CN106197252A - The method for building up of ground deformation position distribution measuring model based on parallel spiral transmission line - Google Patents

Abstract

The invention discloses the method for building up of ground deformation position distribution measuring model based on parallel spiral transmission line, including: parallel spiral transmission line is equally divided into some sections by (1), every section is stretched, utilizes otdr measurement device, collect reflected voltage signal；(2) reflected voltage signal is carried out pretreatment；(3) reflected voltage signal through step (2) pretreatment is carried out feature extraction, it is thus achieved that characteristic vector；(4) using described characteristic vector as input layer data, using stretch position as output layer data, least square method supporting vector machine model is set up.When the present invention utilizes parallel spiral transmission line to deform upon, its impedance can occur respective change, from parallel spiral transmission line one end input pulse signal, detect its reflected voltage signal, because reflected voltage signal comprises positional information, the forecast model of least square method supporting vector machine is set up, for accurately measuring deformation position based on this principle.

Description

The foundation of ground deformation position distribution measuring model based on parallel spiral transmission line Method

Technical field

The present invention relates to ground deformation test technical field, particularly relate to ground deforming position based on parallel spiral transmission line Put the method for building up of distribution measuring model.

Background technology

China is the multiple country of geological disaster, and geological disaster causes grievous injury to people's life's property frequently And massive losses.The local ground deformation that the geological movement such as landslide, surface subsidence causes is a kind of important disaster omen Phenomenon, is accurately positioned ground deformation position significant for the preventing and controlling of China's geological disaster.

Chinese scholars has made extensive work for the monitoring of ground surface deformation position, and monitoring means mainly has following several Class: (1) GPS technology and total powerstation observation technology.This kind of technology, by layouting in a large number on ground surface, measures three-dimensional coordinate change Thus obtain ground deformation position, but due to cost intensive of layouting, it is difficult to cover whole monitored area, big discharge observation can be caused Blind area.(2) optical fiber sensing technology.But fiber-draw amount is little, and can not significantly kinking, at the ground that deformation quantity is bigger In deformation, easily it is pulled off.(3) stay-supported sensing technology.Multiple sensor cloth are become network measure ground deflection by this technology, The certainty of measurement of deformation quantity can be reached 0.1mm, but not can determine that the particular location deformed upon.

CN102522148A discloses a kind of parallel spiral transmission line sensor, outer close around one layer at circular section silicon rubber bar Having the cable of two mutually insulateds, two cables to form helix, two cables are formed to wrap up in outside helix and are surrounded by silicon rubber Gum cover pipe, termination matching impedance Z L of two cables, another termination time domain reflectometer of two cables, but do not have Concrete deformation position method of testing is disclosed.

Summary of the invention

The invention provides the method for building up of ground deformation position distribution measuring model based on parallel spiral transmission line, profit Ground deformation position can be accurately tested with the model set up.

The method for building up of ground deformation position distribution measuring model based on parallel spiral transmission line, described parallel spiral biography Defeated line includes the elastic insulator at center, two parallel spiral signal transmission being wrapped in mutually insulated on described elastic insulator Line, and the insulation protective jacket of outer layer, two signal transmssion line one end are unsettled, and the other end connects otdr measurement device；

Described method for building up includes:

(1) parallel spiral transmission line is equally divided into some sections, every section is stretched, utilize otdr measurement to fill Put, collect reflected voltage signal；

(2) reflected voltage signal is carried out pretreatment；

(3) reflected voltage signal through step (2) pretreatment is carried out feature extraction, it is thus achieved that characteristic vector；

(4) using described characteristic vector as input layer data, using the distance between stretch position and starting point as output layer Data, set up least squares support vector machines model.

When parallel spiral transmission line does not has deformation, its characteristic impedance is fixing state, after deforming upon, deformation position Characteristic impedance change.At the Single port input pulse signal of parallel spiral transmission line, when impedance changes, pulse is believed Number reflected voltage signal change the most therewith, reflected voltage signal contains the characteristic information of deformation position.The present invention is just Based on this principle, establish least squares support vector machines model.

Described parallel spiral transmission line is the longest, and the ground deformation range that can be detected by is the biggest, generally, described parallel The length of helical transmission line is more than 2 meters.

For reducing the baseline drift of reflected voltage signal, need the signal collected is carried out pretreatment, described pretreatment Use variable standardization and normalization method.

The reflected voltage signal gathered can comprise more data, and all data constitute a vector.Owing to being local elongation, The only characteristic impedance of stretcher strain position just can change, and the most only part data contain parallel spiral transmission line impedance The characteristic information of change, needs to use characteristic variable extraction algorithm by feature information extraction out.

Preferably, described feature extraction uses successive projection method.

The kernel function of described least squares support vector machines model is gaussian radial basis function, least square-support vector Machine model particularly as follows:

$y\left(x\right)=\underset{k=1}{\overset{N}{\Σ}}{\mathrm{\α}}_{k}K(x,x_k)+b$

B is bias, and N is sample size, K (x, x_k) it is gaussian radial basis function, a_kLagrangian multiplier, y is defeated Going out layer data, x is input layer data, x_kFor sample input layer data.

The frequency acquisition of otdr measurement device is the highest, and the data that each reflected voltage signal comprises are the most, just calculates The biggest, general 10-1000MHz, the embodiment of the present invention is 100MHz,

Described pulse signal amplitude is relevant with the power of detection signal, and generally its amplitude is ± 1-100V, this Bright embodiment is ± 10V, and pulse width is 1-100ns, and the embodiment of the present invention is 10ns.

When the present invention utilizes parallel spiral transmission line to deform upon, its impedance can occur respective change, from parallel spiral biography Defeated line one end input pulse signal, detects its reflected voltage signal, because reflected voltage signal comprises positional information, based on this principle Set up the forecast model of least squares support vector machines, for accurately measuring deformation position.

Accompanying drawing explanation

Fig. 1 is the structural representation of the parallel spiral transmission line of the present invention.

Fig. 2 is the present invention parallel spiral transmission line local deformation structural representation.

The reflected voltage signal that when Fig. 3 is the stretching of parallel spiral transmission line diverse location, TDR measuring instrument collects.

Detailed description of the invention

As depicted in figs. 1 and 2, parallel spiral transmission line 6, including cylindrical elastic insulator 3, elastic insulator 3 The parallel spiral winding in surface signal transmssion line 1,2, outside is provided with insulation sleeve 4.Signal transmssion line 1,2 mutually insulated, can select Enamel-covered wire, the diameter of signal transmssion line 1,2 is probably at 0.25mm.The material of elastic insulator 3 and insulation sleeve 4 is silica gel, elastic The diameter of insulator 3 and insulation sleeve 4 is 3.5mm and 5.5mm respectively.

As in figure 2 it is shown, parallel spiral transmission line is carried out local elongation, at stretch position 7, deformation occurs.Signal transmssion line 1,2 one end are unsettled, and the other end connects otdr measurement device 5.Otdr measurement device is by pulse signal generation circuit, letter Number modulate circuit and data acquisition circuit are constituted, and the pulse signal amplitude that pulse signal generation circuit sends is ± 10V, pulse width Degree is 10ns.Signal conditioning circuit has the effect of isolation, is avoided that reflected voltage signal is accepted to interfere by pulse signal.Instead Penetrate voltage signal to amplify and filtering through signal conditioning circuit, data acquisition circuit collection send to host computer.Data acquisition current collection The employing frequency on road is 100MHz, and the data every time gathered are 250.

Measurement model method for building up of the present invention is specific as follows:

A (), with end that parallel spiral transmission line and otdr measurement device connect as starting point, does a mark every 10cm Note, carries out one-off drawing between every two adjacent labellings, uses two to step up wood fixture, often by two mark positions during stretching Secondary stretching 2cm.When stretching, use otdr measurement device 5 (TDR measuring instrument) to gather the anti-of parallel spiral transmission line 6 every time Penetrate voltage signal, and send the signal to host computer, host computer data processed and show stretch position.

B parallel spiral transmission line total length that () present invention uses is 5 meters, stretches once every 10cm, obtains 49 groups altogether Reflected voltage signal, often group signal packet is containing 250 data, and one has 250 × 49 data, Fig. 3 be parallel spiral transmission line not The reflected voltage signal that during co-located stretching, TDR measuring instrument collects.

C (), for the impact reducing instrument state, reflected voltage signal measurement is brought by experimental situation change, needs instead Penetrate voltage signal and carry out pretreatment.Common Preprocessing Algorithm includes smothing filtering, derivative correction, multiplicative scatter correction, variable Standardization and normalization etc..The present embodiment first uses variable standardization method for eliminating the baseline drift of reflected voltage signal, so Reflected voltage data are processed by rear employing normalization method, to eliminate the random error of reflected voltage signal.

D () stretching every time is measured the reflected voltage signal obtained and is comprised 250 data, owing to being local elongation, and only have The characteristic impedance of stretcher strain position just can change, and therefore in these 250 data, only part contains parallel spiral biography The characteristic information of defeated line impedence change, needs to use characteristic variable extraction algorithm by feature information extraction out, and the present embodiment is adopted Characteristic variable extraction algorithm be successive projection algorithm.The basic ideas of successive projection algorithm are: randomly choose 49 groups of reflections In voltage signal 36 groups, as modeling collection, remain 13 groups as checking collection.By modeling collection data are carried out projection mapping structure Produce new characteristic variable set, set up multiple linear regression (MLR) model, then profit successively according to these characteristic variable set It is estimated with verifying that MLR model is predicted the outcome by collection data, thus selects the feature containing bottom line redundancy Variables collection, i.e. characteristic vector.After feature variables selection, each group of reflected voltage signal comprises 27 data.

(e) pretreatment and the later reflected voltage signal of feature variables selection as input layer data, with stretch position with Distance between starting point, as output layer data, according to modeling collection, sets up least squares support vector machines model.If modeling collection D ={ (x₁,y₁),(x₂,y₂),…,(x_k,y_k), 1≤k≤N, in the present embodiment, N is 36.Input layer data x in set_k∈R^N, defeated Go out layer data y_k∈R；Then a nonlinear function is utilizedBy x_kIt is mapped to higher dimensional space and sets up regression model:

In above formula, b is bias, w ∈ R^NFor weight vector.The Function Fitting of least squares support vector machines model is asked Topic can be converted into and solve below equation:

$\min J(w,e)=\frac{1}{2}{w}^{T}w+\frac{1}{2}\mathrm{\γ}\underset{k=1}{\overset{n}{\Σ}}{e}_k^2---\left(2\right)$

Above formula constraints is:

Wherein, e_kFor error variance,_γIt is regularization parameter, controls the punishment degree beyond error sample,For kernel mappings function.Changing above formula to dual spaces, the form obtaining Lagrangian is:

Lagrangian multiplier a in formula_k∈ R is referred to as supported value, variable each to above formula ask partial derivative can obtain with Lower equation:

After variable w and e is iterated elimination, system of linear equations can be obtained:

$\left[\begin{array}{cc}0& {\stackrel{\→}{1}}^T\\ {\stackrel{\→}{1}}^T& {\mathrm{\Ω}}_{k,l}+{\mathrm{\γ}}^{-1}I\end{array}\right]\left[\begin{array}{c}b\\ \mathrm{\α}\end{array}\right]=\left[\begin{array}{c}0\\ y\end{array}\right]---\left(5\right)$

In formula

Wherein K (x_k,x_l) it is kernel function, the present embodiment uses RBF function, and computing formula is as follows:

$K(x,x_k)=\exp \left\{\frac{-\left|\right|x-x_k|{|}^2}{2{\mathrm{\σ}}^2}\right\}---\left(7\right)$

In formula (7), σ is the radius of RBF kernel function, and kernel function σ affects the sample distribution complexity in feature space.? Model of fit to least squares support vector machines (LS-SVM) is:

$y\left(x\right)=\underset{k=1}{\overset{N}{\Σ}}{\mathrm{\α}}_{k}K(x,x_k)+b---\left(8\right)$

After selecting RBF kernel function, LS-SVM model also needs to determine regularization parameter γ and RBF kernel functional parameter σ.Utilize Tunelssvm function in Matlab software LS-SVM tool kit, uses complete way of search, and Search Range is set to [10^-6, 10⁶], it being calculated γ and σ and be respectively 6448 and 166287, the trainlssvm function in recycling LS-SVM tool kit just may be used To set up least squares support vector machines model.Remaining 13 groups of reflected voltage signal are used for entering forecast model as checking collection Row assessment, assessment parameter includes root-mean-square error (RMSE) and the coefficient of determination (R²), the present embodiment least squares support vector machines RMSE and R of model²It is respectively 2.390 and 0.993.

In order to contrast, using modeling collection input layer data as independent variable, modeling collection output layer data as dependent variable, Set up linear offset minimum binary forecast model:

Y=A*X+b

Wherein Y is parallel spiral transmission line stretch position, and X is reflected voltage signal, and A is coefficient, and b is constant.

F () parallel spiral transmission line deformation position on-line measurement: the certain point of fixing parallel spiral transmission line, gathers this point Position stretches Time Domain Reflectometry voltage signal when more than 2 centimetres, after this signal is carried out pretreatment and feature extraction, generation respectively Enter to least squares support vector machines model with partial least square model, be calculated the deformation position of parallel spiral transmission line Put, the results are shown in Table 1.

Table 1

As can be seen from Table 1, supporting vector machine model records parallel spiral transmission line deformation position and actual value More closely, be better than partial least square model.This is because, parallel spires contains distributed inductance and distributed capacitor, when During high-frequency signal effect, distributed capacitor and distributed inductance occupy leading position in impedance variation, and this means that when flat During the stretching of row helix, reflected voltage signal comprises the nonlinear transformations of a large amount of impedance variation.Compared to linear offset minimum binary Model, supporting vector machine model can preferably utilize these nonlinear transformations.

In actual applications, for improving the coupling of parallel spiral transmission line 6 and Rock And Soil, first parallel spiral biography is being buried underground The Rock And Soil of defeated line 6 digs a wide 30cm, the ditch of deep 30cm, is then the size of 10cm by cross-sectional diameter, pours strip into Hands pinches plastic yellow sand cement mixing slurry.Again parallel spiral transmission line 6 is imbedded, not yet solidify at yellow sand cement mixing slurry Time, the ground dug up is covered up, so when, after the slurry solidification of yellow sand cement mixing, parallel spiral transmission line 6 will be by one layer Body of cement is wrapped up.Can drive body of cement when Rock And Soil displacement, body of cement can rupture and drive parallel spiral transmission line 6 Stretching, measurement system just can detect the deformation of this section of Rock And Soil.

Claims (8)

1. the method for building up of ground deformation position distribution measuring model based on parallel spiral transmission line, described parallel spiral transmission Line includes the elastic insulator at center, two parallel spiral signal transmission being wrapped in mutually insulated on described elastic insulator Line, and the insulation protective jacket of outer layer, two signal transmssion line one end are unsettled, and the other end connects otdr measurement device；

Described method for building up includes:

(1) parallel spiral transmission line is equally divided into some sections, every section is stretched, utilize otdr measurement device, adopt Collection obtains reflected voltage signal；

(2) reflected voltage signal is carried out pretreatment；

(3) reflected voltage signal through step (2) pretreatment is carried out feature extraction, it is thus achieved that characteristic vector；

(4) using described characteristic vector as input layer data, using the distance between stretch position and starting point as output layer data, Set up least squares support vector machines model.

2. method for building up as claimed in claim 1, it is characterised in that described parallel spiral length of transmission line is more than 2 meters.

3. method for building up as claimed in claim 1, it is characterised in that described pretreatment uses variable standardization and normalization Method.

4. method for building up as claimed in claim 1, it is characterised in that described feature extraction uses successive projection method.

5. method for building up as claimed in claim 1, it is characterised in that the data acquiring frequency of described otdr measurement device For 10-1000MHz.

6. method for building up as claimed in claim 1, it is characterised in that the amplitude of described pulse signal is ± 1-100V, pulse Width is 1-100ns.

7. method for building up as claimed in claim 1, it is characterised in that the core letter of described least squares support vector machines model Number is gaussian radial basis function.

8. method for building up as claimed in claim 1, it is characterised in that every segment length is 5-15cm, carries out every section stretching 1- 3cm。