CN113138978B - Beidou data filling and deformation prediction method for urban differential settlement monitoring - Google Patents

Abstract

The invention provides a Beidou data filling and deformation prediction method for urban differential settlement monitoring, which comprises the following steps of: s1: carrying out high-precision calculation on the settlement monitoring data based on the Beidou to obtain a Beidou monitoring calculation result; s2: carrying out optimal data prediction on the missing data in the Beidou monitoring and resolving result to obtain a prediction result; s3: filling missing data according to the prediction result to obtain a continuous Beidou monitoring and resolving result; s4: segmenting continuous Beidou monitoring and resolving results, fitting each segment of data to obtain a settlement monitoring curve, and completing filling and deformation prediction of Beidou data. The invention provides a Beidou data filling and deformation prediction method for urban differential settlement monitoring, which solves the problem of incomplete Beidou monitoring data.

Description

Beidou data filling and deformation prediction method for urban differential settlement monitoring

Technical Field

The invention relates to the technical field of Beidou monitoring data processing, in particular to a Beidou data filling and deformation prediction method for urban differential settlement monitoring.

Background

In urban ground subsidence monitoring applications, the monitoring is constantly affected by the external environment. For example: the construction site is close to, traffic and transportation loads, newly built structures penetrate existing structures, sundries are stacked nearby monitoring stations along residents, and the like, and the stress state and bearing capacity of ground foundations can be changed under the action of the factors, so that additional settlement of the monitoring stations can be inevitably caused, and deformation monitoring accuracy is affected. In addition, because urban ground subsidence areas easily span different geological and geomorphic units and different climate zones, the urban ground subsidence areas inevitably pass through poor geological areas such as soft soil areas, underground goafs and the like, and even though foundation strengthening treatment in the forms of pile foundations, pile plate structures and the like is carried out in urban construction, subsidence deformation still can be generated in later long-term monitoring. In recent years, due to the transitional exploitation of groundwater in the areas such as the long triangle area, the North China plain and the Fenwei basin of China, the groundwater level is continuously reduced, large-area surface subsidence is caused, and structures in the areas also follow the large-scale subsidence deformation. The foundation settlement inevitably generates deformation even with the structures, changes the initial geometric shape and position of the buildings and the structures in the city, forms structural deformation, and reduces the service life of the structure.

The Beidou satellite positioning technology is a global positioning system with completely independent intellectual property rights in China, and has higher stability and safety. Particularly, the settlement monitoring of the heavy engineering oriented to the high-speed railway can keep reliable long-term performance, but due to the fact that the Beidou monitoring data inevitably generate data loss phenomenon under the disturbance of various factors such as signal interference, communication loss, data quality, external load and construction, the data is incomplete. Meanwhile, data such as synchronous vibration, track irregularity and the like need to be monitored as further data support, so that the necessity of data filling is caused. How to scientifically fill data and predict deformation has realistic necessity for the safe operation of urban infrastructure.

In the prior art, as disclosed in 2019-07-23, an infrastructure settlement monitoring method and system based on SAR data and GNSS data, with publication number of CN110044327A, the InSAR technology and the Beidou GNSS technology are combined, so that settlement displacement data corresponding to an imaging period in the whole coverage area can be obtained, the monitoring precision can reach millimeter level, the precision requirement on monitoring peristaltic disaster bodies can be met, and the data cannot be filled.

Disclosure of Invention

The invention provides a Beidou data filling and deformation prediction method for urban differential settlement monitoring, which aims to overcome the technical defect of incomplete Beidou monitoring data.

In order to solve the technical problems, the technical scheme of the invention is as follows:

a Beidou data filling and deformation prediction method for urban differential settlement monitoring comprises the following steps:

s1: carrying out high-precision calculation on the settlement monitoring data based on the Beidou to obtain a Beidou monitoring calculation result;

s2: carrying out optimal data prediction on the missing data in the Beidou monitoring and resolving result to obtain a prediction result;

s3: filling missing data according to the prediction result to obtain a continuous Beidou monitoring and resolving result;

s4: segmenting continuous Beidou monitoring and resolving results, fitting each segment of data to obtain a settlement monitoring curve, and completing filling and deformation prediction of Beidou data.

Preferably, the Beidou monitoring and resolving result comprises millimeter-level relative displacement in the north-south direction of the short base line, millimeter-level relative displacement in the east-west direction of the short base line and millimeter-level relative displacement in the vertical direction of the short base line.

Preferably, in step S2, the prediction result is obtained by:

s2.1: carrying out optimal data prediction on the missing data by adopting data positioned before the missing data to obtain first optimal predicted data;

s2.2: arranging the data after the missing data in reverse order;

s2.3: performing optimal data prediction on the missing data by using the data arranged in the reverse order to obtain second optimal predicted data;

s2.4: arranging the second optimal predicted data in reverse order;

s2.5: and calculating according to the first optimal prediction data and the second optimal prediction data which are arranged in the reverse order to obtain a prediction result.

Preferably, the LSTM model is used for carrying out optimal data prediction on the missing data.

Preferably, in step S2.5, the prediction result is calculated by the following formula:

wherein S is _i A is the prediction result corresponding to the ith missing data _i B for the first optimal predicted data corresponding to the ith missing data _i L is the number of missing data, which is the second most optimal predicted data corresponding to the ith missing data.

Preferably, before fitting each piece of data, the method further comprises a step of removing common mode errors, specifically:

dividing the Beidou monitoring and resolving results of M adjacent monitoring stations into N sections, fitting the data of each section into straight lines respectively,

if the slope values of the fit straight lines of all monitoring stations in a certain segment are similar, the data of the segment is considered to have common mode errors, otherwise, the data of the segment do not have common mode errors,

the first order term of the fitted line is subtracted from the data with common mode error to remove the common mode error.

Preferably, whether the slopes of the fitting straight lines of the monitoring stations are similar is judged by the following steps:

selecting one monitoring station from M adjacent monitoring stations, taking absolute values of slopes of fitting straight lines of N segments of data of the monitoring stations, and then calculating a first slope mean value r _ave1 ，

The slope of the fitting straight line of each monitoring station in each section is calculated respectively,

if it is

Judging that the data of the j-th section has a common mode error, otherwise, judging that the data of the j-th section does not have the common mode error;

wherein,,

the slope of the fitted straight line of each monitoring station in the j-th section is respectively shown.

Preferably, the specific steps of fitting the data of each segment are as follows:

s4.1: selecting a monitoring station to be fitted after common mode errors are removed, fitting each piece of data into a straight line by using a least square method, and calculating the slope of the fitted straight line of each piece of data;

s4.2: combining two adjacent straight lines with similar slopes;

s4.3: performing secondary curve fitting on each section after merging to obtain a plurality of sections of curves;

s4.4: and fitting the right part of each section of curve to the middle part of the right end point of the section where the right part of each section of curve is positioned and the leftmost point of the next section of curve, and obtaining the sedimentation monitoring curve.

Preferably, in step S4.2, it is determined whether the slopes of two adjacent straight lines are similar by:

taking absolute value of slope of each fitting straight line of the monitoring station selected in the step S4.1 and then calculating a second slope average value r _ave2 ，

If the slopes of two adjacent straight lines a and b satisfy

|r _a -r _b |<α*r _ave2

Judging that the slopes of the straight line a and the straight line b are similar, otherwise, judging that the slopes of the straight line a and the straight line b are not similar;

wherein r is _a ,r _b The slopes of the straight line a and the straight line b are respectively, alpha is the judging coefficient with similar slope,

1.2≤α≤1.8。

preferably, α=1.5.

Compared with the prior art, the technical scheme of the invention has the beneficial effects that:

the invention provides a Beidou data filling and deformation prediction method for urban differential settlement monitoring, which is characterized in that missing data in Beidou monitoring and resolving results are optimally predicted to obtain prediction results, the missing data are filled according to the prediction results, then a complete and continuous settlement monitoring curve is obtained through data fitting, and reliable data support is provided for building deformation.

Drawings

FIG. 1 is a flow chart of an embodiment of the present invention;

FIG. 2 is a schematic diagram of the basic unit of the LSTM model of the present invention;

FIG. 3 is a schematic diagram of a network structure of NAR model according to the present invention;

FIG. 4 is a schematic diagram showing the effect of LSTM model prediction in the present invention;

FIG. 5 is a schematic diagram showing the predictive effect of NAR model in the present invention;

FIG. 6 is a schematic diagram showing the predictive effect of the propset model according to the present invention;

FIG. 7 is a schematic representation of a near fit in the present invention;

fig. 8 is a schematic representation of the effect of the final fit in the present invention.

Detailed Description

The drawings are for illustrative purposes only and are not to be construed as limiting the present patent;

for the purpose of better illustrating the embodiments, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the actual product dimensions;

it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.

The technical scheme of the invention is further described below with reference to the accompanying drawings and examples.

Example 1

As shown in fig. 1, the Beidou data filling and deformation prediction method for urban differential settlement monitoring comprises the following steps:

s1: carrying out high-precision calculation on the settlement monitoring data based on the Beidou to obtain a Beidou monitoring calculation result;

in actual implementation, the Beidou monitoring and resolving result relies on double-difference resolving of a short baseline carrier phase, networking monitoring is generally carried out in one hour, the time sequence can be considered after data are accumulated for at least six months, and long-term monitoring is defined as at least monitoring networking for more than 2 years;

s2: carrying out optimal data prediction on the missing data in the Beidou monitoring and resolving result to obtain a prediction result;

in actual implementation, performing optimal data prediction based on a self-defined machine learning model;

s3: filling missing data according to the prediction result to obtain a continuous Beidou monitoring and resolving result;

s4: segmenting continuous Beidou monitoring and resolving results, fitting each segment of data to obtain a settlement monitoring curve, and completing filling and deformation prediction of Beidou data.

Example 2

More specifically, the Beidou monitoring and resolving result comprises millimeter-level relative displacement in the north-south direction of the short base line, millimeter-level relative displacement in the east-west direction of the short base line and millimeter-level relative displacement in the vertical direction of the short base line.

In the specific implementation process, the Beidou monitoring and resolving result comprises three variables, namely horizontal (north and south, east and west) and vertical, and the deformation directions of the three variables are different.

More specifically, in step S2, the prediction result is obtained by:

s2.1: carrying out optimal data prediction on the missing data by adopting data positioned before the missing data to obtain first optimal predicted data;

s2.2: arranging the data after the missing data in reverse order;

s2.3: performing optimal data prediction on the missing data by using the data arranged in the reverse order to obtain second optimal predicted data;

s2.4: arranging the second optimal predicted data in reverse order;

s2.5: and calculating according to the first optimal prediction data and the second optimal prediction data which are arranged in the reverse order to obtain a prediction result.

More specifically, the LSTM model is adopted to conduct optimal data prediction on missing data.

In the specific implementation process, the prediction effects of an LSTM model, an NAR model, an ARIMA model and a Prophet model are compared:

long and short termMemory (LSTM) is a special Recurrent Neural Network (RNN) mainly to solve the problems of gradient extinction and gradient explosion during Long-sequence training. In short, LSTM is able to perform better in longer sequences than normal RNNs. The basic unit of the LSTM hidden layer is called a Memory Block (Memory Block), and the structure of the Memory Block is shown in fig. 2. The memory block comprises 3 gates (input gate (i), output gate (o) and forget gate (f)) and a memory unit (c), the sign # -represents the addition of two vectors, the sign

Representing the point multiplication of two vectors, σ represents the sigmoid activation function, and tanh is the hyperbolic tangent activation function. The input gate determines how the input layer information is transferred to the memory unit; the forget gate decides how to keep the history information; the output gate determines how the information of the memory module is transferred to the memory block at the next moment. Let the time series of coordinate values be x= { x _t T=1, 2,3, … n, where x _t Is the coordinate value at time t. Given the sequence length l of the neural network, the parameter represents the coordinate value x of the use time length l _t ，x _t+1 ,…,x _t+l-1 The sequence predicts the coordinate value of the next moment. In this embodiment, for the selection of LSTM neural network parameters, the data experiment of the peripheral monitoring station is combined, and the training parameters are iterated: the total number of memory blocks is CELL_SIZE=100, the sequence length is TIME_STEP=72 (the sequence length is preferably selected to be a multiple of 24 because the monitoring data is one hour, so that the neural network can learn the rule of daily periodicity), the loss function is set as the mean square error, the training is carried out by using the Adam optimization algorithm, the learning rate lr=0.001, the iteration threshold is 0.0003, the maximum iteration number T=500, and the initial memory c ₀ =0, initial output s ₀ =0. In actual implementation, the number of storage blocks, the sequence length, the iteration threshold and the maximum iteration number can be properly adjusted according to the size of the processed data and the owned computing resources, and the example parameters only prove the reliability of the method model only for the experiment.

A nonlinear autoregressive neural network model (Nonlinear Autoregressive, NAR) is typically used as a tool for the prediction of time series. A NAR network is typically made up of an input layer, an hidden layer, and an output layer, and a delay function, the basic structure of which is shown in fig. 3. Where y (t) is the output of the neural network, i.e. the coordinate value at time t, the NAR network can be expressed as:

y(t)＝f(y(t-1),y(t-2),…,y(t-d))

wherein: t represents the moment; d represents the delay order, i.e. the past d values y (t-1), y (t-2), …, y (t-d) are used to predict the next value y (t), the delay order d determining the number of inputs to the neural network. In prediction, the delay function returns the output value to the input layer.

ARIMA (Auto-Regressive Integrated Moving Average) model, a differential integrated moving average autoregressive model, also known as an integrated moving average autoregressive model (movement may also be referred to as sliding), is one of the methods of time series predictive analysis. The ARIMA (p, d, q) model contains p autoregressive terms and q moving average terms, and can be expressed as:

wherein the symbols are: p and q are the autoregressive and moving average orders of the model; phi and theta are undetermined coefficients that are not zero; epsilon _t Is an independent error term; x is X _t The time sequence is stable, normal and zero-mean after d-order difference. The determination of the parameters d, p, q may use bayesian information criteria (Bayesian information criterion), which are defined as:

BIC＝-2ln(M)+ln(n)*k

where M is the maximum likelihood under the model, n is the number of data, and k is the number of variables of the model. For an ARIMA model for a given parameter, a smaller BIC value indicates that the model is closer to reality. However, when the ARIMA model is applied to the Beidou settlement monitoring data, the BIC value is found to be too small, and if the BIC value is a negative value, the situation that the model is over-fitted is indicated, and the model cannot be used for prediction. It can be determined that the result predicted by this model is indeed almost a straight line, so the ARIMA model is not applicable.

The propset model, facebook, opens the algorithm propset for a time series prediction in 2017. The propset algorithm is based on time series decomposition and machine-learned fitting. The time series decomposition divides the time series y (t) into several parts, namely a trend term g (t), a period term s (t), a holiday term (t) and an error term e _t . I.e.

y(t)＝g(t)+s(t)+h(t)+∈ _t

The propset algorithm obtains a predicted value of the time sequence by fitting the terms and then finally accumulating the terms.

A 1200 pieces of data without deletion were selected, the first 800 pieces were used as training sets, the last 400 were used as test sets, and the different models in the test sets were represented as shown in fig. 4, 5, and 6.

In this embodiment, the root mean square error RMSE (Root MeanSquare Error) is used to evaluate the model effect, and the calculation formula is as follows:

wherein y is _t And

the real value and the predicted value of the coordinate data at the time T are respectively, and T is the number of the data in the test data set.

And (3) taking a plurality of groups of data experiments, calculating whether the average value of the RMSE value of LSTM is 0.00108, the average value of the RMSE value of NAR is 0.00128, and the average value of the RMSE value of Prophet is 0.00114 or LSMT is good, so that the LSTM model is adopted to optimally predict the missing data.

More specifically, in step S2.5, the predicted result is calculated by the following formula:

wherein S is _i A is the prediction result corresponding to the ith missing data _i B for the first optimal predicted data corresponding to the ith missing data _i L is the number of missing data, which is the second most optimal predicted data corresponding to the ith missing data.

More specifically, before fitting each piece of data, the method further comprises a step of removing common mode errors, specifically:

dividing the Beidou monitoring and resolving results of M adjacent monitoring stations into N sections, fitting the data of each section into straight lines respectively,

in the implementation process, similar deformation or interference exists in monitoring stations with similar layout, and the common deformation or interference is a solution in one hour, so that 24 or 48 data in each section are selected in a segmented mode, and the common mode error of daily periodicity can be eliminated;

if the slope values of the fit straight lines of all monitoring stations in a certain segment are similar, the data of the segment is considered to have common mode errors, otherwise, the data of the segment do not have common mode errors,

the first order term of the fitted line is subtracted from the data with common mode error to remove the common mode error.

More specifically, whether the slopes of the fitting straight lines of all the monitoring stations are similar is judged by the following steps:

selecting one monitoring station from M adjacent monitoring stations, taking absolute values of slopes of fitting straight lines of N segments of data of the monitoring stations, and then calculating a first slope mean value r _ave1 ，

The slope of the fitting straight line of each monitoring station in each section is calculated respectively,

if it is

Judging that the data of the j-th section has a common mode error, otherwise, judging that the data of the j-th section does not have the common mode error;

wherein,,

each of the monitoring stationsThe slope of the fitted line at section j.

More specifically, the specific steps of fitting each piece of data are as follows:

s4.1: selecting a monitoring station to be fitted after common mode errors are removed, fitting each piece of data into a straight line by using a least square method, and calculating the slope of the fitted straight line of each piece of data;

s4.2: combining two adjacent straight lines with similar slopes;

in actual implementation, two adjacent straight lines with similar slopes are respectively set as a straight line a and a straight line b, and if the former straight line a is combined with other straight lines c, the results of combining the straight line b with the straight line a and the straight line c are combined;

s4.3: performing secondary curve fitting on each section after merging to obtain a plurality of sections of curves;

s4.4: and fitting the right part of each section of curve to the middle part of the right end point of the section where the right part of each section of curve is positioned and the leftmost point of the next section of curve, and obtaining the sedimentation monitoring curve.

In a specific implementation process, step S4.4 specifically includes: setting the ith section of sequence

Length n, i+1th section->

Length m, assuming->

I.e. the case shown in fig. 7, i.e. we need to fit the i-th piece of data to the i+1-th piece of data. Is provided with->

For the i-th sequence we only process the right half of the sequence, since the left half of the sequence is processed identically to the right half of the i-1 sequence. Set its length

(if n is odd then +.>

Rounding, here all ∈ ->

Representation) of the sequence +.>

The sequence after treatment is->

Then:

i.e. the rightmost data is closer to dl, the smaller the fitting value is to the left, and the decreasing is at a quadratic speed, so as to ensure that the processed sequence is still a smooth curve.

The left half part of the (i+1) th segment sequence is similarly processed to ensure that the data at the junction of the two segments are all

The discontinuity is eliminated and the curve is smooth after treatment. By doing so between every two sequences, the fitting is finally completed, as shown in fig. 8.

More specifically, in step S4.2, it is determined whether the slopes of two adjacent straight lines are similar or not by:

taking absolute value of slope of each fitting straight line of the monitoring station selected in the step S4.1 and then calculating a second slope average value r _ave2 ，

If the slopes of two adjacent straight lines a and b satisfy

|r _a -r _b |<α*r _ave2

Judging that the slopes of the straight line a and the straight line b are similar, otherwise, judging that the slopes of the straight line a and the straight line b are not similar;

wherein r is _a ,r _b The slopes of the straight line a and the straight line b are respectively, alpha is the judging coefficient with similar slope,

1.2≤α≤1.8。

in the implementation process, when alpha is smaller than 1.2, some segments with similar change trend are not combined; when α is greater than 1.8, some segments that appear to vary in trend are combined, so that α is defined to be between 1.2 and 1.8.

More specifically, α=1.5.

In the implementation process, the best combination effect can be obtained when alpha=1.5.

It is to be understood that the above examples of the present invention are provided by way of illustration only and not by way of limitation of the embodiments of the present invention. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are desired to be protected by the following claims.

Claims (6)

1. The Beidou data filling and deformation prediction method for urban differential settlement monitoring is characterized by comprising the following steps of:

s1: carrying out high-precision calculation on the settlement monitoring data based on the Beidou to obtain a Beidou monitoring calculation result;

s2: carrying out optimal data prediction on the missing data in the Beidou monitoring and resolving result to obtain a prediction result;

s3: filling missing data according to the prediction result to obtain a continuous Beidou monitoring and resolving result;

s4: segmenting continuous Beidou monitoring and resolving results, fitting each segment of data to obtain a settlement monitoring curve, and completing filling and deformation prediction of Beidou data;

in step S2, the prediction result is obtained by:

s2.1: carrying out optimal data prediction on the missing data by adopting data positioned before the missing data to obtain first optimal predicted data;

s2.2: arranging the data after the missing data in reverse order;

s2.3: performing optimal data prediction on the missing data by using the data arranged in the reverse order to obtain second optimal predicted data;

s2.4: arranging the second optimal predicted data in reverse order;

s2.5: calculating according to the first optimal prediction data and the second optimal prediction data arranged in the reverse order to obtain a prediction result;

and adopting an LSTM model to perform optimal data prediction on the missing data.

2. The urban differential settlement monitoring-oriented Beidou data filling and deformation prediction method according to claim 1, wherein Beidou monitoring and resolving results comprise short-baseline north-south millimeter-level relative displacement, short-baseline east-west millimeter-level relative displacement and short-baseline vertical millimeter-level relative displacement.

3. The method for filling and predicting deformation of Beidou data for urban differential settlement monitoring according to claim 1, wherein in step S2.5, a prediction result is obtained by calculating by the following formula:

wherein S is _i A is the prediction result corresponding to the ith missing data _i B for the first optimal predicted data corresponding to the ith missing data _i L is the number of missing data, which is the second most optimal predicted data corresponding to the ith missing data.

4. The method for filling and predicting deformation of Beidou data for urban differential settlement monitoring according to claim 1, wherein the method further comprises the step of removing common mode errors before fitting each piece of data, and is specifically as follows:

dividing the Beidou monitoring and resolving results of M adjacent monitoring stations into N sections, fitting the data of each section into straight lines respectively,

if the slope values of the fit straight lines of all monitoring stations in a certain segment are similar, the data of the segment is considered to have common mode errors, otherwise, the data of the segment do not have common mode errors,

subtracting a first order term of the fitting straight line from the data with the common mode error to remove the common mode error;

the specific steps of fitting the data of each section are as follows:

s4.1: selecting a monitoring station to be fitted after common mode errors are removed, fitting each piece of data into a straight line by using a least square method, and calculating the slope of the fitted straight line of each piece of data;

s4.2: combining two adjacent straight lines with similar slopes;

s4.3: performing secondary curve fitting on each section after merging to obtain a plurality of sections of curves;

s4.4: fitting the right part of each section of curve to the right end point of the section where the right part of each section of curve is positioned and the middle part of the leftmost point of the next section of curve to obtain a sedimentation monitoring curve;

in step S4.2, it is determined whether the slopes of two adjacent straight lines are similar by:

taking absolute value of slope of each fitting straight line of the monitoring station selected in the step S4.1 and then calculating a second slope average value r _ave2 ，

If the slopes of two adjacent straight lines a and b satisfy

|r _a -r _b |<α*r _ave2

Judging that the slopes of the straight line a and the straight line b are similar, otherwise, judging that the slopes of the straight line a and the straight line b are not similar; wherein r is _a ，r _b The slopes of the straight line a and the straight line b are respectively, alpha is the judging coefficient with similar slope,

1.2≤α≤1.8。

5. the urban differential settlement monitoring-oriented Beidou data filling and deformation prediction method according to claim 4 is characterized by judging whether the slopes of fitting straight lines of all monitoring stations are similar or not through the following steps:

selecting one monitoring station from M adjacent monitoring stations, taking absolute values of slopes of fitting straight lines of N segments of data of the monitoring stations, and then calculating a first slope mean value r _ave1 ，

The slope of the fitting straight line of each monitoring station in each section is calculated respectively,

judging that the data of the j-th section has a common mode error, otherwise, judging that the data of the j-th section does not have the common mode error; wherein,,

the slope of the fitted straight line of each monitoring station in the j-th section is respectively shown.

6. The urban differential settlement monitoring-oriented Beidou data filling and deformation prediction method of claim 4, wherein alpha=1.5.

Images