CN119474644A

CN119474644A - A small sample regression prediction method for tunnel blasting effect

Info

Publication number: CN119474644A
Application number: CN202411512790.1A
Authority: CN
Inventors: 陈培帅; 刘哲; 唐启; 杨钊; 姬付全; 曹昂; 杨林; 唐湘隆; 彭松林; 于锦; 周伟; 袁青
Original assignee: Rail Transit Branch Of China Communications Construction Co ltd; CCCC Second Harbor Engineering Co
Current assignee: Rail Transit Branch Of China Communications Construction Co ltd; CCCC Second Harbor Engineering Co
Priority date: 2024-10-28
Filing date: 2024-10-28
Publication date: 2025-02-18

Abstract

The application provides a small sample regression prediction method for tunnel blasting effect, which comprises the following steps of obtaining feature data before blasting and blasting effect data after blasting, carrying out data normalization processing on the data, analyzing and selecting input features of a regression prediction model by utilizing gray correlation degree, establishing a generated countermeasure network model, carrying out expansion and expansion on a training data set, establishing a blasting effect multi-core Gaussian process regression prediction model, searching super parameters of the prediction model based on a sparrow algorithm integrated and optimized by multiple strategies, and predicting a new blasting effect based on the prediction model under the optimal parameters. The application effectively solves the problems of difficult collection of blasting engineering data, more samples with excellent effect and less samples with poor effect of the data while the number of the samples is small. The prediction precision and the prediction efficiency of the blasting effect are improved, the generalization capability of the blasting effect prediction is improved to the greatest extent, and the construction safety and the efficiency of engineering projects are ensured.

Description

Tunnel blasting effect small sample regression prediction method

Technical Field

The invention belongs to the technical field of tunnel blasting, and particularly relates to a small sample regression prediction method for tunnel blasting effect.

Background

The drilling and blasting method is widely applied to mountain tunnel engineering by virtue of the advantages of high efficiency, economy, flexibility, wide application range and the like. However, due to different blasting designs, tunnel structures and surrounding rock properties, the effects of explosion impact on the walls of blast holes are different, and bad blasting effects such as tunnel super-underexcavation, insufficient settlement of step blasting piles, large blocks and the like can be generated, so that the construction safety and the construction efficiency are influenced. Therefore, the method has the advantages that the blasting effect is predicted and judged before blasting operation, and the blasting design is timely and reasonably adjusted, so that the method has important engineering value.

Aiming at the problem of blasting effect prediction, most methods only use characteristic data before blasting and a deep learning algorithm to predict the characteristic parameters of the blasting effect. However, the blasting engineering data is difficult to collect, the number of samples is small, and meanwhile, the data has the unbalanced characteristics of more excellent effect samples and less poor effect samples. Deep learning algorithms often require a large amount of data to train and adjust weight parameters and paranoid parameters in the neural network so that the model can extract the characteristics and modes of the data. This is in contradiction with the data characteristics of the blasting engineering, which makes it difficult to obtain an optimal prediction model. Therefore, how to fully consider the characteristics of small samples of the blasting process characteristic data and realize accurate prediction of the blasting effect of the tunnel is a technical problem to be solved in the field.

Disclosure of Invention

The invention aims to solve the technical problems in the background and provides a small sample regression prediction method for tunnel blasting effect, which comprises the following steps:

s1, collecting characteristic data before blasting and blasting effect data after blasting;

S2, data normalization is carried out, and gray correlation degree analysis is utilized to select input features of a regression prediction model;

S3, establishing and generating an countermeasure network model, and carrying out augmentation and expansion on a training data set;

s4, establishing a multi-core Gaussian process regression prediction model of the blasting effect;

S5, searching a prediction model super-parameter based on a sparrow algorithm of multi-strategy integrated optimization;

S6, predicting the blasting effect.

In a preferred embodiment, in step S1, the collected characteristic data before blasting is characteristic data related to blasting effect, including blasting parameters, surrounding rock parameters, tunnel design, etc., which may include, but are not limited to, surrounding rock category, rock compressive strength, rock integrity coefficient, cross-sectional area, drilling depth, external insertion angle, blasting unit consumption, grooving angle, and peripheral hole spacing, which are expressed by the following matrix:

Wherein n represents the number of collected pre-blasting feature data pieces, and m represents the dimension of the collected pre-blasting feature;

The collected blasting effect data is characteristic data capable of reflecting the blasting effect quality, and specifically comprises, but is not limited to, super-underexcavated quantity, footage, half-porosity and blockiness, and is represented by the following matrix form:

wherein n represents the number of the collected blasting effect characteristic data, and p represents the dimension of the collected blasting effect characteristic;

the data normalization method in the step S2 is as follows:

Wherein x _norm is the normalized data value, x _norm E [0,1], x is the original data value, and x _max,x_min is the maximum value and the minimum value of the parameter sequence respectively.

In a preferred embodiment, the specific process of selecting the model input feature in step S2 includes:

first, the range is calculated by comparing the characteristic column with the reference characteristic column:

Wherein a and b are minimum and maximum differences of two poles, y _kp is the kth line data of the p-th column reference feature, and x _im is the kth line data of the m-th column comparison feature;

calculating the association degree:

Wherein r _mp is the association degree between the comparison feature of the mth column and the reference feature of the p column, namely the association degree between the feature parameter before blasting and the blasting effect. ρ is the resolution factor;

In the preferred scheme, the step S2 also comprises the steps of arranging the association degree of the features before blasting according to the sequence from big to small, setting feature selection dimensions, and taking the comparison feature with the front association degree as input feature data after association degree screening according to the set feature selection dimensions Ni;

and dividing the screened input characteristic data and blasting effect data into a training data set and a testing data set.

In a preferred scheme, the generating countermeasure network model established in the step S3, the generating network comprises a one-dimensional convolution layer, a full connection layer and an activation function layer;

The generator network inputs noise vectors which are in the dimension of (Ni+No) and accord with Gaussian distribution, and outputs generated samples in the dimension of (Ni+No), the discriminator network comprises two full-connection layers and a Sigmoid activation function layer, and the output of the network is a scalar of 0 to 1, which represents the probability estimation of the discriminator network on the input data vector being real data, namely represents the distribution similarity of the input data and the real data;

The loss of the training process discriminator network is formed by the sum of the real blasting data and the loss of the generated sample, the loss of the generated network is calculated by judging whether the data generated by the generator is real data or not through the discriminator, and the loss function is a binary cross entropy loss function:

BCELoss=-(ylog(p)+(1-y)log(1-p)),

and in the formula, y is a real label, the real blasting data corresponding label is 1, and the generated sample corresponding label is 0;p which is the output probability estimation of the identifier network Sigmoid activation function layer.

In the preferred scheme, in the process of the game training of the GAN, the generation network gradually improves the capability of generating the realistic data, and the discriminator improves the capability of discriminating whether the data is a real blasting sample. After the maximum training times are reached, the generated model is stored. And the original training data set is amplified by using the generated sample of the generated model, so that an amplified data set containing the generated sample and the real sample is obtained.

In a preferred scheme, the blasting effect gaussian process regression prediction model in step S4 is established by:

Gaussian process regression is a regression method of a Bayesian inferred nonparametric type, and the regression model expression is:

y=f(x)+ε;

Wherein x represents the input pre-blasting characteristic, y represents the actual observed blasting effect, epsilon is a noise variable, epsilon obeys to F (x) is the predicted blasting effect, assuming it follows the gaussian process f (x) to GP (m (x), k (x, x ^*)), the mean function of which is determined by the training data, and the covariance function is determined by the kernel function:

Wherein x and x ^* represent any two different pieces of pre-blasting characteristic data, m is a mean function, E is a mathematical expectation, and k is a covariance function;

The prior distribution of the actual observed blasting effect y can thus be obtained as:

Let X be the pre-burst feature data in the training set, where k=k (X, X) =k _ij be a positive definite covariance matrix of size n×n, K _ij＝k(x_i,x_j) is used to measure the correlation between X _i and X _j, k=k (X ^*,X^*);K_*＝K(X^*, X) is an n×1 order covariance matrix between the test point and the training set, and I _n is an n order identity matrix. The posterior distribution of predicted values is thus obtained:

in the formula, Mean of predicted values is represented, cov (f _*) is variance.

In a preferred scheme, further, the blasting effect multi-core gaussian process regression prediction model in step S4, and the fusion kernel function construction process includes:

(1) The RBF kernel is used for extracting data local characteristics, and the covariance function is as follows:

(2) Matern kernel, which is a popularization of RBF kernel, is used for extracting local correlation characteristics, and the covariance function is:

(3) A linear kernel for describing the linear correlation between the data, the covariance function being:

(4) A polynomial core for describing a nonlinear relationship between data, the covariance function being:

delta ₁～δ₉ is a nuclear function super parameter to be set;

the multiple kernel functions constitute a fused kernel function as follows:

k_M(x_i,x_j)＝δ₁₀·k_RBF(x_i,x_j)+δ₁₁·k_Matern(x_i,x_j)+δ₁₂·k_l(x_i,x_j)+δ₁₃·k_P(x_i,x_j),

Wherein delta ₁₀～δ₁₃ is covariance weight of different kernel functions;

finally, the augmented data set comprising the generated samples and the real samples is divided into a training set, a verification set and a test set, wherein the training set is used for training a multi-core Gaussian process regression prediction model, the verification set is used for adjusting super parameters of the optimal model and selecting the optimal model, the test set is used for evaluating the performance and generalization capability of the model, and the model performance evaluation indexes are as follows:

Wherein MAE, RMSE, R is the mean absolute error, root mean square error, and decision coefficient, y _i、y′_i is the true sample value and the predicted value, The average value of the samples is shown.

In a preferred scheme, the optimized sparrow algorithm is integrated by multiple strategies in the step S5, wherein the optimized strategies comprise a chaotic mapping mechanism, dynamic self-adaptive weights, an improved scout position updating mode, a fused cauchy variation and random reverse learning strategy, and the specific search model hyper-parameter process comprises the following steps:

S501, setting super-parameters delta ₁～δ₁₃ in a fusion kernel function as search solutions of a sparrow algorithm, setting polynomial kernel super-parameters delta ₉ as the highest times of polynomials, setting the value range as [1,2,3,4,5,6], setting the value range of other super-parameters as [0,1], and setting the number of discoverers, the occupation ratio of the scouts and a safety threshold;

s502, initializing the sparrow population position through an ICMIC chaotic mapping model:

Wherein Z is a generated chaotic sequence, x _i is an ith individual of the sparrow population generated according to the chaotic mapping sequence, x _lb、x_ub is the lower limit and the upper limit of the range of each individual in the search dimension, and the polynomial super-parameter delta ₉ still needs to be rounded after initialization.

S503, calculating the fitness of individuals in the population, sorting and determining the optimal individuals and the worst individuals according to the fitness, and calculating the fitness value of each sparrow in the population according to the following formula:

f_FIT=RMSE+MAE+(1-R2);

s504, updating the position of the finder by using a finder position updating mode with self-adaptive weight:

Wherein ω is the dynamic self-adaptive weight of the position change of the finder, k is the control weight value range of [0,2]; T _max represents the maximum iteration number of the iteration; representing the position information of the ith sparrow when the j-dimensional iteration number is t; The method comprises the steps of respectively representing optimal fitness and worst fitness values of population individuals when iteration times are t, wherein R epsilon [0,1] represents an early warning value, ST epsilon [0.5,1] represents a safety threshold, sparrows perform global search and foraging when R < ST, and perform random walk with regular distribution when R > ST;

S505, updating the position of the joiner:

in the formula, The position with the optimal fitness value in the current discoverer; The method comprises the steps of obtaining a current global fitness worst position, wherein A ⁺＝A^T(AA^T)^-1, A represents a single column vector with the same dimension as a sparrow individual, an internal element is formed by a 1-1 random set, n is the population size, when i is less than or equal to n/2, a user actively follows a finder to move towards a better foraging position, and when i is more than n/2, the user gets rid of the current worse foraging position.

S506, randomly selecting m multiplied by n sparrows as the scouts, wherein m is the ratio of the scouts in the population, and undertaking an early warning task, wherein the position of the scouts is out of control easily due to the fact that the traditional scout updates a formula, and deviates from an optimal position;

the scout location is updated using the following optimization:

beta is a random number obeying normal distribution with mean value of 0 and variance of 1;

the improved position updating formula shows that if the sparrow individual is positioned at the current optimal position, the sparrow individual flies to any random position between the optimal position and the worst position, otherwise, the sparrow individual can fly to any random position between the sparrow individual and the optimal position;

S507, introducing random factors, utilizing a random reverse learning avoidance algorithm to fall into a local optimal solution and improving population diversity:

in the formula, The method is a reverse solution of the optimal solution under the t-th iteration, wherein x _ub、x_lb is an upper and lower bound, and r is a random value from 0 to 1; Representing the individual position of the optimal solution individual after random reverse learning under t iterations; b ₁ is an information exchange control parameter;

s508, simultaneously introducing a Cauchy variation strategy to improve the tendency of the algorithm to fall into a local optimal solution due to the competition of the joiner and the discoverer for food:

wherein cauchy (0, 1) is a standard Cauchy distribution;

S509, selecting to update the optimal individual location by a dynamic selection policy:

When rand (1, e) > P _s, selecting the position update based on the random reverse learning strategy, otherwise, selecting the position update based on the Cauchy variation disturbance strategy;

S510, judging whether the maximum iteration times are reached, continuing the next step when the maximum iteration times are reached, otherwise repeating S53;

S511, outputting the optimal individual position, namely the optimal super-parameter combination of the blasting effect multi-core Gaussian process regression prediction model.

In the preferred scheme, in step S6, the blasting effect prediction is to use the multi-core gaussian process regression prediction model established by the optimal super-parameter combination output in step S5, and obtain the blasting effect by inputting new data.

The method has the advantages of high accuracy, simple structure of a prediction model, strong model generalization capability and the like aiming at the prediction of the small sample of the blasting effect, and is specific to the following steps:

1) Aiming at the characteristics of more normal evaluation samples, less abnormal evaluation samples, less data volume and the like presented by the blasting data, the generated countermeasure network is used for supplementing the data set, so that the training difficulty of the prediction model is reduced;

2) Aiming at the problems of high dimensionality of the small sample and insufficient nonlinear prediction precision, multi-core Gaussian process regression is used, so that the generalization capability and the prediction precision of the prediction model are improved;

3) Aiming at the problem of mixed nuclear super-parameter setting in the multi-nuclear Gaussian process regression prediction model, the sparrow algorithm optimized by multi-strategy integration is used for searching the optimal prediction super-parameter, so that the prediction precision of the prediction model is further enhanced.

Drawings

Fig. 1 is a flow chart overall.

FIG. 2 is a graph of gray correlation analysis.

Fig. 3 generates an augmented data architecture diagram for an antagonism network acquisition.

Fig. 4 is a diagram of an iterative process of optimal fitness.

FIG. 5 is a graph comparing predicted effects.

Detailed Description

As shown in fig. 1 to 5, a regression prediction method for a small sample of a tunnel blasting effect is implemented by the following steps:

S1.1, selecting characteristic data related to blasting effect before blasting according to expert experience, wherein the characteristic data comprise blasting parameters, surrounding rock parameters, tunnel design and the like, and the characteristic data can comprise, but are not limited to, surrounding rock types, rock compressive strength, rock integrity coefficients, cross-sectional areas, drilling depths, external insertion angles, blasting unit consumption, grooving angles, peripheral hole spacing and the like. And is represented by the following matrix form:

where n represents the number of collected pre-blast feature data pieces and m represents the dimension of the collected pre-blast feature.

S1.2, collecting characteristic data which can reflect the quality of the blasting effect after blasting, wherein the characteristic data can specifically comprise but not be limited to super-underexcavated quantity, footage, half-porosity, blockiness and the like. And is represented by the following matrix form:

Where n represents the number of collected blast effect feature data and p represents the dimension of the collected blast effect feature.

S2.1 normalizes the collected data:

S2.2, taking the characteristics before blasting as a comparison characteristic column, taking the effect data after blasting as a reference characteristic column, and calculating the association degree of each characteristic before blasting and the blasting effect through gray association degree analysis. Comparing the characteristic column with the reference characteristic column to calculate the range:

Where a and b are the minimum and maximum differences of the poles, y _kp is the kth line data of the p-th column reference feature, and x _im is the kth line data of the m-th column comparison feature.

S2.3, calculating the association degree:

Wherein r _mp is the association degree between the comparison feature of the mth column and the reference feature of the p column, namely the association degree between the feature parameter before blasting and the blasting effect. ρ is a resolution factor, typically 0.5.

S2.4, the association degrees of the features before blasting are arranged in a sequence from big to small (see figure 2), feature selection dimensions are set, and comparison features with the previous association degrees are taken as input feature data after association degree screening according to the set feature selection dimensions Ni. The screened input characteristic data (dimension Ni) and blasting effect data (dimension No) are divided into a training data set and a test data set.

S3.1, establishing and generating an countermeasure network model. The generating network is composed of a one-dimensional convolution layer, a full connection layer and an activation function layer. The generator network inputs noise vectors which are in the dimension of (Ni+No) and accord with Gaussian distribution, and outputs generated samples in the dimension of (Ni+No), and the discriminator network comprises two full-connection layers and a Sigmoid activation function layer.

S3.2, training a discriminator network and a generation network. The loss of the training process arbiter network is constituted by the sum of the real burst data and the loss of the generated samples. Loss of the generation network is calculated by the arbiter as to whether the data generated by the generator is real data. The loss function is a binary cross entropy loss function:

BCELoss=-(ylog(p)+(1-y)log(1-p))

Where y is the true label, which is 1 if true blast data, and 0 if generated samples. p is the output probability estimation of the arbiter network Sigmoid activation function layer.

And S3.3, saving the generator network. And inputting noise vectors conforming to Gaussian distribution to generate a generated sample to amplify the original data set, so as to obtain an amplified data set. The specific structure of the architecture for generating the augmented data for the countermeasure network is shown in fig. 3.

S4.1, establishing a Gaussian process regression model:

y=f(x)+ε

Wherein x and x ^* represent any two different pieces of pre-blasting characteristic data, m is a mean function, E is a mathematical expectation, and k is a covariance function.

S4.2, obtaining prior distribution of the actual observed blasting effect y:

Let X be the pre-burst feature data in the training set, where k=k (X, X) =k _ij be a positive definite covariance matrix of size n×n, K _ij＝k(x_i,x_j) is used to measure the correlation between X _i and X _j, k=k (X ^*,X^*);K_*＝K(X^*, X) is an n×1 order covariance matrix between the test point and the training set, and I _n is an n order identity matrix.

S4.3, obtaining posterior distribution of predicted values:

S4.4, constructing a fusion kernel function:

1) The RBF kernel is used for extracting data local characteristics, and the covariance function is as follows:

2) Matern kernel, which is a popularization of RBF kernel, is also used for extracting local correlation characteristics, and the covariance function is:

3) A linear kernel for describing the linear correlation between the data, the covariance function being:

4) A polynomial core for describing a nonlinear relationship between data, the covariance function being:

Wherein delta ₁～δ₉ is the hyper-parameter of the kernel function to be set.

The multiple kernel functions constitute a fused kernel function as follows:

k_M(x_i,x_j)＝δ₁₀·k_RBF(x_i,x_j)+δ₁₁·k_Matern(x_i,x_j)+δ₁₂·k_l(x_i,x_j)+δ₁₃·k_P(x_i,x_j)

where δ ₁₀～δ₁₃ is the covariance weight of the different kernel functions.

S4.5, the augmented data set comprising the generated sample and the real sample is divided into a training set and a verification set, wherein the training set is used for training the multi-core Gaussian process regression prediction model, and the verification set is used for adjusting super parameters of the optimal model and selecting the optimal model. The model performance evaluation index is as follows:

S5.1, setting the super-parameter delta ₁～δ₁₃ in the fusion kernel function as search solution of the sparrow algorithm. The polynomial core super parameter delta ₉ is the highest degree of the polynomial, and the value range is set as [1,2,3,4,5,6]. The value range of the rest super parameters is set as [0,1]. The number of discoverers, the duty ratio of the scouts and the safety threshold are set.

S5.2, initializing the sparrow population position through an ICMIC chaotic mapping model:

Wherein Z is a generated chaotic sequence, x _i is an ith individual of the sparrow population generated according to the chaotic mapping sequence, and x _lb、x_ub is the lower limit and the upper limit of the range of each individual in the search dimension respectively. In particular, polynomial super-parameters delta ₉ still need rounding after initialization.

And S5.3, calculating the fitness of individuals in the population. And (3) establishing a multi-core Gaussian process regression prediction model with the procedure in the step S4 according to individual parameters of sparrows, and inputting an augmentation data set for training and verification. And calculating performance indexes and fitness functions, and sequencing according to the fitness, and determining the optimal individuals and the worst individuals. The fitness value of each sparrow in the population is calculated according to the following formula:

f_FIT=RMSE+MAE+(1-R2)

S5.4 updating the finder position:

Wherein ω is the dynamic self-adaptive weight of the position change of the finder, k is the control weight value range of [0,2]; T _max represents the maximum iteration number of the iteration; representing the position information of the ith sparrow when the j-dimensional iteration number is t; the method comprises the steps of respectively representing optimal fitness and worst fitness values of population individuals when iteration times are t, representing early warning values by R epsilon [0,1], representing a safety threshold value by ST epsilon [0.5,1], carrying out global search foraging by sparrows when R is less than ST, carrying out random walk by the sparrows in a regular too-distributed mode when R is more than ST, and carrying out random number obeying normal distribution by Q.

S5.5 updating the enrollee location:

S5.6, randomly selecting m multiplied by n sparrows as the scouts, wherein m is the ratio of the scouts in the population, carrying out early warning tasks and updating the positions of the scouts:

Is the optimal individual position when the iteration algebra is t, and beta is a random number obeying normal distribution with the mean value of 0 and the variance of 1. The improved position updating formula shows that if the sparrow individual is located at the current optimal position, the sparrow individual can fly to any random position between the optimal position and the worst position, otherwise, the sparrow individual can select to fly to any random position between the sparrow individual and the optimal position.

S5.7, introducing a random factor, utilizing random reverse learning to avoid the algorithm to fall into a local optimal solution and improving population diversity:

in the formula, The method is a reverse solution of the optimal solution under the t-th iteration, wherein x _ub、x_lb is an upper and lower bound, and r is a random value from 0 to 1; And b ₁ is an information exchange control parameter, which represents the individual position of the optimal solution individual after random reverse learning under t iterations.

S5.8, introducing a Cauchy mutation strategy to improve the tendency of the algorithm to fall into a locally optimal solution due to the competition of the participants and the discoverers for food:

wherein cauchy (0, 1) is a standard Cauchy distribution;

S5.9 selecting to update the optimal individual location by means of a dynamic selection strategy:

If rand (1, e) > P _s, selecting the position update based on the random reverse learning strategy, otherwise, selecting the position update based on the Kex variation disturbance strategy.

S5.10, judging whether the maximum iteration number is reached, repeating S5.3 if the maximum iteration number is not reached, and continuing if the maximum iteration number is reached. The optimal fitness iteration process is shown in fig. 4.

And S5.11, outputting an optimal individual position, namely an optimal super-parameter combination of the blasting effect multi-core Gaussian process regression prediction model, and establishing the blasting effect multi-core Gaussian process regression optimal prediction model under the optimal super-parameter combination through the processes S4.1-S4.5.

S6, storing an optimal prediction model of blasting effect multi-core Gaussian process regression under an optimal super-parameter combination, inputting a test data set to obtain blasting effect prediction of the test data set, and comparing the prediction precision of a commonly used prediction model SVM (support vector machine), BP (back propagation neural network), RF (random forest), GPR (single-core Gaussian process regression) and the patent Model (MGPR) (see figure 5).

The foregoing embodiments are merely for illustrating the technical solution of the present invention, but not for limiting the same, and although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that modifications may be made to the technical solution described in the foregoing embodiments or equivalents may be substituted for parts of the technical features thereof, and that such modifications or substitutions do not depart from the spirit and scope of the technical solution of the embodiments of the present invention in essence.

Claims

1. A small sample regression prediction method for tunnel blasting effect, characterized by comprising the following steps:

S1. Collect characteristic data before blasting and blasting effect data after blasting;

S2, data normalization, and use grey relational analysis to select input features of the regression prediction model;

S3. Establish a generative adversarial network model to augment and expand the training data set;

S4. Establish a multi-kernel Gaussian process regression prediction model for blasting effect;

S5. Sparrow algorithm search prediction model hyperparameters based on multi-strategy integrated optimization;

S6. Predict blasting effects.

2. According to claim 1, a small sample regression prediction method for tunnel blasting effect is characterized in that: in step S1, the characteristic data collected before blasting is characteristic data related to blasting effect selected according to expert experience, including blasting parameters, surrounding rock parameters, tunnel design and other aspects, specifically including: surrounding rock type, rock mass compressive strength, rock mass integrity coefficient, cross-sectional area, drilling depth, external insertion angle, blasting unit consumption, trenching angle and peripheral hole spacing, which are expressed in the following matrix form:

In the formula, n represents the number of feature data collected before blasting, and m represents the dimension of the features collected before blasting;

The collected post-blasting effect data are characteristic data that can reflect the quality of the blasting effect, including but not limited to: over-excavation and under-excavation, footage, half-hole rate and block size, which are expressed in the following matrix form:

In the formula, n represents the number of blasting effect feature data collected, and p represents the dimension of the blasting effect features collected;

The data normalization method in step S2 is:

In the formula, x _norm is the normalized data value, x _norm ∈[0,1]; x is the original data value; x _max , x _min represent the maximum and minimum values of the parameter column respectively.

3. According to the small sample regression prediction method for tunnel blasting effect in claim 1, the specific process of selecting the model input features in step S2 comprises:

First, calculate the range from the comparison feature column to the reference feature column:

Where a and b are the minimum and maximum differences between the two poles, y _kp is the k-th row data of the reference feature in the p-th column, and x _im is the k-th row data of the comparison feature in the m-th column;

Calculate the correlation:

Where r _mp is the correlation between the comparison feature in the mth column and the reference feature in the pth column, that is, the correlation between the characteristic parameters before blasting and the blasting effect, and ρ is the resolution coefficient.

4. According to the small sample regression prediction method of tunnel blasting effect in claim 1, the feature is that: step S2 further comprises: arranging the correlation degrees of each pre-blasting feature in descending order, setting the feature selection dimension, and taking the comparative features with the highest correlation degree as the input feature data after correlation degree screening according to the set feature selection dimension Ni;

The filtered input feature data and blasting effect data are divided into training data set and test data set.

5. According to the small sample regression prediction method of tunnel blasting effect in claim 1, it is characterized in that: the generative adversarial network model established in step S3, the generative network includes a one-dimensional convolutional layer, a fully connected layer and an activation function layer;

The input of the generator network is a noise vector of (Ni+No) dimension and conforming to the Gaussian distribution, and the output is a generated sample of (Ni+No) dimension; the discriminator network contains two fully connected layers and a Sigmoid activation function layer. The network output is a scalar from 0 to 1, which indicates the probability estimate of the discriminator network that the input data vector is real data, that is, the distribution similarity between the input data and the real data;

The loss of the discriminator network in the training process is composed of the sum of the losses of the real burst data and the generated samples; the loss of the generated network is calculated by whether the discriminator believes that the data generated by the generator is real data, and the loss function is the binary cross entropy loss function:

BCELoss=-(ylog(p)+(1-y)log(1-p)),

Where y is the true label, the label corresponding to the real blasting data is 1, and the label corresponding to the generated sample is 0; p is the output probability estimate of the Sigmoid activation function layer of the discriminator network.

6. According to claim 1, a small sample regression prediction method for tunnel blasting effect is characterized in that: during the adversarial game training process of GAN, the generative network gradually improves its ability to generate realistic data, and the discriminator improves its ability to discriminate whether the data is a real blasting sample; after reaching the maximum number of training times, the generative model is saved, and the generated samples of the generative model are used to augment the original training data set to obtain an augmented data set containing generated samples and real samples.

7. According to the small sample regression prediction method for tunnel blasting effect in claim 1, the characteristic is that the blasting effect Gaussian process regression prediction model establishment process in step S4 is:

Gaussian process regression is a Bayesian inference non-parametric regression method, and its regression model expression is:

y＝f(x)+ε；

In the formula, x represents the input pre-blasting feature, y represents the actual observed blasting effect, ε is the noise variable, and ε obeys f(x) is the predicted burst effect, assuming that it obeys the Gaussian process f(x)~GP(m(x),k(x,x ^* )), whose mean function is determined by the training data and covariance function is determined by the kernel function:

Where x and x ^* represent any two different characteristic data before blasting, m is the mean function, E is the mathematical expectation, and k is the covariance function;

The prior distribution of the actual observed explosion effect y is obtained as follows:

Assume that X is the pre-blast feature data in the training set, where K = K(X, X) = _kij is a positive definite covariance matrix of size n×n, _kij = k( _xi , _xj ) is used to measure the correlation between _xi and _xj ; k = k(X ^* , X ^* ); K _* = K(X ^* , X) is the n×1 order covariance matrix between the test point and the training set; _In is the n-order unit matrix, and the posterior distribution of the predicted value is obtained:

In the formula, represents the mean of the predicted values, and cov(f _* ) represents the variance.

8. According to the small sample regression prediction method of tunnel blasting effect in claim 1, it is characterized in that: further, the blasting effect multi-kernel Gaussian process regression prediction model in step S4, the fusion kernel function construction process includes:

The RBF kernel is used to extract local features of data, and its covariance function is:

The Matérn kernel is a generalization of the RBF kernel and is used to extract local correlation characteristics. Its covariance function is:

Linear kernel is used to describe the linear correlation between data. The covariance function is:

Polynomial kernel is used to describe the nonlinear relationship between data. The covariance function is:

Where δ ₁ ~δ ₉ are the kernel function hyperparameters that need to be set;

Multiple kernel functions form a fusion kernel function in the following way:

k _M (x _i ,x _j )=δ ₁₀ ·k _RBF (x _i ,x _j )+δ ₁₁ ·k _Matern (x _i ,x _j )+δ ₁₂ ·k _l (x _i ,x _j )+δ ₁₃ ·k _P (x _i ,x _j ),

Where δ ₁₀ ~δ ₁₃ are the covariance weights of different kernel functions;

Finally, the augmented dataset containing generated samples and real samples is divided into training set, validation set and test set. The training set is used to train the multi-kernel Gaussian process regression prediction model, the validation set is used to tune the model's hyperparameters and select the best model, and the test set is used to evaluate the model performance and generalization ability. The model performance evaluation indicators are:

Where MAE, RMSE, and R2 represent the mean absolute error, root mean square error, and determination coefficient, respectively; y _i and y ′ _i represent the true sample value and the predicted value, respectively. Represents the sample mean.

9. According to claim 1, a small sample regression prediction method for tunnel blasting effect is characterized in that: the multi-strategy integrated optimization sparrow algorithm in step S5, the optimization strategy includes: chaos mapping mechanism, dynamic adaptive weight, improved scout position update method, fusion Cauchy mutation and random reverse learning strategy, the specific search model hyperparameter process includes:

S501, set the hyperparameters δ ₁ to δ ₁₃ in the fusion kernel function as the search solution of the sparrow algorithm, the polynomial kernel hyperparameter δ ₉ as the highest degree of the polynomial, the value range is set to [1,2,3,4,5,6], the value range of the remaining hyperparameters is set to [0,1], set the number of discoverers, the proportion of scouts and the safety threshold;

S502. Initialize the position of the sparrow population through the ICMIC chaotic mapping model:

Where Z is the generated chaotic sequence, _xi is the i-th individual of the sparrow population generated according to the chaotic mapping sequence, _xlb and _xub are the lower and upper limits of the range of each individual in the search dimension, and the polynomial hyperparameter _δ9 still needs to be rounded after initialization;

S503, calculating the fitness of individuals in the population, sorting them according to the fitness and determining the best and worst individuals, and the fitness value of each sparrow in the population is calculated according to the following formula:

f _FIT = RMSE + MAE + (1-R2);

S504: Use a finder location update method with an adaptive weight to update the finder location:

Where ω is the dynamic adaptive weight of the discoverer position change; k is the control weight value range [0,2]; T _max represents the maximum number of iterations; It represents the position information of the i-th sparrow when the number of iterations in the j-dimension is t; They represent the optimal fitness and the worst fitness of the individuals in the population when the number of iterations is t; R∈[0,1] represents the warning value; ST∈[0.5,1] represents the safety threshold. When R<ST, the sparrow conducts a global search for food. When R>ST, the sparrow walks randomly with a normal distribution; Q is a random number that obeys a normal distribution;

S505, update the joiner's position:

In the formula, It is the position with the best fitness value among the current discoverers; is the position with the worst global fitness at present; A ⁺ = ^AT (AA ^T ) ^-1 ; A represents a single column vector of the same dimension as the individual sparrow, and its internal elements are randomly grouped by 1 and -1; n is the population size, when i≤n/2, the joiner will actively follow the discoverer to move to a better foraging position; when i＞n/2, the joiner will get rid of the current poor foraging position;

S506. Randomly select m×n sparrows as scouts, where m is the proportion of scouts in the population, to undertake the early warning task. The traditional scout update formula easily causes the scout position to be out of control and deviate from the optimal position.

Update the scout position using the following optimizations:

is the optimal individual position when the iteration number is t; β is a random number that obeys a normal distribution with a mean of 0 and a variance of 1;

The improved position update formula means that when the sparrow is at the current optimal position, it will fly to any random position between the optimal position and the worst position; otherwise, the sparrow will choose to fly to any random position between itself and the optimal position;

S507, introduce random factors and use random reverse learning to avoid the algorithm from falling into the local optimal solution and improve population diversity:

In the formula, is the reverse solution of the optimal solution in the tth iteration; x _ub and x _lb are the upper and lower bounds; r is a random value between 0 and 1; represents the individual position of the optimal solution individual after random reverse learning under t iterations; b ₁ is the information exchange control parameter;

S508. At the same time, the Cauchy mutation strategy is introduced to improve the tendency of the algorithm to fall into the local optimal solution due to the competition between joiners and discoverers for foraging:

Where cauchy(0,1) is the standard Cauchy distribution;

S509, select and update the optimal individual position through a dynamic selection strategy:

When rand(1,e)＞ _Ps , the position update based on the random reverse learning strategy is selected, otherwise the position update based on the Cauchy mutation perturbation strategy is selected;

S510, determine whether the maximum number of iterations has been reached, and proceed to the next step if the maximum number of iterations has been reached, otherwise repeat S53;

S511. Output the optimal individual position, that is, the optimal hyperparameter combination of the multi-kernel Gaussian process regression prediction model for blasting effect.

10. A small sample regression prediction method for tunnel blasting effect according to claim 1, characterized in that: in step S6, the blasting effect prediction is a multi-kernel Gaussian process regression prediction model established using the optimal hyperparameter combination output in step S5, and the blasting effect is predicted by inputting new data.