CN109885977A

CN109885977A - A method for predicting bank slope deformation

Info

Publication number: CN109885977A
Application number: CN201910227298.2A
Authority: CN
Inventors: 夏敏; 任光明; 李天斌
Original assignee: Chengdu Univeristy of Technology
Current assignee: Chengdu Univeristy of Technology
Priority date: 2019-03-25
Filing date: 2019-03-25
Publication date: 2019-06-14

Abstract

The invention discloses a method for predicting bank slope deformation, comprising the following steps: step 1: collecting displacement monitoring data, decomposing the total displacement into different response components according to the action forms of various deformation influencing factors; step 2: decomposing the displacement time series, eliminating The "unstable" component of the time series is to eliminate the trend item displacement; Step 3: Use the grey GM(l,1) model to extract the trend item displacement, and the BP neural network model analyzes the deviation amount displacement induced by the reservoir water or rainfall. The predicted value of the total landslide displacement can be obtained by superposing the displacement of the trend term and the periodic term. The advantage of the present invention lies in that: based on the landslide deformation mechanism and evolution stage, and considering the contribution of each inducing factor to the landslide displacement, a grey-neural network displacement prediction model based on multiple inducing factors is established, and the model can be applied to the external dynamic factors. At the same time, it can realize not only medium and long-term prediction of landslide deformation, but also short-term and impending landslide prediction.

Description

A kind of bank slope Deformation Prediction method

Technical field

The present invention relates to Landslide Prediction technical field, in particular to a kind of bank based on Grey Neural Network built-up pattern Slope Deformation Prediction method.

Background technique

Due to landslide formation condition, the complexity of risk factor, diversity and its randomness of variation, so as to cause landslide The extremely difficult capture of multidate information, the prediction on landslide are universally acknowledged problems.

Slide prediction method is roughly divided into experimental forecast method, statistical analysis forecasting procedure, nonlinear prediction method, physics Mechanics forecasting procedure and Comprehensive prediction method.Wherein, statistical analysis forecasting procedure, which mainly passes through, introduces different statistical models, comes It is fitted landslide displacement-time graph, main representative has grey GM (l, 1) model, biological growth model (Verhulst mould Type), curvilinear regression analysis model, nonlinear multivariable relevant function method and exponential smoothing etc..However Landslide Deformation develops by all More extraneous factors influence, such as rainfall, earthquake, the variation of library water, and existing statistical model such as grey GM (l, 1) model can not be anti- Reflect displacement with extraneous factor changing rule, and the generation of displacement of inclined plane be control by slopes itself geologic structure condition and outside Boundary's risk factor it is coefficient as a result, thus carried out by landslide displacement-time graph that traditional statistical model is fitted Displacement prediction is unreasonable.Therefore, Prediction of Displacement in Landslide should consider each risk factor to landslide position based on deformation evolution mechanism The Prediction of Displacement in Landslide model under more risk factor effects is established in the contribution of shifting, this is also theory and practice significance of the invention Place.

The prior art one related to the present invention

Landslide System is a gray system, and gray system refers to INFORMATION OF INCOMPLETE or insufficient system^[1].Grey Systems Theory mainly by the generation to system " part " Given information, exploitation, the valuable information of extraction, is realized to system row For correct understanding^[2].Gray model is will likely be after irregular original deformation data Accumulating generation, so that it becomes relatively there is rule Differential Equation Model is resettled after the generation ordered series of numbers of rule.So grey GM (1,1) model is actually generation series model.Cause And the obtained data of GM (1,1) model must could use after inverse accumulated generating restores, GM (1,1) model is exponential model, The deformation time series that can increase for index of coincidence are predicted.

If the displacement data ordered series of numbers for the time intervals such as original that come down are as follows:

X⁽⁰⁾=(x⁽⁰⁾(1),x⁽⁰⁾(2),…,x⁽⁰⁾(k)…,x⁽⁰⁾(n)) (k=1,2 ..., n)

In formula: n is the total number of monitoring data, and k is time serial number, x⁽⁰⁾(k) be the kth moment displacement monitoring data.

It utilizesTo X⁽⁰⁾Ordered series of numbers makees one-accumulate and generates (AGO) transformation:

X⁽¹⁾=(x⁽¹⁾(1),x⁽¹⁾(2),…,x⁽¹⁾(k)…,x⁽¹⁾(n))

To x⁽¹⁾Establish single order albinism differential equation^[3]:

Undetermined coefficient a, b is solved using least square method:

Wherein:

Available GM (1,1) grey forecasting model are as follows:

Regressive (difference) operation is carried out to model value and obtains original series simulation and forecast value:

The shortcomings that prior art one

Currently with, there are apparent defect, being mainly manifested in: not in grey GM (1,1) model prediction bank slope Study on Deformation The Prediction of Displacement in Landslide model established based on Landslide Deformation mechanism of Evolution is unreasonable；Do not consider bank slope displacement by extraneous factor The prediction model that the rule of influence is established is unreasonable；It is unreasonable that prediction model is not established by the Landslide Deformation stage.Meanwhile ash Application of color GM (1, the 1) model in Slip moinitoring predominantly carries out mid-and-long term forecasting to Landslide Deformation, but to cunning Slope is short-term and to face sliding prediction precision poor, or even cannot be applicable in, therefore urgently improve to make and be both applicable to come down The mid-and-long term forecasting of deformation, and can be adapted to short-term and face sliding prediction.

Bibliography

[1] Liu Sifeng, Guo Tianbang gray system theory and application break a seal: publishing house, He'nan University, and 1991；

[2] comparison application of Li Xiuzhen, Kong Jiming, Wang Chenghua grey GM (1,1) Residual Error Modified Model in Landslide Prediction [J] mountain research, 2007,25 (6): 741-746；

[3] Wang Jianfeng Quantitative prediction of landslide using S-curve [J] Chinese Geological Disasters and prevention and treatment journal, 2003,14 (2): 1-8。

Summary of the invention

The present invention in view of the drawbacks of the prior art, provides a kind of bank slope Deformation Prediction method, can effectively solve above-mentioned Problem of the existing technology.

In order to realize the above goal of the invention, the technical solution adopted by the present invention is as follows:

A kind of bank slope Deformation Prediction method, comprising the following steps:

Step 1: collecting bank slope displacement monitoring data, displacement total amount is decomposed into according to each factors influencing deformation action mode Different response component, since the generation of displacement of inclined plane is to be controlled by slopes itself geologic structure condition and extraneous risk factor It is coefficient as a result, displacement is divided into following four, and can be indicated with following model:

A_t=t_t+c_t+s_t+ε_t

In formula: t_t、c_t、s_t、ε_tRespectively deterministic trend term, periodic term, pulsation item and uncertain random change Amount；

Step 2: displacement time series decompose.Resolution of displacement is carried out using moving average method, by the trend term in total displacement Reject, it is a kind of data processing method of filtering, with linear multi moving average method be rely on, reject time series " no Stablize " ingredient, find out sequences y₁, y₂... y_nSeveral average values early period and later period average value, to construct a new sequence Column, and this new sequence is smoother.Its mathematic(al) representation are as follows:

It is 3 sliding averages if k=1；K=2 is 5 sliding averages, and so on.F resulting in this way_tRandom rise Volt is reduced than original data, so that curve is more smooth, because of referred to herein as smoothed data.Meanwhile taking its residual error are as follows:

e_t=y_t-f_t

After sliding average, frequent random fluctuation in data is filtered, shows smooth variation tendency, the variation of random error Process, the i.e. bias of trend term, thus landslide periodic term displacement is obtained, periodic term displacement is made jointly by multiple influence factors Complex nonlinear sequence.

Step 3: with grey GM (l, 1) model extraction trend term be displaced, BP neural network model analysis by library water load or The bias displacement that rainfall induces, finally, the total position in landslide can be obtained in trend term and the superposition of periodic term displacement prediction value Move predicted value.

Further, steps are as follows for BP neural network model analysis bias displacement:

One typical BP neural network consists of three layers, i.e. input layer, hidden layer and output layer, realizes between each layer complete Connection.Each layer is made of several neuron nodes, and the output valve of each neuron node is by input value, action function and threshold Value determines that principle is as follows:

1) network inputs sample mode, is set are as follows:

A_i=(x_i1,x_i2,…,x_in) (i=1,2 ..., m)

In formula: m is mode of learning logarithm, and n is input layer unit number.

2), corresponding output vector are as follows:

B_i=(y_i1,y_i2,…,y_ik) (i=1,2 ..., m)

In formula: m output mode number corresponding with input pattern, k are input layer unit number.

3) input of hidden layer each unit, is calculated:

Wherein: w_ijFor the continuous power of input layer to middle layer；θ_ajFor the threshold value of implicit function unit, j is the nerve of hidden layer First number.

4), with S_jThe output of hidden layer each unit is calculated by lower array function as independent variable:

(S type function)

(tangsig function)

5), information flows to output layer from input layer, calculates input, the output of output layer unit:

Y_i=f (L_i)

In formula: Y_iFor the output of output layer, v_jiFor middle layer to output layer connection weight；γ_iFor output layer unit threshold value；f () is the function of S.

6), according to the connection weight and threshold value between the size automatic adjustment output layer of error, hidden layer and output layer, repeatedly Training, until making the global error of network tend to minimum E.

After E, which is less than a certain precision of prediction, to be required, show that the network has been learnt well, so that it may according to new input value pair Bank slope deformation is predicted.

Compared with prior art the present invention has the advantages that

Developing for Landslide Deformation is influenced (such as rainfall, earthquake, library water change) by many extraneous factors, solves grey GM (1,1) statistical model can not reflect that displacement with the changing rule of extraneous factor, that is, considers each risk factor to landslide displacement Contribution, is established Prediction of Displacement in Landslide model, and by the Landslide Deformation stage based on Landslide Deformation mechanism of Evolution with La Xiwa For fruit foretells deformable body, the displacement of bank slope platform 1# section representativeness monitoring point is carried out using Prediction of Displacement in Landslide model pre- It surveys, the results showed that using the GM of the displacement prediction built-up pattern set stage by stage to bank slope foundation based on bank slope deformation impact factor The advantages of (1,1) model and BP neural network model, foretelling bank slope deformation more suitable for similar fruit is to have obvious deformation evolution rank The displacement prediction of the deformed slope of section property.Meanwhile it is (i.e. grey based on grey GM (1,1) model after BP neural network model optimization Color-neural network displacement prediction built-up pattern) it may be not only suitable for being influenced by outer power (rainfall, the variation of library water etc.) factor Landslide Deformation prediction, and can realize the mid-and-long term forecasting and short-term forecast for Landslide Deformation.

Detailed description of the invention

Fig. 1 m- rate of displacement curve graph when being profiling observation point of the embodiment of the present invention；

Fig. 2 is Guo Bu bank slope of embodiment of the present invention area daily rain amount observational data curve graph；

Fig. 3 is the displacement prediction value and measured value curve graph in each water storage stage of the embodiment of the present invention；

Fig. 3 a is the first water storage periodical trend item displacement prediction and actual measurement；

Fig. 3 b is the first water storage stage bias displacement prediction and actual measurement；

Fig. 3 c is the second water storage periodical trend item displacement prediction and actual measurement；

Fig. 3 d is the second water storage stage bias displacement prediction and actual measurement；

Fig. 3 e is third water storage periodical trend item displacement prediction and actual measurement；

Fig. 3 f is third water storage stage bias displacement prediction and actual measurement；

Fig. 3 g is the 4th water storage periodical trend item displacement prediction and actual measurement；

Fig. 3 h is the 4th water storage stage bias displacement prediction and actual measurement.

Specific embodiment

To make the objectives, technical solutions, and advantages of the present invention more comprehensible, below in conjunction with attached drawing and embodiment is enumerated, The present invention is described in further details.

A kind of bank slope Deformation Prediction method based on Grey Neural Network built-up pattern, which comprises

Step 1: collecting bank slope displacement monitoring data, displacement total amount is decomposed into according to each factors influencing deformation action mode Different response component, since the generation of displacement of inclined plane is to be controlled by slopes itself geologic structure condition and extraneous risk factor It is coefficient as a result, thus displacement can be divided into following four, and can be indicated with following model:

A_t=t_t+c_t+s_t+ε_tFormula 5

In formula: t_t、c_t、s_t、ε_tRespectively deterministic trend term, periodic term, pulsation item and uncertain random change Amount.Trend term in displacement time series is usually the increasing function of time, is controlled by slopes itself geologic structure condition, drop The other factors control periodic term displacement such as rain, reservoir level variation.Therefore, landslide displacement sequence can be expressed as trend term and periodic term position The sum of move.

e_t=y_t-f_tFormula 7

1) network inputs sample mode, is set are as follows:

A_i=(x_i1,x_i2,…,x_in) (i=1,2 ..., m) formula 8

In formula: m is mode of learning logarithm, and n is input layer unit number.

2), corresponding output vector are as follows:

B_i=(y_i1,y_i2,…,y_ik) (i=1,2 ..., m) formula 9

3) input of hidden layer each unit, is calculated:

Y_i=f (L_i) formula 14

Application example of the present invention is as follows:

By taking the power station La Xiwa fruit foretells deformable body as an example, fruit foretells bank slope rate of displacement and reservoir level correlation (figure 1), i.e. displacement bias is mainly influenced by reservoir level, since monitoring in addition to there is a small amount of rainfall in the summer of, in 2009 in 2012 (Fig. 2), bank slope area totality rainfall is smaller, therefore, uses moving average method by resolution of displacement for trend term incremental at any time Displacement and the bias displacement induced by the load of library water, rainfall, using the variable quantity of reservoir level as BP neural network model The main input factor, rainfall as the secondary input factor, consider day Reservoir Water Level amount, moon reservoir level accumulated change respectively Amount, daily rainfall and moon accumulation rainfall foretell bank slope displacement bias to fruit as the input impact factor of BP neural network model It is analyzed, establishes the bank slope of reservoir Deformation Prediction in each water storage stage based on more impact factors such as reservoir level, rainfall Grey Neural Network built-up pattern.

To on July 15th, 2014 from monitoring on August 15th, 2009, comprehensive displacement accumulation curve and rate of displacement curve are special It was in four stages that sign, which shows bank slope Displacement Development obviously:

First stage: " rate of displacement obviously increases the stage ", the displacement curve angle of contingence are in increased trend (2009.08.15-2010.2.22 water storage level is raised to 2420m for the first time, then falls after rise to 2400m)；

Second stage: " rate of displacement first reduces the stage ", an opposite stage deform, and rate of displacement is in decreasing trend (2010.2.23-2011.4.28 water storage level is stablized after being raised to 2430m from 2400m to 2430m)；

Phase III: " rate of displacement second reduces the stage ", (2011.4.29-2012.12.30 water storage level is lifted from 2430m Stablize after rising to 2448m to 2448m), rate of displacement is generally less than 10mm/d；

Fourth stage: " rate of displacement tends towards stability the stage ", rate of displacement are generally less than (1~2) mm/d (2013.1.1- 2014.7.15 water storage level is stablized to 2448m), and rate of displacement tends towards stability.

Below by taking No. 79 points as an example, bank slope displacement time series are divided into the above four-stage, establish based on reservoir level, The Grey Neural Network built-up pattern of the bank slope of reservoir Deformation Prediction in each water storage stage of more impact factors such as rainfall:

Trend item parts and deviation are decomposed into using displacement time series of the moving average method to the aforementioned four water storage stage Part is measured, using GM (1,1) models fitting trend term, by day Reservoir Water Level amount, moon reservoir level accumulated change amount, daily rainfall And moon accumulation rainfall is fitted as the input factor pair displacement bias of BP neural network model, bank slope 2009 9 The displacement measured value and predicted value result such as Fig. 3 in the moon~in November, 2013.

Displacement measured value shows the model using BP neural network training with predicted value comparison result, to training data Fitting degree it is high, except GM (1, l) model is poor to the data fitting degree in the first water storage stage outer, generally GM (1, l) mould The predicted value error of type and BP neural network model is lower than 10%, and displacement prediction curve is consistent with measured curve trend, fitting effect Fruit is good, therefore, using the displacement prediction built-up pattern set stage by stage to bank slope foundation based on deformation impact factor GM (1,1) The advantages of model and BP neural network model, foretelling bank slope deformation more suitable for similar fruit has obvious deformation evolution interim The displacement prediction of deformed slope.

According to monitoring data of displacement, analysis shows, bank slope rate of deformation tends towards stability at present, most of monitoring point rate of displacement Less than 1~2mm/d, bank slope deformation is in the opposite stabilization sub stage, and therefore, fourth stage displacement prediction built-up pattern is suitable for 2014 Bank slope Deformation Prediction since year, with the displacement in the 4th water storage stage (01 month~in November, 2013 in 2013), Reservoir Water Level amount And the trained BP neural network model of rainfall product data is the prediction model of bias part, chooses the forecast sample time below Section is 01 month~in June, 2014 in 2014, using the accumulative displacement-time for GM (l, 1) this time of model prediction established The trend term of sequence, GM (l, 1) model of No. 79 monitoring point fourth stages are as follows:

To accumulate day Reservoir Water Level amount, moon reservoir level accumulated change amount, daily rainfall and the moon rainfall as BP nerve The input factor of network model is predicted the bias of accumulative displacement-time series using BP model has been trained, most The trend term of prediction is added to the total displacement amount up to prediction with bias afterwards.Based on the Grey Neural Network combination established The displacement prediction result such as table 1 of the bank slope platform 1# section representativeness monitoring point of model prediction.As shown in Table 1, with predicted time Increase, prediction error is gradually increased, and the prediction effect of in January, 2014~3 month section is relatively preferable.Since bank slope becomes at this stage Shape rate totally tends towards stability, and prediction technique proposed by the present invention, the selection number of roller can be used in later period bank slope displacement prediction According to prediction model is established, displacement prediction accuracy is higher.

Table 1 is displaced prediction result (part) (position based on the platform 1# profile monitoring point of Grey Neural Network built-up pattern It moves: mm)

Those of ordinary skill in the art will understand that the embodiments described herein, which is to help reader, understands this hair Bright implementation method, it should be understood that protection scope of the present invention is not limited to such specific embodiments and embodiments.Ability The those of ordinary skill in domain disclosed the technical disclosures can make its various for not departing from essence of the invention according to the present invention Its various specific variations and combinations, these variations and combinations are still within the scope of the present invention.

Claims

1. a method for predicting bank slope deformation, is characterized in that, comprises the following steps:

Step 1: Collect the monitoring data of the slope displacement, and decompose the total displacement into different response components according to the action forms of various deformation influencing factors. Since the slope displacement is controlled by the geological structure conditions of the slope itself and the combined action of external induced factors , the displacement is divided into the following four terms, and can be represented by the following model:

A _t =t _t +c _t +s _t +ε _t

where: t _t , _ct , s _t , and ε _t are deterministic trend items, periodic items, pulsation items and uncertain random variables, respectively;

Step 2: Decomposition of displacement time series; the displacement decomposition is carried out by the moving average method, and the trend item in the total displacement is eliminated. "Unstable" component, obtain several pre-averages and post-averages of the sequence y ₁ , y ₂ , ... y _n , so as to construct a new sequence, and this new sequence is relatively smooth; its mathematical expression is :

If k=1, it is a 3-point moving average; k=2, it is a 5-point moving average, and so on; the random fluctuation of the obtained f _t is reduced compared with the original data, making the curve smoother, so it is called smooth data; At the same time, take its residual as:

e _t =y _t -f _t

After the moving average, the frequent random fluctuations in the data are filtered out, showing a smooth change trend, the change process of the random error, that is, the deviation of the trend term, thus obtaining the landslide periodic term displacement. The periodic term displacement is affected by multiple influencing factors. complex nonlinear sequence;

Step 3: Use the grey GM(l,1) model to extract the trend term displacement, BP neural network model analyzes the offset displacement induced by reservoir water loading or rainfall, and finally, superimpose the trend term and periodic term displacement prediction values to get Predicted value of total landslide displacement.

2. The method according to claim 1, characterized in that: adopting the BP neural network model to analyze the offset displacement induced by reservoir water loading or rainfall, and its concrete steps are as follows:

A typical BP neural network consists of three layers, namely the input layer, the hidden layer and the output layer, and each layer is fully connected; each layer is composed of several neuron nodes, and the output value of each neuron node is determined by the input. The value, action function and threshold are determined, and the principle is as follows:

1), set the network input sample mode as:

A _i =(x _i1 ,x _i2 ,...,x _in ) (i=1,2,...,m)

In the formula: m is the logarithm of learning mode, n is the number of input layer units;

2), the corresponding output vector is:

B _i =(y _i1 ,y _i2 ,...,y _ik ) (i=1,2,...,m)

where m is the number of output modes corresponding to the input mode, and k is the number of input layer units;

3) Calculate the input of each unit of the hidden layer:

Where: w _ij is the continuous weight from the input layer to the middle layer; θ _aj is the threshold of the hidden function unit, j is the number of neurons in the hidden layer;

4), with S _j as the independent variable, calculate the output of each unit of the hidden layer through the following function:

(Sigmoid function)

(tangsig function)

5) The information flows from the input layer to the output layer, and the input and output of the output layer unit are calculated:

Y _i =f(L _i )

In the formula: Y _i is the output of the output layer, v _ji is the connection weight from the middle layer to the output layer; γ _i is the output layer unit threshold; f() is the function of S;

6) Automatically adjust the connection weights and thresholds between the output layer, the hidden layer and the output layer according to the size of the error, and train repeatedly until the global error of the network tends to the minimum value E;

When E is less than a certain prediction accuracy requirement, it indicates that the network has been learned, and the bank slope deformation can be predicted according to the new input value.