Environmental monitoring data anomaly detection method, medium and system

Abstract

The invention provides an environment monitoring data anomaly detection method, a medium and a system, and belongs to the technical field of electrical digital data process.The method comprises the steps that firstly, an environment monitoring data original matrix is constructed, time sequence segmentation is conducted, and a rule matrix and an anomaly matrix are obtained through matrix decomposition; calculating a historical mean value, a historical standard deviation, a spatial feature and a time feature of the environmental parameters, designing an anomaly detection optimization equation set based on the features, including an environmental parameter weight equation, an anomaly threshold equation and a comprehensive anomaly degree equation, and calculating an anomaly detection result by calculating an environmental parameter weight vector, a dynamic anomaly threshold matrix and a comprehensive anomaly degree index. Finally, dynamic optimization of anomaly detection is achieved through the gain vector and the feedback vector, anomaly detection of real-time environment monitoring data is completed, and the technical problem that in the prior art, an environment monitoring data anomaly detection method is difficult to give consideration to dynamic relevance of multi-dimensional environment parameters and the accuracy of anomaly detection at the same time is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electronic digital data processing, and in particular, relates to a method, medium and system for detecting anomalies in environmental monitoring data. Background Art

[0002] Environmental monitoring is an important means to ensure ecological security and environmental quality, involving real-time monitoring of environmental parameters in multiple dimensions such as air quality, water quality, soil, and noise. Traditional methods for detecting anomalies in environmental monitoring data mainly include anomaly detection methods based on statistical analysis, anomaly detection methods based on machine learning, and anomaly detection methods based on expert systems. Statistical analysis methods use variance analysis, correlation analysis, and other methods to identify outliers by establishing a data distribution model; machine learning methods use classifiers or clustering models to learn the normal patterns of data to detect anomalies; and expert systems make anomaly judgments based on a preset rule base. These methods have achieved certain results in detecting anomalies in single environmental parameters.

[0003] However, the existing technology has the following problems: First, environmental monitoring data is multidimensional and highly correlated, and there are complex mutual influence relationships between various environmental parameters. The anomaly of a single parameter often causes a chain reaction of other parameters, and most of the existing methods conduct independent analysis on a single parameter and cannot effectively capture the dynamic correlation between parameters; second, environmental parameters are affected by multiple external factors such as seasonal changes and weather conditions, showing obvious spatiotemporal variation characteristics, and the existing methods mostly use static thresholds or fixed models, which are difficult to adapt to the dynamic change characteristics of environmental parameters; third, there is a contradiction between the sensitivity and reliability of anomaly detection. Excessive sensitivity can easily lead to false alarms, while too low sensitivity may miss important anomalies, and the existing methods lack an effective adaptive adjustment mechanism.

[0004] Based on the above problems, how to achieve accurate and reliable anomaly detection based on the dynamic correlation of multi-dimensional environmental parameters has become a key technical problem that needs to be solved in the field of environmental monitoring. Existing technologies are difficult to take into account both the dynamic correlation of multi-dimensional environmental parameters and the accuracy of anomaly detection, which restricts the performance improvement of environmental monitoring systems. Summary of the invention

[0005] In view of this, the present invention provides an environmental monitoring data anomaly detection method, medium and system, which can solve the technical problem in the prior art that the environmental monitoring data anomaly detection method is difficult to simultaneously take into account the dynamic correlation of multi-dimensional environmental parameters and the accuracy of anomaly detection.

[0006] The present invention is implemented as follows: In a first aspect, the present invention provides an environmental monitoring data anomaly detection method comprising the following steps: constructing an original matrix of environmental monitoring data, wherein the original matrix of environmental monitoring data includes monitoring values ​​of multiple environmental parameters at different time points; performing time series segmentation on the original matrix of environmental monitoring data to obtain a training data submatrix; utilizing a matrix decomposition algorithm to decompose the training data submatrix into a regular matrix and an abnormal matrix; calculating the historical mean and historical standard deviation of each environmental parameter according to the regular matrix; calculating the spatial gradient, spatial correlation, and spatial distribution characteristics of each environmental parameter based on the historical mean; calculating the fluctuation range, sampling frequency, and change rate of each environmental parameter based on the historical standard deviation; obtaining the time continuity, typical fluctuation cycle, and seasonal change amplitude of each environmental parameter based on the regular matrix; utilizing an anomaly detection optimization equation group to calculate the degree of anomaly of data points in the anomaly matrix, wherein the anomaly detection optimization equation group includes an environmental parameter weight equation, an anomaly threshold equation, and a comprehensive anomaly degree equation; constructing a gain vector based on a comprehensive anomaly degree index; utilizing the gain vector to calculate a feedback vector; utilizing the feedback vector to optimize and adjust the anomaly detection optimization equation group; utilizing the optimized anomaly detection optimization equation group to perform anomaly detection on real-time environmental monitoring data.

[0007] Among them, the original matrix of environmental monitoring data is an m×n dimensional matrix, where m is the number of environmental parameters and n is the number of time sampling points; the regularity matrix represents the normal change pattern of environmental parameters; the abnormality matrix represents the abnormal fluctuation components of environmental parameters; the gain vector represents the deviation degree of abnormal data; and the feedback vector represents the correction coefficient of abnormal judgment.

[0008] Among them, in the step of constructing the original matrix of environmental monitoring data, the environmental monitoring data is obtained through the data acquisition system, and the sensor array is used to collect data including temperature, humidity, PM2.5, The environmental parameters of concentration and noise are recorded in real time according to the preset sampling frequency, and the collected data is stored in a two-dimensional matrix form.

[0009] Among them, in the step of segmenting the original matrix of environmental monitoring data into time series, the sliding window method is used for segmentation. The sliding window size is set to 24 hours, and the window sliding step is 1 hour. The representativeness of each sub-matrix is ​​determined by calculating its mean, variance, skewness and kurtosis.

[0010] Among them, in the step of decomposing the training data sub-matrix using the matrix decomposition algorithm, the robust principal component analysis algorithm is used for decomposition, the eigenvalues ​​and eigenvectors are obtained through the singular value decomposition method, and the principal components with a cumulative contribution rate of 85% are selected.

[0011] Among them, in the step of calculating the spatial characteristics of environmental parameters based on historical means, the central difference method is used to calculate the spatial gradient, the Pearson correlation coefficient method is used to construct the correlation matrix, and the Kriging interpolation method is used to construct the parameter spatial distribution model.

[0012] Among them, in the step of analyzing the dynamic characteristics of environmental parameters according to historical standard deviations, the percentile method is used to determine the upper and lower limits of parameter fluctuations, the Fourier transform method is used to analyze the signal spectrum, and the first-order difference method is used to calculate the rate of change.

[0013] Among them, in the step of analyzing the time characteristics of environmental parameters based on the regularity matrix, the autocorrelation function method is used to analyze the time correlation, the Mexican hat wavelet is used to analyze the typical fluctuation cycle, and the seasonal decomposition algorithm is used to extract the seasonal component.

[0014] A second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored. When the program instructions are run in a computer, they are used to execute the above-mentioned method for detecting anomalies in environmental monitoring data.

[0015] The third aspect of the present invention provides an environmental monitoring data anomaly detection system, comprising the above-mentioned computer-readable storage medium, the system is any one of a computer, a server, and a single-chip microcomputer, the computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

[0016] Compared with the prior art, the present invention provides an environmental monitoring data anomaly detection method, medium and system. The environmental monitoring data anomaly detection method proposed by the present invention constructs the original matrix of environmental monitoring data and separates it into a regular matrix and an anomaly matrix by matrix decomposition. On this basis, an optimization equation group including environmental parameter weight equation, anomaly threshold equation and comprehensive anomaly degree equation is designed to realize comprehensive analysis and dynamic optimization of multi-dimensional environmental parameter anomaly characteristics. The method of the present invention makes full use of the spatiotemporal characteristic information of environmental parameters and establishes a complete parameter feature extraction and analysis system.

[0017] The present invention solves the problems existing in traditional methods through multiple innovative technical means: by constructing a decomposition framework of regularity matrix and anomaly matrix, the normal change pattern and the abnormal fluctuation component are effectively separated; by designing the environmental parameter weight equation, the multi-dimensional characteristics such as historical mean, fluctuation range, sampling frequency, time continuity and spatial correlation are comprehensively considered to realize the adaptive evaluation of the importance of different environmental parameters; through the dynamic optimization of the anomaly threshold equation, the seasonal change characteristics of environmental parameters, typical fluctuation cycle and spatial gradient information are combined to establish a flexible anomaly judgment standard; through the multi-feature fusion of the comprehensive anomaly degree equation, a comprehensive evaluation of the abnormal state is realized.

[0018] The present invention successfully solves the problem of unifying the dynamic correlation of multi-dimensional parameters and detection accuracy in anomaly detection of environmental monitoring data. The introduction of gain vector and feedback vector provides an adaptive optimization mechanism for anomaly detection, so that the detection system can be continuously adjusted and improved according to the actual monitoring effect. While maintaining high detection accuracy, the method of the present invention effectively reduces the false alarm rate and improves the reliability and practicality of the environmental monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the method of the present invention.

[0020] Figure 2 It is a scatter plot of submatrix feature distribution in Example 2.

[0021] Figure 3 This is a typical abnormal event characteristic analysis diagram in Example 2. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0023] like Figure 1 FIG. 1 is a flowchart of a method for detecting anomalies in environmental monitoring data provided by the first aspect of the present invention. The method comprises the following steps: S01, constructing an original matrix of environmental monitoring data, wherein the original matrix of environmental monitoring data includes monitoring values ​​of multiple environmental parameters at different time points; S02, segmenting the original matrix of environmental monitoring data into time series to obtain a training data sub-matrix; S03, using a matrix decomposition algorithm to decompose the training data submatrix into a regular matrix and an abnormal matrix; S04. Calculate the historical mean and historical standard deviation of each environmental parameter according to the regularity matrix; S05. Calculate the spatial gradient, spatial correlation, and spatial distribution characteristics of each environmental parameter based on the historical mean; S06. Calculate the fluctuation range, sampling frequency, and change rate of each environmental parameter according to the historical standard deviation; S07. Obtaining the time duration, typical fluctuation cycle, and seasonal variation amplitude of each environmental parameter based on the regularity matrix; S08, calculating the degree of abnormality of the data points in the abnormal matrix using an abnormality detection optimization equation group, wherein the abnormality detection optimization equation group includes an environmental parameter weight equation, an abnormality threshold equation, and a comprehensive abnormality degree equation; S09, using the environmental parameter weight equation to calculate an environmental parameter weight vector, wherein the environmental parameter weight equation input includes the historical mean, the fluctuation range, the sampling frequency, the time duration, and the spatial correlation; S10, using the abnormal threshold equation to calculate a dynamic abnormal threshold matrix, wherein the abnormal threshold equation input includes the environmental parameter weight vector, the historical standard deviation, the seasonal variation amplitude, the typical fluctuation cycle, and the spatial gradient; S11, using the comprehensive anomaly degree equation to calculate a comprehensive anomaly degree index, wherein the comprehensive anomaly degree equation input includes the environmental parameter weight vector, the dynamic anomaly threshold matrix, the anomaly matrix, the change rate, and the spatial distribution characteristics; S12, constructing a gain vector based on the comprehensive abnormality index; S13, calculating a feedback vector using the gain vector; S14, optimizing and adjusting the abnormality detection optimization equation group using the feedback vector; S15. Perform anomaly detection on the real-time environmental monitoring data using the optimized anomaly detection optimization equation group.

[0024] Among them: the original matrix of environmental monitoring data is an m×n dimensional matrix, m is the number of environmental parameters, and n is the number of time sampling points; the regularity matrix represents the normal change pattern of environmental parameters; the anomaly matrix represents the abnormal fluctuation components of environmental parameters; the gain vector represents the degree of deviation of abnormal data; the feedback vector represents the correction coefficient of anomaly judgment; the environmental parameter weight vector represents the importance of different environmental parameters; the dynamic anomaly threshold matrix represents the anomaly judgment standard of multiple environmental parameter combinations; the comprehensive anomaly index represents the overall evaluation result of environmental anomalies.

[0025] The specific implementation of the above steps is described in detail below. The specific implementation of step S01 is to obtain environmental monitoring data through a data acquisition system, using a sensor array to collect data including temperature, humidity, PM2.5, Environmental parameters such as concentration and noise are recorded in real time at the preset sampling frequency, and the collected data is stored in a two-dimensional matrix. The rows of the matrix represent different environmental parameters, and the columns represent different time points. The data collection time span is recommended to be no less than 30 days, and the sampling interval is recommended to be no more than 10 minutes. For different environmental parameters, corresponding data preprocessing methods are used, including outlier removal, data normalization, and missing value filling. Outlier removal uses the 3 times standard deviation method, data normalization uses the minimum and maximum normalization method, and missing value filling uses the moving average interpolation method. Finally, a complete original matrix of environmental monitoring data is constructed. The purpose of this step is to establish a standardized data foundation to ensure the accuracy of subsequent analysis.

[0026] The specific implementation method of step S02 is to use the sliding window method to segment the original matrix of environmental monitoring data into time series, set the sliding window size to 24 hours, and the window sliding step length to 1 hour. For the data in each sliding window, a submatrix is ​​extracted, and the submatrix keeps the number of rows of the original matrix unchanged, and the number of columns corresponds to the window size. The representativeness of each submatrix is ​​determined by calculating its statistical characteristics, which include mean, variance, skewness and kurtosis. The submatrix with the closest statistical characteristics to the overall distribution is selected as training data. It is recommended to select 3 to 5 representative submatrices to form a training data set, and the cross-validation method is used to evaluate the representativeness and integrity of the submatrix to ensure that the selected training data can fully reflect the changing characteristics of environmental parameters. The purpose of this step is to obtain representative training samples and provide a reliable data basis for subsequent matrix decomposition.

[0027] The specific implementation method of step S03 is to decompose the training data submatrix using a robust principal component analysis algorithm. First, the training data submatrix is ​​centralized, the covariance matrix is ​​calculated, and the singular value decomposition method is used to obtain the eigenvalues ​​and eigenvectors. The number of principal components is determined according to the size of the eigenvalues. Generally, the principal components with a cumulative contribution rate of 85% are selected, and the corresponding eigenvectors are formed into a conversion matrix. The regular matrix is ​​obtained by principal component reconstruction, and the difference between the original matrix and the regular matrix is ​​calculated to obtain the abnormal matrix. The iterative optimization method is used to improve the decomposition accuracy, and the convergence threshold is set to 0.001. The maximum number of iterations is 100 times, and the decomposition effect is evaluated by calculating the reconstruction error. The purpose of this step is to decompose the environmental data into regular components and abnormal components, providing a basis for subsequent feature extraction.

[0028] The specific implementation method of step S04 is to calculate the statistical characteristics of environmental parameters based on the regular matrix. For each environmental parameter in the regular matrix, the sliding window method is used to calculate the historical mean, and the window size is set to 24 hours. The historical mean sequence of the parameter is obtained by calculating the arithmetic mean of the data in the window, and the exponentially weighted moving average method is used to smooth the historical mean sequence. The smoothing coefficient is set to 0.2, and the standard deviation is calculated based on the smoothed historical mean sequence. The historical standard deviation of each environmental parameter is calculated by the method of the sum of squared deviations divided by the degree of freedom. For environmental parameters of different scales, relative standard deviations are used for standardization. The purpose of this step is to obtain the basic statistical characteristics of environmental parameters and provide a statistical benchmark for anomaly detection.

[0029] The specific implementation method of step S05 is to use the law matrix to analyze the spatial characteristics of environmental parameters. First, the spatial gradient of environmental parameters is calculated, and the parameter change rate between adjacent monitoring points is calculated by the central difference method. The spatial gradient is expressed as the parameter change within a unit distance. Then, the spatial correlation between environmental parameters is calculated, and the correlation matrix is ​​constructed by the Pearson correlation coefficient method. The correlation coefficient threshold is set to 0.6. According to the correlation matrix, the parameter combination with significant correlation is identified. Finally, the spatial distribution characteristics of environmental parameters are analyzed, and the spatial distribution model of parameters is constructed by the Kriging interpolation method. The spatial autocorrelation index is calculated, and the spatial variation law is analyzed by the variation function to establish the spatial distribution feature vector. The purpose of this step is to obtain the spatial characteristics of environmental parameters and provide a judgment basis for the spatial dimension for anomaly detection.

[0030] The specific implementation method of step S06 is to analyze the dynamic characteristics of environmental parameters based on historical standard deviations. First, the fluctuation range is calculated, and the upper and lower limits of the fluctuation of the parameters are determined by the percentile method. The 15th and 85th percentiles are taken as the normal fluctuation range. The proportion of the fluctuation range to the historical mean is calculated as the fluctuation intensity index. Then, the sampling frequency characteristics are analyzed based on the sampled data, and the signal spectrum is analyzed by the Fourier transform method to determine the main frequency component and harmonic component of the signal. The entropy value of the frequency distribution is calculated to characterize the complexity of the sampling frequency. Finally, the rate of change of the environmental parameters is calculated, and the first-order difference method is used to calculate the change amount of adjacent time points. The least squares method is used to fit the change trend to obtain the change rate vector of each parameter. The purpose of this step is to obtain the dynamic change characteristics of environmental parameters and provide a judgment basis for the time dimension for anomaly detection.

[0031] The specific implementation method of step S07 is to analyze the time characteristics of environmental parameters based on the regularity matrix, first calculate the time persistence, use the autocorrelation function method to analyze the time correlation of parameter values, determine the significance level of the autocorrelation coefficient to be 0.05, calculate the correlation length as the time persistence index, then use the wavelet transform method to analyze the typical fluctuation cycle, select the Mexican hat wavelet as the basis function, determine the main periodic component by the modulus maximum of the wavelet coefficient, and set the periodic significance threshold to 0.8. Finally, analyze the seasonal change characteristics, use the seasonal decomposition algorithm to extract the seasonal component, calculate the amplitude and phase of the seasonal change, and construct the seasonal feature vector. The purpose of this step is to obtain the periodic characteristics of environmental parameters and provide a judgment basis for the time series pattern for anomaly detection.

[0032] The specific implementation method of step S08 is to construct an anomaly detection optimization equation group, first construct an environmental parameter weight equation, use the hierarchical analysis method to determine the importance of parameters, establish a judgment matrix through expert scoring, calculate the characteristic vector as the initial weight, and then construct an anomaly threshold equation, use the adaptive threshold method, dynamically adjust the threshold based on the statistical characteristics of historical data, consider seasonal and cyclical effects, and finally construct a comprehensive anomaly degree equation, use the weighted summation method to combine multidimensional features into a single indicator, and ensure the comparability of features of different dimensions through normalization. The purpose of this step is to establish a scientific anomaly detection mathematical model and realize the effective fusion of multidimensional features.

[0033] The specific implementation method of step S09 is to calculate the environmental parameter weight vector. First, the importance score of the parameter is calculated based on the historical mean, and the information contribution of the parameter is evaluated by the information entropy weight method. The precision parameter in the entropy weight calculation is set to 0.001. Then, the stability score of the parameter is calculated according to the fluctuation range and sampling frequency, and the coefficient of variation method is used to evaluate the fluctuation degree of the parameter. Then, the persistence score of the parameter is calculated based on the time continuity, and the exponential decay function is used to quantify the continuous impact. Finally, the correlation score of the parameter is calculated according to the spatial correlation, and the network centrality analysis method is used to evaluate the influence of the parameter. The scores are combined through the multi-criteria decision-making method to obtain the final environmental parameter weight vector. The purpose of this step is to determine the importance of each environmental parameter in anomaly detection and realize differentiated parameter processing.

[0034] The specific implementation method of step S10 is to calculate the dynamic anomaly threshold matrix. First, the environmental parameter weight vector is applied to the historical standard deviation, and the benchmark threshold is calculated by the weighted standard deviation method. The weight normalization coefficient is set to 1. Then, the benchmark threshold is adjusted according to the seasonal variation amplitude, and the benchmark threshold is modulated by a periodic function. The modulation depth is proportional to the seasonal intensity. Then, the threshold variation law is refined according to the typical fluctuation cycle, and the multi-scale threshold method is used to process fluctuations of different time scales. Finally, the threshold distribution is corrected based on the spatial gradient, and the distance weighted method is used to calculate the spatial correlation correction coefficient to obtain the dynamic anomaly threshold matrix considering the spatiotemporal characteristics. The purpose of this step is to establish an adaptive anomaly judgment standard and improve the accuracy of anomaly detection.

[0035] The specific implementation method of step S11 is to calculate the comprehensive abnormality index. First, the environmental parameter weight vector is multiplied by the dynamic abnormality threshold matrix, and the weighted threshold matrix is ​​obtained by matrix multiplication. The normalization coefficient is set to 1. Then the data in the abnormal matrix is ​​compared with the weighted threshold matrix, and the abnormality of each parameter is calculated by the over-limit detection method. The abnormality calculation adopts the relative deviation method, and then the abnormality is corrected according to the rate of change. The abnormality size is adjusted by the rate weight method. The adjustment coefficient is exponentially related to the rate size. Finally, the spatial consistency coefficient is calculated based on the spatial distribution characteristics, and the spatial continuity of the abnormality is evaluated by the local similarity method to obtain the abnormality index that comprehensively considers the multi-dimensional characteristics. The purpose of this step is to achieve quantitative evaluation of environmental anomalies and provide a reliable basis for abnormality determination.

[0036] The specific implementation of step S12 is to construct a gain vector. First, the degree of anomaly is calculated according to the comprehensive anomaly index, and the degree of anomaly is mapped to the interval of 0 to 1 using the fuzzy membership method. The membership function uses the S-type function, and then the cumulative effect of the anomaly is calculated. The exponential integral method is used to evaluate the continuous impact of the anomaly. The time attenuation coefficient is set to 0.1. Then, the propagation characteristics of the anomaly are analyzed, and the network diffusion model is used to evaluate the spatial spread of the anomaly. The diffusion coefficient is proportional to the spatial correlation. Finally, the intensity, persistence and diffusion of the anomaly are comprehensively considered to construct a gain vector reflecting the degree of anomaly deviation. The purpose of this step is to quantify the impact of the anomaly and provide a basis for the optimization of anomaly detection.

[0037] The specific implementation method of step S13 is to calculate the feedback vector, first evaluate the detection effect based on the gain vector, and use the accuracy evaluation method to calculate the detection performance index, the performance index includes precision and recall, then analyze the false detection situation, use the confusion matrix method to count the false alarms and missed alarms, calculate the false positive rate and missed positive rate, then calculate the correction coefficient according to the detection effect, use the adaptive learning method to adjust the detection parameters, and set the learning rate to 0.01, and finally construct a feedback vector containing the detection performance and correction information. The purpose of this step is to evaluate the effect of anomaly detection and provide feedback information for the optimization of the equation group.

[0038] The specific implementation method of step S14 is to optimize the anomaly detection equation group. First, the environmental parameter weights are updated according to the feedback vector, and the weight coefficients are optimized by the gradient descent method. The step size parameter is set to 0.1. Then the anomaly threshold calculation method is adjusted, and the threshold parameters are updated by the Bayesian optimization algorithm. The optimization goal is to minimize the misjudgment rate. Then, the comprehensive anomaly calculation method is optimized, and the feature combination weights are adjusted by the genetic algorithm. The population size is set to 50, the number of iterations is 100, and finally the parameter configuration of the detection equation group is updated according to the optimization results. The purpose of this step is to improve the accuracy and adaptability of anomaly detection and achieve continuous optimization of the detection method.

[0039] The specific implementation method of step S15 is to realize real-time anomaly detection. First, the real-time environmental monitoring data is preprocessed, and the data is normalized by online standardization method. The processing delay is less than 1 second. Then, the optimized detection equation group is used for calculation, and the parallel computing method is used to improve the processing efficiency. It supports the processing capacity of thousands of data points per second. Then, the anomaly detection results are generated, and the multi-level alarm mechanism is used for risk classification. The alarm threshold is divided into three levels, namely 0.7, 0.8 and 0.9. Finally, the detection results are sent to the monitoring system, and the relevant personnel are notified by real-time push, and the abnormal data is saved in the database. The purpose of this step is to realize real-time anomaly detection of environmental monitoring data to ensure timely detection and handling of abnormal situations.

[0040] The original matrix of environmental monitoring data is specifically expressed as follows: ; In the formula, Indicates The environmental parameters are Monitoring value at each time point; is the number of environmental parameters; is the number of time sampling points.

[0041] The data preprocessing process includes outlier removal, normalization and missing value filling, which are specifically expressed as follows: The outliers were eliminated by using the 3 times standard deviation method: ; ; ; In the formula, For the The mean value of environmental parameters; For the The standard deviation of each environmental parameter; The outlier mark is 1 for outliers and 0 for normal values. The outliers were eliminated using the 3 times standard deviation method, which is based on the characteristics of the normal distribution. Values ​​outside the 99.7% confidence interval are considered outliers; Normalization uses the minimum and maximum normalization method: ; In the formula, is the normalized value; For the The minimum value of an environmental parameter; For the The maximum value of an environmental parameter.

[0042] Missing values ​​are filled using moving average interpolation: ; In the formula, is the value after filling; For the moving window size, the recommended value is 6; is the time index within the window, It refers to The environmental parameters are The monitoring value at each time point. Moving average interpolation takes into account the continuity of the time series, and the selection of the window size k is based on the sampling frequency of the data; The statistical features in the time series segmentation are calculated as follows: ; ; ; ; In the formula, is the element in the submatrix; is the number of columns of the submatrix; is the submatrix mean; is the submatrix variance; is the submatrix skewness; is the kurtosis of the submatrix. The statistical characteristic calculation of the submatrix includes the first to fourth order moments, which fully reflects the distribution characteristics of the data.

[0043] The specific calculation of the matrix decomposition process is as follows: First, the training data sub-matrix is ​​centered: ; In the formula, is the centered matrix; is the original matrix List.

[0044] Compute the covariance matrix: ; In the formula, is the covariance matrix; for The transposed matrix of .

[0045] Singular value decomposition process: ; In the formula, is a left singular matrix; is a diagonal matrix of singular values; is a right singular matrix. Singular value decomposition realizes the orthogonal decomposition of data, which is convenient for extracting the main change mode.

[0046] The principal component selection is based on the cumulative contribution rate: ; In the formula, For the front The cumulative contribution rate of the principal components; For the The calculation of the cumulative contribution rate takes into account the integrity of information retention, and the threshold is set to 85% to ensure the reconstruction accuracy; Regular matrix reconstruction: ; In the formula, is the reconstructed regular matrix; For the left singular vectors; For the singular values; For the right singular vectors.

[0047] Abnormal matrix calculation: ; In the formula, is the abnormal matrix; It is a noise compensation item, and the recommended value range is 0.001~0.01.

[0048] The historical mean is calculated using the sliding window method: ; In the formula, For the The parameters are The historical mean of the time period; is the normalization coefficient; is the attenuation coefficient, ranging from 0.8 to 0.95; is the window size. The historical mean calculation introduces a time decay term, which reflects the importance of recent data.

[0049] Historical standard deviation calculation: ; In the formula, For the The parameters are The historical standard deviation of the time.

[0050] The spatial gradient is calculated using the central difference method: ; In the formula, For the The parameters in The gradient of a spatial position; is the spatial sampling interval; is the gradient correction term, and its value range is 0.01~0.05.

[0051] Spatial correlation calculation: ; In the formula, For the Parameters and The correlation coefficient of the parameters. The Pearson correlation coefficient is used for correlation calculation, which effectively measures the degree of linear correlation between parameters.

[0052] The calculation process of the fluctuation characteristics is as follows: Fluctuation range calculation: ; In the formula, For the The parameters are The intensity of fluctuations at each moment; For the The 85th percentile of the parameters; For the The 15th percentile of the parameter; It is a fluctuation correction term, and its value range is 0.05~0.1. The fluctuation intensity is calculated using the percentile method to avoid the influence of outliers.

[0053] Sampling frequency characteristic calculation: ; ; In the formula, For the Spectral density function of parameters; is the frequency distribution entropy; For the frequency components; is the total number of frequency components.

[0054] Rate of change calculation: ; In the formula, For the The parameters are The rate of change of time; is the sampling time interval; is the time attenuation coefficient, and its value range is 0.001~0.01. Indicates Environmental parameters at time Monitoring value of Indicates Environmental parameters at time The monitoring value, i.e. time The value at the previous time point.

[0055] Time duration calculation: ; ; In the formula, For the The autocorrelation function of the parameters; is the correlation length; For time delay; is the maximum delay time.

[0056] Fluctuation cycle analysis: ; ; In the formula, is the wavelet transform coefficient; is the Mexican hat wavelet function; is the scale parameter; is the translation parameter; is the wavelet power spectrum. Wavelet transform has multi-resolution analysis capability and is suitable for extracting the characteristics of non-stationary signals.

[0057] Seasonal variation characteristics: ; In the formula, For the seasonal components of the parameters; is the amplitude; For the cycle; is the phase; is a random disturbance term, and its value range is 0~0.1.

[0058] The specific calculation of the anomaly detection optimization equation group is as follows: Environmental parameter weight equation: ; ; ; ; ; In the formula, For the The weight of the parameters; is information entropy; Score for stability; Score for persistence; Score for relevance; is the combination coefficient, satisfying ; is the attenuation coefficient, ranging from 0.1 to 0.5; is a weight correction term, and its value range is 0.01~0.05. The weight equation comprehensively considers the four aspects of information volume, stability, persistence and relevance, and realizes the multi-dimensional evaluation of parameter importance.

[0059] Anomaly threshold equation: ; In the formula, For the Parameters at position ,time The abnormal threshold value of is the adjustment coefficient, and its value range is 0~1; are the maximum values ​​of amplitude, power spectrum and gradient, respectively; is the threshold correction term, and its value range is 0.1~0.3. The anomaly threshold equation introduces the influence of seasonality, periodicity and spatial distribution, and improves the adaptability of the threshold.

[0060] Comprehensive anomaly degree equation: ; In the formula, is the comprehensive abnormality; is the corresponding value in the regularity matrix; is the influence coefficient, and its value range is 0~1; is the maximum rate of change; is the spatial second-order derivative; is the maximum value of the second-order derivative; is the anomaly correction term, and its value range is 0.05~0.15. The comprehensive anomaly equation combines the amplitude deviation, change rate and spatial distribution characteristics, and realizes the fusion of multi-dimensional anomaly features.

[0061] Gain vector calculation: ; In the formula, For the The gain value of the parameter; is the baseline abnormality, with a value of 0.5; is the sensitivity coefficient, ranging from 5 to 10; is the cumulative coefficient, ranging from 0.001 to 0.01; It is the gain correction term, and its value range is 0.01~0.05. Indicates Environmental parameters at time The comprehensive abnormality index. Indicates Environmental parameters at time The comprehensive abnormality index.

[0062] Feedback vector calculation: ; In the formula, For the The feedback value of the parameter; are the number of true positives, false positives, and false negatives, respectively; is the total number of samples; is the penalty coefficient, ranging from 1 to 5; It is a feedback correction term with a value range of 0.01~0.05. The feedback vector is constructed based on precision and recall, and a penalty term is introduced to constrain misjudgment, thereby improving the reliability of detection.

[0063] Detection equation group optimization process: ; ; In the formula, Indicates The weight of an environmental parameter reflects the importance of the environmental parameter. is the updated weight; is the learning rate, the value is 0.1; is the loss function, Indicates Environmental parameters at location The comprehensive abnormality degree is the result of the comprehensive abnormality degree equation; is the expected abnormality; To optimize the correction term, the value range is 0.001~0.01.

[0064] Real-time anomaly detection process: ; ; In the formula, Standardize data for real-time; is the warning level; It is a standardization correction item, and its value range is 0.01~0.05.

[0065] A second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored. When the program instructions are run in a computer, they are used to execute the above-mentioned method for detecting anomalies in environmental monitoring data.

[0066] The third aspect of the present invention provides an environmental monitoring data anomaly detection system, comprising the above-mentioned computer-readable storage medium, the system is any one of a computer, a server, and a single-chip microcomputer, the computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

[0067] Specifically, the principle of the present invention is: the technical principle of the present invention is based on the multi-dimensional feature analysis and dynamic optimization theory of environmental monitoring data. First, the structured expression of data is realized through matrix decomposition, and the environmental monitoring data is decomposed into a regular matrix reflecting the normal change law and an abnormal matrix reflecting abnormal fluctuations. This decomposition method is consistent with the physical nature of environmental parameter changes, because environmental parameters usually contain two components: stable periodic changes and random abnormal fluctuations.

[0068] The anomaly detection optimization equations designed by the present invention reflect the adaptability and robustness of the system. The environmental parameter weight equation establishes a quantitative evaluation mechanism for the importance of parameters by analyzing the historical performance, dynamic characteristics and spatial characteristics of the parameters. This evaluation fully considers the multidimensional properties of environmental parameters, so that the system can perform differentiated processing according to the actual degree of influence of the parameters. The anomaly threshold equation integrates information such as seasonal changes, periodic characteristics and spatial distribution of environmental parameters into the threshold calculation, realizing the dynamic adjustment of the judgment criteria. This adjustment mechanism enables the system to adapt to the time-varying characteristics of environmental parameters. The comprehensive anomaly degree equation realizes a comprehensive evaluation of the abnormal state through multi-source information fusion. The equation combines the anomaly matrix information with the changing characteristics of environmental parameters, thereby improving the reliability of anomaly detection.

[0069] A specific embodiment 1 of the present invention is provided below, and the specific implementation method of each step in this embodiment 1 is described in detail as follows.

[0070] The specific implementation of step S01 is to obtain environmental monitoring data through a data acquisition system, using a sensor array to collect data including temperature, humidity, PM2.5, The real-time values ​​of environmental parameters such as concentration and noise are recorded according to the preset sampling frequency, and the collected data is stored in a two-dimensional matrix. The original matrix of environmental monitoring data is specifically expressed as: , where Indicates The environmental parameters are The monitoring value at each time point, is the number of environmental parameters, is the number of time sampling points. The data preprocessing process includes outlier removal: , , ; Normalization processing: ; Missing value filling: The sampling interval is recommended to be no more than 10 minutes, the data collection time span is no less than 30 days, and the window size k is recommended to be 6. This step ensures the data quality and reliability of subsequent analysis through a standardized data processing process.

[0071] The specific implementation of step S02 is to use the sliding window method to segment the original matrix of environmental monitoring data into time series, set the sliding window size to 24 hours, and the window sliding step to 1 hour. For each data in the sliding window, extract the sub-matrix and calculate the statistical features: , , , . Select 3 to 5 sub-matrices whose statistical characteristics are closest to the overall distribution to form a training data set to ensure that the selected training data can fully reflect the changing characteristics of environmental parameters.

[0072] The specific implementation of step S03 is to use a robust principal component analysis algorithm to decompose the training data submatrix. First, perform centralization processing: , calculate the covariance matrix: , perform singular value decomposition: The principal component selection is based on the cumulative contribution rate: , regular matrix reconstruction: , the abnormal matrix calculation: .in is a noise compensation term. The recommended value range is 0.001 to 0.01, the convergence threshold is 0.001, and the maximum number of iterations is 100.

[0073] The specific implementation method of step S04 is to calculate the statistical characteristics of the environmental parameters based on the regularity matrix. The historical mean value calculation adopts the sliding window method: ,in is the normalization coefficient, is the attenuation coefficient, ranging from 0.8 to 0.95, is the window size. Historical standard deviation calculation: Through these statistical characteristics, the changing patterns and fluctuation characteristics of environmental parameters can be effectively reflected.

[0074] The specific implementation of step S05 is to use the regularity matrix to analyze the spatial characteristics of environmental parameters. The spatial gradient calculation adopts the central difference method: ,in is the gradient correction term, ranging from 0.01 to 0.05. Spatial correlation calculation: , the threshold of correlation coefficient is set to 0.6. The spatial distribution model of parameters is constructed by Kriging interpolation method, the spatial autocorrelation index is calculated, the spatial variation law is analyzed by the variation function, and the spatial distribution feature vector is established.

[0075] The specific implementation of step S06 is to analyze the dynamic characteristics of environmental parameters based on historical standard deviation. The fluctuation range is calculated using the percentile method: ,in For the The 85th percentile of the parameters, is the 15th percentile, is the fluctuation correction term, and its value range is 0.05 to 0.1. The sampling frequency characteristic is calculated using the Fourier transform method: , . Change rate calculation: ,in is the time attenuation coefficient, ranging from 0.001 to 0.01. These calculations can fully reflect the dynamic change characteristics of environmental parameters.

[0076] The specific implementation of step S07 is to analyze the time characteristics of environmental parameters based on the regularity matrix. The time continuity calculation adopts the autocorrelation function method: , ,in is the time delay, and the significance level of the autocorrelation coefficient is 0.05. The wavelet transform method is used for the fluctuation period analysis: , , the Mexican hat wavelet is selected as the basis function, and the cycle significance threshold is set to 0.8. Calculation of seasonal change characteristics: ,in is a random disturbance term, ranging from 0 to 0.1.

[0077] The specific implementation of step S08 is to construct an abnormality detection optimization equation group. Environmental parameter weight equation: ,in is the information entropy, For the stability score, For persistence score, is the correlation score, the combination coefficient satisfy , is the weight correction term, and its value range is 0.01 to 0.05. Abnormal threshold equation: ,in is the adjustment coefficient, ranging from 0 to 1. is the threshold correction term, and its value range is 0.1 to 0.3. Comprehensive abnormality equation: ,in is the influence coefficient, ranging from 0 to 1. It is an abnormality correction item, and its value range is 0.05 to 0.15.

[0078] The specific implementation of step S09 is to calculate the environmental parameter weight vector. Information entropy calculation: , stability score calculation: , persistence score calculation: ,in is the attenuation coefficient, ranging from 0.1 to 0.5, and the correlation score is calculated as: The final environmental parameter weight vector is obtained by combining the scores of various items through the multi-criteria decision-making method.

[0079] The specific implementation method of step S10 is to calculate the dynamic anomaly threshold matrix. First, the environmental parameter weight vector is applied to the historical standard deviation, and the weighted standard deviation method is used to calculate the benchmark threshold. The benchmark threshold is adjusted according to the seasonal variation amplitude, and the benchmark threshold is modulated by a periodic function, and the modulation depth is proportional to the seasonal intensity. The threshold variation law is refined according to the typical fluctuation cycle, and the multi-scale threshold method is used to process fluctuations of different time scales. Based on the spatial gradient correction threshold distribution, the distance weighted method is used to calculate the spatial correlation correction coefficient to obtain the dynamic anomaly threshold matrix considering the spatiotemporal characteristics. The threshold calculation process fully considers the spatiotemporal characteristics and dynamic variation laws of the environmental parameters.

[0080] The specific implementation of step S11 is to calculate the comprehensive abnormality index. First, the environmental parameter weight vector is multiplied by the dynamic abnormality threshold matrix, and the weighted threshold matrix is ​​calculated by matrix multiplication, and the normalization coefficient is set to 1. Then the data in the abnormality matrix is ​​compared with the weighted threshold matrix, and the abnormality of each parameter is calculated by the over-limit detection method. The abnormality calculation formula is: ,in is the influence coefficient, ranging from 0 to 1. is the anomaly correction term, with a value range of 0.05 to 0.15. Then, the anomaly is corrected according to the rate of change, and the rate weight method is used to adjust the anomaly size. The adjustment coefficient is exponentially related to the rate size. Finally, the spatial consistency coefficient is calculated based on the spatial distribution characteristics, and the local similarity method is used to evaluate the spatial continuity of the anomaly, and the anomaly index that comprehensively considers multi-dimensional features is obtained.

[0081] The specific implementation of step S12 is to construct a gain vector. First, the degree of abnormality is calculated according to the comprehensive abnormality index, and the abnormality is mapped to the interval of 0 to 1 using the fuzzy membership method. The gain vector calculation formula is: ,in is the baseline abnormality, with a value of 0.5. is the sensitivity coefficient, ranging from 5 to 10. is the cumulative coefficient, ranging from 0.001 to 0.01, is the gain correction term, ranging from 0.01 to 0.05. Then the cumulative effect of the anomaly is calculated, and the exponential integration method is used to evaluate the continuous impact of the anomaly, with the time attenuation coefficient set to 0.1. Then the propagation characteristics of the anomaly are analyzed, and the network diffusion model is used to evaluate the spatial spread of the anomaly. The diffusion coefficient is proportional to the spatial correlation. Finally, the intensity, persistence and diffusion of the anomaly are comprehensively considered to construct a gain vector reflecting the degree of anomaly deviation.

[0082] The specific implementation of step S13 is to calculate the feedback vector. First, the detection effect is evaluated based on the gain vector, and the detection performance index is calculated using the accuracy evaluation method. The feedback vector calculation formula is: ,in are the number of true positives, false positives, and false negatives, respectively, is the total number of samples, is the penalty coefficient, ranging from 1 to 5. is the feedback correction term, with a value range of 0.01 to 0.05. Then analyze the false detection situation, use the confusion matrix method to count the false positives and false negatives, and calculate the false positive rate and false negative rate. Then calculate the correction coefficient according to the detection effect, use the adaptive learning method to adjust the detection parameters, and set the learning rate to 0.01. Finally, construct a feedback vector containing the detection performance and correction information.

[0083] The specific implementation of step S14 is to optimize the anomaly detection equation group. First, the environmental parameter weights are updated according to the feedback vector, and the weight coefficients are optimized using the gradient descent method. The optimization formula is: , ,in is the learning rate, the value is 0.1, To optimize the correction term, the value range is 0.001 to 0.01. Then adjust the abnormal threshold calculation method, use the Bayesian optimization algorithm to update the threshold parameters, and the optimization goal is to minimize the misjudgment rate. Then optimize the comprehensive abnormality calculation method, use the genetic algorithm to adjust the feature combination weight, set the population size to 50, and the number of iterations to 100. Finally, update the parameter configuration of the detection equation group according to the optimization results.

[0084] The specific implementation method of step S15 is to realize real-time anomaly detection. First, the real-time environmental monitoring data is preprocessed, and the data is normalized using an online standardization method. The preprocessing formula is: ,in is a standardized correction term with a value range of 0.01 to 0.05. Then the optimized detection equation group is used for calculation, and the parallel computing method is used to improve the processing efficiency. Then the abnormal detection results are generated, and the multi-level alarm mechanism is used for risk classification. The alarm level determination formula is: Finally, the test results are sent to the monitoring system, and relevant personnel are notified in real time, and the abnormal data is saved in the database.

[0085] In order to better understand and implement the present invention, the following provides an example 2 of a specific application scenario of the present invention: A research team conducted an application study on the environmental monitoring anomaly detection system based on the present invention in an industrial park. The industrial park covers an area of ​​about 10 square kilometers and has 50 environmental monitoring points, each of which is equipped with temperature, humidity, PM2.5, PM10, CO, , 、NOx、 , wind speed, wind direction, noise and other 12 environmental parameters sensors with a sampling frequency of 5 minutes.

[0086] The collection and preprocessing of environmental monitoring data are shown in Table 1: Table 1 Statistical table of environmental monitoring data preprocessing results

[0087] After preprocessing, the original data matrix The dimension is 12×432000. Using 24 hours as the window size and 1 hour as the sliding step, 4 representative sub-matrices are selected as training data. The statistical characteristics of the sub-matrices are shown in Table 2: Table 2 Statistical characteristics of training data submatrices

[0088] Figure 2 This is a multi-dimensional feature scatter plot that shows the statistical feature distribution of four training data sub-matrices. The X-axis represents the mean, the Y-axis represents the standard deviation, the size of the dot represents the skewness, and the color depth represents the kurtosis. Each dot represents a sub-matrix. This visualization method can intuitively show the feature differences of different sub-matrices. Mathematical symbols are used in the figure. , and To represent the corresponding statistics, professionalism and accuracy are ensured. Robust principal component analysis is performed on the training data, and 6 principal components with a cumulative contribution rate of 85% are selected to reconstruct the regular matrix and the anomaly matrix The historical mean and historical standard deviation of environmental parameters calculated based on the law matrix are shown in Table 3: Table 3 Historical mean and standard deviation of environmental parameters

[0089] Figure 3This is a combined chart that mainly shows the characteristic indicators of 5 typical abnormal events (E001-E005). The bar chart corresponding to the Y axis on the left shows the comprehensive abnormality ( ), gain value( ) and feedback value ( ) three characteristic indicators, the Y axis and the red line on the right show the change of the alarm level. In this way, the multi-dimensional characteristics of each abnormal event and its corresponding alarm level can be clearly displayed. The results of spatial correlation analysis show that the correlation coefficient between PM2.5 and PM10 is 0.86, The correlation coefficient with NOx is 0.78, and the two sets of parameters have significant spatial correlation. Based on the spatiotemporal characteristics of environmental parameters, the calculated parameter weight vector is shown in Table 4:

[0090] Table 4 Environmental parameter weight vector

[0091] The typical abnormal events and their characteristics detected by the system during operation are shown in Table 5: Table 5 Characteristics of typical abnormal events

[0092] Traditional environmental monitoring anomaly detection mainly adopts fixed threshold method and single parameter statistical analysis method, which has the following problems: 1. It cannot effectively handle the correlation between multiple parameters; 2. The threshold setting lacks dynamic adaptability; 3. The false alarm rate and missed alarm rate are high; 4. It cannot identify complex anomalies. The present invention constructs an anomaly detection method for environmental monitoring data, uses robust principal component analysis to achieve effective data decomposition, realizes accurate identification of anomalies through multi-dimensional feature fusion, and improves the accuracy of detection through adaptive optimization. In this embodiment, the average detection accuracy of the system reaches 92.3%, the false alarm rate is reduced to 4.5%, and the missed alarm rate is reduced to 3.2%. Compared with the traditional method, the detection accuracy is improved by 15.6%, the false alarm rate is reduced by 8.7%, and the missed alarm rate is reduced by 7.5%. The method of the present invention has significant advantages in the early warning and precise positioning of environmental abnormal events, and provides effective technical support for environmental monitoring and management.

[0093] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 6 and 7 below.

[0094] Table 6 Variable explanation table (Part I)

[0095] Table 7 Variable explanation table (Part II)

[0096] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting anomalies in environmental monitoring data, characterized in that: The method comprises the following steps: constructing an original matrix of environmental monitoring data, wherein the original matrix of environmental monitoring data includes monitoring values ​​of multiple environmental parameters at different time points; performing time series segmentation on the original matrix of environmental monitoring data to obtain a training data submatrix; and using a matrix decomposition algorithm to decompose the training data submatrix into a regular matrix and an abnormal matrix; Calculate the historical mean and historical standard deviation of each environmental parameter based on the regularity matrix; calculate the spatial gradient, spatial correlation, and spatial distribution characteristics of each environmental parameter based on the historical mean; calculate the fluctuation range, sampling frequency, and change rate of each environmental parameter based on the historical standard deviation; Based on the regularity matrix, the time persistence, typical fluctuation cycle, and seasonal variation range of each environmental parameter are obtained; the degree of abnormality of the data points in the abnormality matrix is ​​calculated using the abnormality detection optimization equation group, which includes the environmental parameter weight equation, the abnormality threshold equation, and the comprehensive abnormality degree equation; Construct a gain vector based on the comprehensive abnormality index; calculate the feedback vector using the gain vector; The feedback vector is used to optimize and adjust the anomaly detection optimization equation group; the optimized anomaly detection optimization equation group is used to perform anomaly detection on real-time environmental monitoring data.

2. The method for detecting anomalies in environmental monitoring data according to claim 1, characterized in that: The original matrix of environmental monitoring data is an m×n dimensional matrix, where m is the number of environmental parameters and n is the number of time sampling points; the regularity matrix represents the normal change pattern of environmental parameters; the abnormality matrix represents the abnormal fluctuation components of environmental parameters; the gain vector represents the degree of deviation of abnormal data; and the feedback vector represents the correction coefficient of abnormal judgment.

3. The method for detecting anomalies in environmental monitoring data according to claim 2, characterized in that: In the step of constructing the original matrix of environmental monitoring data, the environmental monitoring data is obtained through the data acquisition system, and the sensor array is used to collect data including temperature, humidity, PM2.5, The environmental parameters of concentration and noise are recorded in real time according to the preset sampling frequency, and the collected data is stored in a two-dimensional matrix form.

4. The method for detecting anomalies in environmental monitoring data according to claim 3, characterized in that: In the step of segmenting the original matrix of environmental monitoring data into time series, the sliding window method is used for segmentation. The sliding window size is set to 24 hours and the window sliding step is 1 hour. The representativeness of each sub-matrix is ​​determined by calculating its mean, variance, skewness and kurtosis.

5. The method for detecting anomalies in environmental monitoring data according to claim 4, characterized in that: In the step of decomposing the training data submatrix using the matrix decomposition algorithm, the robust principal component analysis algorithm is used for decomposition, the eigenvalues ​​and eigenvectors are obtained through the singular value decomposition method, and the principal components with a cumulative contribution rate of 85% are selected.

6. The method for detecting anomalies in environmental monitoring data according to claim 5, characterized in that: In the step of calculating the spatial characteristics of environmental parameters based on historical means, the central difference method is used to calculate the spatial gradient, the Pearson correlation coefficient method is used to construct the correlation matrix, and the Kriging interpolation method is used to construct the parameter spatial distribution model.

7. The method for detecting anomalies in environmental monitoring data according to claim 6, characterized in that: In the step of analyzing the dynamic characteristics of environmental parameters based on historical standard deviations, the percentile method is used to determine the upper and lower limits of parameter fluctuations, the Fourier transform method is used to analyze the signal spectrum, and the first-order difference method is used to calculate the rate of change.

8. The method for detecting anomalies in environmental monitoring data according to claim 7, characterized in that: In the step of analyzing the temporal characteristics of environmental parameters based on the regularity matrix, the autocorrelation function method is used to analyze the temporal correlation, the Mexican hat wavelet is used to analyze the typical fluctuation cycle, and the seasonal decomposition algorithm is used to extract the seasonal component.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, they are used to execute the method for detecting anomalies in environmental monitoring data according to any one of claims 1 to 8.

10. An environmental monitoring data anomaly detection system, characterized in that: The system comprises the computer-readable storage medium as claimed in claim 9, wherein the system is any one of a computer, a server, and a single-chip microcomputer, the computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

Images