Water quality monitoring intelligent early warning method and system based on multi-dimensional data analysis

Abstract

The application discloses a water quality monitoring intelligent early warning method and system based on multi-dimensional data analysis, and belongs to the technical field of water quality monitoring. The method comprises the following steps: processing missing values and noises through a dynamic weighted data filling algorithm to obtain complete water quality data; adopting an improved t-SNE algorithm for multi-parameter collaborative dimension reduction to map high-dimensional water quality parameters to a three-dimensional feature space; constructing a dynamic watershed partition model based on a Delaunay triangular net and DEM data; calculating a spatial anomaly score through a weighted local outlier factor; combining STL decomposition and a Transformer model for time anomaly detection; realizing multi-scale fluctuation detection by using wavelet packet decomposition; fusing the spatial, time and parameter dimensions to construct a three-dimensional judgment matrix and generate a comprehensive anomaly score; and triggering a multi-level early warning mechanism according to the score result. The application can monitor water quality changes in real time, and timely issue a warning when an anomaly occurs, thereby guaranteeing water quality safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water quality monitoring, and particularly relates to a water quality monitoring intelligent early warning method and system based on multi-dimensional data analysis. BACKGROUND

[0002] Water resources are an important basis for human survival and social and economic development, and the quality of water directly affects the ecological environment, public health and industrial production. With the acceleration of industrialization and urbanization, water quality monitoring has become an important means to ensure water resources security. Traditional water quality monitoring methods rely on manual periodic sampling and on-site detection, which have the disadvantages of poor timeliness, high cost and limited coverage, and are easily affected by external environments. Existing technologies mostly use static analysis methods, which lack comprehensive analysis of the spatio-temporal dynamic changes of water quality data, specifically in that: (1) data missing and noise processing are simple and static, which is difficult to adapt to water quality dynamic fluctuations; (2) analysis methods are mostly limited to a single dimension, which breaks the spatio-temporal correlation and weakens the ability to capture complex nonlinear patterns and multi-scale cycles; (3) the early warning mechanism relies on fixed thresholds, which has low sensitivity and is difficult to early detect gradual or sudden abnormalities in multiple parameter coordination, and lacks fine-grained response capabilities. Therefore, how to improve the accuracy, real-time and automation of the water quality monitoring system has become a technical problem to be solved. SUMMARY

[0003] The present application aims to provide a water quality monitoring intelligent early warning method and system based on multi-dimensional data analysis, which comprehensively considers the spatio-temporal distribution characteristics of data, dynamically adjusts the data filling method, uses advanced data dimension reduction and analysis algorithms, and combines a multi-level early warning mechanism to realize accurate monitoring and real-time early warning of water quality changes.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0005] In a first aspect, the present application provides a water quality monitoring intelligent early warning method based on multi-dimensional data analysis, comprising:

[0006] S1, dynamic weighted data filling: obtaining water quality parameter data of each monitoring point, filling and predicting missing values and noise of water quality monitoring to obtain complete water quality parameter data;

[0007] S2, multi-parameter collaborative dimension reduction: according to the distance between each monitoring point, using an improved t-SNE algorithm to map high-dimensional water quality parameter data to a three-dimensional feature space to obtain three-dimensional feature space coordinates;

[0008] S3, dynamic watershed partitioning: according to the distance threshold, the Delaunay triangulation network is constructed to obtain the adjacency relationship between the monitoring points, the Voronoi diagram is generated combined with the hydrological connectivity constraint, and the cell weight is dynamically adjusted based on the DEM to obtain the watershed spatial partitioning model reflecting the terrain weight;

[0009] S4, spatial outlying degree calculation: according to the weighted local outlier factor, the alarm threshold of each monitoring point is dynamically set to obtain the spatial anomaly score of each monitoring point;

[0010] S5, mixed time series decomposition: according to the STL decomposition method, the water quality time series is decomposed, the Transformer model is input, and the rainfall period function is fused to perform time anomaly detection and scoring to obtain the time dimension anomaly score;

[0011] S6: multi-scale fluctuation detection: db4 wavelet basis function is used for wavelet packet decomposition, and the energy entropy of each frequency band is calculated. When the high frequency band energy entropy suddenly increases, the fluctuation warning is triggered;

[0012] S7: three-dimensional judgment matrix construction: a three-dimensional matrix is constructed by fusing spatial outlying degree, time fluctuation degree and parameter coordination degree based on principal component analysis, and a comprehensive anomaly score is obtained by constructing a dynamic threshold surface equation;

[0013] S8: setting warning level and triggering condition: according to the comprehensive anomaly score and the persistence condition, a multi-level response mechanism is triggered.

[0014] In some embodiments, the step S1 of filling and predicting missing values and noise of water quality monitoring specifically comprises:

[0015] The historical data fluctuation amplitude is obtained by calculating the difference between the current time water quality parameter of each monitoring point and the previous time;

[0016] The noise covariance matrix is dynamically adjusted according to the historical data fluctuation amplitude, and the specific adjustment formula is:

[0017]

[0018] Wherein, Q t is the process noise covariance matrix at time t, Q t is the process noise covariance matrix at time t-1, α is the forgetting factor, the value is between 0.85 and 0.95, ω is the sliding window width, is the transpose of state increment at time t, reflecting the change of state at time t relative to the previous time, ΔX t is the change vector of historical data, and ω is the sliding window width;

[0019] Through parameter sensitivity analysis, the sensitivity of each parameter to observation noise is evaluated, the weight is adjusted, and the observation noise matrix is obtained.

[0020] The Kalman filtering algorithm is used to predict and fill in the missing water quality data by combining the dynamically adjusted noise covariance matrix and the observation noise matrix.

[0021] In some embodiments, the step S2 specifically comprises:

[0022] The water quality parameters of each monitoring point are constructed into a vector, the weighted Mahalanobis distance is used to calculate the distance between each monitoring point, and the commonly used Euclidean distance in t-SNE is replaced; the perplexity parameter is selected as the square root of the number of monitoring points, the t-SNE algorithm based on the weighted Mahalanobis distance is used for dimension reduction, and the water quality data is mapped from high dimension to three-dimensional space after the t-SNE dimension reduction.

[0023] In some embodiments, the step S3 specifically comprises:

[0024] For each monitoring site, the distance between all sites in the basin is calculated, and the distance threshold D is set according to the basin area and the number of sites, and the formula is:

[0025]

[0026] Under the calculated distance threshold, the Delaunay triangular network algorithm is used to connect the sites to form the adjacency relationship between the sites; the hydrological connectivity constraint is introduced, and for the dams or other barriers in the basin, the connection between them is automatically disconnected; based on the Delaunay triangular network, the Voronoi diagram is generated, the cell weight is adjusted combined with the DEM data of the basin, and the weight formula is:

[0027]

[0028] wherein, h i is the average elevation of the cell, h avg is the average elevation of the basin, and k is an adjustment coefficient, which is 0.01-0.05.

[0029] In some embodiments, the step S5 specifically comprises:

[0030] According to the STL decomposition method, the water quality time series is decomposed, and the decomposition results include: trend term T(t), seasonal term S(t) and residual term R(t);

[0031] The trend term T(t) and the residual term R(t) are input into the Transformer model, combined with the basin rainfall periodic function P(t)=sin(2πt / 365)+0.3sin(2πt / 7), and the data is subjected to anomaly detection, and the calculation formula of the anomaly score is:

[0032] Scoret = λ · |R(t)| + (1 - λ) · Attention Weight (t)

[0033] wherein λ is 0.6 to 0.8, Attention Weight is the abnormal attention weight calculated by the Transformer model.

[0034] In some embodiments, the step S6 triggers a fluctuation warning when the high-frequency band energy entropy of >0.5 Hz suddenly increases by more than 2 times the standard deviation of the historical average.

[0035] In a second aspect, the present application provides a water quality monitoring intelligent early warning system based on multi-dimensional data analysis, comprising:

[0036] A data acquisition and processing module is configured to acquire water quality parameter data of each monitoring point in real time, perform dynamic weighted data filling and multi-parameter collaborative dimension reduction, and output complete water quality data and three-dimensional feature space coordinates.

[0037] A watershed partitioning module is configured to construct a dynamic watershed spatial partitioning model.

[0038] A spatial and temporal analysis module is configured to calculate a weighted local outlier factor and dynamically set an alarm threshold, perform time anomaly detection through a Transformer model that decomposes and fuses a rainfall period function, and output a spatial anomaly score and a time dimension anomaly score.

[0039] A fluctuation detection module is configured to perform wavelet packet decomposition and frequency band energy entropy analysis, and trigger a fluctuation warning when the high-frequency band energy entropy suddenly increases.

[0040] A comprehensive judgment module is configured to construct a three-dimensional judgment matrix and calculate a comprehensive anomaly score through a dynamic threshold surface equation.

[0041] An early warning execution module is configured to trigger a multi-level response mechanism according to the comprehensive score.

[0042] In some embodiments, the data acquisition and processing module comprises:

[0043] A data acquisition module is configured to acquire water quality parameter data of each monitoring point in real time.

[0044] A dynamic filling module is configured to perform dynamic weighted data filling and output complete water quality data.

[0045] A dimension reduction processing module is configured to perform multi-parameter collaborative dimension reduction and generate three-dimensional feature space coordinates.

[0046] Based on the above technical solutions, the embodiments of the present application can at least produce the following technical effects:

[0047] The application realizes high-precision data preprocessing through dynamic weighted data filling, improved t-SNE dimension reduction and dynamic watershed partitioning; combined with weighted local outlier factor, STL-Transformer time series decomposition and wavelet packet energy entropy analysis, a space-time-parameter three-dimensional judgment matrix is constructed, which significantly improves the abnormal detection sensitivity; based on the comprehensive score triggering multi-level early warning mechanism, the single over-limit alarm to risk precise grading control is realized, which can monitor water quality changes in real time and timely issue early warning when abnormal occurs, providing intelligent protection for water quality safety. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, below will briefly introduce the drawings needed to be used in the embodiments or prior art description, obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the structure shown in the drawings.

[0049] Figure 1 is the flow chart of the embodiment of the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely below, obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on that the combination of technical solutions can be realized by those skilled in the art, when the combination of technical solutions appears contradictory or cannot be realized, it should be considered that the combination of technical solutions does not exist, and is not within the protection scope required by the present application.

[0051] The purpose of the present application is realized by the following technical solutions:

[0052] Embodiment 1, as shown in the figure, the present embodiment provides a water quality monitoring intelligent early warning method based on multi-dimensional data analysis, specifically including the following steps: Figure 1

[0053] Step 1: dynamic weighted data filling

[0054] Adaptive Kalman filtering algorithm based on sliding window is adopted, the purpose is to improve the accuracy and reliability of data by filling and noise processing of water quality monitoring data.

[0055] (1) obtaining original water quality monitoring data

[0056] ​Collect water quality parameter data from various monitoring sites, including but not limited to pH, COD, dissolved oxygen, etc. These data may have missing values or noise and need to be processed.

[0057] (2) Calculate the fluctuation amplitude of historical data

[0058] For each monitoring point, calculate the change vector of historical data, which is the difference between the current time water quality parameter and the previous time. The formula is:

[0059] ΔX t =X t -X t-1

[0060] Calculate the data change range in a certain time window around the monitoring point t, and select the variance calculated by the sliding window method.

[0061] (3) Dynamically adjust the noise covariance matrix Q

[0062] According to the calculated fluctuation amplitude of historical data, dynamically adjust the noise covariance matrix Q, and the specific adjustment formula is:

[0063]

[0064] Where Q t is the process noise covariance matrix at time t, Q t is the process noise covariance matrix at time t-1, α is the forgetting factor (value between 0.85 and 0.95), ω is the sliding window width (recommended 7 to 30 days), is the transpose of the state increment at time t, which is generally obtained by actual system state observation, estimation, etc., reflecting the change of state at time t relative to the previous time.

[0065] (4) Difference in observation noise matrix R setting

[0066] Set different observation noise matrices R for different water quality parameters such as pH, COD, etc. Through parameter sensitivity analysis, evaluate the sensitivity of each parameter to observation noise, and then adjust its weight.

[0067] (5) Kalman filtering for data filling

[0068] Use Kalman filtering algorithm combined with dynamically adjusted noise covariance matrix Q and observation noise matrix R to predict and fill the missing water quality data.

[0069] Step 2: Multi-parameter collaborative dimension reduction

[0070] (1) Calculate the weighted Mahalanobis distance between monitoring points

[0071] The water quality parameters of each monitoring point are constructed into a vector, and the weighted Mahalanobis distance is used to calculate the distance between each monitoring point, replacing the commonly used Euclidean distance in t-SNE. The formula of weighted Mahalanobis distance is:

[0072]

[0073] where X i , X j are the vectors of water quality parameters of i, j two monitoring points, and W is the weighted matrix determined according to the water quality standard limit, reflecting the importance of different water quality parameters.

[0074] (2) Set the parameters of t-SNE

[0075] The perplexity parameter is selected as the square root of the number of monitoring points (for example: ), to ensure the stability of the dimension reduction process. The t-SNE algorithm based on weighted Mahalanobis distance is used for dimension reduction.

[0076] (3) Generate a three-dimensional feature space

[0077] After t-SNE dimension reduction, the water quality data is mapped from high-dimensional space to three-dimensional space. Ensure that the reduced data can retain more than 90% of the original data variance, and provide an effective feature space for subsequent analysis.

[0078] Step 3: Dynamic watershed partitioning

[0079] (1) Calculate the distance threshold

[0080] For each monitoring site, calculate the distance between all sites in the watershed. According to the watershed area and the number of sites, set the distance threshold D:

[0081]

[0082] (2) Construct Delaunay triangulation

[0083] Under the calculated distance threshold, use the Delaunay triangulation algorithm to connect the sites, forming the adjacency relationship between the sites.

[0084] (3) Introduce hydrological connectivity constraints

[0085] For dams or other barriers in the watershed, automatically disconnect them to ensure that the connectivity of the watershed meets the actual situation.

[0086] (4) Generate Voronoi diagram and adjust weights

[0087] Based on the Delaunay triangulation, the Voronoi diagram is generated. Then, combined with the digital elevation model (DEM) data of the watershed, the cell weight is adjusted, and the weight formula is:

[0088]

[0089] where h i is the average elevation of the cell, h avg is the average elevation of the watershed, and k is the adjustment coefficient, taking values between 0.01 and 0.05.

[0090] Step 4: Spatial Outlier Calculation

[0091] (1) Calculation of Weighted Local Outlier Factor

[0092] For each monitoring point, the weighted local outlier factor (WLOF) is calculated. The calculation formula of WLOF is:

[0093]

[0094] where x ij is the value of the jth parameter of the ith station, ω j is the ecological importance weight of parameter j, μ j and σ j are the mean and standard deviation of parameter j, and N k is the number of neighboring points.

[0095] (2) Dynamic Alarm Threshold Calculation

[0096] The alarm threshold of each monitoring point is dynamically calculated, set as the larger value of the 95th percentile value of historical data and 3 times the absolute deviation of the median, to ensure timely detection of abnormal situations.

[0097] Step 5: Mixed Time Series Decomposition

[0098] (1) Time Series Decomposition

[0099] The original water quality time series data is decomposed using the STL (seasonal, trend, and residual decomposition) method. The decomposition results include: trend term T(t), seasonal term S(t), and residual term R(t).

[0100] (2) Abnormality Detection Using Transformer

[0101] The trend term T(t) and residual term R(t) are input into the Transformer model, combined with the watershed rainfall periodic function P(t) = sin(2πt / 365) + 0.3sin(2πt / 7), to perform abnormality detection on the data.

[0102] (3) Abnormality Score Calculation

[0103] The calculation formula of the anomaly score is:

[0104] Score t = λ · |R(t)| + (1 - λ) · Attention Weight (t)

[0105] Wherein, λ is a weight coefficient, usually takes the value of 0.6 to 0.8, Attention Weight is the abnormal attention weight calculated by the Transformer model.

[0106] Step 6: Multi-scale fluctuation detection

[0107] (1) Wavelet packet decomposition

[0108] The water quality signal is decomposed using db4 wavelet basis function, and the decomposition layer is determined according to the sampling frequency of the data (5 layers for daily data and 7 layers for hourly data).

[0109] (2) Calculate the energy entropy of the frequency band

[0110] For each frequency band, calculate its energy entropy, the formula is:

[0111]

[0112] Wherein, E b is the energy entropy, C b is the energy component, E total represents the total energy.

[0113] (3) Fluctuation early warning trigger

[0114] Calculate the energy entropy of each frequency band. When the energy entropy of high frequency band (> 0.5Hz) increases more than 2 times of the standard deviation of historical mean, trigger the fluctuation early warning.

[0115] Step 7: Construction of three-dimensional judgment matrix

[0116] (1) Construction of three-dimensional matrix

[0117] According to the spatial outlying degree, time fluctuation degree and parameter coordination degree, a three-dimensional matrix is constructed. The spatial outlying degree is calculated by WLOF, the time fluctuation degree is analyzed by STL-Transformer model, and the parameter coordination degree is calculated by principal component analysis.

[0118] (2) Dynamic threshold surface equation

[0119] Set the dynamic threshold surface equation, use the coefficients a, b, c, d calculated by the historical data training to calculate the threshold surface:

[0120] F(x, y, z) = a x 2 + b y 2 + c z 2 + d x y z

[0121] where T is an adjustable threshold value, and the default value is 75.

[0122] Step 8: Set the warning level and trigger condition

[0123] According to the results of the decision matrix, set the warning level and trigger condition, as shown in the following table:

[0124] Level Triggering condition Response action Level I 70 < Score < 80 for 3 hours Automatic start of adjacent site review detection Level II 80 < Score < 90 or single point increase of 50% Triggering of unmanned aerial vehicle cruise sampling Level III Score ≥ 90 or multi-point coordination anomaly in the basin Activation of emergency tracing model and notification of regulatory authorities

[0125] Example 2

[0126] This embodiment provides a water quality monitoring intelligent warning system based on multi-dimensional data analysis, which includes:

[0127] Data acquisition and processing module: for real-time acquisition of water quality parameter data of each monitoring point, dynamic weighted data filling and multi-parameter collaborative dimension reduction, output of complete water quality data and three-dimensional feature space coordinates;

[0128] Basin partitioning module: construct a dynamic basin spatial partitioning model;

[0129] Spatial and temporal analysis module: for calculating weighted local outlier factor and dynamically setting alarm threshold, performing time anomaly detection through STL decomposition and fusion of rainfall period function Transformer model, outputting spatial anomaly score and time dimension anomaly score;

[0130] Fluctuation detection module: for performing wavelet packet decomposition and frequency band energy entropy analysis, and triggering fluctuation warning when high frequency band energy entropy suddenly increases;

[0131] Comprehensive judgment module: for constructing a three-dimensional decision matrix and calculating a comprehensive anomaly score through a dynamic threshold surface equation;

[0132] Warning execution module: trigger a multi-level response mechanism according to the comprehensive score.

[0133] In this embodiment, the data acquisition and processing module includes:

[0134] Data acquisition module: for real-time acquisition of water quality parameter data of each monitoring point;

[0135] Dynamic filling module: for performing dynamic weighted data filling and outputting complete water quality data;

[0136] Dimension reduction processing module: for realizing multi-parameter collaborative dimension reduction and generating three-dimensional feature space coordinates.

[0137] The above shows and describes the basic principles and main features of the present application and the advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A water quality monitoring intelligent early warning method based on multidimensional data analysis, characterized in that, include: S1. Dynamic weighted data filling: Obtain water quality parameter data from each monitoring point, fill in and predict missing values ​​and noise in water quality monitoring to obtain complete water quality parameter data; S2. Multi-parameter collaborative dimensionality reduction: Based on the distance between each monitoring point, the improved t-SNE algorithm is used to map the high-dimensional water quality parameter data to a three-dimensional feature space to obtain the three-dimensional feature space coordinates. S3. Dynamic watershed zoning: Based on the distance threshold, a Delaunay triangulation network is constructed to obtain the adjacency relationship between monitoring points. Combined with hydrological connectivity constraints, a Voronoi diagram is generated, and the cell weights are dynamically adjusted based on the DEM to obtain a watershed spatial zoning model that reflects the topographic weight. S4. Spatial outlier calculation: The alarm threshold for each monitoring point is dynamically set according to the weighted local outlier factor to obtain the spatial anomaly score of each monitoring point. S5. Hybrid temporal decomposition: Decompose water quality time series according to STL decomposition method, input into Transformer model, and integrate rainfall periodic function to detect and score time anomalies, and obtain time dimension anomaly score; S6: Multi-scale fluctuation detection: Wavelet packet decomposition is performed using the db4 wavelet basis function to calculate the energy entropy of each frequency band. When the energy entropy of the high frequency band suddenly increases, a fluctuation warning is triggered. S7: Construction of a three-dimensional judgment matrix: A three-dimensional matrix is ​​constructed by integrating spatial outlier degree, temporal volatility and parameter synergy based on principal component analysis, and a comprehensive anomaly score is obtained by constructing a dynamic threshold surface equation; S8: Set warning levels and trigger conditions: Trigger a multi-level response mechanism based on the comprehensive anomaly score and persistence conditions.

2. The intelligent early warning method for water quality monitoring based on multidimensional data analysis according to claim 1, characterized in that, The step S1 of filling in and predicting missing values ​​and noise in water quality monitoring specifically includes: By calculating the difference between the water quality parameters at the current moment and the previous moment at each monitoring point, the fluctuation range of historical data can be obtained. The noise covariance matrix is ​​dynamically adjusted based on the historical data fluctuation amplitude. The specific adjustment formula is as follows: Where, q t q is the process noise covariance matrix at time t. t Let be the process noise covariance matrix at time t-1, α be the forgetting factor with a value between 0.85 and 0.95, and ω be the sliding window width. ΔX is the transpose of the state increment at time t, reflecting the change in state at time t relative to previous values. t Let ω be the change vector of historical data, and ω be the width of the sliding window. By conducting parameter sensitivity analysis, the sensitivity of each parameter to observation noise is evaluated, and its weights are adjusted to obtain the observation noise matrix. The Kalman filter algorithm is used in conjunction with the dynamically adjusted noise covariance matrix and the observation noise matrix to predict and fill in missing water quality data.

3. The intelligent early warning method for water quality monitoring based on multidimensional data analysis according to claim 1, characterized in that, Step S2 specifically includes: The water quality parameters of each monitoring point are constructed into a vector, and the distance between each monitoring point is calculated using weighted Mahalanobis distance, replacing the Euclidean distance commonly used in t-SNE. The perplexity parameter is selected as the square root of the number of monitoring points, and dimensionality reduction is performed using the t-SNE algorithm based on weighted Mahalanobis distance. After dimensionality reduction by t-SNE, the water quality data is mapped from high dimension to three-dimensional space.

4. The intelligent early warning method for water quality monitoring based on multidimensional data analysis according to claim 1, characterized in that, Step S3 specifically includes: For each monitoring station, the distance between all stations within the watershed is calculated. Based on the watershed area and the number of stations, a distance threshold D is set using the following formula: Under the calculated distance threshold, the Delaunay triangulation algorithm is used to connect stations, establishing adjacency relationships between them. Hydrological connectivity constraints are introduced to automatically disconnect dams or other barriers within the watershed. Based on the Delaunay triangulation, a Voronoi diagram is generated, and cell weights are adjusted using the watershed's DEM data. The weight formula is as follows: Among them, h i h is the average elevation of the cell. avg The average elevation of the basin is denoted by , and k is an adjustment coefficient with a value ranging from 0.01 to 0.

05.

5. The intelligent early warning method for water quality monitoring based on multidimensional data analysis according to claim 1, characterized in that, Step S5 specifically includes: The water quality time series was decomposed according to the STL decomposition method. The decomposition results include: trend term T(t), seasonal term S(t), and residual term R(t). The trend term T(t) and residual term R(t) are input into the Transformer model, and anomaly detection is performed on the data using the watershed rainfall periodic function P(t) = sin(2πt / 365) + 0.3sin(2πt / 7). The anomaly score is calculated using the following formula: Score t =λ·|R(t)|+(1-λ)·Attention Weight (t) Where λ is the weight coefficient, with a value ranging from 0.6 to 0.8, Attention Weight The anomaly attention weights calculated for the Transformer model.

6. The intelligent early warning method for water quality monitoring based on multidimensional data analysis according to claim 1, characterized in that, In step S6, when the energy entropy of the high-frequency band >0.5Hz suddenly increases by more than twice the standard deviation of the historical average, a fluctuation warning is triggered.

7. A water quality monitoring intelligent early warning system based on multidimensional data analysis, used to implement the method described in any one of claims 1-6, characterized in that, include: Data acquisition and processing module: used to acquire water quality parameter data from each monitoring point in real time, perform dynamic weighted data filling and multi-parameter collaborative dimensionality reduction, and output complete water quality data and three-dimensional feature space coordinates; Watershed partitioning module: Constructs a dynamic watershed spatial partitioning model; Spatial-temporal analysis module: used to calculate weighted local outlier factors and dynamically set alarm thresholds, and to detect temporal anomalies by decomposing and fusing the Transformer model of the rainfall periodic function using STL, outputting spatial anomaly scores and temporal dimension anomaly scores; Fluctuation detection module: used to perform wavelet packet decomposition and frequency band energy entropy analysis, triggering fluctuation warning when the high frequency band energy entropy suddenly increases; Comprehensive Judgment Module: Used to construct a three-dimensional judgment matrix and calculate a comprehensive anomaly score using a dynamic threshold surface equation; Early warning execution module: Triggers a multi-level response mechanism based on the comprehensive score.

8. The intelligent water quality monitoring and early warning system based on multidimensional data analysis according to claim 7, characterized in that, The data acquisition and processing module includes: Data acquisition module: used to acquire water quality parameter data from each monitoring point in real time; Dynamic data filling module: Used to perform dynamic weighted data filling and output complete water quality data; Dimensionality reduction module: Used to achieve multi-parameter collaborative dimensionality reduction and generate three-dimensional feature space coordinates.

Images