A method and device for quality control and calibration of emission factors of industrial source atmospheric pollutants

Abstract

The application relates to the technical field of atmospheric pollutant emission characterization, and provides an emission factor quality control and verification method and device for industrial source atmospheric pollutants, which comprises the following steps: calculating an emission factor based on online monitoring data of an atmospheric pollutant continuous emission monitoring system and production load information; performing multi-dimensional data quality preliminary screening on basic emission characteristic parameters used for calculating the emission factor in a first stage to obtain a first-stage judgment result; if a current production capacity utilization rate is lower than a preset threshold, determining a quality control level of the emission factor based on the first-stage judgment result; if the current production capacity utilization rate is not lower than the preset threshold and the first-stage judgment result meets a preset credibility condition, performing Bayesian iteration statistical anomaly fine screening on the emission factor to obtain a second-stage recognition result; and determining the quality control level of the emission factor based on the first-stage judgment result and the second-stage recognition result, and performing graded processing on emission factors with different quality control levels; and the application improves the reliability of the emission factor.

Description

Technical Field

[0001] This application relates to the field of atmospheric pollutant emission characterization technology, and more specifically, to a method and apparatus for quality control and verification of emission factors of industrial source atmospheric pollutants. Background Technology

[0002] Currently, in the field of air pollution control, industrial source air pollutant emission factors are the core basic parameters for compiling air pollutant emission inventories and assessing emission reduction effects. With the popularization of Continuous Emission Monitoring System (CEMS) technology, real-time calculation of local emission factors based on equipment-level online monitoring data is gradually becoming an important data source for the dynamic updating of air pollutant emission inventories. However, in current emission inventory compilation practices, the quality control and verification of local emission factors is still relatively lacking. Existing research shows that emission factor quality control and verification methods mainly have two types of limitations: The first type is based on fixed statistical thresholds. This method is simple and effective, but once the threshold is fixed, it remains unchanged for a long time. When changes in the operating conditions of production equipment cause the overall distribution of emission factors to drift, the fixed threshold may not be suitable, and the percentile threshold setting has an inherent contradiction between false positives and false negatives; The second type is based on rule verification or machine learning methods. These methods usually treat CEMS data quality problems and extreme values ​​of emission factor statistics as a single link, without hierarchical distinction, which leads to misjudgment. In summary, existing methods have room for improvement in areas such as hierarchical processing of data quality verification and statistical anomaly identification, dynamic adaptive capability of statistical thresholds, and systematic nature of multidimensional data quality verification. Summary of the Invention

[0003] The purpose of this application is to provide a method and apparatus for quality control verification of emission factors of industrial source air pollutants, so as to realize two-stage hierarchical progressive quality control verification, which solves both the problem of data reliability and the problem of factor distribution drift, thereby improving the reliability of emission factors.

[0004] Firstly, a method for quality control verification of emission factors of industrial source air pollutants is provided, which may include: Based on online monitoring data and production load information from the continuous emission monitoring system for air pollutants, the emission factors of industrial sources are calculated; the online monitoring data includes basic emission characteristic parameters used to calculate the emission factors. In the first stage, a multi-dimensional data quality screening is performed on the basic emission characteristic parameters to obtain the first stage judgment result; If the current capacity utilization rate is lower than the preset threshold, the quality control level of the emission factor is determined based on the first-stage judgment result; the current capacity utilization rate is determined according to the production load information. If the current capacity utilization rate is not lower than the preset threshold and the first stage judgment result meets the preset credibility condition, then proceed to the second stage: perform Bayesian iterative statistical anomaly screening on the emission factors that have entered the second stage to obtain the second stage identification result, and determine the quality control level of the emission factors based on the first stage judgment result and the second stage identification result. Different emission factors at different quality control levels are subject to tiered treatment.

[0005] In one possible implementation, the first stage involves performing a multi-dimensional data quality screening on the basic emission characteristic parameters to obtain the first-stage judgment result, including: The basic emission characteristic parameters are evaluated using binary judgments on the judgment factors in multiple dimensions to obtain judgment values ​​for each dimension. Based on the preset weights of each dimension, the judgment values ​​are weighted and summed to obtain the quality control score. The quality control score will be used as the first-stage judgment result.

[0006] In one possible implementation, the multiple dimensions include four aspects: parameter integrity, physical consistency, channel consistency, and time series mutation; wherein: The parameter integrity level includes criteria for missing parameters, zero concentration, and zero denominator. The physical consistency level includes the determination factors of equal inflow and outflow, outflow exceeding inflow, and matching flue gas generation. The channel consistency level includes determination factors for multi-channel inbound differences and multi-channel outbound differences; The time series mutation level includes the determination factors for changes in emission factors and changes in pollution production coefficients.

[0007] In one possible implementation, Bayesian iterative statistical anomaly screening is performed on the emission factors entering the second stage to obtain the second-stage identification results, including: Obtain the natural logarithm of the emission factor at the current moment as the current observation value; Apply a forgetting discount to the accuracy of the historical posterior distribution of the previous time step, and use the distribution with the forgetting discount as the current prior distribution; Perform a conjugate normal update based on the current observation and the current prior distribution to obtain the posterior distribution at the current time. Based on the two-tailed probability of the posterior distribution of the current observation at the current time, the second-stage identification result of the current observation is determined. The initial prior distribution of the Bayesian iterative statistical anomaly screening is pre-established based on historical reliable emission factor samples.

[0008] In one possible implementation, the forgetting discount controls the equivalent memory length through a preset forgetting factor, so that the weight of historical information exceeding the equivalent memory length is reduced. The posterior distribution at the current moment is used as the prior distribution at the next moment to form an iterative closed loop; The second-stage identification results include low-value statistical anomalies, high-value statistical anomalies, and statistically normal values. The low-value statistical anomalies are when the emission factor is lower than the lower limit of the dynamic threshold interval, and the high-value statistical anomalies are when the emission factor is higher than the upper limit of the dynamic threshold interval.

[0009] In one possible implementation, the initial prior distribution is established as follows: Historical emission factor samples with capacity utilization rates higher than the preset threshold and first-stage judgment results meeting preset credibility conditions are obtained. A normal prior distribution is established by taking the natural logarithm of the historical emission factor samples, and the observation variance is estimated by the same historical emission factor samples. The observation variance is used as a fixed parameter in the conjugate normal update.

[0010] In one possible implementation, the method further includes: When the number of historical emission factor samples is less than a preset threshold, a resampling method is used to randomly sample the historical emission factor samples with replacement. After repeating this process multiple times, the empirical distribution of the obtained statistics is used to replace the normal prior distribution.

[0011] In one possible implementation, the method further includes: When the first-stage judgment result meets the preset confidence conditions and the second-stage judgment is a statistical anomaly, the statistical anomaly is classified and processed according to its duration and degree of deviation: For isolated minor anomalies, retain the original emission factors and tagged in the log; For continuous or significant anomalies, the alternative emission factor is calculated using the operating condition matching method and marked as requiring attention. For persistent anomalies, a temporary alternative emission factor is calculated using an extended time window operating condition matching method, which is then marked as under investigation and triggers manual investigation.

[0012] In one possible implementation, the quality control levels include four levels: The first level is considered trustworthy, corresponding to the first stage judgment result meeting the preset trustworthy conditions, and the second stage identification result being statistically normal. The second level is "needs attention", which corresponds to the quality control score in the first stage being the first preset value, or the identification result in the second stage being statistically abnormal. The third level is suspicious, corresponding to the quality control score of the first stage being within the range of the second and third preset values; Level 4 is unreliable, corresponding to a quality control score in the first stage that is not less than the fourth preset value; Among them, the first preset value is less than the second preset value, less than the third preset value, and less than the fourth preset value.

[0013] Secondly, an emission factor quality control and verification device for industrial source air pollutants is provided, the device may include: The calculation unit is used to calculate the emission factor of industrial sources based on online monitoring data and production load information from the continuous emission monitoring system for air pollutants; the online monitoring data includes basic emission characteristic parameters used to calculate the emission factor. The initial screening unit is used to perform multi-dimensional data quality initial screening on the basic emission characteristic parameters in the first stage to obtain the first stage judgment result. The determining unit is used to determine the quality control level of the emission factor based on the first-stage determination result if the current capacity utilization rate is lower than a preset threshold; the current capacity utilization rate is determined according to the production load information. The fine screening unit is used to enter the second stage if the current capacity utilization rate is not lower than a preset threshold and the first stage judgment result meets the preset credibility condition: perform Bayesian iterative statistical anomaly fine screening on the emission factors that enter the second stage to obtain the second stage identification result; The determining unit is further configured to determine the quality control level of the emission factor based on the first stage judgment result and the second stage identification result; The processing unit is used to perform graded processing on emission factors at different quality control levels.

[0014] This application provides a method and apparatus for quality control verification of emission factors of industrial source air pollutants, comprising: calculating emission factors of industrial sources based on online monitoring data and production load information from a continuous emission monitoring system for air pollutants; in the first stage, performing multi-dimensional data quality screening on the basic emission characteristic parameters used to calculate the emission factors to obtain the first stage judgment result; if the current capacity utilization rate is lower than a preset threshold, determining the quality control level of the emission factors based on the first stage judgment result; if the current capacity utilization rate is not lower than the preset threshold and the first stage judgment result meets the preset credibility condition, then proceeding to the second stage: performing Bayesian iterative statistical anomaly screening on the emission factors entering the second stage to obtain the second stage identification result, and determining the quality control level of the emission factors based on the first stage judgment result and the second stage identification result, and performing graded processing on emission factors with different quality control levels. This application, through a two-stage hierarchical progressive processing logic, solves both the problem of data reliability and the problem of factor distribution drift, thereby improving the reliability of emission factors. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating a method for quality control and verification of emission factors of industrial air pollutants provided in this application embodiment; Figure 2 A schematic diagram of the structure of an emission factor quality control and verification device for industrial source air pollutants provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application. Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art. The terms "first," "second," and similar terms used in this application do not indicate any order, quantity, or importance, but are only used to distinguish different components. The terms "comprising" or "including," etc., mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects. The terms "connection," "coupled," or "linked," etc., are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0018] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0019] Figure 1 This is a flowchart illustrating a method for quality control verification of emission factors of industrial source air pollutants, provided in an embodiment of this application. The method employs a two-stage, hierarchical, progressive processing logic to achieve a layered distinction between data quality verification and statistical anomaly identification. Figure 1 As shown, the method may include: Step S110: Calculate the emission factor of industrial sources based on online monitoring data and production load information from the continuous emission monitoring system for air pollutants.

[0020] Continuous Emission Monitoring System (CEMS) online monitoring data is collected and preprocessed on an hourly basis. For example, hourly averages are taken from minute-level raw data to smooth out random noise. Each industry selects a caliber based on its own process characteristics, such as per unit product, per unit raw material, or per unit of electricity generation, to calculate two core indicators: emission factor and pollution generation coefficient. Taking the coal-fired power generation industry as an example, the emission factor EF and pollution generation coefficient... The calculation formula is as follows:

[0021]

[0022] Among them, EF is the emission factor, which reflects the amount of pollutants emitted per unit of electricity generated, and the unit is g / kWh; The pollution generation coefficient reflects the amount of pollutants generated per unit of electricity generated, and its unit is g / kWh; The hourly average concentration of pollutants at the outlet is expressed in mg / m³. Here, Q represents the hourly average concentration of pollutants at the inlet, in mg / m³; Q represents the hourly average flue gas flow rate, in m³ / h; and P represents the hourly average unit load, in MW. The online monitoring data for inlet and outlet pollutant concentrations and flue gas flow rates used to calculate emission factors and pollution generation coefficients can be directly obtained from the CEMS system. Unit load information (or "production load information"), i.e., the hourly average unit load P, can be obtained from the enterprise's distributed control system for pollution control (DCS system).

[0023] It should be understood that the coal-fired power generation industry is only used as an illustrative example. The method in this embodiment has industry-wide applicability, and other industries can refer to the same calculation logic, replacing P in the denominator with the capacity or output indicators applicable to their industry. For example, in the cement industry, it can be replaced with clinker output, and in the steelmaking industry, it can be replaced with crude steel output, etc.

[0024] Step S120: In the first stage, a multi-dimensional data quality screening is performed on the basic parameters of emission characteristics used to calculate the emission factor to obtain the first stage judgment result.

[0025] The first phase primarily identifies quality control issues with the fundamental emission characteristic parameters used to calculate emission factors, such as missing parameters, physical inconsistencies, or equipment malfunctions. The multi-dimensional data quality screening covers multiple levels, including parameter completeness, physical consistency, channel consistency, and time-series abrupt changes, to comprehensively assess data reliability.

[0026] The results of the first stage of assessment can be expressed as quality control scores or other indicators that can quantify the reliability of the data. This first stage of processing addresses the question of the data's inherent reliability, providing a credible data foundation for subsequent anomaly identification.

[0027] Step S130: Determine whether the current capacity utilization rate is lower than the preset threshold.

[0028] If the current capacity utilization rate is lower than the preset threshold, then proceed to step S140; If the current capacity utilization rate is not lower than the preset threshold and the first-stage judgment result meets the preset credibility condition, then proceed to step S150.

[0029] The current capacity utilization rate is determined based on production load information, such as the ratio of the current production load P to the rated production load. A preset threshold is used as a switch to control the start and stop of the second stage, with a default value of 20%. This preset threshold is a configurable parameter, and different industries can adjust it according to their actual process characteristics; this application does not impose any limitations on it.

[0030] When the capacity utilization rate is lower than the preset threshold, it indicates that the industrial source is in the start-up, shutdown, or extremely low capacity stage. At this time, the emission characteristics are usually unstable, and the emission factors fluctuate drastically, which is not statistically significant. Therefore, in this case, the second stage is not activated, and step S140 is executed directly.

[0031] When the capacity utilization rate is not lower than the preset threshold, the emission factor EF distribution is relatively stable. If the first stage judgment result meets the preset credibility condition, such as the quality control score is 0, then the second stage is entered for fine screening, i.e., step S150 is executed.

[0032] By introducing a preset threshold for capacity utilization as a phased start-stop switch, statistical misjudgments caused by emission factor fluctuations under extremely low capacity utilization are effectively avoided, and the robustness of anomaly identification is improved.

[0033] Step S140: Determine the quality control level of the emission factor based on the results of the first stage judgment.

[0034] When capacity utilization is below a preset threshold, the quality control level is determined directly based on the initial screening results of multi-dimensional data quality in the first stage, and corresponding grading processing is performed. Since emission characteristics are unstable under extremely low capacity utilization, skipping the second stage of statistical anomaly screening avoids misjudging normal operating fluctuations as statistical anomalies, thereby reducing false detections.

[0035] Step S150: Perform Bayesian iterative statistical anomaly screening on the emission factors entering the second stage to obtain the second stage identification results.

[0036] The second stage primarily identifies whether emission factors reflect true emission levels, i.e., whether the factor distribution has drifted or whether the current value is at a statistical extreme. The Bayesian iterative statistical anomaly screening establishes an initial prior distribution based on emission factor samples whose historical capacity utilization rate has not fallen below a preset threshold and whose first-stage judgment results meet preset credibility conditions. Through iterative updates, the statistical threshold dynamically adapts to changes in operating conditions. The second-stage identification results include low-value statistical anomalies, high-value statistical anomalies, and statistically normal values.

[0037] By employing a hierarchical, progressive processing approach in the first and second stages, misjudgments caused by conflating data quality issues with statistical extreme values, as is common in existing technologies. For example, when CEMS data suffers from equipment malfunctions leading to data quality problems, the first stage can identify and intercept the issue; while when the CEMS data itself is reliable, but the emission factor value is at a statistically extreme level due to production anomalies, the second stage will handle the identification. The two stages have a clear division of labor and are seamlessly integrated.

[0038] Step S160: Based on the first-stage judgment results and the second-stage identification results, determine the quality control level of the emission factor; The quality control level can be divided into four levels: Level 1 is reliable, corresponding to the first-stage judgment result meeting the preset reliability conditions and the second-stage identification result being statistically normal; Level 2 is of concern, corresponding to the first-stage quality control score being the first preset value, or the second-stage identification result being statistically abnormal; Level 3 is doubtful, corresponding to the first-stage quality control score being within the range of the second and third preset values; and Level 4 is unreliable, corresponding to the first-stage quality control score being not less than the fourth preset value. The first preset value is less than the second preset value, less than the third preset value, and less than the fourth preset value.

[0039] For example, a trustworthy level corresponds to a quality control quality score of 0 and is statistically normal; a level requiring attention corresponds to a quality control quality score of 2 or is statistically abnormal; a questionable level corresponds to a quality control quality score of 3 to 4; and an untrustworthy level corresponds to a quality control quality score greater than or equal to 5.

[0040] Step S170: Perform grading treatment on emission factors of different quality control levels.

[0041] For emission factors at different quality control levels, differentiated tiered processing is implemented: emission factors at the reliable level are directly entered into the database for use; emission factors at the attention level are available but logged, and a review request is pushed out; emission factors at the suspicious level are marked for updating, and recalculated after a time limit for investigation; emission factors at the unreliable level are unusable, and an alarm is triggered immediately. Through tiered processing, closed-loop management of abnormal data is achieved, improving the data quality of dynamic updates to the emission inventory.

[0042] In some embodiments, before calculating the emission factor of an industrial source based on online monitoring data and production load information from a continuous emission monitoring system for air pollutants, the original time series of each monitoring parameter in the online monitoring data can be obtained. These monitoring parameters include at least the concentration of particulate matter (dust), the concentration of gaseous pollutants, and the temperature, pressure, velocity, and oxygen content of the flue gas. A timestamp alignment operation is performed on the original time series, uniformly interpolating the data of different monitoring parameters onto an equally spaced time grid. For each time series of the interpolated and aligned monitoring parameters, a smoothing filter based on local weighted regression is performed to remove high-frequency measurement noise. Based on the smoothed and filtered time series of each monitoring parameter, the instantaneous value of the emission factor at each moment is calculated according to a preset emission factor calculation formula. This implementation eliminates the data asynchrony problem caused by differences in sampling frequencies of different monitoring devices and the interference of high-frequency noise on the calculation of instantaneous emission factors, improving the accuracy and stability of the calculated emission factor values ​​and providing a more reliable data foundation for subsequent quality control judgments.

[0043] In some embodiments, the first stage performs a multi-dimensional data quality screening on the basic emission characteristic parameters used to calculate the emission factor, obtaining the first stage judgment result, including: The emission characteristic basic parameters used to calculate the emission factor are binary-coded across multiple dimensions to obtain the judgment value for each dimension. Based on the preset weights of each dimension, the judgment values ​​are weighted and summed to obtain the quality control score. The quality control score is used as the first-stage judgment result.

[0044] Binary judgment refers to classifying each dimension's judgment factor as normal or abnormal; for example, a judgment value of 0 indicates normal, and a judgment value of 1 indicates abnormal. The advantage of this binary processing is that it transforms complex physical contradictions and data defects into unified quantitative indicators, facilitating subsequent mathematical operations and logical aggregation. By assigning different preset weights to each dimension and performing weighted summation, a quality control score that comprehensively reflects the severity of data quality defects can be obtained. This score can intuitively reflect the severity of data quality problems. This multi-dimensional data quality initial screening covers four levels: parameter integrity, physical consistency, channel consistency, and time series mutations. These four levels comprehensively cover the key quality control points of CEMS data from acquisition, physical logic, to dynamic changes.

[0045] The parameter integrity aspect can include criteria for missing parameters, zero concentration, and zero denominator. The missing parameter criterion is used to check whether any key parameter (such as concentration, flow rate, or load) is missing or invalid. The zero concentration criterion is used to check whether the inlet or outlet concentration is less than or equal to 0, because in actual physical processes, pollutant concentrations cannot be zero or negative; this usually indicates sensor malfunction or calibration error. The zero denominator criterion is used to check whether the production load is less than or equal to 0 but the flue gas flow rate is greater than 0; this state will cause errors in the division operation in the emission factor calculation formula.

[0046] The physical consistency level includes judgment factors for equal inlet and outlet concentrations, outlet concentration exceeding inlet concentration, and flue gas generation matching. The equal inlet and outlet concentration judgment factor checks for situations where the inlet concentration is greater than 0 and the inlet concentration equals the outlet concentration. In normal desulfurization and denitrification processes, the outlet concentration should be lower than the inlet concentration; if they are equal, it indicates abnormal monitoring data or that the equipment is not operating. The outlet concentration exceeding inlet concentration judgment factor checks whether the outlet concentration is greater than the inlet concentration plus a set range and within the reaction temperature window. Under normal operating conditions, the outlet concentration should not exceed the inlet concentration unless under specific physicochemical reaction conditions. The flue gas generation matching judgment factor checks whether the flue gas generation per unit of product or per unit of fuel consumption exceeds a reasonable range when the operating load is higher than the set value. This reflects the physical correspondence between combustion efficiency and flue gas volume.

[0047] The channel consistency level includes decision factors for multi-channel inlet difference and multi-channel outlet difference. The multi-channel inlet difference decision factor is used to check whether the maximum difference in inlet concentration of each channel in a multi-channel configuration exceeds a set range. The multi-channel outlet difference decision factor is used to check whether the maximum difference in outlet concentration of each channel in a multi-channel configuration exceeds a set range. The decision factors at the channel consistency level are applicable to systems with multi-channel CEMS configurations. When the system is a single-channel system, the determination of these two decision factors is automatically skipped to avoid misjudgments caused by system configuration differences.

[0048] The time series abrupt change level includes criteria for emission factor jumps and pollution factor jumps. The emission factor jump criterion is used to check if the rate of change of the emission factor between adjacent time points is greater than 100%. The pollution factor jump criterion is used to check if the rate of change of the pollution factor between adjacent time points is greater than 100%. These two criteria are used to capture instantaneous abnormal fluctuations in the data. It should be understood that for systems with only online export monitoring and no online import monitoring, criteria involving import parameters such as equal import and export, export exceeding import, multi-channel import differences, multi-channel export differences, and pollution factor jumps will be automatically skipped.

[0049] To illustrate this application more clearly, the following table lists the judgment logic and corresponding preset weights for each dimension: Table 1

[0050] In Table 1, dimension D1 is parameter missing, determined by any missing or invalid key parameter, with a preset weight of 5 and a category of completeness. Dimension D2 is zero concentration, determined by either the inlet or outlet concentration being less than or equal to 0, with a preset weight of 5 and a category of completeness. Dimension D3 is a denominator of zero, determined by production load being less than or equal to 0 but flue gas flow being greater than 0, with a preset weight of 4 and a category of completeness. Dimension D4 is equal inlet and outlet, determined by inlet being greater than 0 and inlet equal to outlet, with a preset weight of 4 and a category of physical consistency. Dimension D5 is outlet exceeding inlet, determined by outlet being greater than inlet plus a set range and within the reaction temperature window, with a preset weight of 3 and a category of physical consistency. Dimension D6 is flue gas generation matching, determined by capacity utilization exceeding a set value, with the unit product flue gas generation or unit fuel consumption flue gas generation exceeding a reasonable range, with a preset weight of 2 and a category of physical consistency. Dimension D7 is multi-channel inlet difference, determined by the maximum difference in inlet concentration between multiple channels being greater than a set range, with a preset weight of 2 and a category of channel consistency. Dimension D8 represents the difference in concentration at the outlets of multiple channels. The judgment logic is that the maximum difference in concentration at the outlets of each channel exceeds a set range when there are multiple channels. The preset weight is 2, and the category is "channel consistency." Dimension D9 represents the jump in emission factors. The judgment logic is that the rate of change of emission factors between adjacent time points exceeds 100%. The preset weight is 2, and the category is "temporal abrupt change." Dimension D10 represents the jump in pollution generation coefficients (based on...). To determine whether a jump has occurred, the judgment logic is that the rate of change of the pollution generation coefficient between adjacent time points is greater than 100%, the preset weight is 2, and the category is time jump.

[0051] Based on the above judgment logic and preset weights, the formula for calculating the quality control score using binary judgment and weighted summation is as follows:

[0052] QC1 represents the quality control score, and D1 to D10 are the judgment values ​​for the aforementioned 10 judgment factors, with values ​​ranging from 0 to 1. The parameters are obtained as follows: the judgment values ​​for D1 to D10 are calculated in real-time by the system based on CEMS online monitoring data. For example, D1 is obtained by checking whether the data field is empty or has an invalid value; D2 is obtained by comparing the concentration value with 0; and D9 and D10 are obtained by calculating the rate of change of the emission factor or pollution generation coefficient between the current time and the previous time. Pollution generation coefficient. Used to test whether the D10 pollution generation coefficient has changed abruptly.

[0053] It should be noted that D7 and D8 are applicable to systems with multi-channel CEMS configurations, and are automatically skipped in single-channel systems. D4, D5, D7, D8, and D10 are applicable to systems where enterprises have online monitoring configured for both imports and exports, and are automatically skipped when the enterprise only has online monitoring configured for exports and not imports. When certain dimensions are automatically skipped, their default value is 0 and they are not included in the weighted summation calculation. This flexible applicability design allows the method in this application to adapt to CEMS systems with different configurations.

[0054] The specific settings for each of the above dimensions can be dynamically adjusted in the system configuration according to the process characteristics of different industries to demonstrate the method's industry applicability and anti-circumvention capabilities. For example, the excess range of D5's outlet to inlet, the operating load setting and the upper and lower limits of the reasonable range for flue gas generation in D6, and the channel difference ranges for D7 and D8 can all be configured according to the process characteristics of different industries. In the coal-fired power generation industry, due to the presence of desulfurization and denitrification facilities, the outlet concentration is usually much lower than the inlet concentration, so the setting range of D5 can be appropriately relaxed; while in the cement industry, due to process differences, the reasonable range for flue gas generation needs to be specifically calibrated based on clinker production and coal consumption. By dynamically adjusting these settings, this application can accurately identify data quality problems in different industries, prevent misjudgments or omissions caused by process differences, and thus establish a multi-dimensional defense depth.

[0055] In an optimized implementation, the preset weights of the above dimensions are not fixed values, but are dynamically adjusted according to the industry type, production process characteristics, and historical operating status of the online monitoring system of the industrial source. The dynamic adjustment method is as follows: the historical volatility of the judgment values ​​of each dimension within a preset period (such as the past 30 days) is statistically analyzed, and the dimension with the greater volatility is assigned a lower weight, and the dimension with the smaller volatility is assigned a higher weight. When it is found that the judgment value of any dimension has not been abnormal within multiple consecutive time windows (such as each day of the past 30 days), the weight of that dimension is gradually reduced at a preset decay rate. When the dimension is abnormally judged again, the weight is restored to the baseline value.

[0056] Specifically, the system maintains a fixed-length sliding history window for each quality dimension (such as parameter integrity, physical consistency, etc.) to store the binary judgment results (normal or abnormal) of the corresponding dimension for each day over the past 30 days.

[0057] At the end of each statistical period, the system iterates through these historical windows, calculating the frequency of jumps or reversals in the judgment values ​​of each dimension, i.e., historical volatility. Dimensions with higher historical volatility indicate that the data in that dimension is unstable and has lower reference value. Therefore, the system assigns a smaller weight to that dimension when calculating the quality control score. Conversely, dimensions with lower historical volatility indicate that the data in that dimension has remained stable over a long period, and the system gives it a larger weight.

[0058] The system maintains a separate counter for consecutive normal days and a baseline weight value stored in the configuration library for each dimension. After initial data quality screening, the system checks the judgment results for each dimension one by one: if a dimension is judged to be without anomalies on a given day, its consecutive normal days are incremented by one; when the cumulative number of days reaches a preset trigger threshold, the current weight of the dimension is multiplied by a decay coefficient slightly less than 1 during daily weight updates, causing it to gradually decline each day, preventing dimensions that are not error-free but are not critical from occupying an excessively high weight share for a long time. The decay process continues until the weight reaches a preset minimum protection limit, at which point the decay stops to prevent it from reaching zero. If the dimension is judged to be abnormal again within a subsequent time window, the system immediately performs two actions: first, it directly resets the counter for consecutive normal days of the dimension to zero; second, it instantly restores the weight of the dimension to the initially set baseline weight value, allowing it to regain its due influence in subsequent weighted calculations, and then re-enters the next round of volatility assessment or decay cycle based on subsequent data performance.

[0059] In some embodiments, Bayesian iterative statistical anomaly screening is performed on the emission factors entering the second stage to obtain the second-stage identification results, including: obtaining the natural logarithm of the emission factors at the current time as the current observation; applying a forgetting discount to the accuracy of the historical posterior distribution at the previous time, and using the distribution after applying the forgetting discount as the current prior distribution; performing a conjugate normal update based on the current observation and the current prior distribution to obtain the posterior distribution at the current time; determining the second-stage identification result of the current observation based on the two-tailed probability of the posterior distribution of the current observation at the current time; wherein, the initial prior distribution of the Bayesian iterative statistical anomaly screening is pre-established based on historical reliable emission factor samples; calculating the two-tailed probability of the current observation under the posterior distribution at the current time, if the emission factor is lower than the lower limit of the dynamic threshold determined by the lower quantile of the posterior distribution, it is determined to be a low-value statistical anomaly, and if the emission factor is higher than the upper limit of the dynamic threshold determined by the upper quantile of the posterior distribution, it is determined to be a high-value statistical anomaly.

[0060] Specifically, the system obtains the emission factor at the current moment. And take its natural logarithm. =ln( The current observation value is used as the logarithmic value. The physical meaning of the logarithmic transformation is that the emission factor usually follows a log-normal distribution. Taking the logarithm can transform it into a normal distribution, thus satisfying the precondition of conjugate normal update. Subsequently, the system applies a forgetting discount to the accuracy of the historical posterior distribution of the previous moment and uses the distribution after applying the forgetting discount as the current prior distribution. The forgetting discount controls the equivalent memory length L through a preset forgetting factor, so that the weight of historical information exceeding the equivalent memory length is decayed; the current posterior distribution forms an iterative closed loop as the prior distribution of the next moment; among them, the identification results of the second stage can include: low-value statistical anomaly, high-value statistical anomaly, and statistical normal. Low-value statistical anomaly is when the emission factor is lower than the lower limit of the dynamic threshold interval, and high-value statistical anomaly is when the emission factor is higher than the upper limit of the dynamic threshold interval. The value range of the forgetting factor λ can be [0.99, 0.9999], and the equivalent memory length L controlled by it is 1 / (1-λ). For example, when λ=0.999875, L is approximately equal to 8000 hours (approximately 1 year).

[0061] In some optimized embodiments, the forgetting factor can be adaptively adjusted in the following way: obtaining the statistical anomaly triggering frequency within a preset time window before the current moment; when the statistical anomaly triggering frequency is lower than a preset low-frequency threshold, increasing the forgetting factor by a first increasing step size to extend the equivalent memory length; when the statistical anomaly triggering frequency is higher than a preset high-frequency threshold, decreasing the forgetting factor by a first decreasing step size to shorten the equivalent memory length; wherein, the adaptive adjustment range of the forgetting factor is also limited within the preset upper and lower boundary range of the forgetting factor to prevent the forgetting factor from exceeding a reasonable range and causing Bayesian iteration instability.

[0062] By applying a forgetting discount, the implicit weight of historical sample i is... Historical information exceeding the equivalent memory length automatically has its weight reduced to zero, thus preventing overconfidence caused by the infinite accumulation of historical information. In a practical system, this iterative process can be directly implemented using a sliding window: maintaining the L most recent equivalent memory length samples entering the second stage. Weighted by time decay Weights are assigned to directly estimate the current distribution parameters. This method is mathematically equivalent to the two-step update, but the calculation is more intuitive. When a new sample enters the window, the weights of older samples exceeding the equivalent memory length automatically fade out. The system performs a conjugate normal update based on the current observations and the current prior distribution to obtain the posterior distribution at the current time step. The update formula is as follows:

[0063]

[0064] in, Let be the mean of the posterior distribution at the current time. Let V be the posterior distribution variance at the current time. The mean of the posterior distribution at the previous time step. Let V be the posterior distribution variance of the previous time step. For the current observation value, Let be the observation variance, and λ be the forgetting factor. Observation variance As a fixed parameter participating in iterative updates, its physical meaning is the inherent uncertainty of the emission factor observations. The system determines the second-stage identification result of the current observation based on the two-tailed probability of the posterior distribution of the current observation at the current time. The dynamic threshold interval is determined by the posterior distribution N( , The 2.5 percentile and 97.5 percentile of ) were determined as follows:

[0065] in, This is the lower limit of the dynamic threshold. This is the upper limit of the dynamic threshold. Let be the mean of the posterior distribution at the current time. Let be the posterior distribution standard deviation at the current time. If < (Left tail probability <2.5%), marked as low-value statistical anomaly; if > (Right tail probability < 2.5%) is marked as a high-value statistical anomaly; otherwise, it is statistically normal. Through dual-sided detection, the system can simultaneously monitor abnormally low and abnormally high values ​​of emission factors, avoiding missed detections caused by single-direction detection.

[0066] It should be noted that the initial prior distribution is established as follows: obtain historical emission factor samples with capacity utilization rates higher than a preset threshold (i.e., the dividing point) and the first-stage judgment results satisfying preset credibility conditions; establish a normal prior distribution by taking the natural logarithm of the historical emission factor samples; estimate the observation variance using the same historical emission factor samples; and use the observation variance as a fixed parameter to participate in the conjugate normal update. A normal prior distribution is established based on historical emission factor samples with capacity utilization rates higher than a preset threshold (i.e., the dividing point) and the first-stage judgment results satisfying preset credibility conditions.

[0067] Specifically, before the system starts, it constructs a normal prior distribution by taking the natural logarithm of a sample of reliable emission factors from historical data whose capacity utilization rate is higher than the threshold:

[0068] Where y is the natural logarithm of the emission factor, and EF is the emission factor. The prior mean, This represents the prior variance. The prior mean is also mentioned. and prior variance All values ​​were obtained from logarithmic estimates of historical reliable samples above the boundary point. Observation variance Historical log-variance estimates from the same sample set are used as fixed parameters in iterative updates. During training set construction, 3σMAD (median absolute deviation) cleaning is performed to remove extreme values ​​in the logarithmic space exceeding three times the median absolute deviation, eliminating interference from outliers that may remain in historical data. By establishing the prior only based on samples above the cutoff point, interference from unstable emission characteristics under extremely low capacity utilization is eliminated, improving the representativeness of the prior distribution.

[0069] When the number of historical emission factor samples is less than the preset threshold, a resampling method is used to randomly draw historical emission factor samples with replacement. After repeating this process multiple times, the empirical distribution of the obtained statistics is used to replace the normal prior distribution.

[0070] When the number of historical emission factor samples n < 2000 hours (approximately 3 months) exceeds a preset threshold and the first-stage judgment result meets the preset credibility conditions, the Bootstrap resampling method is used to enhance the robustness of the prior distribution. Specifically, multiple regenerated samples are obtained through resampling with replacement, and the logarithmic mean is calculated for each regenerated sample. The empirical distribution of these logarithmic means is used to estimate the mean and variance of the normal prior. The system randomly selects n samples with replacement from the initial credible sample set, calculates the logarithmic mean, and repeats this process B = 1000 times. The empirical distribution of B logarithmic means is used as the prior. When n ≥ 2000 hours, direct estimation is sufficient. When the number of historical samples is less than a preset threshold, the Bootstrap resampling method is used to establish the mathematical equivalence and robustness advantages of the empirical distribution replacing the normal prior distribution, preventing overconfidence due to insufficient cold-start data.

[0071] In some embodiments, when the first-stage determination result is that a preset confidence condition is met and the second-stage determination is a statistical anomaly, a graded processing is performed based on the duration and degree of deviation of the statistical anomaly, including: For isolated minor anomalies, the original emission factors are retained and logged. For continuous or significant anomalies, alternative emission factors are calculated using the operating condition matching method and marked as requiring attention. For persistent anomalies, temporary alternative emission factors are calculated using the operating condition matching method with an expanded time window, marked as under investigation, and manual investigation is triggered. Isolated minor anomalies are defined as those where the duration of the statistical anomaly is less than or equal to a first duration threshold and the deviation is less than a first deviation threshold. Continuous anomalies are defined as those where the duration of the statistical anomaly is greater than the first duration threshold but less than or equal to a second duration threshold. Significant anomalies are defined as those where the duration of the statistical anomaly is less than or equal to the first duration threshold but the deviation is greater than or equal to the first deviation threshold. Persistent anomalies are defined as those where the duration of the statistical anomaly is greater than the second duration threshold. The first and second duration thresholds are preset values; for example, the first duration threshold is 1 hour and the second duration threshold is 5 hours. The deviation is the percentage deviation of the emission factor at the current moment from the mean of the posterior distribution at the current moment; for example, the first deviation threshold is 20%.

[0072] Specifically, when the quality control score in the first stage is 0 (i.e., the first stage judgment result meets the preset credibility conditions) and the second stage identification result is a low-value statistical anomaly or a high-value statistical anomaly, the system does not simply remove the data. Instead, it constructs a two-dimensional defense matrix based on the duration and degree of deviation of the statistical anomaly and executes a differentiated hierarchical processing strategy. For example: (1) For isolated minor anomalies, it refers to an emission factor anomaly lasting less than 1 hour and the current emission factor If the deviation from the mean of the experimental distribution at the current moment is less than 20%, such an anomaly is most likely a normal operating condition fluctuation. The system retains the original emission factor and directly enters it into the database, only marking the log for subsequent traceability.

[0073] (2) For continuous or major anomalies, which refer to the situation where the emission factor is abnormal for 2 to 5 hours or the deviation is greater than or equal to 20%, the original value of such anomalies is unreliable. The system uses the operating condition matching method to calculate the alternative emission factor. At the same time, the data sample is marked as needing attention and pushed to the enterprise for verification requirements such as the calibration status of the continuous emission monitoring system equipment and the operating parameters of the waste gas treatment facility.

[0074] (3) For continuous abnormalities, which refers to the situation where the emission factor is abnormal for 6 hours or more, the system will upgrade the quality control level to suspicious. First, the operating condition matching method with an expanded time window will be used to calculate the temporary alternative emission factor. The data sample will be marked as under investigation, and the operation status of the continuous emission monitoring system equipment and waste gas treatment facilities will be manually investigated within a limited time.

[0075] The operating condition matching method may include: taking the median of credible emission factor samples within the most recent time window whose load rate difference from the current emission factor does not exceed a preset range, as the alternative emission factor; when the number of credible emission factor samples meeting the conditions is less than the preset number, it degenerates into linear interpolation of adjacent credible emission factor samples. The core idea of ​​the operating condition matching method is to find historical normal data under similar operating conditions to replace the current abnormal data. For continuous or large anomalies, a 24-hour time window is used, and the formula for calculating the alternative emission factor is:

[0076] in, To replace emission factors, For historical reliable emission factor samples, The current sample load rate, This refers to the current time. Specifically, it uses the median emission factor of reliable samples within the most recent 24 hours that differ from the current sample load rate by no more than 5%. If there are fewer than 5 reliable samples in the same load segment, it degenerates into linear interpolation of adjacent reliable samples; if there are no emission factor samples entering the second stage before or after, it degenerates into the median emission factor of all reliable samples in the most recent 24 hours.

[0077] For persistent anomalies, a 200-hour time window is used, with the calculation formula being the same as above, only the time window is replaced with... .

[0078] It's important to understand that prolonged periods of anomalies may indicate systemic problems in the continuous emission monitoring system, such as zero-point drift. Data from the same operating conditions within a 24-hour window could also be contaminated by the same issue. Therefore, interpolation using historical normal data from a larger time window (e.g., 200 hours) is necessary to ensure the reliability of the substitute values. After manual investigation, the system executes different correction paths based on the investigation results: If the investigation reveals a correctable equipment problem, such as a zero-point drift caused by a long-term lack of calibration in the continuous emission monitoring system, the emission factors for the corresponding period are recalculated using the corrected online monitoring data, replacing the temporary substitute emission factors. If the investigation confirms that the online monitoring data cannot reflect the actual emissions, such as a continuous emission monitoring system equipment malfunction, the temporary substitute emission factors are retained, and the label is adjusted to indicate equipment failure. If the investigation confirms that the online monitoring equipment is operating normally, the emission monitoring data is accurate, and the abnormal emission factors are caused by the abnormal operation of the waste gas treatment facilities, the original emission factors are replaced, and the label from the investigation is removed. Through this closed-loop processing logic, the system can not only identify anomalies but also provide corresponding data correction schemes based on the causes of the anomalies, ensuring the continuity and accuracy of the emission inventory data.

[0079] The quality control level can be divided into four levels: Level 1 is reliable, corresponding to the first stage judgment result meeting the preset reliable conditions and the second stage identification result being statistically normal; Level 2 is of concern, corresponding to the first stage quality control quality score being the first preset value, or the second stage identification result being statistically abnormal; Level 3 is doubtful, corresponding to the first stage quality control quality score being within the range of the second and third preset values; Level 4 is unreliable, corresponding to the first stage quality control quality score being not less than the fourth preset value; wherein, the first preset value is less than the second preset value, less than the third preset value, and less than the fourth preset value.

[0080] In one example, the four-level quality control grading table is as follows: Table 2

[0081] Based on the quality control scores of the first stage and the identification results of the second stage, the quality control level of emission factor data is divided into four levels. Level 1 is reliable, corresponding to a quality control score of 0 in the first stage and a statistically normal identification result in the second stage; this level of data can be directly entered into the database for use. Level 2 requires attention, corresponding to a quality control score of 2 in the first stage or a statistically abnormal identification result in the second stage; this level of data is usable but requires logging and a review request. Level 3 is suspicious, corresponding to a quality control score of 3 to 4 in the first stage; this level of data is marked for updating and will be recalculated after a limited time for investigation. Level 4 is unreliable, corresponding to a quality control score of 5 or greater in the first stage; this level of data is unusable and an immediate alarm is triggered. Through this four-level quality control classification, the system can manage data at different quality levels in a differentiated manner, ensuring the direct usability of high-quality data while effectively intercepting and issuing warnings for low-quality data, thus improving the overall data quality of the dynamic update of the emission inventory.

[0082] It should be noted that the determination of the credibility level in Table 2 requires that the first stage QC1=0 and the second stage identification result be statistically normal. When the capacity utilization rate is lower than the preset threshold, the second stage is not executed, and no credibility level is generated. Samples with QC1=0 are conservatively downgraded to the level requiring attention, rather than being directly discarded. This can also be understood as follows: when the capacity utilization rate is lower than the preset threshold, the second stage is not executed, and samples with QC1≥1 are processed according to the level requiring attention.

[0083] In a specific example, a detailed explanation is provided using a particular application scenario and case study in the coal-fired power generation industry. Taking a 330MW coal-fired generating unit as an example, the online monitoring data of its continuous emission monitoring system for a certain hour is as follows: Nitrogen oxides (NOx) inlet concentration at channel A is 385 mg / m³, and at channel B is 410 mg / m³; NOx outlet concentration at channel A is 32 mg / m³, and at channel B is 28 mg / m³; flue gas flow rate Q is 1,200,000 m³ / h; and unit load P is 300MW. The system first calculates the emission factor of the industrial source based on the online monitoring data from the continuous emission monitoring system. Since this unit has a multi-channel configuration, the system takes the average concentration of outlet channels A and B, 30 mg / m³, as the emission factor. The average concentration of inlet channels A and B, 397.5 mg / m³, was taken as... Perform the calculation. The calculation process for the emission factor EF is: EF = (30 × 1, 200, 000) / (300 × ... =0.120 g / kWh. If calculated directly using the outlet concentration of channel A as 32 mg / m³, then EF = (32 × 1, 200, 000) / (300 × Pollution generation coefficient = 0.128 g / kWh. The calculation process is as follows: =(385×1,200,000) / (300× = 1.540 g / kWh.

[0084] The system then proceeds to the first stage, performing a multi-dimensional preliminary screening of the online monitoring data for data quality. The system sequentially performs binary judgments on 10 judgment factors across four levels: parameter completeness, physical consistency, channel consistency, and time-series abrupt changes. In the hourly data, all key parameters are complete and valid; both inlet and outlet concentrations are greater than 0; the unit load is greater than 0; the inlet concentration is greater than the outlet concentration and within the normal reaction temperature window; the unit product flue gas generation is within a reasonable range; the differences between inlet and outlet concentrations across multiple channels are within the set range; and the rate of change of emission factors and pollution production coefficients at adjacent times does not exceed 100%. Therefore, all parameters from D1 to D10 are judged as 0, and the weighted sum yields a quality control score QC1 = 0, meeting the preset reliability conditions.

[0085] Next, the system determines whether the current capacity utilization rate is lower than a preset threshold. The current capacity utilization rate is 300 / 330 = 90.9%, which is not lower than the default preset threshold of 20%. Therefore, the system enters the second stage, performing Bayesian iterative statistical anomaly screening on the emission factors. The system obtains the natural logarithmic value of the emission factors at the current moment. =ln(0.128)=-2.056 as the current observation value. Before system startup, an initial log-normal prior distribution N(-2.049, 0.092) was established based on historical reliable samples, with observation variance... =0.092, forgetting factor λ=0.999875. The system applies a forgetting discount to the accuracy of the historical posterior distribution at the previous time step and performs a conjugate normal update with the current observation. The updated posterior distribution mean at the current time step is... =-2.050, variance =0.046. Based on this posterior distribution, the system calculates the dynamic threshold interval: lower limit =exp(-2.050-2×0.215)=0.084, upper limit =exp(-2.050+2×0.215)=0.198. Since the current emission factor EF=0.128 is within the range [0.084, 0.198], the system determines that its two-tailed probability is normal, and the identification result is statistically normal. Finally, based on the first-stage judgment results and the second-stage identification results, the system determines that the quality control level of this emission factor is Level 1 reliable, and the emission factor of 0.128 g / kWh and the pollution generation coefficient of 1.540 g / kWh can be directly put into storage for use.

[0086] To further verify the robustness and universality of the method in this application, six coal-fired power generating units (capacity ranging from 335MW to 1038MW) were used as industrial source cases for statistical verification, totaling 45,883 hours of sample. In the initial screening of data quality across 10 dimensions in the first phase, the number of triggering samples and the trigger rate for each dimension are as follows: D1 Missing parameter triggers: 2172, trigger rate: 4.7%; D2 Zero concentration triggers: 626, trigger rate: 1.4%; D3 Zero denominator triggers: 6180, trigger rate: 13.5%; D4 Equal input and output triggers: 0, trigger rate: 0.00%; D5 Outlet exceeding inlet triggers: 7, trigger rate: 0.02%; D6 Matching flue gas generation triggers: 11377, trigger rate: 24.8%; D7 Multi-channel inlet difference triggers: 1192, trigger rate: 2.6%; D8 Multi-channel outlet difference triggers: 1192, trigger rate: 2.6%; D9 Emission factor jump triggers: 174, trigger rate: 3.8%; D10 Pollution generation coefficient jump triggers: 781, trigger rate: 1.7%. Data shows that the trigger rate for matching flue gas generation in D6 was the highest, reaching 24.8%, indicating that the flue gas generation per unit product exceeding the reasonable range is the most common emission characteristic parameter data quality problem in the coal-fired power plants in this case. This application effectively intercepted data anomalies caused by flue gas flow measurement deviations by setting this dimension.

[0087] In the second stage of Bayesian iterative statistical anomaly screening, the system used a capacity utilization rate of 20% as the cutoff point for validation. Of the 45,883 samples, 13,403 were below the cutoff point; these samples were directly output based on the results of the first stage and did not proceed to the second stage. Of the 32,074 samples above the cutoff point, 31,162 were deemed credible in the first stage, and 912 statistical anomalies were detected in the second stage, resulting in an anomaly rate of 2.8%. The data indicates that the 13,403 samples with a capacity utilization rate below 20% are unsuitable for statistical screening due to unstable emission characteristics; these samples were directly classified based on the results of the first stage. The anomaly rate of 2.8% among the credible samples above the cutoff point demonstrates that the cutoff point control mechanism can effectively distinguish between steady-state and non-steady-state operating conditions, reducing the interference of non-steady-state data on statistical anomaly detection. Based on the combined results of the two-stage assessment, the final distribution of the four quality control levels is as follows: 70.1% reliable, 21.3% requiring attention, 1.6% questionable, and 7.3% unreliable. The "requiring attention" level includes samples with a quality control score of 2 in the first stage or those identified as statistically abnormal in the second stage; the "questionable" level includes samples with a quality control score of 3 to 4 in the first stage or those whose statistical abnormalities persisted for more than 6 hours. The percentages of samples at each level reflect the distribution of the tiered results of the two-stage hierarchical progressive verification mechanism adopted in this application.

[0088] Corresponding to the above method, this application also provides an emission factor quality control and verification device for industrial source air pollutants, such as... Figure 2 As shown, the device includes: The calculation unit 210 is used to calculate the emission factor of industrial sources based on online monitoring data and production load information from the continuous emission monitoring system for air pollutants; the online monitoring data includes basic emission characteristic parameters used to calculate the emission factor. The initial screening unit 220 is used to perform multi-dimensional data quality initial screening on the basic emission characteristic parameters in the first stage to obtain the first stage judgment result; The determining unit 230 is used to determine the quality control level of the emission factor based on the first-stage determination result if the current capacity utilization rate is lower than a preset threshold; the current capacity utilization rate is determined according to the production load information. The fine screening unit 240 is used to enter the second stage if the current capacity utilization rate is not lower than a preset threshold and the first stage judgment result meets the preset credibility condition: perform Bayesian iterative statistical anomaly fine screening on the emission factors that enter the second stage to obtain the second stage identification result; The determining unit 230 is further configured to determine the quality control level of the emission factor based on the first stage determination result and the second stage identification result, and to perform grading processing on emission factors with different quality control levels.

[0089] The functions of each functional unit of the emission factor quality control and verification device for industrial source air pollutants provided in the above embodiments of this application can be realized through the above methods and steps. Therefore, the specific working process and beneficial effects of each unit in the emission factor quality control and verification device for industrial source air pollutants provided in the embodiments of this application will not be repeated here.

[0090] This application also provides an electronic device, such as... Figure 3 As shown, it includes a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340.

[0091] Memory 330 is used to store computer programs; When the processor 310 executes the program stored in the memory 330, it performs the following steps: Based on online monitoring data and production load information from the continuous emission monitoring system for air pollutants, the emission factors of industrial sources are calculated; the online monitoring data includes basic emission characteristic parameters used to calculate the emission factors. In the first stage, a multi-dimensional data quality screening is performed on the basic emission characteristic parameters to obtain the first stage judgment result; If the current capacity utilization rate is lower than the preset threshold, the quality control level of the emission factor is determined based on the first-stage judgment result; the current capacity utilization rate is determined according to the production load information. If the current capacity utilization rate is not lower than the preset threshold and the first stage judgment result meets the preset credibility condition, then proceed to the second stage: perform Bayesian iterative statistical anomaly screening on the emission factors that have entered the second stage to obtain the second stage identification result, and determine the quality control level of the emission factors based on the first stage judgment result and the second stage identification result, and perform graded processing on the emission factors with different quality control levels.

[0092] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0093] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0094] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0095] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0096] The implementation methods and beneficial effects of the various components of the electronic device in the above embodiments for solving the problem can be found in [reference needed]. Figure 1 The steps in the illustrated embodiments are used to implement the electronic device. Therefore, the specific working process and beneficial effects of the electronic device provided in this application will not be repeated here.

[0097] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform the emission factor quality control verification method for industrial source air pollutants described in any of the above embodiments.

[0098] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute the emission factor quality control verification method for industrial source air pollutants described in any of the above embodiments.

[0099] Those skilled in the art will understand that the embodiments in this application can be provided as methods, systems, or computer program products. Therefore, the embodiments in this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments in this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] This application describes embodiments of methods, apparatus (systems), and computer program products according to embodiments of this application with reference to flowchart illustrations and / or block diagrams. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0101] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0102] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0103] Although preferred embodiments have been described in this application, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of this application.

[0104] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims in this application and their equivalents, then this application also intends to include these modifications and variations.

Claims

1. A method for quality control and verification of emission factors of industrial source air pollutants, characterized in that, The method includes: Based on online monitoring data and production load information from the continuous emission monitoring system for air pollutants, the emission factors of industrial sources are calculated; the online monitoring data includes basic emission characteristic parameters used to calculate the emission factors. In the first stage, a multi-dimensional data quality screening is performed on the basic emission characteristic parameters to obtain the first stage judgment result; If the current capacity utilization rate is lower than the preset threshold, the quality control level of the emission factor is determined based on the first-stage judgment result; the current capacity utilization rate is determined according to the production load information. If the current capacity utilization rate is not lower than the preset threshold and the first stage judgment result meets the preset credibility condition, then proceed to the second stage: perform Bayesian iterative statistical anomaly screening on the emission factors that have entered the second stage to obtain the second stage identification result, and determine the quality control level of the emission factors based on the first stage judgment result and the second stage identification result. Different emission factors at different quality control levels are subject to tiered treatment.

2. The method as described in claim 1, characterized in that, In the first stage, a multi-dimensional data quality screening is performed on the basic emission characteristic parameters to obtain the first stage judgment results, including: The basic emission characteristic parameters are evaluated using binary judgments on the judgment factors in multiple dimensions to obtain judgment values ​​for each dimension. Based on the preset weights of each dimension, the judgment values ​​are weighted and summed to obtain the quality control score. The quality control score will be used as the first-stage judgment result.

3. The method as described in claim 2, characterized in that, The multiple dimensions include four aspects: parameter integrity, physical consistency, channel consistency, and time series mutation; among which: The parameter integrity level includes criteria for missing parameters, zero concentration, and zero denominator. The physical consistency level includes the determination factors of equal inflow and outflow, outflow exceeding inflow, and matching flue gas generation. The channel consistency level includes determination factors for multi-channel inbound differences and multi-channel outbound differences; The time series mutation level includes the determination factors for changes in emission factors and changes in pollution production coefficients.

4. The method as described in claim 1, characterized in that, Bayesian iterative statistical anomaly screening was performed on the emission factors entering the second stage to obtain the second-stage identification results, including: Obtain the natural logarithm of the emission factor at the current moment as the current observation value; Apply a forgetting discount to the accuracy of the historical posterior distribution of the previous time step, and use the distribution with the forgetting discount as the current prior distribution; Perform a conjugate normal update based on the current observation and the current prior distribution to obtain the posterior distribution at the current time. Based on the two-tailed probability of the posterior distribution of the current observation at the current time, the second-stage identification result of the current observation is determined. The initial prior distribution of the Bayesian iterative statistical anomaly screening is pre-established based on historical reliable emission factor samples.

5. The method as described in claim 4, characterized in that, The forgetting discount controls the equivalent memory length through a preset forgetting factor, so that the weight of historical information exceeding the equivalent memory length is reduced. The posterior distribution at the current moment is used as the prior distribution at the next moment to form an iterative closed loop; The second-stage identification results include low-value statistical anomalies, high-value statistical anomalies, and statistically normal values. The low-value statistical anomalies are when the emission factor is lower than the lower limit of the dynamic threshold interval, and the high-value statistical anomalies are when the emission factor is higher than the upper limit of the dynamic threshold interval.

6. The method as described in claim 4, characterized in that, The initial prior distribution is established as follows: Historical emission factor samples with capacity utilization rates higher than the preset threshold and first-stage judgment results meeting preset credibility conditions are obtained. A normal prior distribution is established by taking the natural logarithm of the historical emission factor samples, and the observation variance is estimated by the same historical emission factor samples. The observation variance is used as a fixed parameter in the conjugate normal update.

7. The method as described in claim 6, characterized in that, The method further includes: When the number of historical emission factor samples is less than a preset threshold, a resampling method is used to randomly sample the historical emission factor samples with replacement. After repeating this process multiple times, the empirical distribution of the obtained statistics is used to replace the normal prior distribution.

8. The method as described in claim 1, characterized in that, The method further includes: When the first-stage judgment result meets the preset confidence conditions and the second-stage judgment is a statistical anomaly, the statistical anomaly is classified and processed according to its duration and degree of deviation: For isolated minor anomalies, retain the original emission factors and tagged in the log; For continuous or significant anomalies, the alternative emission factor is calculated using the operating condition matching method and marked as requiring attention. For persistent anomalies, a temporary alternative emission factor is calculated using the operating condition matching method with an extended time window, and the process is marked as under investigation and a manual investigation is triggered. The isolated minor anomaly is defined as a statistical anomaly whose duration is less than or equal to a first duration threshold and whose deviation is less than a first deviation threshold; the continuous anomaly is defined as a statistical anomaly whose duration is greater than the first duration threshold and less than or equal to a second duration threshold; the major anomaly is defined as a statistical anomaly whose duration is less than or equal to the first duration threshold but whose deviation is greater than or equal to the first deviation threshold; and the persistent anomaly is defined as a statistical anomaly whose duration is greater than the second duration threshold.

9. The method as described in claim 2, characterized in that, The quality control levels include four levels: The first level is considered trustworthy, corresponding to the first stage judgment result meeting the preset trustworthy conditions, and the second stage identification result being statistically normal. The second level is "needs attention", which corresponds to the quality control score in the first stage being the first preset value, or the identification result in the second stage being statistically abnormal. The third level is suspicious, corresponding to the quality control score of the first stage being within the range of the second and third preset values; Level 4 is unreliable, corresponding to a quality control score in the first stage that is not less than the fourth preset value; Among them, the first preset value is less than the second preset value, less than the third preset value, and less than the fourth preset value.

10. A quality control and verification device for emission factors of industrial source air pollutants, characterized in that, The device includes: The calculation unit is used to calculate the emission factor of industrial sources based on online monitoring data and production load information from the continuous emission monitoring system for air pollutants; the online monitoring data includes basic emission characteristic parameters used to calculate the emission factor. The initial screening unit is used to perform multi-dimensional data quality initial screening on the basic emission characteristic parameters in the first stage to obtain the first stage judgment result. The determining unit is used to determine the quality control level of the emission factor based on the first-stage determination result if the current capacity utilization rate is lower than a preset threshold; the current capacity utilization rate is determined according to the production load information. The fine screening unit is used to enter the second stage if the current capacity utilization rate is not lower than a preset threshold and the first stage judgment result meets the preset credibility condition: perform Bayesian iterative statistical anomaly fine screening on the emission factors that enter the second stage to obtain the second stage identification result; The determining unit is further configured to determine the quality control level of the emission factor based on the first stage judgment result and the second stage identification result; The processing unit is used to perform graded processing on emission factors at different quality control levels.

Images