An atmospheric monitoring data anomaly identification method, system and program product

Abstract

The application provides an atmospheric monitoring data anomaly identification method, system and program product. The method comprises: obtaining to-be-identified atmospheric monitoring data; pre-filtering the to-be-identified atmospheric monitoring data by using a preset audit experience rule library to obtain candidate abnormal data; identifying abnormal identification data from the candidate abnormal data by using an anomaly diagnosis classification model; updating a feature sample library based on the abnormal identification data and a confirmation feedback result corresponding to the abnormal identification data, and the updated feature sample library is used to optimize the anomaly diagnosis classification model. It can be understood that the method improves the accuracy and comprehensiveness of identifying abnormal data in atmospheric monitoring data.

Description

Technical Field

[0001] This application relates to the field of environmental monitoring technology, and in particular to a method, system, and program product for identifying anomalies in atmospheric monitoring data. Background Technology

[0002] With increasingly stringent requirements for ecological and environmental protection, atmospheric environmental monitoring networks can collect atmospheric monitoring data. Furthermore, this data can be used for environmental assessments, such as calculating pollutant concentrations, locating pollution sources, and determining air quality levels.

[0003] Currently, monitoring stations are widely deployed in key pollution source areas such as chemical industrial parks. These stations continuously collect atmospheric monitoring data, including time-series data on pollutants such as inorganic odors and volatile organic compounds, using deployed online atmospheric monitoring equipment. This equipment operates around the clock, generating massive amounts of multi-dimensional monitoring data, providing crucial data support for environmental regulation.

[0004] It is understandable that some anomalies will exist in the collected atmospheric monitoring data. For example, the fluctuation pattern of isobutane data monitored within a certain period may not match its actual variation characteristics. Therefore, it is necessary to verify the validity of the atmospheric monitoring data and remove the anomalies.

[0005] For a long time, the verification of the validity of atmospheric monitoring data has relied primarily on the traditional business model of initial screening by auditors based on their experience, combined with spectral confirmation by on-site maintenance personnel. Faced with the massive amounts of time-series concentrations and chromatograms generated by various monitoring stations around the clock, this purely manual screening model is not only time-consuming and labor-intensive, but also highly dependent on the subjective experience of auditors. Especially when dealing with complex operating conditions such as small baseline drifts over long periods or concurrent fluctuations of multiple components, auditors are prone to overlooking potentially abnormal data, making it difficult to ensure the comprehensiveness of monitoring data quality control and the timeliness of audit warnings.

[0006] In recent years, the industry has gradually introduced software early warning modules to assist in data quality control. However, current automated anomaly identification systems generally suffer from problems such as single judgment dimensions and insufficient depth of mechanism. Furthermore, existing automatic judgment methods are often static and fixed, unable to incorporate the experience of auditors and feedback from on-site verification by operations and maintenance personnel, thus limiting the accuracy and comprehensiveness of anomaly data identification. Summary of the Invention

[0007] This application provides a method, system, and program product for identifying anomalies in atmospheric monitoring data, in order to solve the problem of limited accuracy and comprehensiveness in identifying anomalies.

[0008] Firstly, this application provides a method for identifying anomalies in atmospheric monitoring data, comprising: acquiring atmospheric monitoring data to be identified; pre-filtering the atmospheric monitoring data to be identified using a pre-defined audit experience rule base to obtain candidate anomaly data; wherein the audit experience rule base is a set of logical rules corresponding to non-emission anomalies obtained based on long-term manual audit experience; identifying anomaly identification data from the candidate anomaly data using an anomaly diagnosis classification model; wherein the anomaly diagnosis classification model is based on a feature sample library, which includes historical atmospheric monitoring data and corresponding historical anomaly audit results; updating the feature sample library based on the anomaly identification data and the corresponding confirmation feedback results, wherein the updated feature sample library is used to optimize the anomaly diagnosis classification model, and the confirmation feedback results are used to indicate whether the anomaly identification data is actual anomaly data.

[0009] Understandably, based on the audit experience rule base and the anomaly diagnosis classification model, potential anomalies can be automatically and multidimensionally identified from the atmospheric monitoring data to be identified. This eliminates the need for auditors to manually review and judge suspected anomalies such as sudden changes, zero values, and abnormal chromatographic integrals. In other words, it eliminates the need for manual review to distinguish between fluctuations in actual operating conditions and non-emission anomalies such as instrument malfunctions, sampling anomalies, and system errors. This ensures that the data obtained after removing potential anomalies from the atmospheric monitoring data is authentic and valid. Furthermore, after identification, the feature sample library is updated based on the anomaly identification data and the corresponding confirmation feedback results. This updated feature sample library is then used to optimize the anomaly diagnosis classification model. Through continuous sample updates and incremental model training, the static nature of automatic identification methods can be avoided, effectively reducing the false alarm and false negative rates in complex chemical scenarios, thereby further improving the accuracy and comprehensiveness of anomaly data identification.

[0010] Furthermore, the audit experience rule base is based on logical rules corresponding to non-emission anomalies derived from long-term human audit experience. Combined with an anomaly diagnosis classification model, it constructs a cascaded architecture based on physicochemical pre-filtering and multi-dimensional waveform feature model deep discrimination. This avoids the common problems of single judgment dimensions and insufficient mechanistic depth in current automated anomaly identification systems during manual audits. Through the pre-examination experience rule constraints at the rule layer, abnormal data conforming to logical rules can be directly intercepted, avoiding the computational power consumption of subsequent models in processing massive non-feature fluctuations. This effectively improves the system's concurrent processing efficiency and real-time early warning capabilities for massive high-frequency monitoring data.

[0011] In one possible implementation of the first aspect above, the logical rule corresponding to non-emission anomalies is a custom-configured site-level discrimination logic. The atmospheric monitoring data to be identified includes real-time online monitoring time-series data of each atmospheric monitoring station. The audit experience rule base includes logical rules corresponding to each atmospheric monitoring station. Furthermore, the atmospheric monitoring data to be identified is pre-filtered using the preset audit experience rule base to obtain candidate anomaly data, including: inputting the real-time online monitoring time-series data of each atmospheric monitoring station into the audit experience rule base in sequence; when the real-time online monitoring time-series data triggers at least one logical rule corresponding to the audit experience rule base, the data in the real-time online monitoring time-series data within the corresponding trigger time window is extracted as candidate anomaly data.

[0012] In one possible implementation of the first aspect mentioned above, the real-time online monitoring time series data includes monitoring factor data of at least one monitoring factor, and the logical rules corresponding to non-emission anomalies include at least one of the following: Single-point concentration mutation and peak accumulation error mapping rule: When the concentration value of the monitoring factor shows a sudden increase or decrease that deviates from the overall smooth trend at adjacent time points, it is determined that the monitoring factor data of the monitoring factor corresponds to a non-emission anomaly caused by suspected chromatographic peak accumulation error or peak window drift; Common species zero value and non-accumulated full peak mapping rule: If the monitoring factor shows a stable detection concentration level within a first preset time period, and a zero value or a value lower than the instrument detection limit appears at the first time point, it is determined that the monitoring factor data of the monitoring factor corresponds to a non-emission anomaly caused by suspected non-accumulated full peak; Low-level fluctuation and baseline interference judgment rule: If the concentration level of the monitoring factor is at a preset low level, and the monitoring factor shows a concentration fluctuation of a first preset range other than zero, and does not have a continuous emission curve... Linear characteristics indicate that the monitoring factor data corresponds to a non-emission anomaly suspected to be caused by instrument baseline interference; Negative concentration values ​​are mapped to instrument malfunctions: When the real-time measured concentration value of a monitoring factor is less than zero, the monitoring factor data is determined to correspond to a non-emission anomaly suspected to be caused by instrument malfunctions; Isomer trend deviation determination rule: If the monitoring factor and other monitoring factors with physicochemical correlations or homologous components do not show a synchronous increase or decrease trend in concentration at the same time point, the monitoring factor data is determined to correspond to a non-emission anomaly suspected to be caused by the failure to accumulate the full peak of a low-concentration factor or the error in the accumulation of the peak of a high-concentration factor; Low values ​​of multiple components during the breakpoint recovery period are mapped to sampling faults: If multiple monitoring factors simultaneously show zero values ​​or are less than the first low threshold during the first effective period after the instrument interruption is restored, the monitoring factor data of multiple monitoring factors are determined to correspond to a non-emission anomaly suspected to be caused by the failure to collect the full sample gas.

[0013] Understandably, by incorporating actual physicochemical properties such as divergent isomer trends and abnormal chromatographic peaks into the rules and models, a correlation was established between fluctuations in atmospheric monitoring data and specific instrument malfunctions. This method overcomes the limitations of relying solely on high or low numerical values ​​for alarms, and can accurately identify data anomalies caused by hardware issues such as chromatographic integration errors or incomplete gas sampling.

[0014] In one possible implementation of the first aspect above, an anomaly identification data is obtained from candidate anomaly data using an anomaly diagnosis classification model, including: extracting multidimensional temporal fluctuation features of the candidate anomaly data; inputting the multidimensional temporal fluctuation features of the candidate anomaly data into the anomaly diagnosis classification model to obtain the anomaly probability confidence level corresponding to the candidate anomaly data; if the anomaly probability confidence level is higher than a preset diagnostic threshold, then the candidate anomaly data is determined to be anomaly identification data.

[0015] In one possible implementation of the first aspect mentioned above, the feature sample library is obtained through the following methods: acquiring historical atmospheric monitoring data, which includes online monitoring numerical sequences corresponding to historical suspected anomaly data; calculating mathematical features characterizing the fluctuation pattern of online monitoring numerical sequences based on the online monitoring numerical sequences corresponding to historical suspected anomaly data; parsing historical audit logs associated with the abnormal period to extract historical anomaly audit results, which include the final validity status marked manually and the manually entered anomaly cause diagnosis text; and structurally associating the online monitoring numerical sequences, validity status, and anomaly cause diagnosis text retrieved under the same audit event and storing them in the feature sample library.

[0016] In one possible implementation of the first aspect mentioned above, the feature sample library is updated based on the anomaly identification data and the corresponding confirmation feedback results. The updated feature sample library is used to optimize the anomaly diagnosis classification model. This includes: structurally summarizing the anomaly identification data and its corresponding associated feature parameters to generate a list of anomalies to be verified, which is then sent to the operation and maintenance interaction terminal to initiate a verification request; receiving the confirmation feedback results returned by the operation and maintenance interaction terminal for the anomaly identification data; classifying and labeling the anomaly identification data based on the confirmation feedback results, and dynamically updating the anomaly identification data and its corresponding labeling results to the feature sample library; monitoring the data increment and running cycle of the feature sample library, and when the preset update conditions are met, incrementally training the anomaly diagnosis classification model based on the updated feature sample library to optimize the model.

[0017] In one possible implementation of the first aspect mentioned above, the list of anomalies to be verified contains multiple anomaly data entries, and an auxiliary diagnostic card is generated for each anomaly data entry. The content of the auxiliary diagnostic card includes at least one of the following: anomaly identification data and its concentration fluctuation curve in the adjacent time period; the category of the triggered audit experience logic rule; the anomaly probability confidence level; and a confirmation instruction prompting the operation and maintenance personnel to retrieve the on-site instrument chromatogram for comparison.

[0018] Understandably, the auxiliary diagnostic card pushed to the operation and maintenance (O&M) end in this embodiment can adaptively overlay smooth trend lines and correlation factor fluctuation trajectories according to specific judgment rules, and prompt O&M personnel to retrieve chromatograms. This mechanism provides a structured basis for on-site O&M troubleshooting and shortens the communication time between data reviewers and on-site O&M personnel.

[0019] In one possible implementation of the first aspect above, the abnormal identification data is classified and labeled based on the confirmation feedback results, and the abnormal identification data and the corresponding labeling results are dynamically updated to the feature sample library. This includes: if a retention instruction is received from the operation and maintenance interaction terminal for the abnormal identification data, it is determined to be a real operating condition fluctuation, and the abnormal identification data is labeled as a negative sample; if a removal instruction with abnormal reason text is received from the operation and maintenance interaction terminal, it is determined to be an instrument failure or integration error, and the abnormal identification data and the abnormal reason text corresponding to the abnormal identification data are labeled as positive samples; the structured data with positive and negative sample labels is dynamically added to the corresponding sub-library of the feature sample library; wherein, the abnormal reason text includes at least one of the following: chromatographic analysis and integration abnormality, instrument sampling and pipeline abnormality, and equipment operation and maintenance abnormality. Chromatographic analysis and integration abnormality includes: baseline interference, peak window drift, unknown contaminant interference, integration error, and failure to integrate the full peak; instrument sampling and pipeline abnormality includes: quality control residue and failure to collect the full sample gas; equipment operation and maintenance abnormality includes: instrument failure, standard calibration, and replacement of consumables.

[0020] In one possible implementation of the first aspect above, the preset update conditions include at least one of the following triggering mechanisms: a time cycle triggering mechanism and a data volume threshold triggering mechanism, wherein the time cycle triggering mechanism is: the system's continuous running time reaches a preset natural update cycle; the data volume threshold triggering mechanism is: the total number of newly added positive and negative samples with confirmation labels in the feature sample library reaches a preset sample increment threshold.

[0021] Secondly, embodiments of this application provide an atmospheric monitoring data anomaly identification system, comprising: a data acquisition module for acquiring atmospheric monitoring data to be identified; a rule base pre-filtering module for pre-filtering the atmospheric monitoring data to be identified using a preset audit experience rule base to obtain candidate anomaly data; wherein the audit experience rule base is based on logical rules corresponding to non-emission anomalies obtained from long-term manual audit experience; a classification identification module for identifying anomaly identification data from the candidate anomaly data using an anomaly diagnosis classification model; wherein the anomaly diagnosis classification model is based on a feature sample library, which includes historical atmospheric monitoring data and corresponding historical anomaly audit results; and a confirmation update module for updating the feature sample library based on the anomaly identification data and the confirmation feedback results corresponding to the anomaly identification data, wherein the updated feature sample library is used to optimize the anomaly diagnosis classification model.

[0022] Thirdly, embodiments of this application provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the first aspect and any of the possible implementations of the first aspect. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of a method for identifying anomalies in atmospheric monitoring data provided in an embodiment of this application.

[0025] Figure 2 According to some embodiments of this application, a schematic diagram of a process for optimizing anomaly diagnosis classification models based on anomaly identification data is shown.

[0026] Figure 3 This is a schematic diagram illustrating the information structure of an auxiliary diagnostic card provided in an embodiment of this application.

[0027] Figure 4 This is a block diagram of another method for identifying anomalies in atmospheric monitoring data provided in an embodiment of this application.

[0028] Figure 5 This is a schematic diagram of an atmospheric monitoring data anomaly identification system provided in an embodiment of this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] As shown in the background section, the accuracy and comprehensiveness of identifying anomalous data in atmospheric monitoring data are limited.

[0031] Therefore, to address the aforementioned issues, this application proposes a method for identifying anomalies in atmospheric monitoring data. This method involves: acquiring atmospheric monitoring data to be identified; pre-filtering the atmospheric monitoring data to be identified using a pre-defined audit experience rule base to obtain candidate anomaly data; wherein the audit experience rule base is based on logical rules corresponding to non-emission anomalies obtained through long-term manual audit experience; identifying anomaly identification data from the candidate anomaly data using an anomaly diagnosis classification model; wherein the anomaly diagnosis classification model is based on a feature sample library, which includes historical atmospheric monitoring data and corresponding historical anomaly audit results; updating the feature sample library based on the anomaly identification data and the corresponding confirmation feedback results; the updated feature sample library is used to optimize the anomaly diagnosis classification model; wherein the confirmation feedback results are used to indicate whether the anomaly identification data is actual anomaly data.

[0032] Understandably, based on the audit experience rule base and the anomaly diagnosis classification model, potential anomalies can be automatically and multidimensionally identified from the atmospheric monitoring data to be identified. This eliminates the need for auditors to manually review and judge suspected anomalies such as sudden changes, zero values, and abnormal chromatographic integrals. In other words, it eliminates the need for manual review to distinguish between fluctuations in actual operating conditions and non-emission anomalies such as instrument malfunctions, sampling anomalies, and system errors. This ensures that the data obtained after removing potentially anomalies from the atmospheric monitoring data is authentic and valid. Furthermore, after identification, the feature sample library is updated based on the anomaly identification data and the corresponding confirmation feedback results. This updated feature sample library is then used to optimize the anomaly diagnosis classification model. Through continuous sample updates and incremental model training, the static nature of automatic identification methods can be avoided, effectively reducing the false alarm and false negative rates in complex chemical scenarios, thereby further improving the accuracy and comprehensiveness of anomaly data identification.

[0033] Furthermore, the audit experience rule base is based on logical rules corresponding to non-emission anomalies derived from long-term human audit experience. Combined with an anomaly diagnosis classification model, it constructs a cascaded architecture based on physicochemical pre-filtering and multi-dimensional waveform feature model deep discrimination. This avoids the common problems of single judgment dimensions and insufficient mechanistic depth in current automated anomaly identification systems during manual audits. Through the pre-examination experience rule constraints at the rule layer, abnormal data conforming to logical rules can be directly intercepted, avoiding the computational power consumption of subsequent models in processing massive non-feature fluctuations. This effectively improves the system's concurrent processing efficiency and real-time early warning capabilities for massive high-frequency monitoring data.

[0034] In some embodiments, the atmospheric monitoring data to be identified includes real-time online monitoring time-series data from various atmospheric monitoring stations, such as online monitoring time-series data uploaded in real time by front-end devices. The real-time online monitoring time-series data includes monitoring factor data for at least one monitoring factor. For example, the monitoring factor may include, but is not limited to, conventional components such as ethane and propane, and related components such as nitric oxide.

[0035] It is understood that the atmospheric monitoring data anomaly identification method provided in this application embodiment can be applied to the scenario of reviewing online monitoring data of volatile organic compounds and other pollutants in key pollution source areas such as chemical industrial parks.

[0036] In some embodiments, the logical rules corresponding to non-emission anomalies may include at least one of the following: single-point concentration mutation and peak error mapping rule, common species zero value and non-accumulated full peak mapping rule, low-level fluctuation and baseline interference judgment rule, concentration negative value and instrument failure mapping rule, isomer trend deviation judgment rule, and multi-component low value and sampling failure mapping rule during the breakpoint recovery period.

[0037] Specifically, the mapping rule between single-point concentration mutations and peak accumulation errors is as follows: When the concentration value of a monitored factor shows a sudden increase or decrease that deviates from the overall smooth trend at adjacent time points, it is determined that the monitored factor data corresponds to a non-emission anomaly caused by suspected chromatographic peak accumulation errors or peak window drift. For example, when the hourly concentration data at a certain time point increases by more than 150% or decreases by more than 80% compared with the moving average of the previous 3 hours, it is initially intercepted and marked as a candidate anomaly. Based on auditing experience, this phenomenon is usually caused by peak accumulation errors or peak window drift during chromatographic analysis.

[0038] Mapping rules for zero values ​​and non-accumulated peaks for common species: If a monitored factor exhibits a stable detection concentration level within a first preset time period, and a value of zero or below the instrument's detection limit appears at the first moment, then the monitored factor data is determined to correspond to a non-emission anomaly caused by suspected non-accumulated peaks. For example, if the monitored factor is a common component such as ethane or propane, and when a common component such as ethane or propane suddenly shows a value of zero or below 0.5 ppb in continuous 24-hour data, an interception alert is issued, prioritizing the cause of the fault as a non-accumulated peak phenomenon caused by baseline elevation or improper threshold setting during chromatographic integration.

[0039] Low-level fluctuations and baseline interference judgment rules: If the concentration level of a monitored factor is at a preset low level, and the monitored factor exhibits a concentration fluctuation outside the first preset range (extra-zero value) without exhibiting continuous emission curve characteristics, then the monitored factor data is judged to correspond to a non-emission anomaly caused by suspected instrument baseline interference. For example, for a monitored factor whose overall concentration level is normally low, if a concentration fluctuation outside the first preset range (minor concentration fluctuation) occurs, such as remaining in an extremely small range of 0 ppb to 0.1 ppb for a long period of time, and without exhibiting continuous emission curve characteristics, then based on audit experience, it is judged to be suspected instrument baseline interference.

[0040] Negative concentration values ​​are mapped to instrument malfunctions according to the following rule: When the real-time measured concentration of a monitored factor is less than zero, the monitored factor data is determined to correspond to a non-emission anomaly caused by a suspected instrument malfunction. For example, if the real-time measured concentration of a monitored factor is less than zero, such as when the measured minute data of nitric oxide is less than 0 ppb, it is determined to be due to negative baseline drift of the instrument, detector malfunction, or calibration zero-point offset.

[0041] Isomer trend deviation judgment rule: If a monitored factor and other monitored factors of physicochemically related isomers or homologous components do not show a synchronous increase or decrease trend in concentration at the same time point, then the monitored factor data corresponding to the monitored factor is determined to be a non-emission anomaly caused by the failure of a low-concentration factor to accumulate a full peak or an error in the accumulation of a high-concentration factor. For example, if monitored factors such as n-butane and isobutane are physicochemically related isomers or homologous components, and the first derivatives of their concentration fluctuations at the same time point have opposite signs, or the absolute value of the difference in short-term concentration change rates exceeds 30%, then it is determined to be a failure of a low-concentration component to accumulate a full peak or an error in the accumulation of a high-concentration component.

[0042] Mapping rule between low values ​​of multiple components and sampling faults during the breakpoint recovery period: If multiple monitoring factors simultaneously show zero values ​​or values ​​below the first low threshold during the first effective analysis period after instrument interruption recovery, then the monitoring factor data of multiple monitoring factors are determined to correspond to non-emission anomalies suspected to be caused by incomplete gas sampling. For example, if multiple monitoring factors with no inherent physicochemical correlation simultaneously show zero values ​​or extremely low values ​​during the first effective analysis period after communication interruption recovery or system maintenance, it is determined to be a suspected case of incomplete gas sampling or incomplete replacement of residual gas in the sampling pipeline.

[0043] Understandably, by incorporating actual physicochemical properties such as divergent isomer trends and abnormal chromatographic peaks into the rules and models, a correlation was established between fluctuations in atmospheric monitoring data and specific instrument malfunctions. This method overcomes the limitations of relying solely on high or low numerical values ​​for alarms, and can accurately identify data anomalies caused by hardware issues such as chromatographic integration errors or incomplete gas sampling.

[0044] Figure 1 According to some embodiments of this application, a schematic diagram of a method for identifying anomalies in atmospheric monitoring data is provided. The specific steps are as follows: S101, acquire atmospheric monitoring data to be identified.

[0045] As mentioned above, the atmospheric monitoring data to be identified includes real-time online monitoring time-series data from various atmospheric monitoring stations, such as online monitoring time-series data uploaded in real time by front-end devices. The real-time online monitoring time-series data includes monitoring factor data for at least one monitoring factor. It is understood that the real-time online monitoring time-series data can be data obtained after cleaning the original real-time online monitoring data, or it can be the original real-time online monitoring data itself; no specific limitation is made here.

[0046] S102, the atmospheric monitoring data to be identified is pre-filtered using a pre-set audit experience rule base to obtain candidate abnormal data. The audit experience rule base is a set of logical rules corresponding to non-emission anomalies, derived from long-term human audit experience.

[0047] In some embodiments, the logical rules corresponding to non-emission anomalies are custom-configured site-level discrimination logic. It is understood that these logical rules are applicable to all atmospheric monitoring stations, and specific discrimination rules can be added for different stations based on their historical audit results, monitoring environment, and pollutant characteristics, or some monitoring factors can be configured differently. For example, all atmospheric monitoring stations can uniformly apply basic logical rules, including rules for mapping single-point concentration mutations and peak accumulation errors, rules for mapping zero values ​​and non-accumulated full peaks for common species, and rules for judging low-level fluctuations and baseline interference. For example, under the premise of uniformly applying the above basic rules, specific discrimination rules can be added for specific monitoring factors at specific stations. For instance, the dichloromethane monitoring factor at atmospheric monitoring station 1 can be configured with a specific discrimination rule; when it shows a non-zero value, it is judged as suspected baseline interference. Other atmospheric monitoring stations' dichloromethane monitoring factors may not be configured with this specific rule, and only the general basic logical rules apply. It is understood that the specific content of the logical rules has been described above and will not be repeated here.

[0048] Specifically, in some implementations, the real-time online monitoring time-series data of each atmospheric monitoring station are sequentially input into the audit experience rule base; when the real-time online monitoring time-series data triggers at least one corresponding logical rule in the audit experience rule base, the data in the real-time online monitoring time-series data within the corresponding trigger time window is extracted as candidate abnormal data.

[0049] For example, taking the case where different atmospheric monitoring stations correspond to the same logical rules, each atmospheric monitoring station refers to the atmospheric monitoring stations within the chemical industrial park platform system. Real-time online monitoring time-series data uploaded by front-end devices belonging to the chemical industrial park platform system is obtained and sequentially input into the audit experience rule base for pre-filtering. If the current online monitoring time-series data triggers any logical rule, the context data before and after the triggering period is extracted and marked as candidate abnormal data for initial interception.

[0050] Understandably, pre-filtering means that when the current real-time monitoring time series data meets the discrimination rules corresponding to any of the following candidate abnormal data, it is judged as candidate abnormal data and intercepted. It should be noted that in this embodiment, the rule judgment result is marked as "candidate", which aims to establish a cascaded architecture from preliminary rule interception to machine learning model fine judgment and then to manual confirmation by operation and maintenance.

[0051] S103, an anomaly identification data is obtained from candidate anomaly data using an anomaly diagnosis and classification model. The anomaly diagnosis and classification model is based on a feature sample library, which includes historical atmospheric monitoring data and corresponding historical anomaly review results.

[0052] Understandably, anomaly diagnosis and classification models can be built using machine learning models (such as random forest classification algorithms, SVM, XGBoost, LightGBM, etc.). To overcome the limitation of single static empirical rules being prone to false positives, anomaly diagnosis and classification models can perform deep feature identification on candidate anomaly data that is blocked due to rule triggering during the pre-filtering stage.

[0053] In some embodiments, multidimensional temporal fluctuation features of candidate abnormal data are extracted; the multidimensional temporal fluctuation features of candidate abnormal data are input into an anomaly diagnosis classification model to obtain the anomaly probability confidence level corresponding to the candidate abnormal data; if the anomaly probability confidence level is higher than a preset diagnostic threshold, the candidate abnormal data is determined to be anomaly identification data.

[0054] For example, multidimensional time-series fluctuation features of candidate abnormal data are extracted and input into a pre-trained abnormality diagnosis classification model for secondary deep discrimination. The abnormality diagnosis classification model matches the concentration curve fluctuation pattern with the correlation features of multiple components and outputs the confidence score of the candidate data belonging to a specific physicochemical fault. This effectively pre-filters false anomalies caused by fluctuations in normal operating conditions and completes the final definition of possible abnormal data, that is, obtains abnormal identification data.

[0055] Specifically, multi-dimensional time-series feature vectors of candidate anomaly data are extracted, including peak width, peak height change rate, and multi-component correlation coefficients. These vectors are then input into a random forest classifier trained on a feature sample library containing massive amounts of historical confirmed information. Internally, the model uses multiple decision tree nodes to perform high-dimensional space matching and voting scoring between the current waveform's feature vector and historical similar anomaly samples. This outputs the probability confidence score that the candidate anomaly data belongs to a specific physicochemical fault (such as integration error, incomplete gas sampling, baseline interference, etc.). If the confidence score is higher than a 90% threshold, it indicates that historical similar features are mostly real faults, and the data is ultimately classified as suspected anomaly data and retained. If the confidence score is lower than the threshold, it is determined that the data is likely a real operating condition fluctuation triggered by a rule error and is filtered out. This model's identification step effectively eliminates false anomalies caused by static rules, significantly improving the accuracy of anomaly identification.

[0056] S104, based on the anomaly identification data and the corresponding confirmation feedback results, update the feature sample library. The updated feature sample library is used to optimize the anomaly diagnosis classification model. The confirmation feedback results are used to indicate whether the anomaly identification data is actual anomaly data.

[0057] In some embodiments, the anomaly identification data and corresponding associated feature parameters are structurally summarized to generate a list of anomalies to be verified. This list is sent to the operation and maintenance interaction terminal to initiate a verification request. The system receives confirmation feedback results from the operation and maintenance interaction terminal regarding the anomaly identification data. Based on the confirmation feedback results, the anomaly identification data is classified and labeled, and the anomaly identification data and corresponding labeling results are dynamically updated to the feature sample library. The system monitors the data increment and running cycle of the feature sample library. When preset update conditions are met, the anomaly diagnosis classification model is incrementally trained based on the updated feature sample library to optimize the model.

[0058] Furthermore, in some implementations, when the identified anomaly data and associated feature parameters are structurally summarized to generate a list of anomalies to be verified, an auxiliary diagnostic card is rendered and pushed to the operations and maintenance (O&M) interface for each anomaly data entry. In addition to highlighting the concentration curve of the suspected anomaly segment, the card automatically overlays the historical normal average curve of the same site and the fluctuation trajectory of associated factors, allowing O&M personnel to quickly compare and locate anomalies. This step will be explained below. Figure 2 A detailed introduction will be provided in the following section.

[0059] The anomaly diagnosis classification model used in step S103 above will be described in detail below. It is understandable that the anomaly diagnosis classification model is trained based on a feature sample database.

[0060] The feature sample library can be obtained in the following ways. It is understood that the feature sample library can include an initial feature sample library and dynamically updated positive and negative sample sub-libraries. The positive and negative sample sub-libraries include anomaly identification data and corresponding labeling results obtained during the anomaly identification process.

[0061] The initial feature sample library can be constructed based on the multidimensional time-series features of historical suspected anomaly data and the finally confirmed physicochemical fault labels. Specifically, historical atmospheric monitoring data is acquired, including online monitoring numerical sequences corresponding to historical suspected anomaly data; based on the online monitoring numerical sequences corresponding to historical suspected anomaly data, mathematical features characterizing the fluctuation pattern of the online monitoring numerical sequences are calculated; historical audit logs associated with the abnormal period are analyzed to extract historical anomaly audit results, including the final validity status after manual labeling and the manually entered anomaly cause diagnosis text; the online monitoring numerical sequences, validity status, and anomaly cause diagnosis text retrieved under the same audit event are structurally associated and stored in the feature sample library.

[0062] For example, long-term historical online monitoring data and synchronous historical manual review logs from various atmospheric monitoring stations in the chemical industrial zone can be obtained. The original time series of monitoring factors such as VOCs components, conventional factors, and inorganic odor in the historical online monitoring data can be cleaned to remove known abnormal segments such as instrument calibration and faults, and extract multi-dimensional time series features, thereby obtaining the online monitoring value sequence corresponding to the historical suspected abnormal data and the corresponding historical abnormality review results.

[0063] Furthermore, to achieve sufficient training and high-precision discrimination of the anomaly diagnosis classification model, this embodiment specifies the parameter configuration for the construction and training process of the anomaly diagnosis classification model: in addition to the preferred random forest classification algorithm, the anomaly diagnosis classification model can also be constructed using ensemble learning algorithms such as SVM, XGBoost, and LightGBM, or deep learning models such as LSTM suitable for time-series feature mining, to adapt to the data scale and computing power conditions of different chemical industrial zones. Regarding the sample partitioning and training mechanism, the system randomly partitions the initial feature sample library and the dynamically updated positive and negative sample sub-libraries into training and test sets at 70% and 30% respectively. In terms of core hyperparameter configuration and optimization, taking the random forest algorithm as an example, the necessary hyperparameter configurations during the training process include: the preferred setting range for the number of decision trees is 50 to 200, the preferred setting range for the maximum depth of the trees is 5 to 20, and the preferred setting range for the minimum number of samples required for internal node subdivision is 2 to 10. To prevent overfitting on specific chemical industrial zone datasets, a grid search combined with 5-fold cross-validation is preferred for iterative optimization of the hyperparameters. During training, accuracy, precision, recall, and F1 score are used as core evaluation metrics. When the model's F1 score on the test set converges continuously and reaches a preset diagnostic accuracy threshold of 0.85 or higher, the current training cycle ends, and the optimal model weight file is output and saved for use by the anomaly diagnosis classification model in the secondary deep discrimination process.

[0064] Figure 2 According to some embodiments of this application, a schematic diagram of a process for optimizing an anomaly diagnosis classification model based on anomaly identification data is shown. The specific steps are as follows: S201, the anomaly identification data and corresponding associated feature parameters are structured and summarized to generate a list of anomalies to be verified. The list of anomalies to be verified is sent to the operation and maintenance interaction terminal to initiate a verification request.

[0065] The list of anomalies to be verified contains multiple abnormal data entries. For each abnormal data entry, an auxiliary diagnostic card is generated. The auxiliary diagnostic card includes at least one of the following: the anomaly identification data and its concentration fluctuation curve in the adjacent time period; the category of the audit experience logic rule triggered; the anomaly probability confidence level; and a confirmation instruction prompting the operation and maintenance personnel to retrieve the on-site instrument chromatogram for comparison.

[0066] For example, the identified abnormal data (i.e., possible abnormal data) and the associated feature parameters related to the abnormal data are structured and summarized to generate a list of anomalies to be verified. For each abnormal data, an auxiliary diagnostic card is generated and pushed to the operation and maintenance interaction terminal. In addition to highlighting the concentration curve of the suspected abnormal segment corresponding to the abnormal data, the card also automatically displays the historical normal average curve of the same site and the fluctuation trajectory of the associated factors, so that operation and maintenance personnel can quickly compare and locate the anomaly.

[0067] In some implementations, the specific interactive implementation of anomaly cause classification and labeling is as follows: the operation and maintenance interaction terminal provides touch controls for retaining and removing confirmed cases; when the operation and maintenance personnel retrieve the underlying original chromatogram for comparison, if it is determined to be a reasonable fluctuation, they click to retain to generate a negative sample; if it is determined to be an abnormal fluctuation, they click to remove and trigger a structured anomaly cause dictionary for selection. The dictionary mainly includes: integration error, baseline interference, incomplete peak integration, standard calibration, etc.

[0068] S202, Receive confirmation feedback results from the operation and maintenance interaction terminal for anomaly identification data.

[0069] In some embodiments, the system receives confirmation feedback results from the operation and maintenance interaction terminal, which combine anomaly identification data (i.e., identified suspected anomaly data) with a chromatogram. It is understood that the confirmation feedback results are used to indicate whether the anomaly identification data is indeed actual anomaly data.

[0070] In some implementations, the confirmation feedback result can be a retention instruction or a rejection instruction. A rejection instruction indicates that the identified anomaly data is actual anomaly data, while a retention instruction indicates that the identified anomaly data is not actual anomaly data (i.e., fluctuations in real operating conditions). Furthermore, when the confirmation feedback result is a rejection instruction, the rejection instruction also includes an anomaly reason text. It is understood that the anomaly reason text can cover dimensions such as chromatographic integrals and instrument sampling.

[0071] The anomaly cause text includes at least one of the following: chromatographic analysis and integration anomaly, instrument sampling and tubing anomaly, and equipment operation and maintenance anomaly. Chromatographic analysis and integration anomalies include: baseline interference, peak window drift, interference from unknown contaminants, integration error, and incomplete peak integration. Instrument sampling and tubing anomalies include: quality control residue and incomplete sample gas sampling. Equipment operation and maintenance anomalies include: instrument malfunction, standard calibration, and replacement of consumables (such as silica gel). Structured data with positive and negative sample labels will be dynamically added to the corresponding sub-library of the feature sample library.

[0072] S203, classify and label the anomaly identification data based on the confirmation feedback results, and dynamically update the anomaly identification data and the corresponding labeling results to the feature sample library.

[0073] In some implementations, if a retention instruction is received from the operation and maintenance interaction terminal for anomaly identification data, it is determined to be a real fluctuation in operating conditions, and the anomaly identification data is marked as a negative sample; if a removal instruction with anomaly cause text is received from the operation and maintenance interaction terminal, it is determined to be an instrument malfunction or integration error, and the anomaly identification data and the anomaly cause text corresponding to the anomaly identification data are marked as positive samples.

[0074] For example, based on the confirmation feedback results, the time series data corresponding to the anomaly identification data are classified and labeled as positive and negative samples. Among them, the real data anomalies and analysis error features that are to be removed according to the operation and maintenance feedback are classified as positive samples, which are the objects to be learned first by the anomaly diagnosis classification model. The false anomalies of complex production condition fluctuations that are to be retained according to the operation and maintenance feedback are classified as negative samples, which are used to help the model correct the identification boundary and reduce the false alarm rate. Finally, the structured labeled data is dynamically added to the corresponding sub-library of the feature sample library.

[0075] S204 monitors the data increment and running cycle of the feature sample library. When the preset update conditions are met, the abnormal diagnosis classification model is incrementally trained based on the updated feature sample library to optimize the model.

[0076] Understandably, the preset update conditions may include at least one of the following triggering mechanisms: time period triggering mechanism and data volume threshold triggering mechanism. The time period triggering mechanism is triggered when the system runs continuously for a preset natural update cycle. The data volume threshold triggering mechanism is triggered when the total number of newly added positive and negative samples with confirmation labels in the feature sample library reaches a preset sample increment threshold.

[0077] In addition, suspected abnormal data (including anomaly identification data and historical suspected abnormal data), confirmation feedback results and model iteration records are uniformly archived to establish an anomaly identification and model iteration historical information database. Combining this historical information database with the latest monitoring data identification results, the operating status of monitoring equipment and model identification accuracy of each atmospheric monitoring station are determined, providing a basis for operation and maintenance optimization and model parameter adjustment.

[0078] Understandably, the auxiliary diagnostic card pushed to the operation and maintenance (O&M) end in this embodiment can adaptively overlay smooth trend lines and correlation factor fluctuation trajectories according to specific judgment rules, and prompt O&M personnel to retrieve chromatograms. This mechanism provides a structured basis for on-site O&M troubleshooting and shortens the communication time between data reviewers and on-site O&M personnel.

[0079] Understandably, the embodiments of this application possess in-depth feature mining and accurate discrimination capabilities for abnormal data phenomena caused by instrument malfunctions such as integration errors, baseline interference, and incomplete gas sampling. Simultaneously, the dynamic rule base and diagnostic model can incorporate the experience of auditors and feedback from on-site verification during operation and maintenance, thereby improving the accuracy and comprehensiveness of the data. Therefore, it can meet strong practical needs and has significant application value.

[0080] Figure 3 According to some embodiments of this application, a schematic diagram illustrating the information structure of an auxiliary diagnostic card is shown. For example... Figure 3 As shown, the auxiliary diagnostic card includes: a multi-dimensional feature visualization area, an intelligent diagnostic information presentation area, and an operation and maintenance confirmation labeling area.

[0081] Multidimensional feature visualization area: Automatically extracts 3-hour time periods before and after suspected abnormal data points as context time windows, generates concentration fluctuation curves containing suspected abnormal data points, and highlights them; at the same time, adaptively overlays and displays the smooth trend line of the data sequence and the fluctuation trajectory of correlation factors with physicochemical correlation.

[0082] Intelligent diagnostic information presentation area: Extracts the specific audit experience rule categories triggered in the pre-filtering stage of this time series segment, such as single-point concentration mutation, isomer trend deviation, etc., as well as the probability confidence of the abnormal identification data belonging to a specific physicochemical fault output by the abnormal diagnostic classification model, such as 90% confidence.

[0083] Operation and maintenance confirmation labeling area: includes a prompt instruction for operation and maintenance personnel to retrieve the chromatograms of the on-site instruments for manual verification, and is configured with a "retain / reject" triage decision control for receiving confirmation feedback, as well as a physicochemical fault cause selection library that pops up in conjunction with the rejection instruction.

[0084] Figure 4According to some embodiments of this application, a block diagram of another method for identifying anomalies in atmospheric monitoring data is provided. As shown in the figure, an audit experience rule base can be constructed based on long-term manual auditing experience, and an initial feature sample library can be constructed by combining historical online monitoring data and audit logs from various atmospheric monitoring stations within the chemical industrial park platform system. Upon obtaining real-time online monitoring data, the data can be cleaned first, and then anomaly identification data can be obtained through the audit experience rule base and anomaly diagnosis classification model. The anomaly identification data is then structured and summarized to generate a list to be audited. The list to be audited and auxiliary diagnostic cards are pushed to the operation and maintenance interaction segment. The time-series data corresponding to the anomaly identification data is labeled according to the removal or retention instructions sent by the operation and maintenance interaction segment. When the label is a positive sample, the structured data corresponding to the positive sample is added to the positive sample sub-library of the feature sample library; when the label is a negative sample, the structured data corresponding to the negative sample is added to the negative sample sub-library of the feature sample library, thereby achieving dynamic updating of the feature sample library. When incremental training is triggered, the anomaly diagnosis analysis model is optimized and evaluated based on the updated feature sample library.

[0085] Understandably, by integrating historical online monitoring data and historical audit logs from various monitoring stations in the chemical industrial park, and flexibly utilizing technologies such as machine learning and data structuring, an audit experience rule base and a dynamic feature sample base are constructed. Combined with an anomaly diagnosis and classification model, accurate anomaly identification of atmospheric monitoring data is achieved. Anomaly confirmation feedback is completed through linkage with the operation and maintenance interactive terminal. Through dynamic sample updates and incremental model training, the anomaly identification capability is continuously optimized, providing technical support for atmospheric monitoring data quality control, operation and maintenance collaboration, and environmental supervision in the chemical industrial park.

[0086] Figure 5 According to some embodiments of this application, an atmospheric monitoring data anomaly identification system is shown. For example... Figure 5 The atmospheric monitoring data anomaly identification system shown includes: a data acquisition module, a rule base pre-filtering module, a classification and identification module, and a confirmation and update module.

[0087] Specifically, the data acquisition module is used to acquire atmospheric monitoring data to be identified.

[0088] The rule base pre-filtering module is used to pre-filter the atmospheric monitoring data to be identified using a preset audit experience rule base to obtain candidate abnormal data. The audit experience rule base is a set of logical rules corresponding to non-emission anomalies, which are obtained based on long-term human audit experience.

[0089] The classification and identification module is used to identify abnormal data from candidate abnormal data using an anomaly diagnosis and classification model. The anomaly diagnosis and classification model is based on a feature sample library, which includes historical atmospheric monitoring data and corresponding historical anomaly review results. The confirmation update module is used to update the feature sample library based on the anomaly identification data and the corresponding confirmation feedback results. The updated feature sample library is used to optimize the anomaly diagnosis classification model.

[0090] Understandably, the specific content corresponding to the functions of each module can be referred to the specific possible implementation methods of each execution step in the above-mentioned atmospheric monitoring data anomaly identification method, and will not be elaborated here.

[0091] For example, the rule base pre-filtering module may include a multi-dimensional data cleaning and database building module, which is used to establish logical rules for corresponding non-emission anomalies based on long-term human review experience.

[0092] For example, the confirmation update module may further include: an auxiliary diagnostic card rendering and interaction module, a sample dynamic classification and labeling module, and a model adaptive incremental training module. The auxiliary diagnostic card rendering and interaction module is used to structurally summarize the anomaly identification data and its corresponding associated feature parameters, generating a list of anomalies to be verified. This list is sent to the operation and maintenance interaction terminal to initiate a verification request. The list may also include rendered auxiliary diagnostic cards, which are also pushed to the operation and maintenance interaction terminal. The sample dynamic classification and labeling module receives the confirmation feedback results from the operation and maintenance interaction terminal; classifies and labels the anomaly identification data based on the confirmation feedback results; and dynamically updates the anomaly identification data and its corresponding labeling results to the feature sample library. The model adaptive incremental training module monitors the data increment and running cycle of the feature sample library. When preset update conditions are met, it performs incremental training on the anomaly diagnosis classification model based on the updated feature sample library to optimize the model.

[0093] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the atmospheric monitoring data anomaly identification method provided in this embodiment. For example, executing... Figure 1 or Figure 4 The method for identifying anomalies in atmospheric monitoring data is shown. It is understood that during the implementation of the atmospheric monitoring data anomaly identification method provided in this application embodiment, the interactive maintenance terminal required for the electronic device can be located on this electronic device or on other electronic devices, which will not be elaborated upon here.

[0094] For example, electronic devices may include, but are not limited to, servers, personal computers, terminal devices, tablets, industrial control computers, embedded devices, edge computing nodes, mobile terminals, smartphones, workstations, and cloud hosts.

[0095] This application also provides a computer-readable medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the atmospheric monitoring data anomaly identification method mentioned in any of the above embodiments of this application.

[0096] This application also provides a computer program product, including: a computer program / instruction, which, when executed by a processor, implements the atmospheric monitoring data anomaly identification method mentioned in any of the above embodiments of this application.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Those skilled in the art should understand that the steps, model algorithms, and parameter configurations for identifying anomalies in atmospheric monitoring data in the above embodiments are merely examples, and can be flexibly adjusted according to the types of monitoring factors and audit requirements of specific chemical industrial parks in practical applications. Furthermore, the system and its functional modules described in this invention can be executed by a processor in the form of pure software programs, or can be equivalently transformed into hardware entities such as logic gates, application-specific integrated circuits, or programmable controllers. Any modifications or equivalent substitutions made to the step parameters within the spirit and principles of this invention, as well as any form of conversion between software logic and hardware physical structure of functional modules, should be covered within the protection scope of this invention.

[0098] It should be noted that all units / modules mentioned in the device embodiments of this application are logical units / modules. Physically, a logical unit / module can be a physical unit / module, a part of a physical unit / module, or a combination of multiple physical units / modules. The physical implementation of these logical units / modules themselves is not the most important factor; the combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed in this application. Furthermore, to highlight the innovative aspects of this application, the above-described device embodiments of this application have not introduced units / modules that are not closely related to solving the technical problems proposed in this application. This does not mean that the above-described device embodiments do not contain other units / modules.

[0099] It should be noted that, in the examples and description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0100] Although this application has been illustrated and described with reference to certain embodiments thereof, those skilled in the art should understand that various changes in form and detail may be made thereto without departing from the spirit and scope of this application.

Claims

1. A method for identifying anomalies in atmospheric monitoring data, characterized in that, include: Acquire atmospheric monitoring data to be identified; The atmospheric monitoring data to be identified is pre-filtered using a pre-set audit experience rule base to obtain candidate abnormal data; wherein, the audit experience rule base is a set of logical rules corresponding to non-emission anomalies, which are obtained based on long-term human audit experience. Anomaly identification data is obtained from the candidate anomaly data using an anomaly diagnosis classification model; wherein, the anomaly diagnosis classification model is based on a feature sample library, which includes historical atmospheric monitoring data and corresponding historical anomaly review results; Based on the anomaly identification data and the corresponding confirmation feedback results, the feature sample library is updated. The updated feature sample library is used to optimize the anomaly diagnosis classification model. The confirmation feedback results are used to indicate whether the anomaly identification data is actual anomaly data.

2. The method according to claim 1, characterized in that, The corresponding non-emission anomaly logic rule is a custom-configured site-level discrimination logic. The atmospheric monitoring data to be identified includes real-time online monitoring time series data of each atmospheric monitoring station. The audit experience rule base includes logical rules corresponding to each atmospheric monitoring station. Furthermore, the process of pre-filtering the atmospheric monitoring data to be identified using a preset audit experience rule base to obtain candidate abnormal data includes: The real-time online monitoring time series data of each of the aforementioned atmospheric monitoring stations are sequentially input into the audit experience rule base; When the real-time online monitoring time-series data triggers at least one corresponding logical rule in the audit experience rule base, the data within the corresponding trigger time window in the real-time online monitoring time-series data is extracted as candidate abnormal data.

3. The method according to claim 2, characterized in that, The real-time online monitoring time series data includes monitoring factor data for at least one monitoring factor, and the logical rules corresponding to non-emission anomalies include at least one of the following: Single-point concentration mutation and peak accumulation error mapping rule: When the concentration value of the monitored factor shows a sudden increase or decrease that deviates from the overall smooth trend at adjacent time points, it is determined that the monitoring factor data of the monitored factor corresponds to a non-emission anomaly caused by suspected chromatographic peak accumulation error or peak window drift. Mapping rule between zero value and non-accumulated full peak for common species: If the monitoring factor shows a stable detection concentration level within the first preset time period, and a zero value or a value lower than the instrument detection limit appears at the first moment, then the monitoring factor data of the monitoring factor is determined to be a non-emission anomaly caused by a suspected non-accumulated full peak. Low-level fluctuation and baseline interference judgment rule: If the concentration level of the monitored factor is at a preset low level, and the monitored factor shows a concentration fluctuation of a first preset range that is different from zero, and does not have the curve characteristics of continuous emission, then the monitoring factor data of the monitored factor is judged to correspond to a non-emission anomaly caused by suspected instrument baseline interference. Concentration Negative Value Mapping Rule: When the real-time measured concentration value of the monitored factor is less than zero, the monitoring factor data of the monitored factor is determined to be a non-emission anomaly caused by a suspected instrument malfunction. Isomer trend deviation judgment rule: If the monitoring factor and other monitoring factors of isomers or homologous components with physicochemical correlation do not show a synchronous increase or decrease trend in concentration at the same time point, then the monitoring factor data of the monitoring factor is determined to be a non-emission anomaly caused by the failure of the full peak of the suspected low concentration factor or the peak accumulation error of the high concentration factor. Mapping rule between low values ​​of multiple components and sampling faults during the breakpoint recovery period: If multiple monitoring factors simultaneously show zero values ​​or are less than the first low threshold during the first effective period after the instrument interruption is restored, then the monitoring factor data of multiple monitoring factors are determined to correspond to non-emission anomalies caused by suspected failure to collect full sample gas.

4. The method according to claim 1, characterized in that, The step of using an anomaly diagnosis classification model to identify anomaly identification data from the candidate anomaly data includes: Extract the multidimensional temporal fluctuation features of the candidate anomaly data; The multidimensional temporal fluctuation features of the candidate anomaly data are input into the anomaly diagnosis classification model to obtain the anomaly probability confidence level corresponding to the candidate anomaly data. If the confidence level of the anomaly probability is higher than a preset diagnostic threshold, then the candidate anomaly data is determined to be anomaly identification data.

5. The method according to claim 1, characterized in that, The feature sample library is obtained through the following methods: Acquire historical atmospheric monitoring data, which includes online monitoring value sequences corresponding to historical suspected anomaly data; Based on the online monitoring numerical sequence corresponding to the historical suspected anomaly data, mathematical characteristics characterizing the fluctuation pattern of the online monitoring numerical sequence are calculated; Analyze the historical audit logs associated with the abnormal period and extract the historical abnormal audit results. The historical abnormal audit results include the final validity status marked by manual identification and the manually entered abnormal reason diagnosis text. The online monitoring numerical sequences, validity status, and abnormal cause diagnostic texts retrieved under the same audit event are structured and associated, and stored in the feature sample library.

6. The method according to claim 1, characterized in that, The step of updating the feature sample library based on the anomaly identification data and the corresponding confirmation feedback results, and using the updated feature sample library to optimize the anomaly diagnosis classification model, includes: The anomaly identification data and corresponding associated feature parameters are structured and summarized to generate a list of anomalies to be verified. The list of anomalies to be verified is sent to the operation and maintenance interaction terminal to initiate a verification request. Receive the confirmation feedback result returned by the operation and maintenance interaction terminal for the anomaly identification data; Based on the confirmation feedback results, the abnormal identification data is classified and labeled, and the abnormal identification data and the corresponding labeling results are dynamically updated to the feature sample library; The system monitors the data increment and running cycle of the feature sample library. When the preset update conditions are met, it performs incremental training on the anomaly diagnosis classification model based on the updated feature sample library to optimize the model.

7. The method according to claim 6, characterized in that, The list of anomalies to be verified contains multiple abnormal data entries, and an auxiliary diagnostic card is generated for each of the abnormal data entries. The contents of the auxiliary diagnostic card include at least one of the following: The anomaly identification data and its concentration fluctuation curves in adjacent time periods; The category of the logic rule that triggers the review experience; Anomaly probability confidence level; The system prompts maintenance personnel to retrieve and compare chromatograms from on-site instruments for verification.

8. The method according to claim 6, characterized in that, The confirmation feedback result is either a retention instruction or a rejection instruction, wherein the rejection instruction indicates that the anomaly identification data is actual anomaly data, and the retention instruction indicates that the anomaly identification data is not actual anomaly data; and when the confirmation feedback result is the rejection instruction, the rejection instruction also includes an anomaly reason text. The method further includes: classifying and labeling the anomaly identification data based on the confirmation feedback result, and dynamically updating the anomaly identification data and the corresponding labeling results to the feature sample library, including: If the maintenance interaction terminal receives the retention instruction for the anomaly identification data, then the anomaly identification data is determined to be non-actual anomaly data, and the anomaly identification data is labeled as a negative sample. If the removal instruction with an anomaly reason text is received from the operation and maintenance interaction terminal, it is determined to be an instrument malfunction or integration error. The anomaly identification data and the corresponding anomaly reason text are then marked as positive samples. The anomaly reason text includes at least one of the following: chromatographic analysis and integration anomaly, instrument sampling and pipeline anomaly, and equipment operation and maintenance anomaly. The chromatographic analysis and integration anomaly includes: baseline interference, peak window drift, unknown contaminant interference, integration error, and incomplete peak integration. The instrument sampling and pipeline anomaly includes: quality control residue and incomplete sample gas sampling. The equipment operation and maintenance anomaly includes: instrument malfunction, standard calibration, and consumable replacement. Structured data with positive and negative sample labels are dynamically appended to the corresponding sub-library of the feature sample library.

9. The method according to claim 6, characterized in that, The preset update conditions include at least one of the following triggering mechanisms: a time cycle triggering mechanism and a data volume threshold triggering mechanism, wherein the time cycle triggering mechanism is: the system's continuous running time reaches a preset natural update cycle; and the data volume threshold triggering mechanism is: the total number of newly added positive and negative samples with confirmation labels in the feature sample library reaches a preset sample increment threshold.

10. An atmospheric monitoring data anomaly identification system, characterized in that, include: The data acquisition module is used to acquire atmospheric monitoring data to be identified. The rule base pre-filtering module is used to pre-filter the atmospheric monitoring data to be identified using a preset audit experience rule base to obtain candidate abnormal data; wherein, the audit experience rule base is a set of logical rules corresponding to non-emission anomalies obtained from long-term human audit experience. The classification and identification module is used to identify abnormal identification data from the candidate abnormal data using an anomaly diagnosis and classification model; wherein, the anomaly diagnosis and classification model is based on a feature sample library, which includes historical atmospheric monitoring data and corresponding historical anomaly review results; The confirmation update module is used to update the feature sample library based on the anomaly identification data and the confirmation feedback results corresponding to the anomaly identification data. The updated feature sample library is used to optimize the anomaly diagnosis classification model.

11. A computer program product, characterized in that, It includes a computer program / instruction, which, when executed by a processor, implements the atmospheric monitoring data anomaly identification method according to any one of claims 1 to 9.

Images