Abnormal value screening method based on particulate matter component reconstruction

Abstract

The invention provides an abnormal value screening method based on particulate matter component reconstruction, and relates to the technical field of environment monitoring. The method comprises the following steps: obtaining standard ion component data, standard carbon component data and standard inorganic element component data; obtaining a plurality of reconstruction component data; target PM2.5 data are determined; determining a doubt reconstruction component data judgment result of the plurality of reconstruction component data at each moment; determining a univariate time sequence abnormal point judgment result of each kind of reconstruction component data at each moment; determining a multivariable time sequence abnormal point judgment result of the multiple reconstruction component data at each moment; and determining whether the reconstruction component data at each moment is abnormal data. According to the method, the component reconstruction characteristics under different pollution levels can be evaluated, and the understanding of the model on the relationship among the particulate matter components is enhanced, so that the stability and accuracy of identifying whether the reconstructed component data is abnormal data are improved, the calculation efficiency is optimized, and environment monitoring auditing personnel are assisted to make decisions.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to an outlier screening method based on particulate matter composition reconstruction. Background Technology

[0002] Particulate matter (PM2.5) monitoring data is often affected by factors such as instrument malfunction, environmental interference, and abnormal data transmission, resulting in outliers. Therefore, it is necessary to review PM2.5 monitoring data. These outliers not only affect the accuracy of the data but may also mislead subsequent data analysis and decision-making. Currently, manual review is the primary method, mainly considering fixed rules and methods, such as monitoring methodology (whether the upper and lower limits of the monitoring method are exceeded, whether the instrument is malfunctioning or there are quality control issues during operation), component logical relationships (such as the proportional relationships between components, anion and cation balance, etc.), and simple statistical methods (correlation coefficients, time-series variation characteristics, etc.). However, when faced with complex environmental data, it is difficult to reflect the overall logical relationship between different factor concentrations and PM2.5 concentration. Furthermore, manual review methods are highly subjective, making it difficult to quantitatively analyze various components in atmospheric particulate matter and to reconstruct these complex data into more representative features, thus limiting the screening of outlier particulate matter data.

[0003] The information disclosed in the background section of this application is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0004] This invention provides a method for identifying air pollution events caused by straw burning based on the fusion of multi-source monitoring data. This method can solve the technical problems that related technologies are unable to quantitatively analyze various components in atmospheric particulate matter and are unable to reconstruct these complex data into more representative features.

[0005] According to a first aspect of the present invention, an outlier screening method based on particulate matter composition reconstruction is provided, comprising: The system acquires ionic component data, carbon component data, and inorganic element component data from multiple moments within a preset time period from the component monitoring station, and performs data preprocessing to obtain standard ionic component data, standard carbon component data, and standard inorganic element component data. Based on standard ionic component data, standard carbon component data, and standard inorganic element component data, multiple reconstructed component data were obtained; Based on multiple reconstructed component data and the location information of component monitoring stations, the target PM was determined. 2.5 data; Based on multiple reconstructed component data and target PM 2.5Data, determining the judgment results of questionable reconstructed component data of multiple reconstructed component data at various time points; Set a first sliding time window within a preset time period, and determine the univariate time series outlier judgment result of each reconstructed component data at each time point based on the first sliding time window; Set a second sliding time window within a preset time period, and base the second sliding time window on the target PM. 2.5 Data, determining the results of multivariate time series outlier judgment for various reconstructed component data at different times; Based on the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of various reconstructed component data at each time point, it is determined whether various reconstructed component data at each time point are outliers.

[0006] According to the present invention, the reconstructed component data includes: organic component data, nitrate component data, sulfate component data, ammonium salt component data, elemental carbon component data, chloride salt component data, crustal material component data, and trace element component data.

[0007] According to the present invention, the target PM is determined. 2.5 Data, including: Based on multiple reconstructed component data, obtain the summation data of the reconstructed components at each time step; The first PM2.5 concentrations at multiple times within a preset time period were obtained from air monitoring stations at the same location as the component monitoring stations. 2.5 Data, the second PM at multiple times within a preset time period for air monitoring stations located at distances less than a preset distance from the component monitoring station. 2.5 Data, and the third PM at multiple times within a preset time period from air monitoring stations located in the same city. 2.5 data; Determine the summation data of the reconstructed components at each time step and compare it with the first PM at the same time step. 2.5 The first difference of the data, the sum of the reconstructed components at each time step, and the second PM at the same time step. 2.5 The second difference of the data, and the summation of the reconstructed components at each time step with the third PM at the same time step. 2.5 The third difference in the data; Based on the first difference, second difference, and third difference, determine the target PM corresponding to the summed data of the reconstructed components at each time step. 2.5 data.

[0008] According to the present invention, determining the judgment result of doubtful reconstructed component data of multiple reconstructed component data at various time points includes: Obtain the summation data of the reconstructed components and the target PM 2.5The first proportional value of the data; Based on the first ratio value, identify outlier data in the reconstructed component summation data and remove outlier data; In the reconstructed component summation data after removing outlier data, based on the corresponding target PM 2.5 Data is used to determine the normal range of the first ratio value; Based on the normal ratio range and the first ratio value, the judgment results of the questionable reconstructed component data of various reconstructed component data at each time point are determined.

[0009] According to the present invention, determining outlier data in reconstructed component summation data includes: Arrange the first proportion values ​​at each time point in order of magnitude to obtain the first proportion value sequence; Based on the first proportional value sequence, determine the 25th percentile and the 75th percentile; Determine the fourth difference between the 25th percentile and the 75th percentile; The first proportion range is determined based on the 25th percentile, the 75th percentile, and the fourth difference. The reconstructed component data corresponding to the summation data of the reconstructed components whose first proportion value does not belong to the first proportion range are identified as outlier data.

[0010] According to the present invention, determining the univariate temporal outlier judgment result for each reconstructed component data at each time step includes: The neighborhood parameters are determined based on the duration of the first sliding time window; When the center time of the first sliding time window is aligned with the t-th time within the preset time period, the local anomaly factor value of the reconstructed component data at the center time within the first sliding time window is determined based on the neighborhood parameter. Based on the comparison between the local anomaly factor values ​​and the preset factor thresholds, the univariate time-series anomaly judgment result of the reconstructed component data at time t is determined.

[0011] According to the present invention, determining the neighborhood parameters based on the duration of the first sliding time window includes: According to the formula k=max(3,min(k_max,floor(w×ρ))) Determine the neighborhood parameter k, where k_max is the preset upper limit, w is the duration of the first sliding time window, ρ is the preset coefficient, min is the minimum value function, max is the maximum value function, and floor is the floor function.

[0012] According to the present invention, determining the results of multivariate temporal outlier judgment for various reconstructed component data at different times includes: When the center time of the second sliding time window aligns with the t-th time within the preset time period, the summation data of the reconstruction components at each time point of the second sliding time window is obtained based on multiple reconstruction component data. Determine the reconstructed component summation data and target PM at each moment of the second sliding time window. 2.5 Pearson correlation coefficient of the data; Based on the comparison between the Pearson correlation coefficient and the preset coefficient threshold, the results of the multivariate time series outlier judgment for the various reconstructed component data at time t are determined.

[0013] According to the present invention, determining whether various reconstructed component data at each time point are anomalous data includes: If the judgment results of the questionable reconstructed component data of the j-th reconstructed component data at time t, the judgment results of the univariate time series outlier and the judgment results of the multivariate time series outlier are all abnormal, the j-th reconstructed component data at time t will be identified as abnormal data.

[0014] According to a second aspect of the present invention, an outlier screening system based on particulate matter component reconstruction is provided, comprising: an acquisition and preprocessing module, which acquires ionic component data, carbon component data and inorganic element component data of component monitoring stations at multiple times within a preset time period in the past, and performs data preprocessing to obtain standard ionic component data, standard carbon component data and standard inorganic element component data; The reconstructed component data module obtains various reconstructed component data based on standard ionic component data, standard carbon component data, and standard inorganic element component data; Target PM 2.5 The data module, based on various reconstructed component data and the location information of component monitoring stations, determines the target PM. 2.5 data; The first judgment result module, based on multiple reconstructed component data and the target PM... 2.5 Data, determining the judgment results of questionable reconstructed component data of multiple reconstructed component data at various time points; The second judgment result module sets a first sliding time window within a preset time period, and determines the judgment result of the univariate time series outlier point of each reconstructed component data at each time point based on the first sliding time window. The third judgment result module sets a second sliding time window within a preset time period, and determines the result based on the second sliding time window and the target PM. 2.5 Data, determining the results of multivariate time series outlier judgment for various reconstructed component data at different times; The abnormal data module determines whether the various reconstructed component data at each time point are abnormal data based on the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of various reconstructed component data at each time point.

[0015] According to a third aspect of the present invention, a computer-readable storage medium is provided having computer program instructions stored thereon, which, when executed by a processor, implement the outlier screening method based on particulate matter composition reconstruction.

[0016] By adopting the above technical solution, the present invention can achieve the following technical effects: According to this invention, multiple reconstructed component data can be obtained based on ionic component data, carbon component data, and inorganic element component data. This allows for the determination of questionable reconstructed component data, and the establishment of a first and second sliding window to determine the univariate time-series outlier judgment results for each type of reconstructed component data, as well as the multivariate time-series outlier judgment results for multiple reconstructed component data, thereby determining whether various reconstructed component data are anomalous. This invention enables quantitative analysis of various components in atmospheric particulate matter and assessment of component reconstruction characteristics under different pollution levels, enhancing the model's understanding of the relationships between particulate matter components. This improves the stability and accuracy of identifying anomalous data, optimizes computational efficiency, meets the needs of real-time processing of online monitoring data, and provides possible causes for outliers, assisting environmental monitoring auditors in decision-making. Furthermore, multiple reconstructed component data can be obtained based on standard ionic component data, standard carbon component data, and standard inorganic element component data, thereby determining the target PM2.5 concentration. 2.5 The data provides the foundation for identifying anomalous data. Furthermore, it can be based on various reconstructed component data and target PM. 2.5 This system identifies questionable reconstructed component data at various time points and sets first and second sliding time windows to determine the univariate and multivariate time-series outlier assessments for each reconstructed component data at each time point. This allows for the determination of whether various reconstructed component data at each time point are anomalous. This improves the stability and accuracy of identifying anomalous reconstructed component data, optimizes computational efficiency, meets the needs of real-time online monitoring data processing, and provides possible causes of outliers, assisting environmental monitoring auditors in decision-making.

[0017] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Other features and aspects of the invention will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort. Figure 1 An exemplary flowchart of outlier screening based on particulate matter composition reconstruction according to an embodiment of the present invention is shown. Figure 2 An exemplary flowchart is shown below for determining the judgment results of questionable reconstructed component data of various reconstructed component data at various times according to an embodiment of the present invention; Figure 3 A block diagram of an outlier screening system based on particulate composition reconstruction according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 An exemplary flowchart of outlier screening based on particulate matter composition reconstruction according to an embodiment of the present invention is shown, the method comprising: Step S1: Obtain ionic component data, carbon component data, and inorganic element component data from multiple moments within a preset time period of the component monitoring station, and perform data preprocessing to obtain standard ionic component data, standard carbon component data, and standard inorganic element component data. Step S2: Based on standard ionic component data, standard carbon component data, and standard inorganic element component data, obtain multiple reconstructed component data; Step S3: Based on the reconstructed component data and the location information of the component monitoring stations, determine the target PM. 2.5 data; Step S4, based on multiple reconstructed component data and target PM 2.5Data, determining the judgment results of questionable reconstructed component data of multiple reconstructed component data at various time points; Step S5: Set a first sliding time window within a preset time period, and determine the univariate time series outlier judgment result of each reconstructed component data at each time point based on the first sliding time window; Step S6: Set a second sliding time window within a preset time period, and determine the target PM based on the second sliding time window. 2.5 Data, determining the results of multivariate time series outlier judgment for various reconstructed component data at different times; Step S7: Based on the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of various reconstructed component data at each time point, determine whether the various reconstructed component data at each time point are outliers.

[0022] The outlier screening method based on particulate matter component reconstruction according to embodiments of the present invention can obtain multiple reconstructed component data based on ionic component data, carbon component data, and inorganic element component data. It then determines the judgment results of questionable reconstructed component data for these multiple reconstructed component data, and sets a first sliding window and a second sliding window to determine the judgment results of univariate time-series outliers for each type of reconstructed component data and the judgment results of multivariate time-series outliers for multiple reconstructed component data, thereby determining whether various reconstructed component data are anomalous. It can quantitatively analyze various components in atmospheric particulate matter and evaluate the component reconstruction characteristics under different pollution levels, enhancing the model's understanding of the relationships between particulate matter components, thereby improving the stability and accuracy of identifying anomalous data, optimizing computational efficiency, meeting the needs of real-time processing of online monitoring data, and providing possible causes of outlier generation to assist environmental monitoring auditors in decision-making.

[0023] Example 1: According to an embodiment of the present invention, in step S1, ionic component data, carbon component data, and inorganic element component data of component monitoring stations at multiple moments within a preset time period are acquired, and data preprocessing is performed to obtain standard ionic component data, standard carbon component data, and standard inorganic element component data. The preset time period can be one year, and the time interval between adjacent moments can be one hour. For example, ionic component data, carbon component data, and inorganic element component data of component monitoring stations in each region are acquired every hour within the past year. The ionic component data can be obtained by ion chromatography, and the main monitoring indicators include SO4. 2- NO3 - Cl - Na + K + Mg 2+ Ca 2+NH4 + The carbon component data can be obtained through thermal-optical correction and thermal-optical attenuation methods, with the main detection indicators including organic carbon (OC) and elemental carbon (EC). The inorganic elemental component data can be obtained through methods such as inductively coupled plasma mass spectrometry (ICP-MS) and X-ray fluorescence spectrometry (XRF), with the main monitoring indicators including K, Ca, V, and Cr. These multiple indicators can represent the concentration of the substance; for example, K represents the potassium ion concentration. Further, by preprocessing the above-mentioned ionic component data, carbon component data, and inorganic elemental component data (i.e., data cleaning and data standardization), the standard ionic component data, standard carbon component data, and standard inorganic elemental component data can be obtained. For example, obviously erroneous or missing data is removed, and the units of all data are unified to μg / m³, with three decimal places. This invention does not limit the data acquisition method or preprocessing method.

[0024] Example 2: According to an embodiment of the present invention, in step S2, multiple reconstructed component data are obtained based on standard ionic component data, standard carbon component data, and standard inorganic element component data. The reconstructed component data includes: organic component data, nitrate component data, sulfate component data, ammonium salt component data, elemental carbon component data, chloride salt component data, crustal material component data, and trace element component data. For example, organic component data = organic carbon (OC) × 1.6, nitrate component data = NO3. - Sulfate component data = SO4 2- Ammonium salt component data = NH4 + Elemental carbon component data = elemental carbon (EC), chloride salt component data = Cl - Crustal material composition data = (Al×2.2) + (Si×2.49) + (Ca2+×1.63) + (Fe×2.42) + (Ti×1.94) Trace element composition data = K + +Mg 2+ +F - +Ba+Cd+Sn+V+Cr+Mn+Co+Ni+Cu+Zn+As+Se+Pb+Sc+P+Na + .

[0025] Example 3: According to an embodiment of the present invention, in step S3, the target PM is determined based on multiple reconstructed component data and the location information of component monitoring stations. 2.5 The data includes: summation data of reconstructed components at various times based on multiple reconstructed component data; and the first PM2.5 concentration at multiple times within a preset time period for air monitoring stations at the same location of component monitoring stations. 2.5Data, the second PM at multiple times within a preset time period for air monitoring stations located at distances less than a preset distance from the component monitoring station. 2.5 Data, and the third PM at multiple times within a preset time period from air monitoring stations located in the same city. 2.5 Data; determine the reconstructed components at each time step, sum the data, and compare it with the first PM at the same time step. 2.5 The first difference of the data, the sum of the reconstructed components at each time step, and the second PM at the same time step. 2.5 The second difference of the data, and the summation of the reconstructed components at each time step with the third PM at the same time step. 2.5 The third difference of the data; based on the first, second, and third differences, determine the target PM corresponding to the summed data of the reconstructed components at each time step. 2.5 data.

[0026] According to an embodiment of the present invention, the sum of the organic component data, nitrate component data, sulfate component data, ammonium salt component data, elemental carbon component data, chloride salt component data, crustal material component data, and trace element component data is the reconstructed component summation data. If an air monitoring station exists at the same location as the component monitoring station, then PM2.5 levels collected by that air monitoring station at multiple times within a preset time period are obtained. 2.5 The concentration of PM2.5 is the first PM2.5 concentration. 2.5 Data, and simultaneously, PM2.5 levels collected at multiple times within a preset time period from air monitoring stations located less than a preset distance (e.g., 20 km) from the component monitoring station. 2.5 The concentration of PM2.5 is the second PM2.5 concentration. 2.5 Data was collected, and PM2.5 levels were obtained from air monitoring stations located in the same city as the component monitoring stations at multiple times within a preset time period. 2.5 The concentration of that is, the third PM. 2.5 Data. The reconstructed component summation data is compared with the first PM at the same time. 2.5 The absolute value of the difference between the data is the first difference. The reconstructed component summation data is compared with the second PM at the same time. 2.5 The absolute value of the difference between the data is the second difference. The reconstructed component summation data is compared with the third PM at the same time. 2.5 The absolute value of the difference between the data is the third difference, and the minimum value of the first, second, and third differences corresponds to PM. 2.5 The data refers to the target PM. 2.5 Data. If any two or all of the first, second, and third differences are the same, the largest PM can be... 2.5 The data was identified as the target PM. 2.5Data. For example, the first and third differences are both 2, but the first PM... 2.5 The data is less than the sum of the reconstructed components, the third PM 2.5 If the data is greater than the sum of the reconstructed components, the third PM can be used. 2.5 The data was identified as the target PM. 2.5 Data. Based on the same processing method, the target PM at each time point can be obtained. 2.5 Data, target PM at each moment 2.5 The data may be based on different PMs 2.5 The data is definite, for example, the target PM at a certain moment. 2.5 The data represents the first PM at the same time. 2.5 Data, the target PM of the previous moment 2.5 The data is the second PM at the same time. 2.5 Data, various PMs 2.5 The unit of the data is also μg / m³.

[0027] In this way, multiple reconstructed component data can be obtained based on standard ionic component data, standard carbon component data, and standard inorganic element component data, thereby determining the target PM. 2.5 Data provides the basic data for identifying anomalous data.

[0028] Example 4: Figure 2 An exemplary flowchart is shown below for determining the results of judging questionable reconstructed component data of various reconstructed component data at various times according to an embodiment of the present invention.

[0029] According to an embodiment of the present invention, in step S4, based on multiple reconstructed component data and the target PM 2.5 Data, determining the judgment results of questionable reconstructed component data for various reconstructed component data at each time point, including: Step S41, obtaining the summation data of reconstructed components and the target PM 2.5 The first scale value of the data; Step S42, based on the first scale value, determine the outlier data in the reconstructed component summation data and remove the outlier data; Step S43, in the reconstructed component summation data after removing the outlier data, according to the corresponding target PM 2.5 The data is used to determine the normal range of the first ratio value; step S44, based on the normal ratio range and the first ratio value, the judgment result of the doubtful reconstructed component data of various reconstructed component data at each time point is determined.

[0030] According to an embodiment of the present invention, in step S41, the reconstructed component summation data and the target PM are obtained. 2.5 The first scale value of the data. The reconstructed component summation data is compared with the target PM at that moment. 2.5The ratio of the data is the first ratio value, thus the first ratio value at each time point can be obtained.

[0031] According to an embodiment of the present invention, in step S42, determining outlier data in the reconstructed component summation data based on a first ratio value and removing outlier data includes: arranging the first ratio values ​​at each time point in ascending order to obtain a first ratio value sequence; determining the 25th percentile and the 75th percentile value based on the first ratio value sequence; determining the fourth difference between the 25th percentile and the 75th percentile values; determining a first ratio range based on the 25th percentile, the 75th percentile, and the fourth difference; and identifying multiple reconstructed component data corresponding to reconstructed component summation data whose first ratio values ​​do not belong to the first ratio range as outlier data.

[0032] According to an embodiment of the present invention, the sequence obtained by arranging the first proportion values ​​at various times in ascending order is the first proportion value sequence. Then, the 25th percentile and the 75th percentile can be determined based on the first proportion value sequence. For example, if there are 8760 first proportion values ​​in the first proportion value sequence, the 2190th first proportion value is the 25th percentile, and the 6570th first proportion value is the 75th percentile. The difference between the 75th percentile and the 25th percentile is the fourth difference. Furthermore, the normal range of the first proportion value can be determined based on the 25th percentile, the 75th percentile, and the fourth difference [75th percentile - 1.5 × fourth difference, 75th percentile + 1.5 × fourth difference]. This is the first proportion range. Based on the above processing, the normal range of the first proportion value can be appropriately expanded to facilitate the screening of extremely abnormal reconstructed component summation data. When the first proportion value corresponding to the reconstructed component summation data does not belong to the first proportion range, it can be considered that the reconstructed component summation data has a significant difference from other reconstructed component summation data. Therefore, the eight reconstructed component data corresponding to the reconstructed component summation data can be identified as outlier data and removed.

[0033] According to an embodiment of the present invention, in step S43, in the reconstructed component summation data after removing outlier data, the corresponding target PM is... 2.5 The data determines the normal range for the first proportion value. Target PM 2.5 The normal range of the first ratio value varies depending on the range of the data. For example, due to the target PM 2.5 When the data is ≤5μg / m³, the target PM 2.5 If the data is too small, the initial ratio value may be too high. Therefore, it can be adjusted according to the target PM. 2.5The data is set to a normal proportion range, and then the judgment result of the questionable reconstructed component data of various reconstructed component data at each time is determined based on whether the first proportion value belongs to the normal proportion range. For example, when the target PM 2.5 When the data is ≤5, the normal range for the summation of the corresponding reconstructed components is ≤12. When the target PM 2.5 When the data is within the range of (5, 15], the normal range of the corresponding first ratio value is [0.5, 2.5]. When the target PM 2.5 When the data is within the range of (15, 35], the normal range of the corresponding first ratio value is [0.6, 1.5]. When the target PM 2.5 When the data is within the range of (35, 50], the normal range of the corresponding first ratio value is [0.8, 1.25]. When the target PM 2.5 When the data is within the range of (50, 75], the normal range of the corresponding first ratio value is [0.8, 1.2]. When the target PM 2.5 When the data is within the range of (75, 150], the normal range of the corresponding first ratio value is [0.85, 1.13]. When the target PM 2.5 When the data is greater than 150, the normal range of the corresponding first ratio value is [0.85, 1.12].

[0034] According to an embodiment of the present invention, in step S44, the judgment result of the questionable reconstructed component data of various reconstructed component data at each time point is determined based on the normal ratio range and the first ratio value. When the target PM is in different ranges... 2.5 When the first ratio value corresponding to the data is not within the normal ratio range, the judgment result of the questionable reconstructed component data of multiple reconstructed component data corresponding to the first ratio value can be determined as "abnormal". That is, all eight reconstructed component data may be abnormal, and when the target PM is in different ranges 2.5 When the first ratio value corresponding to the data is within the normal ratio range, the judgment result of the questionable reconstructed component data of the multiple reconstructed component data corresponding to the first ratio value can be determined as "normal", thereby determining the judgment result of the questionable reconstructed component data of the multiple reconstructed component data at each time.

[0035] Example 5: According to an embodiment of the present invention, in step S5, a first sliding time window is set within a preset time period, and the univariate temporal anomaly judgment result of each reconstructed component data at each time point is determined based on the first sliding time window, including: determining a neighborhood parameter based on the duration of the first sliding time window; when the center time of the first sliding time window is aligned with the t-th time point within the preset time period, determining the local anomaly factor value of the reconstructed component data at the center time point within the first sliding time window based on the neighborhood parameter; and determining the univariate temporal anomaly judgment result of the reconstructed component data at the t-th time point based on the comparison result between the local anomaly factor value and a preset factor threshold.

[0036] According to an embodiment of the present invention, the first sliding time window can be a time window with a duration of 5 to 12 time units, wherein the time unit is 1 hour. When the center time of the first sliding time window is aligned with the t-th time within a preset time period, the data of the first sliding time window corresponding to the t-th time can be obtained. For example, when the first sliding time window is a time window of 5 time units, the data of the first sliding time window corresponding to the t-th time is a reconstructed component data of five times: t-2, t-1, t, t+1, and t+2.

[0037] According to an embodiment of the present invention, determining the neighborhood parameter based on the duration of the first sliding time window includes: determining the neighborhood parameter k according to formula (1). k=max(3,min(k_max,floor(w×ρ)))(1) Where k_max is the preset upper limit, w is the duration of the first sliding time window, ρ is the preset coefficient, min is the minimum value function, max is the maximum value function, and floor is the floor function.

[0038] According to an embodiment of the present invention, in formula (1), ρ is a preset coefficient, and the value range can be [0.3, 0.5]. floor(w×ρ) is the floor of (w×ρ). For example, the duration w of the first sliding time window is 5, the preset coefficient ρ is 0.32, and (w×ρ) is 1.6. Floor of 1.6 is 1. min(k_max,floor(w×ρ)) represents the minimum value among (k_max,floor(w×ρ). For example, if k_max is 3 and floor(w×ρ) is 1, then min(k_max,floor(w×ρ)) is 1. max(3,min(k_max,floor(w×ρ))) represents the maximum value among (3,min(k_max,floor(w×ρ))) and can be used as the neighborhood parameter. For example, if min(k_max,floor(w×ρ)) is 1, then max(3,min(k_max,floor(w×ρ))) is 3, and the neighborhood parameter k is 3.

[0039] According to an embodiment of the present invention, when the center time of the first sliding time window aligns with the t-th time within a preset time period, the local anomaly factor value of the reconstructed component data at the center time within the first sliding time window is determined based on the neighborhood parameter. When the center time of the first sliding time window aligns with the t-th time within the preset time period, Euclidean distance can be used as the distance metric between the reconstructed component data within the first sliding time window. The corresponding neighborhood parameter k is then substituted into the Local Anomaly Factor (LOF) algorithm to obtain the local anomaly factor value of the component reconstructed data at the t-th time. For example, substituting the neighborhood parameter k corresponding to the organic component data at the t-th time into the LOF algorithm yields the local anomaly factor value of the organic component data at the t-th time.

[0040] According to an embodiment of the present invention, the univariate time-series outlier judgment result of the reconstructed component data at time t is determined based on the comparison result between the local anomaly factor value and the preset factor threshold. The preset factor threshold can be in the range of [1.8, 2.5]. When the local anomaly factor value is greater than the preset factor threshold, the univariate time-series outlier judgment result of the reconstructed component data at time t can be determined as "anomaly". For example, if the preset factor threshold is 2, and the local anomaly factor value of the organic component data at time t is 3, then the univariate time-series outlier judgment result of the organic component data at time t is "anomaly".

[0041] Furthermore, the first sliding time window can be slid so that its center time aligns with the (t+1)th time within a preset time period. This allows the local anomaly factor value at the (t+1)th time to be determined using the same method as above, thereby determining the univariate time series outlier judgment result. This process can continue until the end time of the first sliding time window aligns with the end time of the preset time period. Based on the same processing method, the univariate time series outlier judgment results for various reconstructed component data can be determined.

[0042] Example 6: According to an embodiment of the present invention, in step S6, a second sliding time window is set within a preset time period, and the second sliding time window and the target PM are used as the basis for the calculation. 2.5 The data determines the results of multivariate temporal outlier judgment for various reconstructed component data at different times, including: when the center time of the second sliding time window aligns with the t-th time within a preset time period, obtaining the summation data of the reconstructed components at each time of the second sliding time window based on the various reconstructed component data; determining the summation data of the reconstructed components at each time of the second sliding time window and the target PM. 2.5 The Pearson correlation coefficient of the data; based on the comparison between the Pearson correlation coefficient and the preset coefficient threshold, the multivariate time series outlier judgment result of the multiple reconstructed component data at time t is determined.

[0043] According to an embodiment of the present invention, the second sliding time window has a relatively long duration, which can be a time window of 12 to 48 time units, wherein the time unit is 1 hour, so as to avoid trend noise interference caused by too short a time window, or short-term anomalies being masked by too long a time window. Similar to the first sliding time window, when the center time of the second sliding time window is aligned with the t-th time within the preset time period, the data of the second sliding time window corresponding to the t-th time can be obtained.

[0044] According to an embodiment of the present invention, when the center time of the second sliding time window aligns with the t-th time within a preset time period, the reconstructed component summation data for each time point within the second sliding time window is obtained, thereby determining the reconstructed component summation data sequence within the second sliding time window and its corresponding target PM. 2.5 The Pearson correlation coefficient between data sequences can describe the relationship between the reconstructed component summation data and the corresponding target PM. 2.5 Consistency of data change trends. When the Pearson correlation coefficient is less than a preset threshold (e.g., 0.7), the reconstructed component summation data is compared with the corresponding target PM. 2.5The data shows significant differences in trend, indicating anomalies. Therefore, the multivariate time-series outlier judgment results for the various reconstructed component data at time t can be identified as "anomalies." This means that all eight reconstructed component data may exhibit anomalies at time t, thus determining the multivariate time-series outlier judgment results for each time point. Further, the second sliding time window can be slid so that its center time aligns with the (t+1)th time point within a preset time period. The Pearson correlation coefficient at time t+1 can then be determined using the same method as above, thereby confirming the multivariate time-series outlier judgment results. This process can continue until the end time of the second sliding time window aligns with the end time of the preset time period.

[0045] Example 7: According to an embodiment of the present invention, in step S7, based on the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of various reconstructed component data at each time point, it is determined whether the various reconstructed component data at each time point are abnormal data. This includes: if the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of the j-th reconstructed component data at time t are all abnormal, the j-th reconstructed component data at time t is determined to be abnormal data. For example, if the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of the organic component data at time t are all "abnormal", the organic component data at time t can be determined to be abnormal data; otherwise, the organic component data at time t is normal data. Thus, it is possible to determine whether the various reconstructed component data at each time point are abnormal data.

[0046] In this way, multiple reconstructed component data and target PM can be used. 2.5 This system identifies questionable reconstructed component data at various time points and sets first and second sliding time windows to determine the univariate and multivariate time-series outlier assessments for each reconstructed component data at each time point. This allows for the determination of whether various reconstructed component data at each time point are anomalous. This improves the stability and accuracy of identifying anomalous reconstructed component data, optimizes computational efficiency, meets the needs of real-time online monitoring data processing, and provides possible causes of outliers, assisting environmental monitoring auditors in decision-making.

[0047] The outlier screening method based on particulate matter component reconstruction according to embodiments of the present invention can obtain multiple reconstructed component data based on ionic component data, carbon component data, and inorganic element component data. It then determines the judgment results of questionable reconstructed component data for these multiple reconstructed component data, and sets a first sliding window and a second sliding window to determine the judgment results of univariate time-series outliers for each type of reconstructed component data and multivariate time-series outliers for multiple reconstructed component data, thereby determining whether various reconstructed component data are anomalous. It can quantitatively analyze various components in atmospheric particulate matter and evaluate the component reconstruction characteristics under different pollution levels, enhancing the model's understanding of the relationships between particulate matter components, thereby improving the stability and accuracy of identifying anomalous data, optimizing computational efficiency, meeting the needs of real-time processing of online monitoring data, and providing possible causes of outlier generation to assist environmental monitoring auditors in decision-making. Furthermore, it can obtain multiple reconstructed component data based on standard ionic component data, standard carbon component data, and standard inorganic element component data, thereby determining the target PM2.5 concentration. 2.5 The data provides the foundation for identifying anomalous data. Furthermore, it can be based on various reconstructed component data and target PM. 2.5 This system identifies questionable reconstructed component data at various time points and sets first and second sliding time windows to determine the univariate and multivariate time-series outlier assessments for each reconstructed component data at each time point. This allows for the determination of whether various reconstructed component data at each time point are anomalous. This improves the stability and accuracy of identifying anomalous reconstructed component data, optimizes computational efficiency, meets the needs of real-time online monitoring data processing, and provides possible causes of outliers, assisting environmental monitoring auditors in decision-making.

[0048] Example 8: Figure 3 An exemplary block diagram of an outlier screening system based on particulate matter composition reconstruction according to an embodiment of the present invention is shown, the system comprising: The acquisition and preprocessing module acquires ionic component data, carbon component data, and inorganic element component data from the component monitoring station at multiple moments within a preset time period in the past, and performs data preprocessing to obtain standard ionic component data, standard carbon component data, and standard inorganic element component data. The reconstructed component data module obtains various reconstructed component data based on standard ionic component data, standard carbon component data, and standard inorganic element component data; Target PM 2.5 The data module, based on various reconstructed component data and the location information of component monitoring stations, determines the target PM. 2.5 data; The first judgment result module, based on multiple reconstructed component data and the target PM... 2.5 Data, determining the judgment results of questionable reconstructed component data of multiple reconstructed component data at various time points; The second judgment result module sets a first sliding time window within a preset time period, and determines the judgment result of the univariate time series outlier point of each reconstructed component data at each time point based on the first sliding time window. The third judgment result module sets a second sliding time window within a preset time period, and determines the result based on the second sliding time window and the target PM. 2.5 Data, determining the results of multivariate time series outlier judgment for various reconstructed component data at different times; The abnormal data module determines whether the various reconstructed component data at each time point are abnormal data based on the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of various reconstructed component data at each time point.

[0049] According to one embodiment of the present invention, a computer-readable storage medium is provided having computer program instructions stored thereon, which, when executed by a processor, implement the outlier screening method based on particulate matter composition reconstruction.

[0050] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0051] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An outlier screening method based on particulate matter composition reconstruction, characterized in that, include: The system acquires ionic component data, carbon component data, and inorganic element component data from multiple moments within a preset time period from the component monitoring station, and performs data preprocessing to obtain standard ionic component data, standard carbon component data, and standard inorganic element component data. Based on standard ionic component data, standard carbon component data, and standard inorganic element component data, multiple reconstructed component data were obtained; Based on multiple reconstructed component data and the location information of component monitoring stations, the target PM was determined. 2.5 data; Based on multiple reconstructed component data and target PM 2.5 Data, determining the judgment results of questionable reconstructed component data of multiple reconstructed component data at various time points; Set a first sliding time window within a preset time period, and determine the univariate time series outlier judgment result of each reconstructed component data at each time point based on the first sliding time window; Set a second sliding time window within a preset time period, and base the second sliding time window on the target PM. 2.5 Data, determining the results of multivariate time series outlier judgment for various reconstructed component data at different times; Based on the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of various reconstructed component data at each time point, it is determined whether various reconstructed component data at each time point are outliers.

2. The outlier screening method based on particulate matter composition reconstruction according to claim 1, characterized in that, The reconstructed component data includes: organic component data, nitrate component data, sulfate component data, ammonium salt component data, elemental carbon component data, chloride salt component data, crustal material component data, and trace element component data.

3. The outlier screening method based on particulate matter composition reconstruction according to claim 1, characterized in that, Based on multiple reconstructed component data and the location information of component monitoring stations, the target PM was determined. 2.5 Data, including: Based on multiple reconstructed component data, obtain the summation data of the reconstructed components at each time step; The first PM2.5 concentrations at multiple times within a preset time period were obtained from air monitoring stations at the same location as the component monitoring stations. 2.5 Data, the second PM at multiple times within a preset time period for air monitoring stations located at distances less than a preset distance from the component monitoring station. 2.5 Data, and the third PM at multiple times within a preset time period from air monitoring stations located in the same city. 2.5 data; Determine the summation data of the reconstructed components at each time step and compare it with the first PM at the same time step. 2.5 The first difference of the data, the sum of the reconstructed components at each time step, and the second PM at the same time step. 2.5 The second difference of the data, and the summation of the reconstructed components at each time step with the third PM at the same time step. 2.5 The third difference in the data; Based on the first difference, second difference, and third difference, determine the target PM corresponding to the summed data of the reconstructed components at each time step. 2.5 data.

4. The outlier screening method based on particulate matter composition reconstruction according to claim 3, characterized in that, Based on multiple reconstructed component data and target PM 2.5 Data, determining the judgment results of questionable reconstructed component data for various reconstructed component data at each time point, including: Obtain the summation data of the reconstructed components and the target PM 2.5 The first proportional value of the data; Based on the first ratio value, identify outlier data in the reconstructed component summation data and remove outlier data; In the reconstructed component summation data after removing outlier data, based on the corresponding target PM 2.5 Data is used to determine the normal range of the first ratio value; Based on the normal ratio range and the first ratio value, the judgment results of the questionable reconstructed component data of various reconstructed component data at each time point are determined.

5. The outlier screening method based on particulate matter composition reconstruction according to claim 4, characterized in that, Based on the first ratio value, outlier data in the reconstructed component summation data are identified, including: Arrange the first proportion values ​​at each time point in order of magnitude to obtain the first proportion value sequence; Based on the first proportional value sequence, determine the 25th percentile and the 75th percentile; Determine the fourth difference between the 25th percentile and the 75th percentile; The first proportion range is determined based on the 25th percentile, the 75th percentile, and the fourth difference. The reconstructed component data corresponding to the summation data of the reconstructed components whose first proportion value does not belong to the first proportion range are identified as outlier data.

6. The outlier screening method based on particulate matter composition reconstruction according to claim 1, characterized in that, A first sliding time window is set within a preset time period, and based on the first sliding time window, the univariate time series outlier judgment results for each reconstructed component data at each time point are determined, including: The neighborhood parameters are determined based on the duration of the first sliding time window; When the center time of the first sliding time window is aligned with the t-th time within the preset time period, the local anomaly factor value of the reconstructed component data at the center time within the first sliding time window is determined based on the neighborhood parameter. Based on the comparison between the local anomaly factor values ​​and the preset factor thresholds, the univariate time-series anomaly judgment result of the reconstructed component data at time t is determined.

7. The outlier screening method based on particulate matter composition reconstruction according to claim 6, characterized in that, Based on the duration of the first sliding time window, the neighborhood parameters are determined, including: According to the formula k=max(3,min(k_max,floor(w×ρ))) Determine the neighborhood parameter k, where k_max is the preset upper limit, w is the duration of the first sliding time window, ρ is the preset coefficient, min is the minimum value function, max is the maximum value function, and floor is the floor function.

8. The outlier screening method based on particulate matter composition reconstruction according to claim 6, characterized in that, Set a second sliding time window within a preset time period, and base the second sliding time window on the target PM. 2.5 Data, determining the results of multivariate time-series outlier identification for various reconstructed component data at different times, including: When the center time of the second sliding time window aligns with the t-th time within the preset time period, the summation data of the reconstruction components at each time point of the second sliding time window is obtained based on multiple reconstruction component data. Determine the reconstructed component summation data and target PM at each moment of the second sliding time window. 2.5 Pearson correlation coefficient of the data; Based on the comparison between the Pearson correlation coefficient and the preset coefficient threshold, the results of the multivariate time series outlier judgment for the various reconstructed component data at time t are determined.

9. The outlier screening method based on particulate matter composition reconstruction according to claim 1, characterized in that, Based on the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results at each time point, determine whether the various reconstructed component data at each time point are outliers, including: If the judgment results of the questionable reconstructed component data of the j-th reconstructed component data at time t, the judgment results of the univariate time series outlier and the judgment results of the multivariate time series outlier are all abnormal, the j-th reconstructed component data at time t will be identified as abnormal data.

10. An outlier screening system based on particulate matter composition reconstruction, characterized in that, include: The acquisition and preprocessing module acquires ionic component data, carbon component data, and inorganic element component data from the component monitoring station at multiple moments within a preset time period in the past, and performs data preprocessing to obtain standard ionic component data, standard carbon component data, and standard inorganic element component data. The reconstructed component data module obtains various reconstructed component data based on standard ionic component data, standard carbon component data, and standard inorganic element component data; Target PM 2.5 The data module, based on various reconstructed component data and the location information of component monitoring stations, determines the target PM. 2.5 data; The first judgment result module, based on multiple reconstructed component data and the target PM... 2.5 Data, determining the judgment results of questionable reconstructed component data of multiple reconstructed component data at various time points; The second judgment result module sets a first sliding time window within a preset time period, and determines the judgment result of the univariate time series outlier point of each reconstructed component data at each time point based on the first sliding time window. The third judgment result module sets a second sliding time window within a preset time period, and determines the result based on the second sliding time window and the target PM. 2.5 Data, determining the results of multivariate time series outlier judgment for various reconstructed component data at different times; The abnormal data module determines whether the various reconstructed component data at each time point are abnormal data based on the judgment results of questionable reconstructed component data, univariate time series outlier judgment results, and multivariate time series outlier judgment results of various reconstructed component data at each time point.

Images