CN102176123A - System and method for predicating and controlling blasting vibration - Google Patents

Abstract

The invention discloses a system and method for predicating and controlling blasting vibration. The system comprises an engineering information acquisition module, an importance degree analysis module and a prediction model module, wherein the engineering information acquisition module is used for acquiring blasting information; the importance degree analysis module is used for processing the information and outputting the importance degree of blasting parameters on the monitoring attribute; and the prediction model module is used for outputting the blasting vibration strength or dominant frequency or duration time. By using the system and method for predicating and controlling blasting vibration, which are disclosed by the invention, the comprehensive and high-precision predication on blasting vibration parameters can be realized or various influence factors can be discriminated, wherein the discriminating range extends from the traditional quantitative factors (such as sectional explosive amount, horizontal distance, total explosive amount, a step height, and the like) to qualitative factors (such as throwing direction, aperture, explosive types, and the like) for importance degree analysis; and the high comprehensiveness is realized; control information can be output efficiently, quickly and stably; related report forms are provided; and the safety of the engineering blasting construction process is ensured.

Description

A kind of blasting vibration forecast and control system and control method thereof

Technical field

The present invention relates to the engineering explosion field, be specifically related to a kind of vibration and forecast system and control method thereof with security control engineering explosion construction.

Background technology

At present, the modern project explosion has become national economy and the indispensable special trade of national defense construction: for example, and massif and excavation of foundation pit during nuclear power is built, tunnel in the traffic engineering and pile foundation excavation, and in the work progresss such as crowded silt in the engineering of marine site, all need to use blasting technique.And a lot of engineerings are all carried out by stages.With nuclear power capital construction is example, and blast working often runs through nuclear power builds first phase, second phase etc., promptly, in the time of the first phase unit generation, may also explode mountain, deep excavation project in not far place, this just requires unit in charge of construction under the prerequisite of guaranteeing the nuclear power installation security, accelerates the blast working progress.But,, will cause security incident, even cause casualties if bad to the control of the vibration situation in the blasting process.

Traditional forecasting procedure based on experimental formula, by section dose and horizontal range, elevation correction experimental formula is also considered difference of elevation, can only obtain oscillation intensity, as vibration acceleration peak value or vibration velocity peak value, yet there is following defective in the experimental formula method: (1) is based on number of samples in essence and is tending towards infinitely-great progressive theory, have only when sampled data is tending towards infinite, the training result of parametric technique just is tending towards real model, obviously, the actual sample number of blasting vibration is limited, is difficult to satisfy this prerequisite; (2) can only forecast oscillation intensity.Existing relevant national standard has required to consider simultaneously oscillation intensity and frequency; (3) restricted application.External conditions such as explosion form, place and landform are had dependence, in case variation has taken place these conditions, these values will change thereupon.Though also have some intelligent algorithms now, can forecast oscillation intensity, dominant frequency and duration simultaneously, as the neural network method of prediction.But, present nerual network technique normally obtains inspiration from biological theory and some academic schools, lack unified mathematical theory basis, the selection of network parameter still need be by means of experience, and the scarcity of training sample makes that also the generalization ability of neural network is relatively poor.

The pinking Forecast And Control Technique need take suitable method that the blasting vibration parameter is carried out comprehensive, high-precision forecast on the one hand, and another side also needs numerous influence factors are screened.Traditional discriminating method can only be analyzed quantitative factor, as section dose, horizontal range, total dose, bench height etc., can't carry out the significance level analysis to qualitative factor such as thrown direction, aperture, explosive type etc.

Summary of the invention

Goal of the invention: in order to overcome the deficiencies in the prior art, the invention provides a kind of system and the control method thereof that can forecast and control the vibration of engineering explosion, the present invention can be efficient, convenient, stable the output control information, and provide relevant form, comprehensive height can be guaranteed the security of engineering explosion work progress.

Technical scheme: for achieving the above object, blasting vibration forecast of the present invention and control system comprise with lower module:

The engineering information collecting module is used to gather explosion information, and explosion information comprises blasting parameter and monitoring attribute; The importance degree analysis module is used for the information of engineering information collecting module collection is handled, and the output blasting parameter is to the importance degree of monitoring attribute; The forecasting model module is used for the information of engineering information collecting module collection is handled, and output blasting vibration intensity or dominant frequency or duration.

The control method of blasting vibration forecast of the present invention and control system may further comprise the steps:

A. obtaining explosion information: i implements blasting operation for the first time and carries out on-site data gathering acceptor according to the blasting parameter of design in advance, obtain the monitoring attribute of this separate explosion: blasting parameter comprises explosive type, aperture, thrown direction, section dose, total dose, hole depth, bench height, difference of elevation, horizontal range and unit consumption etc., and the monitoring attribute comprises amplitude (particle vibration acceleration or velocity peak values), dominant frequency and duration; Ii change blasting parameter carries out the blasting operation second time and carries out on-site data gathering acceptor, obtains the monitoring attribute of this separate explosion; Iii continues the change blasting parameter acceptor is carried out blasting operation and carries out on-site data gathering, and the explosion number of times is 15～20 times, obtains the monitoring attribute of each explosion;

B. with the blasting parameter of each separate explosion operation among the step a and corresponding monitoring attribute input engineering information collecting module, set up engineering data base;

C. the importance degree analysis module is handled the information of engineering information collecting module collection, and the output blasting parameter is to the importance degree of monitoring attribute;

D. the forecasting model module is handled the information of engineering information collecting module collection, and output blasting vibration intensity or dominant frequency or duration;

E. the result who obtains with step c or d is a foundation, controls the blasting parameter of next blasting operation.

Described step c may further comprise the steps:

(1) obtaining of training dataset: 1-1 accesses job number from engineering data base; 1-2 selects big gun time numbering; 1-3 need to select the blasting parameter of analysis; 1-4 imports early warning value;

(2) pre-service of training data: 2-1 qualitative parameter assignment, as explosive type (ammonium nitrate fuel oil mixture and emulsion [explosive assignment respectively are 0 and 1), the aperture (aperture that experiment sample is occurred according to 76mm, 76/89mm, 89mm, 89/115mm, 115mm, 150mm, 230mm order successively assignment be 0,1,2 ...), thrown direction (according to forward, positive side, side direction, dorsal part to, dorsad the thrown direction that occurs in to experiment sample of order successively assignment 0,1,2 ...); 2-2 makes normalized respectively to all blasting parameters and monitoring attribute;

(3) training is screened device based on the rough set of particulate optimization:

3-1 breakpoint number h chooses in [2,4], and initial value is 2;

The initialization of 3-2 population, total number of particles order are W, 150≤W≤200, and the quantitative parameter of participating in training in the blasting parameter is for treating discrete attribute number T; The particle position attribute is the vector of h*T, the position of each numeral breakpoint in the vector; Breakpoint location information with each particle is cut apart experiment sample, forms the form of decision table, calculates the dependency degree of blasting parameter to the monitoring attribute, as the adaptive value of each particle;

3-3 chooses preceding L from high to low successively according to dependency degree from all particles individual as elite's particle, with i elite's particle P _iAs i Provisional Center, the initial value of i equals 1;

Whether 3-4 judges i smaller or equal to L, and "No" is execution in step 3-5 then, and "Yes" is execution in step 3-7 then;

3-5h adds 1;

Whether 3-6 judges h smaller or equal to 4, and "Yes" is execution in step 3-1 then, and "No" is then restarted;

3-7 chooses from all particles and elite's particle P according to formula (1) _iGrain M particle before from the near to the remote, set up interim small sample population:

Hypothesis matrix X _iAnd X _jRepresent i particle and j particle position information matrix respectively, utilize Euclidean distance definition X _iAnd X _jBetween the grain distance:

D(i，j)＝||X _i-X _j|| (1)

M particle position information matrix adjusted according to formula (2):

$x_{\mathrm{id}}^{t+1}=w\·x_{\mathrm{id}}^t+c_1{r}_1(P_{\mathrm{id}}-x_{\mathrm{id}}^t)+c_2{r}_2(P_{\mathrm{gd}}-x_{\mathrm{id}}^t)---\left(2\right)$

In the formula: i=1, L, n; D=1, L, D; T is an iterations; W is an inertia weight, when k＞0, and w=0.40, otherwise w=0.95; c ₁, c ₂Be the study factor; r ₁, r ₂Be the random number between (0,1);

For i particle, the graded value k of the corresponding adaptive value of double iteration _iFor:

$k_i=\frac{Y_i^{t+1}-Y_i^t}{D_i(t+1,t)}---\left(3\right)$

In the formula: D _i(t+1 t) is the grain distance of the corresponding t+1 of i particle and the t time iteration,

With

Adaptive value when being the corresponding t+1 of i particle and the t time iteration;

If 3-8 finds satisfied particle, training finishes, otherwise execution in step 3-9;

3-9h adds 1, re-executes step 3-4;

(4) the output blasting parameter is to the importance degree of monitoring attribute.

Described steps d may further comprise the steps:

(5) obtaining of training dataset: 5-1 accesses job number from engineering data base; 5-2 selects big gun time numbering, the individual effectively big gun of a total N, N 〉=10;

(6) pre-service of training dataset: the data sample is done normalization or intervalization processing, and blasting parameter and monitoring attribute be corresponding matrix { X//x respectively ₁, x ₂, L, x _NAnd { Y/y ₁, y ₂, L, y _N}

(7) training of forecasting model;

The initialization of 7-1 population, total number of particles order are W, and interim population total number is M, and the attribute of particle position and speed is respectively penalty factor C and spread factor σ ², match or forecast precision are as the adaptive value of each particle;

7-2 provides the initial value of forecasting model parameter:

Insensitive coefficient initial value ε=0.8*10 ^-4The spread factor initial value

Penalty factor scope: [C ₁, C ₂], wherein

C ₂=2* (max (Y)-min (Y)); Initial value C ₃=min (C ₁, C ₂);

7-3 chooses preceding L from high to low successively according to dependency degree from all particles individual as elite's particle, with i elite's particle P _iAs i Provisional Center, the initial value of i equals 1

Whether 7-4 judges i smaller or equal to L, and "No" is execution in step 7-5 then, and "Yes" is execution in step 7-6 then;

7-5 gives insensitive coefficient ε assignment again, comes back to step 7-1;

7-6 extracts three samples as the self-checking sample in the training sample internal random, and the residue sample is as training sample;

7-7 chooses from all particles and elite's particle P according to formula (1) _iGrain M particle before from the near to the remote, set up interim small sample population; M particle position information matrix adjusted according to formula (2)～(3);

7-8 is end condition simultaneously: the relative error f of test samples ₁Variance and f with training sample ₂If reach end condition, execution in step 8, otherwise execution in step 7-9;

7-9i adds 1, re-executes step 7-4;

(8) check of forecasting model: new test samples is input to model, if predicted value and measured value differ at allowed band f ₃In, execution in step (9), otherwise come back to step 7-2;

(9) input blasting parameter forecasts the monitoring attribute.

Beneficial effect: blasting vibration forecast of the present invention and control system and control method thereof, both can carry out comprehensively the blasting vibration parameter, high-precision forecast, also can screen numerous influence factors, its examination scope from traditional quantitative factor (as the section dose, horizontal range, total dose, bench height etc.) analyze, extend to qualitative factor (as thrown direction, the aperture, explosive type etc.) carry out the significance level analysis, comprehensive height, can be efficient, convenient, stable output control information, and relevant form is provided, guarantee the security of engineering explosion work progress.

Description of drawings

Fig. 1 is the workflow diagram of blasting vibration forecast of the present invention and control system;

Fig. 2 is the examination process flow diagram that device is screened in training among Fig. 1 based on the rough set of particulate optimization;

Fig. 3 is the process flow diagram of the training and the check of forecasting model among Fig. 1.

Embodiment

Below in conjunction with accompanying drawing the present invention is done further explanation.

As shown in Figure 1, blasting vibration forecast of the present invention and control system comprise with lower module: the engineering information collecting module, be used to gather explosion information, and explosion information comprises blasting parameter and monitoring attribute; The importance degree analysis module is used for the information of engineering information collecting module collection is handled, and the output blasting parameter is to the importance degree of monitoring attribute; The forecasting model module is used for the information of engineering information collecting module collection is handled, and output blasting vibration intensity or dominant frequency or duration.

Be example with the massif excavation below, describe the control method of blasting vibration forecast of the present invention and control system in detail.

Embodiment 1:

A. obtaining explosion information: i implements blasting operation for the first time and carries out on-site data gathering massif according to the blasting parameter of design in advance, blasting parameter is thrown direction, section dose, total dose, horizontal range, bench height, difference of elevation and aperture, obtain the monitoring attribute of this separate explosion, the monitoring attribute is the vibration acceleration peak value; Ii change blasting parameter carries out the blasting operation second time and carries out on-site data gathering massif, obtains the monitoring attribute of this separate explosion; Iii continues the change blasting parameter massif is carried out blasting operation and carries out on-site data gathering, and effective big gun of explosion is 18 times, and big gun is numbered 001～018, and the monitoring attribute that obtains each explosion is the vibration acceleration peak value, is recorded in the table 1.

Table 1

The blasting vibration forecast of present embodiment and the inner parameter of control system are set: W=200, M=50, L=6; Two advanced warning grades are set, are respectively 0.018g and 0.025g (wherein controlling threshold value is 0.03g).

Following step comprises as shown in Figure 1:

B. with the blasting parameter of each separate explosion operation among the step a and corresponding monitoring attribute input engineering information collecting module, set up engineering data base; C. the importance degree analysis module is handled the information of engineering information collecting module collection, and the output blasting parameter is to the importance degree of monitoring attribute; D. the forecasting model module is handled the information of engineering information collecting module collection, and output blasting vibration intensity or dominant frequency or duration; E. the result who obtains with step c or d is a foundation, controls the blasting parameter of next blasting operation.

Specifically, step c may further comprise the steps: the obtaining of (1) training dataset: 1-1 accesses job number from engineering data base; 1-2 selects big gun time numbering; 1-3 need to select the blasting parameter of analysis; 1-4 imports early warning value; (2) pre-service of training data: 2-1 qualitative parameter assignment, as thrown direction (according to forward, positive side, side direction, dorsal part to, dorsad the thrown direction that occurs in to experiment sample of order successively assignment 0,1,2 ...); 2-2 makes normalized respectively to all blasting parameters and monitoring attribute; (3) training is screened device based on the rough set of particulate optimization: as shown in Figure 2,3-1 breakpoint number h chooses in [2,4], and initial value is 2; The initialization of 3-2 population, the total number of particles order is W, the quantitative parameter of participating in training in the blasting parameter is for treating discrete attribute number T; The particle position attribute is the vector of h*T, the position of each numeral breakpoint in the vector; Breakpoint location information with each particle is cut apart experiment sample, forms the form of decision table, calculates the dependency degree of blasting parameter to the monitoring attribute, as the adaptive value of each particle; 3-3 chooses preceding L from high to low successively according to dependency degree from all particles individual as elite's particle, with i elite's particle P _iAs i Provisional Center, the initial value of i equals 1; Whether 3-4 judges i smaller or equal to L, and "No" is execution in step 3-5 then, and "Yes" is execution in step 3-7 then; 3-5h adds 1; Whether 3-6 judges h smaller or equal to 4, and "Yes" is execution in step 3-1 then, and "No" is then restarted; 3-7 chooses from all particles and elite's particle P according to formula (1) _iGrain M particle before from the near to the remote, set up interim small sample population:

Hypothesis matrix X _iAnd X _jRepresent i particle and j particle position information matrix respectively, utilize Euclidean distance definition X _iAnd X _jBetween the grain distance:

D(i，j)＝||X _i-X _j|| (1)

M particle position information matrix adjusted according to formula (2):

$x_{\mathrm{id}}^{t+1}=w\·x_{\mathrm{id}}^t+c_1{r}_1(P_{\mathrm{id}}-x_{\mathrm{id}}^t)+c_2{r}_2(P_{\mathrm{gd}}-x_{\mathrm{id}}^t)---\left(2\right)$

In the formula: i=1, L, n; D=1, L, D; T is an iterations; W is an inertia weight, when k＞0, and w=0.40, otherwise w=0.95; c ₁, c ₂Be the study factor; r ₁, r ₂Be the random number between (0,1);

For i particle, the graded value k of the corresponding adaptive value of double iteration _iFor:

$k_i=\frac{Y_i^{t+1}-Y_i^t}{D_i(t+1,t)}---\left(3\right)$

In the formula: D _i(t+1 t) is the grain distance of the corresponding t+1 of i particle and the t time iteration,

With

Adaptive value when being the corresponding t+1 of i particle and the t time iteration;

If 3-8 finds satisfied particle, training finishes, otherwise execution in step 3-9; 3-9h adds 1, re-executes step 3-4; (4) the output blasting parameter is to the importance degree of monitoring attribute.

In the present embodiment, the blasting parameter of output is as follows to the importance degree of monitoring attribute: thrown direction-0.100, aperture-0.107, section dose-0.098, total dose-0.174, bench height-0.131, horizontal range-0.184, difference of elevation-0.207.The pairing numerical value of each blasting parameter is big more, illustrates that its numerical value is high more.

Embodiment 2:

At first, the blasting vibration forecast of present embodiment and the inner parameter of control system are set: W=200, M=50, L=6, f ₁=10%, f ₂=0.00001, f ₃=0.00001.

The input training sample: 18 big guns in the table 1, blasting parameter and monitoring attribute are with embodiment 1, and other step is as follows:

(5) obtaining of training data module: 5-1 accesses job number from engineering data base; 5-2 selects big gun time numbering, a total N big gun; (6) pre-service of training dataset: the data sample is done normalization or intervalization processing, and blasting parameter and monitoring attribute be corresponding matrix { X/x respectively ₁, x ₂, L, x _NAnd { Y/y ₁, y ₂, L, y _N}

(7) training of forecasting model, as shown in Figure 3:

The initialization of 7-1 population, total number of particles order are W, and interim population total number is M, and the attribute of particle position and speed is respectively penalty factor C and spread factor σ ², match or forecast precision are as the adaptive value of each particle;

7-2 provides the initial value of forecasting model parameter:

Insensitive coefficient initial value ε=0.8*10 ^-4The spread factor initial value

Penalty factor scope: [C ₁, C ₂], wherein

C ₂=2* (max (Y)-min (Y)); Initial value C ₃=min (C ₁, C ₂);

7-3 chooses preceding L from high to low successively according to dependency degree from all particles individual as elite's particle, with i elite's particle P _iAs i Provisional Center, the initial value of i equals 1; Whether 7-4 judges i smaller or equal to L, and "No" is execution in step 7-5 then, and "Yes" is execution in step 7-6 then; 7-5 gives insensitive coefficient ε assignment again, comes back to step 7-1; 7-6 extracts three samples as the self-checking sample in the training sample internal random, and the residue sample is as training sample; 7-7 chooses from all particles and elite's particle P according to formula (1) _iGrain M particle before from the near to the remote, set up interim small sample population; M particle position information matrix adjusted according to formula (2)～(3); 7-8 is end condition simultaneously: the relative error f of test samples ₁Variance and f with training sample ₂If reach end condition, execution in step 8, otherwise execution in step 7-9; 7-9i adds 1, re-executes step 7-4;

The training result of forecasting model: spread factor=0.19966, insensitive coefficient=0.00008, penalty coefficient=0.034315, relative error=5.6768%, variance and=0.000002.

(8) check of forecasting model: new test samples (" 019～026 " ten big guns in the table 2) is input to model, if predicted value and measured value differ at allowed band f ₃In, execution in step (9), otherwise come back to step 7-2;

Table 2

Forecasting model assay: relative error=3.365%, variance and=0.000000408.

(9) input blasting parameter, forecast the monitoring attribute:

I forecasts the sample input for the first time: thrown direction=forward, section dose=116kg, total dose=3128kg, horizontal range=666.8m, bench height=10.5m, difference of elevation=120.9m, aperture=89mm.

The output predicted value of native system: 0.0140.Measured value: 0.0135.

II forecasts the sample input for the second time: thrown direction=positive side, section dose=147kg, total dose=4143kg, horizontal range=677.0m, bench height=12.0m, difference of elevation=47.5m, aperture=89mm.

The output predicted value of device: 0.0164.Measured value: 0.0158.

III forecasts the sample input for the third time: thrown direction=forward, section dose=221kg, total dose=3689kg, horizontal range=731.2m, bench height=11.8m, difference of elevation=48.2m, aperture=115mm.

The output predicted value of device: 0.0162.Measured value: 0.0169.

As can be seen from the above-described embodiment: blasting vibration forecast of the present invention and control system and control method thereof, both can carry out comprehensively the blasting vibration parameter, high-precision forecast, also can screen numerous influence factors, its examination scope from traditional quantitative factor (as the section dose, horizontal range, total dose, bench height etc.) analyze, extend to qualitative factor (as thrown direction, the aperture, explosive type etc.) carry out the significance level analysis, comprehensive height, can be efficient, convenient, stable output control information, and relevant form is provided, guarantee the security of engineering explosion work progress.

Claims (4)

1. a blasting vibration forecasts and control system, it is characterized in that comprising with lower module:

The engineering information collecting module: be used to gather explosion information, explosion information comprises blasting parameter and monitoring attribute;

Importance degree analysis module: be used for the information of engineering information collecting module collection is handled, and the output blasting parameter is to the importance degree of monitoring attribute;

The forecasting model module: be used for the information of engineering information collecting module collection is handled, and output blasting vibration intensity or dominant frequency or duration.

2. the described blasting vibration of claim 1 forecasts and the control method of control system, it is characterized in that: may further comprise the steps:

A. obtain explosion information:

I implements blasting operation for the first time and carries out on-site data gathering acceptor according to the blasting parameter of design in advance, obtains the monitoring attribute of this separate explosion;

Ii change blasting parameter carries out the blasting operation second time and carries out on-site data gathering acceptor, obtains the monitoring attribute of this separate explosion;

Iii continues the change blasting parameter acceptor is carried out blasting operation and carries out on-site data gathering, and the explosion number of times is 15～20 times, obtains the monitoring attribute of each explosion;

B. with the blasting parameter of each separate explosion operation among the step a and corresponding monitoring attribute input engineering information collecting module, set up engineering data base;

C. the importance degree analysis module is handled the information of engineering information collecting module collection, and the output blasting parameter is to the importance degree of monitoring attribute;

D. the forecasting model module is handled the information of engineering information collecting module collection, and output blasting vibration intensity or dominant frequency or duration;

E. the result who obtains with step c or d is a foundation, controls the blasting parameter of next blasting operation.

3. the control method of blasting vibration forecast according to claim 2 and control system, it is characterized in that: described step c may further comprise the steps:

(1) obtaining of training dataset:

1-1 accesses job number from engineering data base;

1-2 selects big gun time numbering;

1-3 need to select the blasting parameter of analysis;

1-4 imports early warning value;

(2) pre-service of training data:

2-1 qualitative parameter assignment;

2-2 makes normalized respectively to all blasting parameters and monitoring attribute;

(3) training is screened device based on the rough set of particulate optimization:

3-1 breakpoint number h chooses in [2,4], and initial value is 2;

The initialization of 3-2 population, total number of particles order are W, 150≤W≤200;

3-3 chooses preceding L from high to low successively according to dependency degree from all particles individual as elite's particle, with i elite's particle P _iAs i Provisional Center, the initial value of i equals 1;

Whether 3-4 judges i smaller or equal to L, and "No" is execution in step 3-5 then, and "Yes" is execution in step 3-7 then;

3-5h adds 1;

Whether 3-6 judges h smaller or equal to 4, and "Yes" is execution in step 3-1 then, and "No" is then restarted;

3-7 chooses from all particles and elite's particle P according to formula (1) _iGrain M particle before from the near to the remote, set up interim small sample population:

Hypothesis matrix X _iAnd X _jRepresent i particle and j particle position information matrix respectively, utilize Euclidean distance definition X _iAnd X _jBetween the grain distance:

D(i，j)＝||X _i-X _j|| (1)

M particle position information matrix adjusted according to formula (2):

$x_{\mathrm{id}}^{t+1}=w\·x_{\mathrm{id}}^t+c_1{r}_1(P_{\mathrm{id}}-x_{\mathrm{id}}^t)+c_2{r}_2(P_{\mathrm{gd}}-x_{\mathrm{id}}^t)---\left(2\right)$

In the formula: i=1, L, n; D=1, L, D; T is an iterations; W is an inertia weight, when k＞0, and w=0.40, otherwise w=0.95; c ₁, c ₂Be the study factor; r ₁, r ₂Be the random number between (0,1);

For i particle, the graded value k of the corresponding adaptive value of double iteration _iFor:

$k_i=\frac{Y_i^{t+1}-Y_i^t}{D_i(t+1,t)}---\left(3\right)$

In the formula: D _i(t+1 t) is the grain distance of the corresponding t+1 of i particle and the t time iteration,

With

Adaptive value when being the corresponding t+1 of i particle and the t time iteration;

If 3-8 finds satisfied particle, training finishes, otherwise execution in step 3-9;

3-9h adds 1, re-executes step 3-4;

(4) the output blasting parameter is to the importance degree of monitoring attribute.

4. 2 described blasting vibrations forecast and the control method of control system that it is characterized in that: described steps d may further comprise the steps as requested:

(5) obtaining of training dataset:

5-1 accesses job number from engineering data base;

5-2 selects big gun time numbering, the individual effectively big gun of a total N, N 〉=10;

(6) pre-service of training dataset: the data sample is done normalization or intervalization processing, and blasting parameter and monitoring attribute be corresponding matrix { X/x respectively ₁, x ₂, L, x _NAnd { Y/y ₁, y ₂, L, y _N}

(7) training of forecasting model;

The initialization of 7-1 population: the total number of particles order is W, and interim population total number is M, and the attribute of particle position and speed is respectively penalty factor C and spread factor σ ², match or forecast precision are as the adaptive value of each particle;

7-2 provides the initial value of forecasting model parameter:

Insensitive coefficient initial value ε=0.8*10 ^-4The spread factor initial value

Penalty factor scope: [C ₁, C ₂], wherein

C ₂=2* (max (Y)-min (Y)); Initial value C ₃=min (C ₁, C ₂);

7-3 chooses preceding L from high to low successively according to dependency degree from all particles individual as elite's particle, with i elite's particle P _iAs i Provisional Center, the initial value of i equals 1

Whether 7-4 judges i smaller or equal to L, and "No" is execution in step 7-5 then, and "Yes" is execution in step 7-6 then;

7-5 gives insensitive coefficient ε assignment again, comes back to step 7-1;

7-6 extracts three samples as the self-checking sample in the training sample internal random, and the residue sample is as training sample;

7-7 chooses from all particles and elite's particle P according to formula (1) _iGrain M particle before from the near to the remote, set up interim small sample population; M particle position information matrix adjusted according to formula (2)～(3);

7-8 is end condition simultaneously: the relative error f of test samples ₁Variance and f with training sample ₂If reach end condition, execution in step 8, otherwise execution in step 7-9;

7-9i adds 1, re-executes step 7-4;

(8) check of forecasting model: new test samples is input to model, if predicted value and measured value differ at allowed band f ₃In, execution in step (9), otherwise come back to step 7-2;

(9) input blasting parameter forecasts the monitoring attribute.

Images