CN104299048A - Method for predicting building vibration level caused by rail transit - Google Patents

Abstract

The invention relates to a method for predicting the building vibration level caused by the rail transit. Rail irregularity amplitude parameters, foundation soil property parameters, train speed parameters, train type parameters and building height parameters are collected to carry out normalization processing, wavelet transform is carried out by selecting the dbN wavelet function and the decomposition scale, sudden change factors and random effects in original signals are eliminated, denoised signals are obtained through reconstruction, the denoised signals serve as input variables of a support vector machine and is used for the training test of the support vector machine, finally, a calibrated SVM prediction model is obtained, the precision of prediction of the building vibration level caused by the rail transit is obviously improved, data support is provided for vibration effect evaluation and vibration reducing and denoising measure selection of urban rail transit development, and an important contribution is made for improving the urban working and living environment. Thus, the method can be widely applied to the technical field of civil engineering prediction.

Description

A kind of vibration of building horizontal forecast method that track traffic causes

Technical field

The present invention relates to a kind of buildings level of vibration Forecasting Methodology, particularly about a kind of vibration of building horizontal forecast method that track traffic causes.

Background technology

Track traffic is large with its freight volume, speed fast, take up an area the advantages such as few, plays more and more important role in transportation, but has also caused serious environmental problem simultaneously, and the evil of shaking caused by environment such as track traffics has been listed in one of seven large public hazards.This is not only because the ambient vibration that traffic system causes increases day by day, but also due to the progress along with society, the requirement of people to modern life quality is also more and more higher, even if for the vibration of same level, may not be problem in the past, and cause the kickback of the public more and more now.So the ambient vibration problem caused by track traffic, as a kind of novel environmental hazard, has become a more and more serious social concern.And predicted orbit traffic accurately all plays a part key on the impact of vibration of building for the construction of the potential level of vibration of timely perception and track traffic, reducing level of vibration in addition for taking appropriate measures also has important references to be worth.

The vibration of building caused by track traffic, by the impact of many factors, comprises track irregularity amplitude, foundation soil character, train speed, type of train, depth of building etc., there is very complicated relation between them.In these data procedures of actual treatment; usually can run into the situation that individual data departs from desired value or a large amount of statistical value; how to reject abnormal data wherein; the relation extracted key message and then find between vibration of building level and these factors, the vibration of building caused with reasonable prediction track traffic is that a difficulty is very large and be badly in need of the technical matters that solves.

Summary of the invention

For the problems referred to above, the object of this invention is to provide the vibration effect degree of a kind of predicted orbit traffic for buildings, thus the vibration of building horizontal forecast method providing the track traffic of Data support to cause for the vibration effect evaluation of Urban Rail Transit Development and the selection of vibration and noise reducing measure.

For achieving the above object, the present invention takes following technical scheme: a kind of vibration of building horizontal forecast method that track traffic causes, it comprises the following steps: 1) carry out raw data investigation for Train Track-roadbed-surrounding formation-building system, and this raw data comprises the parameter of the parameter of track irregularity amplitude, the parameter of foundation soil character, the parameter of train speed, the parameter of type of train and depth of building; 2) parameter each in raw data is normalized, and obtains each parameter after normalized; 3) each parameter after normalized is carried out respectively successively wavelet transformation decompose, by every for each parameter one deck respectively correspondence resolve into a low frequency signal and multiple high-frequency signal; 4) low frequency signal obtained for each parameter and high-frequency signal carry out denoising Processing, removing abnormal data; 5) respectively corresponding for each parameter after a denoising Processing low frequency signal and multiple high-frequency signal are added settling signal reconstruct in conjunction with the Weight of its decomposition scale, obtain each signal of weighting; 6) adopt Radial basis kernel function as the kernel function of SVM model, each signal based on weighting is demarcated the regularization term parameter C in kernel function and error term ε by gridding method, thus obtains the SVM model of demarcation.

Described step 2) comprise following content: formula parameter each in raw data be normalized one by one is as follows:

${\tilde{S}}_i=2\×\frac{S_i-S_{\min }}{S_{\max }-S_{\min }}-1$

Wherein, the value after normalized, S _iraw value, S _minthe minimum value in sample sequence, S _maxit is the maximal value in sample sequence.

Described step 4) comprise following content: the low frequency signal obtained for each parameter and high-frequency signal carry out denoising Processing, and its process is as follows: suppose that the model of the one-dimensional signal of a Noise is as follows:

s(k)＝f(k)+λ·e(k) k＝0,1,...,N-1

Wherein, f (k) is the useful signal after denoising, and s (k) is noisy signal, the noise signal of the signal that e (k) is denoising, and λ is weight, N represent each parameter carry out wavelet decomposition after low, the total number of high frequency.

Described step 6) comprise following content: the training sample of each signal as SVM model getting 70% weighting, for demarcating regularization term parameter C and error term ε, the span of C is (0,100), the span of ε is (0,0.4), and whether all the other are 30% correct as test sample book inspection SVM model, and adopt mean square deviation to carry out evaluating the precision of prediction of SVM model to test result, its mean square deviation formula is as follows:

$\mathrm{RMSE}={\left[\frac{{\mathrm{\Σ}}_{i=1}^n(V_i-{\hat{V}}_i)}{n}\right]}^{1/2}$

Wherein, represent a grade predicted value of shaking, V _irepresent and to shake a grade observed value, n is the quantity of grade observed value of shaking.

The present invention is owing to taking above technical scheme, it has the following advantages: the present invention is first by collecting the parameter of track irregularity amplitude, the parameter of foundation soil character, the parameter of train speed, the parameter of type of train and the parameter of depth of building are normalized, then dbN wavelet function and decomposition scale is selected to carry out wavelet transformation, eliminate the Mutagen in original signal and Random Effect, the signal after de-noising is obtained by reconstruct, and using the input variable of the signal after de-noising as support vector machine, training for support vector machine is tested, the SVM forecast model finally obtaining demarcating is to the precision of the vibration of building horizontal forecast significantly improving track traffic and cause, for the vibration effect evaluation of Urban Rail Transit Development and the selection of vibration and noise reducing measure provide Data support, important contribution has been made for improving urban work living environment.Therefore, the present invention can be widely used in civil engineering work electric powder prediction.

Accompanying drawing explanation

Fig. 1 is process schematic of the present invention

Fig. 2 is Train Track-roadbed-surrounding formation-surface of contact-building system model schematic of the present invention

Fig. 3 is the test site floor map that embodiments of the invention adopt

Fig. 4 is the wavelet noise process schematic that the present invention successively decomposes

Fig. 5 carries out three layers of wavelet noise process schematic decomposed for embodiment

Fig. 6 is the actual measurement mean value comparison diagram of predicted value and the vibration of building adopting the present invention to obtain

Embodiment

Below in conjunction with accompanying drawing and instantiation, illustrate the present invention further, these examples 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.

As shown in Figure 1, the vibration of building horizontal forecast method that a kind of track traffic of the present invention causes, it comprises the following steps:

1) as shown in Figure 2, raw data investigation is carried out for Train Track-roadbed-surrounding formation-building system, this raw data comprises the parameter of the parameter of track irregularity amplitude, the parameter of foundation soil character, the parameter of train speed, the parameter of type of train and depth of building, and wherein foundation soil character comprises foundation soil character damping ratio and foundation soil shear wave velocity.

In this example, in test site floor map as shown in Figure 3, test site is positioned near capital-wide line Shijiazhuang, subjects is a floor masonry residential building, building 6, buildings edge and nearest track space are from being only 14m, those skilled in the art are according to the data provided in Fig. 2 and Fig. 3, and method of data capture conveniently just can obtain corresponding raw data, does not describe in detail at this.

2) parameter each in raw data is normalized, and each parameter after obtaining normalized between 0 and 1, avoid there is dimension because of raw data or certain one-dimensional data is excessive and parameter calibration that is that cause is inaccurate, so that final SVM model prediction resultant error is larger.

Formula parameter each in raw data be normalized one by one is as follows:

${\tilde{S}}_i=2\×\frac{S_i-S_{\min }}{S_{\max }-S_{\min }}-1---\left(1\right)$

Wherein, the value after normalized, S _iraw value, S _minthe minimum value in sample sequence, S _maxit is the maximal value in sample sequence.Because parameters all in raw data all adopt formula (1) to process, therefore distinguish no longer one by one.

3) each parameter after normalized is carried out respectively N layer wavelet transformation to decompose, by every for each parameter one deck respectively correspondence resolve into a high-frequency signal A (N) and N number of low frequency signal D (1), D (2) ..., D (N), namely every one deck obtain a high-frequency signal A (N) and N number of low frequency signal D (1), D (2) ..., D (N), thus original signal is resolved into respectively a high-frequency signal and multiple low frequency signal;

As shown in Figure 4, in this example: N is the progression (number of plies) that wavelet transformation decomposes; As N=1, there is step phenomenon, differed greatly with original track irregularity amplitude signal in the track irregularity amplitude sequence indention after de-noising, can not fluctuating factor in reflected signal; As N>5, clearly can find out that the image of the track irregularity amplitude signal after denoising Processing is by excess smoothness, lost the part validity feature of original track irregularity amplitude sequence; As N=2 and N=4, after de-noising, the part stage of sequence and original series has larger difference; When N=3, the figure obtained can the fluctuation characteristic of reasonable reflection track irregularity amplitude signal.

4) low frequency signal obtained for each parameter and high-frequency signal carry out denoising Processing, removing abnormal data, thus improve the precision of SVM model prediction;

The low frequency signal obtained for each parameter and high-frequency signal carry out denoising Processing, and its process is as follows: suppose that the model of the one-dimensional signal (treating the signal of denoising) of a Noise is as follows:

s(k)＝f(k)+λ·e(k) k＝0,1,...,N-1 (2)

Wherein, f (k) is the useful signal after denoising, and s (k) is noisy signal, and the noise signal of the signal that e (k) is denoising, λ is weight.

Utilize Matlab (Matrix & Laboratory, matrix factory) middle wthresh (x, h, t) function, the threshold value process that threshold values is t is set to signal, wherein x is the matrix that will carry out threshold value setting, and h represents that threshold values selection rule is hard threshold values, and t gets famous threshold values form because progression (number of plies) with decompose after low, the total number of high frequency is identical, therefore in formula N represent each parameter carry out wavelet decomposition after low, the total number of high frequency, thus reject the random mutation factor comprised in noisy each parameter, remove noise, obtain each parameter effectively low, high-frequency signal.

5) respectively corresponding for each parameter after a denoising Processing low frequency signal and multiple high-frequency signal are added settling signal reconstruct in conjunction with the Weight of its decomposition scale, obtain each signal of weighting;

In the present embodiment, as shown in Figure 5, signal reconstruction is the common technology means of those skilled in the art to the process of signal reconstruction, therefore no longer describes in detail.

6) adopt Radial basis kernel function as the kernel function of SVM model, each signal based on weighting is demarcated the regularization term parameter C in kernel function and error term ε by gridding method, the span of C is (0,100), the span of ε is (0,0.4), thus obtain demarcate SVM model;

In above-described embodiment, generally get the training sample of each signal as SVM model of 70% weighting, for demarcating regularization term parameter C and error term ε, whether all the other are 30% correct as test sample book inspection SVM model, and usually adopt mean square deviation to carry out evaluating the precision of prediction of SVM model to test result, its mean square deviation formula is as follows:

$\mathrm{RMSE}={\left[\frac{{\mathrm{\Σ}}_{i=1}^n(V_i-{\hat{V}}_i)}{n}\right]}^{1/2}---\left(3\right)$

Wherein, represent a grade predicted value of shaking, V _irepresent and to shake a grade observed value, n is the quantity of grade observed value of shaking.

In the present embodiment, regularization term parameter C be 3.9821 and error term ε be 0.0064.

In order to verify validity of the present invention, as shown in Figure 6, the situation of actual measurement mean value and predicted value is contrasted, grade level that what the present invention obtained shake and grade level of shaking that actual measurement obtains very close, compared with objectively responding the propagation law vibrated between the soil body-buildings, there is higher precision, demonstrate the feasibility on vibration of building horizontal forecast that wavelet support vector machines causes in track traffic.

In sum, the present invention is first by collecting track irregularity amplitude, foundation soil character, train speed, type of train, depth of building data, then select wavelet function and decomposition scale to carry out wavelet transformation, eliminate the Mutagen in original signal and Random Effect, obtain the signal after de-noising by reconstruct.Using the input variable of the signal after de-noising as support vector machine, the training for support vector machine is tested, and finally obtains reliable forecast model.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (5)

1. the vibration of building horizontal forecast method that causes of track traffic, it comprises the following steps:

1) carry out raw data investigation for Train Track-roadbed-surrounding formation-building system, this raw data comprises the parameter of the parameter of track irregularity amplitude, the parameter of foundation soil character, the parameter of train speed, the parameter of type of train and depth of building;

2) parameter each in raw data is normalized, and obtains each parameter after normalized;

3) each parameter after normalized is carried out respectively successively wavelet transformation decompose, by every for each parameter one deck respectively correspondence resolve into a low frequency signal and multiple high-frequency signal;

4) low frequency signal obtained for each parameter and high-frequency signal carry out denoising Processing, removing abnormal data;

5) respectively corresponding for each parameter after a denoising Processing low frequency signal and multiple high-frequency signal are added settling signal reconstruct in conjunction with the Weight of its decomposition scale, obtain each signal of weighting;

6) adopt Radial basis kernel function as the kernel function of SVM model, each signal based on weighting is demarcated the regularization term parameter C in kernel function and error term ε by gridding method, thus obtains the SVM model of demarcation.

2. the vibration of building horizontal forecast method that causes of a kind of track traffic as claimed in claim 1, is characterized in that: described step 2) comprise following content:

Formula parameter each in raw data be normalized one by one is as follows:

${\tilde{S}}_i=2\×\frac{S_i-S_{\min }}{S_{\max }-S_{\min }}-1$

Wherein, the value after normalized, S _iraw value, S _minthe minimum value in sample sequence, S _maxit is the maximal value in sample sequence.

3. the vibration of building horizontal forecast method that causes of a kind of track traffic as claimed in claim 1, is characterized in that: described step 4) comprise following content:

The low frequency signal obtained for each parameter and high-frequency signal carry out denoising Processing, and its process is as follows: suppose that the model of the one-dimensional signal of a Noise is as follows:

s(k)＝f(k)+λ·e(k) k＝0,1,…,N-1

Wherein, f (k) is the useful signal after denoising, and s (k) is noisy signal, the noise signal of the signal that e (k) is denoising, and λ is weight, N represent each parameter carry out wavelet decomposition after low, the total number of high frequency.

4. the vibration of building horizontal forecast method that causes of a kind of track traffic as claimed in claim 2, is characterized in that: described step 4) comprise following content:

The low frequency signal obtained for each parameter and high-frequency signal carry out denoising Processing, and its process is as follows: suppose that the model of the one-dimensional signal of a Noise is as follows:

s(k)＝f(k)+λ·e(k) k＝0,1,…,N-1

Wherein, f (k) is the useful signal after denoising, and s (k) is noisy signal, the noise signal of the signal that e (k) is denoising, and λ is weight, N represent each parameter carry out wavelet decomposition after low, the total number of high frequency.

5. the vibration of building horizontal forecast method that causes of a kind of track traffic as claimed in claim 1 or 2 or 3 or 4, is characterized in that: described step 6) comprise following content:

Get the training sample of each signal as SVM model of 70% weighting, for demarcating regularization term parameter C and error term ε, the span of C is (0,100), the span of ε is (0,0.4), and whether all the other are 30% correct as test sample book inspection SVM model, and adopt mean square deviation to carry out evaluating the precision of prediction of SVM model to test result, its mean square deviation formula is as follows:

$\mathrm{RMSE}={\left[\frac{{\mathrm{\Σ}}_{i=1}^n(V_i-{\hat{V}}_i)}{n}\right]}^{1/2}$

Wherein, represent a grade predicted value of shaking, V _irepresent and to shake a grade observed value, n is the quantity of grade observed value of shaking.