CN103353295B - A kind of method of accurately predicting dam dam body vertical deformation amount - Google Patents

Abstract

The invention discloses a kind of method of accurately predicting dam dam body vertical deformation amount; the method is the method for the accurately predicting dam dam body vertical deformation amount based on degree of stability; be specially:? 1) Monitoring Data collection, 2) set up <i>i</iGreatT.Gr eaT.GT(<i>i</iGr eatT.GreaT.GT=1; 2 ... <i>m</iGreatT.Gr eaT.GT; <i>i</iGreatT.Gr eaT.GT >=2) plant the mathematical model, 3 of Forecasting Methodology) calculate the degree of stability, 4 of various Forecasting Methodology) determine combined method model weight coefficient, 5) set up based on degree of stability dam vertical deformation amount forecast model; Use the method greatly can improve the precision of prediction of dam vertical deformation amount.Through a large amount of case history Analysis of application result: the inventive method is than other Forecasting Methodology, and the precision of prediction of deflection will improve 20%-70%.The present invention can keep the continuity of model accuracy, and make accurately predicting to relating to safe development of deformation trend, take measures in advance, to preventing, dam safety accident is significant.There is significantly society and economic worth.

Description

A kind of method of accurately predicting dam dam body vertical deformation amount

Technical field

The invention belongs to the geodesy technical field in Surveying Science and Technology subject, particularly relate to the Forecasting Methodology of dam vertical deformation amount.

Background technology

Along with the exploitation of waterpower resourses, the scale of dam is increasing, and the geologic condition of dam site also becomes increasingly complex.Therefore, the safety problem of dam merits attention.Show according to whole nation third time statistics, end 2003, China's dam-break accident has 3481.So, after dam builds up, in order to understand the running status of dam, ensureing the safe operation of dam, all must carry out safety monitoring.The generation of dam accident is not accidental, generally all there is progressively evolution from quantitative change to qualitative change.By the treatment and analysis to dam deformation data, the real-time working can grasping dam is in time dynamic, effectively reduces engineering risk, reduces dam accident.Can say, Dam Deformation Monitoring is an important content in engineering survey field.What is more important utilizes the data of monitoring to make prediction to relating to safe development of deformation trend, takes measures in advance, prevents the generation of security incident.

At present, predict that dam vertical deformation metering method mainly concentrates on based on the theoretical method of Settlement Mechanism, the random statistical method of Corpus--based Method theory and method 3 aspects based on artificial intelligence.3 kinds of technique study angles are different, and prediction principle also has basic difference, but all serves good effect in Forecast Settlement.But sum up, there is following deficiency:

(1) theoretical method is based upon on detailed engineering geological investigation basis, integrated structure internal characteristics and associated external influence factor carry out computational analysis, require that clear and definite, all kinds of parameter of Settlement Mechanism is accurate, but due to the various complexity of the factor affecting sedimentation, Settlement Mechanism is more difficult to be understood fully, all kinds of parameter is because of test condition, sampling method restriction also more difficult Obtaining Accurate, theoretical method can only be described from single factor based on ideal hypothesis or experience conclusion settlement prediction, thus cause Calculation results and actual conditions to there is bigger difference;

(2) random statistical method is based on measured data, have quick, calculate the advantages such as simple, but the funtcional relationship can only set up in modeling process between settling amount and single or multiple factor, all kinds of influence factor relation each other can not be taken into full account, be difficult to the settlement prediction adapting to complex condition;

(3) for artificial intelligence approach, although be comparatively applicable to the prediction of sedimentation nonlinear system, calculation of complex, computing time is long, and result of calculation is unstable.

For the above deficiency that these methods exist, the present invention, by advantages such as the stability of research prediction vertical deformation metering method and combination techniques, proposes to adopt certain workflow to carry out accurately predicting dam dam body vertical deformation amount.Essence of the present invention is organically combined the flow process of above-mentioned three kinds of methods according to regulation, achieves the mutual supplement with each other's advantages of above various method, greatly can improve the precision of prediction of dam vertical deformation amount.By the actual monitoring data at scene, the deformation characteristics of research and application object and Changing Pattern thereof, utilize the data of monitoring to make accurately predicting to relating to safe development of deformation trend, take measures in advance, to preventing, dam safety accident is significant.

Summary of the invention

Goal of the invention: for above-mentioned existing Problems existing and deficiency, the object of this invention is to provide a kind of method of accurately predicting dam dam body vertical deformation amount, and the method predicts that the precision of prediction of dam deflection is higher, it is more convenient to use.

Technical scheme: for achieving the above object, the present invention by the following technical solutions: a kind of method of accurately predicting dam dam body vertical deformation amount, comprises the following steps:

The first step, Monitoring Data is divided into two parts in chronological order: a part is time preceding learning data M, and the number of M must be more than or equal to 10; Another part is the check data J of all the other times;

Second step, learning data M is utilized to set up the mathematical model of i kind Forecasting Methodology, wherein i=1,2 ... m; I >=2;

3rd step, calculate the degree of stability of various Forecasting Methodology:

A, dam time series monitor value are y _t(t=1,2, Λ n), the individual event forecast model different according to the m kind of second step foundation, y _it=(i=1,2, Lm) is i-th kind of Single model in the learning value of t or predicted value, calculates i-th kind of Single model at t period error e by formula (1) _it:

e _it＝y _t-y _it（1）

The precision A of i-th kind of Single model in the t phase is calculated by formula (2) _it:

$A_{\mathrm{it}}=\left\{\begin{array}{cc}1,& 0\&le;\left|\frac{e_{\mathrm{it}}}{y_t}\right|\&le;\mathrm{\&alpha;}\\ 1-\left|\frac{e_{\mathrm{it}}}{y_t}\right|,& \mathrm{\&alpha;}<\left|\frac{e_{\mathrm{it}}}{y_t}\right|<\mathrm{\&beta;}\\ 0,& \left|\frac{e_{\mathrm{it}}}{y_t}\right|\&GreaterEqual;\mathrm{\&beta;}\end{array}\right.---\left(2\right)$

Wherein, 0≤α < β≤1, α acquiescence gets 0 value, and β acquiescence gets 1 value;

If first b i-th kind of Single model carry out N phase study, then carry out T phase prediction, by formula (3)

Calculate the average study precision ε of i-th kind of model _i, the consensus forecast precision η of i-th kind of model _i:

${\mathrm{\&epsiv;}}_i=\underset{t=1}{\overset{N}{\mathrm{\&Sigma;}}}{A}_{\mathrm{it}}/N(i=\mathrm{1,2},\mathrm{\&Lambda;m})---\left(3\right)$

${\mathrm{\&eta;}}_i=\underset{t=N+1}{\overset{N+T}{\mathrm{\&Sigma;}}}{A}_{\mathrm{it}}/T(i=\mathrm{1,2},\mathrm{\&Lambda;m})$

C, then, calculates i-th kind of model stability degree S by formula (4) _i:

$S_i=\frac{{\mathrm{\&eta;}}_i}{|{\mathrm{\&eta;}}_i-{\mathrm{\&epsiv;}}_i|+\mathrm{\&sigma;}}---\left(4\right)$

σ is infinitely small arbitrarily;

4th step, determine combined method model weight coefficient:

Make w _ibe the weight of i-th kind of model shared by m kind Single model, then built-up pattern weight coefficient is defined as by formula (5):

$w_i=\frac{s_i}{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}s}---\left(5\right)$

5th step, according to formula (6), set up the dam vertical deformation amount forecast model based on degree of stability:

$y_t=\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{w}_i{y}_{\mathrm{it}}---\left(6\right)$

In formula, y _tfor the dam body vertical deformation amount predicted value of t, w _ifor the weight of i-th kind of model in built-up pattern, y _itit is the dam body vertical deformation amount predicted value of i-th kind of model t.

As preferably, described in second step, the mathematical model of Forecasting Methodology comprises stepwise regression method forecast model, grey GM(1,1) method forecast model, BP neural net method forecast model.

Beneficial effect: compared with prior art, the present invention has the following advantages: 1, the precision of prediction of dam vertical deformation amount is high, and the prediction period of Dam Deformation Monitoring is expanded.Through a large amount of case history Analysis of application result, the present invention is at forecast period, and comparatively stepwise regression method, gray method and BP neural net method, the precision that predicts the outcome improves 73.1%, 90.6% and 26.4% respectively.After precision of prediction improves, dam vertical deformation amount predict the outcome more close to real true, dam safety is run and the guiding value that scents a hidden danger more obvious; 2, economic benefit is obvious.The present invention can keep the continuity of model accuracy, and make accurately predicting to relating to safe development of deformation trend, take measures in advance, to preventing, dam safety accident is significant.There is significantly society and economic worth.

Embodiment

Below in conjunction with specific embodiment, illustrate the present invention further, these embodiments should be understood only be not used in for illustration of the present invention and limit the scope of the invention, after having read the present invention, the amendment of those skilled in the art to the various equivalent form of value of the present invention has all fallen within the application's claims limited range.

When describing specific implementation process, in conjunction with certain concrete engineering example, the inventive method is elaborated.

(1) case history background introduction

Certain the concrete dam monolith dam body monitoring materials of nearly a year 80 phase, adopt multipoint displacement meter to monitor this dam body vertical deformation amount, monitor (time, temperature, compressive stress, pore water pressure, pore water head, upper pond level etc.) environment parameter that this monolith is corresponding simultaneously.The measured data of 80 phases is as shown in table 1:

Certain Dam Deformation Monitoring measured data of table 1

(2) Data classification

Using 50 issues before 80 phase Monitoring Data according to as learning data M, rear 30 issues are according to as check data J.

(3) individual event Forecasting Methodology

This project example selects stepwise regression method, grey GM(1 respectively, 1) method, BP neural net method etc. are as individual event Forecasting Methodology.

1) stepwise regression method forecast model

Stepwise regression analysis is carried out to this Dam Deformation Monitoring measured data, final Environment variable: 1. time (D)/sky, 2. temperature (T)/° C, 3. pore water head (H)/m.

Set up deflection forecast model:

y＝a ₀+a ₁·D+a ₂·T+a ₃·H（7）

In formula, y perpendicular displacement; a _ifor the undetermined parameter (totally 4) of regression model; D is the cumulative time; T is temperature; H is pore water head absolute altitude.

Specific embodiment, substitutes into (7) formula one by one by M Monitoring Data, and can obtain M error equation, error equation general formula is:

v _i＝a ₀+a ₁·D _i+a ₂·T _i+a ₃·H _i-y _i

Being write as matrix form is:

$\underset{M\&times;1}{V}=\underset{M\&times;4}{A}.\underset{4\&times;1}{X}-\underset{M\&times;1}{y}---\left(8\right)$

By " least square method " principle, the estimated value of 4 undetermined parameter X in formula (7) can be obtained:

$\underset{4\&times;1}{X}=\left(\begin{array}{c}{a}_0\\ a_1\\ M\\ a_3\end{array}\right)={\left(A^{T}A\right)}^{-1}\&CenterDot;A^{T}y---\left(9\right)$

Specific embodiment, the Monitoring Data of 50 learning datas (in table 1 front 50 points) is substituted into (7) formula one by one, 50 error equations can be obtained, according to " least square method " principle, obtained the estimated value of 4 undetermined parameters in formula (7) by formula (9), the results are shown in Table 2.

Table 2 undetermined parameter a _iresult of calculation table

a 0＝-56.281

a 1＝0.055

a 2＝0.390

a 3＝0.528

Gradual regression analysis model data processed result is as shown in table 3:

Table 3a Gradual regression analysis model data processed result (unit: mm)

Gradual regression analysis model data processed result precision is as shown in table 3b:

Table 3b Gradual regression analysis model data processed result accuracy table

2) grey GM(1,1) method forecast model

Grey GM(1 is set up to the learning data in deformation measurement data, 1) model treatment.

Specific embodiment, by 50 learning datas (in table 1 front 50 points), through grey GM(1,1) process, Time Created, response function was:

y _t+1＝263.016881e ^0.026231t-261.966881

Grey GM(1,1) model data result is as shown in table 4:

Table 4 grey GM(1,1) model data result (unit: mm)

Grey GM(1,1) precision of model data result is as shown in table 5:

Table 5 grey GM(1,1) model data result accuracy table

3) BP neural net method forecast model

BP neural network model is set up to the learning data in deformation measurement data.Specific embodiment, 50 learning datas (in table 1 front 50 points) are carried out the training of BP neural network model, and parameter configuration is as follows:

Model structure: 6 × 18 × 1.That is: the input layer number of parameters of network is 6, and 6 input parameters are respectively: time/sky, temperature/DEG C, compressive stress/Mpa, pore water pressure/Kpa, pore water head/m and upper pond level/m; The hidden layer number of parameters of network is 18; The output layer number of parameters of network is 1, is dam body vertical deformation amount y _it.

BP neural network model data processed result is as shown in table 6:

Table 6BP neural network model data processed result (unit: mm)

The precision of BP neural network model data processed result is as shown in table 7:

Table 7BP neural network model data processed result accuracy table

4) determine based on degree of stability combined weights coefficient

According to based on degree of stability weight coefficient defining method, select Gradual regression analysis model, grey GM(1,1) model and BP neural network model as single model, set up built-up pattern.Set up through model and calculate with data analysis, choice accuracy factor-alpha and β are default value, and namely α gets 0 value, and β gets 1 value.The weight coefficient obtaining built-up pattern according to formula (1), formula (2), formula (3), formula (4) and formula (5) is:

w ₁＝0.275703，w ₂＝0.048617，w ₃＝0.675680

In this built-up pattern, w ₁for the weight coefficient of Gradual regression analysis model, w ₂for grey GM(1,1) weight coefficient of model, w ₃for the weight coefficient of BP neural network model.

Obviously, w ₁+ w ₂+ w ₃=1,0≤w _i≤ 1, i=1,2,3.

5) the dam vertical deformation amount forecast model based on degree of stability is set up

According to the dam vertical deformation amount forecast model that formula (6) is set up be:

y _t＝0.275703y _1t+0.048617y _2t+0.675680y _3t

Built-up pattern data processed result based on degree of stability power is as shown in table 8:

Table 8 built-up pattern data processed result (unit: mm)

Precision based on the built-up pattern data processed result of degree of stability weight coefficient defining method is as shown in table 9:

Table 9 is based on the built-up pattern data processed result accuracy table of degree of stability weight coefficient defining method

This example has 50 learning samples, and 30 inspection (prediction) samples, utilize forecast sample can evaluate the prediction effect of distinct methods.With the valuation of standard deviation evaluate its precision:

$\widehat{\mathrm{\&sigma;}}=\sqrt{\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{(y_t^{*}-y_t)}^2/n}---\left(10\right)$

In formula, be predicting the outcome of t phase distinct methods, y _tbe the measured value of t phase, n is the number of check point.Know by the definition of standard deviation, the valuation of the standard deviation of check point less, precision is higher, shows that prediction effect is better.Assay is in table 10.Compared with the BP neural network that precision is best, the precision of the inventive method prediction dam body vertical deformation amount can improve 26.4%.

Table 10 distinct methods testing accuracy comparison sheet

Claims (2)

1. a method for accurately predicting dam dam body vertical deformation amount, comprises the following steps:

The first step, Monitoring Data is divided into two parts in chronological order: a part is time preceding learning data M, and the number of M must be more than or equal to 10; Another part is the check data J of all the other times;

Second step, learning data M is utilized to set up the mathematical model of i kind Forecasting Methodology, wherein i=1,2 ... m; I >=2;

3rd step, calculate the degree of stability of various Forecasting Methodology:

A, dam time series monitor value are y _t, t=1,2 ... n, the individual event forecast model different according to the m kind of second step foundation, y _iti=1,2 ... m is i-th kind of Single model in the learning value of t or predicted value, calculates i-th kind of Single model at t period error e by formula (1) _it:

e _it＝y _t-y _it(1)

The precision A of i-th kind of Single model in the t phase is calculated by formula (2) _it:

$A_{it}=\left\{\begin{array}{cc}1,& 0\&le;\left|\frac{e_{it}}{y_t}\right|\&le;\mathrm{\&alpha;}\\ 1-\left|\frac{e_{it}}{y_t}\right|,& \mathrm{\&alpha;}<\left|\frac{e_{it}}{y_t}\right|<\mathrm{\&beta;}\\ 0,& \left|\frac{e_{it}}{y_t}\right|\&GreaterEqual;\mathrm{\&beta;}\end{array}\right.---\left(2\right)$

Wherein, 0≤α < β≤1, α acquiescence gets 0 value, and β acquiescence gets 1 value;

If first b i-th kind of Single model carry out N phase study, then carry out T phase prediction, calculate the average study precision ε of i-th kind of model by formula (3) _i, the consensus forecast precision η of i-th kind of model _i:

${\mathrm{\&epsiv;}}_i=\underset{t=1}{\overset{N}{\&Sigma;}}{A}_{it}/N,i=1,2,...m$

(3)

${\mathrm{\&eta;}}_i=\underset{t=N+1}{\overset{N+T}{\&Sigma;}}{A}_{it}/T,i=1,2,...m$

C, then, calculates i-th kind of model stability degree S by formula (4) _i:

$S_i=\frac{{\mathrm{\&eta;}}_i}{|{\mathrm{\&eta;}}_i-{\mathrm{\&epsiv;}}_i|+\mathrm{\&sigma;}}---\left(4\right)$

σ is infinitely small arbitrarily;

4th step, determine combined method model weight coefficient:

Make w _ibe the weight of i-th kind of model shared by m kind Single model, then built-up pattern weight coefficient is defined as by formula (5):

$w_i=\frac{s_i}{\underset{i=1}{\overset{m}{\&Sigma;}}s}---\left(5\right)$

5th step, according to formula (6), set up the dam vertical deformation amount forecast model based on degree of stability:

$y_t=\underset{i=1}{\overset{m}{\&Sigma;}}{w}_i{y}_{it}---\left(6\right)$

In formula, y _tfor the dam body vertical deformation amount predicted value of t, w _ifor the weight of i-th kind of model in built-up pattern, y _itit is the dam body vertical deformation amount predicted value of i-th kind of model t.

2. the method for accurately predicting dam dam body vertical deformation amount according to claim 1, it is characterized in that: the mathematical model of Forecasting Methodology described in second step comprises stepwise regression method forecast model, grey GM (1,1) method forecast model, BP neural net method forecast model.