CN112884198B - Method for predicting dam crest settlement of panel dam by combining threshold regression and improved support vector machine - Google Patents

Method for predicting dam crest settlement of panel dam by combining threshold regression and improved support vector machine Download PDF

Info

Publication number
CN112884198B
CN112884198B CN202110043580.2A CN202110043580A CN112884198B CN 112884198 B CN112884198 B CN 112884198B CN 202110043580 A CN202110043580 A CN 202110043580A CN 112884198 B CN112884198 B CN 112884198B
Authority
CN
China
Prior art keywords
dam
settlement
support vector
regression
prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110043580.2A
Other languages
Chinese (zh)
Other versions
CN112884198A (en
Inventor
温立峰
张海洋
李炎隆
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian University of Technology
Original Assignee
Xian University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xian University of Technology filed Critical Xian University of Technology
Priority to CN202110043580.2A priority Critical patent/CN112884198B/en
Publication of CN112884198A publication Critical patent/CN112884198A/en
Application granted granted Critical
Publication of CN112884198B publication Critical patent/CN112884198B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2462—Approximate or statistical queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/08—Construction
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
    • Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Software Systems (AREA)
  • Operations Research (AREA)
  • Tourism & Hospitality (AREA)
  • Development Economics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Databases & Information Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Mathematical Optimization (AREA)
  • Game Theory and Decision Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Fuzzy Systems (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Educational Administration (AREA)
  • Artificial Intelligence (AREA)
  • Quality & Reliability (AREA)
  • Evolutionary Biology (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)

Abstract

The invention discloses a method for predicting dam crest settlement of a face plate dam by combining threshold regression and improving a support vector machine, which specifically comprises the following steps: step 1: obtaining sedimentation data; step 2: processing the data; step 3: determining a support vector machine kernel function; step 4: adopting a heuristic particle swarm optimization algorithm to optimize and select a regularization parameter C, a polynomial kernel parameter d and a radial basis kernel parameter g of the support vector regression machine; step 5: establishing a dam crest settlement prediction model combining threshold regression and an improved support vector machine; step 6: combining the established combined threshold regression and improving a face rockfill dam top settlement prediction model of the support vector machine. The method solves the problem that the dam crest settlement cannot be accurately predicted according to the existing empirical formula in the prior art.

Description

Method for predicting dam crest settlement of panel dam by combining threshold regression and improved support vector machine
Technical Field
The invention belongs to the technical field of hydraulic and hydroelectric engineering construction, and particularly relates to a method for predicting dam crest settlement of a panel dam by combining threshold regression and an improved support vector machine.
Background
The concrete face rockfill dam is a dam type which is filled by local earth materials, stones or mixtures through methods of throwing, filling, rolling and the like, and has the characteristics of simple structure, low engineering cost, good adaptive deformation performance, low requirement on foundations, high construction speed and the like, so that the concrete face rockfill dam is the dam type with the highest competitive power and is widely applied to hydraulic and hydroelectric engineering construction. Along with the great investment of the country in water conservancy infrastructure and the superiority of the face rockfill dam, the dam type gradually develops to an ultra-high dam type in recent years, and meanwhile, the deformation control of the dam body also faces a plurality of difficulties, in particular to the aspects of stability of the dam body, cracking of the face rockfill dam and the like.
The deformation control of the dam body is a most critical consideration factor for the construction of the face rockfill dam, and how to effectively and reasonably evaluate and control the deformation of the face rockfill dam is an important factor for determining the further development of the face rockfill dam. One of the main reasons for causing the instability of the dam body is that the dam crest subsides, and the excessive sinking displacement of the dam crest can cause the problems of panel deformation, cracks, seepage and the like, thereby seriously threatening the safety of the dam body. In conventional designs, the dam top settlement is not considered to exceed 2% of the dam height, but the permanent settlement displacement of the dam top cannot be accurately estimated because of creep after the dam is built. If the dam crest settlement can be predicted, the potential danger of the dam body can be timely dealt with, and the loss and the harm are reduced. Therefore, the prediction of the dam crest settlement is particularly important for guiding the design, operation and stability of the dam.
The current method for predicting the dam top settlement of the face rockfill dam mainly comprises a numerical calculation method, a centrifugal model test method and a traditional experience prediction method. The calculation parameters required by the calculation of the numerical model are very dependent on the test result, and the scaling effect in the test inevitably affects the accuracy of the rock-fill test result. The centrifugal model test technology is currently mainly used for clay core wall rock-fill dams, and the centrifugal model test is applied less in the research of the mechanical properties of the face rock-fill dams. In addition, the high cost of the centrifugal model test method, the handling of model boundaries, the simplification of the model, and other experimental limitations have limited their application to some extent. The current design of the face rockfill dam is still mainly based on engineering judgment and engineering experience, so that a method for predicting the settlement of the top of the face rockfill dam is important.
Some scholars at home and abroad put forward a model for predicting dam crest settlement. For example, early Lawton and Lester established a simple empirical prediction formula between dam top settlement and dam height based on 11 dam instance data. Sowers research the dam top settlement of 14 face rockfill dams, and establish an empirical prediction formula for the dam top late settlement considering the dam height and the measurement time. These empirical prediction formulas take into account fewer deformation control factors, often build a direct empirical formula between deformation characteristics and dam height, measurement time, and employ fewer dam instances. Based on the deficiency of the empirical prediction formula, the Clements consider the influence of the measurement time, and an empirical prediction formula between dam crest settlement and dam height is established based on the measured data of 68 dams. This calculation equation only considers one factor of the dam height, while dam settlement is affected by a number of factors, such as dam height, void ratio, time, form factor, vertical compression modulus, etc. Since these parameters also interact, it is apparent that dam top settlement cannot be accurately predicted using such empirical formulas. In order to overcome the defects of the traditional empirical prediction method, intelligent and machine learning algorithms are continuously used for predicting the deformation of the face rockfill dam. Li Jinfeng, yang Qigui and the like research the application of the neural network model in the settlement deformation prediction of the rock-fill body of the face rockfill dam in the construction period; kim establishes an intelligent prediction model for predicting relative dam top settlement by adopting an artificial neural network method based on actual measurement dam top settlement data of 30 face rockfill dams. The neural network model for predicting the deformation of the concrete face rockfill dam is easy to sink into a local minimum value, the convergence speed of the learning process is low, and meanwhile, the quantity and the quality of the learning samples have great influence on a final result. Therefore, further intensive research is necessary for a method for predicting dam crest settlement.
Disclosure of Invention
The invention aims to provide a method for predicting dam crest settlement of a panel dam by combining threshold regression and improving a support vector machine, which solves the problem that dam crest settlement cannot be accurately predicted according to the existing empirical formula in the prior art.
The technical scheme adopted by the invention is that the method for predicting the dam crest settlement of the panel dam of the support vector machine by combining threshold regression comprises the following steps:
step 1: obtaining sedimentation data
Step 2: processing data
Calculating correlation coefficients between a dam crest settlement prediction index CS and six prediction factors respectively, and taking a factor with the largest phase relation number as a threshold variable; calculating the determined threshold variable according to a multiple threshold regression analysis theory, and determining a corresponding threshold value of the threshold variable so as to establish a corresponding sub-data set;
step 3: determining support vector machine kernel functions
Forming a new mixed kernel function by using kernel functions with different characteristics in a linear combination mode;
step 4: adopting a heuristic particle swarm optimization algorithm to optimize and select a regularization parameter C, a polynomial kernel parameter d and a radial basis kernel parameter g of the support vector regression machine;
step 5: establishing a dam crest settlement prediction model combining threshold regression and an improved support vector machine;
step 6: combining the established combined threshold regression and improving a face rockfill dam top settlement prediction model of the support vector machine.
The present invention is also characterized in that,
the specific process of the step 1 is as follows:
selecting n dam heights X which all contain predictors from an actual measurement database of the constructed concrete face rockfill dam top settlement predictors 1 Porosity X 2 Valley shape factor X 3 Foundation condition X 4 Strength X of rock-fill 5 Run time X 6 And an example of a dam top settlement prediction index CS; and carrying out dimensionless treatment on the dam crest settlement prediction index CS in each example, namely Y=CS/H, namely taking the ratio of the dam crest settlement deformation to the dam height.
The specific process of the step 2 is as follows:
step 2.1: calculating dam crest settlement prediction index Y and dam height X of each example by using SPSS software 1 Porosity X 2 Valley shape factor X 3 Foundation condition X 4 Strength X of rock-fill 5 Run time X 6 Correlation coefficient among the n samples, and the prediction factor X with the maximum correlation coefficient in the n sample samples imax ＝{X i1 ,X i2 ,…,X in The values of the data are arranged from small to large, and the constructed example database is also related to X imax The sequence of the dam crest settlement prediction indexes Y is rearranged correspondingly to obtain a new sequence Y' of the dam crest settlement prediction indexes Y;
step 2.2: performing two-division on a new sequence Y' of the dam crest settlement prediction index Y, and respectively marking the two divided sections as: y '(1, k), Y' (k+1, n), where k is an arbitrary division point; calculating to obtain the total variance of the new sequence Y' of the dam crest settlement prediction index Y
Figure BDA0002896277050000041
The sum of the group variances divided into two sections is +.>
Figure BDA0002896277050000042
Inter-group variance B divided into two sections 2 ＝V 2 -S 2 ；
Step 2.3: judging coefficient F= [ B ] by using significance level 2 (n-2)]/S 2 Checking the significance difference between two groups of data which divide the new sequence Y 'of the dam crest settlement prediction index Y into two sections, performing L=n-1 times of two divisions on the sequence Y', calculating n-1F values, and marking as: f (F) 1 (x i )，F 2 (x i )，…，F n-1 (x i ) Selecting
Figure BDA0002896277050000043
The corresponding arbitrary partition point k is the optimal partition point of the ith factor and is marked as k * The method comprises the steps of carrying out a first treatment on the surface of the The m optimal dividing points obtained through calculation are m threshold values corresponding to the m optimal dividing points, and the m threshold values divide the whole instance database into m+1 sub-databases.
The specific process of the step 3 is as follows:
global kernel function K to be of local nature poly And a local kernel function K with global characteristics rbf In linear combination K mix ＝ηK poly +(1-η)K rbf Form of (a) constitutes a new class of mixed kernel functions K which meet the Mercer theorem mix 。
The specific implementation process of the step 4 is as follows:
(1) Reading sample data, and normalizing the sample data to a [0,1] interval;
(2) Determining particle number pop=20, learning factor parameter c 1 And an acceleration learning factor c 2 The learning step length is adjusted, and the learning step length is respectively initialized and set as c 1 ＝1.4、c 2 =1.6, maximum algebra maxgen=100, regularization parameter C with a variation range of [0.0001,100 ]]The radial basis parameters g vary within the range of [0.0001,1000 ]]The polynomial kernel parameter d takes a default value of 3 and randomly generates the initial position of each particle (C, g)
Figure BDA0002896277050000051
And generates an initial speed +.>
Figure BDA0002896277050000052
(3) Regression training is carried out on each particle, the mean square error of K-fold cross validation is taken as the adaptation value of the particle, and the initial position is taken as the individual extremum position P of each particle best The position with the optimal adaptation value is taken as the global optimal position g of the particle swarm best ；
(4) After training the initial population, the method is carried out according to the formula
Figure BDA0002896277050000053
Figure BDA0002896277050000054
Updating the position and the speed of the particles, and enabling the updated numerical value to be in a set range;
wherein t is the number of evolutionary algebra,
Figure BDA0002896277050000055
is the C, g-dimensional component of the ith particle flight velocity vector at the t-th iteration,/->
Figure BDA0002896277050000056
Is the C, g-th dimensional component of the ith particle position vector at the t-th iteration; omega is a weight coefficient for balancing global and local searching capability, is generally used for adjusting the searching range of a space and is initialized to 1; c 1 And c 2 For two different acceleration learning factors, adjusting the learning step length to be 1.4 and 1.6 in an initialized mode respectively; r is (r) 1 And r 2 Is interval [0,1]]The uniform random number in the search module is used for adjusting the randomness of the search;
(5) The adaptation value of the particles in each iteration is calculated through regression training, if the adaptation value of the particles is superior to the original individual extremum, the current adaptation value is set as the individual extremum, the current position is set as the individual extremum position, and otherwise, the original value is reserved;
(6) Comparing the adaptation value of the current population with the original global extremum, setting the current adaptation value as the global extremum, setting the current position as the global extremum position, otherwise, keeping the original value;
(7) Performing adaptive transformation on the particles to prevent the particles from falling into a local optimum;
(8) Returning to the step (4), stopping iteration until the maximum iteration number is met or the required error is met, and outputting the global optimal value position at the moment;
(9) And obtaining a global optimal value and then carrying out regression prediction.
The specific process in the step 5 is as follows:
respectively analyzing and processing the divided m+1 sub-data sets, training dam top settlement data in each sub-data set to build a model, wherein the operation of building the model by training the obtained data considers the prediction factor dam height, foundation conditions, rockfill strength, porosity, valley shape factor and running time to obtain a prediction model of the dam top settlement prediction index Y combined with threshold regression and an improved support vector machine, and after model training is completed, storing the model into a model file by a save_model method for subsequent prediction.
The specific implementation process of the step 6 is as follows:
when the prediction factor dam height, foundation conditions, rock-fill strength, porosity, valley shape factor and running time of the face rockfill dam project to be built are known, the stored dam top settlement prediction model file combined with threshold regression and improved support vector machine is loaded first, and dam top settlement of the face rockfill dam to be built is predicted.
The beneficial effects of the invention are as follows:
according to the dam crest settlement prediction method, through the control function of the threshold variable, after prediction factor data is given, different prediction equations are used under different conditions according to the judgment control function of the threshold value of the threshold variable, so that various phenomena similar to jumping and mutation are attempted to be explained. Secondly, by comparing the performance of different kernel functions, a self-adaptive mixed kernel function is constructed, and the main parameters of the support vector regression machine are optimized and selected by adopting a particle swarm intelligent optimization algorithm. In order to further improve generalization capability and accuracy of the model, from the perspective of segment modeling, clustering division is carried out on dam top settlement by utilizing a multi-threshold regression theory, so that a face rockfill dam top settlement prediction model combining threshold regression and an improved support vector machine is established. The dam crest settlement prediction model combining threshold regression and an improved support vector machine is a nonlinear time sequence model, and can effectively describe complex phenomena such as variability, quasi-periodicity, piecewise dependence and the like. Introducing a threshold concept through a threshold regression model, and ensuring the robustness and wide applicability of the accuracy in the model prediction process by utilizing the control action of the threshold; the main parameters of the support vector regression machine are optimized and selected by constructing a self-adaptive mixed kernel function and adopting a particle swarm intelligent optimization algorithm, so that the generalization capability and accuracy of the model are further improved.
Drawings
FIG. 1 is a flow chart of the method of the invention for predicting dam top settlement of a face dam in combination with threshold regression and improved support vector machine;
FIG. 2 is a graph of a locally improved support vector machine prediction using multiple threshold regression in accordance with the present invention;
fig. 3 is a schematic flow chart of an implementation of a Particle Swarm (PSO) algorithm in the method of the present invention.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
The invention provides a method for predicting dam crest settlement of a panel dam by combining threshold regression and improving a support vector machine, which is shown in fig. 1 and specifically comprises the following steps:
step 1: obtaining sedimentation data
Based on a large amount of existing literature data, collecting maximum dam top settlement data and detailed construction information of n face rockfill dams built in the past 50 years, wherein the data size is more than 87;
the specific process of the step 1 is as follows:
selecting n dam heights X which all contain predictors from an actual measurement database of the constructed concrete face rockfill dam top settlement predictors 1 Porosity X 2 Valley shape factor X 3 Foundation condition X 4 Strength X of rock-fill 5 Run time X 6 And an example of a dam top settlement prediction index CS; and carrying out dimensionless treatment on the dam crest settlement prediction index CS in each example, namely Y=CS/H, namely taking the ratio of the dam crest settlement deformation to the dam height.
Step 2: processing data
Calculating correlation coefficients between a dam crest settlement prediction index CS and six prediction factors respectively, and taking a factor with the largest phase relation number as a threshold variable; calculating the determined threshold variable according to a multiple threshold regression analysis theory, and determining a corresponding threshold value of the threshold variable so as to establish a corresponding sub-data set;
the specific process of the step 2 is as follows:
step 2.1: calculating dam crest settlement prediction index Y and dam height X of each example by using SPSS software 1 Porosity X 2 Valley shape factor X 3 Foundation condition X 4 Strength X of rock-fill 5 Run time X 6 Correlation coefficient among the n samples, and the prediction factor X with the maximum correlation coefficient in the n sample samples imax ＝{X i1 ,X i2 ,…,X in Sequentially from small to largeColumn, also with X for the constructed instance database imax The sequence of the dam crest settlement prediction indexes Y is rearranged correspondingly to obtain a new sequence Y' of the dam crest settlement prediction indexes Y;
step 2.2: performing two-division on a new sequence Y' of the dam crest settlement prediction index Y, and respectively marking the two divided sections as: y '(1, k), Y' (k+1, n), where k is an arbitrary division point; calculating to obtain the total variance of the new sequence Y' of the dam crest settlement prediction index Y
Figure BDA0002896277050000081
The sum of the group variances divided into two sections is +.>
Figure BDA0002896277050000091
Inter-group variance B divided into two sections 2 ＝V 2 -S 2 ；
Step 2.3: judging coefficient F= [ B ] by using significance level 2 (n-2)]/S 2 Checking the significance difference between two groups of data which divide the new sequence Y 'of the dam crest settlement prediction index Y into two sections, performing L=n-1 times of two divisions on the sequence Y', calculating n-1F values, and marking as: f (F) 1 (x i )，F 2 (x i )，…，F n-1 (x i ) Selecting
Figure BDA0002896277050000092
The corresponding arbitrary partition point k is the optimal partition point of the ith factor and is marked as k * The method comprises the steps of carrying out a first treatment on the surface of the The m optimal dividing points obtained through calculation are m threshold values corresponding to the m optimal dividing points, and the m threshold values divide the whole instance database into m+1 sub-databases.
The specific parameters are shown in table 1, the first threshold value is determined to be 0.57, the second threshold value is determined to be 1.09, the third threshold value is determined to be 1.45, and the corresponding dam heights are 57m, 109m and 145m respectively. In addition, according to the common practice of the international committee on dams (ICOLD), the face rockfill dams can be divided into low dams (< 30 m), medium dams (30 m < h <70 m) and high dams (> 70 m) according to dam heights, and many students also consider 100m and 150m as the standard for dividing the high face rockfill dams according to dam heights, so that the threshold value is matched with the common practice of dividing the face rockfill dams according to dam heights.
TABLE 1 face rockfill dam deformation characteristics database partitioning
Figure BDA0002896277050000093
Step 3: determining support vector machine kernel functions
Forming a new mixed kernel function by using kernel functions with different characteristics in a linear combination mode; the mixed kernel function has the advantages of a global kernel function polynomial function and a local kernel function radial basis function, and the action of the global kernel function polynomial function and the local kernel function radial basis function on the mixed kernel function can be adjusted through a weight coefficient factor eta;
the specific process of the step 3 is as follows:
adopting a method for linearly combining kernel functions, and forming a new type of self-adaptive mixed kernel function which meets the Mercer theorem by using the kernel functions with different characteristics in a linear combination mode;
consider a method that includes all 6 (predictor dam height X 1 Porosity X 2 Valley shape factor X 3 Foundation condition X 4 Strength X of rock-fill 5 Run time X 6 ) The predictor, the Matlab software is used to determine the kernel function of the support vector machine in m+1 sub-sample areas divided by m threshold values, namely, in each sub-database obtained in step 2, and a linear combination kernel function method is adopted to make the global kernel function K with local characteristics poly And a local kernel function K with global characteristics rbf In linear combination K mix ＝ηK poly +(1-η)K rbf Form of (a) constitutes a new class of mixed kernel functions K which meet the Mercer theorem mix The combined kernel function has the global kernel function K poly And local kernel function K rbf And the magnitude of the actions of the weight coefficient factors eta on the combined kernel function can be adjusted to construct an adaptive mixed kernel function. Finally, after debugging, when the eta value corresponding to each dam high section is shown in table 2, the learning precision and generalization capability of the support vector machine model are optimized.
Table 2 face rockfill dam deformation characteristics database corresponding eta values for each dam high section
Figure BDA0002896277050000101
Step 4: adopting a heuristic particle swarm optimization algorithm to optimize and select a regularization parameter C, a polynomial kernel parameter d and a radial basis kernel parameter g of the support vector regression machine;
based on the self-adaptive mixed kernel function determined in the step 3, a heuristic particle swarm optimization algorithm is adopted to optimize and select parameters of a support vector regression machine so as to improve learning accuracy and popularization generalization performance of the dam crest settlement prediction model. The accuracy, generalization performance of the support vector machine model depends mainly on the parameter combination (C, d, g), where C is the regularization parameter, d is the polynomial kernel parameter, and g is the radial basis kernel parameter. The algorithm steps of the particle swarm optimization support vector machine (PSO-SVR) optimization three parameters are as follows, as shown in FIG. 3:
(1) Reading sample data, and normalizing the sample data to a [0,1] interval;
(2) Determining particle number pop=20, learning factor parameter c 1 And an acceleration learning factor c 2 The learning step length is adjusted, and the learning step length is respectively initialized and set as c 1 ＝1.4、c 2 =1.6, maximum algebra maxgen=100, regularization parameter C with a variation range of [0.0001,100 ]]The radial basis parameters g vary within the range of [0.0001,1000 ]]The polynomial kernel parameter d takes a default value of 3 and randomly generates the initial position of each particle (C, g)
Figure BDA0002896277050000111
And generates an initial speed +.>
Figure BDA0002896277050000112
(3) Regression training is carried out on each particle, the mean square error of K-fold cross validation is taken as the adaptation value of the particle, and the initial position is taken as the individual extremum position P of each particle best Optimizing the adaptation valueThe position is taken as the global optimal position g of the particle swarm best ；
(4) After training the initial population, the method is carried out according to the formula
Figure BDA0002896277050000113
Figure BDA0002896277050000114
Updating the position and the speed of the particles, and enabling the updated numerical value to be in a set range;
wherein t is the number of evolutionary algebra,
Figure BDA0002896277050000115
is the C, g-dimensional component of the ith particle flight velocity vector at the t-th iteration,/->
Figure BDA0002896277050000116
Is the C, g-th dimensional component of the ith particle position vector at the t-th iteration; omega is a weight coefficient for balancing global and local searching capability, is generally used for adjusting the searching range of a space and is initialized to 1; c 1 And c 2 For two different acceleration learning factors, adjusting the learning step length to be 1.4 and 1.6 in an initialized mode respectively; r is (r) 1 And r 2 Is interval [0,1]]The uniform random number in the search module is used for adjusting the randomness of the search;
(5) The adaptation value of the particles in each iteration is calculated through regression training, if the adaptation value of the particles is superior to the original individual extremum, the current adaptation value is set as the individual extremum, the current position is set as the individual extremum position, and otherwise, the original value is reserved;
(6) Comparing the adaptation value of the current population with the original global extremum, setting the current adaptation value as the global extremum, setting the current position as the global extremum position, otherwise, keeping the original value;
(7) Performing adaptive transformation on the particles to prevent the particles from falling into a local optimum;
(8) Returning to the step (4), stopping iteration until the maximum iteration number is met or the required error is met, and outputting the global optimal value position at the moment;
(9) And obtaining a global optimal value and then carrying out regression prediction.
Finally, through optimization selection of a particle swarm optimization algorithm, when regularization parameters C, polynomial kernel parameters d and radial basis kernel parameters g of each dam high section are shown in a table 3, learning accuracy and popularization generalization performance of the dam crest settlement prediction model are best.
Table 3C, d, g values corresponding to each dam high section of rock-fill dam deformation characteristics database
Figure BDA0002896277050000121
Step 5: establishing a dam crest settlement prediction model combining threshold regression and an improved support vector machine;
the specific process in the step 5 is as follows:
respectively analyzing and processing the divided m+1 sub-data sets, training dam top settlement data in each sub-data set to build a model, wherein the operation of building the model by training the obtained data considers the prediction factor dam height, foundation conditions, rockfill strength, porosity, valley shape factor and running time to obtain a prediction model of the dam top settlement prediction index Y combined with threshold regression and an improved support vector machine, and after model training is completed, storing the model into a model file by a save_model method for subsequent prediction. And (3) determining a support vector machine kernel function in the step (3) and optimally selecting support vector machine parameters in the step (4) in the threshold value interval obtained in the step (2), and finally obtaining the segmented dam crest settlement prediction model parameters of each dam high interval as shown in the table 4.
Table 4 dam crest settlement prediction model parameters combining threshold regression and improved support vector machine
Figure BDA0002896277050000131
Step 6: combining the established combined threshold regression with an improved support vector machine face rockfill dam top settlement prediction model;
the specific implementation process of the step 6 is as follows:
when the prediction factor dam height, foundation conditions, rock-fill strength, porosity, valley shape factor and running time of the face rockfill dam project to be built are known, the stored dam top settlement prediction model file combined with threshold regression and improved support vector machine is loaded first, and dam top settlement of the face rockfill dam to be built is predicted.
The method of the invention adopts n examples with detailed data, and reveals the statistical rule of the deformation characteristics of the face rockfill dam through statistical analysis. The threshold regression and the improved support vector machine are combined to establish a face-plate rock dam crest settlement prediction model, so that the following steps are obtained:
(1) The settlement of the top of the face rockfill dam is affected by the height of the dam, foundation conditions, rockfill strength, valley shape, porosity, and the running time of the dam. Wherein, the dam height, foundation condition and rock-fill strength are the main influencing factors for dam top settlement. The dam or the dam with lower rock-fill strength on the overburden foundation has obviously larger dam settlement deformation and longer stabilization time. The water retention effect can significantly affect the dam deformation characteristics.
(2) The predicted value and the measured value of the dam top settlement prediction model of the face rockfill dam built by the invention are relatively consistent. Compared with most existing prediction models, the prediction model established by the invention has obvious advantages, because the model has comprehensive consideration factors and relatively more used examples, and secondly, because the deformation characteristics of the face rockfill dam often show different variation trends in different dam height regions, the training samples are clustered and divided by the multi-threshold regression analysis, so that the improved support vector machine prediction model for different dam height regions is established, the defect of unstable global regression algorithm is overcome, the prediction accuracy is better, and the method is more suitable for the prediction of the subsidence of the face rockfill dam top.

Claims (2)

1. The method for predicting dam crest settlement of the support vector machine face plate dam by combining threshold regression is characterized by comprising the following steps of:
step 1: obtaining sedimentation data
The specific process of the step 1 is as follows:
selecting n dam heights X which all contain predictors from an actual measurement database of the constructed concrete face rockfill dam top settlement predictors 1 Porosity X 2 Valley shape factor X 3 Foundation condition X 4 Strength X of rock-fill 5 Run time X 6 And examples of dam top settlement predictors CS, wherein the predictors are the dam height X 1 Porosity X 2 Valley shape factor X 3 Foundation condition X 4 Strength X of rock-fill 5 Run time X 6 Six predictors are obtained; carrying out dimensionless treatment on the dam crest settlement prediction index CS in each example, namely Y=CS/H, namely taking the ratio of the dam crest settlement deformation to the dam height;
step 2: processing data
Calculating correlation coefficients between a dam crest settlement prediction index CS and six prediction factors respectively, and taking a factor with the largest phase relation number as a threshold variable; calculating the determined threshold variable according to a multiple threshold regression analysis theory, and determining a corresponding threshold value of the threshold variable so as to establish a corresponding sub-data set;
the specific process of the step 2 is as follows:
step 2.1: calculating dam crest settlement prediction index Y and dam height X of each example by using SPSS software 1 Porosity X 2 Valley shape factor X 3 Foundation condition X 4 Strength X of rock-fill 5 Run time X 6 Correlation coefficient among the n samples, and the prediction factor X with the maximum correlation coefficient in the n sample samples imax ＝{X i1 ,X i2 ,…,X in The values of the data are arranged from small to large, and the constructed example database is also related to X imax The sequence of the dam crest settlement prediction indexes Y is rearranged correspondingly to obtain a new sequence Y' of the dam crest settlement prediction indexes Y;
step 2.2: performing two-division on a new sequence Y' of the dam crest settlement prediction index Y, and respectively marking the two divided sections as: y '(1, k), Y' (k+1, n), where k is an arbitrary division point; calculating to obtain a new sequence Y' of dam crest settlement prediction indexes YTotal variance of
Figure FDA0004203137970000021
The sum of the group variances divided into two sections is +.>
Figure FDA0004203137970000022
Inter-group variance B divided into two sections 2 ＝V 2 -S 2 ；
Step 2.3: judging coefficient F= [ B ] by using significance level 2 (n-2)]/S 2 Checking the significance difference between two groups of data which divide the new sequence Y 'of the dam crest settlement prediction index Y into two sections, performing L=n-1 times of two divisions on the sequence Y', calculating n-1F values, and marking as: f (F) 1 (x i )，F 2 (x i )，…，F n-1 (x i ) Select F k* (x i )＝max{F 1 (x i ),F 2 (x i ),…,F n-1 (x i ) Any partition point k corresponding to the first factor is the optimal partition point of the ith factor, and is marked as k * The method comprises the steps of carrying out a first treatment on the surface of the The m optimal dividing points obtained through calculation are m threshold values corresponding to the m optimal dividing points, the m threshold values divide the whole instance database into m+1 sub-databases, and Matlab software is used for determining a support vector machine kernel function in the m+1 sub-databases;
step 3: determining support vector machine kernel functions
Forming a new mixed kernel function by using kernel functions with different characteristics in a linear combination mode;
the specific process of the step 3 is as follows:
global kernel function K to be of local nature poly And a local kernel function K with global characteristics rbf In linear combination K mix ＝ηK poly +(1-η)K rbf Form of (a) constitutes a new class of mixed kernel functions K which meet the Mercer theorem mix ；
Step 4: adopting a heuristic particle swarm optimization algorithm to optimize and select a regularization parameter C, a polynomial kernel parameter d and a radial basis kernel parameter g of the support vector regression machine;
the specific implementation process of the step 4 is as follows:
(1) Reading sample data, and normalizing the sample data to a [0,1] interval;
(2) Determining particle number pop=20, learning factor parameter c 1 And an acceleration learning factor c 2 The learning step length is adjusted, and the learning step length is respectively initialized and set as c 1 ＝1.4、c 2 =1.6, maximum algebra maxgen=100, regularization parameter C with a variation range of [0.0001,100 ]]The radial basis parameters g vary within the range of [0.0001,1000 ]]The polynomial kernel parameter d takes a default value of 3 and randomly generates the initial position of each particle (C, g)
Figure FDA0004203137970000031
And generates an initial speed +.>
Figure FDA0004203137970000032
(3) Regression training is carried out on each particle, the mean square error of K-fold cross validation is taken as the adaptation value of the particle, and the initial position is taken as the individual extremum position P of each particle best The position with the optimal adaptation value is taken as the global optimal position g of the particle swarm best ；
(4) After training the initial population, the method is carried out according to the formula
Figure FDA0004203137970000033
Figure FDA0004203137970000034
Updating the position and the speed of the particles, and enabling the updated numerical value to be in a set range;
wherein t is the number of evolutionary algebra,
Figure FDA0004203137970000035
is the C, g-th dimensional component of the ith particle flight velocity vector at the t-th iteration,
Figure FDA0004203137970000036
is the C, g-th dimensional component of the ith particle position vector at the t-th iteration; omega is a weight coefficient for balancing global and local searching capability, is generally used for adjusting the searching range of a space and is initialized to 1; c 1 And c 2 For two different acceleration learning factors, adjusting the learning step length to be 1.4 and 1.6 in an initialized mode respectively; r is (r) 1 And r 2 Is interval [0,1]]The uniform random number in the search module is used for adjusting the randomness of the search;
(5) The adaptation value of the particles in each iteration is calculated through regression training, if the adaptation value of the particles is superior to the original individual extremum, the current adaptation value is set as the individual extremum, the current position is set as the individual extremum position, and otherwise, the original value is reserved;
(6) Comparing the adaptation value of the current population with the original global extremum, setting the current adaptation value as the global extremum, setting the current position as the global extremum position, otherwise, keeping the original value;
(7) Performing adaptive transformation on the particles to prevent the particles from falling into a local optimum;
(8) Returning to the step (4), stopping iteration until the maximum iteration number is met or the required error is met, and outputting the global optimal value position at the moment;
(9) Carrying out regression prediction after obtaining a global optimal value;
step 5: establishing a dam crest settlement prediction model combining threshold regression and an improved support vector machine;
the specific process in the step 5 is as follows:
respectively analyzing and processing the divided m+1 sub-data sets, training dam top settlement data in each sub-data set to build a model, wherein the operation of building the model by training the obtained data considers the prediction factor dam height, foundation conditions, rockfill strength, porosity, valley shape factor and running time to obtain a prediction model of a dam top settlement prediction index Y combined with threshold regression and an improved support vector machine, and after model training is finished, storing the model into a model file by a save_model method for subsequent prediction;
step 6: combining the established combined threshold regression and improving a face rockfill dam top settlement prediction model of the support vector machine.
2. The method for predicting dam crest settlement of a panel dam by combining threshold regression and improved support vector machine as claimed in claim 1, wherein the specific implementation process of the step 6 is as follows:
when the prediction factor dam height, foundation conditions, rock-fill strength, porosity, valley shape factor and running time of the face rockfill dam project to be built are known, the stored dam top settlement prediction model file combined with threshold regression and improved support vector machine is loaded first, and dam top settlement of the face rockfill dam to be built is predicted.
CN202110043580.2A 2021-01-13 2021-01-13 Method for predicting dam crest settlement of panel dam by combining threshold regression and improved support vector machine Active CN112884198B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110043580.2A CN112884198B (en) 2021-01-13 2021-01-13 Method for predicting dam crest settlement of panel dam by combining threshold regression and improved support vector machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110043580.2A CN112884198B (en) 2021-01-13 2021-01-13 Method for predicting dam crest settlement of panel dam by combining threshold regression and improved support vector machine

Publications (2)

Publication Number Publication Date
CN112884198A CN112884198A (en) 2021-06-01
CN112884198B true CN112884198B (en) 2023-06-09

Family

ID=76045580

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110043580.2A Active CN112884198B (en) 2021-01-13 2021-01-13 Method for predicting dam crest settlement of panel dam by combining threshold regression and improved support vector machine

Country Status (1)

Country Link
CN (1) CN112884198B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113705900A (en) * 2021-08-30 2021-11-26 西安理工大学 Method for predicting deflection of face plate dam
CN114494979B (en) * 2022-03-29 2022-07-22 杭州定川信息技术有限公司 Method for video recognition of ecological flow discharge

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107357966A (en) * 2017-06-21 2017-11-17 山东科技大学 A kind of surrounding rock of actual mining roadway stability prediction and appraisal procedure
CN107958309A (en) * 2017-12-07 2018-04-24 河海大学 A kind of high dam multi-point deformetion condition combines forecasting procedure
CN108549770A (en) * 2018-04-13 2018-09-18 西安理工大学 The adaptive inversion method of Parameters for Rockfill Dams based on QGA-MMRVM
CN111191191A (en) * 2019-12-26 2020-05-22 南昌大学 Construction method of combined model for accurately predicting deformation effect of concrete dam
CN111199069A (en) * 2020-01-06 2020-05-26 西安理工大学 Face rockfill dam crest settlement experience prediction method based on threshold regression theory

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106443297B (en) * 2016-09-07 2018-05-22 江南大学 The decision tree SVM method for diagnosing faults of photovoltaic diode Clamp three-level inverter

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107357966A (en) * 2017-06-21 2017-11-17 山东科技大学 A kind of surrounding rock of actual mining roadway stability prediction and appraisal procedure
CN107958309A (en) * 2017-12-07 2018-04-24 河海大学 A kind of high dam multi-point deformetion condition combines forecasting procedure
CN108549770A (en) * 2018-04-13 2018-09-18 西安理工大学 The adaptive inversion method of Parameters for Rockfill Dams based on QGA-MMRVM
CN111191191A (en) * 2019-12-26 2020-05-22 南昌大学 Construction method of combined model for accurately predicting deformation effect of concrete dam
CN111199069A (en) * 2020-01-06 2020-05-26 西安理工大学 Face rockfill dam crest settlement experience prediction method based on threshold regression theory

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
A Prediction Model for Deformation Behavior of Concrete Face Rockfill Dams Based on the Threshold Regression Method;Yanlong Li, Haiyang Zhang, Lifeng Wen & Ning Shi;Springer;全文 *
基于PSO-SVM模型的拱坝坝变形预测研究;刘敬洋;刘何稚;朱凯;陈晨;张凤山;;三峡大学学报(自然科学版)(01);全文 *
基于SAPSO-RVM的大坝变形预测模型;唐琪 包腾飞 李月娇 屠立峰;万方/2015年全国大坝安全监测技术与应用学术交流会论文集;全文 *

Also Published As

Publication number Publication date
CN112884198A (en) 2021-06-01

Similar Documents

Publication Publication Date Title
Kirkby A two-dimensional simulation model for slope and stream evolution
CN112884198B (en) Method for predicting dam crest settlement of panel dam by combining threshold regression and improved support vector machine
CN103729550A (en) Multi-model integrated flood forecasting method based on propagation time clustering analysis
CN108549770A (en) The adaptive inversion method of Parameters for Rockfill Dams based on QGA-MMRVM
CN101344389A (en) Method for estimating tunnel surrounding rock displacement by neural network
CN106096203B (en) A kind of substrate construction method of hinge dam
CN115062389B (en) Reservoir gate scheduling method, device and equipment for front-dam sediment removal
CN112100927A (en) Prediction method for slope deformation and soft soil foundation settlement based on GA-BP neural network
CN106677763A (en) Oil well dynamic liquid level prediction method based on dynamic integrated modeling
CN104881707A (en) Sintering energy consumption prediction method based on integrated model
CN114417631B (en) Modeling method of infusion area water delivery and distribution system based on observation data
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
CN106295869A (en) A kind of based on the building settlement Forecasting Methodology improving unbiased function
CN106929615B (en) A kind of control method of blast furnace crucibe numerical simulation and resolution system
CN108196951A (en) GPU runoff simulations distributed scheduling system and method
CN114418234A (en) Power battery manufacturing capacity online prediction method based on reinforcement learning
CN113486556A (en) Improved efficient automatic history fitting method for oil and gas reservoir
CN111199069B (en) Threshold regression theory-based empirical prediction method for dam top settlement of rock-fill dam
CN116486028A (en) Self-adaptive inverse distance weighted interpolation method for three-dimensional geological structure modeling
CN116258275A (en) Foundation pit deformation multivariate time sequence prediction method based on ensemble learning
CN116503378A (en) Concentrate grade prediction method for bimodal CNN secondary transfer learning
CN108133286B (en) Underground water multi-target calculation method based on ground settlement substitution model
CN108446413A (en) A kind of optimization assay method of injection forming club-footed pile stake diameter
CN109117517A (en) A kind of base pit dewatering optimum design method based on genetic algorithm and groundwater model
CN114460639B (en) Shale oil reservoir permeability prediction method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant