Application of multi-dimensional and multi-level analysis method in concrete dam safety monitoring and analysis system platform

Abstract

The application provides a method for applying a multi-dimension multi-level analysis method to a concrete dam safety monitoring analysis system platform, which comprises the following steps: step 1, classifying different dimensions and different types of monitoring physical quantities obtained by monitoring a reservoir dam by using an SVM classification principle, and judging whether the monitoring physical quantities are gross errors by using a Lade criterion; step 2, performing first layer data analysis on the processed different dimensions and different types of monitoring physical quantities, realizing data compilation, and displaying on the platform; step 3, performing second layer data analysis on the structural analysis and calculation of the dam by combining inversion analysis and using strength theory; step 4, establishing a monitoring model to perform third layer data analysis, and guaranteeing short-term prediction accuracy; and step 5, establishing a medium and long term prediction monitoring model to perform fourth layer data analysis, and guaranteeing long-term prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dam safety monitoring system, in particular to the application method of multi-dimensional and multi-level analysis method in concrete dam safety monitoring analysis system platform. BACKGROUND

[0002] The existing dam safety online monitoring system is generally composed of three parts: observation sensors, telemetry hub and automatic safety monitoring microcomputer system. For the automatic safety monitoring microcomputer system, the built-in safety monitoring analysis system platform is the main functional module for dam safety monitoring data processing, compilation, analysis and prediction.

[0003] Traditionally, the monitoring analysis system platform has the following problems: first, the ability to identify gross errors of monitoring data is weak, it can detect obvious outliers well, but the judgment ability of other types of errors is weak; second, the data information mining ability is not enough, except for unified eigenvalue analysis, statistical quantity calculation and effect quantity trend display, there is no deeper display of data connotation information, and the data compilation depth is low; third, the long-term prediction uses the multivariate regression algorithm based on the least square principle, which is prone to overfitting in data fitting, and the accuracy is guaranteed in short-term prediction, but the accuracy is poor in medium and long-term prediction. SUMMARY

[0004] The present application aims to provide an application method of multi-dimensional and multi-level analysis method in concrete dam safety monitoring analysis system platform, which can further improve the reliability of monitoring data, fully mine and display the data connotation information, and has the functions of autonomous supervision, learning and prediction, thereby effectively improving the evaluation and prediction functions of the monitoring data analysis system platform for dam safety monitoring.

[0005] In order to achieve the above technical features, the present application is implemented as follows: the application method of multi-dimensional and multi-level analysis method in concrete dam safety monitoring analysis system platform comprises the following steps:

[0006] Step 1: classifying different dimensions and different types of monitoring physical quantities obtained by reservoir dam monitoring by using SVM classification principle, and judging whether the monitoring physical quantity is a gross error by using the Lade criterion;

[0007] Step 2: performing first layer data analysis on the processed different dimensions and different types of monitoring physical quantities, realizing data compilation, and displaying on the platform;

[0008] Step 3: performing second layer data analysis on the structural analysis and calculation of the dam by combining with the inversion analysis and using the strength theory;

[0009] Step 4: performing third layer data analysis by establishing a monitoring model to ensure the accuracy of short-term prediction;

[0010] Step 5, long-term prediction monitoring model is established to carry out fourth layer data analysis, and long-term prediction accuracy is ensured. In the process of using SVM classification principle in step 1, the specific requirements of the SVM classifier are as follows:

[0011] ①According to the change characteristics of the reservoir dam monitoring physical quantity, the kernel function selection in the SVM classifier is "polynomial kernel function + Gaussian radial basis kernel function";

[0012] ②In order to ensure the accuracy of data classification and avoid overfitting phenomenon, the penalty term is introduced into the hyperplane optimization objective function of the SVM classifier Wherein, C is a penalty parameter, C > 0; ξ i is the ith relaxation variable; m is the total number of relaxation variables;

[0013] ③According to the principle that the error obeys normal distribution, the sample unbiased consistent estimate σ is used instead of σ in Lagrange criterion , wherein, S is the sample standard deviation; n is the total number of samples; is the estimated value of the SVM model; is the sample mean.

[0014] The first layer data analysis in step 2 specifically includes:

[0015] ①Analysis of characteristic value of monitoring physical quantity: including annual maximum value, annual minimum value, annual amplitude and annual mean value of monitoring physical quantity, through the analysis of characteristic value size and change and occurrence time of each year of each monitoring physical quantity, the consistency and rationality of monitoring physical quantity are judged;

[0016] ②Correlation analysis of monitoring physical quantity: through the correlation diagram and correlation calculation accuracy between monitoring physical quantity and environmental quantity, between similar and non-similar measuring points, the influence degree of environmental quantity and the correlation characteristics between measuring points are judged;

[0017] ③Analysis of spatial distribution of monitoring physical quantity: the distribution law of monitoring physical quantity along the typical profile and the longitudinal profile along the dam axis, and the change law along the river direction and elevation are analyzed, the change law of monitoring physical quantity under different perspectives is revealed, and the spatial distribution rationality is judged by comparing with the design calculation results and classical engineering cases;

[0018] ④Comparison analysis of monitoring physical quantity and specification / design value: the measured value of dam body concrete, anchor cable and anchor rod, pressure steel pipe stress and strain value and dam foundation pressure is compared with the design limit or material strength limit value, and the preliminary evaluation of structural state is made.

[0019] The second layer data analysis in step 3 specifically includes:

[0020] ①Select the dam and dam foundation part displacement observation points to inverse the dam and dam foundation material parameters, the requirements that these measuring points have high sensitivity, strong reliability, obvious change rule characteristics, record the dam and dam foundation material parameters obtained by inversion as: E1 and E2;

[0021] ②In the stress concentration parts of the dam body, multi-directional strain gauges are arranged, according to the stress-strain conversion formula combined with the inverted dam foundation material parameters, the strain of the monitoring part is converted into stress, and compared with the yield stress / limit stress of the material, the structural safety of the dam at the present stage is analyzed;

[0022] ③For long-term monitoring of the dam, based on the time series of strain values of the key monitoring parts, the time series of material parameters E are obtained by regular inversion of material parameters, and the least square method is used to predict the degradation limit of material parameters E; according to the strain change range and the degradation limit of material parameters E, the stress change limit of the part is calculated, and the future safety of the dam structure is predicted and evaluated.

[0023] The third layer data analysis in step 4 specifically includes:

[0024] ①The traditional monitoring model includes: statistical model, deterministic model, mixed model and intelligent algorithm-based monitoring model, considering the actual properties of the project and the second layer material parameter inversion basis, the deterministic model is used in the third layer data analysis to establish the functional relationship between the influencing factors and the effect quantity; Taking displacement effect as an example, under external load, the displacement δ of any observation part of the dam and dam foundation can be divided into water pressure component f H (t), temperature component f T (t) and time-dependent component f θ (t) caused by concrete or rock mass material creep and plastic deformation, that is:

[0025] δ=f H (t)+f T (t)+f θ (t) (1)

[0026] Based on the second layer material parameter inversion results, the water pressure component and temperature component are calculated by using the established finite element model, and the time-dependent component is calculated by using the statistical model;

[0027] ②The displacement deterministic model of the measuring point is expressed by the following formula:

[0028]

[0029] In the formula: is the displacement value calculated by the deterministic model; b1, b2, b3, b0 are coefficient constants; c1, c2, c3 are coefficient constants; δ H is the water pressure component displacement calculated by the finite element method; δ Tθ is the 0.01 times of time t; k is a coefficient constant;

[0030] The parameters in the model are obtained by using the least square method.

[0031] The fourth layer data analysis in the step 5 specifically comprises:

[0032] ①The most important reason why the medium and short-term monitoring model has low accuracy in long-term prediction is that the model has poor nonlinear modeling capability and the model has no supervised learning capability; in order to guarantee the long-term prediction accuracy of the model, the SVM algorithm commonly used in machine learning is introduced again in the fourth layer to establish a monitoring prediction model;

[0033] ②According to the measured monitoring data, the third layer deterministic model is introduced to obtain the estimated values f H (t), f T (t) and f θ (t) of f H1 (t), f T1 (t) and f θ1 (t) respectively; the SVM model is introduced for f θ1 (t), and a regression prediction model between time t and f θ1 (t) is established; the relaxation coefficient, the penalty coefficient and the kernel function in the SVM model are continuously adjusted to improve the prediction accuracy of the model, and the estimated value f θ2 (t) of f θ1 (t) in the SVM model is obtained.

[0034] ③The medium and long-term prediction monitoring model is established:

[0035]

[0036] In the formula, d1 and d2 are model correction coefficients respectively, d0 is a constant term; f θ1 (t) is the estimated value of the time effect of the deterministic model;

[0037] In the model, d1 and d2 are calculated by a genetic algorithm, and the objective function is minimized.

[0038] The present application has the following beneficial effects:

[0039] 1. The present application improves the rough error identification capability of the monitoring data in the traditional dam safety monitoring and analysis system platform, and can better identify other types of errors in addition to obvious outliers.

[0040] ​2, The application improves information mining of different dimension monitoring data, introduces spatial distribution analysis and comparison with standard value and typical engineering cases in addition to traditional eigenvalue analysis and correlation analysis, and improves data information mining display capability.

[0041] 3, The application combines inversion analysis and strength theory, further judges safety reliability of the hydraulic structure from structural mechanics and material mechanics, and provides various ways for judging safe and stable operation.

[0042] 4, The application introduces an SVM model, fully mines change trend of various nonlinear factors contained in the time-effect component f θ (t), guarantees medium and long term monitoring of each measuring point of the model, and has guaranteed accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0043] The application will be further described below in combination with the drawings and embodiments.

[0044] Figure 1 It is a flow chart of application of the multi-dimension and multi-level analysis method of the application on a dam safety monitoring platform.

[0045] Figure 2 It is rough error identification and data processing result display realized on a matlab GUI interface.

[0046] Figure 3 It is first layer data processing realized on a matlab GUI interface.

[0047] Figure 4 It is third layer data processing realized on a matlab GUI interface. DETAILED DESCRIPTION

[0048] Embodiment 1:

[0049] Referring to Figures 1-4 , the application is a multi-dimension and multi-level analysis method applied in a concrete dam safety monitoring analysis system platform, which comprises the following steps:

[0050] Step 1, different dimensions and different types of monitoring physical quantities obtained by reservoir dam monitoring are classified by using SVM classification principle, and whether the monitoring physical quantity is a rough error is judged by using the Lardy criterion;

[0051] In order to ensure the wide applicability of the SVM classifier, the above steps require:

[0052] ①According to the change characteristics of the reservoir dam monitoring physical quantity, the kernel function in the SVM classifier is selected as "polynomial kernel function + Gaussian radial basis kernel function";

[0053] ②To ensure the accuracy of data classification and not to appear over-fitting phenomenon, the penalty term is introduced into the optimization objective function of the hyperplane in SVM classifier wherein, C is a penalty parameter, C > 0; ξ i is the i-th relaxation variable; m is the total number of relaxation variables;

[0054] ③According to the principle that the error obeys normal distribution, the sample unbiased consistent estimate is used to replace σ in the Lagrange criterion, wherein, S is the sample standard deviation; n is the total number of samples; is the estimated value of the SVM model; is the sample mean.

[0055] Step 2, the first layer data analysis is carried out on the processed different dimensions and different types of monitoring physical quantities, the data is compiled, and the platform is displayed;

[0056] The first layer data analysis specifically includes:

[0057] ①Monitoring physical quantity eigenvalue analysis: including the maximum value, the minimum value, the annual amplitude and the annual mean value of the monitoring physical quantity, through the analysis of the size and change of the characteristic value of each year of each monitoring physical quantity and the appearance time, the consistency and rationality of the monitoring physical quantity are judged;

[0058] ②Monitoring physical quantity correlation analysis: through the correlation diagram and correlation calculation accuracy between the monitoring physical quantity and the environmental quantity, between the same type and non-same type measuring points, the influence degree of the environmental quantity and the correlation characteristics between the measuring points are judged;

[0059] ③Monitoring physical quantity spatial distribution analysis: the distribution law of monitoring physical quantity along the typical profile and the longitudinal profile along the dam axis, and the change law along the river direction and the elevation are analyzed, the change law of monitoring physical quantity under different perspectives is revealed, and the spatial distribution rationality is judged by comparing with the design calculation results and classical engineering cases;

[0060] ④Monitoring physical quantity and specification / design value comparison analysis: the measured values of the dam body concrete, anchor cable and anchor rod, pressure steel pipe stress and strain value and dam foundation pressure are compared with the design limit or material strength limit value, and the preliminary evaluation of the structural state is made.

[0061] Step 3, the second layer data analysis is carried out on the structural analysis and calculation of the dam by combining with the inversion analysis and using the strength theory;

[0062] The second layer data analysis specifically includes:

[0063] ①Selecting the displacement observation points of the dam and the dam foundation to carry out inversion on the dam body and dam foundation material parameters, requiring that the measuring points have the characteristics of high sensitivity, strong reliability and obvious change law, and recording the dam body and dam foundation material parameters obtained by inversion as E1 and E2; ​

[0064] ②In the stress concentration parts of the dam body, multi-directional strain gauges are arranged. According to the stress-strain conversion formula and the material parameters obtained by inversion, the strain of the monitoring part is converted into stress, which is compared with the yield stress / limit stress of the material to analyze the structural safety of the dam at the present stage;

[0065] ③For a long-term monitored dam, based on the time series of strain values of the key monitoring parts, the time series of material parameters E are obtained by regular inversion of material parameters, and the degradation limit of material parameters E is predicted by using the least square method. According to the strain change range and the degradation limit of material parameters E, the stress change limit of the part is calculated to predict and evaluate the future safety of the dam structure.

[0066] Step 4, establish a monitoring model for third layer data analysis to ensure short-term prediction accuracy;

[0067] The third layer data analysis specifically includes:

[0068] ①The traditional monitoring model includes: statistical model, deterministic model, hybrid model and monitoring model based on intelligent algorithm. Considering the actual properties of the project and the basis of the second layer material parameter inversion, the deterministic model is used in the third layer data analysis to establish the functional relationship between the influencing factors and the effect quantity. Taking displacement effect as an example, under external load, the displacement δ of any observation part of the dam and dam foundation can be divided into water pressure component f H (t), temperature component f T (t) and time-dependent component f θ (t) caused by concrete or rock mass material creep and plastic deformation, i.e.:

[0069] δ=f H (t)+f T (t)+f θ (t)(1)

[0070] Based on the results of the second layer material parameter inversion, the water pressure component and the temperature component are calculated using the established finite element model, and the time-dependent component is calculated using statistical model.

[0071] ②The displacement deterministic model of the measuring point is expressed by the following formula:

[0072]

[0073] In the formula: is the displacement value calculated by the deterministic model; b1, b2, b3, b0 are coefficient constants; c1, c2, c3 are coefficient constants; δ H is the water pressure component displacement calculated by finite element; δ T is the temperature component displacement calculated by finite element; θ is 0.01 times of time t; k is a coefficient constant;

[0074] The parameters in the model are obtained by using the least square method principle.

[0075] Step 5, establishing a medium and long term prediction monitoring model to perform fourth layer data analysis, and guaranteeing long term prediction accuracy;

[0076] The fourth layer data analysis specifically includes:

[0077] 1. The most important reason why the medium and short term monitoring model has low accuracy in long term prediction is that the model has poor non-linear module capability and no supervision learning capability; in order to guarantee the long term prediction accuracy of the model, the SVM algorithm commonly used in machine learning is introduced again in the fourth layer to establish a monitoring prediction model;

[0078] 2. According to the measured monitoring data, the third layer deterministic model is introduced to obtain the estimated values f H (t), f T (t) and f θ (t) of f H1 (t), f T1 (t) and f θ1 (t) respectively; the SVM model is introduced to f θ1 (t), and a regression prediction model between time t and f θ1 (t) is established; by continuously adjusting the relaxation coefficient, the penalty coefficient and the kernel function in the SVM model, the prediction accuracy of the model is improved, and the estimated value f θ1 (t) of f θ2 (t) in the SVM model is obtained;

[0079] 3. Establishing a medium and long term prediction monitoring model:

[0080]

[0081] In the formula, d1 and d2 are model correction coefficients respectively, d0 is a constant term; f θ1 (t) is the estimated value of the time effect of the deterministic model;

[0082] The d1 and d2 in the model are calculated by a genetic algorithm, and the objective function is minimized.

[0083] Principle of the application:

[0084] ​For large hydraulic structures, the problems of dam safety monitoring data processing are: large amount of monitoring data, low efficiency of gross error identification, engineering experience-based data analysis, obvious nonlinear characteristics of monitoring data, and shallow information mining of monitoring data. In the method, the classifier and the regression prediction module in the machine learning SVM method are applied in data gross error identification and medium and long term monitoring, respectively. Since the SVM classification principle is to rely on the support vector of the data edge to make classification decision, it has obvious advantages in analyzing large amount of data. Secondly, according to the related theory of functional analysis, as long as there is a kernel function K(x i ,x j ) that satisfies the Mercer condition, the low-dimensional nonlinear problem can be mapped to high-dimensional to linearize, so the SVM theory also has advantages in processing nonlinear data. Thirdly, the computer program is introduced to automatically identify gross error, which can obviously improve the efficiency of data gross error identification compared with the previous manual identification through engineering experience. Fourthly, the space distribution analysis of observation data, comparison with the standard value, comparison with typical engineering cases and strength calculation are introduced, which can obviously improve the data information mining and display ability compared with the traditional eigenvalue analysis method.

Claims

1. A method for applying multi-dimensional, multi-level analysis to a concrete dam safety monitoring and analysis system platform, characterized in that, It includes the following steps: Step 1: Classify the monitoring physical quantities of different dimensions and types obtained from reservoir dam monitoring using the SVM classification principle, and use the Laida criterion to determine whether the monitoring physical quantities are gross errors. Step 2: Perform first-level data analysis on the processed monitoring physical quantities of different dimensions and types, realize data compilation, and display it on the platform; Step 3: Combine inversion analysis with strength theory to perform a second layer of data analysis on the structural analysis calculation of the dam; Step 4: Establish a monitoring model to conduct third-level data analysis to ensure short-term forecast accuracy; Step 5: Establish a medium- to long-term forecasting and monitoring model to conduct fourth-level data analysis and ensure long-term forecast accuracy; The second-level data analysis in step 3 specifically includes: ① Select some displacement observation points of the dam and dam foundation to invert the material parameters of the dam body and dam foundation. These observation points are required to have high sensitivity, high reliability and obvious variation law. The inverted material parameters of the dam body and dam foundation are denoted as E1 and E2. ② Multi-dimensional strain gauges are installed at stress concentration points in the dam body. Based on the stress-strain conversion formula and the dam foundation material parameters obtained by inversion, the strain at the monitored location is converted into stress, and compared with the yield stress / ultimate stress of the material at that location to analyze the structural safety of the dam at the current stage. ③ For dams with long monitoring periods, based on the time series of strain values ​​at key monitoring locations, the time series of material parameters E1 and E2 of the dam body and foundation are obtained by periodically inverting the material parameters. The deterioration limits of material parameters E1 and E2 of the dam body and foundation are predicted using the least squares method. The stress change limit of the location is calculated based on the strain variation range and the deterioration limits of material parameters E1 and E2, and the future safety of the dam structure is predicted and evaluated.

2. The application method of the multi-dimensional and multi-level analysis method according to claim 1 in the safety monitoring and analysis system platform of concrete dams, characterized in that, In step 1, during the SVM classification process, the specific requirements for the SVM classifier are as follows: ① Based on the characteristics of changes in physical quantities monitored by reservoir dams, the kernel function selected in the SVM classifier is "polynomial kernel function + Gaussian radial basis kernel function"; ② To ensure accurate data classification and prevent overfitting, a penalty term is introduced into the hyperplane optimization objective function of the SVM classifier. ,in, C The penalty parameter is C > 0; For the first i One slack variable; m This represents the total number of slack variables. ③ Based on the principle that the error follows a normal distribution, σ in the Laida criterion is estimated using the unbiased combination of samples. Instead, among them, S The standard deviation is the sample standard deviation. n The total number of samples; The estimated value for the SVM model; This is the sample mean.

3. The application method of the multi-dimensional and multi-level analysis method according to claim 1 in the safety monitoring and analysis system platform of concrete dams, characterized in that, The first-level data analysis in step 2 specifically includes: ①Analysis of characteristic values ​​of monitored physical quantities: This includes the maximum, minimum, annual variation and annual average of monitored physical quantities within the year. By analyzing the magnitude and change of characteristic values ​​of each monitored physical quantity in each year and the time of occurrence, the consistency and rationality of the monitored physical quantities can be judged. ② Correlation analysis of monitored physical quantities: By analyzing the correlation diagrams and correlation calculation accuracy between monitored physical quantities and environmental quantities, and between similar and dissimilar measuring points, the influence of environmental quantities and the correlation characteristics between measuring points can be determined. ③ Spatial distribution analysis of monitored physical quantities: Analyze the distribution patterns of monitored physical quantities in typical profiles and longitudinal profiles along the dam axis, as well as the variation patterns along the river direction and elevation, reveal the variation patterns of monitored physical quantities from different perspectives, and compare them with design calculation results and classic engineering cases to judge the rationality of spatial distribution; ④ Comparative analysis of monitored physical quantities with standard / design values: The stress and strain values ​​of dam concrete, anchor cables and anchor rods, pressure steel pipes, and dam foundation pressure are compared with the measured values ​​and the design limits or material strength limits to make a preliminary evaluation of the structural behavior.

4. The application method of the multi-dimensional and multi-level analysis method according to claim 1 in the concrete dam safety monitoring and analysis system platform, characterized in that, The third-level data analysis in step 4 specifically includes: ① Traditional monitoring models include: statistical models, deterministic models, hybrid models, and monitoring models based on intelligent algorithms. Considering the actual properties of the project and the inversion of the second-level material parameters, a deterministic model is used in the third-level data analysis to establish the functional relationship between influencing factors and effect quantities. Displacement effect quantities are used as an example; under external loads, displacement occurs at any observation point on the dam and dam foundation. It can be divided into water pressure components. Temperature components And the aging component caused by creep and plastic deformation of concrete or soil materials. ,Right now: （1） Based on the inversion results of the second layer of material parameters, the water pressure component and temperature component are calculated using the established finite element model, and the aging component is calculated using a statistical model. ② The deterministic model of the displacement of the measuring point is expressed by the following formula: （2） In the formula: Calculate displacement values ​​for a deterministic model; , , , The coefficient is a constant; , , The coefficient is a constant; For finite element method calculation of water pressure component displacement; For finite element method calculation of temperature component displacement; It is 0.01 times the time t; The coefficient is a constant; The parameters in the model are obtained using the least squares method.

5. The application method of the multi-dimensional and multi-level analysis method according to claim 4 in the safety monitoring and analysis system platform of concrete dams, characterized in that, The fourth layer of data analysis in step 5 specifically includes: ① The reason why the short- and medium-term monitoring model has low accuracy in long-term prediction is mainly because the model has poor nonlinear simulation ability and lacks supervised learning ability. In order to ensure the long-term prediction accuracy of the model, the SVM algorithm commonly used in machine learning is introduced again in the fourth layer to establish a monitoring and prediction model. ② Based on the measured monitoring data, a third-layer deterministic model is introduced to obtain the following results: , and The estimated value , and ,right Introducing the SVM model, establishment time t and The regression prediction model between SVMs is improved by continuously adjusting the relaxation coefficient, penalty coefficient, and kernel function to enhance the model's prediction accuracy. Estimators in SVM models ; ③ Establish a medium- to long-term forecasting and monitoring model: （3） In the formula: and These are the model correction coefficients, For constant terms; In the model , The objective function, calculated using a genetic algorithm, is set as follows: Minimum is sufficient. Displacement is generated at any observation point of the dam or dam foundation.

Images