A system and method for identifying abnormal data of fixed pollution sources based on deep learning

Abstract

The present invention relates to the field of pollution source data identification, and specifically to a system and method for identifying abnormal data of fixed pollution sources based on deep learning, comprising a decision tree model construction module, an effective decision tree model analysis module, an abnormal pollution source early warning module, and a second suspicious pollution source analysis module; the decision tree model construction module is used to analyze the relationship between pollution source attributes and categories to construct a decision tree model that can distinguish the categories of fixed pollution sources; the effective decision tree model analysis module is used to verify that the decision tree model with the highest accuracy is selected as the effective decision tree model for classifying fixed pollution sources; the abnormal pollution source early warning module is used to analyze the pollution data in each type of fixed pollution source by using a K-means clustering analysis algorithm and to warn the abnormal pollution source; the second suspicious pollution source analysis module is used to re-warn the associated second suspicious pollution source when there is any fixed pollution source early warning that meets the abnormal response interval during real-time monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of pollution source data identification, and specifically to a system and method for identifying abnormal data of fixed pollution sources based on deep learning. Background Art

[0002] Fixed pollution sources refer to various industries, enterprises, production facilities, fixed equipment, etc. that emit atmospheric pollutants, referred to as fixed sources. Fixed sources are coal-fired, oil-fired, gas-fired boilers and industrial furnaces, as well as waste gas generated in the production process of petrochemicals, metallurgy, building materials, etc., which is discharged into the air through exhaust pipes; fixed pollution source data mainly consists of online pollution source monitoring, pollution discharge permits, environmental statistics, pollution census and other data. The existing technology for monitoring fixed pollution source data is mostly based on the validity, authenticity and traceability of the data, and the data itself is fully processed with modern data analysis methods; but at the same time, the diversity of pollution source data is fundamentally due to the unclear classification of pollution sources, and when abnormal data appears, it is not sensitive to the type of pollution source to which the data belongs and the pollution sources that may be associated and affected. As a result, many existing pollution source abnormal data identifications are based on the identification of abnormal events that have occurred, and cannot effectively use data analysis of abnormal events to achieve forward-looking warnings, so as to reduce the risk of pollution source abnormalities. Summary of the invention

[0003] The purpose of the present invention is to provide a system and method for identifying abnormal data of fixed pollution sources based on deep learning to solve the problems raised in the above-mentioned background technology.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for identifying abnormal data of fixed pollution sources based on deep learning, comprising the following analysis steps:

[0005] Step S100: obtaining online monitoring data recorded by each fixed pollution source in the same monitoring period, and analyzing the relationship between pollution source attributes and categories based on the online monitoring data to construct a decision tree model that can distinguish the categories of fixed pollution sources;

[0006] Step S200: obtaining a number of decision tree models constructed using a number of decision tree classification algorithms, verifying and selecting the decision tree model with the highest accuracy as an effective decision tree model for classifying fixed pollution sources;

[0007] Step S300: Based on each type of fixed pollution source in the effective decision tree model, the pollution data in each type of fixed pollution source is analyzed using the K-means cluster analysis algorithm to warn of abnormal pollution sources;

[0008] Step S400: extracting the warning response data of abnormal pollution sources within the monitoring period to construct an abnormal response interval corresponding to the fixed pollution source, and re-warning the associated second suspicious pollution source when there is any fixed pollution source warning that meets the abnormal response interval during real-time monitoring.

[0009] Furthermore, building a decision tree model that can distinguish the categories of fixed pollution sources includes the following steps:

[0010] Step S110: divide the online monitoring data into training data and test data according to the ratio a:b, and a≥b. The decision tree model refers to a tree structure composed of non-leaf nodes, branches and leaf nodes; each non-leaf node represents a test on a feature attribute, each branch represents the output of this feature attribute in a certain value range, and each leaf node stores a category;

[0011] Step S120: based on the online monitoring data in the training data, the characteristic attributes of the online monitoring data are marked, the corresponding characteristic attributes are selected using the decision tree algorithm starting from the root node, and the output branch is selected according to the value corresponding to the online monitoring data until the leaf node is reached, and the category stored in the leaf node is used as the decision result;

[0012] Step S130: traverse all online monitoring data in the training data until all characteristic attributes are stored in corresponding non-leaf nodes, forming a decision tree model that can distinguish the categories of fixed pollution sources.

[0013] Further, step S200 includes the following steps:

[0014] Step S210: The decision tree model includes a number of decision paths, where a decision path refers to a path starting from a root node, passing through corresponding branches until reaching a leaf node;

[0015] Step S220: Substitute the same test data into each decision tree model respectively, traverse and match the test data with the same characteristic attributes corresponding to the root node in each decision tree model, locate the decision path that satisfies the branch output value and the characteristic attributes in the non-leaf node before reaching the leaf node based on the online monitoring value corresponding to the test data as the target decision path, mark the category corresponding to the leaf node in the target decision path as the target category; compare the target category with the fixed pollutant category corresponding to the test data;

[0016] When the target category is the same as the fixed pollutant category corresponding to the test data, the decision tree model outputs a correct judgment signal; when the target category is different from the fixed pollutant category corresponding to the test data, the decision tree model outputs an incorrect judgment signal;

[0017] Step S230: extract the number of correct signal judgments and the number of wrong signal judgments corresponding to the output of the i-th decision tree model Fi , using the formula;

[0018] Q i =T i / (T i +F i )

[0019] Calculate the decision accuracy Q of the i-th decision tree model i ; Select the maximum decision accuracy rate Q max The corresponding decision tree model serves as an effective decision tree model for the classification of fixed pollution sources.

[0020] The purpose of analyzing the effective decision tree model is to select the classification model that has the highest degree of accurate matching with the current pollution source data, so that the classification reliability of the pollution source data characteristics is higher, and the error rate caused by the difficulty in determining the K value in the subsequent K-means analysis method can be minimized.

[0021] Further, step S300 includes the following steps:

[0022] Step S310: Determine the number of cluster centers K based on the categories of fixed pollution sources in the effective decision tree model, and randomly select K pollution sources in the data set as initial cluster centers. The data set refers to all online monitoring data recorded for each type of fixed pollution source;

[0023] Step S320: Calculate the distances from the online monitoring data to the K cluster centers in sequence, select the center with the smallest cluster, and divide the calculated online monitoring data into corresponding categories; calculate the center point of each cluster, and update the center point to the new cluster center point of the corresponding category;

[0024] Step S330: Repeat steps S310 to S320 continuously until the data division stops changing and the loop is stopped, so as to realize intelligent abnormal monitoring of pollution sources and classify outliers as abnormal pollution sources; outliers refer to points in the data set whose numerical differences compared with other data points are greater than the difference threshold; based on the abnormal pollution sources, the corresponding fixed pollution sources are determined and early warnings are issued.

[0025] Furthermore, step S400 includes the following steps:

[0026] The early warning response data includes the number of early warning responses and the pollution value required for early warning. The monitoring period refers to the shortest monitoring time that includes at least one early warning response for each type of fixed pollution source; the pollution value required for early warning refers to the value corresponding to the outlier output in the K-means analysis.

[0027] Obtain the maximum value W of the pollution value W required for early warning for each type of fixed pollution source record within the monitoring period max and the minimum value W min, and constitute the abnormal response interval A corresponding to the fixed pollution source, A=[W min ,W max ];

[0028] When the pollution source data acquired through real-time monitoring belongs to the abnormal response interval A corresponding to any fixed pollution source, an early warning signal is transmitted, and a second early warning is issued for the associated second suspected pollution source;

[0029] The associated second suspected contamination source includes the following analysis process:

[0030] Sort each type of fixed pollution source by the number of warnings from large to small, and select the key decision path recorded by the fixed pollution source with the highest ranking. The key decision path refers to the path with the most non-leaf nodes when there are multiple decision paths for the same fixed pollution source; when the path with the most non-leaf nodes is not unique, select any one of them as the key decision path;

[0031] Extract other decision paths that have the same path part as the key decision path and correspond to different fixed pollution source categories as the decision paths to be analyzed, and sort the decision paths to be analyzed from large to small according to the number of non-leaf nodes contained in the same path part to generate a first sequence;

[0032] Based on the first sequence, record the number of non-leaf nodes D contained in the decision path to be analyzed 1 The number of warning responses of fixed pollution sources corresponding to the decision path to be analyzed P 1 Calculate the difference P of the early warning response when each decision path to be analyzed corresponds to a fixed pollution source with the key decision path 0 , P 0 =P 2 -P 1 , where P 2 Indicates the total number of early warning responses corresponding to fixed pollution sources in the key decision-making path;

[0033] And constitute the data set R corresponding to the fixed pollution source for each decision path to be analyzed in the first sequence i , R i =(D 1i ,P 0i ), where R i represents the i-th data group in the first sequence, D 1i represents the number of non-leaf nodes contained in the i-th decision path to be analyzed in the first sequence, P 0i Indicates the difference in early warning response between the i-th decision path to be analyzed and the key decision path corresponding to the fixed pollution source in the first sequence; obtains M data groups in the first sequence, i≤M; uses the formula:

[0034] r=Cov(D,P) / (σD*σP),

[0035] Calculate the correlation index r between the number of non-leaf nodes in the first sequence and the difference in early warning response, where

[0036]

[0037] D 0 Indicates that the data group contains M D 1i The average value, P 0 Indicates that the data group contains M P 0i The average value of

[0038] When |r|≥r 0 When the number of non-leaf nodes in the decision path to be analyzed is output, there is a correlation with the difference in early warning response, r 0 Indicates the preset correlation index threshold; the existence of a correlation relationship means that when the fixed pollution source corresponding to the key decision path is warned, the risk possibility of abnormal warning of the fixed pollution source corresponding to other decision paths with a high overlap with the key decision path is higher or lower;

[0039] When r>0, the fixed pollution source corresponding to the last decision path to be analyzed in the first sequence is selected as the second suspicious pollution source; when r<0, the fixed pollution source corresponding to the first decision path to be analyzed in the first sequence is selected as the second suspicious pollution source.

[0040] The purpose of analyzing the second suspected pollution source is to capture other pollution sources that may have related anomalies when any pollution source has an abnormality, which provides managers with the time cost of handling anomalies. At the same time, it minimizes the irreversibility of handling pollution sources caused by abnormalities that have occurred when real-time monitoring uses cluster analysis, and improves the accuracy of analysis in the early warning interval, so that the recognition system can process and identify abnormal data in a diverse, comprehensive and intelligent manner.

[0041] A system for identifying abnormal data of fixed pollution sources based on deep learning, comprising a decision tree model building module, an effective decision tree model analysis module, an abnormal pollution source early warning module and a second suspicious pollution source analysis module;

[0042] The decision tree model building module is used to analyze the relationship between pollution source attributes and categories to build a decision tree model that can distinguish fixed pollution source categories;

[0043] The effective decision tree model analysis module is used to verify and select the decision tree model with the highest accuracy as the effective decision tree model for the classification of fixed pollution sources;

[0044] The abnormal pollution source warning module is used to analyze the pollution data in each type of fixed pollution source and warn of abnormal pollution sources using the K-means cluster analysis algorithm;

[0045] The second suspicious pollution source analysis module is used to re-warn the associated second suspicious pollution source when there is any fixed pollution source warning that meets the abnormal response interval during real-time monitoring.

[0046] Furthermore, the effective decision tree model analysis module includes a data classification unit, a data traversal unit, a signal output unit and a decision accuracy calculation unit;

[0047] The data classification unit is used to divide the online monitoring data into training data and test data according to the proportion;

[0048] The data traversal unit is used to substitute the same test data into each decision tree model, traverse and match the test data with the same characteristic attributes corresponding to the root node in each decision tree model; locate the decision path that satisfies the branch output value and the characteristic attributes in the non-leaf node before reaching the leaf node based on the online monitoring value corresponding to the test data as the target decision path, mark the category corresponding to the leaf node in the target decision path as the target category; compare the target category with the fixed pollutant category corresponding to the test data;

[0049] The signal output unit is used to output a corresponding signal based on the comparison result;

[0050] The decision accuracy calculation unit is used to calculate the decision accuracy based on the output signal, and select the decision tree model corresponding to the maximum decision accuracy as the effective decision tree model for the classification of fixed pollution sources.

[0051] Further, the second suspected pollution source analysis module includes an abnormal response interval analysis unit, a key decision path determination unit, a first sequence generation unit and a correlation analysis unit;

[0052] The abnormal response interval analysis unit is used to obtain the maximum and minimum values ​​of the pollution values ​​required for early warning for each type of fixed pollution source record within the monitoring period, and to form the abnormal response interval corresponding to the fixed pollution source;

[0053] The key decision path determination unit is used to sort each type of fixed pollution source from large to small according to the number of warnings, and select the key decision path recorded by the fixed pollution source with the first ranking;

[0054] The first sequence generating unit is used to extract other decision paths that have the same path part as the key decision path and have different corresponding fixed pollution source categories as the decision paths to be analyzed, and sort the decision paths to be analyzed from large to small according to the number of non-leaf nodes contained in the same path part to generate a first sequence;

[0055] The association analysis unit is used to analyze the second suspected pollution source with an associated relationship based on the first sequence.

[0056] Further, the correlation relationship analysis unit includes an early warning response difference calculation unit, a data group construction unit, a correlation index calculation unit and an early warning analysis unit;

[0057] The early warning response difference calculation unit is used to calculate the early warning response difference when each decision path to be analyzed corresponds to a fixed pollution source with the key decision path;

[0058] The data group construction unit is used to construct a data group corresponding to a fixed pollution source for each decision path to be analyzed in the first sequence;

[0059] The correlation index calculation unit is used to calculate the correlation index between the number of non-leaf nodes in the first sequence and the difference between the early warning response;

[0060] The early warning analysis unit is used to output the correlation relationship between the number of non-leaf nodes of the decision path to be analyzed and the early warning response difference when the absolute value of the correlation index is greater than or equal to the correlation index threshold, and to issue a re-warning based on the second suspicious pollution source corresponding to the positive and negative outputs of the correlation index.

[0061] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention constructs a classifiable decision tree model through online monitoring recorded by each fixed pollution source within the same monitoring period, and selects an effective decision tree model based on verification, and uses the K-means clustering analysis algorithm to warn of abnormal pollution sources after classification, thereby solving the possibility of high errors caused by the uncertainty of the K value in the clustering analysis method, and making the early warning analysis more accurate; at the same time, the present application analyzes other fixed pollution sources that may be associated based on an early warning, and makes forward-looking considerations before the pollution source actually has an abnormality, thereby providing managers with the time cost of handling anomalies, while minimizing the irreversibility of handling pollution sources caused by abnormalities that have occurred when clustering analysis is used for real-time monitoring, and improving the accuracy of analysis in the early warning interval, thereby making the recognition system's processing and identification of abnormal data diverse, comprehensive, and intelligent. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0063] Figure 1 It is a structural schematic diagram of a system and method for identifying abnormal data of fixed pollution sources based on deep learning in the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] See also Figure 1 The present invention provides a technical solution: a method for identifying abnormal data of fixed pollution sources based on deep learning, comprising the following analysis steps:

[0066] Step S100: obtaining online monitoring data recorded by each fixed pollution source in the same monitoring period, and analyzing the relationship between pollution source attributes and categories based on the online monitoring data to construct a decision tree model that can distinguish the categories of fixed pollution sources;

[0067] Step S200: obtaining a number of decision tree models constructed using a number of decision tree classification algorithms, verifying and selecting the decision tree model with the highest accuracy as an effective decision tree model for classifying fixed pollution sources;

[0068] Step S300: Based on each type of fixed pollution source in the effective decision tree model, the pollution data in each type of fixed pollution source is analyzed using the K-means cluster analysis algorithm to warn of abnormal pollution sources;

[0069] Step S400: extracting the warning response data of abnormal pollution sources within the monitoring period to construct an abnormal response interval corresponding to the fixed pollution source, and re-warning the associated second suspicious pollution source when there is any fixed pollution source warning that meets the abnormal response interval during real-time monitoring.

[0070] Building a decision tree model that can distinguish between stationary pollution source categories involves the following steps:

[0071] Step S110: divide the online monitoring data into training data and test data according to the ratio a:b, and a≥b. The decision tree model refers to a tree structure composed of non-leaf nodes, branches and leaf nodes; each non-leaf node represents a test on a feature attribute, each branch represents the output of this feature attribute in a certain value range, and each leaf node stores a category;

[0072] Step S120: based on the online monitoring data in the training data, the characteristic attributes of the online monitoring data are marked, and the corresponding characteristic attributes are selected from the root node using the decision tree algorithm, and the output branch is selected according to the numerical value corresponding to the online monitoring data until the leaf node is reached, and the category stored in the leaf node is used as the decision result; the decision tree algorithm includes the ID3 algorithm, the C4.5 algorithm and the CART algorithm; the selection of the characteristic attribute corresponding to the node is based on the algorithm to select the optimal partition attribute, indicating the feature with the strongest classification ability;

[0073] Step S130: traverse all online monitoring data in the training data until all characteristic attributes are stored in corresponding non-leaf nodes, forming a decision tree model that can distinguish the categories of fixed pollution sources.

[0074] The decision tree model is a model obtained by analyzing the same training data based on different algorithms.

[0075] Step S200 includes the following steps:

[0076] Step S210: The decision tree model includes a number of decision paths, where a decision path refers to a path starting from a root node, passing through corresponding branches until reaching a leaf node;

[0077] Step S220: Substitute the same test data into each decision tree model respectively, traverse and match the test data with the same characteristic attributes corresponding to the root node in each decision tree model, locate the decision path that satisfies the branch output value and the characteristic attributes in the non-leaf node before reaching the leaf node based on the online monitoring value corresponding to the test data as the target decision path, mark the category corresponding to the leaf node in the target decision path as the target category; compare the target category with the fixed pollutant category corresponding to the test data;

[0078] When the target category is the same as the fixed pollutant category corresponding to the test data, the decision tree model outputs a correct judgment signal; when the target category is different from the fixed pollutant category corresponding to the test data, the decision tree model outputs an incorrect judgment signal;

[0079] Step S230: extract the number of times T that the i-th decision tree model outputs the correct judgment signal i and the number of false signals F i , using the formula;

[0080] Q i =T i / (T i +F i )

[0081] Calculate the decision accuracy Q of the i-th decision tree model i ; Select the maximum decision accuracy rate Q maxThe corresponding decision tree model is used as an effective decision tree model for classifying the fixed pollution sources;

[0082] The purpose of analyzing the effective decision tree model is to select the classification model that has the highest degree of accurate matching with the current pollution source data, so that the classification reliability of the pollution source data characteristics is higher, and the error rate caused by the difficulty in determining the K value in the subsequent K-means analysis method can be minimized.

[0083] Step S300 includes the following steps:

[0084] Step S310: Determine the number of cluster centers K based on the categories of fixed pollution sources in the effective decision tree model, and randomly select K pollution sources in the data set as initial cluster centers. The data set refers to all online monitoring data recorded for each type of fixed pollution source;

[0085] Step S320: Calculate the distances from the online monitoring data to the K cluster centers in sequence, select the center with the smallest cluster, and divide the calculated online monitoring data into corresponding categories; calculate the center point of each cluster, and update the center point to the new cluster center point of the corresponding category;

[0086] Step S330: Repeat steps S310 to S320 continuously until the data division stops changing and the loop is stopped, so as to realize intelligent abnormal monitoring of pollution sources and classify outliers as abnormal pollution sources; outliers refer to points in the data set whose numerical differences compared with other data points are greater than the difference threshold; based on the abnormal pollution sources, the corresponding fixed pollution sources are determined and early warnings are issued.

[0087] Step S400 includes the following steps:

[0088] The early warning response data includes the number of early warning responses and the pollution value required for early warning. The monitoring period refers to the shortest monitoring time that includes at least one early warning response for each type of fixed pollution source; the pollution value required for early warning refers to the value corresponding to the outlier output in the K-means analysis.

[0089] Obtain the maximum value W of the pollution value W required for early warning for each type of fixed pollution source record within the monitoring period max and the minimum value W min , and constitute the abnormal response interval A corresponding to the fixed pollution source, A=[W min ,W max ];

[0090] When the pollution source data acquired through real-time monitoring belongs to the abnormal response interval A corresponding to any fixed pollution source, an early warning signal is transmitted, and a second early warning is issued for the associated second suspected pollution source;

[0091] The associated second suspected contamination source includes the following analysis process:

[0092] Sort each type of fixed pollution source by the number of warnings from large to small, and select the key decision path recorded by the fixed pollution source with the highest ranking. The key decision path refers to the path with the most non-leaf nodes when there are multiple decision paths for the same fixed pollution source; when the path with the most non-leaf nodes is not unique, select any one of them as the key decision path;

[0093] Extract other decision paths that have the same path part as the key decision path and correspond to different fixed pollution source categories as the decision paths to be analyzed, and sort the decision paths to be analyzed from large to small according to the number of non-leaf nodes contained in the same path part to generate a first sequence;

[0094] Based on the first sequence, record the number of non-leaf nodes D contained in the decision path to be analyzed 1 The number of warning responses of fixed pollution sources corresponding to the decision path to be analyzed P 1 Calculate the difference P of the early warning response when each decision path to be analyzed corresponds to a fixed pollution source with the key decision path 0 , P 0 =P 2 -P 1 , where P 2 Indicates the total number of early warning responses corresponding to fixed pollution sources in the key decision-making path;

[0095] And constitute the data set R corresponding to the fixed pollution source for each decision path to be analyzed in the first sequence i , R i =(D 1i ,P 0i ), where R i represents the i-th data group in the first sequence, D 1i represents the number of non-leaf nodes contained in the i-th decision path to be analyzed in the first sequence, P 0i Indicates the difference in early warning response between the i-th decision path to be analyzed and the key decision path corresponding to the fixed pollution source in the first sequence; obtains M data groups in the first sequence, i≤M; uses the formula:

[0096] r=Cov(D,P) / (σD*σP),

[0097] Calculate the correlation index r between the number of non-leaf nodes in the first sequence and the difference in early warning response, where

[0098]

[0099] D 0 Indicates that the data group contains M D 1i The average value, P 0 Indicates that the data group contains M P 0i The average value of

[0100] When |r|≥r 0 When the number of non-leaf nodes in the decision path to be analyzed is output, there is a correlation with the difference in early warning response, r 0 It indicates the preset correlation index threshold, which is generally set to 0.5. The existence of a correlation relationship means that when the fixed pollution source corresponding to the key decision path is warned, the risk possibility of abnormal warning of fixed pollution sources corresponding to other decision paths with high overlap with the key decision path is higher or lower;

[0101] When |r| <r 0 When , there is no correlation between the number of non-leaf nodes in the decision path to be analyzed and the difference in warning response; when there is no correlation, it is still necessary to screen out outliers based on K-means clustering analysis for warning.

[0102] When r>0, the fixed pollution source corresponding to the last decision path to be analyzed in the first sequence is selected as the second suspicious pollution source; when r<0, the fixed pollution source corresponding to the first decision path to be analyzed in the first sequence is selected as the second suspicious pollution source.

[0103] r>0 indicates positive correlation, that is, the more non-leaf nodes overlap, the greater the difference in warning response. When the key decision path corresponds to the warning response of a fixed pollution source, the more fixed pollution sources with overlapping paths are less likely to need warning. Therefore, when there is a positive correlation, the fixed pollution source corresponding to the decision path to be analyzed with the least number of overlaps should be selected as the pollution source most likely to be warned. Similarly, the same analysis applies to r<0.

[0104] The purpose of analyzing the second suspected pollution source is to capture other pollution sources that may have related anomalies when any pollution source has an abnormality, which provides managers with the time cost of handling anomalies. At the same time, it minimizes the irreversibility of handling pollution sources caused by abnormalities that have occurred when real-time monitoring uses cluster analysis, and improves the accuracy of analysis in the early warning interval, so that the recognition system can process and identify abnormal data in a diverse, comprehensive and intelligent manner.

[0105] A system for identifying abnormal data of fixed pollution sources based on deep learning, comprising a decision tree model building module, an effective decision tree model analysis module, an abnormal pollution source early warning module and a second suspicious pollution source analysis module;

[0106] The decision tree model building module is used to analyze the relationship between pollution source attributes and categories to build a decision tree model that can distinguish fixed pollution source categories;

[0107] The effective decision tree model analysis module is used to verify and select the decision tree model with the highest accuracy as the effective decision tree model for the classification of fixed pollution sources;

[0108] The abnormal pollution source warning module is used to analyze the pollution data in each type of fixed pollution source and warn of abnormal pollution sources using the K-means cluster analysis algorithm;

[0109] The second suspicious pollution source analysis module is used to re-warn the associated second suspicious pollution source when there is any fixed pollution source warning that meets the abnormal response interval during real-time monitoring.

[0110] The effective decision tree model analysis module includes a data classification unit, a data traversal unit, a signal output unit and a decision accuracy calculation unit;

[0111] The data classification unit is used to divide the online monitoring data into training data and test data according to the proportion;

[0112] The data traversal unit is used to substitute the same test data into each decision tree model, traverse and match the test data with the same characteristic attributes corresponding to the root node in each decision tree model; locate the decision path that satisfies the branch output value and the characteristic attributes in the non-leaf node before reaching the leaf node based on the online monitoring value corresponding to the test data as the target decision path, mark the category corresponding to the leaf node in the target decision path as the target category; compare the target category with the fixed pollutant category corresponding to the test data;

[0113] The signal output unit is used to output a corresponding signal based on the comparison result;

[0114] The decision accuracy calculation unit is used to calculate the decision accuracy based on the output signal, and select the decision tree model corresponding to the maximum decision accuracy as the effective decision tree model for the classification of fixed pollution sources.

[0115] The second suspected pollution source analysis module includes an abnormal response interval analysis unit, a key decision path determination unit, a first sequence generation unit and a correlation analysis unit;

[0116] The abnormal response interval analysis unit is used to obtain the maximum and minimum values ​​of the pollution values ​​required for early warning for each type of fixed pollution source record within the monitoring period, and to form the abnormal response interval corresponding to the fixed pollution source;

[0117] The key decision path determination unit is used to sort each type of fixed pollution source from large to small according to the number of warnings, and select the key decision path recorded by the fixed pollution source with the first ranking;

[0118] The first sequence generating unit is used to extract other decision paths that have the same path part as the key decision path and have different corresponding fixed pollution source categories as the decision paths to be analyzed, and sort the decision paths to be analyzed from large to small according to the number of non-leaf nodes contained in the same path part to generate a first sequence;

[0119] The association analysis unit is used to analyze the second suspected pollution source with an associated relationship based on the first sequence.

[0120] The correlation relationship analysis unit includes an early warning response difference calculation unit, a data group construction unit, a correlation index calculation unit and an early warning analysis unit;

[0121] The early warning response difference calculation unit is used to calculate the early warning response difference when each decision path to be analyzed corresponds to a fixed pollution source with the key decision path;

[0122] The data group construction unit is used to construct a data group corresponding to a fixed pollution source for each decision path to be analyzed in the first sequence;

[0123] The correlation index calculation unit is used to calculate the correlation index between the number of non-leaf nodes in the first sequence and the difference between the early warning response;

[0124] The early warning analysis unit is used to output the correlation relationship between the number of non-leaf nodes of the decision path to be analyzed and the early warning response difference when the absolute value of the correlation index is greater than or equal to the correlation index threshold, and to issue a re-warning based on the second suspicious pollution source corresponding to the positive and negative outputs of the correlation index.

[0125] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0126] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying abnormal data of fixed pollution sources based on deep learning, characterized in that: The analysis steps include: Step S100: obtaining online monitoring data recorded by each fixed pollution source in the same monitoring period, and analyzing the relationship between pollution source attributes and categories based on the online monitoring data to construct a decision tree model that can distinguish the categories of fixed pollution sources; Step S200: obtaining a number of decision tree models constructed using a number of decision tree classification algorithms, verifying and selecting the decision tree model with the highest accuracy as an effective decision tree model for classifying fixed pollution sources; The step S200 includes the following steps: Step S210: the decision tree model includes a plurality of decision paths, wherein the decision path refers to a path starting from a root node, passing through corresponding branches until reaching a leaf node; Step S220: Substitute the same test data into each decision tree model respectively, traverse and match the test data with the same characteristic attributes corresponding to the root node in each decision tree model, locate the decision path that satisfies the branch output value and the characteristic attributes in the non-leaf node before reaching the leaf node based on the online monitoring value corresponding to the test data as the target decision path, mark the category corresponding to the leaf node in the target decision path as the target category; compare the target category with the fixed pollutant category corresponding to the test data; When the target category is the same as the fixed pollutant category corresponding to the test data, the decision tree model outputs a correct judgment signal; when the target category is different from the fixed pollutant category corresponding to the test data, the decision tree model outputs an incorrect judgment signal; Step S230: extract the number of times T that the i-th decision tree model outputs the correct judgment signal i and the number of false signals F i , using the formula; Q i =T i / (T i +F i ) Calculate the decision accuracy Q of the i-th decision tree model i ; Select the maximum decision accuracy rate Q max The corresponding decision tree model is used as an effective decision tree model for classifying the fixed pollution sources; Step S300: Based on each type of fixed pollution source in the effective decision tree model, the pollution data in each type of fixed pollution source is analyzed using the K-means cluster analysis algorithm to warn of abnormal pollution sources; The step S300 includes the following steps: Step S310: Determine the number of cluster centers K based on the categories of fixed pollution sources in the effective decision tree model, and randomly select K pollution sources in the data set as initialization cluster centers, wherein the data set refers to all online monitoring data recorded for each type of fixed pollution source; Step S320: Calculate the distances from the online monitoring data to the K cluster centers in sequence, select the center with the smallest cluster, and divide the calculated online monitoring data into corresponding categories; calculate the center point of each cluster, and update the center point to the new cluster center point of the corresponding category; Step S330: Repeat steps S310 to S320 until the data division stops changing, so as to realize intelligent abnormal monitoring of pollution sources and classify outliers as abnormal pollution sources; the outliers refer to points in the data set whose values ​​differ from other data points by a value greater than a difference threshold; determine the corresponding fixed pollution sources based on the abnormal pollution sources and issue warnings; Step S400: extracting the warning response data of the abnormal pollution source within the monitoring period to construct the abnormal response interval corresponding to the fixed pollution source, and re-warning the associated second suspicious pollution source when there is any fixed pollution source warning that meets the abnormal response interval during real-time monitoring; The step S400 includes the following steps: The warning response data includes the number of warning responses and the pollution value required for warning. The monitoring period refers to the shortest monitoring time that includes at least one warning response for each type of fixed pollution source; the pollution value required for warning refers to the value corresponding to the outlier output in the K-means analysis; Obtain the maximum value W of the pollution value W required for early warning for each type of fixed pollution source record within the monitoring period max and the minimum value W min , and constitute the abnormal response interval A corresponding to the fixed pollution source, A=[W min ,W max ]; When the pollution source data acquired through real-time monitoring belongs to the abnormal response interval A corresponding to any fixed pollution source, an early warning signal is transmitted, and a second early warning is issued for the associated second suspected pollution source; The associated second suspected pollution source includes the following analysis process: Sort each type of fixed pollution source from large to small according to the number of warnings, and select the key decision path recorded by the fixed pollution source with the first ranking. The key decision path refers to the path with the most non-leaf nodes when there are multiple decision paths for the same fixed pollution source; when the path with the most non-leaf nodes is not unique, select any one as the key decision path; Extract other decision paths that have the same path part as the key decision path and correspond to different fixed pollution source categories as the decision paths to be analyzed, and sort the decision paths to be analyzed from large to small according to the number of non-leaf nodes contained in the same path part to generate a first sequence; Based on the number of non-leaf nodes D1 contained in the decision path to be analyzed and the number of warning responses P1 of the decision path to be analyzed corresponding to the fixed pollution source, the difference P0 of the warning response when each decision path to be analyzed and the key decision path correspond to the fixed pollution source is calculated, P0 = P2-P1, where P2 represents the total number of warning responses of the key decision path corresponding to the fixed pollution source; And constitute the data set R corresponding to the fixed pollution source for each decision path to be analyzed in the first sequence i , R i =(D 1i ,P 0i ), where R i represents the i-th data group in the first sequence, D 1i represents the number of non-leaf nodes contained in the i-th decision path to be analyzed in the first sequence, P 0i Indicates the difference in early warning response between the i-th decision path to be analyzed and the key decision path corresponding to the fixed pollution source in the first sequence; obtains M data groups in the first sequence, i≤M; uses the formula: r=Cov(D,P) / (σD*σP) Calculate the correlation index r between the number of non-leaf nodes in the first sequence and the difference in early warning response, where D0 means that the data set contains M D 1i The average value, P 0 Indicates that the data group contains M P 0i The average value of When |r|≥r0, the number of non-leaf nodes in the decision path to be analyzed is correlated with the difference in early warning response, where r0 represents a preset correlation index threshold; When r>0, the fixed pollution source corresponding to the last decision path to be analyzed in the first sequence is selected as the second suspicious pollution source; when r<0, the fixed pollution source corresponding to the first decision path to be analyzed in the first sequence is selected as the second suspicious pollution source.

2. According to claim 1, a method for identifying abnormal data of fixed pollution sources based on deep learning is characterized in that: The construction of a decision tree model capable of distinguishing the types of fixed pollution sources comprises the following steps: Step S110: divide the online monitoring data into training data and test data according to the ratio a:b, and a≥b, the decision tree model refers to a tree structure composed of non-leaf nodes, branches and leaf nodes; each non-leaf node represents a test on a feature attribute, each branch represents the output of this feature attribute in a certain value range, and each leaf node stores a category; Step S120: based on the online monitoring data in the training data, the characteristic attributes of the online monitoring data are marked, the corresponding characteristic attributes are selected using the decision tree algorithm starting from the root node, and the output branch is selected according to the value corresponding to the online monitoring data until the leaf node is reached, and the category stored in the leaf node is used as the decision result; Step S130: traverse all online monitoring data in the training data until all characteristic attributes are stored in corresponding non-leaf nodes, forming a decision tree model that can distinguish the categories of fixed pollution sources.

3. A system for identifying abnormal data of fixed pollution sources based on deep learning, applying a method for identifying abnormal data of fixed pollution sources based on deep learning as described in any one of claims 1-2, characterized in that: It includes a decision tree model building module, an effective decision tree model analysis module, an abnormal pollution source early warning module and a second suspicious pollution source analysis module; The decision tree model building module is used to analyze the relationship between pollution source attributes and categories to build a decision tree model that can distinguish fixed pollution source categories; The effective decision tree model analysis module is used to verify and select the decision tree model with the highest accuracy as the effective decision tree model for classifying fixed pollution sources; The abnormal pollution source warning module is used to analyze the pollution data in each type of fixed pollution source using the K-means cluster analysis algorithm and warn of abnormal pollution sources; The second suspicious pollution source analysis module is used to re-warn the associated second suspicious pollution source when there is any fixed pollution source warning that meets the abnormal response interval during real-time monitoring.

4. According to claim 3, a system for identifying abnormal data of fixed pollution sources based on deep learning is characterized in that: The effective decision tree model analysis module includes a data classification unit, a data traversal unit, a signal output unit and a decision accuracy calculation unit; The data classification unit is used to divide the online monitoring data into training data and test data according to the proportion; The data traversal unit is used to substitute the same test data into each decision tree model respectively, and traverse and match the test data with the same feature attributes corresponding to the root node in each decision tree model; Based on the online monitoring values ​​corresponding to the test data, the decision path that satisfies the branch output value and the characteristic attributes in the non-leaf node before reaching the leaf node is located as the target decision path, and the category corresponding to the leaf node in the target decision path is marked as the target category; the target category is compared with the fixed pollutant category corresponding to the test data; The signal output unit is used to output a corresponding signal based on the comparison result; The decision accuracy calculation unit is used to calculate the decision accuracy based on the output signal, and select the decision tree model corresponding to the maximum decision accuracy as the effective decision tree model for the classification of the fixed pollution sources.

5. According to claim 4, a system for identifying abnormal data of fixed pollution sources based on deep learning is characterized in that: The second suspected pollution source analysis module includes an abnormal response interval analysis unit, a key decision path determination unit, a first sequence generation unit and a correlation analysis unit; The abnormal response interval analysis unit is used to obtain the maximum and minimum pollution values ​​required for early warning for each type of fixed pollution source record within the monitoring period, and to form an abnormal response interval corresponding to the fixed pollution source; The key decision path determination unit is used to sort each type of fixed pollution source from large to small according to the number of warnings, and select the key decision path recorded by the fixed pollution source with the highest ranking; The first sequence generating unit is used to extract other decision paths that have the same path part as the key decision path and have different corresponding fixed pollution source categories as the decision paths to be analyzed, and sort the decision paths to be analyzed from large to small according to the number of non-leaf nodes contained in the same path part to generate a first sequence; The association analysis unit is used for analyzing a second suspected pollution source having an associated relationship based on the first sequence.

6. The system for identifying abnormal data of fixed pollution sources based on deep learning according to claim 5 is characterized in that: The association relationship analysis unit includes an early warning response difference calculation unit, a data group construction unit, an association index calculation unit and an early warning analysis unit; The warning response difference calculation unit is used to calculate the warning response difference when each decision path to be analyzed corresponds to a fixed pollution source with the key decision path; The data group construction unit is used to construct a data group corresponding to a fixed pollution source for each decision path to be analyzed in the first sequence; The correlation index calculation unit is used to calculate the correlation index between the number of non-leaf nodes in the first sequence and the difference between the early warning response; The early warning analysis unit is used to output the correlation relationship between the number of non-leaf nodes of the decision path to be analyzed and the early warning response difference when the absolute value of the correlation index is greater than or equal to the correlation index threshold, and to issue a re-warning based on the second suspicious pollution source corresponding to the positive and negative outputs of the correlation index.

Images