CN109886460A - The prediction technique of tunnel subsidence time series based on adaboost - Google Patents
The prediction technique of tunnel subsidence time series based on adaboost Download PDFInfo
- Publication number
- CN109886460A CN109886460A CN201910039479.2A CN201910039479A CN109886460A CN 109886460 A CN109886460 A CN 109886460A CN 201910039479 A CN201910039479 A CN 201910039479A CN 109886460 A CN109886460 A CN 109886460A
- Authority
- CN
- China
- Prior art keywords
- time series
- adaboost
- prediction
- sample
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The prediction technique for the tunnel subsidence time series based on adaboost that the invention discloses a kind of, the following steps are included: 1) acquire the ground settlement data of tunnel upper, each collection point successively records settlement values in chronological order, obtains the time series data of one-dimensional；2) time series data of the one-dimensional is expanded into multidimensional data using the equal reforms breath method in gray theory；3) training set is inputted into the weak learner of more bases, initialization error rate, the number of iterations and sample weights sample original training set according to sample weights distribution, and use the more weak learners of base of sample data training；4) weight undated parameter is arranged according to sample error rate, the sample weights is updated according to the weight undated parameter；5) successive ignition is carried out according to the number of iterations, exports final strong learner:Wherein, t is current iteration, and T is the number of iterations, and β is weight undated parameter, and x is the time, and prediction result f (x) is Predicted settlement value.
Description
Technical field
The present invention relates to technical field of medical equipment, more specifically, being related to a kind of tunnel subsidence based on adaboost
The prediction technique of time series.
Background technique
Tunnel subsidence problem not only affects the development of urban track traffic, while to the security of the lives and property of city dweller
There is great threat.Thus accurately forecasting research is carried out to the sedimentation in tunnel to have a very important significance.It learns both at home and abroad
Person has carried out a large amount of research to the settlement prediction in tunnel.Research method can be roughly divided into two types: theoretical calculation empirical method and
Measured data analytic approach.Theoretical calculation empirical method is using Peck empirical formula method as representative, including numerical analysis method, numerical simulation
Method, half economics analysis method and theory of stochastic model etc..
Measured data analytic approach is divided into based on statistical method and machine learning method.The especially research of machine learning
Using, for solve tunnel subsidence forecasting problem provide many new resolving ideas.It has been proposed that using FInite Element to tunnel
The building on periphery carries out settlement research, is then studied using neural network various sedimentation situations, is finally obtained to tunnel
The prediction conclusion of road sedimentation.
Time series refers to that the observation of a certain variable in system is ranked into a sequence in chronological order, it is by system
In the overall result that influences of various other factors by the research to this sequence, the variation characteristic and development that can analyze things become
Gesture, rule.Time series has been widely used in many aspects, the financial shares such as tunnel subsidence in engineering, in economic field
Ticket etc..
The validity of theoretical calculation empirical method depends on the reasonability of soil model and the acquisition methods of Soil Parameters, thus
The generalization of theoretical calculation empirical method is poor；And measured data analytic approach validity depends on a large amount of sample data, thus it is general
The property changed is preferable, but not high to complicated nonlinear regression problem precision of prediction.
Summary of the invention
In view of this, the present invention propose it is a kind of the stronger time series single-dimensional data of randomness and complexity it is non-linear
The prediction technique of the tunnel subsidence time series based on adaboost in the case of regression problem exists for solving the prior art
The not high-tech problem of precision of prediction in these cases.
The prediction technique for the tunnel subsidence time series based on adaboost that the present invention provides a kind of, including following step
It is rapid:
1) the ground settlement data of tunnel upper are acquired, each collection point successively records settlement values in chronological order, obtains
To the time series data of one-dimensional；
2) time series data of the one-dimensional is expanded into multidimensional data using the equal reforms breath method in gray theory；
3) training set is inputted into the weak learner of more bases, initialization error rate, the number of iterations and sample weights are weighed according to sample
Redistribution samples original training set, and uses the more weak learners of base of sample data training；
4) relative error is calculated, and it is compared with the threshold value for measuring precision of prediction, to obtain sample error rate, root
Weight undated parameter is set according to the sample error rate, the sample weights are updated according to the weight undated parameter；
5) successive ignition is carried out according to the number of iterations, exports final strong learner:
Wherein, t is current iteration, and T is the number of iterations, and β is weight undated parameter, and x is
Time, prediction result f (x) are Predicted settlement value.
Optionally, the weak learner of more bases is same type of multiple weak learners or different types of multiple weak
Practise device.
Optionally, different types of multiple weak learners include the weak learner of SVR, the weak learner of BP and ELM weak
Learner.
Optionally, the time series data of the one-dimensional is expanded into multidimensional data, it is specific as follows: every time using first three
The 4th point of point prediction, one-dimensional sample data, which is expanded, becomes three-dimensional samples data.
Optionally, using root-mean-square error, mean absolute error or/and mean absolute percentage error to prediction result into
Row evaluation.
Optionally, the update sample weights are with the following method:
Wherein Zt is normalization factor, guarantees updated Dt+1 and for 1；
Right value update parameter betat=εt h, h desirable 1,2,3；Dt is sample weights.
Optionally, pass through repeatedly training and adjust the threshold value for measuring precision of prediction, with adjustment and Optimization Prediction result.
Optionally, it is evaluated according to prediction result, adjusts the type of more weak learners of base.
Using the present invention, compared with prior art, it is single to have the advantage that the present invention is predicted for current tunnel subsidence
The problems such as model or statistical method precision of prediction are lower, generalization is poor proposes to promote base using the training of Adaboost boosting algorithm
Learning model, and weight obtain a strong learning model, in the forecasting research to tunnel subsidence whether precision of prediction or
It will be better than single study prediction model in generalization, can be effectively applied in Practical Project to the pre- of tunnel subsidence
It surveys.
Detailed description of the invention
Fig. 1 is the first emulation schematic diagram of the prediction technique of the tunnel subsidence time series based on adaboost；
Fig. 2 is the second emulation schematic diagram of the prediction technique of the tunnel subsidence time series based on adaboost.
Specific embodiment
The preferred embodiment of the present invention is described in detail below in conjunction with attached drawing, but the present invention is not restricted to these
Embodiment.The present invention covers any substitution made in the spirit and scope of the present invention, modification, equivalent method and scheme.
In order to make the public have thorough understanding to the present invention, it is described in detail in the following preferred embodiment of the present invention specific
Details, and the present invention can also be understood completely in description without these details for a person skilled in the art.
The present invention is more specifically described by way of example referring to attached drawing in the following passage.It should be noted that attached drawing is adopted
With more simplified form and using non-accurate ratio, only to facilitate, lucidly aid in illustrating the embodiment of the present invention
Purpose.
AdaBoost (adaptive boosting) is a kind of iteration boosting algorithm, and core concept is according to iteration
The learning sample weight that number is constantly updated determines the initial all trained samples of strong learner being made of different weak learners
This weight is endowed equal value, and as training iteration sample weights constantly update, the big sample weights of prediction error will increase
Add, the small sample weights of prediction error will reduce, and just increase weak learner in this way to it is difficult to predict the concerns of sample to realize
The weighting integration of weak learner is finally obtained unique strong study according to sample weights, prediction error etc. by the prediction of higher precision
Device model.AdaBoost iteration boosting algorithm solves complicated nonlinear regression problem by way of two classification.
The present invention provides a kind of prediction technique of tunnel subsidence time series based on adaboost, comprising the following steps:
1) the ground settlement data of tunnel upper are acquired, each collection point successively records settlement values in chronological order, obtains
To the time series data of one-dimensional；
2) time series data of the one-dimensional is expanded into multidimensional data using the equal reforms breath method in gray theory；
3) training set is inputted into the weak learner of more bases, initialization error rate, the number of iterations and sample weights are weighed according to sample
Redistribution samples original training set, and uses the more weak learners of base of sample data training；
4) relative error is calculated, and it is compared with the threshold value for measuring precision of prediction, to obtain sample error rate, root
Weight undated parameter is set according to the sample error rate, the sample weights are updated according to the weight undated parameter；
5) successive ignition is carried out according to the number of iterations, exports final strong learner:
Wherein, t is current iteration, and T is the number of iterations, and β is weight undated parameter, and x is
Time, prediction result f (x) are Predicted settlement value.
The weak learner of more bases is same type of multiple weak learners or different types of multiple weak learners.
Different types of multiple weak learners include the weak learner of SVR, the weak learner of BP and the weak learner of ELM.
SVR is developed from SVM support vector machines, is had very strong generalization ability, can preferably be solved the small of many learning methods
The problems such as sample, overfitting, high-dimensional and Local Minimum；BP neural network have stronger adaptivity, self-organization, from
It overcomes the deficiency of traditional feedback method by study property and non-linear mapping capability, has obtained more and more extensive answer in recent years
With.The more traditional gradient descent algorithm of ELM extreme learning machine has faster convergence rate, faster precision of prediction, and is easy to
Obtain global optimum.
Pseudo-code of the algorithm is as follows:
Input: training set X=(x1, x2, x3, y1), (x2, x3, x4, y2) ..., (xm, xm+1, xm+2, ym) }；
Three kinds of base learning models L1, L2, L3；
Exercise wheel number T=3.
Measure the threshold value φ, sample weights Dt of precision of prediction
Process:
1:
2:for t=1,2, T do
3: dividing training set, training base learner Lt:f according to sample weights Dtt(x)→y,
And calculate relative error
4: calculate error rate:When
5: βt=εt h, (h=1,2,3)；
6:
7:end
Output:
The time series data of the one-dimensional is expanded into multidimensional data, it is specific as follows: to use first three point prediction every time
4th point, one-dimensional sample data, which is expanded, becomes three-dimensional samples data.
Prediction result is evaluated using root-mean-square error, mean absolute error or/and mean absolute percentage error.
It chooses 3 general evaluation criterions of regression model to evaluate the performance of each method, RMSE (Root Mean
Squared Error) root-mean-square error, MAE (Mean Absolute Error) mean absolute error, MAPE (mean
Absolute percent error) mean absolute percentage error, calculation formula is as follows:
The update sample weights are with the following method:
Wherein Zt is normalization factor, guarantees updated Dt+1 and for 1；
Right value update parameter betat=εt h, h desirable 1,2,3；Dt is sample weights.
Pass through repeatedly training and adjust the threshold value for measuring precision of prediction, with adjustment and Optimization Prediction result.
It is evaluated according to prediction result, adjusts the type of more weak learners of base.As shown in table 1, it illustrates
Adaboost integrates Multiple Models Algorithm and SVR, BPnetwork and ELM performance compare, it can be seen that Adaboost integrates multimode
The prediction accuracy of type algorithm is higher than single model.Wherein, table 2 illustrates Adaboost and integrates multi-model process and Adaboost
Integrated SVR, Adaboost integrate the integrated ELM method performance of BP network, Adaboost and compare, it can be seen that Adaboost collection
It is higher than at the prediction accuracy of Multiple Models Algorithm and integrates single model.Meanwhile suitable model can be selected by this method.Table
1 and 2 have carried out the selection of collection point simply by the mode of experiment, in order to guarantee the objectivity of result, therefrom intercept a number of segment
According to as signal, data are not screened.
1 Adaboost of table integrates Multiple Models Algorithm and SVR, BP network and ELM performance compares
Table 1 performance comparison of Adaboost integrated multi-model
algorithm with SVR,BP network and ELM
2 Adaboost of table integrate multi-model process and Adaboost integrate SVR, Adaboost integrate BP network,
Adaboost integrates ELM method performance and compares
Table 2 Performance comparison of Adaboost integrated multi-model
method with Adaboost integrated SVR,
Adaboost integrated BP network and Adaboost integrated ELM method
Fig. 1 and Fig. 2 illustrates the simulation result of the prediction technique of the tunnel subsidence time series based on adaboost, respectively
Using two points as signal.As shown in the figure, prediction result of the strong fallout predictor prediction output as the application, the desired output are
Ideal value or desired value, psosvr, psobp, elm respectively indicate the weak learner of SVR, the weak learner of BP and the weak learner of ELM.Figure
In as it can be seen that strong fallout predictor prediction output closest to desired output, learner weaker than single SVR, the weak learner of BP or weak of ELM
It is more acurrate to practise device, is consistent with the result in table 1 and 2.
Although embodiment is separately illustrated and is illustrated above, it is related to the common technology in part, in ordinary skill
Personnel apparently, can be replaced and integrate between the embodiments, be related to one of embodiment and the content recorded is not known, then
It can refer to another embodiment on the books.
Embodiments described above does not constitute the restriction to the technical solution protection scope.It is any in above-mentioned implementation
Made modifications, equivalent substitutions and improvements etc., should be included in the protection model of the technical solution within the spirit and principle of mode
Within enclosing.
Claims (8)
1. a kind of prediction technique of the tunnel subsidence time series based on adaboost, it is characterised in that: the following steps are included:
1) the ground settlement data of tunnel upper are acquired, each collection point successively records settlement values in chronological order, obtains list
The time series data of dimension；
2) time series data of the one-dimensional is expanded into multidimensional data using the equal reforms breath method in gray theory；
3) training set is inputted into the weak learner of more bases, initialization error rate, the number of iterations and sample weights, according to sample weights point
Cloth samples original training set, and uses the more weak learners of base of sample data training；
4) relative error is calculated, and it is compared with the threshold value for measuring precision of prediction, to obtain sample error rate, according to institute
Sample error rate setting weight undated parameter is stated, the sample weights are updated according to the weight undated parameter；
5) successive ignition is carried out according to the number of iterations, exports final strong learner:
Wherein, t is current iteration, and T is the number of iterations, and β is weight undated parameter, and x is the time,
Prediction result f (x) is Predicted settlement value.
2. the prediction technique of the tunnel subsidence time series according to claim 1 based on adaboost, it is characterised in that:
The weak learner of more bases is same type of multiple weak learners or different types of multiple weak learners.
3. the prediction technique of the tunnel subsidence time series according to claim 2 based on adaboost, it is characterised in that:
Different types of multiple weak learners include the weak learner of SVR, the weak learner of BP and the weak learner of ELM.
4. the prediction technique of the tunnel subsidence time series according to claim 1 based on adaboost, it is characterised in that:
The time series data of the one-dimensional is expanded into multidimensional data, specific as follows: the 4th point of first three point prediction is used every time,
One-dimensional sample data, which is expanded, becomes three-dimensional samples data.
5. the prediction technique of -4 tunnel subsidence time series based on adaboost described in any one according to claim 1,
It is characterized by: being commented using root-mean-square error, mean absolute error or/and mean absolute percentage error prediction result
Valence.
6. the prediction technique of the tunnel subsidence time series according to claim 5 based on adaboost, it is characterised in that:
The update sample weights are with the following method:
Wherein Zt is normalization factor, guarantees updated Dt+1 and for 1；Weight
Undated parameter βt=εt h, h desirable 1,2,3；Dt is sample weights.
7. the prediction technique of the tunnel subsidence time series according to claim 6 based on adaboost, it is characterised in that:
Pass through repeatedly training and adjust the threshold value for measuring precision of prediction, with adjustment and Optimization Prediction result.
8. the prediction technique of the tunnel subsidence time series according to claim 5 based on adaboost, it is characterised in that:
It is evaluated according to prediction result, adjusts the type of more weak learners of base.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910039479.2A CN109886460A (en) | 2019-01-16 | 2019-01-16 | The prediction technique of tunnel subsidence time series based on adaboost |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910039479.2A CN109886460A (en) | 2019-01-16 | 2019-01-16 | The prediction technique of tunnel subsidence time series based on adaboost |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109886460A true CN109886460A (en) | 2019-06-14 |
Family
ID=66926126
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910039479.2A Pending CN109886460A (en) | 2019-01-16 | 2019-01-16 | The prediction technique of tunnel subsidence time series based on adaboost |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886460A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113626918A (en) * | 2021-08-10 | 2021-11-09 | 哈尔滨工业大学 | Basic settlement prediction method based on time-weighted gray system theory |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160033536A1 (en) * | 2010-02-10 | 2016-02-04 | Rules-Based Medicine, Inc. | Novel groups of biomarkers for diagnosing alzheimer's disease |
| CN106548233A (en) * | 2016-10-26 | 2017-03-29 | 南京邮电大学 | A kind of flexible measurement method based on the 4 CBA contents for improving AdaBoost algorithms |
| CN108363844A (en) * | 2018-01-26 | 2018-08-03 | 大连理工大学 | A kind of aero-engine start-up course delivery temperature prediction technique |
-
2019
- 2019-01-16 CN CN201910039479.2A patent/CN109886460A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160033536A1 (en) * | 2010-02-10 | 2016-02-04 | Rules-Based Medicine, Inc. | Novel groups of biomarkers for diagnosing alzheimer's disease |
| CN106548233A (en) * | 2016-10-26 | 2017-03-29 | 南京邮电大学 | A kind of flexible measurement method based on the 4 CBA contents for improving AdaBoost algorithms |
| CN108363844A (en) * | 2018-01-26 | 2018-08-03 | 大连理工大学 | A kind of aero-engine start-up course delivery temperature prediction technique |
Non-Patent Citations (2)
| Title |
|---|
| D. P. SOLOMATINE: "《a boosting algorithm for regression problems》", 《IEEE》 * |
| 李伟等: "《等维新息SVＲ 模型对隧道沉降时间序列的预测研究》", 《中国计量大学学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113626918A (en) * | 2021-08-10 | 2021-11-09 | 哈尔滨工业大学 | Basic settlement prediction method based on time-weighted gray system theory |
| CN113626918B (en) * | 2021-08-10 | 2022-04-15 | 哈尔滨工业大学 | Basic settlement prediction method based on time-weighted gray system theory |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Davò et al. | Post-processing techniques and principal component analysis for regional wind power and solar irradiance forecasting | |
| CN101480143B (en) | Method for predicating single yield of crops in irrigated area | |
| CN109659933A (en) | A kind of prediction technique of power quality containing distributed power distribution network based on deep learning model | |
| CN107886161A (en) | A kind of global sensitivity analysis method for improving Complex Information System efficiency | |
| Deepa et al. | Multiclass model for agriculture development using multivariate statistical method | |
| CN107292098A (en) | Medium-and Long-Term Runoff Forecasting method based on early stage meteorological factor and data mining technology | |
| CN106886846A (en) | A kind of bank outlets' excess reserve Forecasting Methodology that Recognition with Recurrent Neural Network is remembered based on shot and long term | |
| CN102183802B (en) | Short-term climate forecast method based on Kalman filtering and evolution modeling | |
| CN110992113A (en) | Neural network intelligent algorithm-based project cost prediction method for capital construction transformer substation | |
| US11874429B2 (en) | High-temperature disaster forecast method based on directed graph neural network | |
| CN106022614A (en) | Data mining method of neural network based on nearest neighbor clustering | |
| CN103049651A (en) | Method and device used for power load aggregation | |
| Saravanan et al. | Forecasting India's electricity demand using artificial neural network | |
| CN108549907A (en) | A kind of data verification method based on multi-source transfer learning | |
| CN107516168A (en) | A kind of Synthetic Assessment of Eco-environment Quality method | |
| CN104318515A (en) | Hyper-spectral image wave band dimension descending method based on NNIA evolutionary algorithm | |
| Swain et al. | Impact of catchment classification on streamflow regionalization in ungauged catchments | |
| CN115689008A (en) | CNN-BilSTM short-term photovoltaic power prediction method and system based on ensemble empirical mode decomposition | |
| CN115392554A (en) | Track passenger flow prediction method based on depth map neural network and environment fusion | |
| Zhao et al. | Short-term microgrid load probability density forecasting method based on k-means-deep learning quantile regression | |
| CN113033081A (en) | Runoff simulation method and system based on SOM-BPNN model | |
| CN105279582B (en) | Super short-period wind power prediction technique based on dynamic correlation feature | |
| Khan et al. | A new hybrid approach of clustering based probabilistic decision tree to forecast wind power on large scales | |
| CN106327392A (en) | Examination admission intelligent prediction method based on big data | |
| CN109886460A (en) | The prediction technique of tunnel subsidence time series based on adaboost |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190614 |