record. When the time interval between two adjacent candidate intervals in the deviation time period set is less than 2 minutes, they are merged into a single traffic deviation record entry. The duration after merging is calculated as the sum of the two segments. If the same outage is split into multiple records due to a brief recovery in the middle, the duration of a single outage will be underestimated. Entries in the traffic deviation record with a duration of several minutes to tens of minutes mostly correspond to short outages caused by human intervention; entries with a duration spanning several hours mostly correspond to equipment failures or power outages. The traffic deviation record distinguishes between the two types of cases using a duration classification field. The classification boundary is taken from the typical duration value of various outage events in historical maintenance records, usually with 30 minutes as the boundary. Short-term interruption entries that repeatedly occur within the same day and whose start times are concentrated in similar time periods in the deviation time set mostly point to the periodic impact of the emission source on the pressure of the sampling pipeline during fixed time periods rather than equipment failure. The flow deviation record adds time clustering labels to multiple short-term interruption entries with a standard deviation of less than 20 minutes for the start time within a single day. This label triggers priority weighting processing during the interruption interval distribution construction stage, and the identification efficiency of periodic emission nodes is significantly improved.

[0024] Atmospheric diffusion levels are determined by identifying calm wind residence periods based on meteorological monitoring data. Wind speed time series data from meteorological monitoring data are statistically analyzed using sliding windows (e.g., 10 minutes) to determine the mean and standard deviation of wind speed within each window. Periods where the mean wind speed is consistently below the calm wind threshold of 0.5 m / s and the standard deviation is simultaneously small are identified as calm wind residence periods. During calm wind residence periods, the atmospheric mixing layer height is further compressed due to radiative cooling, limiting vertical dispersion of pollutants, and the near-surface concentration accumulation rate is significantly higher than during periods of normal wind speed. Atmospheric diffusion levels are determined based on the Pasqual stability classification, combined with solar radiation intensity and surface wind speed from meteorological monitoring data. Low wind speeds during sunny days correspond to levels A to B, while low wind speeds during cloudy days or nighttime correspond to levels E to F. The atmospheric diffusion levels for each time period are coded and entered into a time series table aligned with the meteorological monitoring data for batch retrieval during the meteorological concentration prediction calculation phase. Low wind speeds combined with weak solar radiation (at night or on cloudy days) are prone to inversions, which trigger the lowest diffusion level in the atmospheric diffusion classification, resulting in the most severe pollutant retention near the ground. Monitoring points around chemical industrial parks often experience radiative cooling after nightfall, and frequently experience F-level stable weather around dawn. Meteorological monitoring data for this period show a long plateau pattern with near-zero wind speeds, and the pollutant enrichment in ambient air samples during this period is significantly higher. The lowest atmospheric diffusion level period highly overlaps temporally with the key sampling batches identified based on flow deviation records. This correspondence is important evidence for identifying near-end source areas. Low wind speed values ​​within the validity period of rain gauge precipitation markers are considered precipitation interference rather than true calm winds, and the corresponding periods are not included in the main atmospheric diffusion classification sequence.

[0025] Step S102: Based on the flow deviation record, identify the periodic pattern of sampling interruption time to form key sampling batches, and use the key sampling batches to trigger concentrated enrichment based on differential pressure abnormal jump to obtain concentrated contamination samples.

[0026] In some embodiments, the step of identifying the periodic pattern of sampling interruption times based on the flow deviation record to form key sampling batches includes: extracting a set of complete interruption event occurrence times from the flow deviation record to generate an interruption time sequence; statistically analyzing the periodic distribution of adjacent interruption time intervals in the interruption time sequence to obtain interruption period characteristics; identifying fixed-time repetitive interruption cluster nodes based on the interruption period characteristics to establish emission node markers; and grouping the periodic interruption corresponding time period batches according to the emission node markers to determine key sampling batches.

[0027] The set of complete interruption event occurrence times is extracted from the flow deviation records to generate an interruption time sequence. Entries of the complete interruption type are filtered out using the duration classification field of each entry in the flow deviation records. A complete interruption is determined by the maximum deviation duration exceeding the lower limit of a complete interruption. The lower limit is based on the sampling pump start-stop response time, typically not less than 1 minute. Deviations shorter than the response time are mostly instantaneous pipeline disturbances rather than actual shutdowns. The start times of each complete interruption entry in the flow deviation records are arranged chronologically to form the interruption time sequence. For flow deviation record entries that overlap with the time period of a site maintenance work order, a maintenance interval label is added to the interruption time sequence. Maintenance shutdowns are not considered emission activities, and the interval values ​​and hit rate statistics of the maintenance interval label entries are handled separately in subsequent stages. After establishing the interruption time series, calendar normalization is performed on each time point to convert absolute times into relative times within a day to support cross-day comparisons. Normalization must distinguish between weekdays and non-weekdays. Some factories stop production on holidays, resulting in sparse entries for non-weekdays. Mixing normalization of the two types of calendar days would dilute the clustering of low-density entries on weekdays. The interruption time series is distinguished between the two types of calendars by work calendar labels. Independent intraday relative time distributions are established for weekday and non-weekday entries. Subsequent interval statistics and hit rate verification are performed separately according to these two sets of distributions. If the interval between multiple interruption times within a single day matches the factory shift rotation cycle, a shift matching label is added to the corresponding entry. The weight of the shift matching label entry is increased in the confidence weighting stage of the interruption cycle feature. Consecutive entries with an interval of less than 5 minutes between adjacent interruption times mostly correspond to multiple pressure disturbances caused by the same emission event and are merged into a single interruption record to avoid the interruption frequency being artificially inflated due to duplicate counting.

[0028] For example, obtaining interruption cycle characteristics by statistically analyzing the periodic distribution of adjacent interruption time intervals in the interruption time sequence includes: forming an interruption interval distribution by statistically analyzing the duration distribution of adjacent interruption event intervals in the interruption time sequence; identifying high-frequency clustered fixed interval peaks in the interruption interval distribution to determine the dominant cycle density distribution; calculating the inverse misalignment correlation coefficient of the interruption density of each cycle based on the dominant cycle density distribution to obtain the emission rhythm consistency; and outputting the interruption cycle characteristics by aggregating the confidence-weighted calculated values ​​of each cycle based on the emission rhythm consistency.

[0029] Interruption interval distribution is formed by statistically analyzing the duration of intervals between adjacent interruption events in the interruption time sequence. The time differences between adjacent entries in the interruption time sequence are arranged into an interval duration sequence. The frequency histogram after grouping with a fixed class interval is the interruption interval distribution. The class interval is a compromise between the sampling time resolution and the shortest expected emission cycle, commonly half an hour. If the class interval is too wide, similar interval values ​​are grouped together, and adjacent periodic peaks overlap; if it is too narrow, the frequency of a single group is too low, and the true peak value is buried in noise. The interval values ​​of maintenance interval annotation entries are excluded when constructing the interruption interval distribution, so that extremely long intervals will not form isolated low-density boxes at the end of the histogram, interfering with the identification of the dominant peak. The interval values ​​corresponding to entries carrying time cluster annotations are weighted and included in the interruption interval distribution, so that the periodic patterns indicated by daily clustering behavior are given priority. The interruption interval distribution is expressed as normalized frequency rather than raw count, thus eliminating the influence of differences in the number of historical batches. The interruption interval distribution of short-term monitoring batches with fewer interruption time sequence entries and long-term batches can be directly compared to determine the position of the dominant peak. Continuous interruptions triggered by the same emission event in the interruption time sequence have been merged during the construction phase, so the interruption interval distribution does not show high-frequency spurious peaks with extremely short intervals. For emission sources with diurnal drift in emission time, their basic periodic peaks show slight broadening rather than sharp single peaks in the interruption interval distribution. The degree of broadening reflects the stability of the emission time. The broadening value is temporarily recorded in the candidate peak label for weighted reference in the peak width field of the dominant period density distribution. Candidate peaks with broadening exceeding 3 class intervals are marked with time instability in the interruption interval distribution. Such peaks are not directly excluded. During the dominant period identification phase, they are reviewed using a relaxed peak threshold, thus preserving the weakly regular true emission cycle.

[0030] The dominant period density distribution is determined by identifying high-frequency clustered peak values ​​in the interruption interval distribution. Segments in the interruption interval distribution with frequencies exceeding the peak value threshold are identified as candidate dominant period peaks. The peak width is capped at two class intervals. Candidate peaks with widths exceeding the limit but carrying unstable time markers are retained after review with a relaxed peak value threshold. Unmarked, excessively wide segments are not considered candidates. The peak frequency must also exceed four times the median frequency of the entire interruption interval distribution to be considered a valid peak. The multiple threshold is calibrated based on the highest frequency of pseudo-periodic peaks in the archive of historical irregular shutdown batches (a random interruption sample library accumulated by the site). All retained dominant period density distributions are significant period peaks supported by actual emission behavior. The symmetry of the peak is quantified by the skewness of the frequency distribution on both sides. Candidate peaks with large skewness often correspond to a systematic shift in emission time in a certain direction. Gradual increase in production load and continuous earlier start-up time are typical examples. The dominant cycle density distribution adds a time drift label to such candidate peaks. The time drift label triggers dynamic phase correction when calculating the emission rhythm fit. The suppression effect of time drift on the correlation coefficient is eliminated by the correction. The emission rhythm fit of wide peak candidate cycles is generally low. Wide peaks are still written into the dominant cycle density distribution as candidates after being labeled with low confidence. The low confidence is the frequency confidence reduction value of the peak. It automatically reduces the ranking weight in the weighted product of the interrupted cycle characteristics instead of directly eliminating it. The weak regularity of the real emission cycle will not be over-screened. The integer multiple harmonic relationships between multiple candidate peaks are approximately determined by the greatest common divisor of the center interval values ​​of each peak. Harmonic peaks are labeled with harmonics in the dominant period density distribution and bound to the basic period peak. The higher the consistency between the frequency of the harmonic peak and the basic period peak, the stronger the regularity of the emission behavior. Harmonic identification effectively prevents the dominant period density distribution from misjudging the harmonic period of the same emission source as an independent emission rhythm, and subsequent emission node marking will not introduce redundant false nodes.

[0031] The emission rhythm fit is obtained by calculating the anti-phase misalignment correlation coefficient of the interruption density of each cycle based on the dominant cycle density distribution. The anti-phase misalignment test uses each candidate dominant cycle of the dominant cycle density distribution as a step size τ, and calculates the normalized correlation coefficient after misaligning the interruption density sequence of the interruption time sequence by the step size τ. The emission rhythm fit is quantified by the following formula: In the formula, t is the sampling batch time number, R(τ) is the emission rhythm fit degree corresponding to the step size τ, n(t) is the interruption density sequence after removing the mean, which is obtained by subtracting the mean from the interruption start count in a single sampling period at each time, and τ is the misalignment step size, which is taken from the value of each candidate period. When the step size is exactly equal to the actual emission period, n(t) and n(t-τ) are highly aligned, and R(τ) is close to 1; when the step size is a half-integer multiple of the actual emission period, the dense periods of the two sequences are completely misaligned, and R(τ) is close to zero or even negative; the curve of R(τ) changing with τ shows a positive maximum at integer multiples of the actual emission period and a minimum at half-integer multiples. The alternating positive and negative shape is the anti-phase misalignment feature. The positive maximum value of R(τ) of each candidate period is written as the emission rhythm fit degree into the attribute field of the candidate period corresponding to the dominant period density distribution. In the dominant periodic density distribution, before calculating R(τ), the n(t) sequence undergoes dynamic phase correction to eliminate the drift trend before participating in the calculation. The corrected emission rhythm fit reflects the intrinsic fit of the emission rhythm and is no longer suppressed by time drift. For factories that emit at fixed times every day, their R(τ) is close to 1 at the 24-hour step and close to zero at the 12-hour step, showing a clear phase misalignment characteristic, and the daily periodic emission pattern is verified. If R(τ) is close to 1 at both the 24-hour and 12-hour steps, it indicates that the emission source uses 12 hours as the basic period. The R(τ) of the 12-hour period of the emission source is preferentially identified as the basic emission rhythm, and the daily periodic peak is treated as the second harmonic of the 12-hour basic period.

[0032] The interruption cycle characteristics are output based on the weighted calculation of the confidence level of each cycle according to the emission rhythm fit. The emission rhythm fit R(τ) of each candidate dominant cycle is multiplied by the high-frequency confidence level of the density distribution peak of the corresponding dominant cycle and then normalized. Candidate cycles with a high proportion of shift fit annotations are further weighted on the product. The candidate cycle with the highest weighted product is identified as the dominant emission cycle. The interruption cycle characteristics are output with the interval value and confidence level of this cycle as the core content. Candidate cycles with emission rhythm fit below 0.6 are downweighted even if the frequency confidence level is high. Although occasional interval clusters can form frequency peaks, they cannot generate stable misaligned correlations. The downweighting process prevents such pseudo-cycles from being identified as dominant emission cycles. There are two reasons for low emission rhythm fit: one is that the interval exists but the emission time is unstable and the correlation is reduced; the other is pure statistical noise. The downweighting of candidate cycles with time drift annotations is smaller than that of unannotated low-fit candidate cycles, thus protecting the weakly regular true emission cycles. When outputting the interrupted cycle feature, the candidate cycles and their confidence levels of the top three weighted products are retained. For multiple cycles, scenarios with multiple emission rhythms at the same monitoring point are retained. The candidate cycles of harmonic annotation are recorded as independent fields in the interrupted cycle feature and are not listed in the ranking of the top three weighted products. They are reserved for the emission node labeling stage to be attached to the corresponding basic cycle nodes with harmonic verification annotation. The harmonics with high emission rhythm consistency are equivalent to independent verification of the basic cycle, and the overall confidence level of the corresponding node is increased accordingly. The interrupted cycle feature has a more robust and reliable conclusion for the identification of periodic emission sources.

[0033] Emission node markers are established based on the identification of recurring interruption clusters at fixed times using interruption cycle features. The interval value corresponding to the dominant cycle in the interruption cycle features is used to back-calculate each interruption time to the cycle phase. The set of interruption times with high phase consistency corresponds to the repeated activation of emission sources at fixed phase points. The coordinates of each high-density phase point on the time axis are the emission nodes. Phase aggregation is performed separately in two sets according to the work calendar, ensuring that the weekday rhythm is not diluted by holiday entries. The accuracy of the dominant cycle in the interruption cycle features determines the accuracy of emission node marker positioning. When there is a deviation in the dominant cycle, the cluster density after phase expansion will gradually disperse with the increase of historical batches. For nodes whose hit rate monotonically decreases with batches, phase drift warnings are added to the emission node markers, indicating that the dominant cycle estimation may be biased and needs to be re-extracted. The time coordinates of emission nodes are matched with time windows one hour before and after in the interruption time sequence. Phase nodes with a hit rate of 70% are confirmed as valid emission nodes and written into the emission node marker. Harmonic verification labels are simultaneously attached to the corresponding basic cycle nodes to increase their confidence. A two-shift factory is a typical example: the dominant peak of the interruption cycle characteristic falls at the 12-hour interval, with one emission node each around 06:00 and 18:00 daily. The 06:00 node corresponds to the morning shift's furnace start-up, and the 18:00 node corresponds to the evening shift's start-up. After the hit rate of the interruption time sequence reached the standard in the past three weeks, both nodes were written into the emission node tag. When there was a lack of hit records due to equipment maintenance shutdown in a certain week, the maintenance interval tag removed the entry for that week from the hit rate statistics, and the identification of effective nodes was not affected by the maintenance batch. The multiple candidate cycles output by the interruption cycle characteristic each generated independent node tags and then merged into the emission node tag, participating in the time period coverage of key sampling batches in parallel. When multiple node time windows overlap on the time axis, the overlapping time period tags have multiple overlapping nodes. The concentration peaks triggered by multiple emission rhythms at the same time are often higher. The multiple overlapping node tags of the emission node tags guide subsequent enrichment resources to concentrate on the above-mentioned high concentration periods.

[0034] The key sampling batches are determined based on the periodic interruptions in the emission node markers. For a two-shift plant, the furnace starts at 06:00 daily. Emission node markers have one-hour time windows before and after 06:00. Batches from 05:30 to 06:30 in the sampling log that overlap with this window and whose flow deviation records show an interruption starting at 06:05 falling within the window are identified as key sampling batches. In the first ten minutes after furnace ignition, the furnace temperature is not yet stable, combustion is incomplete, and the pollutant concentration in the initial emission stage is several times higher than during stable combustion, with a low background interference, making it the optimal time window for differential pressure enrichment. Historical batches with missing flow deviation records within the emission node marker time window are replaced by adjacent batches with interruption records. These supplementary batches are marked as estimated batches. The differential pressure threshold applicable to the estimated batches during the enrichment stage is further relaxed, as detailed in the concentrated pollutant sample generation process. The key sampling batches are arranged in ascending order of the deviation between the start time of the interruption and the coordinates of the emission node marker. The batch with the smallest deviation has the highest degree of agreement between the sampling time and the emission node, and is given priority in the differential pressure enrichment stage. When the effective node time windows of multiple emission node markers do not overlap, the key sampling batches of each node are screened separately and then merged. The merged key sampling batches are grouped and managed according to the node number. After multi-wavelength differential detection, the component differences of each group of concentrated pollution samples can be compared at different time periods to help determine whether the emission behavior at each time period comes from the same emission source or different process units.

[0035] Concentrated contaminant samples are obtained by triggering concentrated enrichment based on abnormal differential pressure spikes in key sampling batches. During the emission node time window, the differential pressure between the inlet and outlet of the sampling pipeline is continuously monitored for each batch of key sampling batches. Under normal sampling conditions, the differential pressure rises slowly with increasing filter membrane load. When the emission source is activated, causing a sudden increase in the upstream gas concentration, the accompanying particulate matter buildup resistance increases, and the differential pressure spikes abnormally within several sampling cycles. An increase exceeding 1.5 times the recent average is initially identified as an abnormal spike. The sampling pump increases the gas flow rate, preferentially enriching the high-concentration samples from the initial emission stage into dedicated enrichment tubes. These batch-by-batch enrichment tubes serve as the physical carriers for concentrated contaminant samples. Abnormal differential pressure spikes must be distinguished from the slow rise caused by cumulative filter blockage: differential pressure from cumulative blockage rises monotonically over time with a stable slope, while spikes caused by emissions occur abruptly within a few frames, with the slope immediately falling back. Enrichment enhancement is confirmed only when the sudden rise slope exceeds 8 times the slow rise slope. Slow rises in differential pressure caused by cumulative filter blockage do not trigger enrichment enhancement mode. Enrichment enhancement mode is prioritized for multi-node overlapping labeling periods, with the trigger threshold lowered by one level. High-concentration peak capture windows during multi-rhythm overlapping periods must be carefully protected. The differential pressure spike threshold for estimated and labeled batches in key sampling batches is appropriately relaxed. Since the estimated batch time deviates from the actual emission node, the initial high-concentration peak may have partially subsided. The relaxed threshold allows even subsided medium-concentration peaks to trigger enrichment enhancement. Each batch of enriched tubes is archived with a batch number and enrichment time as double keys. The enriched tubes of all key sampling batches are collected to form a concentrated pollution sample. The batch number of the concentrated pollution sample is bound to the node number marked by the emission node. The source tracing analysis in the multi-wavelength differential detection stage is carried out according to this batch-node index.

[0036] Step S103: Perform multi-wavelength differential optical detection on the concentrated pollutant sample to generate the original absorption intensity, identify the multi-component covariance break time period based on the original absorption intensity to determine the interference correction amount, and calculate the equivalent concentration deviation of each pollutant component based on the interference correction amount to establish a quantitative detection result.

[0037] Specifically, concentrated contamination samples are subjected to multi-wavelength differential optical detection to generate raw absorption intensities. Concentrated contamination samples carrying sampling information from different batches are sequentially fed into the optical detection cell via a spectrophotometer. Light emitted from a broadband light source is decomposed into a continuous spectrum covering the characteristic absorption wavelengths of the target contaminant. The negative logarithm of the ratio of the transmitted light intensity at each wavelength to the reference light intensity constitutes the absorbance value for that frame. The frames of spectra are arranged in the detection sequence to form the basic data layer of raw absorption intensities. The relationship between the absorbance value at the characteristic wavelength of each component and the concentration and absorption cross-section of each component is described by the following formula: In the formula Σ i This represents the summation of all target pollutant components (with i as the component index) component by component, where k is the wavelength index; A(λ k ) represents the wavelength λ kThe measured absorbance at point c is the original absorption intensity when the frame value is used; i σ(i,λ) represents the equivalent column density (optical path integral concentration) of the i-th component. k ) represents the i-th component in λ k The molar absorption cross section is determined by a standard gas calibration experiment; ε(λ) k The background value at that wavelength is subtracted after measurement with a clean air baseline before each batch is tested. When multiple components coexist, each component contributes to the cross-absorption at the characteristic wavelength of the other. The absorbance of each wavelength in the original absorption intensity is the superposition of the absorption of multiple components rather than the pure signal of a single component. Differential processing eliminates the broadband background by using the difference in absorbance of the reference wavelength near the characteristic wavelength. The narrow-band characteristic absorption of each component is separated from the broadband scattering, and covariance analysis directly uses this as input. The absorbance of high-concentration batches in concentrated contaminated samples may exceed the linear response range of the detection cell. The original absorption intensity is marked with saturation for frames with absorbance exceeding 1.8. If the nonlinearity of the saturation segment is mixed into the covariance analysis, it will fabricate non-existent covariance breaks.

[0038] In some embodiments, the step of identifying the multi-component covariance break period and determining the interference correction amount for the original absorption intensity includes: calculating the real-time covariance degree of each pollutant component using the original absorption intensity to obtain the component covariance coefficient; identifying the period when the component covariance coefficient suddenly drops below a threshold to determine the covariance break window; extracting the pure absorption signal of the single component dominant absorption based on the covariance break window to obtain a pure reference absorption value; and collecting the cross-sensitivity correction coefficients of each component based on the pure reference absorption value to output the interference correction amount.

[0039] The covariance coefficients of each pollutant component are obtained by calculating the real-time covariance degree of each component using the original absorption intensity. For example... Figure 2 As shown, the time series of covariance coefficients for each component pair constitute multidimensional covariance sequences such as component covariance coefficient AB sequence 1 and component covariance coefficient AC sequence 2. The absorbance time series of characteristic wavelengths of each component is extracted pairwise using a sliding window. The normalized correlation coefficient is calculated after removing the mean from the two component sequences within the window. Let x and y be the absorbance sequences of the two components of the same component pair after removing the mean within the current window. The component covariance coefficient ρ is calculated using the following formula: In the formula, ρ is the component covariance coefficient of the component pair in the current window, ranging from -1 to 1. A value close to 1 indicates that the two components are highly synchronously covariant, approaching zero indicates that the two components are linearly uncorrelated and fluctuate independently, and approaching -1 indicates that the two components are inversely covariant (when one component increases, the other decreases synchronously); m is the frame number within the window; Σm represents the summation of all valid frames within the window frame by frame; x(m) and y(m) are the absorbance of the two components in the m-th frame after removing the mean, respectively, taken from the current frame value at the characteristic wavelength of the component corresponding to the original absorption intensity; if the norm of the mean-removed sequence of a component within the window is lower than the threshold η (i.e., the concentration of the component is approximately constant within the current window), then the corresponding ρ is marked as invalid and excluded from the time series main link calculation. The time series of covariance coefficients of all component pairs after normalization are arranged side by side according to their numbers to form a multidimensional component covariance coefficient matrix time series. The window length is taken as the shortest effective emission period of the emission source, usually 20 minutes. If it is too short, noise will dominate, and if it is too long, the interrupted period will be smoothed and diluted, resulting in insufficient sudden drop amplitude. Low-concentration batches exhibit high noise levels, leading to a relaxation of the rapid rate threshold by approximately one-quarter (implemented in the trigger segment identification stage). During batch switching, incomplete rinsing of the detection pool results in background jumps in several frames before and after the switch. Baseline perturbation annotations are added to the corresponding frames, which are then excluded from the main sequence calculation along with the saturated frames. The entire break identification link implements silent protection for the annotated frames, and rate calculation and window determination both avoid these frames, recalculating with only valid frames.

[0040] For example, determining the covariance break window for the period when the component covariance coefficient identification coefficient drops sharply to below a threshold includes: calculating the rate of decrease in the covariance coefficient time series based on the component covariance coefficient to obtain the covariance decrease rate; identifying the set of time moments of the rate of sudden increase exceeding the threshold for the covariance decrease rate to determine the sudden drop triggering segment; calculating the effective duration of the single component dominance period in the sudden drop triggering segment to obtain the break window period; and eliminating pseudo break windows locked by residual common mode in the break window period to form the covariance break window.

[0041] The covariant descent rate is obtained by calculating the descent rate in the covariant coefficient time series based on the component covariant coefficients. The covariant coefficient time series of each component pair is obtained by dividing the difference between adjacent frames by the frame interval to obtain the frame-by-frame change rate. The absolute values ​​of the negative rates are arranged in frame order to form a covariant descent rate sequence. The larger the covariant descent rate, the more drastic the decrease of the covariant coefficient at that moment, and the higher the probability of covariance break. The set of time moments of rate surge exceeding the threshold in the covariant descent rate sequence is identified to determine the sudden drop trigger segment. Inter-frame noise of the component covariant coefficients will introduce random spikes into the covariant descent rate. Isolated high values ​​of a single frame are replaced by the average values ​​of the preceding and following frames. True covariance break produces high rates for multiple consecutive frames, and a single high value is insufficient to support a valid sudden drop identification. The coefficient changes near the baseline perturbation label frame in the component covariant coefficients are affected by background jumps, and the corresponding covariant descent rate at this point appears as an unreal descent rate. A silent window is set near the baseline perturbation label frame for rate calculation. The rate value within the window is replaced by linear interpolation of the effective frames at both ends. Baseline jumps caused by batch switching are not included in the covariant descent rate time series. The calculation frequency of the covariant decline rate is consistent with the update frequency of the sliding window of the component covariance coefficients. Frame-by-frame advancement provides near real-time basis for the rapid location of the sudden drop trigger segment. After the emission scenario switch, the covariant decline rate increases significantly within several sampling periods, and the identification delay of the sudden drop trigger segment generally does not exceed twice the length of the sliding window. When the emission process changes from organic solvent evaporation to sulfur-containing gas release, the covariant decline rate of organic and sulfur-containing components increases sharply to more than ten times the normal fluctuation within several frames after the switch. The number of frames for the rate increase is positively correlated with the transition time of the emission switch. In contrast, the covariant decline rate caused by the natural fluctuation of atmospheric background concentration changes slowly and with a lower amplitude. The morphological characteristics of the two types of cases in the covariant decline rate time series are completely different.

[0042] The set of time moments for sudden increases in the covariant rate of change exceeding the threshold is used to determine the sudden drop trigger segment. Frames with a covariant rate of change exceeding the threshold are identified as sudden drop trigger frames. A segment with multiple consecutive trigger frames constitutes a sudden drop trigger event. The start and end frame coordinates of the event and the corresponding peak value of the covariant rate of change together form a record of the sudden drop trigger segment. For low-concentration batches with a relaxed threshold, the sudden increase threshold is lowered by one-quarter of the aforementioned range before comparison, resulting in fewer missed events of weak signal batches. The sudden increase threshold is taken as the 95th quantile of the distribution of the covariant rate of change during historical normal covariant periods. If the threshold is too low, even normal covariant fluctuations can trigger false sudden drops; if the threshold is too high, weak covariant break events are easily missed. The value must be determined after a compromise based on the false detection rate and missed detection rate of historical monitoring batches. In the covariant rate of decline, isolated high values ​​have been replaced by the mean. The sudden drop triggering segment does not contain single-frame events caused by noise. In the sudden drop triggering segment, when the interval between two adjacent triggering events is less than 3 frames, they are merged into a single event, with the first triggering frame as the starting point and the last triggering frame as the ending point. The same break process will not be divided into multiple triggering events due to a brief rate drop in the middle. The sudden increase threshold is reassessed at the beginning of each quarter based on the distribution of the covariant rate of decline during the recent normal covariant period. When the concentration of atmospheric background components increases seasonally, the absolute magnitude of the rate of decline caused by the same emission switching changes accordingly. If the threshold remains fixed for a long period, there will be more false detections in high background seasons and more missed detections in low background seasons. The criteria for determining the sudden drop triggering segment must be adjusted synchronously with the background level of the current season.

[0043] The effective duration of the single-component dominance period is calculated in the sudden drop triggering segment to obtain the break window period. After the initial frame of each event in the sudden drop triggering segment, the covariance coefficient of the corresponding component pair is continuously tracked. A segment where the coefficient is below the covariance coefficient break threshold line 8 (0.3) for multiple consecutive frames is identified as the single-component dominance period 5. The number of frames of the dominance period is multiplied by the sampling period to convert it into the effective duration. The initial frame and the effective duration together define the time boundary of the break window period 4. The sudden drop triggering segment only marks the trigger time of the break. The break window period, based on this, characterizes the actual coverage of the single-component dominance period. The former is the basis for the latter's location, and the latter is the temporal constraint for extracting the pure reference absorbance value. In fast-fluctuation scenarios, the covariance coefficient recovers within a few frames after a sudden drop. The sudden drop triggering segment shows the minimum sample size where the triggering event occurs, but the effective duration of the break window period is less than 25 frames. A short window annotation is added to the corresponding period. The pure reference absorbance value of the short window annotation segment is replaced by the estimation of adjacent historical batches during the extraction stage. Break window periods with sufficient effective duration have a higher weight than short window annotation segments during the pure reference extraction stage. After the dedicated sulfide emission pipeline was opened, the sulfur-containing component dominated absorption for more than ten minutes. The corresponding break window period had sufficient effective duration and was classified as highly available. However, during the production line switchover, the concentrations of each component fluctuated rapidly and synchronously. The dominance of a single component lasted only a few frames before being covered by the coexistence of new multi-component components. The corresponding break window period had insufficient sample size and was classified as short window. The break window period was classified by duration to reflect the availability of each segment. The segment with higher availability contributed more to the estimation quality of the interference correction matrix. The signal-to-noise ratio of the corresponding component covariance coefficient time series jointly determined the final fusion weight.

[0044] During the fracture window period, pseudo-fracture windows locked by residual common mode are eliminated to form covariant fracture windows. Pseudo-fracture elimination is a prerequisite for ensuring the extraction quality of pure reference absorbance at point 6. The inclusion of missed pseudo-fractures will systematically skew the estimation of cross-sensitivity correction coefficients of each component, resulting in the interference correction matrix carrying implicit bias for a long time. During fracture window period 4, each candidate window is first identified by pseudo-fractures. When the absorbance of both components is in the near-zero concentration range within the window, the covariance coefficient is artificially suppressed due to the lack of correlation, forming pseudo-fracture elimination segment 7. The concentration approaching zero during the detection cell rinsing stage when switching batches is a typical scenario. When the absolute value of the absorbance of the target component at the characteristic wavelength within the candidate window is less than 0.02, it is identified as the near-zero concentration pseudo-fracture elimination segment 7, and the candidate pseudo-fracture elimination segment 7 is excluded from covariant fracture window 3. During fracture window period 4, the overall fluctuation labeling window for the simultaneous decrease of covariance coefficients of all components is also excluded. Overall disturbances such as light source jitter or airflow pulses are fundamentally different from the local fractures of single-component emission switching. Windows with simultaneous decrease of covariance coefficients of multiple components cannot extract effective pure references. After removing two types of candidates, the remaining windows are classified according to their duration and written into covariant fracture window 3. The confidence level of each entry is determined by combining the duration, signal strength, and false fracture identification results. When the sulfide-specific pipeline is opened alone, the sulfur-containing component dominates absorption for more than ten minutes, corresponding to the highest confidence level in covariant fracture window 3. Extended frames where the covariance coefficients at the beginning and end of the window body are still in the rising transition zone are marked as single-component fracture extensions in covariant fracture window 3. These extension segments are usable but must be downweighted. The seasonally effective proportion of covariant fracture window 3 varies seasonally with atmospheric background concentration, and the sampling strategy and correction matrix update frequency are adjusted accordingly.

[0045] Pure reference absorbance values ​​are obtained by extracting the pure signal of the single-component dominant absorption based on the covariant fracture window. The characteristic wavelength absorbance sequence of the target component in the single-component dominant period 5 within each effective window of the covariant fracture window 3 is extracted from the original absorption intensity. The absorbance of each frame within the window is weighted by the signal-to-noise ratio and used as the representative value of the pure reference absorbance extraction point 6. The saturated labeled frames have zero weight and are not included in the weighting. When the sulfur-containing emission pipeline in the chemical industrial park is opened independently, the absorbance of sulfur dioxide within the covariant fracture window continuously increases in the initial stage of valve opening, with a monotonically increasing trend in the absorbance time series, while the concentration is still dynamically changing. A linear detrending process is performed before weighted averaging, and the signal-to-noise ratio weighted average of the detrended residual time series is used as the representative value of the pure reference absorbance. If the trend component is not eliminated, the high value in the rising segment will cause a systematic bias in the mean, and the estimation of the cross-sensitivity of sulfur dioxide to adjacent components will be systematically higher than the true level. The pure reference absorbance of short-window labeled segments is estimated by substituting the weighted average of reference values ​​of the same component from adjacent historical batches. This estimated substitute value is assigned only a low fusion weight during the interference correction phase, and fracture events with insufficient samples do not significantly affect the estimation of correction matrix elements. For covariant fracture windows carrying segments with single-component fracture extension labels, their pure reference absorbance values ​​are listed separately, and the fusion weight of the extended segment extraction results is lower than that of the main segment. The pure reference absorbance values ​​of all valid covariant fracture windows are summarized by component number. The sample size of each component's pure absorbance reference library continuously expands with the operational cycle, and the estimation stability of the interference correction matrix improves synchronously with the increase in sample size.

[0046] The interference correction is output by accumulating cross-sensitivity correction coefficients for each component based on the pure reference absorbance. The pure reference absorbance has already eliminated the cross-absorption contributions of other components. These coefficients are then substituted into the multi-wavelength absorbance equations to deduce the measured absorption cross-sections of each component at each wavelength. The difference between the measured values ​​and the standard gas calibration values ​​quantifies the deviation of the current atmospheric composition from standard conditions; this deviation is the core source of the interference correction's cross-sensitivity correction coefficients. A larger sample size of pure reference absorbance results in a more stable estimation of the interference correction matrix. With a small sample size, the matrix elements will be affected by the random errors of single-event breakage events. The interference correction iteratively merges the pure reference absorbance from multiple breakage events using a time-reverse decay weighted approach, with a weight half-life of approximately half a month. Reference values ​​from recent events take precedence over historical events, and the correction matrix can be updated in a timely manner according to the seasonal changes in atmospheric composition. When pure reference absorbance values ​​are derived from historical batches at different atmospheric diffusion levels, the atmospheric background composition varies between batches with large differences in diffusion levels. Mixing and merging these batches will reduce the representativeness of the interference correction. Pure reference absorbance values ​​are stratified by atmospheric diffusion level and merged within the same layer. Different levels maintain independent versions of interference correction values, and quantitative detection results use the corresponding version for each detection batch based on the diffusion level at that time. When monitoring stations experience both winter and summer, the background concentration of sulfur dioxide in the atmosphere is higher in winter due to coal-fired heating. Its contribution to the cross-absorption of characteristic peaks of other components in the ultraviolet band is correspondingly increased, and the corresponding correction coefficient of the winter version of the interference correction is about 20% higher than that of the summer version. If the same version is used for both seasons, the concentration of some components in the summer quantitative detection results will be systematically overestimated. Stratified version management of interference correction values ​​is the core guarantee for the stable accuracy of quantitative detection results across seasons.

[0047] Quantitative detection results are established by calculating the equivalent concentration deviation of each pollutant component based on interference correction. The interference correction coefficient matrix and the original absorption intensity multi-wavelength absorbance matrix are solved simultaneously. The net absorbance of each component after deducting the contribution of cross-absorption is divided by the corresponding component molar absorption cross section and optical path to convert it into equivalent concentration. The equivalent concentration of each component is normalized based on the batch gas flow rate. The enrichment enhancement mode increases the flow rate. If normalization is not performed, the actual concentration will be systematically overestimated. The condition number of the interference correction coefficient matrix reflects the numerical stability of the decoupling operation. When the characteristic wavelengths of multiple components highly overlap, the condition number is too large, and direct inversion amplifies the numerical error. For batches with a condition number exceeding 50, the quantitative detection results are solved using regularization. The regularization parameter is determined based on the measured signal-to-noise ratio of the original absorption intensity, effectively controlling the numerical error and preserving the resolution of component concentration differences. Batches solved using regularization, along with batches with low signal-to-noise ratios, are marked with a solution uncertainty in the quantitative detection results. The equivalent concentration of batches marked with solution uncertainty is usable, but the solution accuracy is limited. The concentration mismatch is handled separately. The interference correction is replaced by the correction coefficient extracted from the measured covariance break period instead of the pure standard library theoretical value. The representativeness of the actual coexisting components in the atmosphere is better than the scheme relying solely on theoretical values. The equivalent concentration of each component in the quantitative detection results is bound to the corresponding emission node marker number. The differences in component contributions at different emission periods are quantified in the node group statistics. After joint analysis of the component concentration time series and atmospheric diffusion level at each node, the proportion of near-end source region contribution and long-distance transport background can be further separated.

[0048] Step S104: Based on the quantitative detection results and the atmospheric diffusion level, analyze the meteorological mismatch characteristics of pollutant concentrations to establish a meteorological decoupling record. Use the meteorological decoupling record to quantify the local contribution ratio of near-end emission intensity to obtain the corrected concentration value. Analyze the time-series periodic law of meteorological correction residuals from the corrected concentration value to output the emission law index.

[0049] In some embodiments, establishing a meteorological decoupling record based on the quantitative detection results and the atmospheric diffusion level to analyze the meteorological mismatch characteristics of pollutant concentrations includes: calculating the predicted diffusion and dilution values ​​of each component based on the quantitative detection results and the atmospheric diffusion level to determine the meteorological predicted concentration; quantifying the measured prediction residuals of the meteorological predicted concentration to obtain the concentration mismatch amount; identifying the downwind consistency characteristics of the mismatch intensity based on the concentration mismatch amount to establish a near-end source area marker; and collecting the persistent deviation distribution of unreported source areas according to the near-end source area marker to generate a meteorological decoupling record.

[0050] Based on quantitative detection results and atmospheric diffusion levels, the predicted diffusion and dilution values ​​of each component are calculated to determine the meteorological predicted concentrations. The Pasquale stability parameter corresponding to the atmospheric diffusion level is used to determine the diffusion parameter s. y With s zBased on the current wind speed and the source emission intensity declared in the pollution discharge permit, the background contribution concentration of each component at the monitoring point is estimated using a Gaussian diffusion model. The meteorological forecast concentration is calculated using the following formula: In the formula, i is the component number, j is the emission source number, and t is the sampling batch time; C pred (i,t) represents the meteorological forecast concentration (in g / m³) of the i-th component at time t. 3 ); C bg (i) represents the regional background concentration of the i-th component (in g / m³). 3 ), estimated from the average upwind concentration over the previous 24 hours; E j (i) represents the emission factor (in g / s) of the i-th component of the j-th declared emission source; G(s) y ,s z ,u,x j ) is the Gaussian diffusion dilution factor (dimensions s / m) under the current diffusion level conditions. 3 ); u represents the wind speed at the current time (in m / s); x j s represents the horizontal distance (in meters) from the j-th emission source to the monitoring point. y s z The horizontal and vertical diffusion widths (in meters) correspond to the diffusion levels. Lower diffusion levels correspond to s. y s z The smaller the value, the higher the predicted concentration contribution value under the same emission intensity. The atmospheric diffusion level codes corresponding to each batch of quantitative detection results are retrieved from the time series table and then input into the dilution factor calculation. The meteorological predicted concentration and the quantitative detection results are aligned moment-by-moment, and the concentration mismatch method has batch-level resolution. When upwind monitoring points are missing, C... bg (i) The average of quantitative detection results from the same historical atmospheric diffusion level period in the same region is used as a substitute, and the substitute batch is marked with a background estimation label. A stricter threshold is set for near-end source area identification. During the prevailing southeasterly winds in spring, emissions from the upwind side of the park cause a continuous increase in background concentration, compressing the difference between quantitative detection results and meteorological predictions, and potentially masking near-end emission signals; C during these seasonally high background periods... bg (i) By replacing the net background value after removing the near-end contribution, independent background concentration parameters are maintained for each season, and the seasonal adaptability of meteorological forecast concentrations is effectively guaranteed.

[0051] Concentration mismatch is obtained by quantifying the residual of measured predictions of meteorological forecast concentrations. The measured equivalent concentration of each component in each batch is subtracted from the corresponding meteorological forecast concentration for each batch to obtain the concentration mismatch for each component. A positive concentration mismatch indicates that the measured concentration is higher than the background prediction level, suggesting additional emission contributions not explained by the forecast. A negative mismatch usually corresponds to an overestimation of the background contribution by the meteorological forecast concentration, or an underestimation of the quantitative result for this batch. The concentration mismatches of all batches are arranged in both temporal and component dimensions to form a concentration mismatch matrix. Meteorological forecast concentrations are sensitive to the accuracy of diffusion parameters; small errors in diffusion parameters during low diffusion levels can cause large deviations in meteorological forecast concentrations. A systematically high concentration mismatch during low diffusion levels does not necessarily represent true near-end emissions. The concentration mismatch is adjusted for low diffusion levels based on the uncertainty of diffusion parameters (approximately ±15%). Only the portion exceeding the upper limit of the adjusted interval is considered a reliable near-end emission signal. In quantitative detection results, uncertain labeled batches are transferred to the corresponding positions of concentration mismatches. The concentration mismatch of uncertain batches is not included in the downwind consistency master sequence analysis, but is kept as an independent auxiliary for reference. False concentration mismatches introduced by insufficient quantitative accuracy are not included in the meteorological decoupling record. If a certain component in the multi-component matrix of concentration mismatches has a persistently high mismatch while the mismatches of other components are close to zero, it indicates that there is a specific near-end emission source for this component rather than a multi-component common source. In this case, the concentration mismatch is supplemented with a single-component high label for the specific mismatch component. Based on this, the near-end source area marking stage performs directional analysis on this component separately, thereby improving the accuracy of the directional identification of specific emissions.

[0052] Near-end source area markers are established based on the downwind consistency characteristics of concentration mismatch intensity identification. The distribution pattern of concentration mismatch at different wind angles over different time periods points to the location of near-end emission sources. If the concentration mismatch is consistently high under a specific wind direction, the upwind direction is most likely to have unreported near-end emission sources. Concentration mismatch is grouped into 16 wind direction sectors, and the mean of each sector constitutes the mismatch intensity distribution with wind direction as the horizontal axis. The wind direction of the distribution peak is the azimuth coordinate of the near-end source area marker. Concentration mismatch of background estimation labeled batches and undetermined labeled batches are not included in this main sequence statistics. Background estimation batches are separately verified according to a stricter consistency threshold before they can be included. A period when the mismatch intensity is consistently significantly higher in one wind direction than in other wind directions is considered a valid activation event for the near-end source region marker. The duration and corresponding wind direction angle are recorded. Multiple activation events belonging to similar wind direction angles are merged into the same near-end source region marker. Components carrying high single-component markers are analyzed separately for directional characteristics. Their concentration mismatch is statistically analyzed independently by sector, and the directional conclusion is listed separately. The direction of specific emissions is not diluted by the multi-component mean. When the concentration mismatch of multiple components is synchronously high in the same period, the mixed component characteristics and the component ratios of the quantitative detection results are used to jointly infer the industry type of the emission source. The mismatch ratio of benzene series compounds and chlorine-containing components in chemical emission sources often coincides with the volatility characteristics of industrial solvents. The mismatch ratio of each component is extracted from the near-end source region marker as an auxiliary fingerprint of the emission source type and attached to the near-end source region marker for reference during the source tracing inventory stage. For azimuths with fewer than 5 effective activation events, the average concentration mismatch of adjacent wind directions is used to supplement the estimation. Low-sample labels are added to the supplementary azimuths. The local contribution ratio of the low-sample labeled azimuths in the concentration value calculation stage is conservatively reduced. Therefore, the azimuths with scarce samples will not overestimate the near-end emission intensity.

[0053] Meteorological decoupling records were generated based on the persistent deviation distribution of unreported source areas using near-end source area markers. An unreported boiler room was located northwest of the monitoring station. During periods of northwesterly winds, the concentration mismatch in the northwest sector of the near-end source area markers remained consistently high. Each effective activation event was weighted by duration. An activation event lasting 4 hours during a northwesterly wind had a higher weight than a short-lived event lasting only 40 minutes. Longer-duration events generally exhibited more stable diffusion conditions and more representative concentration mismatches. The meteorological decoupling records primarily consist of the weighted average mismatch amount and the distribution of mismatch duration across various azimuth angles. In the first month, with only 3 northwesterly wind activation events, the average concentration mismatch was skewed by random fluctuations in diffusion levels of individual batches, reaching a deviation of up to 40%. After accumulating more than 20 activations in the third month, the average fluctuation converged to less than 8%, and the emission contribution from the unreported boiler room was stably represented in the northwest azimuth parameters of the meteorological decoupling records. The intensity difference of concentration mismatch at different atmospheric diffusion levels is stored in layers in the meteorological decoupling records. During the concentration value calculation stage, the mismatch distribution parameters of the corresponding layer are called according to the diffusion level at that time, and the systematic bias introduced by the mixed estimation of high and low diffusion levels can be fundamentally avoided. The mismatch distribution parameters of each direction in the meteorological decoupling records are accumulated with historical batches and then weighted and fused with new data in reverse decay over time. The near-end source area marker can be updated synchronously with actual dynamic changes and is not fixed to the initial estimate. The update of quantitative detection results only triggers the incremental update of the corresponding azimuth parameters in the meteorological decoupling records rather than a full recalculation, so the update burden under large-scale historical batches can be controlled.

[0054] The local contribution ratio of near-end emission intensity is quantified using meteorological decoupling records to obtain corrected concentration values. Concentration mismatch in each time period of the meteorological decoupling records is separated into local contribution and transport background based on windward consistency. When multiple points are deployed, the concentration mismatch at downwind points is consistently higher than that at upwind points; the difference is partly attributed to local near-end emissions. The difference between upwind and downwind concentration mismatch is divided by the measured total concentration to obtain the local contribution ratio. The trend of this ratio over time reflects the dynamic characteristics of near-end emission intensity. The local contribution ratio at low-sample labeled locations is conservatively reduced before being included in the calculation. The near-end contribution obtained by multiplying the local contribution ratio by the measured concentration of each component in the quantitative detection results is equivalent to the estimated value of the total measured concentration minus the transport background; both form the corrected concentration value. The quality of the mismatch distribution in the meteorological decoupling records directly determines the reliability of the corrected concentration value. During periods with high mismatch distribution noise, the corrected concentration value is labeled with high uncertainty. The corrected concentration value with high uncertainty is weighted less during the emission regularity index extraction stage and included in the residual periodic analysis. Since a single monitoring point cannot obtain the concentration difference between upwind and downwind directions, the meteorological decoupling record is replaced by the difference in dilution factor caused by wind speed changes at different times of the day. The quantitative detection results during the daytime when the wind speed is relatively high represent the regional background concentration level after sufficient dilution, while the nighttime when the wind speed is relatively low represent the near-surface accumulation of local emissions under inversion and low diffusion conditions. The difference between the two types of time periods is converted into the equivalent local contribution ratio of a single point. The representativeness of the corrected concentration value in the single-point scenario is lower than that of the multi-point scenario. The corresponding batch is marked with a single-point estimation. The treatment method for this type of batch is detailed in the index generation stage.

[0055] In some embodiments, the step of analyzing the time-series periodicity of meteorological correction residuals from the corrected concentration value to output an emission pattern index includes: calculating a residual deviation sequence after meteorological diffusion correction using the corrected concentration value to obtain a corrected residual sequence; analyzing the clustering rhythm of the residuals on the time axis to obtain a residual periodic rhythm; identifying continuous emission patterns with positive residual periodic bias based on the residual periodic rhythm to form an emission time density; and aggregating emission pattern intensity weights for each time period according to the emission time density to generate an emission pattern index.

[0056] The corrected residual sequence is obtained by calculating the meteorological diffusion-corrected residual deviation sequence using the corrected concentration values. The residual deviation values ​​for each batch are obtained by subtracting the trend baseline estimated using multi-day moving averages from the time series of each component of the corrected concentration values. The residual deviations of all batches are arranged chronologically to form the corrected residual sequence. The smoothing window of the trend baseline is taken as an integer multiple of the typical emission cycle of the emission source, commonly six times. The baseline should only capture slowly changing background drift, and the window should be offset from the emission cycle by an integer multiple to avoid including regular emission fluctuations in the trend estimate. The corrected residual sequence deviation values ​​corresponding to batches with high uncertainty in the corrected concentration values ​​are labeled with low confidence. Low confidence deviation values ​​are weighted less in the residual periodic rhythm analysis, as noise in periods with insufficient meteorological decoupling accuracy is insufficient to broaden the residual distribution or obscure the true periodic signal. The corrected concentration values ​​should approach zero during non-emission periods, and the deviation of the corrected residual sequence during these periods should be lower than the baseline noise level. If the concentration is expected to remain high during non-emission periods, it likely corresponds to an overestimation of the background in meteorological decoupling records for that period. In this case, the corrected residual sequence value is replaced with the average deviation of other low-concentration mismatch periods in the same batch, and a background bias label is added. False biases introduced by background overestimation are not included in the residual periodic analysis. The corrected residual sequences of the multi-component time series of corrected concentration values ​​form a multi-component residual matrix. When the multi-component residuals synchronously exhibit the same periodic rhythm, the confidence level of the emission periodic signal is higher than that of the single-component signal. The in-phase intensity of the co-peak values ​​of the residuals of each component is used as a multi-component verification field and is transmitted to the residual periodic rhythm analysis along with the corrected residual sequence. Ultimately, this plays a role in the multi-component phase comparison of the emission time density, effectively distinguishing between single-pollutant specific emissions and multi-component co-source emissions.

[0057] The residual periodic rhythm is obtained by analyzing the clustering rhythm of the residuals on the time axis of the corrected residual sequence. The batch deviation values ​​of each batch of the corrected residual sequence and the time coordinates are used together in the time series autocorrelation analysis. The autocorrelation coefficient corresponding to the time step τ reflects the statistical association between two batches of the corrected residual sequence at a distance of τ. The coefficient is significantly higher at a certain τ, indicating that the residuals have a repeated clustering pattern at that time interval. The residual periodic rhythm is based on the set of significant τ values. The low confidence labeled deviation values ​​of the corrected residual sequence are not included in the calculation of the main autocorrelation sequence. Instead, they are replaced by interpolation of the previous and subsequent effective batches and used to construct the basic covariance matrix. The low confidence batches do not introduce artificial periodic peaks. The autocorrelation curve of the corrected residual sequence shows positive maxima at integer multiples of the real emission cycle and local minima at half-integer multiples. The alternating positive and negative patterns are consistent with the anti-phase misalignment characteristics of the S102 emission rhythm in terms of physical meaning. The data sources of the two analysis paths are independent of each other, but they point to the same emission time series pattern, which mutually confirms and improves the reliability of the residual periodic rhythm. If the smoothing window of the trend baseline is close to the actual emission cycle, the baseline will absorb part of the average concentration within the emission cycle, thus suppressing the amplitude of the corrected residual sequence and the autocorrelation peak. The residual periodic rhythm tests the divisibility relationship between the smoothing window and the candidate cycle for candidate cycles with low autocorrelation peaks. If a divisibility relationship exists, the trend baseline is re-estimated and replaced with a window that is an integer multiple of the candidate cycle. After the smoothing suppression is eliminated, the autocorrelation peak of the corrected residual sequence recovers accordingly. The candidate cycle of the residual periodic rhythm will not be systematically missed due to baseline estimation bias.

[0058] The emission density at specific times is formed by identifying the continuous emission patterns with positive residual periodic bias based on the residual periodic rhythm. The dominant emission cycle determined by the residual periodic rhythm projects historical batches onto a unified periodic phase coordinate. After summing the corrected residual sequence deviation values ​​of each batch near the same phase point, the average value is statistically calculated according to the phase resolution given by the residual periodic rhythm. The phase-mean curve constitutes the concentration fluctuation pattern within a complete emission cycle. The peak phase of the curve corresponds to the concentrated emission period. The continuity and concentration of the peak phase, i.e., the spatial distribution of the emission density at specific times, is quantified by the proportion of batches with positive deviation within a unit phase width on the phase axis. The background bias annotation batches of the corrected residual sequences have been replaced upstream, and the phase distribution of emission time density is not affected by spurious biases from background overestimation. The bias values ​​of the corrected residual sequences in a certain phase interval are consistently close to zero with minimal fluctuations, mostly corresponding to fixed shutdown and maintenance periods of emission sources. The emission time density shows a near-zero groove in this phase interval, and the groove boundary is marked as a typical switching moment for emission start-up and shutdown. After comparing with the time coordinates marked by the S102 emission node, the consistency of the two analysis paths is verified, and the dominant period estimation of the residual periodic rhythm is also back-calibrated with the verification results. The phase distribution of the multi-component corrected residual sequences is highly consistent in the same peak interval, confirming that this phase corresponds to a single near-end emission source. The systematic shift of the peak phase of different components exceeds 1 hour, mostly corresponding to two types of components from different emission behaviors. The emission time density adds multi-source mixed annotations for this situation, and the emission regularity index is processed according to the annotation groups when outputting, and the independent emission time series characteristics of each component are effectively preserved.

[0059] An emission pattern index is generated by weighting the emission pattern intensity for each time period based on the emission time density. The emission pattern index is calculated by normalizing the proportion of batches with positive deviations in emission time density for each phase interval by multiplying it by the mean amplitude of the corrected residual sequence for the corresponding interval: In the formula, φ represents the phase coordinate of the emission cycle; I(φ) is the emission regularity index at phase φ; D(φ) is the proportion of batches with positive deviation in emission density at phase φ; r(φ) is the frame mean of the positive deviation of the corrected residual sequence for that phase interval (non-positive frames are counted as zero); the product of the two is normalized to the maximum value across all phases, and the emission regularity index takes a value between 0 and 1. When the product across all phases is zero, I(φ) is uniformly defined as 0; phase intervals close to 1 correspond to core emission periods with high emission intensity and high batch repetition rate, while phase intervals close to 0 correspond to non-emission or extremely weak emission periods. Low-confidence labeled batches are weighted and reduced when calculating D(φ), and the statistical impact of high-uncertainty batches on the emission regularity index is minimized; for phase intervals with a high proportion of single-point estimated labeled batches, the confidence level of the emission regularity index is downgraded, and the single-point source proportion is simultaneously recorded in the index attributes to understand its source limitations when tracing the source. In the phase intervals of multi-source mixed annotations in the emission time density, the emission regularity index is calculated separately for each component and output in parallel according to component number without merging. The independent emission time sequence fingerprints of each component are left for categorized reference in the source tracing list. The deviation between the peak phase of the emission regularity index and the coordinates of the S102 emission node marker time serves as a consistency indicator between the two paths. When the deviation is less than 45 minutes, they corroborate each other, and the confidence in identifying near-end emission sources is enhanced; a large deviation indicates that there is a systematic deviation in a certain path, which needs to be backtracked for verification. The update of the corrected concentration value triggers the incremental update of the emission regularity index for the corresponding phase. With the accumulation of historical batches, the peak phase of the emission regularity index gradually converges to a stable estimate of the actual emission start and stop times.

[0060] Step S105: Based on the emission regularity index, identify the baseline drift characteristics during low emission periods and output the instrument deviation index. Screen for periods of periodic deviation exceeding the threshold for the instrument deviation index to obtain a reliable dataset. Construct a multi-pollutant emission source tracing list from the reliable dataset and generate a detection and analysis report.

[0061] In some embodiments, the step of identifying baseline drift characteristics during low-emission periods based on the emission pattern index and outputting an instrument deviation index includes: determining a low-emission prediction window from the periodic time when the emission concentration is expected to approach zero, identified by the emission pattern index; obtaining a baseline measured distribution by measuring the baseline values ​​of the instrument within the statistical window period of the low-emission prediction window; determining the direction and magnitude of the systematic non-zero offset based on the baseline measured distribution to obtain a baseline drift characteristic quantity; and generating an instrument deviation index by analyzing the sensor collaborative drift mode during the emission pattern verification period based on the baseline drift characteristic quantity.

[0062] Low-emission prediction windows are determined by identifying cyclical periods where emission concentrations are expected to approach zero using the emission regularity index. Industrial park emission sources typically follow fixed production rhythms. Periods such as nighttime shutdowns for maintenance and complete shutdowns during holidays repeatedly show stable low values ​​in the full-phase distribution of the emission regularity index, providing objective conditions for instruments to collect true baseline readings under conditions where near-end emission contributions approach zero. Periods where the index value is below 0.1 for six consecutive phase intervals in the full-phase time series of the emission regularity index are identified as low-emission segments. The corresponding historical batch times are projected onto the current and future times to form a candidate set of times for low-emission prediction windows. The candidate set must also be cross-verified with the atmospheric diffusion level. Under stable weather conditions, even if the emission source is shut down, the pollutants accumulated in the previous period have not yet fully diffused and dissipated in the atmosphere, and the measured concentration may still be higher than the instrument baseline level. If the park shuts down around midnight, and the atmospheric diffusion level remains below level D that night, the pollutant clouds formed by previous emissions accumulate and remain near the ground. The measured concentration does not show a significant drop in several hours after the shutdown. The baseline reading for such periods is raised by the residual atmospheric concentration and does not meet the criteria for identifying an effective low-emission prediction window. The intersection of the low-emission segment of the emission regularity index and the atmospheric diffusion levels A to C can be identified as an effective low-emission prediction window. This ensures that the extraction of the instrument baseline value is carried out under conditions where both near-end emission contributions and atmospheric residual concentrations are at low levels. When monitoring stations are first deployed, the historical batch accumulation of emission regularity indices is not yet sufficient, and the confidence level of the low-emission segment is relatively low. In the initial operation phase, the low-emission prediction window uses manually marked known shutdown batches as the initial seed batch. As the confidence level of the emission regularity index continues to improve, it gradually transitions to a fully automatic identification mode.

[0063] The baseline measured distribution is obtained by analyzing the instrument's measured baseline values ​​within the statistical window period of the low emission prediction window. The equivalent concentrations of each component in the quantitative detection results corresponding to the effective time of the low emission prediction window are extracted batch by batch, and the frequency distribution is statistically analyzed in intervals with a resolution matching the measurement range. The frequency histograms of each component constitute a multi-component matrix of the baseline measured distribution. Batches labeled as "seeds" are included in the statistics with reduced weights. After the automatic identification of batches accumulates to a sufficient quantity, the statistical proportion of seed batches will decrease accordingly. In a typical low emission prediction window at 3:00 AM when the park is completely shut down and the wind speed is moderate, the readings of each component fluctuate randomly around zero when the instrument is drift-free. The baseline measured distribution has a peak center at a low concentration close to zero, is symmetrical, and has a narrow peak shape. After six months of operation, the UV detection channel light source ages, and the corresponding component readings at 3:00 AM remain consistently positive. The peak center of the baseline measured distribution shifts to a non-zero position around 0.03. If aging continues to accelerate, the peak position moves monotonically between monthly batch subsets, exhibiting a trend of drift. In the low-emission prediction window, batches with diffusion levels near the C-level criticality have higher measured concentrations due to residual atmospheric background. The high-concentration tails of the baseline measured distribution are largely contributed by the residual background of these batches. After truncating the high-concentration tails, the baseline measured distribution participates in drift analysis as the main distribution. The truncation threshold is the main peak position plus 2.5 times the standard deviation. The version number of each component matrix of the baseline measured distribution is bound to the corresponding low-emission prediction window batch number. Each new valid window batch is incrementally updated without requiring a full reconstruction. Under long-term operating conditions, the continuous tracking calculation burden of the instrument bias index does not increase with the accumulation of historical batches.

[0064] Based on the measured baseline distribution, the direction and magnitude of the systematic non-zero offset are determined to obtain the baseline drift characteristic quantity. The amount by which the weighted mean of the main distribution of each component in the measured baseline distribution deviates from zero is the current systematic offset of that component. The time series of offsets for each historical low emission prediction window batch reveals the evolution pattern of the drift. The trend slope of the time series reflects the baseline drift rate. A significantly non-zero slope corresponds to trend drift, while a slope close to zero and a persistently high absolute value of the offset corresponds to fixed-bias drift. These two types fall on the slope and intercept terms of the baseline drift characteristic quantity, respectively. The offset of the measured baseline distribution of each component is fitted with the corresponding batch time using linear regression. The fitted slope and intercept constitute the baseline drift characteristic quantity of that component. The mean of the fitted residuals quantifies the degree of fit of the measured baseline distribution to the linear drift assumption. A large residual indicates the presence of a nonlinear component in the drift. Nonlinear drift labels are added to the baseline drift characteristic quantity. During the instrument bias index generation stage, piecewise fitting is used, and the staged changes in the drift rate are effectively identified. Comparing the slope direction and amplitude of multi-component baseline drift characteristic quantities can reveal aging differences in each detection channel. A significantly higher drift slope for a particular component compared to others often corresponds to accelerated aging of optical components at the detection wavelength. In the ultraviolet (UV) band detection channel, the radiation intensity of the UV light source decreases monthly after long-term operation, and the reference light intensity standard shifts accordingly. The corresponding component's baseline drift characteristic quantity slope reaches more than three times that of the visible light channel, while the visible light channel, sharing the same detection cell, has a stable slope. The difference in slope between the two channels clearly points to UV light source aging rather than detection cell contamination. Adding a rapid aging label to this component triggers an upward adjustment in the priority of instrument deviation index updates.

[0065] The instrument bias index is generated based on the sensor-coordinated drift mode during the emission pattern verification period, analyzed using baseline drift characteristics. The emission pattern verification period is defined as the batch interval after the emission pattern index has fully converged and the low emission prediction window identification confidence level reaches 0.8. During this period, the drift characteristics of the measured baseline distribution are most statistically representative. Coordination is quantified by the correlation coefficient matrix of the drift slope of all components in the baseline drift characteristics. Components with high correlation coefficients often share the same optical path or detection module, exhibiting highly synchronized drift evolution. The instrument bias index groups highly coordinated drifting components into the same drift source for combined correction, thus avoiding the overfitting risk caused by independent correction of components with multiple degrees of freedom. The instrument bias index mainly consists of the estimated current drift offset of each component and the corresponding correction confidence level, and the offset correction amount of each component at the current moment is estimated using the following formula: In the formula, i is the component number, and t is the batch time calculated from the start of the emission law verification period; B corr (i,t) represents the correction estimate of the i-th component at time t; a(i) is the slope of the baseline drift feature fitting, representing the trend drift rate; b(i) is the fitting intercept, representing the fixed bias component; δ nl(i,t) represents the piecewise correction term for the nonlinear drift labeling batch, which is non-zero only during nonlinear labeling and is estimated by piecewise fitting residual interpolation. The correction amount B for each component of the instrument deviation index. corr (i,t) is subtracted from the measured concentration of the corresponding batch in the reliable dataset screening stage. The net concentration after subtraction represents the true near-end emission contribution after removing the systematic bias of the instrument. The correction accuracy of the instrument bias index is directly related to the validity of the reliable dataset and the accuracy of the traceability conclusion of the detection analysis report.

[0066] A reliable dataset was obtained by screening for periods of periodic deviation exceeding the threshold in the instrument deviation index. The correction values ​​B for each component of the instrument deviation index were also determined. corr The absolute value of (i,t) is compared batch by batch with the correction upper limit (e.g., 8% of the range). Batches with values ​​below the upper limit are considered to be in good instrument condition and have reliable measurement results. The reliable dataset includes the quantitative detection results of such batches minus B. corr The net concentration sequence after (i,t); batches exceeding the upper limit are considered to have exceeded the instrument deviation threshold, are excluded from the reliable dataset, and trigger equipment maintenance warnings. Operators use this information to check the detection optical path and sensor status. Periodic and random exceedances of the instrument deviation index correction amount must be handled separately. Periodic exceedances often correspond to systematic biases caused by a fixed operational event. A typical example is the brief baseline jump caused by the automatic zero-gas calibration operation performed at midnight on the first day of each month. These batches can be retrospectively corrected after being marked with calibration labels and then reinstated into the reliable dataset. Random exceedances often correspond to sudden instrument failures. These batches cannot be corrected, are permanently excluded from the reliable dataset, and trigger fault alarms. The uncertainty of the correction amount for the nonlinear drift labeling component of the instrument deviation index is relatively large near the piecewise fitting node. The reliable dataset uses a stricter exceedance judgment for batches of this type of component within one week before and after the piecewise node. The error in the estimation of the correction amount at the node should neither exclude reliable batches by misjudging them nor include unreliable batches in the reliable dataset. The reliable dataset is based on the batch time and net concentration sequence of each component. It includes phase annotations corresponding to emission regularity indices, atmospheric diffusion level codes, and near-end source area marker location information as extended attributes. During the source tracing inventory construction stage, the detection and analysis report classifies and aggregates the reliable dataset according to emission time period, diffusion conditions, and emission location.

[0067] A multi-pollutant emission source tracing inventory is constructed from a reliable dataset to generate a detection and analysis report. The net concentration sequences of multiple components in each batch of the reliable dataset are grouped according to the emission pattern index phase. The component concentration ratios within the peak intervals of each phase constitute the multi-pollutant emission fingerprints for that emission period. These emission fingerprints are matched against an industry emission characteristic fingerprint database (compiled from industry emission spectrum data and historical detection archives). If the benzene-to-toluene ratio falls within the typical range of the printing industry, it matches printing solvent emissions; if the proportion of sulfur-containing components is high, it matches coal-fired boiler emissions, thus inferring the most likely emission source industry type. This is then compared with the mismatch ratio auxiliary fingerprints retained in the near-end source area markers. If the two sets of fingerprints match, the confidence of the industry inference is strengthened; if they differ, it indicates a deviation in the component analysis link, requiring backtracking and verification. The industry inference results for each emission period are combined with the corresponding near-end source area marker directional coordinates to form spatial-component association entries in the tracing inventory. In reliable datasets, the component concentrations of multi-source mixed-label batches are grouped by source and retained as independent time series. The emission fingerprints of mixed batches are then grouped and used for industry inference of their respective emission sources. This effectively eliminates the misjudgment path caused by the overlapping proportions of components from different sources, which could lead to fingerprints deviating from single-industry characteristics. The consistency verification results of the emission pattern index peak phase and emission node markers corresponding to the reliable dataset are written into the detection and analysis report. Emission periods with high consistency are marked with dual-path verification in the report, while periods with low consistency indicate discrepancies between the two analysis paths, requiring manual verification. The detection and analysis report summarizes the emission periods, component concentration statistics, industry fingerprint similarity, and dual-path verification status of each near-end emission source, with quantitative estimation of emission contribution as the core. Emission periods with fewer than 10 batches in the reliable dataset are marked with low-sample confidence. The strength of the conclusions and the confidence level in the detection and analysis report are consistently matched, providing regulatory authorities with emission pattern archives with quantitative evidence.

[0068] To implement the above-described method embodiments, an automatic sampling and analysis method for air pollutants is proposed to achieve the corresponding functions and technical effects. See also... Figure 3 , Figure 3 This diagram illustrates a structural block diagram of an automatic air pollutant sampling and analysis device 300 according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The automatic air pollutant sampling and analysis device 300 provided in this embodiment includes: Data acquisition module 301 is used to collect ambient air samples and meteorological monitoring data, collect abnormal deviation periods of sampling flow based on the ambient air samples to obtain flow deviation records, and identify the characteristics of calm wind residence periods based on the meteorological monitoring data to determine the atmospheric diffusion level. The batch screening module 302 is used to identify the periodic pattern of sampling interruption time based on the flow deviation record to form key sampling batches, and to use the key sampling batches to trigger concentrated enrichment based on differential pressure abnormal jump to obtain concentrated contaminated samples; The spectral detection module 303 is used to perform multi-wavelength differential optical detection on the concentrated pollutant sample to generate the original absorption intensity, identify the multi-component covariance break time period based on the original absorption intensity to determine the interference correction amount, and calculate the equivalent concentration deviation of each pollutant component based on the interference correction amount to establish a quantitative detection result. Meteorological correction module 304 is used to establish a meteorological decoupling record based on the quantitative detection results and the atmospheric diffusion level to analyze the meteorological mismatch characteristics of pollutant concentration, use the meteorological decoupling record to quantify the local contribution ratio of near-end emission intensity to obtain the corrected concentration value, and analyze the time-series periodic law of meteorological correction residual from the corrected concentration value to output the emission law index. The report generation module 305 is used to identify baseline drift characteristics during low emission periods based on the emission pattern index, output an instrument deviation index, screen for periodic offset exceeding the threshold period based on the instrument deviation index to obtain a reliable dataset, and construct a multi-pollutant emission source tracing list from the reliable dataset to generate a detection and analysis report.

[0069] The aforementioned automatic air pollutant sampling and analysis device 300 can implement an automatic air pollutant sampling and analysis method according to the above-described method embodiments. The options in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining contents of this application's embodiments can be referred to the contents of the above method embodiments, and will not be repeated in this embodiment.

[0070] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

Claims

1. An automatic sampling and analysis method for air pollutants, characterized in that, include: Ambient air samples and meteorological monitoring data are collected. Based on the time periods of abnormal sampling flow deviation in the ambient air samples, flow deviation records are obtained. Based on the meteorological monitoring data, the atmospheric diffusion level is determined by identifying the characteristics of calm wind residence periods. Based on the periodic pattern of sampling interruption times identified in the flow deviation records, key sampling batches are formed. These key sampling batches are then used to trigger concentrated enrichment based on abnormal differential pressure spikes to obtain concentrated contaminated samples. The concentrated pollutant sample is subjected to multi-wavelength differential optical detection to generate the original absorption intensity. The multi-component covariance break time period is identified based on the original absorption intensity to determine the interference correction amount. The equivalent concentration deviation of each pollutant component is calculated based on the interference correction amount to establish a quantitative detection result. Based on the quantitative detection results and the atmospheric diffusion level, a meteorological decoupling record is established to analyze the meteorological mismatch characteristics of pollutant concentration. The meteorological decoupling record is used to quantify the local contribution ratio of near-end emission intensity to obtain a corrected concentration value. The meteorological correction residual time-series periodic law is analyzed from the corrected concentration value to output the emission law index. Based on the emission pattern index, baseline drift characteristics during low emission periods are identified, and an instrument deviation index is output. Periodic offset exceeding the threshold is screened for the instrument deviation index to obtain a reliable dataset. The reliable dataset is used to construct a multi-pollutant emission source tracing list and generate a detection and analysis report.

2. The method according to claim 1, characterized in that, The flow deviation record obtained based on the abnormal deviation period of the ambient air sample collection includes: The sampling flow sequence is generated by using the instantaneous flow rate change node of the ambient air sample detection sampling pump. The flow deviation distribution is obtained by calculating the coupling slope between gas path resistance and flow rate attenuation for the sampled flow rate sequence; Based on the aforementioned flow deviation distribution, aberrant, ultra-stable zero-fluctuation intervals are identified to form a set of deviation time periods; The set of deviation periods is used to collect the duration of the maximum deviation within the time period and output the flow deviation record.

3. The method according to claim 1, characterized in that, The step of identifying the periodic pattern of sampling interruption times based on the traffic deviation record to form key sampling batches includes: Extract the set of complete interruption event occurrence times from the traffic deviation records to generate an interruption time sequence; The interruption period characteristics are obtained by statistically analyzing the periodic distribution of adjacent interruption time intervals in the interruption time sequence. Based on the interruption cycle characteristics, identify cluster nodes with repetitive interruptions at fixed times and establish emission node markers; The key sampling batches are determined based on the periodic interruption batches corresponding to the emission node markers.

4. The method according to claim 1, characterized in that, The step of identifying the multi-component covariance breakage period based on the original absorption intensity and determining the interference correction amount includes: The covariance coefficients of each pollutant component are obtained by calculating the real-time covariance degree of each pollutant component using the original absorption intensity. The covariance break window is determined by the period during which the identification coefficient of the component covariance coefficient drops sharply below the threshold. Based on the covariant fracture window, a pure reference absorbance value is obtained by extracting the pure signal of the dominant single-component absorption; Based on the pure reference absorbance, the cross-sensitivity correction coefficients of each component are collected, and the interference correction amount is output.

5. The method according to claim 1, characterized in that, The establishment of a meteorological decoupling record based on the quantitative detection results and the atmospheric diffusion level analysis of pollutant concentration meteorological mismatch characteristics includes: Based on the quantitative detection results and the atmospheric diffusion level, the predicted diffusion and dilution values ​​of each component are calculated to determine the meteorological predicted concentration. The concentration mismatch is obtained by quantifying the measured prediction residuals of the meteorological predicted concentration. A near-end source region marker is established based on the downwind consistency characteristics of the concentration mismatch intensity. Meteorological decoupling records are generated by aggregating the persistent deviation distribution of unreported source regions based on the near-end source region markings.

6. The method according to claim 1, characterized in that, The step of analyzing the meteorological correction residual time-series periodicity from the corrected concentration value to output the emission pattern index includes: The corrected residual sequence is obtained by calculating the meteorological diffusion-corrected residual sequence using the corrected concentration value; The residual periodic rhythm is obtained by analyzing the clustering rhythm of the residuals on the time axis of the modified residual sequence. Based on the residual periodic rhythm, the continuous emission pattern of positive residual period deviation is identified to form the emission time density; An emission pattern index is generated by weighting the emission pattern intensity of each time period based on the emission time density.

7. The method according to claim 1, characterized in that, The method of identifying baseline drift characteristics during low-emission periods based on the emission pattern index and outputting an instrument deviation index includes: The low emission prediction window is determined by identifying the periodic periods during which the emission concentration is expected to approach zero from the emission regularity index. The baseline measured distribution is obtained by measuring the baseline values ​​of the instrument during the statistical window period of the low emission prediction window. Based on the measured distribution of the substrate, the direction and magnitude of the systematic non-zero offset are determined to obtain the baseline drift characteristic quantity; Based on the baseline drift characteristics, the emission law was analyzed to verify the sensor-coordinated drift mode and generate the instrument deviation index.

8. The method according to claim 3, characterized in that, The step of statistically analyzing the periodic distribution of adjacent interruption time intervals in the interruption time sequence to obtain interruption period characteristics includes: The interruption interval distribution is formed by statistically analyzing the duration distribution of adjacent interruption events using the interruption time sequence. The dominant periodic density distribution is determined by identifying high-frequency clustered fixed-interval peaks in the interruption interval distribution. Based on the dominant period density distribution, the anti-phase misalignment correlation coefficient of the period interruption density is calculated to obtain the emission rhythm fit degree; Based on the emission rhythm consistency, the confidence-weighted values ​​of each cycle are collected to output the interruption cycle characteristics.

9. The method according to claim 4, characterized in that, The determination of the covariance break window during the period when the identification coefficient of the component covariance coefficient drops sharply below the threshold includes: The covariant decrease rate is obtained by calculating the decrease rate of the covariant coefficient in the time series based on the covariant coefficient of the components. The set of time moments for the sudden increase in the rate of decrease of the covariant rate is used to determine the sudden drop trigger segment; The effective duration of the single-component dominance period in the sudden drop triggering segment is calculated to obtain the fracture window period. During the fracture window period, pseudo-fracture windows locked by residual common mode are eliminated to form covariant fracture windows.

10. An automatic sampling and analysis device for air pollutants, characterized in that, For implementing the automatic sampling and analysis method for air pollutants as described in any one of claims 1 to 9, the apparatus comprises: The data acquisition module is used to collect ambient air samples and meteorological monitoring data, collect flow deviation records based on the abnormal deviation period of the sampling flow from the ambient air samples, and determine the atmospheric diffusion level based on the characteristics of the calm wind residence period identified by the meteorological monitoring data. The batch screening module is used to identify the periodic pattern of sampling interruption times based on the flow deviation record to form key sampling batches, and to use the key sampling batches to trigger concentrated enrichment based on differential pressure abnormal jumps to obtain concentrated contaminated samples; The spectral detection module is used to perform multi-wavelength differential optical detection on the concentrated pollutant sample to generate the original absorption intensity, identify the multi-component covariance break time period based on the original absorption intensity to determine the interference correction amount, and calculate the equivalent concentration deviation of each pollutant component based on the interference correction amount to establish a quantitative detection result. The meteorological correction module is used to analyze the meteorological mismatch characteristics of pollutant concentration based on the quantitative detection results and the atmospheric diffusion level, establish a meteorological decoupling record, use the meteorological decoupling record to quantify the local contribution ratio of near-end emission intensity to obtain the corrected concentration value, and analyze the time series periodic law of meteorological correction residual from the corrected concentration value to output the emission law index. The report generation module is used to identify baseline drift characteristics during low-emission periods based on the emission pattern index, output an instrument deviation index, screen for periodic offset exceeding the threshold period based on the instrument deviation index to obtain a reliable dataset, and construct a multi-pollutant emission source tracing list from the reliable dataset to generate a detection and analysis report.

Images