Embodiment
Further is made to technical scheme below in conjunction with the accompanying drawings and by embodiment of railway traffic engineering constructionExplanation.
It is that track traffic material risk source monitors schematic diagram in real time as shown in Figure 1, is the great wind of track traffic as shown in Figure 2Dangerous source method for real-time monitoring flow chart, track traffic material risk source method for real-time monitoring specifically includes following steps：
(1) three dimensional environmental model and BIM construction management platforms are set up using 3DGIS+BIM technologies；
Three dimensional environmental model is comprising in terrain data, urban landscape data, geological information, underground pipeline information and projectPortion's information.GIS-Geographic Information System (GIS) technology, BIM (BIM) technology.
(2) material risk source is monitored, and upload the data to 3DGIS+BIM construction management platforms；
Material risk source data is gathered in real time using automatic monitoring means.
(3) risk source is managed；
Early warning value is set to BIM construction managements platform, when Monitoring Data exceedes early warning value, the automatic bright aobvious alarm of system.
(4) adaptive grey Fourier Index Prediction Model is set up；
Based on gray system theory, improved by background value, particular value conversion and residual GM Technology Integration etc. threeMethod, sets up prediction of the adaptive grey Fourier Index Prediction Model realization to material risk source time sequence.
Background value improves the application by integral term, eliminates the error that traditional background value formula is produced；Specific solution conversion is logicalCross to set up and GMm (1,1) models of concept are minimized and based on reinforcing up-to-date information based on prediction curve and real data deviationGMn (1,1) model of concept, is fitted prediction curve；Residual GM, which is integrated, passes through Fourier space and exponential smoothing amendmentThe periodicity residual error of sequence and randomness residual error, to promote the reliability of forecast model.
(5) engineering construction material risk source time sequence is predicted.
All monitoring results are analyzed and handled based on adaptive grey Fourier Index Prediction Model, to engineeringConstruction risk source carries out the real-time estimate of time series.
Step 1 specific implementation method is as follows：
(1.1) terrain data drawing three-dimensional relief model is obtained；
(1.2) urban landscape data drawing three-dimensional city model is passed through；
(1.3) 3DGIS+BIM construction management platforms are set up.
Step 2 specific implementation method is as follows：
(2.1) automatic monitoring data；
By automatic monitoring system, subway tunnel delta data under external force is gathered, data pass through optical fiber or netThe automatic real-time Transmission of network gives BIM construction management platforms.Required monitoring item, monitoring mode and measuring point are laid as shown in table 1.
Table 1
Automatic monitoring system based on platform has following advantage：
1. automate：, can after some Initialize installations and given monitoring plan is carried out (such as observation interval, issue)Full-automatic observation is strictly performed according to plan, and records initial data automatically.
2. it is intelligent：Realize unattended deformation monitoring, and the energy encountered problems with certain adaptation environment and processingPower.At a time when some monitoring point is kept off during such as the observation of certain phase, instrument is unable to reading, and software meeting automatic control survey robot is tastedExamination measurement several times, if still not all right, is voluntarily retried every a period of time again, until observation is normal.
3. ripe data processing method：Locate after being carried out to the initial data of collection according to the data processing model of settingReason, removes the various errors introduced in measurement process, and finally fitting reflects the data for best suiting practical distortion.
4. data storage and management：A large amount of original observed datas of collection are carried out using SQL Server database technologysStorage and management, and data are inquired about and analyzed.
5. various result output：Platform can carry out form to the data after initial data and analyzing and processing, draw deformationConditional curve, and the curvilinear figure of drafting can be printed out.
6. automatic real-time early warning, alarm：According to setting limit difference to more than allow limit difference deformation point carry out real-time early warning,Alarm, and show the value that deformation is transfinited.
Step 3 specific implementation method is as follows：
(3.1) risk source is managed；
The monitoring measurement Con trolling index proposed according to designing unit, by the alert status of monitoring point in work progress by seriousDegree is ascending to be divided into three-level：Yellow monitoring and warning, orange monitoring and warning and red monitoring and warning.
1. yellow monitoring and warning：" dual control " index (variable quantity, rate of change) exceedes the 70% of monitoring measurement controlling valueWhen, or one of dual control index is when exceeding the 85% of monitoring measurement controlling value；
2. orange monitoring and warning：One of when " dual control " index exceedes the 85% of monitoring measurement controlling value, or dual control indexDuring more than monitoring measurement controlling value；
3. red monitoring and warning：" dual control " index exceedes monitoring measurement controlling value, or actual measurement rate of change occurs drasticallyDuring growth.
The positional information of monitoring point is added in the 3DGIS+BIM models of structure, can be quick to monitoring point by modelPositioning, when being broken down so as to monitoring device or no longer meeting monitoring accuracy requirement, equipment replacement can be completed in time, it is ensured that monitoringThe authenticity of data.
(3.2) Risk-warning；
By setting early warning line on platform, once Monitoring Data is exceeded, platform can propose alarm automatically.
Warning information list display the monitoring point quantity of each project, red early warning number, orange warning number, yellow early warningKeep count of, the detailed report of each early warning can be checked.
The specific implementation of step 4 and step 5 is as follows：
By setting up adaptive grey Fourier Index Prediction Model, the reality of time series is carried out to engineering construction risk sourceWhen predict.Model is specifically described below to set up and analysis calculating process.
(1) GM (1,1) modeling and pre- flow gauge
Assuming that there is a time series to be X(0)=(x(0)(1), x(0)(2) ..., x(0)(n)), its be positive number, it is equidistant n whenBetween put upper data and constituted.The present invention illustrates GM (1,1) modeling and pre- flow gauge based on this time sequence.
Step 1：To X(0)Generated
Original series X(0)Kenel be probably in disorder, it is difficult to find out the data of its rule of development.In order to which its is implicitRule excavates out, can pass through many generation techniques and original in disorder sequence is carried out into one or many generations, to reduce dataRandomness, and it is regular to lift its.Wherein, one-accumulate generation is the most frequently used a kind of technology in GM (1,1) model, and it is transportedIt is
Wherein, X(1)=(x(1)(1), x(1)(2) ..., x(1)(n) sequence after once generating) is represented, is represented herein onceSequence after Accumulating generation.
Step 2：Carry out smooth than examining
To ensure GM (1,1) model modelings and the precision and reliability of prediction, modeling sequence must assure that certain lightSlide than ρ (k), its formula is：
The result of smooth ratio, which need to fall within feasible interval, can just ensure the precision and reliability of modeling and prediction.
Step 3：Level is carried out than examining
Level is than the operational formula of inspection：
Likewise, level also needs to fall within feasible interval than the result of inspection can just ensure the acceptable degree of forecast model.
Step 4：To X(1)Make close to generation
In grey differential equation formula, x(1)(t=k) value will be seated x(1)And x (k-1)(1)(k) (because x between(1)(t) it is monotonically increasing function), therefore can x(1)(t=k) it will be indicated as：
z(1)(k)=α x(1)(k)+(1-α)x(1)(k-1)
Wherein, α ∈ [0,1].In general, if without particular requirement, the mode close to average generation, i.e. α=0.5 can be adopted.
Step 5：Estimate the parameter of GM (1,1) basic model
The basic conception of GM (1,1) model, is exactly that the sequence X after generation is fitted using grey differential equation(1), with progress afterContinuous modeling and prediction.Grey differential equation is：
Wherein, α is referred to as development coefficient, is referred to as grey input quantity.However, grey differential equation system is continuous function, it is necessary to pass throughBy means of the mode in generation, using the shadow equation of grey difference equation, that is, GM (1,1) model, the parameter of grey differential equation is estimated, thenIn generation, returns grey differential equation modeling.The grey difference equation of GM (1,1) model is：
x(0)(k)+az(1)(k)=b
In GM (1,1) model, original series and formation sequence are discrete series, it is therefore necessary to by the side by means of generationFormula, is borrowed for the differential equation with difference equation, and it is borrowed is for mode：
x(1)(t)=z(1)(k)
Then, grey difference equation and least squares method are utilized, you can obtain the estimate of basic model.Order：
It can obtain：
Step 6：Determine grey differential equation and time response formula
After a and the decision of b values, grey differential equation formula is taken back, and ask the general solution of its single order linear differential equation to be：
The primary condition of GM (1,1) model is set in (1, x(1)(1)), can obtain specific solution is：
Also known as time response formula.As 1≤t≤n, the result of gained is referred to as the analogue value；Work as t>During n, its result is referred to as pre-Measured value.Therefore, GM (1,1) p step predicted values are：
Step 7：Try to achieve X(1)Simulation and forecasting sequence
By X(1)Simulation and forecasting sequence arrange and be：
ArriveFor the simulation assessed value of GM (1,1) model；ArriveFor GM (1,1) mouldThe forecast assessment value of type.
Step 8：It is rightCarry out inverse generation in the hope of
, will using the mode of inverse generationIt is reduced toSimulation and forecasting sequence for original series.ReductionMode be generating function inverse function.So that one-accumulate is generated as an example, its inverse generation is the method that inverse one-accumulate is generated,I.e.：
Step 9：Examine error
The error of original series kth point is defined as：
ε=(ε (2), ε (3) ..., ε (n))
For error sequence, error-checking criterion includes：
1. error sum of squares：
S=εTε
2. relative error：
3. average relative error：
(2) background value improves
Background value is the important ring of influence precision of prediction in GM (1,1) arithmetic logic, and the adjustment of wherein α values is background valueKey parameter.It is below the computing flow of improvement GM (1,1) Model Background value：
Because grey differential equation is a kind of exponential function, therefore the present embodiment makes x(1)(t) it is：
x(1)(t)=AeBt+C
Then, formula is substituted into
It can obtain：
T is equal to k, k-1 and k-2 Fen Do and substitutes into formula, can be obtained：
x(1)(k)=AeBk+C
x(1)(k-1)=AeB(k-1)+C
x(1)(k-2)=AeB(k-2)+C
It can be calculated：
B=lnx(0)(k)-lnx(0)(k-1)
B values generation time formula , The can be obtained into A is
A values and B values are substituted into formula, and calculate C values and is：
Finally, B values and C values are substituted into formula, the background value after must can improving is：
In the arithmetic logic that new background value is substituted into GM (1,1) model, you can GM (1,1) model after being improved.To avoid confusion, alleged GM (1,1) is GM (1,1) model after improving below.
(3) specific solution conversion
It is the running concept for showing GM (1,1) model as shown in Figure 3, wherein, x(1)(1) x is arrived(1)(n) it is original seriesOne-accumulate formation sequence, is the modeling foundation of GM (1,1) model mainly.It can be seen that in figure, influence the key of precision of predictionPoint is whether forecasting sequence can fall in Probability Area.However, due to the unstability of System Development, GM (1,1) model differsIt is fixed to fall in Probability Area every time.
(4) residual GM Technology Integration
Effectively to grasp the pulsation of System Development, the present embodiment also integrates residual error in addition to improving background value, changing specific solutionCorrection technique, including the periodic error of Fourier space (Fourier series) modified grey model is included, and useExponential smoothing (exponential smoothing technique) corrects remaining random error, and is named as EFGMm(1,1) model and EFGMn (1,1) model, to lift the self adaptive of grey prediction.
Assuming that the time series for having a pen data is x(0)=(x(0)(1), x(0)(2) ..., x(0)(n)), the present embodiment is with thisThe arithmetic logic of EFGMm (1,1) models and EFGMn (1,1) model is described as follows for target.Due to EFGMm (1,1) models withOnly in the kenel of residual sequence, different (EFGMm (1,1) residual sequence has n to the difference of EFGMn (1,1) model calculation logicData, because its not using any specified point as specific solution where；EFGMn (1,1) is then to have n-1, because it using nth point is used as spyFixed solution), but the correctness of calculation is had no effect on, elastic it can adjust.For purposes of illustration only, it is following with EFGMm (1,1) model exemplified by.
Step 1：System Development trend is grasped using GMm (1,1), and calculates its residual error；
X is calculated using GMm (1,1) model(0)One-step prediction valueAnd a residual sequence E '(0)For：
E′(0)=(ε '(0)(1),ε′(0)(2),...,ε′(0)(n))
Step 2：Utilize residual sequence of Fourier space amendment；
Now, is extracted using Fourier space and lies in a residual sequence E '(0)In periodicity speciality, that is, pass throughFollowing equation is fitted a residual sequence E '(0)。
Wherein, T is a residual sequence E '(0)Length, be n-1.Least squares method deploys result：
C=(PTP)-1PTE′(0)
Wherein：
A residual error simulated series, which can be obtained, is：
The concept of Fourier space is a residual sequence ε '(0)Frequency spectrum (frequency spectra) is converted into,And extract its low-frequency part.In addition, Fourier space can sift out high-frequency part (being the part of noise), for oneThe simulation of secondary residual sequence has good effect, and can obtain quadratic residue sequence E "(0)For：
E″(0)=(ε "(0)(2),ε″(0)(3),...,ε″(0)(n))
Step 3：Utilization index exponential smoothing amendment quadratic residue sequence；
Now, utilization index exponential smoothing, which is extracted, lies in quadratic residue sequence E "(0)In randomness speciality, that is, pass throughFollowing equation fitting quadratic residue sequence E "(0)
Wherein, φ is smoothing factor (smoothing coefficient), and 0<φ<1.
The present embodiment tries to achieve the smoothing factor value minimum with quadratic residue sequence error by optimized technology herein, i.e.,Solve the minimum value of following formula.
It can be further broken into：
Therefore, function can will be optimized to be write as：
After obtaining optimum value, you can try to achieve the one-step prediction value of quadratic residue sequence, be：
Finally combine above correction value, you can the one-step prediction value for trying to achieve original series is：
Above-described algorithm is a circulation, through the technology of rolling, can constantly be predicted and update the systemThe pulsation of development.
The present embodiment sets up two new-type Grey Prediction Model , Fen Do for GMm (1,1) models and GMn (1,1) model.
(1) GMm (1,1) model
GMm (1,1) model is to be solved using least squares method based on simulated series and model the pre- of sequence deviation minimumModel is surveyed, this mode is not with X(1)On any component as primary condition, but entered with the optimized angle of series model precisionRow prediction, has suitable benefit to the lifting of precision of prediction.
The calculation logic of GMm (1,1) model is as follows, coefficient value to be ordered is set into unknown herein, and set up following calculating formula：
Represented in the way of matrix, be：
Order：
Therefore, the estimate that can calculate c using least squares method is：
WillGeneration, which returns formula and can obtain the time response formula of GMm (1,1) model, is：
Therefore, the one-step prediction value of original series is：
Through the running of above-mentioned formula, you can obtain the grey forecasting model minimized based on modeling data deviation.
(2) GMn (1,1) model
Traditional GM (1,1) model is with sequence X(1)One-component as gray model primary condition, it is but thisMode is to the utilization of fresh information and insufficient.If can be with X(1)N-th of component as primary condition, then the utilization of fresh information is justCan be more abundant, the trend of sequence is grasped can be more accurate, and prediction curve can be also substantially improved and falls probability in Probability Area.In consideration of it, GMn (1,1) model is then that specific solution is set in into (n, x(1)(n) on), that is, one newest in simulated seriesNumerical value.
The calculation logic of GMn (1,1) model is as follows.By primary condition (n, x(1)(n) the general of grey differential equation formula) is substituted intoIn solution, you can the time response formula for obtaining GMn (1,1) model is：
Therefore, the 1 step predicted value that also can obtain GMn (1,1) model is：
Therefore, X can further be derived(0)1 step predicted value be：
According to the running of above-mentioned formula, you can the gray model in update based on primary condition is obtained, to lift ashGrasp ability of the color model to future trend.
GM (1,1) models have been formally the forecast model with reliability and practicality on numerous documents,.However,It should also be appreciated that GM (1,1) models are unsatisfactory for the prediction effect of non-stationary system developmental sequence from document.By to passingThe improvement of GM (1,1) Model Background value calculation of uniting, and the specific solution of conversion fall within the probability of Probability Area to increase, after improvementGMm (1,1) models or GMn (1,1) models grasp ability for the trend of turning point and also increase therewith, then be aided with residual GMEtc. technology, then the error of turning point can be effectively corrected, to promote the precision of prediction of material risk source Monitoring Data.
(5) engineering construction material risk source time sequence is predicted.
All monitoring results are analyzed and handled based on adaptive grey Fourier Index Prediction Model, to engineeringConstruction risk source carries out the real-time estimate of time series.When predicted value exceedes early warning value, the automatic bright aobvious alarm of system.