Flood embankment deformation monitoring method based on integrated model
By constructing an integrated model for monitoring flood control dike deformation based on water pressure, temperature, and time-dependent components, and using random sampling to screen and calculate the probability distribution and weights of sub-model parameters, the overfitting problem caused by unreasonable model weights was solved, thereby improving the accuracy and generalization ability of flood control dike deformation monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SICHUAN UNIV
- Filing Date
- 2023-11-01
- Publication Date
- 2026-06-19
AI Technical Summary
In existing methods for monitoring the deformation of flood control dikes, the ensemble model's predictive performance is reduced and overfitting occurs due to unreasonable weighting of the model.
The probability distribution and ensemble weights of the deformation monitoring sub-model parameters are updated by random sampling. By constructing multiple sub-models based on water pressure, temperature and aging components, error indices are calculated using particle parameters and weights are assigned reasonable weights to establish an ensemble model.
It improves the model's generalization ability, mitigates the overfitting problem of traditional ensemble models, reduces the monitoring misjudgment rate, and achieves more accurate monitoring of flood control dike deformation.
Smart Images
Figure CN117494510B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of dam deformation monitoring, and specifically to a method for monitoring the deformation of flood control dikes based on an integrated model. Background Technology
[0002] Flood control dike deformation is characterized by its ease of monitoring and high accuracy, making it the most direct and reliable reflection of the dike's operational status. Establishing a robust and reliable monitoring and early warning model based on its deformation monitoring data is one of the important means to grasp the operational status of flood control dikes in real time and reduce operational risks. Currently, flood control dike deformation monitoring often uses statistical models and hybrid models. Both models assume that flood control dike deformation consists of water pressure, temperature, and time-dependent components, and that each component exists in multiple different forms.
[0003] The statistical model is based entirely on historical monitoring data, while the hybrid model combines the statistical model with finite element method calculations to construct the water pressure components. The parameters of each model are often calculated using different mathematical methods such as multiple linear regression and stepwise regression. Different forms of deformation monitoring models built using different mathematical tools also exhibit certain differences in data information mining performance. To fully leverage the advantages of each sub-model, establishing an integrated monitoring model by reasonably weighting the sub-models is an effective approach.
[0004] Currently, methods such as genetic algorithms and extreme learning machines are often used to determine the weights of each sub-model by minimizing the fitting error of the ensemble model. However, these methods are prone to assigning unreasonable weights to the sub-models, leading to more severe overfitting in the ensemble model and reduced prediction performance. Summary of the Invention
[0005] To address the aforementioned shortcomings in existing technologies, the flood control dike deformation monitoring method based on an integrated model provided by this invention solves the problem that the model prediction performance is reduced due to unreasonable weight assignment to the model in existing methods.
[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:
[0007] A method for monitoring the deformation of flood control dikes based on an integrated model is provided, characterized by the following steps:
[0008] S1. Obtain multiple deformation monitoring sub-models of the flood control dike when different temperature and pressure components, water pressure components, and failure aging components are selected;
[0009] S2. Obtain the measured deformation monitoring dataset, and update the probability distribution p(θ|data) of the deformation monitoring sub-model parameters and the ensemble weights W(Mj|data) of the deformation monitoring sub-model using random sampling:
[0010]
[0011]
[0012] Where, p ε (θ|data) represents the probability distribution of model parameters calculated using a population that satisfies the population error threshold ε; data represents the measured deformation monitoring dataset; data * The dataset is used to fit the model; ε is the population error threshold; Δ(data, data) * ) represents the error index; M t M represents any deformation monitoring sub-model in the population; j For the target sub-model; θ (t) The particle parameters θ are required to meet the error specifications. * represents the randomly sampled particle parameters; T represents the total number of particle parameters in the population; I(·) represents the indicator function;
[0013] S3. Construct an integrated model for flood control dike deformation monitoring based on the integration weights of the deformation monitoring sub-models.
[0014]
[0015] in, These are the predicted values from the sub-model;
[0016] S4. Obtain flood control embankment deformation monitoring data and calculate flood control embankment deformation using the flood control embankment deformation monitoring integrated model. And calculate the monitoring and early warning interval. Both r1 and r2 are quantile values;
[0017] S5. Determine whether the deformation monitoring data of the flood control embankment is within the monitoring and early warning range. If so, the flood control embankment is operating normally; otherwise, issue an early warning of abnormal operation of the flood control embankment.
[0018] The beneficial effects of this invention are as follows: This solution calculates the ensemble weights of the sub-models and the probability distribution of each model parameter by randomly sampling and screening the sub-models and their parameters. This method avoids assigning high weights to sub-models with high fitting accuracy but poor prediction performance, and can assign reasonable weights to each sub-model to give full play to the advantages of the sub-models. The ensemble model established by this method has better overall performance than each sub-model and traditional ensemble models.
[0019] Furthermore, step S2 further includes:
[0020] S21. Initialize the total number of particle parameters, population error threshold, error index, parameter perturbation kernel k(·), initial weights, initial parameter probability distributions, and convergence conditions for each deformation monitoring sub-model.
[0021] S22. Select the deformation monitoring sub-model according to the initial weight of the deformation monitoring sub-model, and determine whether the current population is the first population. If yes, proceed to step S23; otherwise, proceed to step S24.
[0022] S23. Generate particle parameters based on the initial probability distribution of parameters, the measured deformation monitoring dataset, and the selected deformation monitoring sub-model, and then proceed to step S25.
[0023] S24. Based on the selected deformation monitoring sub-model, the previous generation population and the parameter perturbation kernel, the particle parameters are generated using formula (1), and then proceed to step S25.
[0024]
[0025] in, Let be the parameter of the i-th particle in population p; Let be the parameter of the j-th particle in population p-1; Let L be the parameter of the t-th particle in population p-1; uniform is the uniform distribution function; p is the population generation; L p-1 The variation of the parameter value of population p-1;
[0026] S25. Based on the selected deformation monitoring sub-model and the generated particle parameters, calculate the model fitting data, and based on the measured deformation monitoring data and the model fitting data, calculate the error index Δ(data,data). * ):
[0027]
[0028] Among them, y i This is actual deformation monitoring data; The model fit value is n; n is the total number of measured deformation monitoring data.
[0029] S26. Determine whether the error index meets the preset error. If yes, proceed to step S27; otherwise, delete the particle parameters generated by the selected deformation monitoring sub-model and return to step S22.
[0030] S27. Accept the particle parameters, assign them weights using formula (2), and increment the number of particle parameters by 1.
[0031]
[0032] in, The weight of the parameter of the i-th particle in the p-th generation population; for Initial probability distribution of parameters; exp(·) is an exponential function; ε p is the population error threshold for the p-th generation population;
[0033] S28. Determine whether the number of particle parameters in the current population has reached the total number of particle parameters in the population. If yes, proceed to step S29; otherwise, return to step S22.
[0034] S29. Normalize the weights of the particle parameters so that the sum of the weights of the parameters of each particle in the current population is 1.
[0035] S210. Determine whether the relative error of the current population meets the convergence condition. If yes, proceed to step S211; otherwise, increment the population size by 1 and return to step S22.
[0036] S211. Calculate the model parameter probability distribution p(θ|data) and the ensemble weight W(M) of each deformation monitoring sub-model using the last generation population. j |data).
[0037] The further effects of the above technical solution are as follows: by estimating the probability distribution of parameters of each sub-model and the ensemble weight of the sub-model through random sampling and filtering, and establishing an ensemble model based on the sub-model weights, the advantages of each sub-model are fully utilized, the generalization ability of the model is improved, the overfitting problem of traditional ensemble models is mitigated, and the misjudgment rate of ensemble model monitoring is reduced.
[0038] Furthermore, the deformation monitoring sub-model is constructed using water pressure components, temperature components, and aging components:
[0039] δ=δ H +δ T +δ θ +C
[0040] Where, δ H δ T δ θ These are the water pressure component, temperature component, and aging component, respectively; C is a constant term.
[0041] The water pressure components include different forms of models:
[0042]
[0043] Among them, a l The coefficients for the l-th water pressure component; H, H min H max These represent the measured water level, minimum water level, and historical maximum water level of the river channel, respectively; EL0 represents the elevation of the foundation surface of the flood control dike.
[0044] The temperature components include different forms of models:
[0045]
[0046]
[0047] Among them, b o PC is the coefficient of the 0th temperature component. o The 0th principal component extracted using principal component analysis; m is the total number of principal components; b1 to b4 are the regression coefficients of the temperature components; s is the temperature factor; q is the date corresponding to the temperature measurement; q0 is January 1st of the year in which the measurement started.
[0048] The time-sensitivity component includes different forms of models:
[0049] δ θ =c1s′+c2 exp(-s′), s′=(qq′0) / 100
[0050] δ θ =c1s′+c2 ln(s′), s′=(qq′0) / 100
[0051] Where c1 and c2 are both time-dependent coefficients; s′ is the time-dependent factor; and q′0 is the start date of the test.
[0052] The further advantages of the above technical solution are as follows: The water pressure component in this solution can correct the error between the material design parameters and the actual parameters by setting the water pressure component coefficient. Principal component analysis is used to select the temperature principal components, which significantly reduces model complexity while ensuring high data information entropy. When measured temperature data of the embankment body is difficult to obtain and the operation period is long, the dam body temperature can be considered to change periodically over time. In this case, periodic trigonometric functions are often used to establish the temperature component of the flood control embankment deformation monitoring model.
[0053] Furthermore, the methods for obtaining quantile values r1 and r2 include:
[0054] Obtain the original deformation observation data δ′(t) of the flood control embankment within the past preset time period, and use the integrated model for flood control embankment deformation monitoring to calculate the fitted value sequence δ(t) corresponding to the original deformation observation data δ′(t);
[0055] Based on the original observed data δ′(t) and the fitted value sequence δ(t), the model residual sequence r(t) is calculated as δ′(t) - δ(t);
[0056] The residual normal distribution f(r) was estimated using the kernel density function in MATLAB software, and the quantile values r1 and r2 corresponding to f(r1) and f(r2) were obtained according to the preset confidence level.
[0057] Furthermore, the preset confidence level is 99%, f(r1) = 0.5%, and f(r2) = 99.5%.
[0058] The further effect of the above technical solution is that the sub-location can be accurately obtained through the above method, so as to ensure that the established model can give a clear monitoring and early warning interval. This monitoring and early warning interval can detect newly collected data in real time and detect dam behavior anomalies in a timely manner. Attached Figure Description
[0059] Figure 1 This is a flowchart of a flood control dike deformation monitoring method based on an integrated model.
[0060] Figure 2 This is a flowchart for calculating the probability distribution of model parameters of multiple deformation monitoring sub-models and the integrated weights of the deformation monitoring sub-models.
[0061] Figure 3 The horizontal displacement process line of the top of the 9th section of a gravity concrete flood control dike.
[0062] Figure 4 River water level and ambient temperature monitoring data process line.
[0063] Figure 5 Two-dimensional finite element model of section 9 of a gravity-type flood control dike for a certain project.
[0064] Figure 6 Integrated weights of representative populations in the monitoring model of deformation of the top of the flood control dike at section 9#.
[0065] Figure 7 Model M6 in the final population (1) The probability distribution of each parameter.
[0066] Figure 8 Integrated Model M9 (1) Model residual distribution diagram.
[0067] Figure 9 Model ensemble weights were calculated using the method proposed in this invention (a), Markov chain Monte Carlo (b), and multiple linear regression-genetic algorithm (c), respectively.
[0068] Figure 10 Integrated Model M9 (1) The integrated model covers 99% of the monitoring range.
[0069] Figure 11 Integrated Model M9 (2) The integrated model covers 99% of the monitoring range.
[0070] Figure 12 Integrated Model M9 (3) The integrated model covers 99% of the monitoring range. Detailed Implementation
[0071] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0072] refer to Figure 1 , Figure 1 A flowchart of a flood control dike deformation monitoring method based on an integrated model is shown; for example... Figure 1 As shown, the method includes steps S1 to S5.
[0073] In step S1, multiple deformation monitoring sub-models of the flood control dike are constructed when different temperature and pressure components, water pressure components, and failure aging components are selected.
[0074] In implementation, this scheme preferably uses a deformation monitoring sub-model constructed from water pressure, temperature, and aging components:
[0075] δ=δ H +δ T +δ θ +C
[0076] Where, δ H δ T δ θ These are the water pressure component, temperature component, and aging component, respectively; C is a constant term.
[0077] The water pressure components include different forms of models:
[0078]
[0079] Among them, a l The coefficients for the l-th water pressure component; H, H min H max These represent the measured water level, minimum water level, and historical maximum water level of the river channel, respectively; EL0 represents the elevation of the foundation surface of the flood control dike.
[0080] The temperature components include different forms of models:
[0081]
[0082]
[0083] Among them, b o PC is the coefficient of the 0th temperature component. oThe 0th principal component extracted using principal component analysis; m is the total number of principal components; b1 to b4 are the regression coefficients of the temperature components; s is the temperature factor; q is the date corresponding to the temperature measurement; q0 is January 1st of the year in which the measurement started.
[0084] The time-sensitivity component includes different forms of models:
[0085] δ θ =c1s′+c2 exp(-s′), s′=(qq′0) / 100
[0086] δ θ =c1 s′+c2 ln(s′), s′=(qq′0) / 100
[0087] Where c1 and c2 are both time-dependent coefficients; s′ is the time-dependent factor; and q′0 is the start date of the test.
[0088] In step S2, the measured deformation monitoring dataset is obtained, and the probability distribution p(θ|data) of the deformation monitoring sub-model parameters and the ensemble weight W(M) of the deformation monitoring sub-model are updated using random sampling. j |data):
[0089]
[0090]
[0091] Where, p ε (θ|data) represents the probability distribution of model parameters calculated using a population that satisfies the population error threshold ε; data represents the measured deformation monitoring dataset; data * The dataset is used to fit the model; ε is the population error threshold; Δ(data, data) * ) represents the error index; M t M represents any deformation monitoring sub-model in the population; j For the target sub-model; θ (t) The particle parameters θ are required to meet the error specifications. * represents the randomly sampled particle parameters; T represents the total number of particle parameters in the population; I(·) represents the indicator function;
[0092] like Figure 2 As shown, in one embodiment of the present invention, step S2 further includes:
[0093] S21. Initialize the total number of particle parameters, population error threshold, error index, parameter perturbation kernel k(·), initial weights, initial parameter probability distributions, and convergence conditions for each deformation monitoring sub-model.
[0094] S22. Select the deformation monitoring sub-model according to the initial weight of the deformation monitoring sub-model, and determine whether the current population is the first population. If yes, proceed to step S23; otherwise, proceed to step S24.
[0095] S23. Generate particle parameters based on the initial probability distribution of parameters, the measured deformation monitoring dataset, and the selected deformation monitoring sub-model, and then proceed to step S25.
[0096] S24. Based on the selected deformation monitoring sub-model, the previous generation population and the parameter perturbation kernel, the particle parameters are generated using formula (1), and then proceed to step S25.
[0097]
[0098] in, Let be the parameter of the i-th particle in population p; Let be the parameter of the j-th particle in population p-1; Let L be the parameter of the t-th particle in population p-1; uniform is the uniform distribution function; p is the population generation; L p-1 The variation of the parameter value of population p-1;
[0099] S25. Based on the selected deformation monitoring sub-model and the generated particle parameters, calculate the model fitting data, and based on the measured deformation monitoring data and the model fitting data, calculate the error index Δ(data,data). * ):
[0100]
[0101] Among them, y i This is actual deformation monitoring data; The model fit value is n; n is the total number of measured deformation monitoring data.
[0102] S26. Determine whether the error index meets the preset error. If yes, proceed to step S27; otherwise, delete the particle parameters generated by the selected deformation monitoring sub-model and return to step S22.
[0103] S27. Accept the particle parameters, assign them weights using formula (2), and increment the number of particle parameters by 1.
[0104]
[0105] in, The weight of the parameter of the i-th particle in the p-th generation population; for Initial probability distribution of parameters; exp(·) is an exponential function; ε p is the population error threshold for the p-th generation population;
[0106] S28. Determine whether the number of particle parameters in the current population has reached the total number of particle parameters in the population. If yes, proceed to step S29; otherwise, return to step S22.
[0107] S29. Normalize the weights of the particle parameters so that the sum of the weights of the parameters of each particle in the current population is 1.
[0108] S210. Determine whether the relative error of the current population meets the convergence condition. If yes, proceed to step S211; otherwise, increment the population size by 1 and return to step S22.
[0109] S211. Calculate the model parameter probability distribution p(θ|data) and the ensemble weight W(M) of each deformation monitoring sub-model using the last generation population. j |data).
[0110] In step S3, an integrated model for monitoring the deformation of the flood control dike is constructed based on the integrated weights of the deformation monitoring sub-model.
[0111]
[0112] in, These are the predicted values from the sub-model;
[0113] In step S4, flood control embankment deformation monitoring data is acquired, and the flood control embankment deformation is calculated using the integrated flood control embankment deformation monitoring model. And calculate the monitoring and early warning interval. Both r1 and r2 are quantile values.
[0114] In one embodiment of the present invention, the method for obtaining quantile values r1 and r2 includes:
[0115] Obtain the original deformation observation data δ′(t) of the flood control embankment within the past preset time period, and use the integrated model for flood control embankment deformation monitoring to calculate the fitted value sequence δ(t) corresponding to the original deformation observation data δ′(t);
[0116] Based on the original observed data δ′(t) and the fitted value sequence δ(t), the model residual sequence r(t) is calculated as δ′(t) - δ(t);
[0117] The residual normal distribution f(r) was estimated using the kernel density function in MATLAB software, and the quantile values r1 and r2 corresponding to f(r1) and f(r2) were obtained according to the preset confidence level.
[0118] The pre-set reliability was 99%, f(r1) = 0.5%, and f(r2) = 99.5%.
[0119] In step S5, it is determined whether the deformation monitoring data of the flood control embankment is within the monitoring and early warning range. If so, the flood control embankment is operating normally; otherwise, an early warning of abnormal operation of the flood control embankment is issued.
[0120] The effectiveness of the flood control embankment deformation monitoring method in this scheme is described below using specific examples:
[0121] Taking a gravity concrete flood control dike with a height of about 14m in a flood control area as an example, and taking the top monitoring point of its No. 9 monitoring section as the research object, the observed displacement of the dike top from May 1, 2018 to August 1, 2020 was collected. The integrated model of flood control dike deformation monitoring was constructed by the method proposed in this scheme, and compared with the calculation results by traditional genetic algorithm and Markov chain Monte Carlo.
[0122] The horizontal displacement process line corresponding to the top monitoring point of monitoring section #9 is as follows: Figure 3 As shown, the process lines of the river water level and ambient temperature along the flood control dike are as follows: Figure 4 As shown in the figure, due to fluctuations in ambient temperature and reservoir water level, the displacement of this measuring point exhibits periodic changes within an annual cycle, with relatively small daily fluctuations. The maximum and minimum measured displacement values are 6.81 mm and 0.18 mm, respectively.
[0123] This scheme is used to establish an integrated model for monitoring the deformation of the flood control dike at section 9#, and the specific steps are as follows:
[0124] (1) Construction of the sub-model for monitoring the deformation of flood control dikes
[0125] 1) Construction of water pressure components. The water pressure components were established using the finite element method. Based on the foundation rock type of the flood control dike, the concrete of the dike body, and the engineering geological conditions, a linear elastic constitutive model was selected to establish a two-dimensional finite element model of the No. 9 monitoring section of the flood control dike. The elastic modulus, density, and Poisson's ratio of each partition of the model are shown in Table 1.
[0126] Table 1 Design parameters of the levee body and foundation at section 9 of the gravity flood control dike.
[0127]
[0128]
[0129] The finite element model has the X-axis along the river, the Y-axis vertical, and the Z-axis transverse to the river. A two-dimensional finite element model of the flood control dike is shown below. Figure 5 As shown. Based on the calculation results of the finite element numerical simulation model under different river water level conditions, the water pressure components of the hybrid model can be established:
[0130] δ H =4.02×10 -4 h-8.74×10 -6 h2 +4.95×10 -8 h 3
[0131] Considering that there may be some discrepancies between the design value of the dam material and the actual parameter value of the project, when using this water pressure component and other components to build a monitoring model, it is necessary to multiply this item by an adjustment coefficient α to correct the deviation between the parameter design value and the actual value.
[0132] 2) Temperature component construction. The temperature components of the flood control dike were constructed using two methods: periodic trigonometric functions and principal component analysis. The principal component analysis was performed by extracting the principal components of four thermometers (T4, T10, T17, and T18) on the backwater side of the dike. The principal component analysis results of the four thermometers are shown in Table 2.
[0133] Table 2. Principal component analysis results for thermometers T4, T10, T17, and T18.
[0134]
[0135] As shown in Table 2, the cumulative contribution rate of the two principal components, PC1 and PC2, reaches 99.42%. Therefore, PC1 and PC2 can be used to construct the temperature component of the concrete flood control dike deformation monitoring model.
[0136] 3) Construction of the time-sensitivity component. The time-sensitivity component is constructed using exponential and logarithmic forms of time-sensitivity.
[0137] (2) Calculate the probability distribution of sub-model parameters and the integrated weight of deformation monitoring sub-model.
[0138] Eight different gravity-based concrete flood control dike deformation monitoring sub-models can be established by combining different forms of water pressure, temperature and time components. The forms of each sub-model are shown in Table 3.
[0139] Table 3 Deformation monitoring model forms
[0140]
[0141]
[0142] The deformation monitoring sub-model M is calculated using random sampling and screening. j The probability distribution of the parameters p(θ|data) and the sub-model M j The integrated weight W(M) j |data). Here, the total number of particle parameters in the population is T = 1000, and the error threshold ε for the first generation population is 50 to ensure that all candidate models are included; for the remaining populations, the mean error of the previous population particles is used as the error threshold and error index Δ(data,data) for the current generation population. *The model fitting segment root mean square error and the perturbation kernel k(·) are used. The uniformly distributed perturbation kernel and the initial model weights W(M) are used. j =1 / 8, the initial probability distribution of the parameter p(θ) adopts a uniform distribution, and the convergence condition is set as the upper limit of the relative error of the population.
[0143] Calculations showed that the 15th population reached the set convergence condition. The typical population sub-model ensemble weights are shown below. Figure 6 The probability distribution of parameters of the sub-model with the largest ensemble weight in the final population is shown in the figure. Figure 7 The point estimates of each parameter of the sub-model can be obtained by using the probability distribution of each parameter of the final population. The estimated values of each model parameter are shown in Table 4.
[0144] Table 4. Parameter estimation results of the deformation monitoring model for section 9 of the flood control dike.
[0145]
[0146]
[0147] 3) Construct an integrated model for monitoring the deformation of gravity-type concrete flood control dikes. Substitute the estimated values of the parameter points of each model into the model to obtain the predicted values of each sub-model.
[0148] (4) Monitoring and early warning interval: Model-based monitoring and early warning indicators. Based on the original observation data δ′(t) of the gravity-type flood control dike deformation and the fitted value sequence δ(t) corresponding to the original deformation observation data δ′(t) calculated using the integrated model for monitoring flood control dike deformation, the model residual sequence r(t) = δ′(t) - δ(t) is calculated. The residual distribution f(r) is estimated using the kernel density function in MATLAB software. The results are as follows: Figure 8 As shown. Based on the model fitted value δ(t), a certain quantile value of the residual distribution, r1 = 1.0025, r2 = 0.8417 (confidence level can be taken as 99%, i.e., f(r1) = 0.5%, f(r2) = 99.5%), is superimposed to obtain the monitoring and early warning interval. When newly collected arch dam deformation data is outside the monitoring and early warning range, the arch dam deformation data is considered abnormal, and the model issues an abnormality warning.
[0149] This example uses the top displacement monitoring data of a dike from May 1, 2018 to May 1, 2020 from an actual project as the model fitting segment, and establishes an integrated model M9 for flood control dike deformation monitoring based on the deformation monitoring method of this scheme. (1) The monitoring data of levee crest displacement from May 2, 2020 to August 1, 2020 were used as the model prediction segment to verify the predictive performance of the established model, and M9 was used as the prediction segment. (1) Compared with the ensemble model M9 established through traditional methods (2) M9 (3)A comparison was made. The model calculation results were evaluated using indicators such as root mean square error, mean absolute error, mean absolute percentage error, correlation coefficient, overfitting coefficient, and false alarm rate. The integrated weights of each sub-model, calculated using the three methods, are as follows: Figure 9 As shown, the performance indicators of each integrated model and sub-model are listed in Tables 5, 6, and 7. The monitoring and early warning intervals of the integrated model are as follows: Figure 10 , Figure 11 and Figure 12 As shown.
[0150] Table 5 Performance Indicators of the Deformation Monitoring Model for Section 9 of the Flood Control Dike (Method Proposed in this Invention)
[0151]
[0152] Table 6 Performance Indicators of the Deformation Monitoring Model for Section 9 of the Flood Control Dike (Markov Chain Monte Carlo Method)
[0153]
[0154] Table 7 Performance Indicators of the Deformation Monitoring Model for Section 9 of the Flood Control Dike (Multiple Linear Regression-Genetic Algorithm)
[0155]
[0156] From Tables 5-7 and Figure 10-12 It can be seen that both the multiple linear regression-genetic algorithm and the Markov chain Monte Carlo method achieved the highest fitting accuracy for their respective sub-models, M8. (2) and M2 (3) Assigning the highest weights results in the ensemble model M9. (2) and M9 (3) Both exhibited varying degrees of overfitting, with overfitting coefficients of 2.89 and 2.31, respectively. Furthermore, the proposed method did not assign the maximum weight to the model with the best fit accuracy; the ensemble model M9... (1) While uncertainty was reduced, overfitting did not occur. The root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), overfitting coefficient (OC), and false alarm rate R0 were all measured. FA The percentages were 0.33%, 0.24%, 8.58%, 0.96%, 0.60%, and 0.00%, respectively, all of which outperformed the other two integrated monitoring models, M9. (2) and M9 (3) and each sub-model.
[0157] It is evident that the integrated model for monitoring flood control dike deformation calculated based on the method proposed in this scheme can better utilize the advantages of each sub-model, assign reasonable weights to each sub-model, and the established integrated model can truly reflect the operating status of the flood control dike, with good predictive performance and promising application prospects.
Claims
1. A method for monitoring the deformation of flood control dikes based on an integrated model, characterized in that, Including the following steps: S1. Obtain multiple deformation monitoring sub-models of the flood control dike when different temperature and pressure components, water pressure components, and failure aging components are selected; S2, obtaining the measured deformation monitoring data set, and updating the probability distribution of the deformation monitoring sub-model parameters in a random sampling screening manner and the deformation monitoring sub-model integration weight : in, To meet the population error threshold ε The probability distribution of model parameters for population computation; data This is a dataset of measured deformation monitoring data; data * Fit the dataset to the model; ε The population error threshold is △( data , data * ) is the error index; M t For any deformable monitoring sub-model in the population; M j For the target sub-model; θ (t) Particle parameters to meet error specifications; θ * These are the parameters of randomly sampled particles; T The total number of particle parameters in the population; I (•) is an indicator function; S3. Construct an integrated model for flood control dike deformation monitoring based on the integration weights of the deformation monitoring sub-models. ; wherein, is the predicted value of the sub-model; S4. Obtain flood control embankment deformation monitoring data and calculate flood control embankment deformation using the flood control embankment deformation monitoring integrated model. And calculate the monitoring and early warning interval [ + r 1, + r 2], r 1 and r Both 2 are quantile values; S5. Determine whether the deformation monitoring data of the flood control embankment is within the monitoring and early warning range. If so, the flood control embankment is operating normally; otherwise, issue an early warning of abnormal operation of the flood control embankment. Step S2 further includes: S21. Initialize the total number of particle parameters, population error threshold, error index, and parameter perturbation kernel of the population. k (•) Initial weights, initial probability distribution of parameters, and convergence conditions for each deformation monitoring sub-model; S22. Select the deformation monitoring sub-model according to the initial weight of the deformation monitoring sub-model, and determine whether the current population is the first population. If yes, proceed to step S23; otherwise, proceed to step S24. S23. Generate particle parameters based on the initial probability distribution of parameters, the measured deformation monitoring dataset, and the selected deformation monitoring sub-model, and then proceed to step S25. S24. Based on the selected deformation monitoring sub-model, the previous generation population and the parameter perturbation kernel, the particle parameters are generated using formula (1), and then proceed to step S25. in, For population p The first in i Each particle parameter; For population p -1 of the j Each particle parameter; For population p -1 of the t Each particle parameter; uniform It is a uniform distribution function; p For the number of generations in a population; L p-1 For population p The range of the parameter value of -1; S25, according to the selected deformation monitoring sub-model and the generated particle parameters, calculating model fitting data, and based on the measured deformation monitoring data and the model fitting data, calculating an error index Δ data , data * ): wherein, y i is the measured deformation monitoring data; is the model fit value; n is the total number of measured deformation monitoring data; S26. Determine whether the error index meets the preset error. If yes, proceed to step S27; otherwise, delete the particle parameters generated by the selected deformation monitoring sub-model and return to step S22. S27. Accept the particle parameters, assign them weights using formula (2), and increment the number of particle parameters by 1. in, For the first p The first generation of the population i Weights of each particle parameter; for Initial probability distribution of parameters; exp (•) is an exponential function; ε p For the first p Population error threshold for a generation; S28. Determine whether the number of particle parameters in the current population has reached the total number of particle parameters in the population. If yes, proceed to step S29; otherwise, return to step S22. S29. Normalize the weights of the particle parameters so that the sum of the weights of the parameters of each particle in the current population is 1. S210. Determine whether the relative error of the current population meets the convergence condition. If yes, proceed to step S211; otherwise, increment the population size by 1 and return to step S22. S211. Calculate the probability distribution of model parameters for each deformation monitoring sub-model using the last generation population. and deformation monitoring sub-model integration weights .
2. The flood control dike deformation monitoring method based on an integrated model according to claim 1, characterized in that, The deformation monitoring sub-model is constructed using water pressure, temperature, and aging components: in, δ H δ T δ θ These are water pressure component, temperature component, and aging component, respectively. C For constant terms; The water pressure components include different forms of models: ， in, a l For the first l Water pressure component coefficient; H , H min , H max These are the measured water level, minimum water level, and historical maximum water level of the river. EL 0 represents the elevation of the foundation surface of the flood control dike; The temperature components include different forms of models: in, b o This is the coefficient of the 0th temperature component; PC o This is the 0th principal component extracted using principal component analysis; m The total number of principal components; b 1~ b 4 represents the regression coefficients of each temperature component; s Temperature factor; q The date corresponding to the temperature measurement; q 0 represents January 1st of the year the test began; The time-sensitivity component includes different forms of models: in, c 1 and c Both 2 are time-dependent coefficients; Time factor; This is the start date of the test.
3. The integrated model based monitoring method of embankment deformation for flood protection according to any one of claims 1-2, characterized in that, Quantile values r 1 and r The acquisition method of 2 comprises: Obtain raw observation data on the deformation of flood control embankments over a preset time period. The deformation monitoring integrated model of the flood control dike was used to calculate the original deformation observation data. The corresponding fitted value sequence ; Based on the original deformation observation data and fitted value sequence Calculate the residual sequence of the model ; The kernel density function of MATLAB software is used to estimate the normal distribution of residuals. f ( r ), obtained based on preset reliability f ( r 1) and f ( r 2) Corresponding quantile values r 1 and r 2.
4. The integrated model based flood embankment deformation monitoring method of claim 3, wherein, The preset reliability is 99%. f ( r 1) = 0.5%, f ( r 2) = 99.5%.