A multi-level serial intelligent auditing method and system for environmental monitoring data and a storage medium

Abstract

This invention proposes a multi-level serial intelligent auditing method, system, and storage medium for environmental monitoring data. The method includes: acquiring multiple environmental data streams from a distributed environmental monitoring network; filtering all environmental data streams to select data carrying a backfill identifier as valid candidate data; performing parallel anomaly detection on the valid candidate data across multiple dimensions to generate a multi-dimensional anomaly feature vector; wherein the multiple dimensions include at least anomalies in the data's numerical characteristics, equipment operating status, station environment, impact of maintenance activities, and impact of equipment anomaly recovery period; inputting the multi-dimensional anomaly feature vector into a preset multi-level decision rule tree; the multi-level decision rule tree performs fusion reasoning based on different combinations of anomaly features across dimensions and preset decision logic to output an audit conclusion. This invention achieves efficient and accurate auditing of backfilled data, significantly reducing the proportion of manual spot checks and the rate of missed and false detections.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a multi-level serial intelligent auditing method, system, and storage medium for environmental monitoring data. Background Technology

[0002] In the field of environmental data auditing, data loss frequently occurs due to sensor malfunctions, communication anomalies, and other reasons. Existing technologies typically identify missing data and then use interpolation and model prediction to recover it. However, these methods only focus on how to generate the recovered data and generally lack effective auditing mechanisms for the recovered data. Whether the recovered data accurately reflects the environmental state, whether it logically conflicts with adjacent data, and whether it introduces significant biases are all subject to a lack of objective and automated judgment.

[0003] Currently, the quality verification of supplementary data still relies heavily on manual spot checks, resulting in inconsistent review standards and low efficiency, making it difficult to meet the real-time processing needs of large-scale environmental data streams. While rule engines can be used to verify some routine data, they struggle to establish accurate review logic for supplementary data in missing scenarios, leading to high rates of missed and false detections, which makes it difficult to guarantee the overall reliability of environmental data. Summary of the Invention

[0004] This invention provides a multi-level serial intelligent auditing method, system, and storage medium for environmental monitoring data to address the problems existing in related technologies. The technical solution is as follows: In a first aspect, embodiments of the present invention provide a multi-level serial intelligent auditing method for environmental monitoring data, including: Multiple environmental data streams from a distributed environmental monitoring network are acquired, and all environmental data streams are filtered to select data carrying a backfill identifier as valid candidate data. Multiple dimensions of anomaly detection are performed in parallel on valid candidate data to generate multi-dimensional anomaly feature vectors. Among them, multiple dimensions include at least the numerical feature anomalies of the data itself, the anomalies of equipment operation status, the anomalies of the station environment, the impact of operation and maintenance activities, and the impact of equipment anomaly recovery period. The multi-dimensional abnormal feature vectors are input into a preset multi-level decision rule tree. The multi-level decision rule tree performs fusion reasoning based on different combinations of abnormal features of each dimension and preset decision logic, and outputs the audit conclusion.

[0005] In one implementation, data filtering of all environmental data streams includes: Based on a pre-defined missing pattern library, the environmental data stream is classified into missing data categories to identify missing data. Missing data is imputed, imputed data is marked with an imputed flag, and data with the imputed flag are selected as valid candidate data.

[0006] In one implementation, the detection of numerical anomalies in the data itself includes multiple anomaly models based on numerical logic, which include at least a zero negative value model, a data constant value model, a general outlier model, an ozone feature model, and a particulate matter feature model.

[0007] In one implementation, the detection of abnormal equipment operation status is achieved by associating with instrument status information, which includes at least one or more of the following: instrument status parameters, quality control records, fault work orders, and real-time alarm information.

[0008] In one implementation, the detection of station building environmental anomalies is achieved based on abnormal signals from sampling equipment and the assessment of station building dynamic environmental parameters, including temperature and humidity.

[0009] In one implementation, the impact of maintenance activities is detected by matching data timestamps with the entry-to-exit time windows in maintenance work orders and access control records to mark the periods of known human activity impact.

[0010] In one implementation, the detection of the impact of equipment abnormality recovery period is used to identify power outage recovery or equipment startup events, and to mark the data within a preset stable period after the event occurs.

[0011] In one implementation, a multi-level decision rule tree performs fusion reasoning based on different combinations of abnormal features across various dimensions and preset decision logic, outputting audit conclusions including: If the multi-dimensional anomaly feature vector contains zero negative value detection features that are less than the rounding lower limit, then the conclusion of invalid data review will be output. If the multi-dimensional anomaly feature vector contains zero negative value detection features within the rounding interval, and does not contain any of the features of abnormal equipment operation status, abnormal station environment, impact of operation and maintenance activities, and impact of equipment anomaly recovery period, then the audit conclusion of the data to be rounded will be output. If the multi-dimensional anomaly feature vector contains zero negative value detection features within the rounding interval, and contains at least one of the features of abnormal equipment operation status, abnormal station environment, impact of operation and maintenance activities, and impact of equipment anomaly recovery period, then the audit conclusion of invalid data will be output. If the multi-dimensional abnormal feature vector contains other numerical abnormal features besides zero and negative values, and does not contain any of the features of abnormal equipment operation status, abnormal station environment, impact of operation and maintenance activities, and impact of equipment abnormality recovery period, then the audit conclusion of the reminder data that needs to be manually reviewed will be output. If the multi-dimensional abnormal feature vector contains other numerical abnormal features besides zero and negative values, and contains at least one of the following features: abnormal equipment operation status, abnormal station environment, impact of operation and maintenance activities, and impact of equipment abnormality recovery period, then the audit conclusion of invalid data will be output.

[0012] Secondly, embodiments of the present invention provide a multi-level serial intelligent auditing system for environmental monitoring data, which executes the multi-level serial intelligent auditing method for environmental monitoring data as described above; the system includes: The data acquisition and processing module is used to acquire various environmental data streams from the distributed environmental monitoring network, and to filter all environmental data streams to select data carrying a backfill identifier as valid candidate data. The multi-dimensional anomaly detection module is used to perform anomaly detection on valid candidate data in parallel across multiple dimensions, generating a multi-dimensional anomaly feature vector. Among these dimensions, at least the numerical features of the data itself are abnormal, the equipment operating status is abnormal, the station environment is abnormal, the impact of operation and maintenance activities is abnormal, and the impact of the equipment anomaly recovery period is abnormal. The decision fusion module has a built-in multi-level decision rule tree, which is used to input multi-dimensional abnormal feature vectors into the preset multi-level decision rule tree. The multi-level decision rule tree performs fusion reasoning based on different combinations of abnormal features of each dimension and preset decision logic, and outputs the review conclusion.

[0013] Thirdly, embodiments of the present invention provide a computer-readable storage medium that stores a computer program, wherein when the computer program is run on a computer, the methods in any of the above-described embodiments are executed.

[0014] The advantages or beneficial effects of the above technical solutions include at least the following: This invention aggregates multiple time-series environmental data streams from a distributed monitoring network. First, it checks and replenishes missing data streams from all sites, adding a replenishment identifier to the replenished data. This filters out candidate data requiring in-depth review, avoiding unnecessary computational overhead on all normal data. For data with replenishment identifiers, this invention performs parallel multi-factor anomaly detection, comprehensively evaluating data quality from multiple dimensions including numerical characteristics of the data itself, equipment operating status, station environment, impact of maintenance activities, and impact of equipment anomaly recovery period, outputting a multi-dimensional anomaly feature vector for each data point. Finally, based on different combinations and priorities of various anomaly factors in a multi-level decision rule tree, it performs fusion reasoning on the multi-dimensional anomaly feature vectors, ultimately outputting a standardized review conclusion.

[0015] Through the aforementioned cascaded screening and multi-dimensional fusion reasoning mechanism, this invention fills the gap in the existing technology regarding the lack of automated review of missing data, especially supplemented data. It achieves efficient and accurate review of supplemented data, significantly reduces the proportion of manual sampling and the rate of missed and false detections, and avoids the waste of computing power caused by unified review of all data. Overall, it improves the automation level, processing efficiency and reliability of environmental data review.

[0016] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0017] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments disclosed in the invention and should not be construed as limiting the scope of the invention.

[0018] Figure 1 This is a flowchart illustrating the multi-level serial intelligent auditing method of the present invention; Figure 2 This is a system architecture block diagram of the multi-level serial intelligent auditing method of the present invention; Figure 3 This is a schematic diagram of the multi-factor abnormality parallel diagnosis steps of the present invention; Figure 4 This is a logical diagram of the multi-level decision rule tree of the present invention; Figure 5 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0019] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0020] Example 1 This embodiment provides a multi-level serial intelligent review method for environmental monitoring data. This method can replace the traditional manual review method, improving review efficiency and the accuracy and reliability of review results.

[0021] refer to Figure 1 As shown, the multi-level serial intelligent auditing method in this embodiment specifically includes the following steps: Step S1: Obtain multiple environmental data streams from the distributed environmental monitoring network, filter all environmental data streams, and select data carrying a backfill identifier as valid candidate data.

[0022] This embodiment aggregates multiple time-series environmental data streams from a distributed environmental monitoring network, including concentration data streams of monitored factors (SO2, CO, NO2, O3, PM2.5). 2.5 PM 10The data includes: status parameter streams of instruments and equipment (such as UV lamp intensity, reaction chamber temperature, reaction chamber pressure, etc.), dynamic parameter streams of the station environment (such as station temperature, station humidity, etc.), and business record streams characterizing external events (such as alarm, power outage, operation and maintenance, quality control, etc.).

[0023] Based on a pre-defined missing pattern library, the system checks and fills in missing data in all environmental data streams, adds a filling mark to the filled data, and completes the first level of data filtering.

[0024] Specifically, the missing pattern library is a predefined set of rules used to classify and identify various types of missing data in environmental data streams. This library is built upon historical data statistical analysis and equipment operation mechanisms. Each missing pattern record contains a set of missing feature conditions and its corresponding missing type label. The condition set for each missing pattern consists of multiple judgment dimensions, including but not limited to: the name and number of missing parameters (e.g., a single parameter continuously missing or multiple parameters simultaneously missing), the continuous length of the missing data on the timeline (e.g., continuous missing for more than 6 hours), the regularity of the time period in which the missing data occurs (e.g., consistently occurring at night or during quality control periods), and whether other related parameters are normal during the missing period.

[0025] During actual matching, the system extracts statistical features of the data stream to be detected by sliding across a time window, such as the missing rate of each parameter, the length of the longest consecutive missing segment, and the synchronization of missing parameters. These features are then compared sequentially with each condition group in the missing pattern library. Only when all preset conditions are met does the system determine that the current data segment matches the missing pattern and output the corresponding missing type label (completely missing or partially missing) and more specific subcategories, such as "completely missing single parameter" or "partially missing multiple parameters followed by recovery." Through this line-by-line comparison, the missing pattern library can transform the complex missing phenomena in the original environmental data stream into structured, computable missing classification results, thus providing a basis for subsequent data imputation and candidate data selection.

[0026] It should be noted that "complete absence" refers to a situation where, within a specific hour, the six conventional pollutant parameters (i.e., SO2, NO2, CO, O3, PM2.5, and PM10) are all absent. 10 PM 2.5The first type of missing data is a state where all hourly concentration values ​​are missing, and all 5-minute raw data constituting that hourly value are also missing. This state indicates that the monitoring equipment has completely stopped working or that data acquisition and transmission have been completely interrupted during that period, and can be automatically associated with the cause of "site power outage". The second type of missing data refers to other data missing situations besides complete missing data, including but not limited to: missing hourly values ​​of some of the six parameters, missing 5-minute raw data of a certain parameter within an hour, or inconsistent missing data of various parameters. This state usually reflects temporary equipment abnormalities, communication interruptions, or single-parameter sensor failures, and can be automatically associated with the cause of "network transmission failure".

[0027] After classifying the missing data status, the next step is to intelligently replenish the missing data and add labels. Specifically, based on the missing data classification results, the system automatically or semi-automatically sends replenishment instructions to the data acquisition device or data management system to replenish the identified missing data. A unified replenishment label is added to the replenished data, ultimately generating a dataset with preliminary cause tags. This dataset includes data with replenishment labels, data with maintenance labels, or other labels, etc., for use in subsequent anomaly detection and review processes.

[0028] The generation and issuance of data replenishment instructions are based on a preset replenishment triggering mechanism. When the data loss status is determined to be "partially missing" or meets the specific loss type of the replenishment strategy, the system automatically generates a replenishment request based on the time window of the loss, the type of missing parameters, and the importance level of the data. This request can be issued to the front-end data acquisition equipment, data aggregation platform, or third-party data replenishment system via an interface protocol to trigger the data replenishment process. The replenishment instruction includes at least the target site identifier, the time range of the missing data, the list of pollutant parameters to be replenished, the replenishment priority, and the replenishment method (such as historical mean interpolation, neighboring site correlation interpolation, model prediction interpolation, or manual data entry).

[0029] Upon receiving a data recovery instruction, the data acquisition equipment or data management system executes the corresponding data recovery operation. The sources of the recovered data include, but are not limited to, historical data cached locally on the device, backup data from the central database, interpolated values ​​based on spatiotemporal correlation algorithms, and manually entered supplementary test data or quality control data by maintenance personnel. After the recovery operation is completed, time-series data with the same format as the original data is generated, filling in the original missing intervals.

[0030] To distinguish it from the original collected data, all data generated through the backfilling operation must be labeled with a unified backfilling identifier. This backfilling identifier, acting as a data quality label, is appended to the backfilled data in the form of metadata fields, status flags, or packet header information. The backfilling identifier can include the following information dimensions: Data recovery status: Mark this data as generated during data recovery; Data interpolation method: Record the source or algorithm type of data interpolation (such as historical interpolation, model prediction, manual entry, etc.); Recover timestamp: Records the specific time of the recovery operation; Backfill confidence: An optional field used to identify the confidence level of the backfilled data.

[0031] Subsequently, each hourly data point in the dataset with preliminary cause markers is parsed to automatically identify predefined identifiers. Predefined identifiers are status marker fields added by equipment, systems, or personnel during data acquisition, transmission, quality control, or preprocessing to characterize the specific source, quality status, or associated events of the data.

[0032] According to the identifier type rule base, identifier data that meets preset conditions is automatically reviewed as invalid data, completing the second level of data filtering. Specifically, the identifier type rule base is a pre-established set of mapping relationships used to define the correspondence between various predefined identifiers and review conclusions. The identifier type rule base includes at least the following two types of rules: Invalid Identification Rules: Define the identification types that can be directly determined as invalid data. When hourly data carries an invalid identification, no further anomaly detection is required; it directly enters the invalid data channel. Examples include PZ (Equipment Failure), PS (Equipment Maintenance), AS (Gas Circuit Maintenance), CZ (Operational Error), CS (Calibration), TSL (Test), PM (Particulate Matter Maintenance), and PF (Particulate Matter Failure). Hourly data carrying these identifications indicates an abnormal data acquisition environment or equipment status; its data quality cannot be guaranteed and is therefore directly determined as invalid data.

[0033] Valid Identification Rules: Define the identification types that can be determined as valid candidate data. When hourly data carries a valid identification, it indicates that although the data is generated manually or by the system, it possesses reliable data quality and can be used as a valid data candidate output. For example, hourly data carrying a backfilling identification is determined to be valid candidate data. After being determined to be valid candidate data, the backfilling identification of the data is canceled (i.e., the backfilling identification is removed from the metadata field), restoring it to the normal data format. The data after the identification is canceled is released to the next stage to participate in subsequent multi-dimensional anomaly detection or multi-level decision fusion processes.

[0034] The second-level data filtering step can intercept a large amount of data carrying clear invalid identifiers at an early stage, preventing it from entering the subsequent complex multi-dimensional anomaly detection process and significantly reducing the consumption of computing resources.

[0035] Step S2: Perform multi-dimensional anomaly detection on the valid candidate data in parallel to generate a multi-dimensional anomaly feature vector; wherein, the multiple dimensions include at least the numerical feature anomalies of the data itself, the anomalies of the equipment operating status, the anomalies of the station environment, the impact of operation and maintenance activities, and the impact of the equipment anomaly recovery period.

[0036] The remaining data after the second-level data filtering is then input in batches into a multi-factor anomaly diagnostic model, such as... Figure 3 As shown, the multi-factor anomaly diagnosis model performs at least five dimensions of anomaly detection in parallel for each input data point, including anomalies in the numerical characteristics of the data itself, anomalies in equipment operating status, anomalies in the station environment, the impact of maintenance activities, and the impact of equipment anomaly recovery period. The multi-factor anomaly diagnosis model outputs one or more anomaly flags or quantized scores for each dimension, and these outputs are concatenated into a fixed-dimensional numerical vector, i.e., a multi-dimensional anomaly feature vector. Specifically: The detection of numerical anomalies in the data itself is performed by multiple parallel or serial anomaly models. Each model receives one or more consecutive environmental data values ​​as input, performs calculations according to its own preset judgment logic, and finally outputs the corresponding anomaly flag or subtype code. The specific execution method is as follows: The zero-negative-value model reads the current data value one by one, first determining whether the data value is less than or equal to zero. If it is greater than zero, both flag bits are directly output as zero, indicating no non-zero negative value anomalies. If it is less than or equal to zero, the current data value is further compared with a preset rounding lower limit threshold. This rounding lower limit threshold is preset according to the parameter type; for example, for environmental concentration parameters, it can be set to a negative instrument detection lower limit. If the current data value is less than the rounding lower limit threshold, the less-than-rounding lower limit flag is set to one, and the flag within the rounding interval is set to zero; if the current data value is between the rounding lower limit threshold and zero (inclusive), the less-than-rounding lower limit flag is set to zero, and the flag within the rounding interval is set to one.

[0037] The constant value model scans continuous data points using a sliding window. The window length is set to a number of consecutive sampling points, such as six consecutive hourly values. The system extracts all valid data values ​​within the current window and calculates their standard deviation. If the standard deviation is less than a preset constant threshold, and all data points within the window are not empty, it is determined to be a constant value anomaly, and the constant value flag is set to 1; otherwise, the constant value flag is set to zero. The window slides one step at a time, such as one sampling interval, and the flag is output independently for each window.

[0038] A variety of statistical methods can be used to perform general outlier models. Taking the three-standard-deviation method as an example, for each data point to be detected, several historical valid data points are taken before that data point, such as the same parameter value in the past 24 hours, and the mean and standard deviation of these historical data are calculated. Then, it is determined whether the absolute value of the difference between the current data point and the mean is greater than three standard deviations. If it is greater, the outlier flag is set to 1, and the outlier score is calculated as the ratio of the current difference to three standard deviations, but with an upper limit of 1; if it is not greater, the outlier flag is set to 0, and the outlier score is zero. The algorithm can also be replaced by a method based on a global percentile threshold, which identifies points below the first percentile or above the ninety-ninth percentile as outliers, with the output format remaining unchanged.

[0039] The ozone characteristic model first constructs a daily variation expectation range based on historical data. It collects effective ozone concentration values ​​for the same hour over the past thirty days, calculates the mean and standard deviation for each hour, and sets the expectation range to the mean plus or minus 1.96 times the standard deviation. During detection, the current hourly ozone concentration value and the current hour index are input. The model determines whether the current concentration value is outside the expected range for that hour and significantly higher than the upper limit of the range, for example, exceeding the upper limit by more than 20%. If these conditions are met, an ozone anomaly flag of 1 is output; otherwise, 0 is output. Furthermore, if an abnormal increase occurs during nighttime, even if it does not exceed the upper limit of the range, it can be directly marked as abnormal according to preset rules.

[0040] The particulate matter characteristic model establishes a diurnal rhythm baseline curve by analyzing historical data, reflecting the typical daily variation pattern of particulate matter concentration. During detection, the current hourly value is compared with the baseline range for the corresponding time period; if the deviation exceeds a dynamic threshold, it is judged as an anomaly. This method can effectively distinguish between real pollution events and anomalies caused by instrument malfunctions.

[0041] After each of the above models has been executed independently, its output Boolean flag and the outlier score output by the general outlier model are concatenated in a fixed order to form a subvector representing the numerical anomaly dimension of the data itself.

[0042] In addition, the detection of abnormal equipment operating status is achieved by associating it with instrument status information. This involves real-time acquisition of instrument status information, including instrument status parameters, quality control records, fault work orders, and real-time alarm information; each parameter value is then compared item by item with the preset normal operating range. If any parameter exceeds the normal range, the equipment is immediately marked as abnormal. Simultaneously, the system searches the quality control record table to determine if there are any non-compliant quality control events near the current data time, such as calibration failures or excessive calibration deviations; if such events exist and the time difference is within the preset allowable range, the equipment is also marked as abnormal. Furthermore, the system queries the fault work order table and the real-time alarm information table to check if the current data timestamp falls within the start and end time of any fault work order record, or if any alarm code is matched. If any of these conditions are met, the output flag for the abnormal equipment operating status dimension is set to one; otherwise, it is set to zero.

[0043] The method for detecting anomalies in the station building environment involves reading abnormal signals from sampling equipment and station building dynamic environmental parameters, including at least temperature, humidity, and voltage. For continuous parameters such as temperature and humidity, the system checks multiple consecutive sampling points in a time sequence. If several consecutive points exceed a preset threshold range (e.g., temperature above 30 degrees Celsius), an environmental anomaly is determined. For sampling equipment status signals, such as pump operating status, pipeline pressure signals, and sampling flow alarm signals, if abnormal status signals such as shutdown, blockage, or leakage are detected, an environmental anomaly is directly determined. If any of the above conditions are met, the output flag for the station building environment anomaly dimension is set to one; otherwise, it is set to zero.

[0044] The method for detecting the impact of operations and maintenance activities involves extracting the planned start and end times recorded in the operations and maintenance work orders, as well as the entry and exit times of personnel in the access control system, to construct time windows for personnel presence. The union of these two timestamps is then taken as the final effective impact period. Next, the timestamp of the data to be detected is intersected with this impact period. If the data timestamp falls within the impact period, the data is determined to be affected by the operations and maintenance activity, and a flag is set to one; otherwise, it is set to zero.

[0045] The method for detecting the impact dimensions of equipment anomaly recovery periods involves monitoring the equipment status event logs to capture power outage recovery events and equipment startup events, recording the occurrence time of each event. For each recovery event, the system counts back a preset stable duration, such as 30 minutes, from the event's occurrence time, marking all data points within this stable duration as the recovery period. The system automatically generates multiple recovery period intervals using a time range coverage approach. When the current data timestamp is input, the system checks if the timestamp falls within any recovery period interval; if so, the output flag is set to one; otherwise, it is set to zero.

[0046] To further improve the accuracy of anomaly diagnosis, this embodiment can also introduce a site-zone collaborative analysis mechanism. First, based on the correlation between the geographical distance between sites and the trend of pollutant concentration changes, monitoring sites are divided into the same area (e.g., using clustering algorithms or spatial correlation analysis). During multi-dimensional anomaly detection, collaborative comparison between sites within the same area is introduced: if data from a certain site shows an anomaly (e.g., PM2.5 concentration),... 2.5 If an anomaly is detected at a single site, and other sites within the same area show no abnormalities, the system tends to determine that the anomaly originates from a device malfunction or localized environmental interference, rather than a regional environmental change. This mechanism effectively reduces misjudgments caused by occasional fluctuations at a single site and improves the reliability of diagnostic results.

[0047] After the above-mentioned multi-dimensional anomaly detection is executed in parallel, the multi-factor anomaly diagnosis model generates a multi-dimensional anomaly feature vector for each input data. This vector contains the detection results of each dimension, such as the anomaly type (zero negative value - less than the rounding lower limit / zero negative value - within the rounding interval / constant value / outlier / no anomaly), equipment anomaly flag (yes / no), station building anomaly flag (yes / no), operation and maintenance impact flag (yes / no), and recovery period flag (yes / no), etc. This multi-dimensional anomaly feature vector serves as the input to the subsequent multi-level decision rule tree, used for fusion reasoning to arrive at the final review conclusion.

[0048] Step S3: Input the multi-dimensional abnormal feature vector into the preset multi-level decision rule tree. The multi-level decision rule tree performs fusion reasoning based on different combinations of abnormal features of each dimension and the preset decision logic, and outputs the audit conclusion.

[0049] In the specific execution of the multi-level decision rule tree, the multi-dimensional anomaly feature vector is first parsed into multiple independent Boolean flags and numerical features, specifically including: flags for values ​​less than the rounding lower limit, flags for values ​​within the rounding interval, flags for constant data values, general outlier flags and outlier scores, ozone anomaly flags, particulate matter anomaly flags, equipment operating status anomaly flags, station environment anomaly flags, maintenance activity impact flags, and equipment anomaly recovery period impact flags. These flags collectively constitute the feature set input to the rule tree. The first six flags belong to the numerical anomaly dimension of the data itself, while the last four flags correspond to the four dimensions of equipment operating status anomaly, station environment anomaly, maintenance activity impact, and equipment anomaly recovery period impact, respectively.

[0050] Multi-level decision rule trees employ a hierarchical condition matching mechanism, such as... Figure 4 As shown, the input features are judged in descending order of preset priority.

[0051] First priority: Determination of data less than the rounding lower limit. The system first checks if the flag indicating data is less than the rounding lower limit is true. If true, regardless of the values ​​of other flags, the reasoning process terminates directly, and the conclusion "Data deemed invalid" is output. This rule corresponds to Rule 1 and has the highest priority.

[0052] Second priority: Zero negative value determination within the rounding interval. If the flag indicating values ​​less than the lower rounding limit is false, the system checks if the flag indicating values ​​within the rounding interval is true. If true, it further checks if at least one of the following four flags is true: equipment operation status abnormality flag, station environment abnormality flag, maintenance activity impact flag, and equipment abnormality recovery period impact flag. If at least one flag is true, it matches rule three and outputs "Data deemed invalid"; if all four flags are false, it matches rule two and outputs "Data requiring rounding".

[0053] Third priority: Judgment of other numerical feature anomalies. If the flag is also false within the rounding interval, the system checks whether at least one other numerical feature anomaly exists. Other numerical feature anomalies include: any of the following: data constant value flag is true, general outlier flag is true, ozone anomaly flag is true, and particulate matter anomaly flag is true. If at least one is true, the system further checks whether at least one of the aforementioned four environmental and equipment anomaly flags is true. If at least one is true, rule five is matched, and "Data invalid" is output; if all four flags are false, rule four is matched, and "Reminder" is output.

[0054] Fourth Priority: Environmental and Equipment Anomaly Judgment Only. If all the above numerical anomaly flags (including less than the rounding lower limit, within the rounding interval, constant value, outlier, ozone anomaly, and particulate matter anomaly) are false, the system will proceed to judge the four environmental and equipment anomaly flags. Specifically, the system checks whether at least one of the equipment operation status anomaly flag, station environment anomaly flag, maintenance activity impact flag, and equipment anomaly recovery period impact flag is true. If at least one is true, the system outputs a "Reminder" audit conclusion, indicating that the data itself is numerically acceptable but external conditions are abnormal and require manual review; if all four flags are false, the system outputs a "Reviewed as Valid Data" conclusion.

[0055] Fifth priority: Default valid judgment. All the above branches cover all possible feature combinations, and there are no uncovered cases. For data that skips all exception checks and has all four environment and device class flags as false, it has fallen into the fourth priority's all-false branch, and valid data is output.

[0056] Through the aforementioned step-by-step matching process, the multi-level decision rule tree fuses and infers anomaly features across five dimensions, achieving a complete mapping from multi-dimensional anomaly feature vectors to standardized review conclusions. Each branch of the rule tree corresponds to a specific feature combination condition, and the priority order is pre-determined by the importance of the specific rule, avoiding misjudgment based on a single dimension and improving review accuracy. In practical applications, the rule tree's judgment logic can be implemented through configuration files or a rule engine, supporting dynamic addition and deletion of rules and adjustment of priorities to meet the needs of different data review scenarios.

[0057] For example, this embodiment performs batch review of hourly monitoring data from 40 air quality monitoring stations in a certain city on a certain day. The system architecture is as follows: Figure 2 As shown, the specific process is as follows: The first step is to collect hourly and 5-minute monitoring data for all factors from all sites on that day, as well as maintenance work order records, access control records, alarm records, instrument status parameters, etc. Table 1 below shows some hourly monitoring data.

[0058] Table 1 Monitoring data of all stations on the same day Site time <![CDATA[PM 2.5 ]]> <![CDATA[PM 10 ]]> <![CDATA[SO2]]> <![CDATA[NO2]]> <![CDATA[O3]]> CO XX site 1 o'clock 6 28 6 14 65 0.7 XX site 2 o'clock 6 30 5 12 71 0.7 XX site 3 o'clock 6 33 5 11 73 0.6 XX site 4 o'clock 6 46 5 12 70 0.7 XX site 5 o'clock 6 28 5 13 68 0.7 XX site 6 o'clock 6 46 6 12 70 0.7 XX site 7 o'clock 5 51 6 13 70 0.7 XX site 8 o'clock 5 67 7 14 66 0.7 XX site 9 o'clock 6 89 8 16 60 0.8 XX site 10:00 6 65 7 16 61 0.8 XX site 11:00 8 102 6 10 75 0.5 XX site 12 o'clock 10 143 6 10 83 0.5 XX site 13:00 14 129 10 12 84 0.5 XX site 2 PM 18 123 9 12 90 0.5 XX site 3 PM 19 108 7 12 96 0.5 XX site 16:00 21 108 9 13 100 0.5 XX site 5 PM 22 102 8 14 101 0.5 XX site 6 PM 22 94 8 15 99 0.5 XX site 7 PM 23 72 7 17 95 0.4 XX site 8 PM 22 77 7 18 92 0.4 XX site 21:00 22 80 7 19 87 0.4 XX site 22:00 21 84 6 18 84 0.4 XX site 11 PM 20 93 7 18 82 0.5 XX site 0 o'clock 19 74 6 18 77 0.5 Step 2: Data Integrity Check. Check for missing data in the hourly and 5-minute data for all stations. If any data is missing, immediately initiate the data recovery process. The recovered data will be marked with a recovery tag. Once recovery is complete, proceed to the next step. If no data is missing, proceed to the next step.

[0059] Step 3: Data Attribute Check. Hourly data from all sites undergoes an identifier check. For hourly data with maintenance identifiers (PZ, PS, AS, CZ, CS, TSL, PM, PF) or other identifiers (BB, B, Hsp, Lsp, H), the system preprocessing result is "Reviewed as Invalid Data," and an invalid (RM) label is applied. For hourly data with a recovery identifier (Re), the system preprocessing result is "Reviewed as Valid Data," and the label is removed. After completing the identifier preprocessing, invalid data is filtered out, and the process proceeds to the next step.

[0060] Step 4: Data accuracy check. The hourly data for all sites is checked for accuracy, and parallel diagnostics are performed for feature anomalies, equipment anomalies, station building anomalies, maintenance quality control period anomalies, and data recovery period anomalies. The core algorithms for each anomaly diagnosis are described below: 1. Feature Anomaly Diagnosis: This includes zero negative value - exceeding the upper limit detection, zero negative value - within the rounding interval detection, data constant value detection, and data outlier detection. Taking the "single-site sudden high outlier algorithm" in the data outlier detection set as an example, its judgment logic is as follows: 1) Check each hourly data point to see if it exceeds the preset upper limit parameter x (x is set based on the historical emission characteristics of regional environmental pollutants, such as SO2, CO, NO2, O3, PM2.5). 2.5 PM 10 The values ​​can be set to 85, 8, 800, 400, 500, and 500 respectively. If the values ​​exceed these values, the entry is marked as "outlier"; otherwise, proceed to the next step.

[0061] 2) Determine if the current hourly value is greater than the preset threshold a (set the corresponding parameter values ​​according to the historical emission characteristics of regional environmental pollutants, such as SO2, CO, NO2, O3, PM). 2.5 PM 10 The values ​​can be set to 30, 2, 30, 80, 75, and 75 respectively. If 'Yes', proceed to the next step; if 'No', end the judgment.

[0062] 3) Calculate the relative deviation between the current hour and the data from the previous n hours (n can be set to 2), using the following formula: Relative deviation = (current hourly value - previous n-hourly value) / previous n-hourly value; Determine if these n relative deviations have the same sign. If 'yes', proceed to the next step; if 'no', end the process.

[0063] 4) Determine whether all n relative deviations are greater than the peak threshold b (set corresponding parameter values ​​based on the historical emission characteristics of regional environmental pollutants, such as SO2, CO, NO2, O3, PM2.5). 2.5 PM 10 The values ​​can be set to 5, 2, 4.5, 2.5, 3, 3 respectively. If so, mark it as "outlier"; otherwise, end the judgment.

[0064] If there are missing data or the data has been identified as abnormal by exceeding the upper limit detection in the previous n hours, the values ​​should be extended forward accordingly; for the data at 1:00 AM and 2:00 AM, the data of the corresponding time period of the previous day should be used for compensation.

[0065] 2. Equipment Anomaly Diagnosis: Based on abnormal instrument status parameters, quality control non-compliance records, instrument fault handling records, and instrument alarm information. Taking abnormal instrument status parameters as an example, the judgment logic is as follows: 1) Retrieve alarm information for the day the data under review is being reviewed, and filter out alarm records in the category of "instrument status exceeds upper and lower limits".

[0066] 2) Based on the alarm time, mark the hourly data of the corresponding time period as "abnormal instrument status parameter". For example, if an alarm for a certain status parameter exceeding the upper or lower limit occurs at 12:13 on the same day, then mark the hourly data at 13:00 as abnormal, and the abnormality type is "abnormal instrument status parameter".

[0067] 3) Record the following fields for the tagged data: anomaly type, anomaly cause, alarm time, alarm device name, name of status parameter exceeding the upper limit, corresponding upper and lower limit values, and hourly value of the status parameter. This information provides objective data support for subsequent anomaly cause tracing and judgment.

[0068] 3. Station building anomaly diagnosis: This includes sampling equipment anomalies and station building environment anomalies. Taking sampling equipment anomalies as an example, the judgment logic is as follows: 1) Retrieve the data acquisition alarm records for the day the data is reviewed, and filter out the alarm types related to the sampling equipment, including but not limited to: abnormal sampling manifold temperature, abnormal sampling manifold humidity, abnormal manifold static pressure, sudden change in manifold static pressure, insufficient heating, heating temperature difference, sample gas temperature exceeding the limit, sample gas humidity exceeding the limit, abnormal sampling flow rate, excessive sampling residence time, sampling tube blockage, backflow of air into the sampling tube, abnormal heating, etc.

[0069] 2) Mark all factor hour data corresponding to the alarm time as "sampling equipment malfunction". For example, if a sampling main tube temperature alarm is generated at 12:23 on the same day, then mark the SO2, CO, NO2, and O3 data at 13:00 as "sampling equipment malfunction". 3) Record the following fields for the marked data: anomaly type, anomaly cause, alarm time, and alarm content, to provide objective data support for subsequent anomaly cause tracing and judgment basis display.

[0070] 4. Operation and Maintenance Quality Control Period Diagnosis: Equipment maintenance may interfere with monitoring data, leading to inaccurate data during maintenance. The judgment logic is as follows: 1) Search for records where there is an operation and maintenance work order on the same day and the "start time filled in by operation and maintenance" is on the same day.

[0071] 2) Retrieve the access control records for the day the data is reviewed.

[0072] 3) If both maintenance work orders and access control records exist on the same day, the following judgments will be made based on the time from when the access control door opens to when the door opens to when the door opens again: Let the time of entry be T hours and XX minutes. The threshold for SO2 inspection task in the maintenance work order is 45. If XX > 45, the data at time T does not need to be marked; if XX ≤ 45, the SO2 hourly data at (T+1) hours needs to be marked as "equipment maintenance".

[0073] Let the time of opening and leaving be Z hours and YY minutes. Today's maintenance work order inspection task is SO2. If YY < 15 hours, then the data at Z hours does not need to be marked; if YY ≥ 15 hours, mark the SO2 hour at (Z+1) hours as "equipment maintenance".

[0074] 4) Record the following fields for the marked data: anomaly type, anomaly cause, maintenance equipment name, maintenance work order number, entry time into the station, and exit time into the station, to provide data support for subsequent anomaly cause tracing and judgment basis display.

[0075] 5. Data recovery period: This includes the quality control recovery period, the operation and maintenance recovery period, and the power outage recovery period. Taking the power outage recovery period as an example, the judgment logic is as follows: 1) Retrieve the power outage records for the day the data is audited.

[0076] 2) Mark all factor hour data corresponding to the power outage end time and the data of the next hour as "power outage recovery period". For example, if a power outage occurs from 11:01 to 13:20 on the same day, mark all factor data at 14:00 and 15:00 as "power outage recovery period".

[0077] 3) Record the following fields for the marked data: anomaly type, anomaly cause, power outage start time, and power outage end time, to provide data support for subsequent anomaly cause tracing and judgment basis display.

[0078] Step 5: Based on the preset multi-level decision rule tree, perform fusion reasoning on the abnormal data and output the preprocessed results. The core logic of the decision rule tree follows rules one to five defined in the technical solution, combines and judges abnormal features of each dimension, and finally outputs standardized review conclusions.

[0079] Example 2 This embodiment provides a multi-level serial intelligent auditing system for environmental monitoring data. The system executes the multi-level serial intelligent auditing method for environmental monitoring data as described in Embodiment 1. The system includes: The data acquisition and processing module is used to acquire various environmental data streams from the distributed environmental monitoring network, and to filter all environmental data streams to select data carrying a backfill identifier as valid candidate data. The multi-dimensional anomaly detection module is used to perform anomaly detection on valid candidate data in parallel across multiple dimensions, generating a multi-dimensional anomaly feature vector. Among these dimensions, at least the numerical features of the data itself are abnormal, the equipment operating status is abnormal, the station environment is abnormal, the impact of operation and maintenance activities is abnormal, and the impact of the equipment anomaly recovery period is abnormal. The decision fusion module has a built-in multi-level decision rule tree, which is used to input multi-dimensional abnormal feature vectors into the preset multi-level decision rule tree. The multi-level decision rule tree performs fusion reasoning based on different combinations of abnormal features of each dimension and preset decision logic, outputs the review conclusion, and provides a manual review window.

[0080] Furthermore, the system also includes: The site zoning configuration module groups sites that are geographically close and exhibit similar pollutant change trends into the same zone, and provides this zone data to the multi-dimensional anomaly detection module. Specifically, it calculates the spatial distance between sites based on Geographic Information System (GIS) data, performs trend similarity analysis using historical pollutant concentration time-series data (e.g., using dynamic time warping, correlation coefficients, or clustering algorithms), and automatically identifies and generates site zoning schemes. The zoning results can be stored in configuration files or database tables, supporting manual review and adjustment. Through site zoning, the multi-dimensional anomaly detection module can incorporate horizontal comparison information within zones, such as identifying whether a particular site's data deviates from the overall trend of the zone, thereby improving the accuracy of anomaly detection.

[0081] The parameter configuration module is used to configure the adjustable parameters, application time periods, and application scope of the multi-factor anomaly diagnostic model. Adjustable parameters include at least: thresholds for anomaly detection algorithms in each dimension (e.g., lower limit for rounding off negative values ​​to zero, duration for constant value determination, and size of the outlier statistical window), normal ranges for equipment status parameters, temperature and humidity thresholds for the station environment, and recovery period stabilization time. The application time period configuration allows for the use of different parameter sets for different time periods (e.g., day / night, seasons, and special events). The application scope configuration supports selectively enabling or disabling specific detection rules based on site, region, pollutant type, or data source. This module is typically provided in the form of a visual configuration interface or configuration file, allowing system administrators or business experts to dynamically adjust model behavior according to actual needs without modifying the underlying code.

[0082] A graphical batch review interface is used to centrally display preprocessed multi-site single-factor 24-hour data line charts and receive review instructions from reviewers. This interface should have at least the following functions: Data visualization: Displays the concentration change curves of the same pollutant factor at multiple sites simultaneously over 24 hours in the form of line charts, supporting filtering and grouping by region, site, or anomaly marker; Anomaly marker overlay: Highlights the abnormal time periods or feature labels (such as zero negative values, constant values, equipment anomalies, etc.) output by the multi-dimensional anomaly detection module on the curve. Batch operation: Allows reviewers to assign "valid", "invalid" or "warning" conclusions to multiple selected curves or time periods in batches; Manual review instruction reception: Reviewers can correct or confirm the automatic review conclusion by clicking, selecting boxes, or entering information, and the system records the manual operation log.

[0083] This interface aims to overcome the limitations of fully automated review, provide an efficient tool for human intervention in complex scenarios, and accumulate human review data for subsequent rule optimization and model iteration.

[0084] In this embodiment, the multi-level serial intelligent auditing system for environmental monitoring data works in concert with multiple modules to jointly construct a complete anomaly diagnosis and auditing system, thereby improving the intelligence level and maintainability of data auditing.

[0085] It should be noted that the functions of each module in this embodiment system can be found in the corresponding descriptions in the above methods, and will not be repeated here.

[0086] Example 3 This embodiment provides an electronic device. Figure 5 A structural block diagram of an electronic device according to an embodiment of the present invention is shown. Figure 5 As shown, the electronic device includes a memory 100 and a processor 200. The memory 100 stores a computer program that can run on the processor 200. When the processor 200 executes the computer program, it implements the multi-level serial intelligent auditing method for environmental monitoring data described in the above embodiment. The number of memories 100 and processors 200 can be one or more.

[0087] The electronic device also includes: The communication interface 300 is used to communicate with external devices and perform data exchange and transmission.

[0088] If the memory 100, processor 200, and communication interface 300 are implemented independently, they can be interconnected via a bus to communicate with each other. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc.

[0089] Optionally, in a specific implementation, if the memory 100, processor 200, and communication interface 300 are integrated on a single chip, then the memory 100, processor 200, and communication interface 300 can communicate with each other through an internal interface.

[0090] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method provided in this invention.

[0091] This invention also provides a chip, which includes a processor for calling and executing instructions stored in a memory, causing a communication device on which the chip is installed to perform the method provided in this invention.

[0092] This invention also provides a chip, including: an input interface, an output interface, a processor, and a memory. The input interface, output interface, processor, and memory are connected through an internal connection path. The processor is used to execute code in the memory. When the code is executed, the processor is used to execute the method provided in this invention.

[0093] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that the processor can be a processor supporting the Advanced Reduced Instruction Set Computing (RISC) machine (ARM) architecture.

[0094] Further, optionally, the aforementioned memory may include read-only memory and random access memory, and may also include non-volatile random access memory. The memory may be volatile or non-volatile, or may include both. Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which serves as an external cache. Many forms of RAM are available by way of example, but not limitation. Examples include static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0095] In the above embodiments, implementation can be achieved, in whole or in part, by software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0096] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0097] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0098] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-level serial intelligent verification method for environmental monitoring data, characterized in that, include: Multiple environmental data streams from a distributed environmental monitoring network are acquired, and all environmental data streams are filtered to select data carrying a backfill identifier as valid candidate data. The valid candidate data are subjected to parallel anomaly detection in multiple dimensions to generate a multi-dimensional anomaly feature vector; wherein, the multiple dimensions include at least the numerical feature anomalies of the data itself, the anomalies of the equipment operating status, the anomalies of the station environment, the impact of operation and maintenance activities, and the impact of the equipment anomaly recovery period. The multi-dimensional abnormal feature vector is input into a preset multi-level decision rule tree. The multi-level decision rule tree performs fusion reasoning based on different combinations of abnormal features in each dimension and preset decision logic, and outputs the audit conclusion.

2. The multi-level serial intelligent verification method for environmental monitoring data according to claim 1, characterized in that, The data filtering of all environmental data streams includes: Based on a preset missing pattern library, the environmental data stream is classified into missing data categories to determine the missing data. The missing data is imputed, an imputed identifier is added to the imputed data, and the data with the imputed identifier is selected as the valid candidate data.

3. The multi-level serial intelligent verification method for environmental monitoring data according to claim 1, characterized in that, The detection of numerical anomalies in the data itself includes multiple anomaly models based on numerical logic. These multiple anomaly models include at least a zero negative value model, a data constant value model, a general outlier model, an ozone feature model, and a particulate matter feature model.

4. The multi-level serial intelligent verification method for environmental monitoring data according to claim 1, characterized in that, The detection of abnormal equipment operation status is achieved by associating with instrument status information, which includes at least one or more of the following: instrument status parameters, quality control records, fault work orders, and real-time alarm information.

5. The multi-level serial intelligent verification method for environmental monitoring data according to claim 1, characterized in that, The detection of station building environmental anomalies is based on abnormal signals from sampling equipment and the assessment of station building dynamic environmental parameters, including temperature and humidity.

6. The multi-level serial intelligent verification method for environmental monitoring data according to claim 1, characterized in that, The detection of the impact of the operation and maintenance activities is achieved by matching the data timestamp with the entry to exit time window in the operation and maintenance work order and access control record to mark the period of known human activity impact.

7. The multi-level serial intelligent verification method for environmental monitoring data according to claim 1, characterized in that, The detection of the impact of the abnormal equipment recovery period is used to identify power outage recovery or equipment startup events, and to mark the data within a preset stable period after the event occurs.

8. The multi-level serial intelligent verification method for environmental monitoring data according to claim 1, characterized in that, The multi-level decision rule tree performs fusion reasoning based on different combinations of abnormal features from various dimensions and preset decision logic, and outputs the following audit conclusions: If the multi-dimensional anomaly feature vector contains a zero negative value detection feature that is less than the rounding lower limit, then the conclusion of invalid data is output. If the multi-dimensional anomaly feature vector contains zero negative value detection features within the rounding interval, and does not contain any of the features of abnormal equipment operation status, abnormal station environment, impact of operation and maintenance activities, and impact of equipment anomaly recovery period, then the audit conclusion of the data to be rounded is output. If the multi-dimensional anomaly feature vector contains a zero negative value detection feature within the rounding interval, and contains at least one feature among abnormal equipment operation status, abnormal station environment, impact of operation and maintenance activities, and impact of equipment anomaly recovery period, then the audit conclusion of invalid data will be output. If the multi-dimensional abnormal feature vector contains other numerical abnormal features besides zero negative values, and does not contain any of the features of abnormal equipment operation status, abnormal station environment, impact of operation and maintenance activities, and impact of equipment abnormal recovery period, then the audit conclusion of the reminder data that needs to be manually reviewed will be output. If the multi-dimensional abnormal feature vector contains other numerical abnormal features besides zero negative values, and contains at least one of the following features: abnormal equipment operation status, abnormal station environment, impact of operation and maintenance activities, and impact of equipment abnormality recovery period, then the audit conclusion of invalid data will be output.

9. A multi-level serial intelligent verification system for environmental monitoring data, characterized in that, The system implements the multi-level serial intelligent verification method for environmental monitoring data as described in any one of claims 1 to 8; the system includes: The data acquisition and processing module is used to acquire various environmental data streams from the distributed environmental monitoring network, and to filter all environmental data streams to select data carrying a backfill identifier as valid candidate data. A multi-dimensional anomaly detection module is used to perform anomaly detection on the valid candidate data in parallel across multiple dimensions, generating a multi-dimensional anomaly feature vector; wherein, the multiple dimensions include at least the numerical feature anomalies of the data itself, the anomalies of the equipment operating status, the anomalies of the station environment, the impact of operation and maintenance activities, and the impact of the equipment anomaly recovery period; The decision fusion module has a built-in multi-level decision rule tree, which is used to input the multi-dimensional abnormal feature vector into the preset multi-level decision rule tree. The multi-level decision rule tree performs fusion reasoning based on different combinations of abnormal features of each dimension and preset decision logic, and outputs the review conclusion.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the multi-level serial intelligent auditing method for environmental monitoring data as described in any one of claims 1 to 8.

Images