A real-time monitoring and management method and system for water conservancy and hydropower projects

Abstract

The present application discloses a real-time monitoring and management method and system for water conservancy and hydropower projects, which relates to the field of water conservancy and hydropower. The method comprises the following steps: collecting environmental data X and effect data Y, constructing a correlation coefficient matrix between the environmental data X and the effect data Y; constructing an anomaly detection algorithm library, and constructing an evaluation index matrix based on the correlation coefficient matrix and the anomaly detection algorithm library; selecting the optimal anomaly detection algorithm for each type of monitoring data based on the evaluation index matrix; performing outlier processing on the collected monitoring data based on the optimal anomaly detection algorithm; grouping the monitoring data after outlier processing according to season and working condition, combining the correlation coefficient matrix, and statistically calculating the mean and standard deviation of each group of data to obtain the mean ±2σ as the first-level warning threshold and the mean ±3σ as the second-level warning threshold; and performing monitoring and management based on the first-level warning threshold and the second-level warning threshold. In view of the lack of flexibility in the threshold setting of water conservancy projects in the prior art, which leads to low monitoring accuracy, the present application improves monitoring accuracy.

Description

Technical Field

[0001] The present application relates to the field of water conservancy and hydropower treatment, and in particular to a real-time monitoring and management method and system for water conservancy and hydropower projects. Background Art

[0002] Water conservancy and hydropower projects are critical national infrastructure, directly impacting flood control, water resource utilization, power supply, and ecological and environmental protection. With the rapid development of my country's water conservancy and hydropower sector, the number of large-scale projects continues to increase, posing increasingly complex challenges to their safe operation and management. These projects are influenced by a variety of environmental factors, including water level fluctuations, rainfall, and seasonal temperature swings. These factors, in turn, reflect their status through effects such as dam deformation and seepage changes. These factors interact to form a complex dynamic system.

[0003] Real-time monitoring is a key means of ensuring the safe operation of water conservancy and hydropower projects. Traditional monitoring systems for water conservancy and hydropower projects deploy various monitoring devices to collect environmental data such as water levels and rainfall, as well as data on effects such as dam deformation and seepage, generating massive streams of monitoring data. This data, after processing and analysis, provides a basis for project safety assessments and early warning decisions. With the development of the Internet of Things and big data technologies, the automation of monitoring equipment and the real-time nature of data collection have been significantly improved, creating conditions for more precise project monitoring.

[0004] However, the effective use of monitoring data still faces many challenges. On the one hand, monitoring data is easily affected by factors such as equipment failure, transmission interference, and environmental changes, resulting in outliers. On the other hand, there is a complex correlation between environmental data and effect data, which varies significantly with seasonal changes and project operating conditions.

[0005] For example, most existing systems use fixed thresholds or simple statistical models to set warning thresholds. These thresholds are often determined based on the overall statistical characteristics of historical data or empirical values, lacking differentiated considerations for different seasons and operating conditions. For example, the normal data fluctuation ranges during the summer flood season and the winter dry season vary significantly, and using a unified threshold will inevitably lead to reduced monitoring accuracy. Summary of the Invention

[0006] In response to the lack of flexibility in threshold setting in the existing technology, which leads to low monitoring accuracy, the present application provides a real-time monitoring and management method and system for water conservancy and hydropower projects. The method selects the optimal anomaly detection algorithm through comprehensive scoring, and generates differentiated warning thresholds based on season-operating condition combinations to improve monitoring accuracy.

[0007] The purpose of this application is achieved through the following technical solutions.

[0008] One aspect of the present application provides a real-time monitoring and management method for a water conservancy and hydropower project, comprising: S1, collecting monitoring data of the water conservancy and hydropower project, the monitoring data including environmental data X and effect data Y, the environmental data X including water level and rainfall; the effect data Y including dam deformation and seepage; S2, constructing a correlation coefficient matrix between the environmental data X and the effect data Y using a multivariate linear regression method; S3, constructing an anomaly detection algorithm library, the anomaly detection algorithm library including a mean method, a median method, a difference method, a Hampel filter method, and a 3σ criterion; S4, constructing an evaluation index matrix based on the correlation coefficient matrix and the anomaly detection algorithm library; S5, selecting an optimal anomaly detection algorithm for each type of monitoring data based on the evaluation index matrix; S6, performing outlier processing on the collected monitoring data based on the optimal anomaly detection algorithm; S7, grouping the monitoring data after outlier processing according to season and operating condition, combining the correlation coefficient matrix, and statistically calculating the mean and standard deviation of each group of data to obtain a mean ± 2σ as a first-level warning threshold and a mean ± 3σ as a second-level warning threshold; and S8, performing monitoring and management based on the first-level warning threshold and the second-level warning threshold.

[0009] Further, S2, a correlation coefficient matrix between environmental data X and effect data Y is constructed using a multivariate linear regression method, including: using Z-score to process environmental data X and effect data Y respectively, wherein environmental data X = {X1, X2}, X1 represents water level data, and X2 represents rainfall data; effect data Y = {Y1, Y2}, Y1 represents dam deformation data, and Y2 represents seepage data; time lag models Y1(t) and Y2(t) of dam deformation data Y1 and seepage data Y2 are constructed respectively; based on the time lag models Y1(t) and Y2(t), the Pearson correlation coefficient r of each environmental data X = {X1, X2} and effect data Y = {Y1, Y2} at different lag times is calculated by a sliding time window method,

[0010] r(X j (ti),Y l (t)),i∈[0,k],j∈{1,2},l∈{1,2}, k is the maximum lag order; i and j represent positive integers};

[0011] For each environmental data X j and effect data Y l Combination, from k+1 Pearson correlation coefficients r(X j (ti),Y l (t)) select the correlation coefficient r with the largest absolute value max and its corresponding optimal lag time t opt , forming a 2×2 time-lag correlation coefficient matrix R, the matrix elements R j,l Contains two parameter values ​​{rmax ,t opt}, where j∈{1,2} represents the environment data type, l∈{1,2} represents the effect data type; R 1,1 represents the maximum correlation coefficient between water level X1 and dam deformation Y1 and its corresponding lag time; R 1,2 represents the maximum correlation coefficient between water level X1 and seepage Y2 and its corresponding lag time; R 2,1 represents the maximum correlation coefficient between rainfall X2 and dam deformation Y1 and its corresponding lag time; R 2,2 represents the maximum correlation coefficient between rainfall X2 and seepage Y2 and its corresponding lag time;

[0012] Furthermore, the time lag models Y1(t) and Y2(t): Where t represents the current time point; k is the maximum lag order, which is determined based on the physical characteristics of water conservancy and hydropower projects and historical data analysis, and the value range is 1-30 days; α1, α2 are the constant terms of each model; β 1i ,β 2i are the regression coefficients of water level data X1 at different lag times in the two effect data models; γ 1i ,γ 2i are the regression coefficients of rainfall data X2 at different lag times in the two effect data models; ε1 and ε2 are the random error terms of each model.

[0013] Furthermore, S4 constructs an evaluation index matrix based on the correlation coefficient matrix and the anomaly detection algorithm library, including: the optimal lag time t in the time lag correlation coefficient matrix R opt , perform lag processing on the environmental data X to form a lagged environmental data set X' that is most correlated with the effect data Y; construct a data set based on the environmental data set X' and the effect data Y, and divide the data set into a training set and a test set; use each anomaly detection algorithm in the anomaly detection algorithm library to perform anomaly detection on the data in the training set; use the test set to calculate the evaluation indicators of the detection results of each anomaly detection algorithm, the evaluation indicators include detection accuracy A, false alarm rate B, missed alarm rate C and robustness index D; construct a 4×5 evaluation indicator matrix E, the matrix elements E i,j Contains the anomaly detection evaluation results, where i∈{1, 2, 3, 4} represents the monitoring data type (X1 water level, X2 rainfall, Y1 dam deformation, Y2 seepage), j∈{1, 2, 3, 4, 5} represents the anomaly detection algorithm type (mean method, median method, difference method, Hampel filter method and 3σ criterion); each matrix element E i,j The four evaluation indicators {A, B, C, D} of the corresponding anomaly detection algorithm for the corresponding data type are stored in .

[0014] Furthermore, S5 selects the optimal anomaly detection algorithm for each monitoring data according to the evaluation index matrix, including: setting the weight coefficient w of each evaluation index in the evaluation index matrix A 、w B 、w C 、w D ; For each monitoring data, calculate the comprehensive score S of each anomaly detection algorithm i,j :S i,j =w A ×A i,j -w B ×B i,j -w C ×C i,j +w D ×D i,j ; Among them, i∈{1, 2, 3, 4} represents the monitoring data type, and j∈{1, 2, 3, 4, 5} represents the anomaly detection algorithm type.

[0015] Furthermore, S5 selects the optimal anomaly detection algorithm for each monitoring data according to the evaluation index matrix, and also includes: for the environmental data X1 and X2, adding a correlation coefficient correction term when calculating the comprehensive score: S i,j '=S i,j ×(1+0.2×|r max |), where r∈{1,2} represents environmental data, |r max | is the absolute value of the maximum correlation coefficient corresponding to the environmental data in the time-lagged correlation coefficient matrix R; for each type of monitoring data, the anomaly detection algorithm with the highest comprehensive score is selected as the optimal algorithm for this data type: Represents effect data Y1 and Y2; i∈{1,2} represents environmental data X1 and X2; where A opt (i) is the optimal anomaly detection algorithm corresponding to the monitoring data type i.

[0016] Further, S7, the mean ± 2σ is obtained as the first-level warning threshold and the mean ± 3σ is obtained as the second-level warning threshold, including: dividing the monitoring data after the outlier processing into four groups according to the season; grouping the monitoring data after the outlier processing according to the working condition according to the water level state, the working condition includes normal water storage working condition and flood season precipitation working condition; forming a season-working condition combination according to each season and working condition combination, including spring-normal water storage, spring-flood season precipitation, summer-normal water storage, summer-flood season precipitation, autumn-normal water storage, autumn-flood season precipitation, winter-normal water storage and winter-flood season precipitation; for each season-working condition combination, respectively calculating the statistical parameters of each monitoring data type, the statistical parameters including the mean μ, standard deviation σ and skewness coefficient s; based on the data of each season-working condition combination, using the optimal lag time t in the time lag correlation coefficient matrix R optCalculate the real-time correlation response coefficient between environmental data X and effect data Y in each set of data According to the statistical parameters of each season-operating condition combination, combined with the calculated real-time correlation response coefficient Calculation of warning threshold: For effect data Y1 and Y2, the calculation formula of warning threshold is: μ±kσ, where k is the warning level coefficient; for environmental data X1 and X2, when |r max When |>0.7, the calculation formula for the warning threshold is: Where: δ is the correlation adjustment coefficient, which is 0.2; when |r max When |≤0.7, the warning threshold calculation formula is: μ±kσ, where k is the warning level coefficient; generate the first-level warning threshold and the second-level warning threshold for each group of data; integrate the warning thresholds under all season-operating condition combinations into the warning threshold matrix Θ, the matrix elements Θ i,j,m represents the warning threshold range of monitoring data type i in season j and working condition m,

[0017] Where: i∈{1, 2, 3, 4} represents the monitoring data type X1 water level, X2 rainfall, Y1 dam deformation, Y2 seepage; j∈{1, 2, 3, 4} represents spring, summer, autumn, and winter respectively; m∈{1, 2} represents normal water storage conditions and flood season precipitation conditions respectively; It represents the first-level warning lower limit of monitoring data type i in season j and working condition m; It represents the upper limit of the first-level warning for monitoring data type i in season j and working condition m; It represents the lower limit of the second-level warning for monitoring data type i in season j and working condition m; It represents the upper limit of the second-level warning for monitoring data type i in season j and working condition m;

[0018] Furthermore, the real-time correlation response coefficient The calculation formula is: Where: p∈{1, 2} represents the environmental data type (X1 water level, X2 rainfall); q∈{1, 2} represents the effect data type (Y1 dam deformation, Y2 seepage); t opt,p,q is the optimal lag time of the corresponding combination in the time-lag correlation coefficient matrix R; μ p and σ p They are environmental data X p Mean and standard deviation in the current season-operating condition combination; μ q and σ q The effect data Y q The mean and standard deviation in the current season-operating condition combination; n is the number of samples in the current season-operating condition combination.

[0019] Furthermore, the first-level warning threshold and the second-level warning threshold are generated for each group of data. The formula is:

[0020] Level 1 warning threshold:

[0021] Level 2 warning threshold:

[0022] Among them, μ i,j,m is the mean value of monitoring data type i in season j and operating condition m; σ i,j,m is the standard deviation of monitoring data type i in season j and operating condition m; i∈{1, 2, 3, 4} represents the monitoring data types X1 water level, X2 rainfall, Y1 dam deformation, and Y2 seepage, respectively; j∈{1, 2, 3, 4} represents spring, summer, autumn, and winter, respectively; m∈{1, 2} represents normal water storage conditions and flood season precipitation conditions, respectively.

[0023] C i,j,m is the correction coefficient, determined according to the data characteristics:

[0024] For effect data (i∈{3,4}) and the data are normally distributed |s i,j,m |<0.5, C i,j,m =1; for effect data (i∈{3,4}) and the data are skewed |s i,j,m When |≥0.5, the upper limit of warning is C i,j,m =1+0.1×s i,j,m , the lower limit of warning is C i,j,m =1-0.1×s i,j,m ;

[0025] For environmental data (i∈{1,2}) and |r max When |>0.7 (strong correlation):

[0026] when When the warning upper limit is Warning lower limit

[0027] when When the warning upper limit is Warning lower limit

[0028] For environmental data (i∈{1,2}) and |r max |≤0.7 (weak correlation), C i,j,m =1;

[0029] is the defined real-time correlation response coefficient; p∈{1,2} represents the environmental data type, when i=1, p=1, when i=2, p=2; q∈{1,2} represents the effect data type with the greatest correlation with the environmental data p; s i,j,m is the skewness coefficient of monitoring data type i in season j and operating condition m; δ is the correlation adjustment coefficient, which is fixed at 0.2; r max is the maximum correlation coefficient between the corresponding environmental data i and the effect data in the time-lagged correlation coefficient matrix R.

[0030] Another aspect of the present application also provides a real-time monitoring and management system for water conservancy and hydropower projects, which is used to implement a real-time monitoring and management method for water conservancy and hydropower projects of the present application.

[0031] Compared with the existing technology, the advantages of this application are:

[0032] Water conservancy and hydropower projects are subject to the complex influence of multiple environmental factors (such as seasonal changes, water level fluctuations, rainfall changes) and engineering conditions. There is a characteristic of obvious time lag relationship between environmental data and effect data. Existing technologies generally use fixed thresholds or simple statistical methods for monitoring, resulting in low warning accuracy, high false alarm rate and inability to adapt to changes in different seasons and working conditions. This application establishes a time lag correlation model for environmental and effect data, implements the selection of the optimal anomaly detection algorithm based on comprehensive scoring, and constructs a differentiated warning threshold matrix for season-working condition combinations. It achieves accurate identification and processing of monitoring data anomalies, precise quantification of the time lag relationship between environmental and effect data, and dynamic adaptive adjustment of warning thresholds, thereby significantly improving the accuracy and reliability of real-time monitoring of water conservancy and hydropower projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present application will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same numbers represent the same structures, wherein:

[0034] Figure 1 is an exemplary flow chart of a real-time monitoring and management method for water conservancy and hydropower projects according to some embodiments of the present application;

[0035] Figure 2 is an exemplary flow chart for constructing a time-lag correlation coefficient matrix R according to some embodiments of the present application;

[0036] Figure 3 is an exemplary flow chart of the method E for constructing an evaluation index matrix according to some embodiments of the present application;

[0037] Figure 4 This is an exemplary flowchart of generating warning thresholds according to some embodiments of the present application. DETAILED DESCRIPTION

[0038] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0039] like Figure 1 As shown, monitoring data of a water conservancy and hydropower project is collected, the monitoring data including environmental data X and effect data Y, the environmental data X including water level and rainfall; the effect data Y including dam deformation and seepage; a correlation coefficient matrix between the environmental data X and the effect data Y is constructed using a multivariate linear regression method; an anomaly detection algorithm library is constructed, the anomaly detection algorithm library including the mean method, median method, difference method, Hampel filter method, and 3σ criterion; an evaluation index matrix is ​​constructed based on the correlation coefficient matrix and the anomaly detection algorithm library; an optimal anomaly detection algorithm for each type of monitoring data is selected based on the evaluation index matrix; outlier processing is performed on the collected monitoring data based on the optimal anomaly detection algorithm; the monitoring data after outlier processing is grouped according to season and operating condition, and the mean and standard deviation of each group of data are statistically calculated based on the correlation coefficient matrix to obtain the mean ± 2σ as the first-level warning threshold and the mean ± 3σ as the second-level warning threshold; monitoring and management are performed based on the first-level warning threshold and the second-level warning threshold.

[0040] S1. Collect monitoring data for water conservancy and hydropower projects; install automated monitoring equipment at key locations on reservoirs and dams, including water level gauges, rain gauges, deformation gauges, and seepage monitoring equipment. Water level gauges are installed at representative locations within the reservoir area and downstream river channels; rain gauges are deployed around the reservoir area and in the upstream catchment area; deformation monitoring equipment (such as GNSS reference stations, total stations, and displacement gauges) is installed on the dam body, focusing on monitoring the dam crest and slope; and seepage monitoring equipment (such as piezometers and seepage flow measurement weirs) is deployed downstream of the dam and in the foundation area.

[0041] Different collection frequencies are set based on project characteristics and monitoring needs. Typically, water level data is collected every hour; rainfall data is accumulated every hour; dam deformation data is collected every 6-12 hours; and seepage data is collected every 6 hours. During flood season or under unusual weather conditions, the collection frequency can be increased, for example, water level and rainfall data can be adjusted to every 30 minutes.

[0042] IoT technology is used to achieve real-time transmission of monitoring data, transmitting it to the monitoring center via wired, wireless, or satellite networks. A tiered data storage strategy is used, with raw data stored in a local database and processed data transmitted to a cloud data center, ensuring data security and efficient access.

[0043] Preprocessing of the collected raw data involves unit conversion, numerical correction, and basic error screening. For example, seepage data is converted from milliliters per minute to liters per day, and dam deformation data is converted from instrument readings to actual displacement. Data that is clearly out of range is also removed.

[0044] The preprocessed data is categorized and stored into two main categories: environmental data (X) and effect data (Y). Environmental data X includes water level X1 and rainfall X2; effect data Y includes dam deformation Y1 and seepage Y2. Each data type is archived according to acquisition time, monitoring point location, and data type to facilitate subsequent analysis.

[0045] like Figure 2 As shown in S2, the time-lag correlation between environmental data and effect data is established by the multiple linear regression method. First, a historical data set of environmental data X and effect data Y for a certain period of time (such as the past three years) is collected; for water level data X1, its historical mean μ1 and standard deviation σ1 are calculated, and the formula is applied Standardize; for rainfall data X2, calculate its historical mean μ2 and standard deviation σ2, and apply the formula Standardize; for the dam deformation data Y1, calculate its historical mean μ3 and standard deviation σ3, and apply the formula Standardize; for the seepage data Y2, calculate its historical mean μ4 and standard deviation σ4, and apply the formula Standardize.

[0046] A standardized environmental data set X' = {X1', X2'} and a standardized effect data set Y' = {Y1', Y2'} are formed; the standardized data have a mean of 0 and a standard deviation of 1, eliminating dimensional differences; the standardization process does not change the correlation between the data, but makes different types of data comparable.

[0047] The appropriate maximum lag order k is determined based on the physical characteristics of the water conservancy and hydropower project. For concrete dams, k = 7 to 14 days is usually selected; for earth-rock dams, k = 14 to 30 days is usually selected. The k value that best reflects the environment-effect relationship can be selected through repeated verification with historical data.

[0048] For the standardized data set, prepare a historical observation sequence of sufficient length; construct a data set for each time point t, including the environmental data at the current time t and the past k time points, as well as the effect data at the current time t; form a sliding data window: {X1'(t),X1'(t-1),......,X1'(tk),X2'(t),X2'(t-1),.....,X2'(tk),Y1'(t),Y2'(t)}.

[0049] The time lag model of the dam deformation data Y1 expresses the relationship between the current deformation value and current and historical environmental factors: Estimate model parameters α1,β using the least squares method 1i ,γ 1i β in the model 1i The coefficient represents the influence of water level data X1 on the current dam deformation i days ago; γ in the model 1i The coefficient represents the intensity of the impact of rainfall data X2 on the current dam deformation i days ago.

[0050] The time lag model of seepage data Y2 expresses the relationship between the current seepage value and current and historical environmental factors: The least squares method is also used to estimate the model parameters α2, β 2i ,γ 2i ; Seepage hysteresis models focus on capturing the time process by which changes in water level and rainfall are transmitted to the seepage system.

[0051] Select an appropriate time window length w (such as 90 days or 180 days), which should be long enough to include seasonal variation characteristics; set the initial window [t-w+1,t] on the time series, including data of w consecutive time points; for each environment-effect combination (X j ,Y l ),j∈{1,2},l∈{1,2}, calculate the Pearson correlation coefficient at different lag times; for lag time i∈[0,k], calculate r(X j (ti),Y l (t)); Pearson correlation coefficient calculation formula:

[0052] where μ x ,μ y X in the window j (ti) and Y l The mean of (t), σ x ,σ y are the standard deviations, and n is the number of samples in the window.

[0053] Slide the window forward by a certain step size (e.g., 1 day) to form a new window [t-w+2, t+1]; repeatedly calculate the correlation coefficient of each lag time in the new window; continue sliding until the entire data range is covered; for each lag time i, take the average of the calculation results of all windows as the final correlation coefficient r(X j (ti),Y l (t)).

[0054] For each environment-effect combination (X j ,Y l), check the correlation coefficients under k+1 different lag times; select the correlation coefficient with the largest absolute value as the maximum correlation coefficient r of the combination max ; Record the corresponding lag time as the optimal lag time t of the combination opt For example, if the correlation coefficient between water level and seepage is 0.78 at a lag of 2 days, and this value is the maximum value among all lag times, then the r of this combination is max =0.78, t opt =2.

[0055] Construct a 2×2 time-lag correlation coefficient matrix R; each element in the matrix R j,l Contains parameter pairs {r max ,t opt}; R 1,1 represents the maximum correlation coefficient and optimal lag time between water level X1 and dam deformation Y1; R 1,2 represents the maximum correlation coefficient and optimal lag time between water level X1 and seepage Y2; R 2,1 represents the maximum correlation coefficient and optimal lag time between rainfall X2 and dam deformation Y1; R 2,2 It represents the maximum correlation coefficient and optimal lag time between rainfall X2 and seepage Y2.

[0056] The matrix R can be used to intuitively understand the impact intensity and time lag characteristics of each environmental factor on the effect index; for example, if R 2,2 = {0.85, 3}, indicating that rainfall has a significant effect on seepage (correlation coefficient 0.85) and that there is a 3-day time lag. If the correlation coefficients for multiple lag times are very close (difference < 0.05), multiple candidate optimal lag times can be recorded for further analysis.

[0057] S3. This implementation builds an algorithm library containing five typical anomaly detection algorithms for processing outliers in water conservancy and hydropower project monitoring data. The algorithm library is designed as a scalable architecture, with each algorithm targeting different types of anomaly characteristics and data distribution characteristics, effectively identifying multiple anomaly patterns.

[0058] Specifically, the mean method selects an appropriate time window length w (e.g., 7 days). For each monitoring data point x(t), a window is formed by taking the w data points before and after it. The mean μ of all data points in the window except the current point is calculated as (x(tw / 2)+...+x(t-1)+x(t+1)+...+x(t+w / 2)) / (w-1). The deviation d from the current data point and the mean is calculated as |x(t)-μ|. A threshold θ is set (usually 2-3 times the standard deviation within the window). If d>θ, x(t) is considered an outlier. The window length w is determined by the data collection frequency; daily data collection is usually set to 5-9 days. The threshold θ is determined using an adaptive method: θ=c×σ, where c is a coefficient (usually 2.5) and σ is the standard deviation within the window.

[0059] Median method: select a time window length w (usually an odd number, such as 7 days), take a total of w data points before and after each data point x(t), and calculate the median of all data in the window Calculate the deviation between the current point and the median d = |x(t)-med|; calculate the median absolute deviation of the deviation within the window Set the threshold θ = c × mad (c is usually 3.0). If d > θ, then x(t) is considered an outlier. The window length w is an odd value, usually 5 to 11 days. The determination coefficient c is adjusted based on the data's volatility, usually between 3.0 and 4.5.

[0060] The difference method detects mutation anomalies by calculating the rate of change of the difference between adjacent data points. Calculate the first-order difference sequence for the time series: diff(t) = x(t) - x(t-1); select the window length w, calculate the statistical characteristics of the historical difference (mean μ diff and standard deviation σ diff ); For the current difference value diff(t), calculate its standardized score Set the threshold θ (usually 2.5-3.5), if z>θ, then x(t) is judged to be an abnormal value. Difference calculation method: You can choose first-order difference or relative rate of change Statistical characteristic calculation cycle: determined according to the seasonal characteristics of the data, usually 30-90 days; judgment threshold θ: adjusted according to the project's sensitivity to mutations, usually 2.5-3.5.

[0061] Hampel filter method, select a window length w (such as 11 days), take a total of w points before and after each data point x(t); calculate the median in the window Calculate the deviation of each point in the window from the median: Calculate the median absolute deviation (madm) = 1.4826 × median(dev); set the decision threshold (θ) = k × madm (k is typically 3). If |x(t) - med| > θ, then x(t) is considered an outlier. The window length (w) is selected based on the frequency and volatility of the data, typically 9-15 days. The scaling factor (1.4826) makes the MADM equivalent to the standard deviation under a normal distribution. The determination coefficient (k) is adjusted based on the required outlier sensitivity, typically 3-5.

[0062] The 3σ criterion calculates the overall mean μ and standard deviation σ of the monitoring data series. For each data point x(t), its standardized score z = |x(t) - μ| / σ is calculated. If z > 3, x(t) is considered an outlier. The judgment threshold can be adjusted to 2σ or 4σ for different monitoring data types. The statistical calculation cycle is determined by the nature of the data and can be calculated globally or with a sliding window. The anomaly judgment threshold is set to 3 as a standard setting and can be adjusted to 2-4 depending on the data type.

[0063] like Figure 3 As shown in S4, based on the correlation coefficient matrix and the anomaly detection algorithm library, the evaluation index matrix is ​​constructed. Specifically, the optimal lag time t for each pair of environment-effect combinations is extracted from the time-lag correlation coefficient matrix R. opt ; Apply the optimal lag time t to the water level data X1 opt,1,1 and t opt,1,2 , forming two lagged sequences X1(tt opt,1,1 ) and X1(tt opt,1,2 ); Apply the optimal lag time t to the rainfall data X2 opt,2,1 and t opt,2,2 , forming two lagged sequences X2(tt opt,2,1 ) and X2(tt opt,2,2 ).

[0064] Integrate the processed lagged environmental data into a new data set X'; for each effect data, select the most relevant lagged environmental data: for Y1, select X1(tt opt,1,1 ) and X2(tt opt,2,1 ); select X1 for Y2(tt opt,1,2 ) and X2(tt opt,2,2 ); forming a lagged environmental dataset X' that is time-aligned with the effect data Y to ensure the correct expression of causal relationships.

[0065] Combine the lagged environmental data set X' and the effect data Y to construct a comprehensive data set D containing all monitoring indicators; the data set D contains the original environmental data, lagged environmental data, and effect data; add a timestamp and monitoring location identifier for each data type. Divide the data set D into a training set Dtrain and a test set Dtest in a ratio of 7:3

[0066] On the training set, the parameters of each anomaly detection algorithm are adjusted for each monitoring data type (water level, rainfall, dam deformation, seepage); grid search or Bayesian optimization methods are used to find the optimal parameter combination for each algorithm; and the F1 score (the harmonic mean of precision and recall) is used as the objective function for parameter optimization.

[0067] Evaluation indicators were calculated, including detection accuracy A: the number of correctly detected anomalies / (the number of correctly detected anomalies + the number of false alarms); false alarm rate B: the number of false alarms / the total number of detected anomalies; false negative rate C: the number of false negatives / the total number of actual anomalies; and robustness index D: the degree to which the algorithm performance remains stable after adding random noise, calculating the consistency ratio between the original result and the result after adding noise.

[0068] Construct an evaluation indicator matrix and construct a 4×5 evaluation indicator matrix E; the row index i∈{1, 2, 3, 4} corresponds to four monitoring data types: water level X1, rainfall X2, dam deformation Y1, and seepage Y2; the column index j∈{1, 2, 3, 4, 5} corresponds to five anomaly detection algorithms: mean method, median method, difference method, Hampel filter method and 3σ criterion.

[0069] Each matrix element E i,j Store four evaluation metrics {A, B, C, D}; for example, E 1,3 Represents the evaluation result of water level data X1 using the difference method; the four indicators in each element are stored in a structured form: {accuracy A, false alarm rate B, omission rate C, robustness D}.

[0070] S5, according to the evaluation index matrix, select the optimal anomaly detection algorithm for each monitoring data. The weight coefficient is determined by combining the analytic hierarchy process (AHP) and the Delphi method. The weight w of the detection accuracy A is A =0.4, reflecting the algorithm's basic ability to correctly identify anomalies; the weight of the false alarm rate B is w B =0.25, reflecting the impact of the algorithm's false alarms; the weight of the missed alarm rate C is w C =0.25, reflecting the risk cost of the algorithm missing real anomalies; the weight of the robustness index D is w D =0.1, reflecting the algorithm's ability to adapt to data fluctuations.

[0071] The weight can be adjusted according to the actual risk level of the project; for high-risk projects, the weight of the missed reporting rate C can be increased (such as w C =0.3~0.35); For monitoring systems with frequent false alarms, the weight of the false alarm rate B can be increased (such as w B =0.3); keep the sum of all weights to 1: w A +w B +w C +w D =1.

[0072] For each monitoring data type i and anomaly detection algorithm j, calculate the comprehensive score S i,j ,

[0073] S i,j =w A ×A i,j -w B ×B i,j -w C ×C i,j +w D ×D i,j , where i∈{1, 2, 3, 4} represents the type of monitoring data, j∈{1, 2, 3, 4, 5} represents the type of anomaly detection algorithm; evaluation index standardization processing: A i,j and D i,j It is a positive indicator with a value range of [0, 1], the larger the better; B i,j and C i,j It is a negative indicator with a value range of [0, 1]. The smaller the value, the better. Therefore, a negative sign is used in the formula.

[0074] For example, the evaluation result of the mean method applied to water level data is {A=0.85, B=0.12, C=0.08, D=0.75}; the comprehensive score S is calculated. 1,1 =0.4×0.85-0.25×0.12-0.25×0.08+0.1×0.75=0.365; repeat the above calculation for all algorithm-data type combinations to form a comprehensive scoring matrix.

[0075] Extract the maximum absolute value of the correlation coefficient of environmental data from the time-lag correlation coefficient matrix R|r max |; For water level data X1, take |r max,1 |=max(|r 1,1 |,|r 1,2 |), which is the maximum correlation coefficient between water level and dam deformation and seepage; for rainfall data X2, take |r max,2 |=max(|r 2,1 |,|r 2,2 |), which is the maximum correlation coefficient between rainfall and dam deformation and seepage.

[0076] Perform correlation weighted correction on the comprehensive scores of environmental data X1 and X2; S i,j '=S i,j ×(1+0.2×|r max |), i∈{1, 2}; the correction coefficient of 0.2 represents the weight of the influence of correlation on the score, which can be adjusted according to the specific conditions of the project (generally ranging from 0.1 to 0.3); the stronger the correlation, the more significant the correction, which reflects the importance of environmental data to the prediction of effect data.

[0077] For example, the maximum correlation coefficient between water level data X1 and effect data |r max,1 |=0.75|; The original comprehensive score S of the water level data using the mean method 1,1 =0.365; Corrected score S 1,1 '=0.365×(1+0.2×0.75)=0.42.

[0078] For the dam deformation data Y1 and seepage data Y2, the optimal algorithm is selected directly based on the original comprehensive score. Select the optimal algorithm for dam deformation data Y1; Select the optimal algorithm for the seepage data Y2.

[0079] For water level data X1 and rainfall data X2, the optimal algorithm is selected based on the revised comprehensive score;

[0080] Select the optimal algorithm for water level data X1; Select the optimal algorithm for rainfall data X2.

[0081] Further comparative verification is conducted on algorithms with similar scores (score difference < 0.05); candidate algorithms are executed on actual data subsets to confirm their performance stability; considering the algorithm execution efficiency, algorithms with lower computational complexity are given priority when performance is comparable.

[0082] S6, process outliers on the collected monitoring data according to the optimal anomaly detection algorithm; further compare and verify algorithms with similar scores (score difference < 0.05); execute candidate algorithms on actual data subsets to confirm their performance stability; consider algorithm execution efficiency and give priority to algorithms with lower computational complexity when performance is comparable; apply the corresponding optimal anomaly detection algorithm to the newly collected monitoring data; use A for water level data X1 opt (1) Algorithm detection anomaly; rainfall data X2 uses A opt (2) Algorithm detection anomaly; dam deformation data Y1 uses A opt (3) Algorithm detection anomaly; seepage data Y2 uses A opt(4) Algorithm for detecting anomalies. Calculate an anomaly score for each monitored data point to indicate the degree of anomaly at that point; set a basic threshold θ0 and an advanced threshold θ1 (θ1>θ0) to implement anomaly classification; data points with anomaly scores less than θ0 are marked as “normal”; data points with anomaly scores between θ0 and θ1 are marked as “mildly abnormal”; data points with anomaly scores greater than θ1 are marked as “severely abnormal”; and save the correspondence between the original data points and the anomaly labels.

[0083] For mildly abnormal data, a strategy of marking but retaining the original value is adopted; for severely abnormal data, a replacement strategy is adopted; different replacement methods are selected according to the data type: for environmental data (X1 and X2): use the predicted values ​​of the time series model (such as ARIMA) to replace; for effect data (Y1 and Y2): use the predicted values ​​of the multivariate regression model based on lagged environmental data to replace; the replaced data are marked as "corrected", while retaining the original value and the basis for correction.

[0084] like Figure 4 As shown in S7, the monitoring data after outlier processing are grouped according to season and working condition. Combined with the correlation coefficient matrix, the mean and standard deviation of each group of data are statistically calculated to obtain the mean ± 2σ as the first-level warning threshold and the mean ± 3σ as the second-level warning threshold.

[0085] First, the monitoring data after outlier processing were divided into seasons according to calendar dates; spring: monitoring data from March 1 to May 31; summer: monitoring data from June 1 to August 31; autumn: monitoring data from September 1 to November 30; winter: monitoring data from December 1 to February 28 / 29 of the following year; a season tag field was established to add the corresponding season identifier (1-4) to each monitoring record.

[0086] The operating status is comprehensively determined based on the water level data X1 and the rainfall data X2; normal water storage conditions (marked as 1): the water level is lower than the warning level and the water level change rate within 72 hours is less than the set threshold (such as 0.5 meters / day), and there is no heavy rainfall in the recent period; flood season precipitation conditions (marked as 2): the water level exceeds the warning level or the water level rise rate within 72 hours exceeds the set threshold, or the cumulative rainfall for 24 consecutive hours exceeds 50 mm; add operating condition mark fields to all monitoring data to ensure that they match the hydraulic status at the time.

[0087] By combining season marks (1-4) and operating condition marks (1-2), 8 season-operating condition combinations are formed; a combination type identifier (1-8) is added to each monitoring record, corresponding to eight combinations: Combination 1: spring-normal water storage (j=1, m=1); Combination 2: spring-flood season precipitation (j=1, m=2); Combination 3: summer-normal water storage (j=2, m=1); Combination 4: summer-flood season precipitation (j=2, m=2); Combination 5: autumn-normal water storage (j=3, m=1); Combination 6: autumn-flood season precipitation (j=3, m=2); Combination 7: winter-normal water storage (j=4, m=1); Combination 8: winter-flood season precipitation (j=4, m=2); For combinations that are not common at a specific time (such as winter-flood season precipitation), there may be fewer samples. In this case, the historical data of the past 3 to 5 years can be integrated to increase the sample size.

[0088] Group statistical calculation: For each monitoring data type (i) under each season-operating condition combination (j, m), calculate the following statistical parameters: mean Where x is the data value within the group, n is the number of samples within the group; standard deviation Skewness coefficient Measures the degree of asymmetry in data distribution. Handling insufficient sample size: Set a minimum sample size threshold, such as 30 data points. For combinations with insufficient sample size, use approximate seasonal-operating condition combinations to supplement. For example, if the "winter-flood season precipitation" sample is insufficient, "autumn-flood season precipitation" or data from the same period of the previous year can be used as supplementary data.

[0089] Statistical parameter verification: Use normality tests (such as the Shapiro-Wilk test) to verify data distribution characteristics; for data with obvious non-normal distribution (test p-value <0.05), mark its skewness characteristics to prepare for subsequent threshold correction; generate a statistical parameter report including the sample size, mean, standard deviation, skewness coefficient and normal distribution test results for each combination.

[0090] Extract the optimal lag time t for each environment-effect combination (p, q) from the time-lag correlation coefficient matrix R opt,p,q ; The corresponding relationship is: (p = 1, q = 1) represents the water level and dam deformation; (p = 1, q = 2) represents the water level and seepage; (p = 2, q = 1) represents the rainfall and dam deformation; (p = 2, q = 2) represents the rainfall and seepage. For each season-operating condition combination (j, m) data, according to the optimal lag time t opt,p,q Process environmental data; for example, if the optimal lag time between water level and seepage is 2 days, then shift the water level data series back by 2 days to align X1(t-2) with Y2(t).

[0091] For each season-condition combination (j, m), calculate the real-time correlation response coefficient between environmental data and effect data Among them, μ p and σ p The environmental data X in the current combination p The mean and standard deviation of q and σ q The effect data Y in the current combination q The mean and standard deviation of ; n is the number of valid paired samples in the current combination.

[0092] For each environmental data type (p), determine the effect data type (q) with the greatest correlation; for example, if the correlation coefficient between water level and dam deformation is greater than the correlation coefficient between water level and seepage, then the most relevant effect data of water level is dam deformation.

[0093] For effect data (i∈{3,4}): If the data is approximately normally distributed If the data is skewed (|s i,j,m |≥0.5): Warning upper limit correction coefficient: C i,j,m =1+0.1×s i,j,m ; Warning lower limit correction coefficient: C i,j,m =1-0.1×s i,j,m .

[0094] If there is a strong correlation with the effect data (|r max |>0.7): When Time: Warning upper limit Warning lower limit when Time: Warning upper limit Warning lower limit If the effect data are weakly correlated (|r max |≤0.7): C i,j,m =1.

[0095] Level 1 warning threshold (k=2): Lower limit: (lower limit); upper limit: (Upper limit).

[0096] Second level warning threshold (k=3): Lower limit: (lower limit); upper limit: (Upper limit).

[0097] For water level data (X1), the upper limit of the second-level warning shall not exceed the design flood level; for dam deformation data (Y1), the upper limit of the second-level warning shall not exceed the design allowable deformation value; if the calculated value exceeds the physical safety limit, the safety limit shall be used as the final threshold.

[0098] Construction of warning threshold matrix Θ: Construct a four-dimensional warning threshold matrix Θ; the matrix dimension is 4×4×2×4, corresponding to monitoring data type (4)×season (4)×operating condition (2)×threshold type (4); each element Θ i,j,m Contains four warning thresholds: For example, Θ 1,2,1 Represents the warning threshold set of water level data under normal water storage conditions in summer.

[0099] Convert the warning threshold matrix into a practical warning threshold table for engineering applications; the table includes: monitoring data type, season, working conditions, first-level warning lower limit, first-level warning upper limit, second-level warning lower limit, and second-level warning upper limit; add threshold effective time and next update time information.

[0100] At the seasonal or operating condition transition point, a threshold transition period of 7-10 days is set; during the transition period, the threshold gradually transitions from one season-operating condition combination to the next combination in a linear proportion, avoiding false alarms caused by sudden changes in the threshold at the transition point.

[0101] S8: Monitor and manage according to the first and second warning thresholds. Collect monitoring data at the frequency specified in S1; apply the optimal anomaly detection algorithm determined in S6 to process outliers; and determine the season-operating condition combination based on the current date and water level.

[0102] Extract the warning thresholds for each monitoring data type under the current season-operating condition combination from the warning threshold matrix Θ; compare the real-time monitoring data with the warning thresholds; and dynamically fine-tune the warning thresholds for environmental data (X1 and X2) based on the real-time correlation response coefficient φ_{p, q}.

[0103] Normal status: monitoring data is within the first-level warning threshold range; first-level warning: monitoring data exceeds the first-level warning threshold but does not exceed the second-level warning threshold; second-level warning: monitoring data exceeds the second-level warning threshold; comprehensive warning: when multiple related monitoring indicators trigger warnings at the same time, the warning level is raised.

[0104] Level 1 early warning response: Automatically generate level 1 early warning alarm information and display it through the system interface; notify on-duty monitoring personnel to pay attention and increase the frequency of data collection at relevant monitoring points; start a short-term prediction model to estimate the trend of indicators in the next 24-72 hours; record the details of the early warning event, including trigger time, monitoring point, warning value, threshold range, etc.

[0105] Level 2 early warning response: Automatically generate level 2 early warning alarm information and notify through multiple channels such as the system interface, mobile phone text messages, and emails; notify the project management person in charge and professional and technical personnel, and organize an emergency response team; increase the data collection frequency of all relevant monitoring points to the highest level; activate the emergency plan and prepare necessary protective measures; conduct on-site inspections of relevant areas to verify the reasons for abnormal monitoring data.

[0106] Warning upgrade and downgrade: Continuous warning conditions: If the monitoring data remains in the warning state for 24 consecutive hours, the warning level will be automatically upgraded; Cross-validation warning: If warnings appear at multiple related monitoring points at the same time, the warning level will be raised; Warning cancellation conditions: The monitoring data returns to the normal range for 48 consecutive hours and no abnormal phenomena are found; Warning downgrade process: The second-level warning must meet the 24-hour stability condition to be downgraded to the first-level warning, and the first-level warning must meet the 48-hour stability condition to be cancelled.

Claims

1. A real-time monitoring and management method for water conservancy and hydropower projects, characterized in that: include: S1, collecting monitoring data of water conservancy and hydropower projects, the monitoring data including environmental data X and effect data Y, the environmental data X including water level and rainfall; the effect data Y including dam deformation and seepage; S2, using the multiple linear regression method to construct the correlation coefficient matrix between environmental data X and effect data Y, including: The Z-score is used to process the environmental data X and the effect data Y respectively, where the environmental data X = {X1, X2}, X1 represents the water level data, X2 represents the rainfall data; the effect data Y = {Y1, Y2}, Y1 represents the dam deformation data, Y2 represents the seepage data; Construct the time lag models Y1(t) and Y2(t) of the dam deformation data Y1 and seepage data Y2 respectively; According to the time lag model Y1(t) and Y2(t), the Pearson correlation coefficient r of each environmental data X = {X1, X2} and effect data Y = {Y1, Y2} at different lag times was calculated by the sliding time window method. j (ti),Y l (t)),i∈[0,k],j∈{1,2},l∈{1,2}, k is the maximum lag order; i and j represent positive integers; For each environmental data X j and effect data Y l Combination, from k+1 Pearson correlation coefficients r(X j (ti),Y l (t)) select the correlation coefficient r with the largest absolute value max and its corresponding optimal lag time t opt , forming a 2×2 time-lag correlation coefficient matrix R; The time lag model Y1(t) and Y2(t) is expressed as: Where t represents the current time point; k is the maximum lag order; α1, α2 are the constant terms of each model; β 1i ,β 2i are the regression coefficients of water level data X1 at different lag times in the two effect data models; γ 1i ,γ 2i are the regression coefficients of rainfall data X2 at different lag times in the two effect data models; ε1 and ε2 are the random error terms of each model; S3, building an anomaly detection algorithm library, wherein the anomaly detection algorithm library includes the mean method, the median method, the difference method, the Hampel filter method and the 3σ criterion; S4, based on the correlation coefficient matrix and the anomaly detection algorithm library, construct an evaluation indicator matrix, including: According to the optimal lag time t in the time-lag correlation coefficient matrix R opt , perform lag processing on the environmental data X to form the lag environmental data set X' that has the greatest correlation with the effect data Y; Construct a data set based on the environmental data set X' and the effect data Y, and divide the data set into a training set and a test set; Use each anomaly detection algorithm in the anomaly detection algorithm library to perform anomaly detection on the data in the training set; The test set is used to calculate the evaluation indicators of the detection results of each anomaly detection algorithm. The evaluation indicators include detection accuracy A, false alarm rate B, missed alarm rate C and robustness index D; Construct a 4×5 evaluation index matrix E, the matrix element E i,j Contains anomaly detection evaluation results; S5, based on the evaluation index matrix, selects the optimal anomaly detection algorithm for each monitoring data, including: Set the weight coefficient w of each evaluation indicator in the evaluation indicator matrix A 、w B 、w C 、w D ; For each monitoring data, calculate the comprehensive score S of each anomaly detection algorithm i,j : S i,j =w A ×A i,j -w B ×B i,j -w C ×C i,j +w D ×D i,j ; S6, performing outlier processing on the collected monitoring data according to the optimal anomaly detection algorithm; S7, after outlier processing, the monitoring data are grouped by season and working condition, and the mean and standard deviation of each group of data are statistically calculated by combining the correlation coefficient matrix. The mean ± 2σ is obtained as the first-level warning threshold and the mean ± 3σ is obtained as the second-level warning threshold; where σ is the standard deviation; S8, monitor and manage according to the first-level warning threshold and the second-level warning threshold.

2. The real-time monitoring and management method for water conservancy and hydropower projects according to claim 1 is characterized in that: S5, selecting the optimal anomaly detection algorithm for each monitoring data, also includes: For environmental data X1 and X2, add the correlation coefficient correction term when calculating the comprehensive score: S i,j '=S i,j ×(1+0.2×|r max |) Among them, |r max | is the absolute value of the maximum correlation coefficient corresponding to the environmental data in the time-lagged correlation coefficient matrix R; For each type of monitoring data, the anomaly detection algorithm with the highest comprehensive score is selected as the optimal algorithm for the corresponding data type: Among them, A opt (i) is the optimal anomaly detection algorithm corresponding to the monitoring data type i.

3. The real-time monitoring and management method for water conservancy and hydropower projects according to claim 2 is characterized in that: S7, the mean ± 2σ is obtained as the first-level warning threshold and the mean ± 3σ is obtained as the second-level warning threshold, including: The monitoring data after outlier processing are divided into four groups according to season: Spring Festival, summer, autumn and winter; According to the water level status, the monitoring data after abnormal value processing are grouped according to the working conditions, which include normal water storage conditions and flood season precipitation conditions; According to each season and working condition combination, a season-working condition combination is formed, including spring-normal water storage, spring-flood season precipitation, summer-normal water storage, summer-flood season precipitation, autumn-normal water storage, autumn-flood season precipitation, winter-normal water storage and winter-flood season precipitation; For each season-operating condition combination, the statistical parameters of each monitoring data type are calculated respectively, and the statistical parameters include mean μ, standard deviation σ and skewness coefficient s; Based on the data of each season-operating condition combination, the optimal lag time t in the time-lag correlation coefficient matrix R is used. opt Calculate the real-time correlation response coefficient between environmental data X and effect data Y in each set of data According to the statistical parameters of each season-operating condition combination, combined with the calculated real-time correlation response coefficient Calculate the warning threshold: For effect data Y1 and Y2, the warning threshold calculation formula is: μ±kσ, where k is the warning level coefficient; For environmental data X1 and X2, when |r max When |>0.7, the calculation formula for the warning threshold is: Where: δ is the correlation adjustment coefficient, which is 0.2; When | r max When |≤0.7, the warning threshold calculation formula is: μ±kσ, where k is the warning level coefficient; Generate the first-level warning threshold and the second-level warning threshold for each group of data; The warning thresholds under all season-operating condition combinations are integrated into the warning threshold matrix Θ, and the matrix elements Θ i,j,m represents the warning threshold range of monitoring data type i in season j and working condition m, in: It represents the first-level warning lower limit of monitoring data type i in season j and working condition m; It represents the upper limit of the first-level warning for monitoring data type i in season j and working condition m; It represents the lower limit of the second-level warning for monitoring data type i in season j and working condition m; Indicates the upper limit of the second-level warning for monitoring data type i in season j and operating condition m.

4. The real-time monitoring and management method for water conservancy and hydropower projects according to claim 3 is characterized by: Real-time correlation response coefficient The calculation formula is: Where: p and q are positive integers; t opt,p,q is the optimal lag time of the corresponding combination in the time-lag correlation coefficient matrix R; μ p and σ p They are environmental data X p Mean and standard deviation in the current season-operating condition combination; μ q and σ q The effect data Y q The mean and standard deviation in the current season-operating condition combination; n is the number of samples in the current season-operating condition combination; p∈{1, 2} represents the environmental data type; q∈{1, 2} represents the effect data type.

5. The real-time monitoring and management method for water conservancy and hydropower projects according to claim 4 is characterized in that: Generate the first-level warning threshold and second-level warning threshold for each group of data using the following formula: Level 1 warning threshold: Level 2 warning threshold: Among them, μ i,j,m is the mean value of monitoring data type i in season j and operating condition m; σ i,j,m is the standard deviation of monitoring data type i in season j and operating condition m; C i,j,m is the correction factor.

Images