Structure monitoring data fusion probability abnormity early warning method based on streaming over-threshold and evidence reasoning

Abstract

The invention discloses a structure monitoring data fusion probability abnormity early warning method based on streaming super-threshold and evidence reasoning. The method comprises the following steps: establishing a historical data set of each sensor as sample data; performing temperature effect separation on the sample data, and extracting response components after temperature effect removal as an initial data set of anomaly detection; determining an initial threshold value; constructing a super-threshold data set; fitting the super-threshold data set, and obtaining posterior probability distribution of shape parameters and scale parameters through Bayesian inference; calculating probability distribution of the early warning trigger value; calculating an anomaly certainty degree index; dynamically updating the probability distribution of the early warning trigger value through a streaming threshold exceeding method; and performing synthesis through an evidence fusion method to obtain a fusion anomaly certainty degree index. According to the method, the fusion anomaly certainty degree can be formed, conflicts between indexes can be better processed, early warning misinformation caused by faults of a single sensor is avoided, and certain data support is provided for maintenance of the structure.

Description

Technical Field

[0001] The present invention belongs to the technical field of structural health monitoring data, and in particular relates to a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold value and evidence reasoning. Background Art

[0002] The main function of structural health monitoring (SHM) is to monitor the state of the structure in real time or periodically through a variety of sensing methods. With the development of intelligent technology and the widespread application of monitoring and sensing methods, structural health monitoring data has become an important basis for structural status assessment and maintenance management of large infrastructure, and structural health monitoring systems have been widely used. Taking bridge structural health monitoring as an example, effective use of monitoring data to conduct abnormal early warning of bridge response is of great significance for maintaining safe operation and timely maintenance of bridges.

[0003] At present, the abnormal warning of bridge response usually adopts setting a certain threshold and issuing an alarm for the response exceeding the threshold. The methods of setting thresholds are mainly divided into methods based on physical material properties and data-driven methods according to the principle. Early threshold setting relied on the analysis of the ultimate state of structural bearing capacity or the ultimate state of normal use, and often used material strength or design live load standard value as the warning judgment threshold. However, the threshold set by this method essentially reflects the ultimate state of the structure, which is generally much higher than the normal monitoring data level. In actual applications, even if an abnormal condition occurs in the bridge structure, it is difficult to trigger the warning.

[0004] On this basis, some scholars proposed to construct thresholds based on statistical analysis of monitoring data from a statistical perspective, and then determine the warning line. Under normal operation, the probability of the monitoring value reaching the set warning line is extremely small. Once the data touches the warning line, it can be inferred that the structure is likely to be abnormal. Extreme value theory (EVT) in the field of statistics is a technical model used to describe and study extreme random events. The EVT framework uses two different methods to model extreme values, namely the block maximum model (BMM) and the peak overthreshold model (POT). The block maximum model (BMM) divides the data into blocks and focuses on analyzing the maximum value within the block, but it will cause data waste. The peak overthreshold (POT) model aims to model data that exceeds a given threshold. Non-maximum values ​​within the block will not be discarded, thereby improving data efficiency. Existing warning line setting methods are often static, which means that the warning line cannot be updated with changes in the structural response level, and it is not appropriate to manually update the threshold frequently. At the same time, the above warning line setting still remains at the level of deterministic methods, without considering the uncertainty factors in the warning process.

[0005] In addition, sensor failure may lead to data loss or errors, thus interfering with the assessment of structural status. Researchers have proposed a variety of methods, including those based on statistical methods, signal processing, and machine learning, to diagnose and isolate sensor failures. However, the above methods all focus on the diagnosis of faults rather than classification, and it is very important to distinguish between sensor failures and structural state abnormalities. In actual situations, situations where real structural abnormal events and sensor failures occur at the same time will also occur with a certain probability, that is, the two are coupled with each other; if this happens, the key to preventing false alarms and misleading conclusions is to accurately identify and distinguish between structural abnormalities and sensor abnormalities, and focus on structural state abnormalities with higher priority. In infrastructure health monitoring, the introduction of decision support methods suitable for handling information conflicts can provide an economical, efficient and universal technical means for abnormality identification and early warning. Summary of the invention

[0006] Purpose of the invention: In order to overcome the deficiencies in the prior art, a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning is provided, which is a multi-sensor fusion anomaly warning method with automatic update of warning values ​​and quantification of the anomaly certainty of structural response in the form of probability. It has the advantages of strong interpretability and good data adaptability, can effectively use data streams to timely update anomaly warning values, and synthesize the anomaly certainty of multiple individual sensors to form a fused anomaly certainty, thereby better handling conflicts between indicators, avoiding false warnings caused by single sensor failures, and providing certain data support for structural maintenance.

[0007] Technical solution: To achieve the above purpose, the present invention provides a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning, comprising the following steps:

[0008] S1: Based on the monitoring data of the sensor, establish the historical data set of each sensor as sample data;

[0009] S2: Separate the temperature effect of the sample data and extract the response components after removing the temperature effect as the initial data set D for anomaly detection;

[0010] S3: Determine the initial threshold value according to the initial data set D;

[0011] S4: Based on the initial threshold, construct a super-threshold data set;

[0012] S5: Use the generalized Pareto distribution to fit the super-threshold data set, and obtain the posterior probability distribution of the shape parameter γ and the scale parameter σ through Bayesian inference;

[0013] S6: Calculate the probability distribution of the warning trigger value according to the posterior probability distribution of the shape parameter γ and the scale parameter σ, and determine the specific form of the probability distribution of the warning trigger value by fitting the sample distribution;

[0014] S7: Calculate the abnormality certainty index based on the probability distribution of the warning trigger value;

[0015] S8: Dynamically update the probability distribution of the warning trigger value through the streaming over-threshold method;

[0016] S9: The abnormality certainty indexes of the selected multiple individual sensors are synthesized through an evidence fusion method to obtain a fused abnormality certainty index.

[0017] Furthermore, the step S1 specifically includes: calculating the Pearson correlation coefficient between multiple sensors, selecting multiple individual sensors with good correlation, and extracting the structural response recorded by the sensors as the basis for establishing a fusion probability anomaly certainty index; selecting monitoring data within a certain period of time for all sensors, resampling, and establishing a historical data set of each sensor as sample data.

[0018] Furthermore, the specific method of separating the temperature effect in step S2 includes: establishing a Bayesian dynamic linear model, decomposing the obtained sample data into a superposition of a trend component, a periodic component and a residual component, and using a Kalman filter to dynamically update the posterior probability of the state parameters of each component, thereby extracting the temperature effect;

[0019] The Bayesian dynamic linear model consists of the observation equation and the state equation. The observation equation represents the state vector θ t and the observation matrix C t The generated observations Y t , and add the observation error V t ; The state equation represents the state vector θ t The evolution in time is governed by the state transfer matrix A t The influence of state error W t ;

[0020] Y t =C t θ t +v t ,v t ~N(0,V t )

[0021] θ t =A t θ t-1 +w t ,w t ~N(0,W t )

[0022] In the prediction stage, the Kalman filter estimates the prior probability distribution of the parameters at time T+1 through the state equation based on the posterior distribution of the hidden state parameters at time T; the posterior probability distribution of the hidden state parameters at time T is

[0023] P(θ t |D t )~N(μ t|t ,∑ t|t )

[0024] Among them, μ t|t and∑ t|t are the mean and covariance matrix of the posterior probability of the state parameters at time t; the prior distribution of the parameters at time T+1 is obtained by converting the posterior distribution through the state equation

[0025] P(θ t+1 |D t )~N(μ t+1|t ,∑ t+1|t )

[0026] Among them, μ t+1|t =A t+1 μ t|t ,Σ t+1|t =A t+1 ∑ t A' t+1 +W t+1 ;

[0027] The predicted value of the monitoring data at time T+1 is

[0028] P(Y t+1 |D t )~N(f t+1 ,Q t+1 )

[0029] Among them, E[Y t+1 |D t ]=C t+1 μ t+1|t ,Var[Y t+1 |D t ]=C t+1 ∑ t+1|t C' t+1 +V t+1 ;

[0030] When the observation data is obtained through the monitoring system, the posterior probability distribution of the state parameters is updated using the predicted value and the observed value.

[0031] P(θ t+1 |D t+1 )~N(μ t+1|t+1 ,∑ t+1|t+1 )

[0032] where t+1|t+1 =μ t+1|t +K t+1 e t+1

[0033] ∑ t+1|t+1 =Σ t+1|t -K t+1 K′ t+1 Q t+1

[0034] e t+1 =Y t+1 -f t+1 ,K t+1 =∑ t+1|t C t+1 / Q t+1 .

[0035] Furthermore, the initial threshold in step S3 includes an initial upper threshold and the initial lower threshold The initial upper threshold is determined by calculating the mean transcendental function MEF. and the initial lower threshold

[0036] Initial upper threshold The calculation includes:

[0037] A1: The probability density function of the generalized Pareto distribution is expressed as:

[0038]

[0039] in, σ>0 represents the shape and scale parameters respectively, and u is a specific threshold. Given a sufficiently large threshold u0, for any threshold u>u0, the mean excess function MEF(u) is expressed as:

[0040]

[0041] There is a linear correlation between the average excess function MEF(u) and the threshold u, and its approximate expression is:

[0042]

[0043] Among them, N u ={i|X i >u} is a sample X1,…,X n The number of observations that exceed a certain threshold u in , is the corresponding excess sample; therefore, the following points:

[0044]

[0045] Depicted in a two-dimensional coordinate system, if With respect to u (u>u0), it changes approximately linearly, so u0 is a suitable threshold;

[0046] A2: Linear fitting of the MEF function curve after each threshold u0, corresponding to Figure X;

[0047] A3: Calculate the root mean square error (RMES) between the linear fitting line and the MEF function curve, and use RMES to quantify the linearity of the MEF function curve;

[0048] The root mean square error RMSE is expressed as:

[0049]

[0050] Among them, y i represents the actual value of the mean excess function MEF(u), Represents the linear fitting value; the smaller the root mean square error RMSE, the better the linearity of the MEF(u) function curve;

[0051] A4: Select the threshold value at the MEF function curve with the smallest root mean square error as the initial upper threshold value

[0052] Initial lower threshold Calculation and Same.

[0053] Furthermore, the construction of the super-threshold data set in step S4 includes:

[0054] Extract the initial data set D that is greater than the initial upper threshold The data constitutes the super-threshold dataset Peaks up ;

[0055] Extract the initial data set D that is less than the initial lower bound threshold The data constitutes the super-threshold dataset Peaks down .

[0056] Furthermore, in step S5, the generalized Pareto distribution is used to fit the super-threshold data set Peaks up and Peaks down , the posterior probability distribution of shape parameter γ and scale parameter σ is obtained through Bayesian inference;

[0057] The core of Bayesian inference is the Bayesian formula, which describes how to update the posterior probability based on existing prior knowledge. Its mathematical expression is:

[0058]

[0059] Among them, θ is the unknown parameter to be inferred, that is, the shape parameter γ and scale parameter σ of the generalized Pareto model, P(θ|Y) is the posterior distribution of the parameter θ; P(Y|θ) is the likelihood function, which represents the possibility of observing data X under the condition that the parameter θ is known; P(θ) is the prior distribution of the parameter θ; P(Y) is the marginal likelihood, which is also the normalization constant; In fact, it is impossible to calculate the posterior distribution analytically, so numerical methods (such as the Markov chain Monte Carlo method) are used to sample the posterior probability distribution of the parameter.

[0060] Furthermore, the step S6 specifically includes:

[0061] B1: Based on the posterior probability distribution of the generalized Pareto distribution γ and the scale parameter σ, the warning trigger value is sampled; the value of the warning trigger based on a certain extreme value level is calculated according to the following formula

[0062]

[0063] Where p is the extreme value level corresponding to a certain return period. According to the design operation period of the bridge, the quantile corresponding to the 95% guarantee rate within the 100-year operation period is usually taken as the value of p;

[0064] B2: After sampling the samples of the warning trigger value, use the lognormal distribution to fit the warning trigger value T p The distribution of f(T p ; α, β), corresponding to Figure X.

[0065] Furthermore, in step S7, based on the warning trigger value T p The distribution of f(T p ; α, β), calculate the normalized abnormality certainty index, the calculation formula is as follows:

[0066]

[0067] Among them, α and β are the probability distribution parameters of the parameterized trigger. The greater the anomaly certainty, the greater the probability that the point is considered an anomaly.

[0068] Furthermore, in step S8, the warning trigger value T is dynamically updated by the streaming over-threshold method. p The distribution of f(T p ; α, β), specifically including:

[0069] C1: If the incoming data X new Less than the initial upper threshold Or greater than the initial lower threshold It is regarded as a normal value and the warning trigger value is not updated;

[0070] C2: If the incoming data X new Greater than the initial upper threshold Or less than the initial lower threshold Then, according to step S7, the normalized abnormality certainty m is calculated. i ;

[0071] C3: If the normalized anomaly certainty m i Meet 0.05 <m i <0.98, then the future flow data X new Add super-threshold dataset Peaks up or Peaks down , and recalculate and update the probability distribution of the warning trigger value through step S6;

[0072] C4: If the normalized anomaly certainty m i Meet m i >0.98, then the future flow data X new Add abnormal data set A up or A down .

[0073] Furthermore, the step S9 specifically includes:

[0074] D1: Determine the recognition framework Θ = {C1, C2}, where C1 represents the data warning state and C2 represents the data normal state;

[0075] D2: Arrange the abnormal certainty indexes of multiple selected individual sensors into a body of evidence;

[0076] The evidence body composed of the abnormal certainty index of multiple individual sensors is as follows:

[0077] E2:[m1(C1),m1(C2),m1(Θ)]

[0078] E2:[m2(C1),m2(C2),m2(Θ)]

[0079] …

[0080] E N :[m N (C1),m N (C2),m N (Θ)]

[0081] Among them, m i (C k ) is assigned to proposition C k confidence; using abnormal certainty as the confidence of proposition C km(C1) is the degree of certainty of the data warning state; m(C2) is the degree of certainty of the data normal state;

[0082] m(C1)=m i ,m(C2)=1-m i

[0083] D3: Calculate the correlation coefficient between each piece of evidence and determine the correlation coefficient between each piece of evidence m i Credibility;

[0084] Use the evidence synthesis formula to synthesize each piece of evidence in the evidence body, and use the synthesized evidence as the fusion anomaly certainty index;

[0085] The similarity factor between two pieces of evidence is defined as

[0086]

[0087] Among them, N A To identify the number of focal elements in the framework; calculate the correlation coefficient r(m i ,m j ) and establish a correlation matrix between the evidence

[0088]

[0089] According to the correlation matrix, determine the evidence body for each piece of evidence m i The credibility of is as follows:

[0090]

[0091] Where R i The value range of is [0,1];

[0092] D4: Introduce the credibility of evidence as a weight to modify the original body of evidence;

[0093] Evidence credibility R i It is used to modify the probability distribution function, considering ∑m(C k )=1, the modified formula is

[0094]

[0095] D5: Calculate the conflict coefficient between evidences;

[0096] Evidence E i and E j The conflict factor is

[0097]

[0098] The average conflict factor between evidences is

[0099]

[0100] Define the uncertainty of evidence as in N is the number of evidences;

[0101] D6: Based on the uncertainty between the evidences, the revised evidence body is synthesized to obtain the fusion anomaly certainty index; the combination rule of the evidence body is

[0102]

[0103] Among them, q(A) is the subset C of the identification frame Θ k The average support of

[0104]

[0105] Beneficial effects: Compared with the prior art, the present invention provides a multi-sensor fusion anomaly warning method with automatic update of warning values ​​and quantification of the abnormality certainty of structural response in the form of probability. It has the advantages of strong interpretability and good data adaptability, and can effectively use data streams to timely update abnormal warning values, and synthesize the abnormality certainty of multiple individual sensors to form a fused abnormality certainty. When the fused abnormality certainty is large, it indicates that the abnormal response of the structure is caused by an abnormal event; when the abnormality certainty index of the individual sensor is large, and the fused abnormality certainty is small, it indicates that the abnormal response of the structure is caused by a sensor failure. Therefore, the present invention can better handle conflicts between indicators, avoid false warnings caused by single sensor failures, ensure the accuracy of the warning, and provide certain data support for the maintenance of the structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 It is a schematic diagram of the process of the present invention;

[0107] Figure 2 The result graph of linear fitting for each threshold value;

[0108] Figure 3 It is the posterior probability distribution diagram of the abnormal warning trigger;

[0109] Figure 4 It is the fitting distribution diagram of the warning trigger value;

[0110] Figure 5 A display diagram of the dynamically updated early warning trigger value obtained based on the method of the present invention;

[0111] Figure 6 This is a display diagram of the abnormal certainty of multi-sensor fusion obtained based on the method of the present invention;

[0112] Figure 7 This is a graph showing the multi-sensor fusion anomaly certainty index obtained based on the method of the present invention. DETAILED DESCRIPTION

[0113] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0114] like Figure 1 As shown, the present invention provides a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold value and evidence reasoning, comprising the following steps:

[0115] S1: Based on the monitoring data of the sensor, establish the historical data set of each sensor as sample data;

[0116] The Pearson correlation coefficient between multiple sensors is calculated, and multiple individual sensors with good correlation are selected. The structural response recorded by the sensors is extracted as the basis for establishing the fusion probability anomaly certainty index; the monitoring data within a certain period of time is selected for all sensors, and after resampling, the historical data set of each sensor is established as the sample data.

[0117] S2: Separate the temperature effect of the sample data and extract the response components after removing the temperature effect as the initial data set D for anomaly detection;

[0118] The specific method of separating the temperature effect includes: establishing a Bayesian dynamic linear model, decomposing the obtained sample data into the superposition of trend components, periodic components and residual components, and using Kalman filtering to dynamically update the posterior probability of the state parameters of each component, thereby extracting the temperature effect;

[0119] The Bayesian dynamic linear model consists of the observation equation and the state equation. The observation equation represents the state vector θ t and the observation matrix C t The generated observations Y t , and add the observation error V t ; The state equation represents the state vector θ t The evolution in time is governed by the state transfer matrix A t The influence of state error W t ;

[0120] Y t =C t θ t +v t ,vt ~N(0,V t )

[0121] θ t =A t θ t-1 +w t ,w t ~N(0,W t )

[0122] In the prediction stage, the Kalman filter estimates the prior probability distribution of the parameters at time T+1 through the state equation based on the posterior distribution of the hidden state parameters at time T; the posterior probability distribution of the hidden state parameters at time T is

[0123] P(θ t |D t )~N(μ t|t ,∑ t|t )

[0124] Among them, μ t|t and∑ t|t are the mean and covariance matrix of the posterior probability of the state parameters at time t; the prior distribution of the parameters at time T+1 is obtained by converting the posterior distribution through the state equation

[0125] P(θ t+1 |D t )~N(μ t+1|t ,∑ t+1|t )

[0126] Among them, μ t+1|t =A t+1 μ t|t ,Σ t+1|t =A t+1 ∑ t A' t+1 +W t+1 ;

[0127] The predicted value of the monitoring data at time T+1 is

[0128] P(Y t+1 |D t )~N(f t+1 ,Q t+1 )

[0129] Among them, E[Y t+1 |D t ]=C t+1 μ t+1|t ,Var[Y t+1 |D t ]=C t+1 ∑ t+1|t C' t+1 +V t+1;

[0130] When the observation data is obtained through the monitoring system, the posterior probability distribution of the state parameters is updated using the predicted value and the observed value.

[0131] P(θ t+1 |D t+1 )~N(μ t+1|t+1 ,∑ t+1|t+1 )

[0132] where t+1|t+1 =μ t+1|t +K t+1 e t+1

[0133] ∑ t+1|t+1 =∑ t+1|t -K t+1 K′ t+1 Q t+1

[0134] e t+1 =Y t+1 -f t+1 ,K t+1 =∑ t+1|t C t+1 / Q t+1 .

[0135] S3: Determine the initial threshold value according to the initial data set D;

[0136] The initial threshold includes the initial upper threshold and the initial lower threshold The initial upper threshold is determined by calculating the mean transcendental function MEF. and the initial lower threshold

[0137] Initial upper threshold The calculation includes:

[0138] A1: The probability density function of the generalized Pareto distribution is expressed as:

[0139]

[0140] in, σ>0 represents the shape and scale parameters respectively, and u is a specific threshold. Given a sufficiently large threshold u0, for any threshold u>u0, the mean excess function MEF(u) is expressed as:

[0141]

[0142] There is a linear correlation between the average excess function MEF(u) and the threshold u, and its approximate expression is:

[0143]

[0144] Among them, N u ={i|X i >u} is a sample X1,…,X n The number of observations that exceed a certain threshold u in , is the corresponding excess sample; therefore, the following points:

[0145]

[0146] Depicted in a two-dimensional coordinate system, if With respect to u (u>u0), it changes approximately linearly, so u0 is a suitable threshold;

[0147] A2: Linearly fit the MEF function curve after each threshold u0. The specific fitting results are as follows: Figure 2 As shown;

[0148] A3: Calculate the root mean square error (RMES) between the linear fitting line and the MEF function curve, and use RMES to quantify the linearity of the MEF function curve;

[0149] The root mean square error RMSE is expressed as:

[0150]

[0151] Among them, y i represents the actual value of the mean excess function MEF(u), Represents the linear fitting value; the smaller the root mean square error RMSE, the better the linearity of the MEF(u) function curve;

[0152] A4: Select the threshold value at the MEF function curve with the smallest root mean square error as the initial upper threshold value

[0153] Initial lower threshold Calculation and Same.

[0154] S4: Based on the initial threshold, construct a super-threshold data set;

[0155] Extract the initial data set D that is greater than the initial upper threshold The data constitutes the super-threshold dataset Peaks up ;

[0156] Extract the initial data set D that is less than the initial lower bound threshold The data constitutes the super-threshold dataset Peaks down .

[0157] S5: Use the generalized Pareto distribution to fit the super-threshold data set, and obtain the posterior probability distribution of the shape parameter γ and the scale parameter σ through Bayesian inference;

[0158] Fitting the super-threshold dataset Peaks using generalized Pareto distribution up and Peaks down , the posterior probability distribution of shape parameter γ and scale parameter σ is obtained through Bayesian inference;

[0159] The core of Bayesian inference is the Bayesian formula, which describes how to update the posterior probability based on existing prior knowledge. Its mathematical expression is:

[0160]

[0161] Among them, θ is the unknown parameter to be inferred, that is, the shape parameter γ and scale parameter σ of the generalized Pareto model, P(θ|Y) is the posterior distribution of the parameter θ; P(Y|θ) is the likelihood function, which represents the possibility of observing data X under the condition that the parameter θ is known; P(θ) is the prior distribution of the parameter θ; P(Y) is the marginal likelihood, which is also the normalization constant; in fact, it is impossible to calculate the posterior distribution analytically, so a numerical method (such as the Markov chain Monte Carlo method) is used to sample the posterior probability distribution of the parameter, as shown in Figure 3 shown.

[0162] S6: Calculate the probability distribution of the warning trigger value according to the posterior probability distribution of the shape parameter γ and the scale parameter σ, and determine the specific form of the probability distribution of the warning trigger value by fitting the sample distribution;

[0163] Specifically include:

[0164] B1: Based on the posterior probability distribution of the generalized Pareto distribution γ and the scale parameter σ, the warning trigger value is sampled; the value of the warning trigger based on a certain extreme value level is calculated according to the following formula

[0165]

[0166] Where p is the extreme value level corresponding to a certain return period. According to the design operation period of the bridge, the quantile corresponding to the 95% guarantee rate within the 100-year operation period is usually taken as the value of p;

[0167] B2: After sampling the samples of the warning trigger value, use the lognormal distribution to fit the warning trigger value T p The distribution of f(T p ; α, β), as Figure 4 shown.

[0168] S7: Based on the warning trigger value T pThe distribution of f(T p ; α, β), calculate the normalized abnormality certainty index, the calculation formula is as follows:

[0169]

[0170] Among them, α and β are the probability distribution parameters of the parameterized trigger. The greater the abnormality certainty, the greater the probability that the point is considered abnormal; the warning trigger value obtained in this embodiment is as follows Figure 5 As shown;

[0171] S8: Dynamically update the warning trigger value T through the streaming threshold method p The distribution of f(T p ; α, β), specifically including:

[0172] C1: If the incoming data X new Less than the initial upper threshold Or greater than the initial lower threshold It is regarded as a normal value and the warning trigger value is not updated;

[0173] C2: If the incoming data X new Greater than the initial upper threshold Or less than the initial lower bound threshold Then, according to step S7, the normalized abnormality certainty m is calculated. i ;

[0174] C3: If the normalized anomaly certainty m i Meet 0.05 <m i <0.98, then the future flow data X new Add super-threshold dataset Peaks up or Peaks down , and recalculate and update the probability distribution of the warning trigger value through step S6;

[0175] C4: If the normalized anomaly certainty m i Meet m i >0.98, then the future flow data X new Add abnormal data set A up or A down .

[0176] S9: The abnormality certainty indexes of the selected multiple individual sensors are synthesized through the evidence fusion method to obtain the fused abnormality certainty index corresponding to the attached Figure 6 and attached Figure 7The fusion abnormality certainty represents the probability that the structural response is caused by an abnormal event. When the fusion abnormality certainty is large, it indicates that the abnormal response of the structure is caused by an abnormal event; when the abnormality certainty index of the individual sensor is large and the fusion abnormality certainty is small, it indicates that the abnormal response of the structure is caused by a sensor failure.

[0177] Specifically include:

[0178] D1: Determine the recognition framework Θ = {C1, C2}, where C1 represents the data warning state and C2 represents the data normal state;

[0179] D2: Arrange the abnormal certainty indexes of multiple selected individual sensors into a body of evidence;

[0180] The evidence body composed of the abnormal certainty index of multiple individual sensors is as follows:

[0181] E2:[m1(C1),m1(C2),m1(Θ)]

[0182] E2:[m2(C1),m2(C2),m2(Θ)]

[0183] …

[0184] E N :[m N (C1),m N (C2),m N (Θ)]

[0185] Among them, m i (C k ) is assigned to proposition C k confidence; using abnormal certainty as the confidence of proposition C k m(C1) is the degree of certainty of the data warning state; m(C2) is the degree of certainty of the data normal state;

[0186] m(C1)=m i ,m(C2)=1-m i

[0187] D3: Calculate the correlation coefficient between each piece of evidence and determine the correlation coefficient between each piece of evidence m i Credibility;

[0188] Use the evidence synthesis formula to synthesize each piece of evidence in the evidence body, and use the synthesized evidence as the fusion anomaly certainty index;

[0189] The similarity factor between two pieces of evidence is defined as

[0190]

[0191] Among them, NA To identify the number of focal elements in the framework; calculate the correlation coefficient r(m i ,m j ) and establish a correlation matrix between the evidence

[0192]

[0193] According to the correlation matrix, determine the evidence body for each piece of evidence m i The credibility of is as follows:

[0194]

[0195] Where R i The value range of is [0,1];

[0196] D4: Introduce the credibility of evidence as a weight to modify the original body of evidence;

[0197] Evidence credibility R i It is used to modify the probability distribution function, considering ∑m(C k )=1, the modified formula is

[0198]

[0199] D5: Calculate the conflict coefficient between evidences;

[0200] Evidence E i and E j The conflict factor is

[0201]

[0202] The average conflict factor between evidences is

[0203]

[0204] Define the uncertainty of evidence as in N is the number of evidences;

[0205] D6: Based on the uncertainty between the evidences, the revised evidence body is synthesized to obtain the fusion anomaly certainty index; the combination rule of the evidence body is

[0206]

[0207] Among them, q(A) is the subset C of the identification frame Θ k The average support of

[0208]

[0209] Based on the above scheme, in order to verify the effectiveness and effect of the method of the present invention, this embodiment uses the method of the present invention to perform abnormal warning analysis on the structural monitoring data of a suspension bridge structure, as follows:

[0210] The structural health monitoring sensors of a suspension bridge selected based on the Pearson correlation index are shown in Table 1:

[0211] Table 1 Correlation coefficients of selected sensors

[0212]

[0213]

[0214] In order to verify the effectiveness of the method of the present invention for integrating the abnormal certainty index, abnormal warnings were carried out on the monitoring data under the following two working conditions:

[0215] Working condition 1: real abnormal state of the structure (vehicle collision);

[0216] Working condition 2: Sensor failure.

[0217] Analysis of numerical results:

[0218] Working condition 1: In this embodiment, the monitoring data of the selected sensor on May 10, 2024 is given an abnormal warning, and the method of step S1 to step S9 is used to dynamically update the abnormal warning value and calculate the abnormal certainty index. The abnormal certainty index at the time of vehicle collision is shown in Table 2.

[0219] Table 2 Abnormal certainty index of selected multiple body sensors

[0220]

[0221] Using the method in step 9, the abnormality certainty indexes of multiple sensors are synthesized to obtain a fusion abnormality certainty index of 0.99. Figure 6 It can be seen that the fusion anomaly certainty index is high, indicating that a real abnormal event has occurred.

[0222] Working condition 2: In this implementation example, the monitoring data of the selected sensor on March 1, 2024 is given an abnormal warning. The method from step S1 to step S9 is used to dynamically update the abnormal warning value and calculate the abnormal certainty index. The abnormal certainty index at the time of sensor failure is shown in Table 3.

[0223] Table 3 Abnormal certainty index of selected multiple body sensors

[0224]

[0225] Using the method of step S9, the abnormality certainty indexes of multiple sensors are synthesized to obtain a fused abnormality certainty index of 0. Figure 7 It can be seen that although the abnormality certainty index of the individual sensor is high, the fusion abnormality certainty index is low, indicating that the abnormal response at this time is caused by sensor failure.

Claims

1. A structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning, characterized in that: The steps include: S1: Based on the monitoring data of the sensor, establish the historical data set of each sensor as sample data; S2: Separate the temperature effect of the sample data and extract the response components after removing the temperature effect as the initial data set D for anomaly detection; S3: Determine the initial threshold value according to the initial data set D; S4: Based on the initial threshold, construct a super-threshold data set; S5: Use the generalized Pareto distribution to fit the super-threshold data set, and obtain the posterior probability distribution of the shape parameter γ and the scale parameter σ through Bayesian inference; S6: Calculate the probability distribution of the warning trigger value according to the posterior probability distribution of the shape parameter γ and the scale parameter σ, and determine the specific form of the probability distribution of the warning trigger value by fitting the sample distribution; S7: Calculate the abnormality certainty index based on the probability distribution of the warning trigger value; S8: Dynamically update the probability distribution of the warning trigger value through the streaming over-threshold method; S9: The abnormality certainty indexes of the selected multiple individual sensors are synthesized through an evidence fusion method to obtain a fused abnormality certainty index.

2. According to claim 1, a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning is characterized in that: The step S1 specifically includes: calculating the Pearson correlation coefficient between multiple sensors, selecting multiple individual sensors with good correlation, extracting the structural response recorded by the sensors as the basis for establishing the fusion probability anomaly certainty index; resampling the monitoring data within the selected time period of all sensors, and establishing the historical data set of each sensor as sample data.

3. According to claim 1, a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning is characterized in that: The specific method of separating the temperature effect in step S2 includes: establishing a Bayesian dynamic linear model, decomposing the obtained sample data into a superposition of a trend component, a periodic component and a residual component, and using a Kalman filter to dynamically update the posterior probability of the state parameters of each component, thereby extracting the temperature effect; The Bayesian dynamic linear model consists of the observation equation and the state equation. The observation equation represents the state vector θ t and the observation matrix C t The generated observations Y t , and add the observation error V t ; The state equation represents the state vector θ t The evolution in time is controlled by the state transfer matrix A t The influence of state error W t ; Y t =C t θ t +v t ,v t ~N(0,V t ) i t =A t i t-1 +w t ,w t ~N(0,W t ) In the prediction stage, the Kalman filter estimates the prior probability distribution of the parameters at time T+1 through the state equation based on the posterior distribution of the hidden state parameters at time T; the posterior probability distribution of the hidden state parameters at time T is P(θ t |D t )~N(μ t|t ,S t|t ) Among them, μ t|t and∑ t|t are the mean and covariance matrix of the posterior probability of the state parameters at time t; the prior distribution of the parameters at time T+1 is obtained by converting the posterior distribution through the state equation P(θ t+1 |D t )~N(μ t+1|t ,∑ t+1|t ) Among them, m t+1|t =A t+1 m t|t ,S t+1|t =A t+1 ∑ t A' t+1 +W t+1 ; The predicted value of the monitoring data at time T+1 is P(Y t+1 |D t )~N(f t+1 ,Q t+1 ) where, E[Y t+1 |D t = C t+1 μ t+1|t , Var[Y t+1 |D t = C t+1 ∑ t+1|t C' t+1 + V t+1 ; When the observation data is obtained through the monitoring system, the posterior probability distribution of the state parameters is updated using the predicted value and the observed value. P(θ t+1 |D t+1 )~N(μ t+1|t+1 ,∑ t+1|t+1 ) where m t+1|t+1 =μ t+1|t +K t+1 e t+1 ∑ t+1|t+1 =∑ t+1|t -K t+1 K t+1 Q t+1 e t+1 =Y t+1 -f t+1 ,K t+1 =∑ t+1|t C t+1 / Q t+1 。 4. According to claim 1, a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning is characterized in that: The initial threshold in step S3 includes an initial upper threshold and the initial lower threshold The initial upper threshold is determined by calculating the mean transcendental function MEF. and the initial lower threshold Initial upper threshold The calculation includes: A1: The probability density function of the generalized Pareto distribution is expressed as: in, σ>0 represents the shape and scale parameters respectively, and u is a specific threshold. Given a threshold u0, for any threshold u>u0, the mean excess function MEF(u) is expressed as: There is a linear correlation between the average excess function MEF(u) and the threshold u, and its approximate expression is: Among them, N u ={i|X i >u} is a sample X1,…,X n The number of observations that exceed a certain threshold u in X'1, X'2, ..., is the corresponding excess sample; therefore, the following points: Depicted in a two-dimensional coordinate system, if With respect to u (u>u0), it changes approximately linearly, so u0 is a suitable threshold; A2: Linear fitting of the MEF function curve after each threshold u0; A3: Calculate the root mean square error (RMES) between the linear fitting line and the MEF function curve, and use RMES to quantify the linearity of the MEF function curve; The root mean square error RMSE is expressed as: Among them, y i represents the actual value of the mean excess function MEF(u), represents the linear fitting value; A4: Select the threshold value at the MEF function curve with the smallest root mean square error as the initial upper threshold value Initial lower threshold Calculation and Same.

5. According to claim 4, a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning is characterized in that: The construction of the super-threshold data set in step S4 includes: Extract the initial data set D that is greater than the initial upper threshold The data constitutes the super-threshold dataset Peaks up ; Extract the initial data set D that is less than the initial lower bound threshold The data constitutes the super-threshold dataset Peaks down .

6. According to claim 5, a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning is characterized in that: In step S5, the generalized Pareto distribution is used to fit the super-threshold data set Peaks up and Peaks down , the posterior probability distribution of shape parameter γ and scale parameter σ is obtained through Bayesian inference; The core of Bayesian inference is the Bayesian formula, which describes how to update the posterior probability based on existing prior knowledge. Its mathematical expression is: Among them, θ is the unknown parameter to be inferred, that is, the shape parameter γ and scale parameter σ of the generalized Pareto model, P(θ|Y) is the posterior distribution of parameter θ; P(Y|θ) is the likelihood function, which represents the possibility of observing data X under the condition that parameter θ is known; P(θ) is the prior distribution of parameter θ; P(Y) is the marginal likelihood, which is also the normalization constant; the posterior probability distribution of the parameter is sampled using numerical methods.

7. According to claim 6, a structural monitoring data fusion probability anomaly warning method based on streaming over-threshold and evidence reasoning is characterized in that: The step S6 specifically includes: B1: Sample the warning trigger value according to the posterior probability distribution of the generalized Pareto distribution γ and the scale parameter σ; The value of the warning trigger based on a certain extreme value level is calculated according to the following formula Among them, p is the extreme level corresponding to a certain return period; B2: After sampling the samples of the warning trigger value, use the lognormal distribution to fit the warning trigger value T p The distribution of f(T p ; α, β).

8. The method for structural monitoring data fusion probability anomaly warning based on streaming over-threshold and evidence reasoning according to claim 7 is characterized in that: In step S7, based on the warning trigger value T p The distribution of f(T p ; α, β), calculate the normalized abnormality certainty index, the calculation formula is as follows: Among them, α and β are the probability distribution parameters of the parameterized trigger.

9. The method for structural monitoring data fusion probability anomaly warning based on streaming over-threshold and evidence reasoning according to claim 8 is characterized in that: In step S8, the warning trigger value T is dynamically updated by the streaming over-threshold method. p The distribution of f(T p ; α, β), specifically including: C1: If the incoming data X new Less than the initial upper threshold Or greater than the initial lower threshold It is regarded as a normal value and the warning trigger value is not updated; C2: If the incoming data X new Greater than the initial upper threshold Or less than the initial lower bound threshold Then, according to step S7, the normalized abnormality certainty m is calculated. i ; C3: If the normalized anomaly certainty m i Meet 0.05 <m i <0.98, then the future flow data X new Add super-threshold dataset Peaks up or Peaks down , and recalculate and update the probability distribution of the warning trigger value through step S6; C4: If the normalized anomaly certainty m i Meet m i >0.98, then the future flow data X new Add abnormal data set A up or A down .

10. The method for structural monitoring data fusion probability anomaly warning based on streaming over-threshold and evidence reasoning according to claim 8 is characterized in that: The step S9 specifically includes: D1: Determine the recognition framework Θ = {C1, C2}, where C1 represents the data warning state and C2 represents the data normal state; D2: Arrange the abnormal certainty indexes of multiple selected individual sensors into a body of evidence; The evidence body composed of the abnormal certainty index of multiple individual sensors is as follows: E2:[m1(C1),m1(C2),m1(Θ)] E2:[m2(C1),m2(C2),m2(Θ)] … E N :[m N (C1),m N (C2),m N (Θ)] Among them, m i (C k ) is assigned to proposition C k The reliability of the proposition C is used as the k m(C1) is the degree of certainty of the data warning state; m(C2) is the degree of certainty of the data normal state; m(C1)=m i ,m(C2)=1-m i D3: Calculate the correlation coefficient between each piece of evidence and determine the correlation coefficient between each piece of evidence m i Credibility; Use the evidence synthesis formula to synthesize each piece of evidence in the evidence body, and use the synthesized evidence as the fusion anomaly certainty index; The similarity factor between two pieces of evidence is defined as Among them, N A To identify the number of focal elements in the framework; calculate the correlation coefficient r(m i ,m j ) and establish a correlation matrix between the evidence According to the correlation matrix, determine the evidence body for each piece of evidence m i The credibility of is as follows: Where R i The value range of is [0,1]; D4: Introduce the credibility of evidence as a weight to modify the original body of evidence; Evidence credibility R i It is used to modify the probability distribution function, considering ∑m(C k )=1, the modified formula is D5: Calculate the conflict coefficient between evidences; Evidence E i and E j The conflict factor is The average conflict factor between evidences is Define the uncertainty of evidence as in N is the number of evidences; D6: Based on the uncertainty between the evidences, the revised body of evidence is synthesized to obtain the fusion anomaly certainty index; The combination rule of the evidence body is Among them, q(A) is the subset C of the identification frame Θ k The average support of

Images