CN102243672A - Gushing operation condition soft sensing modeling method based on hybrid multiple models in shield tunneling process - Google Patents

Abstract

The invention discloses a gushing operation condition soft sensing modeling method based on hybrid multiple models in a shield tunneling process. Hybrid multiple models include the following four models: 1, a data-driven exploration information and construction data-based soil texture soft sensing model established by using a belief rule base method; 2, a data-driven seepage flow soft sensing model established by using a support vector machine method; 3, a data-driven permeability coefficient model established by using the support vector machine method; and 4, a gushing operation condition soft sensing model established by using a simplification mechanism. The soil texture soft sensing model, the seepage flow soft sensing model and the permeability coefficient model are input of the gushing operation condition soft sensing model. The modeling method disclosed by the invention has the advantages that a hybrid modeling method combining mechanism modeling and data-driven modeling is adopted, and a model modeled by using the modeling method is simple and has high interpretability, high reliability and good extrapolation.

Description

Based on the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model

Technical field

The present invention relates to a kind of based on the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model.

Background technology

Shield method is to build one of fruitful construction method of city subterranean tunnel, plays a significant role in China's process of modernization.In recent years, China's big and medium-sized cities shield-tunneling construction development is very fast, successively has tens Tunnel Engineering to go into operation successively.Earth pressure balanced shield, EPBS is applied widely because of it, the low widespread use by the time in tunnel piercings such as moisture weak soil, soft rock, hard rock and mixing stratum of cost.

Fig. 1 has shown that the shield structure sends out the typical process flow of construction.Balancing earth-pressure shielding machine utilization knife cutting the place ahead soil body foremost.The potpourri passing hole channel of soil that cuts and water enters native cabin.Soil in native cabin, aqueous mixtures accumulate gradually and form certain supporting pressure.The spiral dumper is got rid of the native aqueous mixtures of bottom in the native cabin from native bilge portion.

In tunneling process, too high and can't produce supporting pressure as the water and soil mixture pressure in the native cabin, a large amount of water is carried silt secretly and is gushed out to form from the spiral dumper and gush, even forms pipeline and collapse, and problems such as surface collapse have a strong impact on shield structure normal construction.For addressing this problem, need in native cabin, add foam and bentonitic clay, with improvement soil nature, the plastic flow state of the potpourri that conserves water and soil.But dosage that adds and time are difficult to hold, and are subjected to the influence of water and soil mixture state and shield-tunneling construction state.

Present construction state is, the manual analysis boring parameter forms the judgement of soil body improvement and instructs the modification operation of the soil body in the native cabin with people's construction experience.The shortcoming of doing like this is:

1, manual analysis is longer time delay, is unfavorable for the strain rate effect of shield driving process.

2, the artificial experience error is bigger, can not accurately judge construction risk.

Summary of the invention

The purpose of this invention is to provide a kind of based on the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model.

In the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model, mix multi-model and be by based on the soil property soft-sensing model of exploration information and construction data, based on the seepage flow soft-sensing model of data-driven, constitute based on the infiltration coefficient soft-sensing model of data-driven with based on the operating mode soft-sensing model of gushing of mechanism, wherein soil property soft-sensing model, seepage flow soft-sensing model, infiltration coefficient soft-sensing model are the inputs of operating mode soft-sensing model of gushing; The concrete steps of modeling method are as follows:

1) select fltting speed, cutterhead rotating speed, cutterhead moment of torsion, cutterhead thrust, be CL the group of local soil type _X, CL _Y, CL _ZOr CL _ADifferent local soil types in, comprise the joint distribution local soil type quantity of specific stratum A and soil layer X, Y and four kinds of soil layers of Z

The group of local soil type is CL _X, CL _Y, CL _ZOr CL _ADifferent local soil types in, only comprise local soil type's quantity of the joint distribution of soil layer X, Y and Z

1 T (t on the tunnel axis _x, t _y, t _z) near geotechnical boring BH _iThe horizontal projection distance at place, the correlation parameter of soil are as auxiliary variable, go out the stochastic distribution probability of specified point place soil layer according to the geologic prospecting information calculations, and then letter rule base method is put in utilization, foundation is based on the soil property soft-sensing model of exploration information and construction data, and the correlation parameter of soil is meant that specific heat at constant volume, the weight of sieving of soil account for the particle diameter of 10% soil, the initial seepage flow coefficient of soil, the original pressure of soil, the compressibility coefficient of soil, the unit weight of soil, the nonuniformity coefficient of soil, the plasticity index of soil, the liquid limit of soil, the initial void ratio of soil;

2) select native specific heat, cutterhead rotating speed, cutterhead moment of torsion, native cabin pressure, thrust, fltting speed, native cabin temperature, spiral dumper rotating speed as auxiliary variable, utilize support vector machine method, set up seepage flow soft-sensing model based on data-driven; Select native cabin pressure, initial void ratio, compressibility of soil, initial soil pressure, the weight of sieving to account for the particle diameter of 10% soil, initial infiltration coefficient, utilize support vector machine method, set up infiltration coefficient soft-sensing model based on data-driven as auxiliary variable;

3) select infiltration coefficient, seepage flow, the weight of sieving to account for the particle diameter of 10% soil, native severe, set up the soft-sensing model of the operating mode of gushing as the input of simplifying mechanism model.

Described step 3) is:

1) calculates seepage pressure according to seepage flow and infiltration coefficient

Wherein K is an infiltration coefficient, and A is a native cabin sectional area, and Q is a seepage flow, d ₁₀' account for the particle diameter of 10% soil for the weight of sieving;

2) calculate soil particle severe in water according to grogs footpath, native unit weight

γ wherein _sBe soil body unit weight, γ is the unit weight of water, d ₁₀' account for the particle diameter of 10% soil for the weight of sieving;

3) the gush judgement of operating mode

If seepage pressure P is greater than soil body deadweight G or near soil body deadweight then very easily gush, if seepage pressure P is much smaller than soil body deadweight G then difficult gushing.

Utilize in the described step 1) and put letter rule base method and be:

According to training data { (x _i, y _i), i=1,2 ..., l} sets up the soil property parameter rule storehouse model of following form

The rule weight is θ _k

k∈{1，…，L}

Wherein

(i ∈ 1 ..., Tk}) be the reference value of i input variable of k bar rule, T _kBe the number of the prerequisite attribute used in the k bar rule, D _Jk(j ∈ 1 ..., N}) be the j kind result of k bar rule,

Be D _JkDegree of confidence, L be in the rule base rule number, regular weight θ _kAnd the output result is D _JkConfidence level Training sample by construction note provides.

Described step 2) support vector machine method is in:

Given training data { (x _i, y _i), i=1,2 ..., l} obtains following nonlinear model by study

$f\left(x\right)=\underset{x_i\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})K(x_i,x)+b,$ Wherein $K(x,x_i)=\exp \{-\frac{\left|\right|x-x_i\left|\right|}{2{\mathrm{\&sigma;}}^2}\}$ Be the radially basic kernel function of Gauss, $b=\frac{1}{N_{\mathrm{NSF}}}\left\{\underset{0<{\mathrm{\&alpha;}}_i<C}{\mathrm{\&Sigma;}}\right[y_i-\underset{x_j\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})K(x_j,x_i)-\mathrm{\&epsiv;}]+\underset{0<{\mathrm{\&alpha;}}_i^{*}<C}{\mathrm{\&Sigma;}}[y_i-\underset{x_j\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})K(x_j,x_i)+\mathrm{\&epsiv;}\left]\right\},$

Be support vector, N _NSFBe support vector quantity, ε is linear insensitive loss parameter.

The present invention compared with prior art has beneficial effect:

1) the unusual service condition model based on mixture model is simple, good reliability;

2) model has adaptivity, can adapt to the non-linear variation in stratum;

3) use algorithm of support vector machine, it is saturated can not to be absorbed in sample.

Description of drawings

Fig. 1 is the typical process flow of shield driving process;

Fig. 2 is based on the operating mode soft-measuring modeling method process flow diagram of gushing in the shield driving process of mixing multi-model.

Embodiment

In the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model, mix multi-model and be by based on the soil property soft-sensing model of exploration information and construction data, based on the seepage flow soft-sensing model of data-driven, constitute based on the infiltration coefficient soft-sensing model of data-driven with based on the operating mode soft-sensing model of gushing of mechanism, wherein soil property soft-sensing model, seepage flow soft-sensing model, infiltration coefficient soft-sensing model are the inputs of operating mode soft-sensing model of gushing; The concrete steps of modeling method are as follows:

1) select fltting speed, cutterhead rotating speed, cutterhead moment of torsion, cutterhead thrust, be CL the group of local soil type _X, CL _Y, CL _ZOr CL _ADifferent local soil types in, comprise the joint distribution local soil type quantity of specific stratum A and soil layer X, Y and four kinds of soil layers of Z

The group of local soil type is CL _X, CL _Y, CL _ZOr CL _ADifferent local soil types in, only comprise local soil type's quantity of the joint distribution of soil layer X, Y and Z

1 T (t on the tunnel axis _x, t _y, t _z) near geotechnical boring BH _iThe horizontal projection distance at place, the correlation parameter of soil are as auxiliary variable, go out the stochastic distribution probability of specified point place soil layer according to the geologic prospecting information calculations, and then letter rule base method is put in utilization, foundation is based on the soil property soft-sensing model of exploration information and construction data, and the correlation parameter of soil is meant that specific heat at constant volume, the weight of sieving of soil account for the particle diameter of 10% soil, the initial seepage flow coefficient of soil, the original pressure of soil, the compressibility coefficient of soil, the unit weight of soil, the nonuniformity coefficient of soil, the plasticity index of soil, the liquid limit of soil, the initial void ratio of soil;

2) select native specific heat, cutterhead rotating speed, cutterhead moment of torsion, native cabin pressure, thrust, fltting speed, native cabin temperature, spiral dumper rotating speed as auxiliary variable, utilize support vector machine method, set up seepage flow soft-sensing model based on data-driven; Select native cabin pressure, initial void ratio, compressibility of soil, initial soil pressure, the weight of sieving to account for the particle diameter of 10% soil, initial infiltration coefficient, utilize support vector machine method, set up infiltration coefficient soft-sensing model based on data-driven as auxiliary variable;

3) select infiltration coefficient, seepage flow, the weight of sieving to account for the particle diameter of 10% soil, native severe, set up the soft-sensing model of the operating mode of gushing as the input of simplifying mechanism model.

Described step 3) is:

1) calculates seepage pressure according to seepage flow and infiltration coefficient

Wherein K is an infiltration coefficient, and A is a native cabin sectional area, and Q is a seepage flow, d ₁₀' account for the particle diameter of 10% soil for the weight of sieving;

2) calculate soil particle severe in water according to grogs footpath, native unit weight γ wherein _sBe soil body unit weight, γ is the unit weight of water, d ₁₀' account for the particle diameter of 10% soil for the weight of sieving;

3) the gush judgement of operating mode

If seepage pressure P is greater than soil body deadweight G or near soil body deadweight then very easily gush, if seepage pressure P is much smaller than soil body deadweight G then difficult gushing.

The stochastic distribution probabilistic method that goes out specified point place soil layer according to the geologic prospecting information calculations in the described step 1) is:

1) the joint distribution probability of soil layer

The joint distribution probability of soil layer should be determined jointly by other soil layer that its adjacent geotechnical boring obtains.Suppose geotechnical boring BH _iIn local soil type's type be X-Y-Z, promptly soil layer is respectively X, Y, Z, then with this geotechnical boring BH _iMay exist specific soil layer A and X, Y, Z distribution probability computing formula linked together to be in the adjacent area:

$p{\left(A\right)}_{{\mathrm{BH}}_i}=\frac{N{\left(A\right)}_{{\mathrm{BH}}_i}}{N{\left(A\right)}_{{\mathrm{BH}}_i}+N{\left(\stackrel{\&OverBar;}{A}\right)}_{{\mathrm{BH}}_i}}$

In the formula:

Be specific soil layer the joint distribution probability,

Be to be CL the group of local soil type _X, CL _Y, CL _ZOr CL _ADifferent local soil types in, comprise specific stratum A and soil layer X, Y and four kinds of soil layers of Z joint distribution local soil type quantity,

Be to be CL the group of local soil type _X, CL _Y, CL _ZOr CL _ADifferent local soil types in, only comprise local soil type's quantity of the joint distribution of soil layer X, Y and Z.

2) the stochastic distribution probability of specified point place soil layer

Specific soil layer can be judged according to geology borehole data data along the situation that the engineering longitudinal axis occurs in the zone, promptly according to the specified point of the engineering vertical section of need research, by soil layer information and the joint distribution rule that analysis closes on geotechnical boring, calculate specific soil layer and have the stochastic distribution probability along the engineering longitudinal axis.Suppose that along the coordinate of 1 T of engineering longitudinal axis be T (t _x, t _y, t _z), the different geotechnical boring column of n local soil type data is arranged near this T, then this some place stochastic distribution probability that specific soil layer A occurs is:

$R{\left(A\right)}_T=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}P{\left(A\right)}_{{\mathrm{BH}}_i}/d_i}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}1/d_i}$

In the formula: R (A) _TBe along 1 T (t on the engineering longitudinal axis _x, t _y, t _z) locate to occur the stochastic distribution probability of specific soil layer A;

It is the joint distribution probability that comprises specific soil layer A near the adjacent geotechnical boring; d _iBe 1 T (t on the tunnel axis _x, t _y, t _z) near geotechnical boring BH _iThe horizontal projection distance at place.

Generally, if geotechnical boring BH _iApart from specified point T (t _x, t _y, t _z) distance located is near more, then the soil layer data information that provides of this boring is reliable more.In order to reflect geotechnical boring BH _iThe geological information that provides is worth, and adopts engineering geology statistical analysis technique definition d _iComputing formula be:

$d_i=\sqrt{{(t_x-x_{\mathrm{BHi}})}^2+{(t_y-y_{\mathrm{BHi}})}^2}$

In the formula: d _iBe specified point T (t _x, t _y, t _z) and geotechnical boring

Between the horizontal projection distance.

Utilize in the described step 1) and put letter rule base method and be:

The distribution probability of certain specific soil layer in some place, and behind the fltting speed in the shield-tunneling construction process, cutterhead thrust, cutterhead moment of torsion, cutterhead rotating speed discretize as the prerequisite property set of rule base, the initial infiltration coefficient of the specific heat of the relevant soil body, the soil body, soil body compressibility coefficient, initial soil pressure, soil body particle diameter, as the output result, set up the rule base model after the unit weight discretize of the soil body.Concrete implementation step is as follows

Usually, the rule of the k bar in traditional rule storehouse form is as follows:

Wherein (i ∈ 1 ..., T _k) be the reference value of i input variable of k bar rule, T _kBe the number of the prerequisite attribute used in the k bar rule, D _j(j ∈ 1 ..., N}) be the result of k bar rule.

Thereby the distribution of various possible outcomes makes rule comprise abundanter information in the rule in order to embody, and on the basis of traditional rule, puts letter rule base method and has adopted a kind of expression mechanism of more approaching reality promptly to put the letter rule.Result and degree of confidence occur with distributed frame in putting the letter rule.

It is as follows to put the letter rule format:

Rule

Weight is θ _kk∈{1，…，L}

D wherein _Jk(j ∈ 1 ..., N}) be the j kind result of k bar rule,

Be D _JkDegree of confidence.L is the number of rule in the rule base.

Based on the method for expressing of putting the letter rule, can qualitative data and quantitative data unification be converted to have that degree of confidence distributes put the letter rule, and can not cause information loss.

Regular Method of Weight Determination is in the rule base:

Given training data { (x _i, y _i), i=1,2 ..., satisfy condition among the l}

Be recorded as a bar, wherein corresponding output result is D _Jk(j ∈ 1 ..., N}) be recorded as a _jBar, and

Then the weight of regular k is ${\mathrm{\&theta;}}_k=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{a}_j/a.$

Definite method of confidence level is in the rule base:

Given training data { (x _i, y _i), i=1,2 ..., satisfy condition among the l}

Be recorded as a bar, wherein corresponding output result is D _Jk(j ∈ 1 ..., N}) be recorded as a _jBar, and

Then the output result of regular k is D _JkDegree of confidence be ${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_{\mathrm{jk}}=a_j/\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{a}_j.$

Described step 2) support vector machine method is in:

According to given training data { (x _i, y _i), i=1,2 ..., l}, wherein x _i∈ R ^dBe the input value of i learning sample, and be a d dimensional vector $x_i={\left[\begin{array}{cccc}{x}_i^1& x_i^2& ...& x_i^d\end{array}\right]}^T,$ y _i∈ R is corresponding desired value.We define linear ε insensitive loss function earlier

${|y-f\left(x\right)|}_s=\left\{\begin{array}{cc}0& |y-f\left(x\right)|\&le;\mathrm{\&epsiv;}\\ |y-f\left(x\right)|-\mathrm{\&epsiv;}& |y-f\left(x\right)|>\mathrm{\&epsiv;}\end{array}\right.---\left(1\right)$

If promptly the difference between the value f (x) of the recurrence estimation function of desired value y and process learn configuration is less than ε, then loss equals 0.

Support vector machine is by the suitable kernel function K (x of definition _i, x _j), will import the sample space nonlinear transformation to another feature space, levy structure recurrence estimation function in the space at this.Select the radially basic kernel function of Gauss

The Gauss radial basis function then is the strong kernel function of locality, and its extrapolability weakens along with the increase of parameter σ.

Therefore at nonlinear learning sample collection { (x _i, y _i), i=1,2 ..., l}, the recurrence estimation function of establishing nonlinear case is

f(x)＝w ^T·φ(x)+b (2)

And then it is right to seek w, b, makes under the constant prerequisite of formula (1) to minimize

Consider simultaneously when constraint condition can not realize slack variable ξ _i,

Introduce, optimization problem is like this

$\left\{\begin{array}{cc}\underset{\mathrm{\&omega;}}{\mathrm{Min}}& \frac{1}{2}{w}^{T}w+C\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\mathrm{\&xi;}}_i+{\mathrm{\&xi;}}_i^{*})\\ s.t.& y_i-w\&CenterDot;\mathrm{\&phi;}\left(x_i\right)-b\&le;\mathrm{\&epsiv;}+{\mathrm{\&xi;}}_i\\ & w\&CenterDot;\mathrm{\&phi;}\left(x_i\right)+b-y_i\&le;\mathrm{\&epsiv;}+{\mathrm{\&xi;}}_i^{*}\\ & {\mathrm{\&xi;}}_i\&GreaterEqual;0\\ & {\mathrm{\&xi;}}_i^{*}\&GreaterEqual;0i=\mathrm{1,2},...,l\end{array}\right.---\left(3\right)$

Utilize method of Lagrange multipliers to find the solution this constrained optimization problem, be constructed as follows Lagrangian function for this reason:

$L_p=\frac{1}{2}{\left|\right|w\left|\right|}^2+C\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\mathrm{\&xi;}}_i+{\mathrm{\&xi;}}_i^{*})-\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i(\mathrm{\&epsiv;}+{\mathrm{\&xi;}}_i-y_i+w\&CenterDot;\mathrm{\&phi;}\left(x_i\right)+b)-$

$\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i^{*}(\mathrm{\&epsiv;}+{\mathrm{\&xi;}}_i^{*}+y_i-w\&CenterDot;\mathrm{\&phi;}\left(x_i\right)-b)-\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\mathrm{\&beta;}}_i{\mathrm{\&xi;}}_i+{\mathrm{\&beta;}}_i^{*}{\mathrm{\&xi;}}_i^{*})---\left(4\right)$

According to Optimum Theory, with L _pRespectively to w, b, ξ _i,

Asking partial differential and making it is 0

$\left\{\begin{array}{c}w=\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})\mathrm{\&phi;}\left(x_i\right)\\ \underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})=0\\ C-{\mathrm{\&alpha;}}_i-{\mathrm{\&beta;}}_i=0\\ C-{\mathrm{\&alpha;}}_i^{*}-{\mathrm{\&beta;}}_i^{*}=0\end{array}\right.---\left(5\right)$

With (5) substitution (4), obtain the antithesis optimization problem

$\left\{\begin{array}{cc}\underset{{\mathrm{\&alpha;}}_i,{\mathrm{\&alpha;}}_i^{*}}{\mathrm{Max}}& [-\frac{1}{2}\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})K(x_i,x_j)-\mathrm{\&epsiv;}\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\mathrm{\&alpha;}}_i+{\mathrm{\&alpha;}}_i^{*})+\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}{y}_i({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})]\\ s.t.& \underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})=0\\ & 0{\&le;\mathrm{\&alpha;}}_i\&le;C\\ & 0\&le;{\mathrm{\&alpha;}}_i^{*}\&le;C\end{array}\right.---\left(6\right)$

Wherein has only partial parameters

Be not 0, they are exactly the support vector (SV) in the problem.Thereby be by the recurrence estimation function that study obtains

$f\left(x\right)=\underset{x_i\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})K(x_i,x)+b---\left(7\right)$

Wherein:

$b=\frac{1}{N_{\mathrm{NSF}}}\left\{\underset{0<{\mathrm{\&alpha;}}_i<C}{\mathrm{\&Sigma;}}\right[y_i-\underset{x_j\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})K(x_j,x_i)-\mathrm{\&epsiv;}]+\underset{0<{\mathrm{\&alpha;}}_i^{*}<C}{\mathrm{\&Sigma;}}[y_i-\underset{x_j\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})K(x_j,x_i)+\mathrm{\&epsiv;}\left]\right\}---\left(8\right)$

N in the following formula _NSFBe support vector quantity.

Seepage flow is identical with the computation process of infiltration coefficient, and its corresponding auxiliary variable sees Table 1.

Terminological interpretation of the present invention:

Group of local soil type and local soil type: the difference of the topsoil kind that obtains according to boring is divided into different cohort CL with geotechnical boring local soil type _i, the different groups' of local soil type stochastic distribution probability can be represented the distribution characteristics of endosexine, engineering region soil.In like manner, may there be the multiple different type SF of local soil type among the same group of local soil type _A, different local soil types type has reflected the combination rule of serial soil layer.

Construction note: construction note is made up of three parts, first is geologic prospecting in early stage data, should write down and comprise the geotechnical boring position, the group of local soil type that geotechnical boring is gathered, local soil type, the specific heat of soil, the weight of sieving accounts for the particle diameter of 10% soil, the initial seepage flow coefficient of soil, the original pressure of soil, the compressibility coefficient of soil, the unit weight of soil, the nonuniformity coefficient of soil, the plasticity index of soil, the liquid limit of soil, the initial void ratio of soil, adopt in artificial exploration record and the entering computer database, data in the second portion shield structure operational process, should write down and comprise the cutterhead moment of torsion, cutterhead thrust, fltting speed, the cutterhead rotating speed, spiral dumper rotating speed, soil cabin pressure, soil cabin temperature, gross thrust, use PLC gathers and is recorded in the real-time data base, the spiral dumper is discharged the data of soil in the third part shield structure operational process, should write down and comprise collection position, acquisition time, the specific heat of soil, the weight of sieving accounts for the particle diameter of 10% soil, the compressibility coefficient of soil, the unit weight of soil, the nonuniformity coefficient of soil, the plasticity index of soil, the liquid limit of soil, the void ratio of soil, gathered by spiral dumper soil outlet, artificial off-line analysis obtains and charges in the Computer Database.

Soft measurement: in the modern industry production run, have the variable of a big class reactor product quality, they are for technology or reason economically, can not directly measure with conventional sensors, but come out by in-line analyzer detection or lab analysis, analytical cycle is long, and it is big to lag behind.But the real-time detection of these variablees for the continuous quiet run that guarantees product quality and process units and the productive capacity etc. of giving full play to device all tool play a very important role.If use the laboratory values that lags behind can't correctly reflect real-time operating mode, the operative employee leaves the economic benefit that excessive leeway also can reduce process units to be on the safe side usually simultaneously.In order to address this problem, soft-measuring technique has been obtained very big progress in recent years.Soft-measuring technique just is to use the process variable and the corresponding product quality analysis value of easy survey, adopt certain modeling method (modelling by mechanism, regression modeling etc.) set up a quality estimation model, under the prerequisite that does not increase hardware investment, estimate product quality in real time by this model.

Mixture model: soft-measuring modeling method can be divided into pure modelling by mechanism method and usually based on the modeling method of data-driven.These two kinds of methods respectively have its relative merits: the former can be from reflecting the rule of process in essence, the reliability height, and extrapolation is good, has interpretation; Shortcoming is that modeling process is more loaded down with trivial details, and for some complex processes, getable mechanism model generally also is through the model after some simplification.The latter is directly according to the direct modeling of I/O data of process, almost need not the priori of process object, but shortcoming also is significantly, is example with the neural network, as a kind of modeling method of black box, pace of learning is slow, promote poor performance, and model do not have interpretation, be difficult to determine that appropriate network architecture and study stop index, cause the over-fitting phenomenon easily, even noise also match may be come in.Because pure modelling by mechanism and caused hybrid modeling thought based on the limitation of these two kinds of methods of data-driven modeling, promptly for there being the process of simplifying mechanism model, can will simplify mechanism model and combine based on the model of data-driven, complement one another, simplify the priori that mechanism model provides, can save training sample for model based on data-driven; Model based on data-driven can compensate the not modeling characteristic of simplifying mechanism model again simultaneously.

The present invention is further illustrated below in conjunction with accompanying drawing:

1, the coupling of the sequential between leading variable and the process variable

The promptly definite leading variable of sequential coupling is with respect to the retardation time of each process variable.Because the existence of various disturbing factors, actual industrial process is difficult to reach desirable stable state, therefore, in order to make soft-sensing model under the situation that various disturbing factors exist, still can estimate the variation tendency of leading variable more exactly, when gathering training sample, must specify the retardation time of each model input variable correspondence.The sequential of each auxiliary variable coupling needs to calculate according to shield structure design parameter and Tunnel Design parameter etc. in the operating mode of gushing in the shield driving process soft-sensing model.

Table 1 is depicted as each auxiliary variable and corresponding identification symbol thereof in the operating mode soft-sensing model of gushing in the shield driving process.

Table 1 auxiliary variable and identifier thereof

Claims (4)

1. one kind based on the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model, it is characterized in that mixing multi-model and be by based on the soil property soft-sensing model of exploration information and construction data, based on the seepage flow soft-sensing model of data-driven, constitute based on the infiltration coefficient soft-sensing model of data-driven with based on the operating mode soft-sensing model of gushing of mechanism, wherein soil property soft-sensing model, seepage flow soft-sensing model, infiltration coefficient soft-sensing model are the inputs of operating mode model of gushing; The concrete steps of modeling method are as follows:

1) select fltting speed, cutterhead rotating speed, cutterhead moment of torsion, cutterhead thrust, be CL the group of local soil type _X, CL _Y, CL _ZOr CL _ADifferent local soil types in, comprise the joint distribution local soil type quantity of specific stratum A and soil layer X, Y and four kinds of soil layers of Z

The group of local soil type is CL _X, CL _Y, CL _ZOr CL _ADifferent local soil types in, only comprise local soil type's quantity of the joint distribution of soil layer X, Y and Z

1 T (t on the tunnel axis _x, t _y, t _z) near geotechnical boring BH _iThe horizontal projection distance at place, the correlation parameter of soil are as auxiliary variable, go out the stochastic distribution probability of specified point place soil layer according to the geologic prospecting information calculations, and then letter rule base method is put in utilization, foundation is based on the soil property soft-sensing model of exploration information and construction data, and the correlation parameter of soil is meant that specific heat at constant volume, the weight of sieving of soil account for the particle diameter of 10% soil, the initial seepage flow coefficient of soil, the original pressure of soil, the compressibility coefficient of soil, the unit weight of soil, the nonuniformity coefficient of soil, the plasticity index of soil, the liquid limit of soil, the initial void ratio of soil;

2) select native specific heat, cutterhead rotating speed, cutterhead moment of torsion, native cabin pressure, thrust, fltting speed, native cabin temperature, spiral dumper rotating speed as auxiliary variable, utilize support vector machine method, set up seepage flow soft-sensing model based on data-driven; Select native cabin pressure, initial void ratio, compressibility of soil, initial soil pressure, the weight of sieving to account for the particle diameter of 10% soil, initial infiltration coefficient, utilize support vector machine method, set up infiltration coefficient soft-sensing model based on data-driven as auxiliary variable;

3) select infiltration coefficient, seepage flow, the weight of sieving to account for the particle diameter of 10% soil, native severe, set up the soft-sensing model of the operating mode of gushing as the input of simplifying mechanism model.

2. according to claim 1 a kind of based on the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model, it is characterized in that described step 3) is:

1) calculates seepage pressure according to seepage flow and infiltration coefficient

Wherein K is an infiltration coefficient, and A is a native cabin sectional area, and Q is a seepage flow, d ₁₀' account for the particle diameter of 10% soil for the weight of sieving;

2) calculate soil particle severe in water according to grogs footpath, native unit weight

γ wherein _sBe soil body unit weight, γ is the unit weight of water, d ₁₀' account for the particle diameter of 10% soil for the weight of sieving;

3) the gush judgement of operating mode

If seepage pressure P is greater than soil body deadweight G or near soil body deadweight then very easily gush, if seepage pressure P is much smaller than soil body deadweight G then difficult gushing.

3. according to claim 1 a kind of based on the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model, it is characterized in that utilizing in the described step 1) and put letter rule base method and be:

According to training data { (x _i, y _i), i=1,2 ..., l} sets up the soil property parameter rule storehouse model of following form

The rule weight is θ _k

k∈{1，…，L}

Wherein

(i ∈ 1 ..., Tk}) be the reference value of i input variable of k bar rule, T _kBe the number of the prerequisite attribute used in the k bar rule, D _Jk(j ∈ 1 ..., N}) be the j kind result of k bar rule,

Be D _JkDegree of confidence, L be in the rule base rule number, regular weight θ _kAnd the output result is D _JkConfidence level

Training sample by construction note provides.

4. according to claim 1 a kind of, it is characterized in that described step 2 based on the operating mode soft-measuring modeling method of gushing in the shield driving process of mixing multi-model) in support vector machine method be:

Given training data { (x _i, y _i), i=1,2 ..., l} obtains following nonlinear model by study $f\left(x\right)=\underset{x_i\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})K(x_i,x)+b,$ Wherein $K(x,x_i)=\exp \{-\frac{\left|\right|x-x_i\left|\right|}{2{\mathrm{\&sigma;}}^2}\}$ Be the radially basic kernel function of Gauss, $b=\frac{1}{N_{\mathrm{NSF}}}\left\{\underset{0<{\mathrm{\&alpha;}}_i<C}{\mathrm{\&Sigma;}}\right[y_i-\underset{x_j\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})K(x_j,x_i)-\mathrm{\&epsiv;}]+\underset{0<{\mathrm{\&alpha;}}_i^{*}<C}{\mathrm{\&Sigma;}}[y_i-\underset{x_j\&Element;\mathrm{SV}}{\mathrm{\&Sigma;}}({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})K(x_j,x_i)+\mathrm{\&epsiv;}\left]\right\},$

Be support vector, N _NSFBe support vector quantity, ε is linear insensitive loss parameter.

Images