A method for constructing a segmented statistical model for safety monitoring data of relatively safe dams

Abstract

The invention discloses a method for constructing a segmented statistical model of safety monitoring data of a partial safety type dam, comprising: S1, based on analyzing the characteristics of dam safety monitoring data, considering the relationship between dam safety monitoring data and dam safety, establishing a relative importance measurement of dam safety monitoring data; S2, after normalizing the relative importance measurement of dam safety monitoring data, using the weighted least squares principle for reference, taking it as a weight factor, and constructing a partial safety type statistical model; S3, using the Pearson correlation coefficient to measure the accuracy of the partial safety statistical model, and dynamically identifying the structural state changes of the dam itself through an online segmentation algorithm, performing online segmentation on the dam safety monitoring data sequence, and discovering and predicting the deformation characteristics of the dam in real time; the invention can discover and predict the temporal changes of the structural state of the dam itself under the influence of external factors, and retain the volatility in the actual monitoring data, the extreme values ​​in the envelope monitoring data, and the ability and characteristics of the analysis results being biased towards safety.

Description

Technical Field

[0001] The invention relates to the technical field of dam operation safety monitoring and management, in particular to a method for constructing a segmented statistical model of safety monitoring data of a relatively safe dam. Background Art

[0002] Dams are one of the core of water conservancy project construction and an important project to improve water resource regulation and allocation. In order to ensure the safe operation of dams, relevant deformation monitoring theories and methods have attracted much attention. Among them, various dam deformation prediction models reflect the safe operation status of dams by exploring the internal relationship between external factors and deformation measurements. They have become an important means of studying dam deformation and provide important technical support for safety management during the safe operation of dams.

[0003] For dam safety monitoring data, statistical models are widely used because of their fast execution speed and high correlation between input variables and corresponding variables. Constructing statistical models and then evaluating, analyzing, diagnosing, predicting and warning the dam safety status is one of the important ways to maximize the value of dam safety monitoring data. With the emergence of various new theories, new ideas and new methods, the accuracy of dam deformation monitoring models has been continuously improved. The combination of intelligent algorithms and statistical models can have a better prediction effect on nonlinear time series such as dam deformation monitoring series.

[0004] However, since the dam's own structure is affected by external factors and will produce complex dynamic changes, the accuracy of the monitoring model will weaken over time. At the same time, due to the good mathematical properties of the square loss function, a series of methods such as the least squares regression model are used to establish a dam safety monitoring model, but this method can only provide data information on the average condition and cannot solve the extreme value changes of the data itself. Summary of the invention

[0005] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a method for constructing a segmented statistical model of safety monitoring data of a relatively safe dam, aiming to consider the relative importance of the characteristic values ​​of the dam safety monitoring data, reflect the importance of the characteristic values ​​of the data itself in establishing the model, and online identify the changes in the structural properties of the dam itself, segment the dam safety monitoring data, and then construct and estimate the segmented statistical model of safety monitoring data of a relatively safe dam.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for constructing a segmented statistical model of partial safety dam safety monitoring data, which comprises the following steps:

[0007] S1. Based on the analysis of the characteristics of dam safety monitoring data, the relationship between dam safety monitoring data and dam safety is considered, and a relative importance measurement of dam safety monitoring data is established to measure the dam safety status mapped by the monitoring data;

[0008] S2. After normalizing the relative importance measurement of dam safety monitoring data, the weighted least squares method is used as the weight factor to construct a partial safety statistical model;

[0009] S3. The Pearson correlation coefficient is used to measure the accuracy of the partial safety statistical model, and the online segmentation algorithm is used to dynamically identify the changes in the dam's own structural properties, perform online segmentation on the dam safety monitoring data series, and discover and predict the dam's deformation characteristics in real time.

[0010] Furthermore, in step S1, a relative importance metric of dam safety monitoring data is established, and the construction method thereof is as follows:

[0011] The dam safety monitoring data is regarded as a set of time series, denoted as Y n ,n=1,2,…,N, within a certain time range, the monitoring data revolves around its mean μ n The relative importance measurement index MI(Y n ) is expressed as:

[0012] MI(Y n )=|Y n -μ n | p ,n=1,2,…,N (1)

[0013] In the formula, MI(Y n ) is the monitoring data Y n The corresponding relative importance, Y n is the nth monitoring data, μ n is the monitoring data Y n The corresponding mean, N is the total number of monitoring data, and p is the relative importance index, which is 1 or 2.

[0014] Furthermore, the step S2 is specifically as follows:

[0015] For the statistical model of dam safety monitoring data, a corresponding partial safety statistical model is established based on the statistical model of deformation monitoring of concrete gravity dams;

[0016] The statistical model of deformation monitoring data of concrete gravity dam is expressed as:

[0017] δ=δ H +δ T +δ θ +ε (2)

[0018] In the formula, δ represents the deformation of the concrete dam, δ H represents the water pressure component in the deformation, δ Trepresents the temperature component in deformation, δ θ represents the time-dependent component in deformation, and ε represents the deformation error;

[0019] Considering only the effect of reservoir water pressure on the deformation of concrete gravity dam, then:

[0020]

[0021] In the formula, a i represents the coefficient to be estimated, H represents the water depth upstream of the reservoir;

[0022] Considering only the effect of periodic temperature on the deformation of concrete gravity dam, then:

[0023]

[0024] In the formula, t is the number of days corresponding to the monitoring time, b 1i and b 2i is the coefficient to be estimated, m is the number of deformation cycles considered per year;

[0025] Considering linear and logarithmic time-dependent components, namely:

[0026] δ θ =c 1 lnθ+c 2 θ (5)

[0027] In the formula, θ is in years, which is the number of years from the monitoring time to the start time of monitoring, c 1 and c 2 is the coefficient to be estimated;

[0028] By expressing formula (2) in the form of a time series and combining it with formula (1), the statistical model of safety monitoring data of partial safety dams can be expressed as:

[0029]

[0030] In the formula, δ n is the deformation of the concrete dam at time n, δ Hn represents the water pressure component in the deformation at time n, δ Tn represents the temperature component of the deformation at time n, δ θn represents the time-dependent component of deformation at time n, ε n represents the deformation error at time n; N is the total number of concrete dam deformation time series; μ n is the monitoring data Y n The corresponding mean, p is the relative importance index, which takes 1 or 2;

[0031] For formula (5), in order to realize the parameter estimation of the model, MI(δ n) is normalized and recorded as ω n , which is used as the weight in the weighted least squares method, namely:

[0032]

[0033] Therefore, the weighted least squares estimation model corresponding to formula (6) can be expressed as:

[0034]

[0035] Where Δ is the monitoring data vector, X is the independent variable vector composed of water pressure, temperature and time, β is the parameter vector composed of the parameters to be estimated corresponding to the water pressure, temperature and time components, E is the error vector, and W is ω n ,n=1,2,…,N,Q W (β) is the cost function of the least squares estimate;

[0036] If X is a full rank matrix and W is a positive definite matrix, the estimation result of β in equation (8) is:

[0037] β=(X′WX) -1 X′WΔ (9)

[0038] Furthermore, the step S3 is specifically as follows:

[0039] Suppose the sequence u=(u 1 ,u 2 ,…,u m ), and u~N(0,σ), where m is the total number of sequences, N(0,σ) represents a normal distribution with an expected value of 0 and a standard deviation of σ; let

[0040] F(1,m)=T(0,m) (10)

[0041] represents the likelihood function that the sequence u is divided into only one segment; and

[0042] F(2,m)=T(1,m-1) (11)

[0043] represents the likelihood function of dividing sequence u into two segments from m, where

[0044] T(h,m)=(mh)lg(S *2 ) (12)

[0045] Where S *2 Represents the sequence u h+1 ,u h+2 ,…,u m If the sample variance of

[0046]

[0047] It means that it is more reasonable to divide u into two segments from m, where δ is the threshold;

[0048] In the above formula, F obeys F α (m,m-1); when the significance level α is given, F α The value of (m,m-1);

[0049] The online segmentation process of the statistical model for safety monitoring data of partial safety dams is as follows:

[0050] (1) Given a dam safety monitoring sequence δ = (δ 1 ,δ 2 ,…δ m ), as the monitoring sequence collects data at the current moment, a statistical model for safety monitoring data of the dam with a partial safety factor is established;

[0051] (2) Calculate the Pearson correlation coefficient of the statistical model of the partial safety dam safety monitoring data at the current moment and use it as the time series u = (u 1 ,u 2 ,…,u m );

[0052] (3) Calculate the sample variance of sequence u respectively Where t=1,2,…,m-1, calculate F(1,m) and F(2,m);

[0053] (4) Given a significance level, the threshold δ can be determined as α The value of (m,m-1) is to eliminate the data measurement error at a certain moment. t +F t+1 +F t+2 When >3δ, segmentation is performed from this point, otherwise return to step (1).

[0054] Beneficial effects of the present invention: Compared with the existing prediction and analysis models, the present invention, based on the analysis of the characteristics of dam safety monitoring data, considers the relationship between dam safety monitoring data and dam safety, measures the relative importance of dam monitoring data, draws on the relevant ideas and technologies of weighted least squares method, and online segmentation recursive algorithm, constructs and estimates a segmented statistical analysis model of safety monitoring data of dam with a bias towards safety, which can discover and predict the temporal changes of the dam's own structural state under the influence of external factors, retains the volatility in actual monitoring data, envelopes the extreme values ​​in monitoring data, and the ability and characteristics of analysis results biased towards safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flowchart of a method for constructing a segmented statistical model for safety monitoring data of a relatively safe dam;

[0056] Figure 2 This is the vertical displacement monitoring process line of a dam;

[0057] Figure 3 Monitoring process lines for upstream water levels;

[0058] Figure 4 Statistical analysis results of different models when p = 2;

[0059] Figure 5 Statistical analysis results of different models when p = 1;

[0060] Figure 6 is the weight of monitoring data when p=1 and p=2;

[0061] Figure 7 Statistical analysis results of different models when p = -1;

[0062] Figure 8 is the weight of monitoring data when p=-1;

[0063] Fig. 9 The prediction results of different models. DETAILED DESCRIPTION

[0064] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0065] Embodiment 1: A method for constructing a segmented statistical model of safety monitoring data of a relatively safe dam, comprising the following steps:

[0066] Step 1: Based on the analysis of the characteristics of dam safety monitoring data, the relationship between dam safety monitoring data and dam safety is considered, and a relative importance measurement of dam safety monitoring data is established to measure the dam safety status mapped by the monitoring data;

[0067] The principle of building safety monitoring is that buildings have an inherent response to changes in loads and environmental quantities. For dams, deformation properties must change under loads such as water pressure, temperature and deadweight. The eigenvalue statistics method is an important method in the analysis of dam safety monitoring data. This method mainly analyzes the characteristic values ​​of the dam safety monitoring data, such as the (cycle) mean, (cycle) maximum, (cycle) minimum, (cycle) amplitude, and so on, and compares them with the design value, test value, model prediction value under the same conditions, and the range of changes over the years. When the monitoring data exceeds the warning value, the cause and the impact on the dam safety should be analyzed. It can be seen that compared with the sample population of the dam safety monitoring data, the characteristic value of the dam safety monitoring data is more important in reflecting the dam safety.

[0068] Usually, the dam safety monitoring data is taken as a set of time series, denoted as Y n,n=1,2,…,N, within a certain time range, the monitoring data revolves around its mean μ n To reflect the importance of each monitoring data, this paper adopts the concept of relative importance. The relative importance measurement index of monitoring data constructed is MI (Y n ) can be expressed as:

[0069] MI(Y n )=|Y n -μ n | p ,n=1,2,…,N (1)

[0070] In the formula, MI(Y n ) is the monitoring data Y n The corresponding relative importance, Y n is the nth monitoring data, μ n is the monitoring data Y n The corresponding mean, N is the total number of monitoring data, and p is the relative importance index, usually 1 or 2.

[0071] Formula (1) shows that the greater the deviation or distance of the monitoring data from its mean, that is, the greater the volatility of the monitoring data, the greater the relative importance of the monitoring data. Since Formula (1) is constructed based on the degree to which the monitoring data reflects the safety of the dam, Formula (1) also means that the greater the volatility of the monitoring data and the more it deviates from its mean, the more it can map out or distinguish the safety status of the dam, which is consistent with the current concept of dam safety monitoring.

[0072] For the mean μ n ,In order to consider the dynamic characteristics or structural change characteristics of the dam safety ,status, this paper proposes to use the sliding average method to ,determine the length of the sliding window which can be determined according to the main ,period of monitoring data.

[0073] Step 2: After normalizing the relative importance measurement of dam safety monitoring data, the weighted least squares method is used as a weight factor to construct a partial safety statistical model;

[0074] Aiming at the statistical model of dam safety monitoring data, taking the statistical model of deformation monitoring of concrete gravity dam as an example, a corresponding partial safety statistical model is established.

[0075] The conventional statistical model of deformation monitoring data of concrete gravity dam can be expressed as:

[0076] δ=δ H +δ T +δ θ +ε (2)

[0077] In the formula, δ represents the deformation of the concrete dam, δH represents the water pressure component in the deformation, δ T represents the temperature component in deformation, δ θ represents the time-dependent component of deformation, and ε represents the deformation error.

[0078] Considering only the effect of reservoir water pressure on the deformation of concrete gravity dam, then:

[0079]

[0080] In the formula, a i represents the coefficient to be estimated, H represents the water depth upstream of the reservoir;

[0081] Considering only the effect of periodic temperature on the deformation of concrete gravity dam, then:

[0082]

[0083] In the formula, t is the number of days corresponding to the monitoring time, b 1i and b 2i is the coefficient to be estimated, and m is the number of deformation cycles considered per year.

[0084] Considering linear and logarithmic time-dependent components, namely:

[0085] δ θ =c 1 lnθ+c 2 θ (5)

[0086] In the formula, θ is usually expressed in years, which is the number of years from the monitoring time to the start time of monitoring, and c 1 and c 2 is the coefficient to be estimated.

[0087] It should be noted that equations (2)-(4) are only the statistical models used in this paper. Generally, the changing patterns and influencing factors of dam monitoring data are relatively complex, and the corresponding statistical model needs to be determined based on the characteristics of the monitoring data and other empirical knowledge.

[0088] By expressing formula (2) in the form of a time series and combining it with formula (1), the statistical model of safety monitoring data of partial safety dams can be expressed as:

[0089]

[0090] In the formula, δ n is the deformation of the concrete dam at time n, δ Hn represents the water pressure component in the deformation at time n, δ Tn represents the temperature component of the deformation at time n, δ θn represents the time-dependent component of deformation at time n, ε nrepresents the deformation error at time n; N is the total number of concrete dam deformation time series; μ n is the monitoring data Y n The corresponding mean, p is the relative importance index, which takes 1 or 2;

[0091] For formula (5), in order to realize the parameter estimation of the model, this paper will draw on the relevant ideas and techniques of weighted least squares method. n ) is normalized and recorded as ω n , which is used as the weight in the weighted least squares method, namely:

[0092]

[0093] Therefore, the weighted least squares estimation model corresponding to formula (6) can be expressed as:

[0094]

[0095] Where Δ is the monitoring data vector, X is the independent variable vector composed of water pressure, temperature and time, β is the parameter vector composed of the parameters to be estimated corresponding to the water pressure, temperature and time components, E is the error vector, and W is ω n ,n=1,2,…,N,Q W (β) is the cost function of the least squares estimate.

[0096] If X is a full rank matrix and W is a positive definite matrix, the estimation result of β in equation (8) is:

[0097] β=(X′WX) -1 X′WΔ (9)

[0098] Step 3: Use the Pearson correlation coefficient to measure the accuracy of the partial safety statistical model, and use the online segmentation algorithm to dynamically identify the structural changes of the dam itself, perform online segmentation on the dam safety monitoring data sequence, and discover and predict the dam deformation characteristics in real time;

[0099] For the segmentation of time series, the segmentation algorithm is required to be able to identify the moment when the system model changes, which is called change point monitoring in statistics. In order to dynamically identify the changes in the structural properties of the dam itself, an online segmentation recursive algorithm is used to segment the dam safety monitoring data series.

[0100] Suppose the sequence u=(u 1 ,u 2 ,…,u m ), and u~N(0,σ), where m is the total number of sequences, N(0,σ) represents a normal distribution with an expected value of 0 and a standard deviation of σ; let

[0101] F(1,m)=T(0,m) (10)

[0102] represents the likelihood function that the sequence u is divided into only one segment; and

[0103] F(2,m)=T(1,m-1) (11)

[0104] represents the likelihood function of dividing sequence u into two segments from m, where

[0105] T(h,m)=(mh)lg(S *2 ) (12)

[0106] Where S *2 Represents the sequence u h+1 ,u h+2 ,…,u m If the sample variance of

[0107]

[0108] It means that it is more reasonable to divide u into two segments from m, where δ is the threshold.

[0109] In the above formula, F obeys F α (m,m-1). When the significance level α is given, F α The value of (m,m-1).

[0110] The online segmentation process of the statistical model for safety monitoring data of partial safety dams is as follows:

[0111] (1) Given a dam safety monitoring sequence δ = (δ 1 ,δ 2 ,…δ m ), as the monitoring sequence collects data at the current moment, a statistical model for safety monitoring data of the relatively safe dam is established.

[0112] (2) Calculate the Pearson correlation coefficient of the statistical model of the partial safety dam safety monitoring data at the current moment and use it as the time series u = (u 1 ,u 2 ,…,u m ).

[0113] (3) Calculate the sample variance of sequence u respectively Where t=1,2,…,m-1, calculate F(1,m) and F(2,m).

[0114] (4) Given a significance level, the threshold δ can be determined as α The value of (m,m-1) is to eliminate the data measurement error at a certain moment. t +F t+1 +Ft+2 When >3δ, segmentation is performed from this point, otherwise return to step (1).

[0115] Example 2: See Figure 1-Figure 7

[0116] Figure 1 The flowchart of the construction of a segmented statistical model of dam safety monitoring data for a partial safety type of the present invention is as follows: a hydropower station is located on a tributary of the Minjiang River in Fujian Province, China, and its main water retaining structure is a concrete wide-slot gravity dam. The main dam is 412m long, with a maximum dam height of 71m and a dam top elevation of 384.5m. It is divided into 22 dam sections. The dam top horizontal displacement monitoring sequence and the corresponding upstream water level monitoring sequence of a dam section are selected, such as Figure 2 and Figure 3 shown.

[0117] The calculation results of the conventional statistical model and the safe statistical model (p = 1, p = 2) are shown in Figure 4 , Figure 5 The corresponding weights are shown in Figure 6 . Among them, the conventional statistical model is equivalent to the result of the partial safety statistical model when p=0.

[0118] Depend on Figure 4 , Figure 5 It can be seen that when the significance level α=0.05 is considered, a partial safety segmented statistical model is established. The conventional statistical model treats the measured value of each monitoring data with equal weight, so its fitting value is closer to the mean of the monitoring data. The partial safety segmented statistical model, however, gives greater weight to the measured values ​​that deviate far from the mean, so its fitting value envelops the fitting value of the conventional statistical model, and at the same time, the online segmentation recursive algorithm is used for segmentation. Therefore, for the maximum value in the monitoring data, the fitting value of the partial safety statistical model is greater than the result of the conventional statistical model, and for the minimum value in the monitoring data, the fitting value of the partial safety statistical model is smaller than the result of the conventional statistical model. At the same time, when p=1 and p=2, the displacement sequence is divided into four ends and modeled separately, and the segment positions are basically close, indicating that in the segmented part, the data characteristics of the dam displacement monitoring sequence change. The use of conventional statistical models for modeling will lead to deviations in model accuracy over time. After segmentation, the sequence is modeled separately, which can integrate the information contained in the changing monitoring data sequence into the statistical model, making the statistical model more practical for changes in the comprehensive properties of the dam.

[0119] From the perspective of data volatility, the sample standard deviation of the actual monitoring data is 0.972mm, the sample standard deviation of the conventional statistical model fitting value is 0.973mm, and the sample standard deviation of the safety-oriented statistical model fitting value is 1.05mm (p=1) and 1.05mm (p=2). It can be seen that the calculation results of the safety-oriented statistical model have greater volatility, more reasonably utilize the change characteristics of the actual monitoring data, and can more completely describe the extreme value changes of the data series.

[0120] On the other hand, let p = -1, the calculation results are shown in Figure 7 , the weight of monitoring data can be found in Figure 8 .

[0121] Depend on Figure 6 and Figure 7 It can be seen that when the p value is negative, it means that the greater the deviation of the measured value from the mean, the smaller its weight, so its fitting value is closer to the mean of the monitoring data. This idea corresponds to the conventional weighted least squares model. This kind of idea is to regard the measured value that deviates more from the mean as a possible outlier or gross error point. From the perspective of volatility, the sample standard deviation corresponding to the fitted value of the model is only 0.714mm, that is, the model greatly weakens the volatility of the actual monitoring data. It is believed that the conventional weighted least squares model is suitable for the preprocessing stage of safety monitoring data to realize the identification or elimination of outliers; while the partial safety statistical model is suitable for the analysis and utilization stage of safety monitoring data, and can realize the prediction, forecast and diagnosis of the safety status of the dam in a partial safety manner.

[0122] right Fig. 9 It can be seen that when p = 2, both models can predict the vertical displacement changes of the dam well, but in terms of the degree of fluctuation, the sample standard deviation of the measured data is 0.757 mm, the sample standard deviation of the predicted value of the conventional statistical model is 0.732 mm, and the sample standard deviation of the predicted value of the partial safety statistical model is 0.774 mm. It can be seen that the results of the partial safety statistical model can avoid the characteristics of the square loss function and more fully reflect the volatility of the actual monitoring data.

[0123] According to the above examples, the method of the present invention can reflect the subjective tendency of dam safety monitoring data modeling method. Based on different cognitions of dam safety monitoring data, different monitoring data (relative) importance measurement indicators can be constructed, and different types of analysis models such as time series and machine learning can also be established. Continuing to study the characterization technology of safety-oriented tendencies, expanding the modeling types of monitoring data, and conducting systematic tests in actual projects are the main directions and contents of future research.

[0124] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for constructing a segmented statistical model for safety monitoring data of a relatively safe dam, characterized by: It includes the following steps: S1. Based on the analysis of the characteristics of dam safety monitoring data, the relationship between dam safety monitoring data and dam safety is considered, and a relative importance measurement of dam safety monitoring data is established to measure the dam safety status mapped by the monitoring data; S2. After normalizing the relative importance measurement of dam safety monitoring data, the weighted least squares method is used as the weight factor to construct a partial safety statistical model; S3. The Pearson correlation coefficient is used to measure the accuracy of the partial safety statistical model, and the online segmentation algorithm is used to dynamically identify the structural changes of the dam itself, and the dam safety monitoring data series is segmented online to discover and predict the deformation characteristics of the dam in real time; In step S1, a relative importance metric of dam safety monitoring data is established, and the construction method is as follows: The dam safety monitoring data is regarded as a set of time series, recorded as , within a certain time range, the monitoring data is around its mean Fluctuation changes, constructing relative importance metrics of monitoring data It is expressed as: （1）； In the formula, To monitor data The corresponding relative importance, For the Monitoring data, To monitor data The corresponding mean, is the total number of monitoring data, is the relative importance index, taking the value of 1 or 2; The step S2 is specifically as follows: For the statistical model of dam safety monitoring data, a corresponding partial safety statistical model is established based on the statistical model of deformation monitoring of concrete gravity dams; The statistical model of deformation monitoring data of concrete gravity dam is expressed as: （2）； In the formula, represents the deformation of the concrete dam, represents the water pressure component in the deformation, represents the temperature component in the deformation, represents the time-dependent component of deformation, Indicates deformation error; Considering only the effect of reservoir water pressure on the deformation of concrete gravity dam, then: （3）； In the formula, represents the coefficient to be estimated, Indicates the water depth upstream of the reservoir; Considering only the effect of periodic temperature on the deformation of concrete gravity dam, then: （4）； In the formula, is the number of days corresponding to the monitoring time, and is the coefficient to be estimated, is the number of deformation cycles considered per year; Considering linear and logarithmic time-dependent components, namely: （5）； In the formula, The unit is years, which is the number of years from the monitoring time to the start time of monitoring. and is the coefficient to be estimated; By expressing formula (2) in the form of a time series and combining it with formula (1), the statistical model of safety monitoring data of partial safety dams can be expressed as: （6）； In the formula, for The deformation of the concrete dam at the moment, express The water pressure component in the momentary deformation, express The temperature component of the momentary deformation, express The time-dependent component of the moment deformation, express Moment deformation error; is the total number of concrete dam deformation time series; To monitor data The corresponding mean, is the relative importance index, taking the value of 1 or 2; for The relative importance of the corresponding For formula (5), in order to realize the parameter estimation of the model, After normalization, it is recorded as , which is used as the weight in the weighted least squares method, namely: （7）； Therefore, the weighted least squares estimation model corresponding to equation (6) can be expressed as: ； In the formula, is the monitoring data vector, is the independent variable vector consisting of water pressure, temperature and time. is the parameter vector composed of the parameters to be estimated corresponding to the water pressure, temperature and time-dependent components, is the error vector, for The weight matrix formed by is the cost function estimated by least squares; like For a full rank matrix, is a positive definite matrix, then in formula (8) The estimated results are: （9）。 2. The method for constructing a segmented statistical model for safety monitoring data of a relatively safe dam according to claim 1, characterized in that: The step S3 is specifically as follows: With sequence ,and ,in is the total number of sequences, The expected value is 0 and the standard deviation is Normal distribution of （10）； Representation sequence The likelihood function is divided into only one segment; （11）； Representation sequence from The likelihood function is divided into two segments, where （12）； Where: Representation sequence If the sample variance of （13）； express from It is more reasonable to divide the punishment into two sections, is the threshold value; In the above formula, obey ; When the significance level is given When The value of The online segmentation process of the statistical model for safety monitoring data of partial safety dams is as follows: (1) Given a dam safety monitoring sequence , as the monitoring sequence collects data at the current moment, a statistical model of partial safety type dam safety monitoring data is established; (2) Calculate the Pearson correlation coefficient of the statistical model of the current partial safety dam safety monitoring data and use it as a time series that changes over time. ; (3) Calculate the sequences separately The sample variance of , ,in ,calculate and ; (4) Given a significance level, the threshold can be determined The value of is to eliminate the data measurement error at a certain moment. If , segment from the position at that moment, otherwise return to step (1).

Images