Automatic identification method for monitoring data abnormity based on air quality station

Abstract

The invention relates to an automatic identification method based on air quality station monitoring data abnormity, and belongs to the technical field of air quality monitoring. The method comprises the following steps: acquiring equipment working conditions and microenvironment interference data based on a station exclusive dynamic acquisition scheme, and eliminating working condition abnormal or extreme interference data through preprocessing; matching a historical abnormal inducement feature library, and separating three types of inducement feature signals of equipment fault, environmental interference and real pollution to form a standardized recognition feature data set; constructing a site scene automatic adaptation model, starting multiple rounds of automatic reasoning, eliminating cross-incentive cross interference, and outputting a preliminary anomaly recognition result; abnormal types are identified through cross-site cross validation, an abnormal conduction chain early warning mechanism is established, and a feature library and inference rules are updated. Automatic accurate identification and timely early warning of monitoring data abnormity are realized, and identification efficiency and reliability are improved.

Description

Technical Field

[0001] This belongs to the field of air quality monitoring technology, specifically involving an automated identification method based on anomalies in air quality monitoring data from air quality monitoring stations. Background Technology

[0002] Air quality monitoring is a core support for environmental management, public health protection, and pollution prevention and control. Air quality monitoring stations, as key carriers of data collection and analysis, directly determine the scientific nature of environmental decisions and the effectiveness of pollution responses through the accuracy of their monitoring data and the timeliness of anomaly identification. With the increasing demand for environmental monitoring, the industry has gradually begun research and development of anomaly identification technologies based on monitoring data. Through basic operations such as data collection, preprocessing, and feature comparison, these technologies attempt to distinguish between factors causing data anomalies, such as equipment malfunctions, environmental interference, and actual pollution, providing a basis for the formulation of subsequent environmental management measures.

[0003] However, existing technologies for identifying anomalies in air quality monitoring data still have many problems that urgently need to be solved, making it difficult to meet the actual needs of accurate and automated identification in complex environments. First, in the data acquisition stage, existing technologies mostly adopt a uniform acquisition scheme, failing to fully consider the differences in regional types of different stations, the diffusion characteristics of surrounding pollution sources, and historical anomaly patterns. This results in a lack of specificity in the acquisition intervals and the combination of core monitoring indicators. For example, industrial area stations and residential area stations have completely different types of pollution sources, yet the same acquisition logic is used, which can easily lead to missing key pollution characteristic data or invalid data redundancy. At the same time, a dual-linkage adjustment mechanism between the microenvironment and equipment operating conditions has not been established during the acquisition process. It is impossible to dynamically adjust the acquisition strategy according to temporary pollution source activities (such as sudden construction dust) or fluctuations in equipment operating conditions (such as unstable voltage of the sampling module), further reducing the effectiveness of the acquired data.

[0004] Secondly, in the data preprocessing stage, the existing interference correction methods are relatively simple, mostly using generalized correction coefficients without generating specific correction rules based on historical data of the same scenario at the site. When faced with scenarios such as monitoring data deviations caused by building obstruction and numerical fluctuations caused by temporary dust, the correction accuracy is insufficient. Furthermore, there is a lack of mapping and correlation between microenvironment and operating condition characteristics, making it impossible to check the cause of data deviation through the matching degree between the two. As a result, some data with abnormal operating conditions or extreme interference cannot be effectively removed, affecting the accuracy of subsequent anomaly identification.

[0005] Furthermore, in the anomaly cause identification and reasoning stage, existing technologies struggle to effectively separate equipment failure, environmental interference, and actual pollution. On one hand, there is a lack of a systematic historical anomaly cause feature storage architecture; feature data is often scattered and not associated with site scenario classification and anomaly propagation processes, failing to provide comprehensive feature support for cause identification. On the other hand, the reasoning process often employs a single-round comparison mode, failing to dynamically adapt cause priorities to real-time scenarios and effectively eliminating cross-cause interference, resulting in a high false positive rate. Simultaneously, anomaly type identification relies solely on single-site data, without cross-site cross-validation, making it difficult to distinguish between regional pollution and single-site equipment failure, further reducing the reliability of the identification results.

[0006] Finally, in terms of early warning mechanism construction, existing technologies are mostly limited to passive response after an anomaly occurs, lacking forward-looking analysis of the anomaly transmission process, and unable to capture early warning features and trigger early warnings based on the cause transmission path; moreover, there is a lack of dynamic update mechanism for feature library and inference rules after the early warning, which means that subsequent anomaly identification still relies on old data and rules, making it difficult to adapt to the constantly changing monitoring environment.

[0007] In summary, existing technologies for identifying anomalies in air quality monitoring data have significant shortcomings in terms of data collection targeting, preprocessing accuracy, effectiveness of cause identification, and early warning foresight. These shortcomings result in low accuracy and automation in anomaly identification, failing to meet the high-efficiency requirements of current environmental monitoring and management. There is an urgent need for a monitoring data anomaly identification method that can achieve site-specific, process-automated, and result-accurate analysis. Summary of the Invention

[0008] To address the aforementioned problems in the existing technology, this invention provides an automated method for identifying anomalies in air quality monitoring data. The objective of this invention can be achieved through the following technical solutions: S1: Based on a site-specific dynamic data acquisition solution, acquire equipment operating condition data and micro-environment interference data; based on the type of micro-environment interference, call the appropriate preprocessing rules, and automatically filter valid data in combination with the real-time fluctuation trend of equipment operating condition data, remove data with abnormal operating conditions and micro-environment, and generate structured data removal and association files. S2: Based on the preprocessed data, match the historical anomaly cause feature library to establish a mapping relationship between the current data and three types of causes: equipment failure, environmental interference, and actual pollution; through the feature differentiation analysis mechanism, automatically separate the feature signals of the three types of causes, and compare the fluctuation pattern of the feature signals with the anomaly threshold to form a standardized air quality automatic identification feature dataset. S3: Construct an automatic site scenario adaptation model, automatically retrieve the priority weight of the corresponding scenario's triggers based on real-time micro-environment data; through the aforementioned automatically recognizable feature dataset, initiate multiple rounds of automatic reasoning operations, and through the matching degree verification of feature signals and trigger labels, automatically eliminate cross-trigger interference and output preliminary anomaly identification results; S4: Cross-validate the preliminary identification results with historical normal data from adjacent sites to identify the anomaly type; establish an anomaly transmission chain early warning mechanism, automatically trigger corresponding anomaly early warning operations based on the anomaly type, and update the cause feature library and inference rules through verified anomaly cases.

[0009] As a preferred technical solution of the present invention, the site-specific dynamic data collection scheme includes: Based on the site area type, the diffusion characteristics of surrounding pollution sources, and historical anomaly patterns, a customized data collection timeline baseline is established, distinguishing the collection interval logic between normal periods and periods of high anomaly, and simultaneously determining a unique combination of core monitoring indicators; a microenvironment-operating condition dual-linkage adjustment link is built, automatically adjusting the collection interval and supplementing the collection of start-up and shutdown indicators based on the monitored temporary pollution source activities and equipment operating condition fluctuations; a pre-processing collaborative mechanism for data collection and preprocessing is established, and the microenvironment and operating condition information associated with the data are simultaneously marked during the collection process.

[0010] Specifically, the preprocessing rules include: Based on different micro-environment interference types, customized correction coefficients generated from historical data of the same scene at the site are invoked, building shading is adapted to the shading angle-data deviation correlation coefficient, and temporary dust is adapted to the diffusion intensity-data correction coefficient; a micro-environment-operating condition feature mapping relationship is constructed simultaneously, and the corrected data is compared with the mapping relationship. When the deviation exceeds the preset range, data tracing and verification are initiated.

[0011] Specifically, the method for removing data with abnormal operating conditions and abnormal microenvironment is as follows: A site-specific operating condition-microenvironment correlation benchmark is constructed. Based on the equipment operating condition parameters and microenvironment characteristic data during the site's historical normal operation period, a dynamic matching model is established to clarify the reasonable fluctuation range of equipment operating condition parameters and the normal characteristic range of corresponding monitoring data under different microenvironment conditions. An anomaly type hierarchical judgment logic is adopted. For operating condition anomalies, the judgment basis is formed by combining the continuous change trend of equipment operating condition parameters with the deviation of the correlation benchmark. For microenvironment anomalies, a synchronization judgment rule is established between the abrupt change state of the correlation microenvironment characteristics and the abrupt change characteristics of the monitoring data. Data to be removed is marked, and the corresponding real-time equipment operating condition parameters and microenvironment characteristic information are recorded synchronously to form a data removal correlation file.

[0012] Specifically, the historical anomaly cause feature database includes: A three-dimensional architecture of cause-scenario-transmission is adopted, in which the cause dimension is subdivided into equipment failure, environmental interference, and actual pollution; the scenario dimension is associated with site-specific scenario classification; and the transmission dimension adds a feature transmission chain from abnormal precursors to manifest abnormalities, forming a multi-dimensional associated feature network. The feature library stores abnormal feature thresholds and synchronously archives abnormal evolution trajectory fragments and corresponding working condition-microenvironment coupled feature groups. A built-in scenario adaptation update mechanism is used to determine the assigned dimension based on the scenario feature matching degree verification of new abnormal cases, synchronously update the feature transmission chain of the corresponding dimension, and coordinate with the interference correction coefficient in the preprocessing stage for optimization.

[0013] Specifically, the feature differentiation parsing mechanism includes: Based on the historical anomaly cause feature library and site scenario classification results, exclusive prior feature anchor points for the three types of causes are extracted through feature clustering. The generation process synchronously associates the operating condition fluctuation patterns and micro-environment change characteristics under the same historical scenario of the site. Based on the characteristics of the causes, a differentiated analysis strategy is adapted. Equipment fault features are decomposed into time-domain periods and combined with operating condition parameter correlation analysis. Environmental interference features are captured by instantaneous mutations and superimposed with the time-series matching of micro-environmental mutations. Real pollution features are analyzed by trend gradient and associated with regional pollution transmission trends. The analyzed features are compared for correlation, overlapping areas are identified, and the dominant features are determined by the occurrence sequence of causes. The feature differences are resolved by simultaneously combining the current micro-environment parameters with the deviations of the same historical scenario.

[0014] Specifically, the method for forming the standardized automated air quality identification feature dataset is as follows: Based on the three types of cause characteristic signals after separation: equipment failure, environmental interference, and actual pollution, the data dimensions are divided by site scenario classification. Under each dimension, the fluctuation period, intensity, and associated working conditions and microenvironment parameters of the characteristic signals are defined as basic fields. Standardize the fields of each dimension, assign exclusive coding rules based on the cause type, and convert the feature signal intensity into a normalized value based on the historical normal data of the site, so as to unify the data units and the expression form of abnormal features. Each feature data point is assigned a source identifier, and the micro-environment snapshot and equipment condition logs of the collection period are linked. At the same time, a three-level index structure of cause-feature-scenario is established to form a structured, traceable, and standardized feature dataset that is adapted to the reasoning process.

[0015] Specifically, the automatic site scenario adaptation model includes: Real-time microenvironmental parameters are categorized and integrated with historical site scene data to construct a multi-dimensional input matrix for scene recognition. Input data is labeled with attributes according to real-time dynamic data and historical static data. A temporary pollution source-feature association table is established, and a feature scene trigger threshold is preset. Based on real-time feature data exceeding the threshold, the feature weight is automatically adjusted by matching the preset weight adjustment coefficient with the magnitude of the exceedance. The anomaly recognition parameters specific to the site scene are called simultaneously, and the mapping relationship between the feature weight adjustment coefficient and the rule parameter is reversed by combining the recognition results of the parameter application.

[0016] Specifically, the multi-round automatic reasoning operation includes the following methods: The first round of scenario-oriented initial inference retrieves the priority cause inference rule set corresponding to the site scenario, compares the standardized feature signals with the feature thresholds and time-series patterns in the rule set, marks the initial screening of abnormal causes and associates them with traceability information; the second round of confidence-enhanced inference calls the inference model trained based on historical cases to review the feature details of the initially screened causes and update the cause confidence; the third round of scenario-condition matching inference combines real-time condition parameters to verify the adaptability of the causes to the current scenario, and finally outputs the inference results with cause type, confidence and traceability chain.

[0017] Specifically, the automatic elimination of cross-cause interference includes the following methods: A cross-causal feature association benchmark library is constructed. Based on the feature overlap data of historical abnormal case causes, the association threshold and time-series synchronization parameters of different causal features are determined. The feature association degree evaluation mechanism is invoked to evaluate the overlap and time-series matching degree between the current feature to be analyzed and each causal feature in real time, and generate association degree values. Features with association degree values ​​exceeding the threshold are marked as cross-interference items and removed, and the confidence of the retained features is adjusted in a synchronous manner.

[0018] Specifically, the method for identifying the anomaly type is as follows: Based on the spatiotemporal correlation characteristics of historical normal data and historical abnormal cases of adjacent sites, a spatiotemporal correlation verification engine is constructed. The preliminary identification results are input into the spatiotemporal correlation verification engine to compare the spatiotemporal synchronization and gradient consistency of abnormal features with the normal baseline. The spatiotemporal correlation verification engine calls the built-in judgment logic library to combine the abnormal distribution of multiple sites, automatically determine the site's regional pollution, equipment failure or environmental interference anomalies, and output the anomaly type result with verification basis.

[0019] Specifically, the method for establishing the abnormal transmission chain early warning mechanism is as follows: Based on historical transmission cases in the historical anomaly cause feature library, combined with scene information output by the site scene automatic adaptation model, and microenvironment and operating condition data, a cause-specific anomaly transmission chain model is constructed. Based on the cause type, the transmission path is sorted out, and the precursor characteristics and time-series transmission thresholds of each transmission node are identified. The transmission chain dynamic monitoring engine is activated to capture precursor characteristic signals in real time and compare them with the model node characteristics and normal feature segments in the standardized air quality automatic identification feature dataset. When the precursor characteristic exceeds the threshold and conforms to the time-series pattern of the transmission path, an early warning is triggered, and emergency intervention rules are retrieved to form an anomaly transmission chain early warning mechanism.

[0020] The beneficial effects of this invention are as follows: (1) Improve the effectiveness of monitoring data and the accuracy of preprocessing: By using a site-specific dynamic acquisition scheme, the acquisition logic is customized based on the site's regional characteristics and historical patterns. A micro-environment-operating condition dual-linkage adjustment link is built simultaneously to dynamically adapt to the acquisition strategy. In the preprocessing stage, a dedicated correction rule based on the site's historical data of the same scenario is invoked to associate the micro-environment and operating condition feature verification data, effectively eliminating invalid data and providing data support for subsequent anomaly identification. (2) Improve the accuracy and automation level of abnormal cause identification: Based on the historical abnormal cause feature library and feature differentiation analysis mechanism, the effective separation of three types of cause features, namely equipment failure, environmental interference and real pollution, is realized; combined with multi-round automatic reasoning operation and cross-site cross verification, cross-cause interference is eliminated, the abnormal type is automatically determined, the probability of cause misjudgment is reduced, and manual intervention is reduced.

[0021] (3) Enhance the foresight and adaptability of anomaly warning: By constructing an anomaly transmission chain warning mechanism with specific triggers, the characteristics of anomalies can be captured in advance and the warning can be triggered in advance to avoid passive response; at the same time, the historical anomaly trigger feature library and reasoning rules are dynamically updated based on new anomaly cases, so that the method can adapt to the changing monitoring environment and continuously ensure the identification and warning effect. Attached Figure Description

[0022] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0023] Figure 1 This is a flowchart illustrating the automated identification method for anomalies in air quality monitoring data based on the present invention. Figure 2 This is an architecture diagram of an automated identification method for anomalies in air quality station monitoring data according to the present invention. Detailed Implementation

[0024] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0025] Please see Figure 1-2 An automated method for identifying anomalies in air quality monitoring data: S1: Based on a site-specific dynamic data acquisition solution, acquire equipment operating condition data and micro-environment interference data; based on the type of micro-environment interference, call the appropriate preprocessing rules, and automatically filter valid data in combination with the real-time fluctuation trend of equipment operating condition data, remove data with abnormal operating conditions and micro-environment, and generate structured data removal and association files. S2: Based on the preprocessed data, match the historical anomaly cause feature library to establish a mapping relationship between the current data and three types of causes: equipment failure, environmental interference, and actual pollution; through the feature differentiation analysis mechanism, automatically separate the feature signals of the three types of causes, and compare the fluctuation pattern of the feature signals with the anomaly threshold to form a standardized air quality automatic identification feature dataset. S3: Construct an automatic site scenario adaptation model, automatically retrieve the priority weight of the corresponding scenario's triggers based on real-time micro-environment data; through the aforementioned automatically recognizable feature dataset, initiate multiple rounds of automatic reasoning operations, and through the matching degree verification of feature signals and trigger labels, automatically eliminate cross-trigger interference and output preliminary anomaly identification results; S4: Cross-validate the preliminary identification results with historical normal data from adjacent sites to identify the anomaly type; establish an anomaly transmission chain early warning mechanism, automatically trigger corresponding anomaly early warning operations based on the anomaly type, and update the cause feature library and inference rules through verified anomaly cases.

[0026] As a preferred technical solution of the present invention, the site-specific dynamic data collection scheme includes: Based on the site area type, the diffusion characteristics of surrounding pollution sources, and historical anomaly patterns, a customized data collection timeline baseline is established, distinguishing the collection interval logic between normal periods and periods of high anomaly, and simultaneously determining a unique combination of core monitoring indicators; a microenvironment-operating condition dual-linkage adjustment link is built, automatically adjusting the collection interval and supplementing the collection of start-up and shutdown indicators based on the monitored temporary pollution source activities and equipment operating condition fluctuations; a pre-processing collaborative mechanism for data collection and preprocessing is established, and the microenvironment and operating condition information associated with the data are simultaneously marked during the collection process.

[0027] In this embodiment, taking an air quality monitoring station in a chemical industrial park of a certain city as an example, the station-specific dynamic data acquisition scheme described in this invention is implemented. The specific data acquisition process is as follows: First, a customized baseline for data collection time was established. The surrounding area of ​​the monitoring stations contains chemical production enterprises and storage facilities. Historical anomalies were mostly concentrated during peak production hours (8:00 AM to 8:00 PM), with anomalies primarily characterized by VOCs exceeding standards and sudden increases in particulate matter concentration. When customizing the baseline, the collection interval for the normal period (8:00 PM to 8:00 AM the following day) was set as a standard interval, while a more intensive collection logic was applied during the peak anomaly period (8:00 AM to 8:00 PM). Simultaneously, a unique combination of core monitoring indicators was determined. In addition to the standard PM2.5 and sulfur dioxide indicators, additional characteristic indicators related to chemical production, such as VOCs components and hydrogen chloride, were added to ensure accurate capture of regionally specific pollution source signals.

[0028] Secondly, the microenvironment-operating condition dual-linkage adjustment link is activated. When the station detects temporary pollution source activity, such as a sudden dust emission during raw material loading and unloading at a company in the park, and the microenvironment sensor captures a signal of instantaneous increase in particulate matter concentration, the link automatically shortens the sampling interval, adjusting from the normal interval to a more frequent sampling frequency, and simultaneously initiates supplementary sampling of total suspended particulate matter. If the equipment operating conditions fluctuate, such as the sampling pump pressure deviating from the normal range, after the operating condition monitoring module sends a feedback signal, the link immediately adjusts the sampling parameters to reduce the impact of sampling flow fluctuations on the data, and simultaneously starts and stops the collection of humidity compensation indicators to avoid humidity changes interfering with the accuracy of particulate matter monitoring data.

[0029] Finally, a pre-processing coordination mechanism is implemented for data acquisition and preprocessing. During data acquisition, each monitoring data point is simultaneously tagged with related information: at the micro-environment level, real-time wind speed, wind direction, and humidity are recorded (e.g., whether the wind direction during dust storms points towards the loading / unloading point); at the equipment operating condition level, sampling pump pressure and sensor power supply voltage are recorded (e.g., the data collected during periods of pressure fluctuation). This tagged information is stored synchronously with the data. In the subsequent preprocessing stage, when calling the dust-adapted correction coefficient, the correction range can be directly determined by associating it with the wind speed data at that time. When investigating abnormal operating condition data, the source can also be quickly traced through the operating condition records without the need for additional data retrieval, achieving efficient integration of data acquisition and preprocessing.

[0030] Through this implementation process, the collected data not only matches the regional pollution characteristics of the chemical industrial park site, but also dynamically adapts to temporary disturbances and changes in equipment operating conditions, while providing accurate correlation data support for subsequent pretreatment.

[0031] Specifically, the preprocessing rules include: Based on different micro-environment interference types, customized correction coefficients generated from historical data of the same scene at the site are invoked, building shading is adapted to the shading angle-data deviation correlation coefficient, and temporary dust is adapted to the diffusion intensity-data correction coefficient; a micro-environment-operating condition feature mapping relationship is constructed simultaneously, and the corrected data is compared with the mapping relationship. When the deviation exceeds the preset range, data tracing and verification are initiated.

[0032] Specifically, the method for removing data with abnormal operating conditions and abnormal microenvironment is as follows: A site-specific operating condition-microenvironment correlation benchmark is constructed. Based on the equipment operating condition parameters and microenvironment characteristic data during the site's historical normal operation period, a dynamic matching model is established to clarify the reasonable fluctuation range of equipment operating condition parameters and the normal characteristic range of corresponding monitoring data under different microenvironment conditions. An anomaly type hierarchical judgment logic is adopted. For operating condition anomalies, the judgment basis is formed by combining the continuous change trend of equipment operating condition parameters with the deviation of the correlation benchmark. For microenvironment anomalies, a synchronization judgment rule is established between the abrupt change state of the correlation microenvironment characteristics and the abrupt change characteristics of the monitoring data. Data to be removed is marked, and the corresponding real-time equipment operating condition parameters and microenvironment characteristic information are recorded synchronously to form a data removal correlation file.

[0033] Specifically, the historical anomaly cause feature database includes: A three-dimensional architecture of cause-scenario-transmission is adopted, in which the cause dimension is subdivided into equipment failure, environmental interference, and actual pollution; the scenario dimension is associated with site-specific scenario classification; and the transmission dimension adds a feature transmission chain from abnormal precursors to manifest abnormalities, forming a multi-dimensional associated feature network. The feature library stores abnormal feature thresholds and synchronously archives abnormal evolution trajectory fragments and corresponding working condition-microenvironment coupled feature groups. A built-in scenario adaptation update mechanism is used to determine the assigned dimension based on the scenario feature matching degree verification of new abnormal cases, synchronously update the feature transmission chain of the corresponding dimension, and coordinate with the interference correction coefficient in the preprocessing stage for optimization.

[0034] Specifically, the feature differentiation parsing mechanism includes: Based on the historical anomaly cause feature library and site scenario classification results, exclusive prior feature anchor points for the three types of causes are extracted through feature clustering. The generation process synchronously associates the operating condition fluctuation patterns and micro-environment change characteristics under the same historical scenario of the site. Based on the characteristics of the causes, a differentiated analysis strategy is adapted. Equipment fault features are decomposed into time-domain periods and combined with operating condition parameter correlation analysis. Environmental interference features are captured by instantaneous mutations and superimposed with the time-series matching of micro-environmental mutations. Real pollution features are analyzed by trend gradient and associated with regional pollution transmission trends. The analyzed features are compared for correlation, overlapping areas are identified, and the dominant features are determined by the occurrence sequence of causes. The feature differences are resolved by simultaneously combining the current micro-environment parameters with the deviations of the same historical scenario.

[0035] Specifically, the method for forming the standardized automated air quality identification feature dataset is as follows: Based on the three types of cause characteristic signals after separation: equipment failure, environmental interference, and actual pollution, the data dimensions are divided by site scenario classification. Under each dimension, the fluctuation period, intensity, and associated working conditions and microenvironment parameters of the characteristic signals are defined as basic fields. Standardize the fields of each dimension, assign exclusive coding rules based on the cause type, and convert the feature signal intensity into a normalized value based on the historical normal data of the site, so as to unify the data units and the expression form of abnormal features. Each feature data point is assigned a source identifier, and the micro-environment snapshot and equipment condition logs of the collection period are linked. At the same time, a three-level index structure of cause-feature-scenario is established to form a structured, traceable, and standardized feature dataset that is adapted to the reasoning process.

[0036] In this embodiment, a mixed-type air quality monitoring station (referred to as station T) in a certain urban traffic and residential area is used as a scenario to demonstrate how it can transform the three types of causal feature signals separated from the original monitoring data into a structured and standardized feature dataset to serve the subsequent automatic reasoning process.

[0037] First, the system categorizes data dimensions based on the "scenario classification" results of station T (e.g., "morning rush hour traffic pollution scenario" and "static and stable residential emissions scenario"). Within each scenario dimension, a set of basic descriptive fields is defined for the three types of causal characteristic signals identified: equipment failure, environmental interference, and actual pollution. Taking the actual pollution characteristic signal under the "morning rush hour traffic pollution scenario" as an example, its basic fields are defined as: fluctuation period (T_p, such as a 2-hour cycle synchronized with traffic flow), intensity (A_p, such as the peak PM2.5 concentration in micrograms per cubic meter), associated operating condition parameters (e.g., sampling pump flow rate Q_s, sensor temperature T_s), and associated microenvironmental parameters (e.g., traffic flow rate Traffic_f, wind speed WS).

[0038] For equipment fault characteristic signals in the same scenario, the following fields are defined: fluctuation period (T_f, such as sensor drift manifested as a linear change lasting for several hours), intensity (A_f, such as the magnitude of signal voltage deviation from the reference value), associated operating condition parameters (such as power supply voltage V, internal temperature T_in), and associated micro-environment parameters (at this time there may be no direct association).

[0039] Next, the system standardizes all fields across all dimensions. The core of this standardization is to unify the units of measurement and the format of expression, thereby eliminating the site-specificity and differences in physical meaning of the original data.

[0040] Dedicated coding rules: Assign a prefix code to each cause type, such as "EF" for equipment failure, "EI" for environmental interference, and "RP" for actual pollution.

[0041] Numerical normalization: Numerical fields such as characteristic signal intensity are converted into normalized values ​​calibrated based on historical normal data of station T. For example, the peak PM2.5 concentration A_p is divided by the standard deviation of PM2.5 concentration σ_p in the historical normal data of the station during the morning rush hour to obtain the normalized intensity value A'_p = A_p / σ_p. Similarly, the voltage deviation amplitude A_f of equipment failure is also normalized by dividing it by its historical normal fluctuation range. This allows characteristic intensities with different physical meanings to be compared on the same scale.

[0042] Finally, to ensure the reliability and traceability of the dataset and to provide direct support for automated reasoning, the system adds rich metadata to each feature data.

[0043] Add a source identification identifier: Each standardized feature data (such as a normalized pollution intensity A'_p) is associated with a unique source identification identifier. This identifier is linked to a snapshot of the microenvironment during its original collection period (such as a set of instantaneous values ​​of wind speed, wind direction, temperature, humidity, traffic flow, etc. at that moment) and equipment operating logs (such as real-time operating records of sampling pumps such as flow rate, voltage, temperature, etc.).

[0044] A three-level index is established: all this information is organized in a three-level index structure of "cause-feature-scenario". For example, the path of a piece of data in the index can be: actual pollution (RP) -> intensity (A'_p) -> morning rush hour traffic pollution scenario. Through this index, the subsequent automatic inference engine can quickly retrieve and compare standardized features of the same cause type in the same scenario, or trace the original environment and equipment status when any feature was generated.

[0045] Through the above process, the raw, heterogeneous monitoring data generated by site T is transformed into a standardized feature dataset with a clear structure, unified dimensions, and traceability. This dataset is compatible with subsequent multi-round automated reasoning processes, laying a solid data foundation for automatically identifying the causes of data anomalies.

[0046] Specifically, the automatic site scenario adaptation model includes: Real-time microenvironmental parameters are categorized and integrated with historical site scene data to construct a multi-dimensional input matrix for scene recognition. Input data is labeled with attributes according to real-time dynamic data and historical static data. A temporary pollution source-feature association table is established, and a feature scene trigger threshold is preset. Based on real-time feature data exceeding the threshold, the feature weight is automatically adjusted by matching the preset weight adjustment coefficient with the magnitude of the exceedance. The anomaly recognition parameters specific to the site scene are called simultaneously, and the mapping relationship between the feature weight adjustment coefficient and the rule parameter is reversed by combining the recognition results of the parameter application.

[0047] Specifically, the multi-round automatic reasoning operation includes the following methods: The first round of scenario-oriented initial inference retrieves the priority cause inference rule set corresponding to the site scenario, compares the standardized feature signals with the feature thresholds and time-series patterns in the rule set, marks the initial screening of abnormal causes and associates them with traceability information; the second round of confidence-enhanced inference calls the inference model trained based on historical cases to review the feature details of the initially screened causes and update the cause confidence; the third round of scenario-condition matching inference combines real-time condition parameters to verify the adaptability of the causes to the current scenario, and finally outputs the inference results with cause type, confidence and traceability chain.

[0048] Specifically, the automatic elimination of cross-cause interference includes the following methods: A cross-causal feature association benchmark library is constructed. Based on the feature overlap data of historical abnormal case causes, the association threshold and time-series synchronization parameters of different causal features are determined. The feature association degree evaluation mechanism is invoked to evaluate the overlap and time-series matching degree between the current feature to be analyzed and each causal feature in real time, and generate association degree values. Features with association degree values ​​exceeding the threshold are marked as cross-interference items and removed, and the confidence of the retained features is adjusted in a synchronous manner.

[0049] Specifically, the method for identifying the anomaly type is as follows: Based on the spatiotemporal correlation characteristics of historical normal data and historical abnormal cases of adjacent sites, a spatiotemporal correlation verification engine is constructed. The preliminary identification results are input into the spatiotemporal correlation verification engine to compare the spatiotemporal synchronization and gradient consistency of abnormal features with the normal baseline. The spatiotemporal correlation verification engine calls the built-in judgment logic library to combine the abnormal distribution of multiple sites, automatically determine the site's regional pollution, equipment failure or environmental interference anomalies, and output the anomaly type result with verification basis.

[0050] In this embodiment, three adjacent air quality monitoring stations in a city's urban area are used as the application scenario. Station A is located in the city center's commercial district, Station B is located in a residential area 1 kilometer east of Station A, and Station C is located in a science and education area 1.5 kilometers west of Station A. The three stations form a triangularly distributed adjacent monitoring network. The specific implementation process is as follows: First, a spatiotemporal correlation verification engine was constructed: historical normal data from the three stations over the past year was collected (excluding data from marked abnormal periods), and historical abnormal cases from the past three years were compiled (including monitoring data anomalies caused by sampling module failure at station A, environmental disturbances caused by temporary construction dust around station B, and regional pollution with PM2.5 exceeding the standard at all three stations simultaneously). Spatiotemporal correlation features were extracted—for example, in the case of regional pollution, the anomaly spread from the upstream northwest station to the southeast at a propagation speed of v, and the concentration gradient showed a pattern of "high in the northwest and low in the southeast"; then, the microenvironmental disturbance data (wind speed v, wind direction, humidity H) from the three stations were combined with the equipment data. The correlation characteristics of operating condition data (sampling pump pressure, sensor power supply voltage), for example, when the equipment at station A fails, the sampling pump pressure often deviates from the operating condition-microenvironment correlation benchmark of "pressure maintained at PMPa when microenvironment humidity is H", and there is no obvious microenvironmental change at that time; based on these data, a spatiotemporal correlation verification engine is built, and a dedicated spatiotemporal judgment logic library for three types of causes is built into the engine to clarify the judgment rules for equipment failure corresponding to "single station anomaly + operating condition parameters deviating from the benchmark + no synchronization signal from adjacent stations", regional pollution corresponding to "multi-station anomaly + spatiotemporal gradient conforming to regional transmission trend", and environmental interference corresponding to "single station anomaly + microenvironmental change".

[0051] At 8:00 AM on a certain workday, Station A detected a sudden increase in PM2.5 concentration, initially identified as "suspected pollution anomaly." After inputting this result into a spatiotemporal correlation verification engine, the engine automatically extracted the anomaly characteristics: the occurrence period was 8:00-10:00 AM, the affected area was limited to Station A, and the associated microenvironmental data was "northeast wind direction at 8:00 AM, wind speed v m / s, no sudden changes in dust or gusts," and the operating condition data was "sampling pump pressure dropped from P1MPa to P2MPa." The engine then compared these data with the normal baselines of Stations B and C, finding that the PM2.5 concentrations at Stations B and C remained within the normal range during the same period, with no synchronous anomaly signals. The spatiotemporal synchronicity and gradient consistency of the anomaly characteristics at Station A were both below the threshold, and the deviation data segment from 8:00-10:00 AM at Station A was synchronously marked.

[0052] The engine further calls the judgment logic library for matching: since only station A has an anomaly, and the sampling pump pressure deviates from the operating condition-microenvironment correlation benchmark, and the adjacent stations B and C have no synchronous anomalies, it meets the judgment rules for equipment failure. Finally, the anomaly type is determined to be "equipment failure", and the result is output with a complete judgment logic chain (single station anomaly + operating condition parameter deviation from benchmark + no synchronous signal from adjacent stations → equipment failure) and related data (snapshot of station A's microenvironment and sampling pump pressure change log from 8-10 am).

[0053] Specifically, the method for establishing the abnormal transmission chain early warning mechanism is as follows: Based on historical transmission cases in the historical anomaly cause feature library, combined with scene information output by the site scene automatic adaptation model, and microenvironment and operating condition data, a cause-specific anomaly transmission chain model is constructed. Based on the cause type, the transmission path is sorted out, and the precursor characteristics and time-series transmission thresholds of each transmission node are identified. The transmission chain dynamic monitoring engine is activated to capture precursor characteristic signals in real time and compare them with the model node characteristics and normal feature segments in the standardized air quality automatic identification feature dataset. When the precursor characteristic exceeds the threshold and conforms to the time-series pattern of the transmission path, an early warning is triggered, and emergency intervention rules are retrieved to form an anomaly transmission chain early warning mechanism.

[0054] In this embodiment, site S is located in an industrial area surrounded by chemical plants, oil storage facilities, and major transportation routes. Historical data indicates that this area is prone to complex pollution events caused by the superposition of multiple pollution sources. Such pollution typically exhibits a clear transmission pattern: it is initially triggered by specific meteorological conditions (such as calm winds and temperature inversions) to produce precursory characteristics, followed by a gradual increase in pollutant concentrations, eventually leading to overt anomalies. To provide early warning, it is necessary to construct a specific anomaly transmission chain model and initiate dynamic monitoring.

[0055] First, historical transmission cases related to site S were extracted from the historical anomaly precipitator feature database. Through pattern analysis of these cases, a specific anomaly transmission chain model for precipitator P (compound pollution) was constructed. This model divides the transmission path into three key nodes: Initial node N_i: corresponds to the pollution incubation stage. Precursor features F_i include specific microenvironmental combinations, such as wind direction WD falling within the range [WD_min, WD_max] (e.g., downstream of the prevailing wind direction), wind speed WS below the threshold WS_th (calm wind conditions), and atmospheric stability index AS exceeding the threshold AS_th (temperature inversion). Simultaneously, operating parameters must be within normal ranges, such as sampling pump pressure P_p within [P_min, P_max], and sensor voltage V_s stable.

[0056] Intermediate node N_m: corresponding to the pollution accumulation stage. The precursor characteristic F_m is that the concentration of characteristic pollutants (such as PM2.5 concentration C_pm25 and VOCs concentration C_vocs) begins to show a trend of increase, and its change gradient ΔC / Δt exceeds the threshold G_th, and the correlation with microenvironmental parameters (such as humidity H and temperature T) is enhanced.

[0057] Explicit node N_e: corresponds to the pollution outbreak stage, characterized by pollutant concentrations continuously exceeding the standard, such as C_pm25> C_max, C_vocs> C_vocs_max, accompanied by multiple indicators showing synchronous anomalies.

[0058] The model clarifies the timing conduction thresholds between nodes: the time window from N_i to N_m is Δt1, and the time window from N_m to N_e is Δt2. For example, Δt1 = 2 hours and Δt2 = 1 hour, indicating that the entire conduction process needs to be completed within 3 hours; otherwise, it is regarded as a false alarm.

[0059] Secondly, the conduction chain dynamic monitoring engine accesses in real time the scenario information (such as the scenario type being "industrial area calm wind pollution"), microenvironment data (WS, WD, H, AS, etc.) and working condition data (P_p, V_s, etc.) output by the site scenario automatic adaptation model, and compares them with the normal feature segments in the standardized air quality automatic recognition feature dataset. Real-time data capture: The engine continuously monitors the data stream. When it detects that the microenvironment parameters meet F_i (such as WD ∈ [WD_min, WD_max] and WS < WS_th), it immediately marks the initial node N_i as activated and starts the timer T1. Node feature comparison: Within the time when T1 < Δt1, the engine locks the pollutant concentration signals (C_pm25, C_vocs) associated with N_m and calculates their gradient changes. If ΔC_pm25 / Δt > G_th and ΔC_vocs / Δt > G_vocs_th, it determines that the intermediate node N_m is activated and starts the timer T2. Conduction rule verification: Within the time when T2 < Δt2, the engine further verifies whether the data conforms to the timing rules of the conduction path, such as whether the increase of C_pm25 and C_vocs is synchronous and has timing consistency with the microenvironment mutation (such as the continuous increase of AS). At the same time, the engine calls the normal feature dataset to exclude the interference of daily fluctuations (such as morning peak traffic pollution).

[0060] When the dynamic monitoring engine of the transmission chain determines that the precursor features (such as F_i and F_m) exceed the preset threshold, and the entire abnormal transmission sequence is fully presented within the time window (Δt1+Δt2) defined by the model, the system will immediately trigger a forward-looking tiered early warning. This process is not a simple alarm, but the starting point for initiating an automated, multi-level emergency response chain: the system first automatically retrieves and executes a preset set of emergency intervention rules. The core measures include: First, pushing structured early warning messages to the environmental monitoring platform. The content not only covers the type of trigger (such as "compound pollution"), but also includes in detail the node status of the current transmission path, the expected impact period calculated based on the time series model, the list of key pollutants and their evolution trend, providing in-depth intelligence for decision-making; Second, implementing dynamic enhancement of the data collection strategy, automatically increasing the data collection frequency of the core site S and geographically associated adjacent sites from The system employs several key mechanisms: First, it upgrades the interval between pollutants to a denser interval and intelligently initiates supplementary sampling for specific secondary pollutants (such as NOx and O3) to capture a more comprehensive pollution profile. Second, it activates a regional collaborative verification mechanism by cross-validating real-time and historical data from adjacent stations to determine whether an anomaly is a localized phenomenon or a regional problem, thus accurately defining the pollution range and transmission trend. Third, it executes precise preliminary intervention commands, directly triggering local response measures based on a rule base. Examples include sending emission reduction recommendations to key polluting enterprises within the warning area or automatically linking with the municipal system to initiate road watering and dust suppression operations in designated areas. After the warning is processed, the system archives the entire process data of this transmission case (including micro-environment snapshots, operating logs, and characteristic signal sequences) into the historical anomaly cause feature library. Through scene feature matching degree verification, the transmission chain model of the cause P is updated, such as optimizing the threshold of the precursor feature F_i or adjusting the time series window Δt1. Simultaneously, the site scene automatic adaptation model reverse-calibrates the feature weight adjustment coefficients to improve the accuracy of future warnings. The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An automated identification method based on anomalies in air quality monitoring data, characterized in that, include: S1: Based on a site-specific dynamic data acquisition solution, acquire equipment operating condition data and micro-environment interference data; Based on the type of micro-environmental interference, the preprocessing rules are adapted and automatically filtered by combining the real-time fluctuation trend of equipment operating data. Data with abnormal operating conditions and micro-environmental abnormalities are removed, and structured data removal and association files are generated. S2: Based on the preprocessed data, match the historical anomaly cause feature library to establish a mapping relationship between the current data and three types of causes: equipment failure, environmental interference, and actual pollution; Through a feature differentiation analysis mechanism, the feature signals of the three types of causes are automatically separated. Based on the fluctuation pattern of the feature signals and the abnormal threshold, a standardized automatic air quality identification feature dataset is formed. S3: Construct an automatic site scenario adaptation model, automatically retrieve the priority weight of the corresponding scenario's triggers based on real-time micro-environment data; through the aforementioned automatically recognizable feature dataset, initiate multiple rounds of automatic reasoning operations, and through the matching degree verification of feature signals and trigger labels, automatically eliminate cross-trigger interference and output preliminary anomaly identification results; S4: Cross-validate the preliminary identification results with historical normal data from adjacent sites to identify the anomaly type; establish an anomaly transmission chain early warning mechanism, automatically trigger corresponding anomaly early warning operations based on the anomaly type, and update the cause feature library and inference rules through verified anomaly cases.

2. The method according to claim 1, characterized in that, The site-specific dynamic data collection solution includes: Based on the site area type, the diffusion characteristics of surrounding pollution sources, and historical anomaly patterns, a customized data collection timeline baseline is established, distinguishing the collection interval logic between normal periods and periods of high anomaly, and simultaneously determining a unique combination of core monitoring indicators; a microenvironment-operating condition dual-linkage adjustment link is built, automatically adjusting the collection interval and supplementing the collection of start-up and shutdown indicators based on the monitored temporary pollution source activities and equipment operating condition fluctuations; a pre-processing collaborative mechanism for data collection and preprocessing is established, and the microenvironment and operating condition information associated with the data are simultaneously marked during the collection process.

3. The method according to claim 1, characterized in that, The preprocessing rules include: Based on different micro-environment interference types, customized correction coefficients generated from historical data of the same scene at the site are invoked, building shading is adapted to the shading angle-data deviation correlation coefficient, and temporary dust is adapted to the diffusion intensity-data correction coefficient; a micro-environment-operating condition feature mapping relationship is constructed simultaneously, and the corrected data is compared with the mapping relationship. When the deviation exceeds the preset range, data tracing and verification are initiated.

4. The method according to claim 1, characterized in that, The specific method for removing data with abnormal operating conditions and microenvironment is as follows: A site-specific operating condition-microenvironment correlation benchmark is constructed. Based on the equipment operating condition parameters and microenvironment characteristic data during the site's historical normal operation period, a dynamic matching model is established to clarify the reasonable fluctuation range of equipment operating condition parameters and the normal characteristic range of corresponding monitoring data under different microenvironment conditions. An anomaly type hierarchical judgment logic is adopted. For operating condition anomalies, the judgment basis is formed by combining the continuous change trend of equipment operating condition parameters with the deviation of the correlation benchmark. For microenvironment anomalies, a synchronization judgment rule is established between the abrupt change state of the correlation microenvironment characteristics and the abrupt change characteristics of the monitoring data. Data to be removed is marked, and the corresponding real-time equipment operating condition parameters and microenvironment characteristic information are recorded synchronously to form a data removal correlation file.

5. The method according to claim 1, characterized in that, The historical anomaly cause feature database includes: A three-dimensional architecture of cause-scenario-transmission is adopted, in which the cause dimension is subdivided into equipment failure, environmental interference, and actual pollution; the scenario dimension is associated with site-specific scenario classification; and the transmission dimension adds a feature transmission chain from abnormal precursors to manifest abnormalities, forming a multi-dimensional associated feature network. The feature library stores abnormal feature thresholds and synchronously archives abnormal evolution trajectory fragments and corresponding working condition-microenvironment coupled feature groups. A built-in scenario adaptability update mechanism is used to determine the assigned dimension based on the scenario feature matching degree verification of new abnormal cases, synchronously update the feature transmission chain of the corresponding dimension, and coordinate with the interference correction coefficient in the preprocessing stage for optimization.

6. The method according to claim 1, characterized in that, The feature differentiation parsing mechanism includes: Based on the historical anomaly cause feature library and site scenario classification results, exclusive prior feature anchor points for the three types of causes are extracted through feature clustering. The generation process synchronously associates the operating condition fluctuation patterns and micro-environment change characteristics under the same historical scenario of the site. Based on the characteristics of the causes, a differentiated analysis strategy is adapted. Equipment fault features are decomposed into time-domain periods and combined with operating condition parameter correlation analysis. Environmental interference features are captured by instantaneous mutation and superimposed with the time-series matching of micro-environmental mutations. Real pollution features are analyzed by trend gradient and associated with regional pollution transmission trends. The analyzed features are compared for correlation, overlapping areas are identified, and the dominant features are determined by the occurrence time sequence of the causes. The feature differences are resolved by simultaneously combining the current micro-environment parameters with the deviations of the same historical scenario.

7. The method according to claim 1, characterized in that, The specific method for forming a standardized automatic air quality identification feature dataset is as follows: Based on the three types of cause characteristic signals after separation—equipment failure, environmental interference, and actual pollution—data dimensions are divided by site scenario classification. Under each dimension, the fluctuation period, intensity, and associated working conditions and microenvironment parameters of the characteristic signals are defined as basic fields. Standardize the fields of each dimension, assign exclusive coding rules based on the cause type, and convert the feature signal intensity into a normalized value based on the historical normal data of the site, so as to unify the data units and the expression form of abnormal features. Each feature data point is assigned a source identifier, and the micro-environment snapshot and equipment condition logs of the collection period are linked. At the same time, a three-level index structure of cause-feature-scenario is established to form a structured, traceable, and standardized feature dataset that is adapted to the reasoning process.

8. The method according to claim 1, characterized in that, The automatic site scenario adaptation model includes: Real-time microenvironmental parameters are categorized and integrated with historical site scene data to construct a multi-dimensional input matrix for scene recognition. Input data is labeled with attributes according to real-time dynamic data and historical static data. A temporary pollution source-feature association table is established, and a feature scene trigger threshold is preset. Based on real-time feature data exceeding the threshold, the feature weight is automatically adjusted by matching the preset weight adjustment coefficient with the magnitude of the exceedance. The anomaly recognition parameters specific to the site scene are called simultaneously, and the mapping relationship between the feature weight adjustment coefficient and the rule parameter is reversed by combining the recognition results of the parameter application.

9. The method according to claim 1, characterized in that, The multi-round automatic reasoning operation specifically includes the following methods: The first round of execution involves scenario-oriented initial inference, retrieving the priority cause inference rule set corresponding to the site scenario, comparing the standardized feature signals with the feature thresholds and time-series patterns in the rule set, marking the initial screening of abnormal causes and associating them with source information; the second round involves confidence-enhanced inference, calling the inference model trained based on historical cases to review the feature details of the initially screened causes and update the cause confidence; the third round involves scenario-condition matching inference, combining real-time condition parameters to verify the adaptability of the causes to the current scenario, and finally outputting the inference results with cause type, confidence, and source chain.

10. The method according to claim 1, characterized in that, The method for automatically eliminating cross-inducing interference includes: A cross-causal feature association benchmark library is constructed. Based on the feature overlap data of historical abnormal case causes, the association threshold and time-series synchronization parameters of different causal features are determined. The feature association degree evaluation mechanism is invoked to evaluate the overlap and time-series matching degree between the current feature to be analyzed and each causal feature in real time, and generate association degree values. Features with association degree values ​​exceeding the threshold are marked as cross-interference items and removed, and the confidence of the retained features is adjusted in a synchronous manner.

11. The method according to claim 1, characterized in that, The specific method for identifying the anomaly type is as follows: Based on the spatiotemporal correlation characteristics of historical normal data and historical abnormal cases of adjacent sites, a spatiotemporal correlation verification engine is constructed. The preliminary identification results are input into the spatiotemporal correlation verification engine to compare the spatiotemporal synchronization and gradient consistency of abnormal features with the normal baseline. The spatiotemporal correlation verification engine calls the built-in judgment logic library to combine the abnormal distribution of multiple sites, automatically determine the site's regional pollution, equipment failure or environmental interference anomalies, and output the anomaly type result with verification basis.

12. The method according to claim 1, characterized in that, The specific method for establishing the abnormal transmission chain early warning mechanism is as follows: Based on historical transmission cases in the historical anomaly cause feature library, combined with scene information output by the site scene automatic adaptation model, and microenvironment and operating condition data, a cause-specific anomaly transmission chain model is constructed. Based on the cause type, the transmission path is sorted out, and the precursor characteristics and time-series transmission thresholds of each transmission node are identified. The transmission chain dynamic monitoring engine is activated to capture precursor characteristic signals in real time and compare them with the model node characteristics and normal feature segments in the standardized air quality automatic identification feature dataset. When the precursor characteristic exceeds the threshold and conforms to the time-series pattern of the transmission path, an early warning is triggered, and emergency intervention rules are retrieved to form an anomaly transmission chain early warning mechanism.

Images