A real-time monitoring method and system for ecological environment quality

Abstract

The application provides an ecological environment quality real-time monitoring method and system, the method comprises the following steps: collecting N-dimensional ecological index time series data greater than two dimensions; setting an extreme value discrimination threshold with a historical quantile for an initial window; dynamically correcting the threshold value of a subsequent window in combination with a tail index, a robust kurtosis and a structure instability scalar of a previous window; calculating an initial dependence coefficient when two dimensions are synchronously over-limit, and obtaining a pair-wise dependence coefficient by combining the remaining dimension ternary co-occurrence frequency correction to construct a preliminary matrix; constructing a penalty matrix based on the normalized spherical distance between dimensions to regularize and obtain a high-dimensional extreme value dependence matrix; and representing the structure instability by calculating the Frobenius norm of the difference value of the adjacent window matrix through rolling calculation, and determining that the ecological environment has a mutation when the difference value exceeds a preset threshold.

Description

Technical Field

[0001] This application belongs to the field of ecological environment, and in particular relates to a method and system for real-time monitoring of ecological environment quality. Background Technology

[0002] In recent years, with the development of the Internet of Things (IoT) for ecological and environmental protection and multi-sensor technologies, ecological and environmental monitoring networks have been able to continuously acquire time-series data covering multiple aspects such as meteorology, hydrology, soil, and biology. Ecosystems exhibit nonlinear, multivariate coupling, and dynamic evolutionary characteristics, and complex spatiotemporal correlations often exist between various ecological and environmental indicators. Under the combined effects of global climate change and human activities, the risk of structural abrupt changes or degradation in ecological and environmental systems is constantly increasing. Systemic anomalies in the ecological environment are often not caused by changes in a single indicator, but rather by a chain reaction formed by the synchronous anomalies and interactions of multiple key indicators under extreme conditions. Therefore, how to utilize high-dimensional, multi-source ecological monitoring time-series data to extract structural change characteristics during the evolution of ecological indicators and promptly identify abnormal fluctuations in system status is a crucial problem that needs to be solved in the field of ecological and environmental monitoring and risk early warning.

[0003] Existing ecological and environmental monitoring and analysis methods typically focus on the average trend or linear correlation of indicators under normal conditions, paying insufficient attention to the synergistic anomalies between multiple ecological indicators under extreme conditions. In actual ecological processes, the extreme value dependencies between indicators may change with external disturbances, seasonal variations, and internal system succession. Relying solely on static thresholds or conventional correlation analysis is insufficient to reflect the tail characteristics of high-dimensional ecological data and the spatial distribution constraints between ecological indicators in a timely manner. Furthermore, existing methods are inadequate in characterizing the synergistic effects when multiple dimensional extreme values ​​occur simultaneously, and are easily affected by local noise or occasional fluctuations of single indicators, leading to insufficient timeliness in perceiving critical states and structural abrupt changes in ecosystems. Therefore, a monitoring method is needed that can identify the evolutionary characteristics of high-dimensional extreme value dependencies and combine them with the ecological connectivity attributes between variables to improve the timeliness and reliability of ecological and environmental abrupt change identification and early warning. Summary of the Invention

[0004] To address the problems of insufficient adaptability of static thresholds, inadequate characterization of extreme value synergy effects, and lagging identification of structural mutation dependence in existing technologies, this invention proposes a real-time monitoring method and system for ecological environment quality.

[0005] In a first aspect, the present invention proposes a method for real-time monitoring of ecological environment quality, comprising the following steps:

[0006] Obtain time-series data of N-dimensional ecological indicators with more than two dimensions; for the initial window, use historical quantiles as the extreme value discrimination thresholds for each dimension; for subsequent windows, calculate the basic thresholds based on the tail exponent and robust kurtosis values ​​of each dimension, and use the dependency structure instability scalar of the previous window to generate an adjustment factor to correct the basic thresholds, thereby obtaining the extreme value discrimination thresholds for the current window, wherein when the previous window is the initial window, a preset initial value is used as the dependency structure instability scalar of the previous window; When any two dimensions simultaneously exceed the extreme value discrimination threshold, the initial dependency coefficient is calculated; the ternary co-occurrence frequency of the remaining N-2 dimensions exceeding the extreme value discrimination threshold is counted to generate a correction term, and the initial dependency coefficient is adjusted using the correction term to obtain the pairwise dependency coefficient; a preliminary matrix is ​​constructed based on the pairwise dependency coefficients of all dimensions; a penalty matrix is ​​constructed based on the normalized spherical distance between dimensions, and the preliminary matrix is ​​regularized using the penalty matrix to obtain the high-dimensional extreme value dependency matrix; The high-dimensional extreme value dependency matrix is ​​generated in a rolling manner, and the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows is calculated as the dependency structure instability scalar of the current window; when the dependency structure instability scalar of the current window exceeds a preset threshold, it is determined that a sudden change has occurred in the ecological environment.

[0007] Optionally, obtaining N-dimensional ecological indicator time-series data (greater than two dimensions) includes: Establish communication channels with each monitoring terminal; Send a data read request containing time span parameters and dimension labels to the monitoring terminal; The system receives response data packets, parses and extracts measured values, and stores the measured values ​​and corresponding timestamps into a time-series database to obtain the time-series data of the N-dimensional ecological indicators.

[0008] Optionally, for the initial window, using historical quantiles as the extreme value discrimination thresholds for each dimension includes: Obtain all historical sampling data for the corresponding dimension and generate an unsorted set of values; The unsorted set of values ​​is sorted in ascending order to obtain an ascending sequence; A preset percentile in the ascending sequence is selected as the extreme value discrimination threshold of the dimension in the initial window.

[0009] Optionally, the step of calculating the basic threshold based on the tail exponent and robust kurtosis value of each dimension for subsequent windows includes: Calculate the quantiles of historical time-series data in each dimension to obtain robust kurtosis values ​​that represent the data distribution characteristics; The tail exponent of historical time series data is calculated using the Hill estimation algorithm. The robust kurtosis value and the tail index are input into a preset basic threshold evaluation model, and the basic threshold is calculated by combining the mean and standard deviation of the dimension.

[0010] Optionally, the step of using the dependency structure instability scalar of the previous window to generate an adjustment factor to correct the basic threshold and obtain the extreme value discrimination threshold of the current window includes: The dimensionless processing is performed on the dependent structure instability scalar, and a dimensionless adjustment factor negatively correlated with the dependent structure instability scalar is generated based on a preset exponential decay function, wherein the value of the dimensionless adjustment factor is greater than 0 and less than or equal to 1. The non-negative threshold reduction amount is calculated based on the difference obtained by subtracting the dimensionless adjustment factor from 1 and the preset positive scale parameter. The extreme value discrimination threshold for the current window is obtained by subtracting the non-negative threshold adjustment amount from the basic threshold.

[0011] Optionally, the step of generating a correction term by statistically analyzing the ternary co-occurrence frequencies of the remaining N-2 dimensions that exceed the extreme value discrimination threshold, and adjusting the initial dependency coefficient using the correction term to obtain the pairwise dependency coefficient, includes: The total frequency is obtained by summing the statistically obtained ternary co-occurrence frequencies. The total frequency is input into a preset logarithmic function, and the output is added to 1 to generate the correction term; The initial dependency coefficient is multiplied by the correction term to obtain the pairwise dependency coefficient.

[0012] Optionally, the step of generating the high-dimensional extreme value dependency matrix through rolling, and calculating the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows as a scalar of dependency structure instability for the current window, includes: For the initial window, its dependent structure instability scalar is set to a preset initial value, and the difference calculation between the high-dimensional extreme value dependency matrix of the previous window is not performed. For the second window and all subsequent windows, calculate the difference between the corresponding elements in the high-dimensional extreme value dependency matrix of the current window and the high-dimensional extreme value dependency matrix of the previous window; Squaring and summing all the differences gives the sum of squares; The Frobenius norm of the difference between the high-dimensional extremum dependence matrices of the adjacent windows is obtained by taking the square root of the sum of squares.

[0013] On the other hand, the present invention also proposes a real-time monitoring system for ecological environment quality, comprising the following modules: The generation module is used to acquire time series data of N-dimensional ecological indicators with more than two dimensions. For the initial window, historical quantiles are used as the extreme value discrimination thresholds for each dimension. For subsequent windows, basic thresholds are calculated based on the tail exponent and robust kurtosis values ​​of each dimension, and the basic thresholds are corrected by generating an adjustment factor using the dependency structure instability scalar of the previous window to obtain the extreme value discrimination threshold of the current window. When the previous window is the initial window, a preset initial value is used as the dependency structure instability scalar of the previous window. The calculation module is used to calculate the initial dependency coefficient when any two dimensions simultaneously exceed the extreme value discrimination threshold; to generate a correction term by counting the ternary co-occurrence frequency of the remaining N-2 dimensions that exceed the extreme value discrimination threshold; to adjust the initial dependency coefficient using the correction term to obtain pairwise dependency coefficients; to construct a preliminary matrix based on the pairwise dependency coefficients of all dimensions; to construct a penalty matrix based on the normalized spherical distance between dimensions; and to perform regularization processing on the preliminary matrix using the penalty matrix to obtain a high-dimensional extreme value dependency matrix. The determination module is used to continuously generate the high-dimensional extreme value dependency matrix, calculate the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows as the dependency structure instability scalar of the current window; when the dependency structure instability scalar of the current window exceeds a preset threshold, it is determined that a sudden change has occurred in the ecological environment.

[0014] Preferably, the step of acquiring N-dimensional ecological indicator time-series data (greater than two dimensions) includes: Establish communication channels with each monitoring terminal; Send a data read request containing time span parameters and dimension labels to the monitoring terminal; The system receives response data packets, parses and extracts measured values, and stores the measured values ​​and corresponding timestamps into a time-series database to obtain the time-series data of the N-dimensional ecological indicators.

[0015] Preferably, for the initial window, using historical quantiles as the extreme value discrimination thresholds for each dimension includes: Obtain all historical sampling data for the corresponding dimension and generate an unsorted set of values; The unsorted set of values ​​is sorted in ascending order to obtain an ascending sequence; A preset percentile in the ascending sequence is selected as the extreme value discrimination threshold of the dimension in the initial window.

[0016] Preferably, the step of calculating the basic threshold based on the tail exponent and robust kurtosis value of each dimension for subsequent windows includes: Calculate the quantiles of historical time-series data in each dimension to obtain robust kurtosis values ​​that represent the data distribution characteristics; The tail exponent of historical time series data is calculated using the Hill estimation algorithm. The robust kurtosis value and the tail index are input into a preset basic threshold evaluation model, and the basic threshold is calculated by combining the mean and standard deviation of the dimension.

[0017] Preferably, the step of using the dependency structure instability scalar of the previous window to generate an adjustment factor to correct the basic threshold and obtain the extreme value discrimination threshold of the current window includes: The dimensionless processing is performed on the dependent structure instability scalar, and a dimensionless adjustment factor negatively correlated with the dependent structure instability scalar is generated based on a preset exponential decay function, wherein the value of the dimensionless adjustment factor is greater than 0 and less than or equal to 1. The non-negative threshold reduction amount is calculated based on the difference obtained by subtracting the dimensionless adjustment factor from 1 and the preset positive scale parameter. The extreme value discrimination threshold for the current window is obtained by subtracting the non-negative threshold adjustment amount from the basic threshold.

[0018] Preferably, the step of generating a correction term by statistically analyzing the ternary co-occurrence frequencies of the remaining N-2 dimensions that exceed the extreme value discrimination threshold, and adjusting the initial dependency coefficient using the correction term to obtain the pairwise dependency coefficient, includes: The total frequency is obtained by summing the statistically obtained ternary co-occurrence frequencies. The total frequency is input into a preset logarithmic function, and the output is added to 1 to generate the correction term; The initial dependency coefficient is multiplied by the correction term to obtain the pairwise dependency coefficient.

[0019] Preferably, the step of generating the high-dimensional extreme value dependency matrix by rolling, and calculating the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows as a scalar of dependency structure instability for the current window, includes: For the initial window, its dependent structure instability scalar is set to a preset initial value, and the difference calculation between the high-dimensional extreme value dependency matrix of the previous window is not performed. For the second window and all subsequent windows, calculate the difference between the corresponding elements in the high-dimensional extreme value dependency matrix of the current window and the high-dimensional extreme value dependency matrix of the previous window; Squaring and summing all the differences gives the sum of squares; The Frobenius norm of the difference between the high-dimensional extremum dependence matrices of the adjacent windows is obtained by taking the square root of the sum of squares.

[0020] This invention can identify extreme value correlations and their variation characteristics among multidimensional ecological indicators. By fusing the tail exponent and robust kurtosis values ​​of ecological data and dynamically correcting the basic threshold using a scalar of previous dependency structure instability, it helps improve the adaptability and accuracy of multidimensional extreme value event identification. When constructing the high-dimensional extreme value dependency matrix, the pairwise dependency coefficients are corrected by combining the extreme value co-occurrence frequency of other dimensions, and a penalty matrix is ​​constructed using inter-dimensional spherical distance for regularization, thereby reducing the impact of spurious correlations with large spatial distances on dependency modeling and improving the stability and interpretability of the high-dimensional extreme value dependency matrix. By calculating the Frobenius norm of the difference between adjacent window matrices, the degree of change in the dependency structure within the ecosystem can be quantified, and sudden changes in the ecological environment can be determined. Attached Figure Description

[0021] Figure 1 A flowchart of the first embodiment; Figure 2 For ecological indicators time series plots; Figure 3 A schematic diagram of the Frobenius norm variation curve for matrix differences; Figure 4 This is a diagram illustrating the performance comparison of the models. Detailed Implementation

[0022] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0023] In the first embodiment, the present invention proposes a method for real-time monitoring of ecological environment quality, such as... Figure 1 As shown, it includes the following steps: S1, acquire N-dimensional ecological indicator time series data with more than two dimensions; for the initial window, use historical quantiles as the extreme value discrimination threshold for each dimension; for subsequent windows, calculate the basic threshold based on the tail exponent and robust kurtosis value of each dimension, and use the dependency structure instability scalar of the previous window to generate an adjustment factor to correct the basic threshold, thereby obtaining the extreme value discrimination threshold of the current window, wherein when the previous window is the initial window, a preset initial value is used as the dependency structure instability scalar of the previous window.

[0024] An IoT sensor network deployed in the monitoring area collects ecological indicators such as temperature, humidity, carbon monoxide concentration, ozone concentration, and fine particulate matter concentration in real time. Missing values ​​are filled using linear interpolation to construct a multidimensional time series matrix with N greater than two dimensions. A sliding window size is set to a preset time step. For the initial window, the 95th percentile of historical data for each dimension is calculated, and this quantile is set as the extreme value threshold for that dimension. For each subsequent sliding window, the tail exponent of each dimension's data is calculated using the Hill estimation algorithm from extreme value theory, and a robust kurtosis value based on the quantile is calculated. The tail exponent and robust kurtosis value are input into a preset basic threshold evaluation model to generate a basic threshold.

[0025] Obtain the dependency structure instability scalar generated in the previous window. If the previous window is the initial window, assign this scalar a preset constant of 0.1. Perform dimensionless processing on the dependency structure instability scalar and generate a dimensionless adjustment factor with a value greater than 0 and less than or equal to 1 based on a preset exponential decay function. Calculate the non-negative threshold reduction amount based on the difference between 1 and the dimensionless adjustment factor and a preset positive scale parameter. Subtract this non-negative threshold reduction amount from the base threshold to obtain the final extremum discrimination threshold for the current window. Therefore, as the dependency structure instability scalar increases, the non-negative threshold reduction amount increases accordingly, causing the extremum discrimination threshold of the current window to decrease relative to the base threshold.

[0026] In an optional embodiment, obtaining N-dimensional ecological indicator time-series data (greater than two dimensions) includes: Establish communication channels with each monitoring terminal; Send a data read request containing time span parameters and dimension labels to the monitoring terminal; The system receives response data packets, parses and extracts measured values, and stores the measured values ​​and corresponding timestamps into a time-series database to obtain the time-series data of the N-dimensional ecological indicators.

[0027] Establish TLS-encrypted communication channels with N IoT monitoring terminals based on TCP / IP or MQTT protocols, for example, N=5, including dimensions such as temperature, humidity, PM2.5, SO2, and NO2, and set a communication timeout threshold of 5000 milliseconds. Generate data read request instructions in JSON or Protobuf format, encapsulating time span parameters, such as "2023-01-01 00:00:00 to 2023-01-31 23:59:59" and target dimension labels, such as "temp_1,hum_1,pm25_1,so2_1,no2_1", and continuously send requests to the monitoring terminals at a frequency of 10Hz through the established communication channels. In the data return phase, receive JSON response data packets from each terminal and extract measured values ​​through key-value pair matching or regular expressions. For example, extract the measured value of 25.4℃ from the temperature sensor. Obtain the Unix timestamp aligned to the current network time protocol, such as 1672531200. Create key-value pairs with the measured values ​​and this timestamp, and write them to a distributed time-series database such as InfluxDB or TimescaleDB using batch insert operations, retaining storage precision to two decimal places. Align and aggregate the discrete data points according to the time dimension to obtain a 5-dimensional structured ecological indicator time-series data table with a sampling frequency of once per minute. The time-series fluctuation curves of representative ecological indicator measured values ​​within a continuous 140-minute period in the monitored area are shown below. Figure 2 As shown.

[0028] In an optional embodiment, the step of using historical quantiles as extreme value discrimination thresholds for each dimension for the initial window includes: Obtain all historical sampling data for the corresponding dimension and generate an unsorted set of values; The unsorted set of values ​​is sorted in ascending order to obtain an ascending sequence; A preset percentile in the ascending sequence is selected as the extreme value discrimination threshold of the dimension in the initial window.

[0029] Specific ecological dimensions, such as PM2.5 concentration, are extracted from a time-series database. This data includes all historical sampling data from the past year or a specified period. After data cleaning to remove null values ​​and outliers, an unsorted set of tens of thousands of measured values ​​(e.g., 10,000 samples) is generated. Using either quicksort or mergesort algorithms, with a time complexity of O(MlogM), where M is the size of the set, all values ​​in this unsorted set are sorted in ascending order, generating a monotonically increasing ascending sequence, such as [12.1, 12.3, 13.5, ..., 156.8]. After obtaining the ascending sequence, the target percentile parameter P=95% is used to calculate the index value Index=M×P. If the calculated index value is an integer, it is mapped to the value at that index position in the ascending sequence; if it is a decimal, it is rounded down and the result is calculated using linear interpolation of the preceding and following elements. For example, when the total historical data M is 10,000, if the system locates the 9,500th element and the extracted value is 85.5, then the scalar value of 85.5 is established as the extreme value threshold for the PM2.5 dimension within the initial time window of the system, such as the first 24 hours after startup. During this period, any real-time collected measured value greater than 85.5 is judged as an extreme ecological event exceeding the threshold.

[0030] In an optional embodiment, the step of calculating the base threshold based on the tail exponent and robust kurtosis value for subsequent windows includes: Calculate the quantiles of historical time-series data in each dimension to obtain robust kurtosis values ​​that represent the data distribution characteristics; The tail exponent of historical time series data is calculated using the Hill estimation algorithm. The robust kurtosis value and the tail index are input into a preset basic threshold evaluation model, and the basic threshold is calculated by combining the mean and standard deviation of the dimension.

[0031] For each dimension of ecological data, a sliding window of historical time-series data of length W is extracted, such as W=1440, representing minute-level data within a day. The quantiles of four key features—10%, 25%, 75%, and 90%—are calculated using a quantile algorithm and denoted as follows: and using the formula The robust kurtosis value was calculated. For example, if the PM2.5 quantiles within a certain window are 15, 25, 65, and 95 μg / m³, then the calculated robust kurtosis values ​​are... =2.0. Select the top 10% of extreme value samples in this dimension of data, such as 144 extremely high values. Use the Hill estimation algorithm to calculate the logarithmic excess mean of the extreme value samples relative to the high quantile threshold, obtaining the tail exponent ξ, which represents the heavy-tailed characteristic of the data. Assuming the maximum extreme value sample is 150 and the baseline threshold is 95, the calculated tail exponent ξ example value is 0.35. The obtained robust kurtosis value... The tail exponent ξ is used as a core parameter, input into a preset baseline threshold to evaluate the model. The model formula is as follows: This involves simultaneously calculating the mean μ of the historical data for the current window, for example, 45.0, and the standard deviation σ, for example, 20.0, and setting pre-tuned model weight hyperparameters α=0.5 and β=1.2. The example values ​​are then substituted into the calculation. ≈99.056. This calculation result is the basic threshold generated after integrating the current data distribution pattern and tail features.

[0032] In an optional embodiment, the step of using the dependency structure instability scalar of the previous window to generate an adjustment factor to correct the base threshold and obtain the extreme value discrimination threshold of the current window includes: The dimensionless processing is performed on the dependent structure instability scalar, and a dimensionless adjustment factor negatively correlated with the dependent structure instability scalar is generated based on a preset exponential decay function, wherein the value of the dimensionless adjustment factor is greater than 0 and less than or equal to 1. The non-negative threshold reduction amount is calculated based on the difference obtained by subtracting the dimensionless adjustment factor from 1 and the preset positive scale parameter. The extreme value discrimination threshold for the current window is obtained by subtracting the non-negative threshold adjustment amount from the basic threshold.

[0033] Extract the dependency structure instability scalar from the output of the previous time window. The scalar is dimensionless using the Min-Max extremum normalization algorithm. This is achieved by utilizing pre-recorded and continuously updated historical maximum values ​​of the scalar. , such as 5.0 and minimum value For example, 0.1 linearly maps the scalar to the standard closed interval [0,1].

[0034] Normalized values Substitute into the preset exponential decay function In this context, the attenuation coefficient λ is a constant set based on environmental sensitivity, preferably within the range of [0.1, 1.0], with an example value of λ = 0.2 taken here. The dimensionless adjustment factor is then calculated. With a value of ≈0.9048, a negative correlation feedback regulation mechanism was constructed in which the adjustment factor decreases exponentially as the instability scalar increases. A preset positive scaling parameter was set. The preset positive scale parameter has the same dimension as the base threshold of the corresponding dimension, and can be the absolute value of the base threshold, the historical standard deviation, or the interquartile range. In this example, we take... =99.056, then the non-negative threshold downsizing amount ≈9.43. The base threshold... Subtracting the non-negative threshold adjustment amount, we obtain the current window extreme value discrimination threshold. ≈89.63. This mechanism increases the threshold down-regulation as the scalar of dependent structural instability increases, thereby improving the sensitivity of identifying potential ecological mutation stages.

[0035] S2, when any two dimensions simultaneously exceed the extreme value discrimination threshold, calculate the initial dependency coefficient; count the ternary co-occurrence frequency of the remaining N-2 dimensions that exceed the extreme value discrimination threshold to generate a correction term, and use the correction term to adjust the initial dependency coefficient to obtain the pairwise dependency coefficient; construct a preliminary matrix based on the pairwise dependency coefficients of all dimensions; construct a penalty matrix based on the normalized spherical distance between dimensions, and use the penalty matrix to perform regularization processing on the preliminary matrix to obtain the high-dimensional extreme value dependency matrix.

[0036] Iterate through all pairwise combinations of the N-dimensional ecological indicators within the current window, generating Boolean sequences for each dimension to indicate whether it exceeds the corresponding extreme value threshold at each sampling time. Perform a logical AND operation on the two Boolean sequences to count the number of times they both exceed the threshold at the same sampling time. Perform a logical OR operation on the two Boolean sequences to count the number of sampling times when at least one of the two dimensions exceeds the threshold. When the number of sampling times is 0, set the initial dependency coefficient to 0. When the number of sampling times is greater than 0, use the ratio of the number of times the threshold is exceeded to the number of sampling times as the initial dependency coefficient.

[0037] For the currently selected pair of dimensions, iterate through the remaining N-2 ecological indicator dimensions, counting the number of times any one of these remaining dimensions and the two dimensions mentioned above simultaneously exceed their respective extreme value discrimination thresholds at the same time, i.e., the ternary co-occurrence frequency. Sum all the ternary co-occurrence frequencies obtained from the iteration, divide by the total time step of the current window, and perform normalization mapping to obtain the co-occurrence enhancement amount with values ​​in the [0,1] interval. When the ternary co-occurrence frequency is 0, the co-occurrence enhancement amount is 0. Multiply the initial dependency coefficient by the result of adding the co-occurrence enhancement amount to obtain the pairwise dependency coefficient. After iterating through all dimension pairs, an N×N symmetric preliminary matrix is ​​generated.

[0038] When constructing the penalty matrix, the deployment latitude and longitude coordinates of the sensors corresponding to each dimension are obtained, and the spherical distance between any two corresponding sensors in any dimension is calculated using the Half-Sinet (Haversine) algorithm. Subsequently, the spherical distance is normalized to the [0,1] interval, and the negative exponential function value of the normalized distance is used as the element at the corresponding position in the penalty matrix to construct an N×N penalty matrix. Finally, the Hadamard product operation is performed between the preliminary matrix and the penalty matrix to attenuate the dependencies between dimensions with large spatial distances, resulting in a high-dimensional extreme value dependency matrix.

[0039] In an optional embodiment, the step of generating a correction term by statistically analyzing the ternary co-occurrence frequencies of the remaining N-2 dimensions that exceed the extreme value discrimination threshold, and adjusting the initial dependency coefficient using the correction term to obtain the pairwise dependency coefficient, includes: The total frequency is obtained by summing the statistically obtained ternary co-occurrence frequencies. The total frequency is input into a preset logarithmic function, and the output is added to 1 to generate the correction term; The initial dependency coefficient is multiplied by the correction term to obtain the pairwise dependency coefficient.

[0040] For the collected N-dimensional ecological data, for example, N=5, labeled as dimensions A, B, C, D, and E, after determining that target dimensions A and B simultaneously exceed the extreme value discrimination threshold within the same time step, the remaining N-2 auxiliary dimensions, namely dimensions C, D, and E, are queried. Within a set time window, the frequency of dimensions A, B, and C simultaneously exceeding the threshold is counted. For example, the frequency of dimensions A, B, and D simultaneously going out of bounds is recorded as 5 times. For example, the frequency of times dimensions A, B, and E simultaneously cross the boundary. For example, record it as 0 times. Summate these N-2 independent ternary co-occurrence frequencies to calculate the total frequency. =8 times. After obtaining the total frequency, the total frequency is substituted into a preset natural logarithmic function model with a smoothing constant for nonlinear transformation. The mathematical expression of this logarithmic function is set as follows: The smoothing term "+1" is used to prevent the logarithmic calculation from becoming meaningless, while the adjustment coefficient η controls the amplification amplitude of the correction term, with a preferred range of [0.01, 0.1]. In this example, the set value η = 0.05 is used. Substituting the example values, ln(8+1)≈2.197 is calculated, resulting in the correction term. =1.10985. The pre-calculated initial dependency coefficient between dimension A and dimension B, assuming its example value is 0.45, is multiplied by this generated correction term to obtain the pairwise dependency coefficient. The calculated pairwise dependency coefficient is approximately 0.4994. When this coefficient exceeds a preset upper limit, a limiting process is applied. This correction mechanism can moderately enhance the dependency expression between the two dimensions of the target in the context of concurrent extreme events in multi-dimensional environmental indicators, thus characterizing systemic ecological pressure.

[0041] S3, the high-dimensional extreme value dependency matrix is ​​generated in a rolling manner, and the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows is calculated as the dependency structure instability scalar of the current window; when the dependency structure instability scalar of the current window exceeds a preset threshold, it is determined that the ecological environment has undergone a sudden change.

[0042] A loop structure is used to slide a window forward along the time axis with a fixed step size. At each new window position, the aforementioned operation is repeated to continuously generate new high-dimensional extreme value dependency matrices. For the initial window, the dependency structure instability scalar is set to a preset initial value, and matrix difference calculation is not performed. For the second window and subsequent windows, after each generation of the current window's matrix, the high-dimensional extreme value dependency matrix of the current window is subtracted from the high-dimensional extreme value dependency matrix of the previous adjacent window to obtain the difference matrix. The Frobenius norm of this difference matrix is ​​calculated, which is the square root of the sum of the squares of all elements in the difference matrix. The calculated Frobenius norm value is assigned to the dependency structure instability scalar of the current window. An empirically preset threshold, trained from historical stationary data, is pre-set in memory. Conditional judgment logic compares the current window's instability scalar with the preset threshold. When the instability scalar is determined to be greater than the preset threshold, system alarm logic is triggered, and a sudden change in the ecological environment is identified.

[0043] In an optional embodiment, the step of generating the high-dimensional extreme value dependency matrix through rolling, and calculating the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows as a scalar measure of the dependency structure instability of the current window, includes: For the initial window, its dependent structure instability scalar is set to a preset initial value, and the difference calculation between the high-dimensional extreme value dependency matrix of the previous window is not performed. For the second window and all subsequent windows, calculate the difference between the corresponding elements in the high-dimensional extreme value dependency matrix of the current window and the high-dimensional extreme value dependency matrix of the previous window; Squaring and summing all the differences gives the sum of squares; The Frobenius norm of the difference between the high-dimensional extremum dependence matrices of the adjacent windows is obtained by taking the square root of the sum of squares.

[0044] After obtaining the newly constructed N×N high-dimensional extreme value dependency matrix within the current time window, labeled as matrix A, and the same-dimensional matrix generated in the previous time window, labeled as matrix B, a residual matrix D representing the change in dependency structure is generated by calculating the difference between the numerical elements at the same row and column indices (i,j) of the two matrices through point-to-point subtraction. For example, for a 3×3 dependency matrix, if the element at index (1,2) in matrix A... =0.50, the element at the corresponding position in matrix B =0.45, then the absolute difference at that position is =0.05. After constructing the residual matrix D, for each difference element in the residual matrix... Perform squaring operations, that is, calculate For example, the square of the difference 0.05 above is 0.0025. (This is followed by a series of seemingly unrelated sentences.) The calculated square values ​​are summed up according to the formula. Obtain the global sum of squares. Assume that within a specific period, the sum of squares of all matrix elements is 0.0144. Perform the square root operation on this global sum of squares, Sum, to calculate... =0.12. The calculated result of 0.12 in this output is the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices between two adjacent time windows, which is used as an instability scalar to represent the degree of abrupt changes in the associated structure. The curve of the Frobenius norm change of the difference between the high-dimensional extreme value dependency matrices of adjacent windows under the sliding time window sequence is shown in the figure. Figure 3 As shown.

[0045] The experimental conditions used a one-year real-time series dataset from an IoT monitoring terminal deployed in a national nature reserve. This dataset contained 525,600 data records with a sampling frequency of minutes, covering five dimensions including temperature, humidity, and fine particulate matter. Forty-eight real-world extreme ecological mutation events were manually labeled within the dataset. The control group used a traditional static extremum-based method, while the experimental group employed the method described in this invention, which calculates the basic threshold based on the tail exponent and robust kurtosis, and corrects the threshold using a scalar method for dependency structure instability characterized by the Frobenius norm of the difference in the high-dimensional extremum dependency matrix. Both models ran on the same hardware platform with the same initial time window size.

[0046] The control group detected 35 real-world extreme events in the test set, with an abnormal event detection rate of 72.9%, and issued 122 false alarms, resulting in a false alarm rate of 12.5%, with an average alarm delay of 38 minutes. The experimental group successfully detected 46 real-world extreme events, achieving an abnormal event detection rate of 95.8%, and issued only 19 false alarms, reducing the false alarm rate to 1.9%, while also shortening the average alarm delay to 9 minutes. Figure 4 As shown.

[0047] This invention quantifies the instability of the dependency structure through the Frobenius norm of the difference between high-dimensional extreme value dependency matrices, and generates a dimensionless adjustment factor and a non-negative threshold downsampling amount based on this instability, enabling the extreme value discrimination threshold to be dynamically adjusted according to changes in the dependency structure. By combining ternary co-occurrence frequencies to enhance and correct paired dependency coefficients, the interference of single-dimensional random fluctuations on mutation identification is reduced, improving the timeliness and accuracy of early warnings, and enhancing the stability of monitoring results.

[0048] In the second embodiment, the present invention also proposes a real-time monitoring system for ecological environment quality, comprising the following modules: The generation module is used to acquire time series data of N-dimensional ecological indicators with more than two dimensions. For the initial window, historical quantiles are used as the extreme value discrimination thresholds for each dimension. For subsequent windows, basic thresholds are calculated based on the tail exponent and robust kurtosis values ​​of each dimension, and the basic thresholds are corrected by generating an adjustment factor using the dependency structure instability scalar of the previous window to obtain the extreme value discrimination threshold of the current window. When the previous window is the initial window, a preset initial value is used as the dependency structure instability scalar of the previous window. The calculation module is used to calculate the initial dependency coefficient when any two dimensions simultaneously exceed the extreme value discrimination threshold; to generate a correction term by counting the ternary co-occurrence frequency of the remaining N-2 dimensions that exceed the extreme value discrimination threshold; to adjust the initial dependency coefficient using the correction term to obtain pairwise dependency coefficients; to construct a preliminary matrix based on the pairwise dependency coefficients of all dimensions; to construct a penalty matrix based on the normalized spherical distance between dimensions; and to perform regularization processing on the preliminary matrix using the penalty matrix to obtain a high-dimensional extreme value dependency matrix. The determination module is used to continuously generate the high-dimensional extreme value dependency matrix, calculate the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows as the dependency structure instability scalar of the current window; when the dependency structure instability scalar of the current window exceeds a preset threshold, it is determined that a sudden change has occurred in the ecological environment.

[0049] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0050] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A real-time monitoring method of ecological environment quality, characterized in that, include: Obtain time-series data of N-dimensional ecological indicators with more than two dimensions; for the initial window, use historical quantiles as the extreme value discrimination thresholds for each dimension; for subsequent windows, calculate the basic thresholds based on the tail exponent and robust kurtosis values ​​of each dimension, and use the dependency structure instability scalar of the previous window to generate an adjustment factor to correct the basic thresholds, thereby obtaining the extreme value discrimination thresholds for the current window, wherein when the previous window is the initial window, a preset initial value is used as the dependency structure instability scalar of the previous window; Calculate the initial dependency coefficient when any two dimensions simultaneously exceed the extreme value discrimination threshold. The co-occurrence frequency of the remaining N minus 2 dimensions exceeding the extreme value discrimination threshold is used to generate a correction term. The initial dependency coefficient is adjusted using the correction term to obtain the pairwise dependency coefficient. A preliminary matrix is ​​constructed based on the pairwise dependency coefficients of all dimensions. A penalty matrix is ​​constructed based on the normalized spherical distance between dimensions. The preliminary matrix is ​​regularized using the penalty matrix to obtain a high-dimensional extreme value dependency matrix. The high-dimensional extreme value dependency matrix is ​​generated in a rolling manner, and the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows is calculated as the dependency structure instability scalar of the current window; when the dependency structure instability scalar of the current window exceeds a preset threshold, it is determined that a sudden change has occurred in the ecological environment.

2. The method of claim 1, wherein, The acquisition of N-dimensional ecological indicator time-series data (greater than two dimensions) includes: Establish communication channels with each monitoring terminal; Send a data read request containing time span parameters and dimension labels to the monitoring terminal; The system receives response data packets, parses and extracts measured values, and stores the measured values ​​and corresponding timestamps into a time-series database to obtain the time-series data of the N-dimensional ecological indicators.

3. The method of claim 1, wherein, For the initial window, historical quantiles are used as the extreme value discrimination thresholds for each dimension, including: Obtain all historical sampling data for the corresponding dimension and generate an unsorted set of values; The unsorted set of values ​​is sorted in ascending order to obtain an ascending sequence; A preset percentile in the ascending sequence is selected as the extreme value discrimination threshold of the dimension in the initial window.

4. The method according to claim 2, characterized in that, For subsequent windows, the basic threshold is calculated based on the tail exponent and robust kurtosis value of each dimension, including: Calculate the quantiles of historical time-series data in each dimension to obtain robust kurtosis values ​​that represent the data distribution characteristics; The tail exponent of historical time series data is calculated using the Hill estimation algorithm. The robust kurtosis value and the tail index are input into a preset basic threshold evaluation model, and the basic threshold is calculated by combining the mean and standard deviation of the dimension.

5. The method according to claim 1, characterized in that, The step of using the dependency structure instability scalar of the previous window to generate an adjustment factor to correct the basic threshold, and obtaining the extreme value discrimination threshold of the current window, includes: The dimensionless processing is performed on the dependent structure instability scalar, and a dimensionless adjustment factor negatively correlated with the dependent structure instability scalar is generated based on a preset exponential decay function, wherein the value of the dimensionless adjustment factor is greater than 0 and less than or equal to 1. The non-negative threshold reduction amount is calculated based on the difference obtained by subtracting the dimensionless adjustment factor from 1 and the preset positive scale parameter. The extreme value discrimination threshold for the current window is obtained by subtracting the non-negative threshold adjustment amount from the basic threshold.

6. The method according to claim 1, characterized in that, The method of generating a correction term by statistically analyzing the ternary co-occurrence frequencies of the remaining N-2 dimensions that exceed the extreme value discrimination threshold, and adjusting the initial dependency coefficient using the correction term to obtain pairwise dependency coefficients, including: The total frequency is obtained by summing the statistically obtained ternary co-occurrence frequencies. The total frequency is input into a preset logarithmic function, and the output is added to 1 to generate the correction term; The initial dependency coefficient is multiplied by the correction term to obtain the pairwise dependency coefficient.

7. The method according to claim 5, characterized in that, The process of generating the high-dimensional extreme value dependency matrix through rolling, and calculating the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows as a scalar measure of the dependency structure instability of the current window, includes: For the initial window, its dependent structure instability scalar is set to a preset initial value, and the difference calculation between the high-dimensional extreme value dependency matrix of the previous window is not performed. For the second window and all subsequent windows, calculate the difference between the corresponding elements in the high-dimensional extreme value dependency matrix of the current window and the high-dimensional extreme value dependency matrix of the previous window; Squaring and summing all the differences gives the sum of squares; The Frobenius norm of the difference between the high-dimensional extremum dependence matrices of the adjacent windows is obtained by taking the square root of the sum of squares.

8. A real-time monitoring system for ecological environment quality, characterized in that, include: The generation module is used to acquire time series data of N-dimensional ecological indicators with more than two dimensions. For the initial window, historical quantiles are used as the extreme value discrimination thresholds for each dimension. For subsequent windows, basic thresholds are calculated based on the tail exponent and robust kurtosis values ​​of each dimension, and the basic thresholds are corrected by generating an adjustment factor using the dependency structure instability scalar of the previous window to obtain the extreme value discrimination threshold of the current window. When the previous window is the initial window, a preset initial value is used as the dependency structure instability scalar of the previous window. The calculation module is used to calculate the initial dependency coefficient when any two dimensions simultaneously exceed the extreme value discrimination threshold. The co-occurrence frequency of the remaining N minus 2 dimensions exceeding the extreme value discrimination threshold is used to generate a correction term. The initial dependency coefficient is adjusted using the correction term to obtain the pairwise dependency coefficient. A preliminary matrix is ​​constructed based on the pairwise dependency coefficients of all dimensions. A penalty matrix is ​​constructed based on the normalized spherical distance between dimensions. The preliminary matrix is ​​regularized using the penalty matrix to obtain a high-dimensional extreme value dependency matrix. The determination module is used to continuously generate the high-dimensional extreme value dependency matrix, calculate the Frobenius norm of the difference between the high-dimensional extreme value dependency matrices of adjacent windows as the dependency structure instability scalar of the current window; when the dependency structure instability scalar of the current window exceeds a preset threshold, it is determined that a sudden change has occurred in the ecological environment.

9. The system according to claim 8, characterized in that, The acquisition of N-dimensional ecological indicator time-series data (greater than two dimensions) includes: Establish communication channels with each monitoring terminal; Send a data read request containing time span parameters and dimension labels to the monitoring terminal; The system receives response data packets, parses and extracts measured values, and stores the measured values ​​and corresponding timestamps into a time-series database to obtain the time-series data of the N-dimensional ecological indicators.

10. The system according to claim 8, characterized in that, For the initial window, historical quantiles are used as the extreme value discrimination thresholds for each dimension, including: Obtain all historical sampling data for the corresponding dimension and generate an unsorted set of values; The unsorted set of values ​​is sorted in ascending order to obtain an ascending sequence; A preset percentile in the ascending sequence is selected as the extreme value discrimination threshold of the dimension in the initial window.

Images