Modeling method of concrete dam displacement monitoring model using measured temperature
By using the SVM model to analyze and optimize the temperature monitoring data of concrete dams, the problem of difficulty in extracting information in massive data is solved, and the performance improvement of the concrete dam displacement monitoring model is achieved.
Patent Information
- Application Number
- CN202210628877.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-06
- Publication Date
- 2025-05-02
- Estimated Expiration
- 2042-06-06
AI Technical Summary
The prior art is difficult to effectively extract information from massive temperature monitoring data and establish accurate temperature factors, resulting in insufficient performance of concrete dam displacement monitoring models.
The support vector machine (SVM) model is used to establish the initial SVM model based on the initial subset of modeling factors, calculate the importance scores of each modeling factor, and gradually eliminate the modeling factor with the lowest importance score, and select a very small number of the most representative temperature measurement points to reduce the total number of modeling factors.
It realizes efficient extraction of effective information from massive temperature monitoring data, reduces the total number of modeling factors, improves the prediction accuracy of the concrete dam displacement monitoring model, and avoids changes and loss of feature information in the principal component method.
Smart Images
Figure CN114969937B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a modeling method, in particular to a modeling method of a concrete dam displacement monitoring model using actually measured temperature. Background Art
[0002] With the improvement of engineering technology, the height and scale of concrete dams are increasing. The Jinping I arch dam has a maximum height of 305m, making it the world's tallest concrete dam. However, the operation and safety management of ultra-high dams still faces many major problems and challenges. Structural health monitoring throughout the life cycle plays an indispensable role in ensuring its safe operation. One of the keys lies in the deep intelligent analysis of massive monitoring data based on mathematical models.
[0003] Among many monitoring effect quantities, displacement is the most intuitive reflection of the structural state of concrete dams. The most classic mathematical monitoring model is the HST model of water pressure-periodic temperature-time 3 causal relationship modeling factors. With the increase of dam height, the hysteresis effect of reservoir water depth and reservoir operation mode on the temperature field of concrete dam becomes more complicated, which makes it difficult for the harmonic function temperature factor in the HST model to accurately reflect the temperature deformation effect of ultra-high concrete dams. In this regard, using the measured temperature data of the dam body to establish the temperature factor is one of the effective methods to improve the performance of the displacement monitoring model of ultra-high concrete dams. However, due to the huge amount of temperature monitoring data, how to extract effective information from the massive temperature monitoring data to establish the temperature factor is an urgent problem to be solved. The temperature measurement points of ultra-high concrete dams are usually hundreds or even thousands. The existing temperature measurement point selection method lacks quantitative evaluation criteria and is mainly determined by engineers' experience or simple correlation analysis, which is greatly interfered by human subjectivity. The method of extracting the principal component of the measured temperature time series of all temperature measurement points of the dam body will lead to the change and loss of the original characteristic information.
[0004] The traditional modeling method of concrete dam monitoring model mainly regards the effect quantity as a linear explicit function of the relevant modeling factors, while engineering practice shows that the relationship between the two is actually a complex nonlinear relationship, and the influencing mechanism is also dynamically changing. In this regard, machine learning models, such as support vector machines (SVM) and artificial neural networks (ANN), have good nonlinear data mining capabilities and are therefore widely used in the field of dam safety monitoring. Summary of the invention
[0005] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a modeling method for a concrete dam displacement monitoring model using measured temperature. The model is established based on the time series of the measured temperature of the preferred concrete dam as the temperature deformation factor, and a very small number of the most representative temperature measuring points are selected from a large number of temperature measuring points. The total number of modeling factors is reduced as much as possible without affecting the prediction performance of the model.
[0006] In order to solve the above technical problems, the technical solution of the present invention is: a modeling method of a concrete dam displacement monitoring model using measured temperature, comprising:
[0007] Step S1: establishing an initial SVM model based on an initial modeling factor set; wherein the initial modeling factor set includes multiple modeling factors, and the multiple modeling factors include a water pressure component, a temperature component, and a time-effect component; wherein the temperature component adopts the measured temperature time series of all effective temperature measurement points of the corresponding dam body;
[0008] Step S2: Calculate the importance score of each modeling factor in the SVM model;
[0009] Step S3: After removing the modeling factor with the lowest importance score, all remaining modeling factors are used to form a new modeling factor set;
[0010] Step S4: Based on the new modeling factor set, a new SVM model is established, and it is determined whether there is only one modeling factor left in the new modeling factor set. If not, return to step S2;
[0011] Step S5: The minimum value of the prediction MSE of all SVM models is recorded as the minimum prediction MSE min , from the minimum prediction MSE min The corresponding SVM model is recursively extended to the later built SVM model to find the first occurrence prediction MSE ≥ (1 + preset threshold) * minimum prediction MSE min The SVM model is recorded as the first-time over-limit SVM model, and the data set composed of all measured temperature time series in the first-time over-limit SVM model is taken as the optimal measured temperature factor set;
[0012] Step S6: using the optimal measured temperature factor set to establish a final SVM model.
[0013] Furthermore, between step S5 and step S6, a measured temperature factor elimination step is also included, and the measured temperature factor elimination step includes:
[0014] The optimal measured temperature factor set is reduced and optimized; wherein the criterion of the reduction and optimization is:
[0015] When the two measured temperature time series process lines are basically overlapped and the modeling factor importance scores are similar, if the one with a smaller importance score is eliminated, the prediction MSE increase of the new SVM model is compared with the minimum prediction MSE min If it does not exceed the preset threshold, the one with the smaller importance score will be eliminated.
[0016] Furthermore, between the measured temperature factor elimination step and step S6, the following steps are further included:
[0017] The measured temperature time series where the optimal measured temperature factor is concentrated at the dam top and the downstream surface of the dam body are processed by wavelet denoising.
[0018] Further, in step S2, the importance score of each modeling factor in the SVM model is expressed as the output f(x i ) with respect to the input factor, the importance score of the rth modeling factor in the current SVM model is defined as:
[0019]
[0020] Where f(x) is the regression function of the current SVM model; n is the total number of training set samples; l is the total number of modeling factors of the current SVM model.
[0021] Further, in step S5, the data set consisting of all measured temperature time series in the first over-limit SVM model is used as the optimal measured temperature factor set, including:
[0022] If the first exceeded SVM model is relative to the minimum prediction MSE min For the corresponding SVM model, the number of modeling factor reductions is greater than the preset number, and the data set composed of all measured temperature time series in the first over-limit SVM model is taken as the optimal measured temperature factor set.
[0023] Furthermore, step S5 also includes:
[0024] If the first exceeded SVM model is relative to the minimum prediction MSE min The corresponding SVM model, the number of modeling factors reduced is not greater than the preset number, and the minimum prediction MSE min The data set composed of all measured temperature time series in the corresponding SVM model is used as the optimal measured temperature factor set.
[0025] Furthermore, in step S1, when the concrete dam is a concrete gravity dam, the corresponding dam body is the dam section where the displacement monitoring point is located; when the concrete dam is a concrete arch dam, the corresponding dam body is the arch crown beam dam section.
[0026] Further, in step S1 and step S4, the method for establishing the SVM model includes:
[0027] The daily values corresponding to the modeling factors are used as the input of the SVM model, and the daily measured displacement is used as the output of the SVM model;
[0028] At the same time point, the monitoring data time series of each modeling factor and measured displacement are divided into two sections, respectively, as a training set and a test set. The training set is used for modeling training of the corresponding SVM model, and the test set is used to calculate the prediction MSE of the corresponding SVM model; wherein,
[0029] The prediction MSE is defined as:
[0030]
[0031] In the formula, δ t and are the measured values of concrete dam displacement and the predicted values of SVM model respectively; k is the total number of samples in the test set.
[0032] Further, the use of the optimal measured temperature factor set to construct a final SVM model includes:
[0033] On the basis of using the optimal measured temperature factor set as the temperature component, the final SVM model is constructed by combining the water pressure component, the time effect component and other causal components of the concrete dam displacement.
[0034] After adopting the above technical scheme, the present invention takes the measured temperature time series of all effective temperature measuring points of the concrete dam body as the measured temperature factor in the displacement monitoring model, and uses the powerful nonlinear data mining ability of the SVM model to analyze the importance and gradually eliminate the initially selected modeling factors. Under the premise of not affecting the prediction performance of the model, the total number of modeling factors of the model is reduced as much as possible, thereby realizing the optimization of the measured temperature deformation factor of the dam body. The present invention selects a very small number of the most representative temperature measuring points from hundreds or even thousands of temperature measuring points of the concrete dam, realizes the efficient extraction of effective information from the massive temperature monitoring data of the dam body, greatly reduces the total number of modeling factors in the displacement monitoring model of the concrete dam established based on the measured temperature, and also solves the practical problem of lack of quantitative evaluation basis in the selection process of the measured temperature factor of the concrete dam; the measured temperature field characterization information extracted by the present invention is the original monitoring information, which avoids the problems of feature information change and loss caused by data fusion in the principal component method, and the present invention also overcomes the defect that the traditional harmonic function temperature factor is difficult to accurately reflect the temperature deformation effect of the ultra-high concrete dam. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 A flow chart of a modeling method according to an embodiment of the present invention;
[0036] Figure 2 This is the layout diagram of the arch dam vertical monitoring system;
[0037] Figure 3 This is the thermometer monitoring arrangement diagram for the arch dam crown beam section - 13th dam section;
[0038] Figure 4 Predict the MSE evolution law during the optimization process of the SVM model modeling factors for each displacement measurement point;
[0039] Figure 5 It is the temperature time series retained after the importance analysis of modeling factors at the displacement measurement point PL13-3;
[0040] Figure 6 This is the temperature time series retained after the displacement measurement point PL13-3 is manually removed;
[0041] Figure 7 The measured temperature time series finally selected for the displacement measurement point PL13-1;
[0042] Figure 8 The measured temperature time series finally selected for the PL13-2 displacement measurement point;
[0043] Fig. 9 The measured temperature time series finally selected for the PL13-4 displacement measurement point;
[0044] Fig.10 The measured temperature time series finally selected for the PL13-5 displacement measurement point;
[0045] Fig.11 This is the temperature time series of the T13-121 temperature measurement point before and after wavelet denoising;
[0046] Fig.12 It is the time series of the radial displacement monitoring value and SVM model fitting (prediction) value of the PL13-1 displacement measuring point;
[0047] Fig.13 It is the time series of the radial displacement monitoring value and SVM model fitting (prediction) value of the PL13-4 displacement measuring point;
[0048] Fig.14 is the prediction MSE of the SVM model. DETAILED DESCRIPTION
[0049] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments in conjunction with the accompanying drawings.
[0050] like Figure 1As shown, a modeling method of a concrete dam displacement monitoring model using measured temperature includes:
[0051] Step S1: establishing an initial SVM model based on an initial modeling factor set; wherein the initial modeling factor set includes multiple modeling factors, and the multiple modeling factors include water pressure component, temperature component, time effect component and other causal components; wherein the temperature component adopts the measured temperature time series of all effective temperature measurement points of the corresponding dam body;
[0052] Step S2: Calculate the importance score of each modeling factor in the SVM model;
[0053] Step S3: After removing the modeling factor with the lowest importance score, all remaining modeling factors are used to form a new modeling factor set;
[0054] Step S4: Based on the new modeling factor set, a new SVM model is established, and it is determined whether there is only one modeling factor left in the new modeling factor set. If not, return to step S2;
[0055] Step S5: The minimum value of the prediction MSE of all SVM models is recorded as the minimum prediction MSE min , from the minimum prediction MSE min The corresponding SVM model is recursively extended to the later built SVM model to find the first occurrence prediction MSE ≥ (1 + preset threshold) * minimum prediction MSE min The SVM model is recorded as the first-time over-limit SVM model. Based on the first-time over-limit SVM model, the optimal measured temperature factor set is obtained;
[0056] The optimal measured temperature factor set is reduced and optimized; wherein the criterion of the reduction and optimization is:
[0057] When the two measured temperature time series process lines are basically overlapped and the modeling factor importance scores are similar, if the one with a smaller importance score is eliminated, the prediction MSE increase of the new SVM model is compared with the minimum prediction MSE min If the value does not exceed the preset threshold, the one with the smaller importance score will be eliminated;
[0058] The measured temperature time series where the optimal measured temperature factor after reduction and optimization is concentrated on the dam top and the downstream surface of the dam body are processed by wavelet denoising.
[0059] Step S6: Using the optimal measured temperature factor set after denoising, establish the final SVM model.
[0060] Wherein, the modeling factor is a causal modeling factor.
[0061] Among them, in step S1, all the effective temperature measuring points of the corresponding dam body, for concrete gravity dam, are all the effective temperature measuring points of the dam section where the displacement monitoring point of the dam body is located; for concrete arch dam, the measured temperature of the arch crown beam dam section is the most representative, so only all the effective temperature monitoring points of the arch crown beam dam section can be selected. An effective temperature measuring point means that the integrity of the measured temperature time series of the measuring point is good, and a small number of missing values can be supplemented by conventional simple interpolation methods.
[0062] In step S2, the importance score of each modeling factor in the SVM model is expressed as the output f(x i ) with respect to the input factor, the importance score of the rth modeling factor in the current SVM model is defined as:
[0063]
[0064] Where f(x) is the regression function of the current SVM model; n is the total number of training set samples; l is the total number of modeling factors of the current SVM model.
[0065] Wherein, in step S1 and step S4, the method for establishing the SVM model includes:
[0066] The daily values corresponding to the modeling factors are used as the input of the SVM model, and the daily measured displacement is used as the output of the SVM model;
[0067] At the same time point, the monitoring data time series of each modeling factor and measured displacement are divided into two sections, respectively, as a training set and a test set. The training set is used for modeling training of the corresponding SVM model, and the test set is used to calculate the prediction MSE of the corresponding SVM model; wherein,
[0068] The prediction MSE is defined as:
[0069]
[0070] In the formula, δ t and are the measured values of concrete dam displacement and the predicted values of SVM model respectively; k is the total number of samples in the test set.
[0071] Among them, in step S5, the prediction MSE of the post-built SVM model is relative to the minimum prediction MSE min The maximum allowable increase, i.e. the preset threshold, can be set to 5% according to the commonly used accuracy requirements of engineering projects; if the relative prediction MSE of the SVM model exceeds the limit for the first time minIf the number of modeling factors of the corresponding SVM model is reduced by more than the preset number, the data set consisting of all measured temperature time series in the first over-limit SVM model is taken as the optimal measured temperature factor set. If the relative prediction MSE of the first over-limit SVM model is min The number of modeling factors corresponding to the SVM model is reduced to a small number and does not exceed the preset number. To improve the accuracy, the minimum prediction MSE can be directly min The corresponding SVM model is used as the optimal model in the process of automatic elimination of modeling factors. min The reason why the corresponding SVM model is recursively extended to the later SVM model is that the later the modeling factors are eliminated, the greater the impact on the prediction performance of the SVM model, and the greater the increase in the prediction MSE caused by the elimination. In this embodiment, the preset number is 5% of the total number of initial modeling factors, and can also be set to other numbers according to actual needs.
[0072] Among them, in step S5, the reason for reducing and optimizing the optimal measured temperature factor set is that: although all unimportant modeling factors before the SVM model prediction MSE increase does not exceed the preset threshold have been effectively eliminated in the modeling factor importance analysis process, there are still a small number of measured temperature factors among the retained modeling factors that have very close importance scores, and the time-varying laws of the corresponding measured temperature time series have great similarity.
[0073] Among them, in step S6, when establishing the final SVM model, in addition to the preferred measured temperature factor as the temperature component, the modeling factors must also include other causal components in the concrete dam displacement monitoring model, such as water pressure component, time effect component, etc.
[0074] The technical solution involved in the above embodiment is described below in conjunction with an optional embodiment.
[0075] Figure 1 It is a flow chart of the modeling method of the concrete dam displacement monitoring model using the measured temperature according to the present embodiment.
[0076] This embodiment is specifically: a concrete hyperbolic arch dam, with a maximum dam height of 305m, is the highest concrete dam built in the world. On December 23, 2013, the dam body concrete was fully poured to the dam top elevation of 1885m. On August 24, 2014, the water level was stored to the normal water level of 1880m. The reservoir water level has been cyclically circulating between the elevations of 1800 and 1880m. The dam body vertical line monitoring system is arranged as follows: Figure 2 shown.
[0077] The dam displacement and temperature monitoring data used in this study are from September 1, 2015 to December 31, 2018, with a sampling frequency of once a day. May 28, 2018 is used as the dividing point, and the two periods of data are used for SVM model training and prediction performance comparison analysis. A total of 5 dam displacement measurement points were selected, which are all the vertical measurement points on the No. 13 arch crown beam dam section. The displacement measurement points are numbered PL13-1, PL13-2, PL13-3, PL13-4 and PL13-5.
[0078] (1) Constructing the initial set of modeling factors
[0079] The arch dam has obvious viscoelastic hysteresis deformation effect during the stable period of high water level of 1880m per year. It is necessary to introduce the hysteresis component characterized by step exponential function on the basis of water pressure component, temperature component and time-effect component. The mathematical expression of the modeling factor in the newly established HETT model is:
[0080]
[0081] Where H is the water depth of the upstream reservoir on that day; T j is the measured temperature value of the jth dam body temperature measuring point on the day, N is the total number of dam body temperature measuring points selected; θ = t / 100, where t is the cumulative number of days from the displacement monitoring day to the start monitoring day; f(τ) is a step function, and its value is 1 when the upstream reservoir water level is in the stable period of 1880m high water level, otherwise it is 0; τ is the duration of the current reservoir water level stable period; a i 、b j , c1, c2, d1 and d2 are model regression coefficients; α=E K1 / η K1 , β=E K2 / η K2 , where E K1 and E K2 is the hysteresis elastic modulus of the dam concrete in the 5-parameter generalized Kelvin model, η K1 and η K2 are the viscosity coefficients, and this time we take α = 0.5283, β = 0.0052 respectively; the two hysteresis components are respectively δ Hv1 =f(τ)·[1-exp(-ατ)] and δ Hv2 =f(τ)·[1-exp(-βτ)] sign.
[0082] In order to compare with the traditional harmonic function temperature factor, the HEST model based on the harmonic function temperature factor is established in this embodiment. The difference between the HEST model and the HETT model is only the temperature factor. The mathematical expression is:
[0083]
[0084] Where b 1i and b 2i is the regression coefficient of the temperature factor of the harmonic function; other symbols have the same meaning as in the mathematical expression of the HETT model.
[0085] The arch-crowned beam dam section is dam section 13, so all effective temperature measurement points of dam section 13 are selected as the initial temperature modeling factor set of temperature deformation. Its arrangement in the dam body is as follows: Figure 3 As shown, except for the three elevation layers at the bottom, five thermometers are buried in each elevation layer along the center line of the dam section, two of which are buried 5 cm inside the upstream and downstream surfaces of the dam body, and the remaining three are arranged at equal intervals between the two thermometers.
[0086] In summary, the initial SVM model has a total of 148 modeling factors, including 4 water pressure components, 2 time-dependent components, 2 lag components, and 140 measured temperature time series. The measured temperature time series is also the measured temperature factor.
[0087] (2) Importance scoring and automatic optimization of measured temperature factors
[0088] First, an initial SVM model containing 148 modeling factors was established for each displacement measurement point, m = 1, where m is the model number. After importance scoring, the modeling factors with the lowest importance scores in the initial SVM models of the displacement measurement points PL13-1, PL13-2, PL13-3, PL13-4 and PL13-5 are T13-14, T13-39, T13-8, T13-9 and T13-59, respectively. These measured temperature factors can be obtained from the input of their initial SVM models. The input factors were eliminated, and an SVM model containing 147 modeling factors was established for each displacement measurement point, m = 2. After the importance of the modeling factors was scored again, the factors with the lowest importance scores among the 147 modeling factors were T13-74, T13-8, T13-133, T13-73 and T13-99. After eliminating them, an SVM model containing 146 modeling factors was established, m = 3, and so on, until the last SVM model containing only one modeling factor, m = 148. In this embodiment, the prediction MSE of the post-built SVM model is relative to the MSE min The maximum allowable increase can be set to 5% according to the commonly used accuracy requirements in engineering.
[0089] Taking the displacement measurement point PL13-1 as an example, the prediction MSE of the 135th SVM model is the smallest. min The prediction MSE of the 135th to 148th SVM models increases successively, and the prediction MSE of the 136th SVM model is 0.0268, which is greater than (1+0.05)*minimum prediction MSE.min Therefore, it can be determined that in the process of automatic elimination of modeling factors, the data set composed of all measured temperature time series in the 135th SVM model of the PL13-1 measuring point is the optimal measured temperature factor set.
[0090] For the PL13-5 displacement measurement point, the prediction MSE of the 129th SVM model is the smallest. min The corresponding SVM model is recursively derived from the subsequent SVM model. The first over-limit SVM model number is 131, and the number of factors of the two models differs by only 2. In order to improve the model accuracy without adding too many modeling factors, the data set composed of all measured temperature time series in the 129th SVM model during the automatic elimination of modeling factors is directly used as the optimal measured temperature factor set.
[0091] During the automatic elimination of modeling factors at each displacement measurement point, the evolution law of the predicted MSE of the SVM model is as follows: Figure 4 As shown in Table 1, the best SVM model number, that is, the model number where the optimal measured temperature factor set is located, is shown in Table 1.
[0092] Table 1 The best SVM model for each displacement point in the process of modeling factor importance analysis
[0093]
[0094] (3) Artificial elimination of a small number of measured temperature factors
[0095] The importance analysis of modeling factors based on SVM has effectively eliminated all unimportant modeling factors before the predicted MSE increase does not exceed the 5% threshold. However, among the modeling factors retained, there are still a small number of measured temperature time series with very similar importance scores. The reason is that the evolution laws of the two are basically the same. Therefore, it is necessary to combine the manual elimination method and perform reduction and optimization again. Taking the PL13-3 displacement measurement point as an example, Figure 5 The measured temperature time series retained after the modeling factor importance analysis of the displacement measuring point can be found. Among the selected temperature measuring points, the temperature time series curves of T13-146 and T13-139, T13-160 and T13-158, T13-65 and T13-70, T13-15 and T13-38 are close to each other. After excluding T13-146, T13-160, T13-65 and T13-15, the predicted MSE increase of the new SVM model did not exceed the 5% threshold, and even became smaller. Therefore, these four measured temperature factors can be effectively eliminated. The results are shown in Figure 6Similarly, after the other displacement measuring points were manually eliminated using the same method, the modeling factors that were finally retained are shown in Table 2. The time series of the dam body temperature measuring points selected for the displacement measuring points PL13-1, PL13-2, PL13-4 and PL13-5 are shown in Table 2. Figure 7 , 8 , 9, and 10.
[0096] Table 2 Modeling factors retained after importance analysis and manual elimination of each displacement measurement point
[0097]
[0098] There are a total of 140 initial measured temperature factors. After optimization by the method of this embodiment, the numbers of measured temperature factors finally selected for the displacement measurement points PL13-1, PL13-2, PL13-3, PL13-4 and PL13-5 are 8, 7, 11, 13 and 11, respectively. The average selection rate of the 140 initial measured temperature factors is 7.1%, achieving efficient compression of massive temperature monitoring data.
[0099] (4) Wavelet denoising of measured temperature deformation factors
[0100] The temperature measurement points located on the downstream surface and the top of the dam are directly affected by solar radiation, and the measured values fluctuate greatly. If they are directly used for SVM modeling, the prediction performance of the model will be seriously affected. Therefore, wavelet denoising should be performed on them first. Taking the T13-121 temperature measurement point located on the downstream surface of the dam as an example, Fig.11 This is the temperature time series of the temperature measuring point before and after wavelet denoising. After removing a few high-frequency components with large fluctuations, the resulting temperature time series process line is relatively smooth and still maintains the original time-varying law.
[0101] (5) Build the final SVM model
[0102] In the process of modeling factor importance analysis, a small number of non-temperature factors were eliminated, but the order of elimination was relatively late, and the purpose of modeling factor importance analysis was to optimize the measured temperature factors. Therefore, when establishing the final SVM model, these eliminated non-temperature factors were still retained in the HETT model, that is, H and H in the mathematical expression of the HETT model were retained. 2 , H 3 , H 4 ,θ,lnθ,δ Hv1 and δ Hv2 ; Among them, H, H 2 , H 3 and H 4 are the four water pressure components, θ and lnθ are the two time-dependent components, δ Hv1 and δ Hv2There are two hysteresis components. The difference between the HETT model and the HEST model used for comparison is only that the temperature factors of the two are different, which are the preferred measured temperature factor of the present invention and the traditional harmonic function temperature factor.
[0103] (6) Model prediction performance evaluation
[0104] In order to compare with the traditional harmonic function temperature factor mathematical monitoring model, the optimal measured temperature factor and harmonic function temperature factor are used as temperature components, combined with water pressure component, hysteresis component and aging component, and then SVM is used to model them, which are marked as HETT and HEST models respectively. The time series of the measured value of radial displacement of PL13-1 displacement measuring point and the SVM model fitting (prediction) value are shown in Figure 2. Fig.12 As shown in the figure, the time series of the measured radial displacement value and the SVM model fitting (prediction) value of the PL13-4 displacement measuring point are as follows: Fig.13 The results obtained at other displacement measurement points are similar. Fig.14 As shown in the figure, overall, the displacement prediction value of the HETT model at each displacement measuring point is closer to the monitoring value, and the average prediction MSE of the five displacement measuring points in the HETT model is only 62.7% of that of the HEST model, indicating that compared with the traditional harmonic function temperature factor, the measured temperature factor of the dam body preferably obtained by the present invention can effectively improve the prediction accuracy of the concrete dam displacement monitoring model, verifying the rationality and scientificity of the present invention.
[0105] The above embodiments are only used to explain the technical solutions of the present invention and verify the efficiency and rationality of the present invention. The embodiments are illustrative rather than restrictive. Any modifications and substitutions made by ordinary technicians in the field under the guidance of the present invention without departing from the purpose and scope of the present invention should be included in the scope of the claims of the present invention.
Claims
1. A modeling method for a concrete dam displacement monitoring model using measured temperature, characterized in that: include: Step S1: establishing an initial SVM model based on an initial modeling factor set; wherein the initial modeling factor set includes multiple modeling factors, and the multiple modeling factors include a water pressure component, a temperature component, and a time-effect component; wherein the temperature component adopts the measured temperature time series of all effective temperature measurement points of the corresponding dam body; Step S2: Calculate the importance score of each modeling factor in the SVM model; Step S3: After removing the modeling factor with the lowest importance score, all remaining modeling factors are used to form a new modeling factor set; Step S4: Based on the new modeling factor set, a new SVM model is established, and it is determined whether there is only one modeling factor left in the new modeling factor set. If not, return to step S2; Step S5: The minimum value of the prediction MSE of all SVM models is recorded as the minimum prediction MSE min , from the minimum prediction MSE min The corresponding SVM model is recursively extended to the later built SVM model to find the first occurrence prediction MSE ≥ (1 + preset threshold) * minimum prediction MSE min The SVM model is recorded as the first-time over-limit SVM model, and the data set composed of all measured temperature time series in the first-time over-limit SVM model is taken as the optimal measured temperature factor set; Step S6: using the optimal measured temperature factor set to establish a final SVM model; In step S2, the importance score of each modeling factor in the SVM model is expressed as the output of the SVM model The partial derivative of the input factor, the first The importance score of a modeling factor is defined as: ； In the formula, is the regression function of the current SVM model; is the total number of samples in the training set; is the total number of modeling factors of the current SVM model; In step S1 and step S4, the method for establishing the SVM model includes: The daily values corresponding to the modeling factors are used as the input of the SVM model, and the daily measured displacement is used as the output of the SVM model; At the same time point, the monitoring data time series of each modeling factor and measured displacement are divided into two sections, respectively, as a training set and a test set. The training set is used for modeling training of the corresponding SVM model, and the test set is used to calculate the prediction MSE of the corresponding SVM model; wherein, The prediction MSE is defined as: ； In the formula, and are the measured value of concrete dam displacement and the predicted value of SVM model respectively; is the total number of samples in the test set; The mathematical expression of the modeling factor in the SVM model is: ； In the formula, is the water depth of the upstream reservoir on that day; For the selected The actual temperature value of each dam body temperature measuring point on the day, is the total number of dam body temperature measurement points selected; ,in It is the cumulative number of days from the displacement monitoring day to the starting monitoring day; is a step function, is the duration of the current reservoir water level stable period; , , , , and is the model regression coefficient; , ,in and is the hysteresis elastic modulus of the dam concrete in the 5-parameter generalized Kelvin model, and are the viscosity coefficients, and the two hysteresis components are expressed as and mark.
2. The modeling method of the concrete dam displacement monitoring model using measured temperature according to claim 1 is characterized in that: Between step S5 and step S6, there is also a step of eliminating the measured temperature factor, and the step of eliminating the measured temperature factor includes: The optimal measured temperature factor set is reduced and optimized; wherein the criterion of the reduction and optimization is: When the two measured temperature time series process lines are close to overlap and the modeling factor importance scores are similar, if the one with a smaller importance score is eliminated, the prediction MSE increase of the new SVM model is compared with the minimum prediction MSE min If it does not exceed the preset threshold, the one with the smaller importance score will be eliminated.
3. The modeling method of the concrete dam displacement monitoring model using the measured temperature according to claim 2 is characterized in that: Between the measured temperature factor elimination step and step S6, the method further includes: The measured temperature time series where the optimal measured temperature factor is concentrated at the dam top and the downstream surface of the dam body are processed by wavelet denoising.
4. The modeling method of the concrete dam displacement monitoring model using measured temperature according to claim 1, characterized in that: In step S5, the data set consisting of all measured temperature time series in the first over-limit SVM model is used as the optimal measured temperature factor set, including: If the first exceeded SVM model is relative to the minimum prediction MSE min For the corresponding SVM model, the number of modeling factor reductions is greater than the preset number, and the data set composed of all measured temperature time series in the first over-limit SVM model is taken as the optimal measured temperature factor set.
5. The modeling method of the concrete dam displacement monitoring model using measured temperature according to claim 4 is characterized in that: Step S5 also includes: If the first exceeded SVM model is relative to the minimum prediction MSE min The corresponding SVM model, the number of modeling factors reduced is not greater than the preset number, and the minimum prediction MSE min The data set composed of all measured temperature time series in the corresponding SVM model is used as the optimal measured temperature factor set.
6. The modeling method of the concrete dam displacement monitoring model using measured temperature according to claim 1, characterized in that: In step S1, when the concrete dam is a concrete gravity dam, the corresponding dam body is the dam section where the displacement monitoring point is located; when the concrete dam is a concrete arch dam, the corresponding dam body is the arch crown beam dam section.
7. The modeling method of the concrete dam displacement monitoring model using measured temperature according to claim 1, characterized in that: The method of using the optimal measured temperature factor set to construct a final SVM model includes: On the basis of using the optimal measured temperature factor set as the temperature component, the water pressure component and the aging component are combined to construct the final SVM model.