CN112597694A - Neural network-based surrounding rock deformation prediction system and prediction method - Google Patents

Neural network-based surrounding rock deformation prediction system and prediction method Download PDF

Info

Publication number
CN112597694A
CN112597694A CN202011303458.6A CN202011303458A CN112597694A CN 112597694 A CN112597694 A CN 112597694A CN 202011303458 A CN202011303458 A CN 202011303458A CN 112597694 A CN112597694 A CN 112597694A
Authority
CN
China
Prior art keywords
neural network
surrounding rock
prediction
firefly
rock deformation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011303458.6A
Other languages
Chinese (zh)
Inventor
李二兵
濮仕坤
段建立
高磊
蔡舒凌
潘越
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Army Engineering University of PLA
Original Assignee
Army Engineering University of PLA
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Army Engineering University of PLA filed Critical Army Engineering University of PLA
Publication of CN112597694A publication Critical patent/CN112597694A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
    • G01B21/00—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant
    • G01B21/32—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant for measuring the deformation in a solid
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A prediction system and a prediction method of surrounding rock deformation based on a neural network utilize monitoring equipment to obtain monitoring data of surrounding rock deformation in a set time range to serve as a learning sample of an algorithm, a non-linear autoregressive dynamic combination neural network artificial intelligence algorithm model based on firefly search algorithm optimization is established, namely, the firefly algorithm is used for carrying out global search on training results of the neural network to determine the optimal solution of delay orders and the number of hidden layer units in initial parameters of the neural network; then, the updated neural network carries out learning training again on the learning sample serving as sample data to construct a surrounding rock deformation prediction model; and setting a prediction range according to the precision of the surrounding rock deformation prediction model, and automatically outputting a surrounding rock deformation prediction value. The network learning efficiency and the prediction precision are improved, and the prediction extrapolation time range is expanded, so that the surrounding rock deformation prediction really and effectively serves scientific research work and engineering practice.

Description

Neural network-based surrounding rock deformation prediction system and prediction method
Technical Field
The invention relates to the technical field of prediction of long-term deformation time sequences of surrounding rocks, in particular to a prediction system and a prediction method for the deformation of the surrounding rocks based on a neural network, particularly to a prediction system and a prediction method for the deformation of the surrounding rocks based on a combined neural network, and particularly relates to a prediction system and a prediction method for the deformation of the surrounding rocks based on the combined neural network when the surrounding rocks are deformed for a long time and are limited by the service life of monitoring instruments and monitoring environmental conditions and are difficult to obtain long-term effective values.
Background
The surrounding rock deformation refers to the change of the shape and volume of rock mass around the underground cavern and the displacement of the cavern wall. Is a general term for the occurrence of rheology, creep, displacement, sedimentation and heaving of the surrounding rock. The displacement field representation of the surrounding rock is commonly used in mechanical analysis. The traditional prediction of surrounding rock deformation is based on mechanical analysis. Due to the coupling effect of multiple factors such as a geotechnical constitutive model, a geotechnical body structure, a geological environment, an engineering structure and the like on the underground engineering, the surrounding rock deformation mechanical calculation is too complex, and a closed solution is difficult to be given through a pure theoretical method. Meanwhile, the factors influencing the deformation of the surrounding rock and the factors influencing the deformation of the surrounding rock have the characteristics of nonlinearity, dispersion, randomness and the like, and the computer-aided numerical simulation method is more suitable for practical working conditions in terms of method and result, but is not suitable for practical engineering due to the fact that parameters are too complicated. Therefore, the deformation of the surrounding rock is difficult to accurately predict no matter theoretical calculation or numerical simulation under the condition of the prior art.
With the application of machine learning algorithm in the field of rock and soil in recent years, prediction is directly carried out on the basis of monitoring data through a regression analysis model, a gray system model, a neural network model, a support vector machine model and the like, and the advantages in the aspects of program implementation and the like are remarkable. However, the single algorithm models still have limitations, are often not high in accuracy, and are only suitable for short-term prediction. Therefore, a combined model suitable for predicting the long-term deformation of the surrounding rock is urgently found.
Disclosure of Invention
In order to solve the problems, the invention provides a prediction system and a prediction method for surrounding rock deformation based on a neural network, so that a combined model suitable for long-term surrounding rock deformation prediction is provided.
In order to overcome the defects in the prior art, the invention provides a solution of a prediction system and a prediction method of surrounding rock deformation based on a neural network, which comprises the following specific steps:
a neural network-based surrounding rock deformation prediction system comprises:
the monitoring equipment is used for inputting the monitored data of the deformation of the surrounding rock into the test terminal;
the testing terminal comprises an NAR dynamic neural network, and the NAR dynamic neural network consists of an input layer, a hidden layer, an output layer and a delay order part.
A prediction method of a neural network-based surrounding rock deformation prediction system comprises the following steps:
acquiring monitoring data of surrounding rock deformation in a set time range by using monitoring equipment as a learning sample of an algorithm, and establishing a firefly search algorithm optimization-based nonlinear autoregressive dynamic combined neural network artificial intelligence algorithm model, namely determining the optimal solution of a delay order and the number of hidden layer units in initial parameters of the neural network by using the global search of the firefly algorithm on training results of the neural network; then, the updated neural network carries out learning training again on the learning sample serving as sample data to construct a surrounding rock deformation prediction model; and setting a prediction range according to the precision of the surrounding rock deformation prediction model, and automatically outputting a surrounding rock deformation prediction value.
The prediction method of the neural network-based surrounding rock deformation prediction system specifically comprises the following steps:
step 1: inputting monitoring data of surrounding rock deformation; the input of the monitoring data of the deformation of the surrounding rock comprises the following steps: and acquiring the radial single-dimensional displacement data of a target point of the monitoring section of the surrounding rock through manual acquisition or automatic monitoring acquisition.
Step 2: initializing a firefly algorithm, and randomly generating a parameter group of a nonlinear autoregressive dynamic neural network with the number of fireflies of N, wherein N is a positive integer; the initializing firefly algorithm randomly generates a parameter group of a nonlinear autoregressive dynamic neural network with the number of fireflies of N, and comprises the following steps: optimizing delay order d and hidden layer unit number n as neural network parameters by using firefly algorithm global searchhThe corresponding search range is: d: 5 to 10, nh: 5-20; the maximum iteration number is set to be 200, the light intensity absorption coefficient gamma is 1.02, the step factor alpha is 0.12, and the maximum attraction beta is set0At 1, the fluorescence luminance I (r) is represented by the formula (1):
I(r)＝I0e-γr (1)
in equation (1): i is0The maximum fluorescence brightness is the fluorescence brightness of the firefly per se under the condition that r is 0; r is the Euclidean distance r between the fluorescent insect i and the fluorescent insect jijIs shown in formula (2):
Figure BDA0002787483500000031
in equation (2): n is the number of fireflies; x is the number ofi,kIs the kth component of firefly i, i and j are positive integers;
the attraction degree β (r) is expressed by the formula (3):
Figure BDA0002787483500000032
in equation (3): beta is a0The maximum attraction degree is the attraction degree of the light source under the condition that r is 0; m is a set parameter;
location update function xi(t +1) is represented by formula (4):
xi(t+1)＝xi(t)+β[xj(t)-xi(t)]+α(rand-1/2) (4)
in equation (4): x is the number ofi(t) is firefly xiThe spatial position after the t-th movement; t is a positive integer and represents the number of iterations; alpha is a step factor and is in [0,1 ]]A constant within a range; rand is a random factor, at [0,1 ]]The medicine is uniformly distributed within the scope.
And step 3: reading sample data by the NAR dynamic neural network, simultaneously reading all firefly individuals, namely network parameters, in the initial population, training and predicting the NAR dynamic neural network, wherein the training and predicting of the NAR dynamic neural network comprises the following steps: the relation between the deformation value y (t) at the moment t and the deformation values y (t-1), y (t-2), … and y (t-d) at the previous d moments is set, and the model is shown in formula (5):
y(t)＝f[y(t-1),y(t-2),…,y(t-d)] (5)
in equation (5): f [ ] is a nonlinear function, d is the delay order;
then, the prediction results of all individuals are transmitted to a firefly algorithm, iterative updating is carried out, the optimal individual, namely a network parameter, is globally searched, wherein the initial position of the firefly is randomly set, and I is initiated0Equal to the objective function value, where the objective function uses mean square error MSE as the determination criterion, which is specifically shown in formula (6):
Figure BDA0002787483500000041
in equation (6): y isiFor prediction of ith time stepValue, yi' is the measured value of the ith time step, i is a positive integer;
respectively calculating I and beta according to a formula (1) and a formula (3), determining the moving direction of the firefly individual by I, calculating the position of the firefly and updating by the formula (5), randomly disturbing the position of the optimal individual, recalculating I and beta according to the updated spatial position of the firefly, outputting the optimal individual in the last iteration as the optimal network parameter serving as the optimal search result when the maximum iteration number is reached, and ending the algorithm; otherwise, adding 1 to the iteration times, and repeatedly executing the search command until the preset maximum iteration times is reached.
And 4, step 4: and (3) returning the optimal network parameters to the NAR dynamic neural network for learning and training again, judging whether the NAR dynamic neural network is good or not according to the prediction effect graph of the NAR dynamic neural network and the bar graph of the error autocorrelation, and returning the NAR dynamic neural network which does not meet the requirements to the step 3 for iterative training until the NAR dynamic neural network is in an ideal state, namely the building of the prediction model of the surrounding rock deformation is completed.
The invention has the beneficial effects that:
the nonlinear autoregressive dynamic neural network adopted by the invention has feedback and memory functions, so that the calculation efficiency and the learning effect on the time series data of the whole process of the deformation of the surrounding rock are better. In order to avoid the blindness of artificial input of initial parameters by the neural network, the time delay order and the number of hidden layer units of the non-linear autoregressive dynamic neural network are globally searched by the firefly algorithm with simple and convenient operation and few parameters, so that the network learning efficiency and the prediction precision are improved, the time range of prediction extrapolation is expanded, and the prediction of the deformation of the surrounding rock really and effectively serves scientific research work and engineering practice.
Drawings
Fig. 1 is a flow chart of a prediction method of a neural network-based surrounding rock deformation prediction system of the present invention.
FIG. 2 is a graph of the measuring point full displacement time course of the surrounding rock according to the embodiment of the invention.
Fig. 3 is a network topology diagram of a NAR dynamical neural network of an embodiment of the present invention.
Fig. 4 is a diagram of the predicted effect of the NAR dynamical neural network of the embodiment of the present invention.
Fig. 5 is a bar graph of the error autocorrelation of the NAR dynamical neural network of an embodiment of the present invention.
FIG. 6 is a diagram showing the comparison of the predicted results of FA-NAR of the present invention and LS-SVM of the prior art.
FIG. 7 is a schematic diagram of the comparison of FA-NAR prediction results and actual monitoring values of the present invention.
FIG. 8 is a graphical illustration of the prediction results of the extrapolated prediction time range of the FA-NAR of the present invention.
Detailed Description
The surrounding rock deformation is one of the most direct mechanical responses of underground engineering excavation disturbance, and is an important basis for evaluating the stability of the rock mass and inverting the mechanical parameters of the rock mass. In the excavation process of underground engineering, not only the stress state is continuously adjusted along with excavation, but also the rock mass shape and the internal state are continuously changed, and the surrounding rock deformation is actually a long-term change process. In practical engineering, generally, the buried monitoring instrument is used for acquiring measured data to provide basic information for surrounding rock stability judgment and deformation analysis. However, due to the limitation of multiple factors such as the service life of monitoring instruments and equipment, the environmental conditions, and the like, long-term effective monitoring data is generally difficult to obtain, and great inconvenience is brought to long-term stability analysis and safety early warning of underground engineering. Therefore, prediction of deformation of surrounding rocks is necessary.
The invention adopts an artificial neural network method optimized by an intelligent algorithm, establishes a neural network combination model capable of reflecting the internal change rule of data through the learning and training of actual monitoring data, makes up the defects of a single algorithm model in precision and prediction range, and can accurately predict the future evolution rule and development trend of the deformation of the surrounding rock.
The invention will be further described with reference to the following figures and examples.
A neural network-based surrounding rock deformation prediction system comprises:
the monitoring equipment is used for inputting the monitored data of the deformation of the surrounding rock into the test terminal; the monitoring device can be a bow micrometer, a horizontal comparator or a strain sensor.
The testing terminal comprises an NAR dynamic neural network, and the NAR dynamic neural network consists of an input layer, a hidden layer, an output layer and a delay order part.
A prediction method of a neural network-based surrounding rock deformation prediction system comprises the following steps:
acquiring monitoring data of surrounding rock deformation in a set time range by using monitoring equipment as a learning sample of an algorithm, and establishing a firefly search algorithm (FA) optimized Nonlinear Autoregressive (NAR) dynamic combination neural network artificial intelligence algorithm model, namely determining the optimal solution of the number of delay orders and hidden layer units in initial parameters of the neural network by using global search of the firefly algorithm on training results of the neural network; then, the updated neural network carries out learning training again on the learning sample serving as sample data to construct a surrounding rock deformation prediction model; and setting a prediction range according to the precision of the surrounding rock deformation prediction model, and automatically outputting a surrounding rock deformation prediction value. The monitoring device can be a bow micrometer, a horizontal comparator or a strain sensor.
The method for predicting the deformation of the surrounding rock is realized based on a neural network combination model, the selected embodiment of the invention is granite underground roadway engineering, and the method for predicting the deformation of the surrounding rock based on the neural network specifically comprises the following steps:
step 1: the roadway deformation mainly considers the problem of plane strain, and monitoring data of surrounding rock deformation are input to a test terminal; the input of the monitoring data of the deformation of the surrounding rock comprises the following steps: and acquiring radial single-dimensional displacement data of a target point of a preset monitoring section of the surrounding rock through manual acquisition or automatic monitoring acquisition. This example collected full displacement data of target points from 2016, 10, 26, 12:00, to 2018, 7, 2, 12:00, and processed into displacement time series data with a time step of 6 hours as a learning training sample, as shown in fig. 2. 70% of sample data is used as a training set to train the neural network, 15% is used as a verification set to verify whether the neural network is over-trained, and the other 15% is used as a test set to judge the quality of the neural network. The manual collection can be that the radial single-dimensional displacement data of a target point of the cross section is monitored manually by using an arch-shaped screw micrometer or a horizontal comparator and is input into a test terminal. The automatic monitoring and acquisition can be realized by acquiring the radial single-dimensional displacement data of a target point of the monitoring section by using a strain sensor arranged on the monitoring section, and the strain sensor is in communication connection with the test terminal, so that the acquired data can be automatically sent to the test terminal.
Step 2: initializing a firefly algorithm, and randomly generating a parameter group of a nonlinear autoregressive dynamic neural network with the number of fireflies of N, wherein N is a positive integer; the initializing firefly algorithm randomly generates a parameter group of a nonlinear autoregressive dynamic neural network with the number of fireflies of N, and comprises the following steps: optimizing delay order d and hidden layer unit number n as neural network parameters by using firefly algorithm global searchhThe corresponding search range is: d: 5 to 10, nh: 5-20; the maximum iteration number is set to be 200, the light intensity absorption coefficient gamma is 1.02, the step factor alpha is 0.12, and the maximum attraction beta is set0At 1, the fluorescence luminance I (r) is represented by the formula (1):
I(r)＝I0e-γr (1)
in equation (1): i is0The maximum fluorescence brightness is the fluorescence brightness of the firefly per se under the condition that r is 0; r is the Euclidean distance r between the fluorescent insect i and the fluorescent insect jijIs shown in formula (2):
Figure BDA0002787483500000081
in equation (2): n is the number of fireflies; x is the number ofi,kIs the kth component of firefly i, i and j are positive integers;
the attraction degree β (r) is expressed by the formula (3):
Figure BDA0002787483500000082
formula (II)(3) The method comprises the following steps: beta is a0The maximum attraction degree is the attraction degree of the light source under the condition that r is 0; m is a set parameter, and generally m is 2;
location update function xi(t +1) is represented by formula (4):
xi(t+1)＝xi(t)+β[xj(t)-xi(t)]+α(rand-1/2) (4)
in equation (4): x is the number ofi(t) is firefly xiThe spatial position after the t-th movement; t is a positive integer and represents the number of iterations; alpha is a step factor and is in [0,1 ]]A constant within a range; rand is a random factor, at [0,1 ]]The medicine is uniformly distributed within the scope.
And step 3: the NAR dynamic neural network reads sample data, reads all firefly individuals, namely network parameters, in an initial population, trains and predicts the NAR dynamic neural network, the NAR dynamic neural network consists of an input layer, an output layer, a hidden layer and a delay order part arranged on the hidden layer, the model structure of the NAR dynamic neural network is shown in figure 3, y (t) on the left represents network input, y (t) on the right represents network output, w is a weight, b is a threshold, and 1:5 represents the delay order. The training and predicting the NAR dynamic neural network comprises the following steps: the relation between the deformation value y (t) at the moment t and the deformation values y (t-1), y (t-2), … and y (t-d) at the previous d moments is set, and the model is shown in formula (5):
y(t)＝f[y(t-1),y(t-2),…,y(t-d)] (5)
in equation (5): f [ ] is a nonlinear function, d is the delay order;
then, the prediction results of all individuals are transmitted to a firefly algorithm, iterative updating is carried out, the optimal individual, namely a network parameter, is globally searched, wherein the initial position of the firefly is randomly set, and I is initiated0Equal to the objective function value, where the objective function uses mean square error MSE as the determination criterion, which is specifically shown in formula (6):
Figure BDA0002787483500000091
in equation (6): y isiIs the predicted value of the ith time step, yi' is the measured value of the ith time step, i is a positive integer;
respectively calculating I and beta according to a formula (1) and a formula (3), determining the moving direction of the firefly individual by I, calculating the position of the firefly and updating by the formula (5), randomly disturbing the position of the optimal individual, recalculating I and beta according to the updated spatial position of the firefly, outputting the optimal individual in the last iteration as the optimal network parameter serving as the optimal search result when the maximum iteration number is reached, and ending the algorithm; otherwise, adding 1 to the iteration times, repeatedly executing the search command until the preset maximum iteration times is reached, taking 9 to the step d when the optimal network parameter setting value of the NAR dynamic neural network is obtained, and setting n to the number of hidden layer unitshAnd taking 16.
And 4, step 4: and returning the optimal network parameters to the NAR dynamic neural network for learning and training again, judging the quality of the NAR dynamic neural network according to specific requirements according to a prediction effect graph and an error autocorrelation bar graph of the NAR dynamic neural network, returning the NAR dynamic neural network which does not meet the requirements to the step 3 for iterative training, and completing the establishment of a prediction model of the deformation of the surrounding rock of the roadway until the required ideal state is achieved. As shown in fig. 4 and 5, the maximum error of the prediction effect of the NAR dynamical neural network is less than 0.002, and the error autocorrelation is within the confidence interval. Fig. 6 compares the NAR dynamic neural network FA-NAR of the present invention with the LS-SVM method of the prior art, and it is obvious that the method of the present invention has a smaller prediction error than the method of the prior art, the deviation of the method of the present invention from the actual monitoring value is also greatly reduced, and the NAR dynamic neural network combination model has obvious advantages in prediction performance.
In order to determine the prediction accuracy of the method and reasonably extrapolate the prediction time range, the FA-NAR combined model is also adopted to carry out learning training on detection data of 181 days in total from 1 month and 1 day in 2017 to 6 months and 30 days in 2017, and the optimal delay order d and the number n of hidden layer units determined by the methodhTake 6 and 20 respectively. The extrapolated predictions are shown in table 1:
TABLE 1
Figure BDA0002787483500000101
Figure BDA0002787483500000111
And extrapolating and predicting the surrounding rock deformation value of 365 days later based on the monitoring data of 181 days, and comparing the surrounding rock deformation value with the measured value to show that the relative error of the prediction result is less than 5 percent. The extrapolation prediction of the embodiment shown in fig. 7 is significantly diverged from the actual displacement trend after 2 months in 2018. In the conservative case, it is considered that the prediction result has reliability when the extrapolated prediction time scale of the surrounding rock deformation prediction model established in the embodiment is close to the training data time scale. In the embodiment, the model is subjected to learning training based on 614 days of monitoring data, and surrounding rock deformation data is predicted for the subsequent 614 days (3 months and 4 days 2020) as shown in fig. 8.
The present invention has been described in an illustrative manner by the embodiments, and it should be understood by those skilled in the art that the present disclosure is not limited to the embodiments described above, but is capable of various changes, modifications and substitutions without departing from the scope of the present invention.

Claims (6)

1. A neural network-based surrounding rock deformation prediction system is characterized by comprising:
the monitoring equipment is used for inputting the monitored data of the deformation of the surrounding rock into the test terminal;
the testing terminal comprises an NAR dynamic neural network, and the NAR dynamic neural network consists of an input layer, a hidden layer, an output layer and a delay order part.
2. A prediction method of a neural network-based surrounding rock deformation prediction system is characterized by comprising the following steps:
acquiring monitoring data of surrounding rock deformation in a set time range by using monitoring equipment as a learning sample of an algorithm, and establishing a firefly search algorithm optimization-based nonlinear autoregressive dynamic combined neural network artificial intelligence algorithm model, namely determining the optimal solution of a delay order and the number of hidden layer units in initial parameters of the neural network by using the global search of the firefly algorithm on training results of the neural network; then, the updated neural network carries out learning training again on the learning sample serving as sample data to construct a surrounding rock deformation prediction model; and setting a prediction range according to the precision of the surrounding rock deformation prediction model, and automatically outputting a surrounding rock deformation prediction value.
3. A prediction method of a neural network-based surrounding rock deformation prediction system is characterized by comprising the following steps:
step 1: inputting monitoring data of surrounding rock deformation;
step 2: initializing a firefly algorithm, and randomly generating a parameter group of a nonlinear autoregressive dynamic neural network with the number of fireflies of N, wherein N is a positive integer;
and step 3: reading sample data by the NAR dynamic neural network, simultaneously reading each firefly individual in the initial population, namely network parameters, and training and predicting the NAR dynamic neural network;
and 4, step 4: and (3) returning the optimal network parameters to the NAR dynamic neural network for learning and training again, judging whether the NAR dynamic neural network is good or not according to the prediction effect graph of the NAR dynamic neural network and the bar graph of the error autocorrelation, and returning the NAR dynamic neural network which does not meet the requirements to the step 3 for iterative training until the NAR dynamic neural network is in an ideal state, namely the building of the prediction model of the surrounding rock deformation is completed.
4. The prediction method of the neural network-based surrounding rock deformation prediction system according to claim 3, wherein the inputting of the monitoring data of the surrounding rock deformation includes: and acquiring the radial single-dimensional displacement data of a target point of the monitoring section of the surrounding rock through manual acquisition or automatic monitoring acquisition.
5. The prediction method of the neural network-based surrounding rock deformation prediction system according to claim 3,the initializing firefly algorithm randomly generates a parameter group of a nonlinear autoregressive dynamic neural network with the number of fireflies of N, and comprises the following steps: optimizing delay order d and hidden layer unit number n as neural network parameters by using firefly algorithm global searchhThe corresponding search range is: d: 5 to 10, nh: 5-20; the maximum iteration number is set to be 200, the light intensity absorption coefficient gamma is 1.02, the step factor alpha is 0.12, and the maximum attraction beta is set0At 1, the fluorescence luminance I (r) is represented by the formula (1):
I(r)＝I0e-γr (1)
in equation (1): i is0The maximum fluorescence brightness is the fluorescence brightness of the firefly per se under the condition that r is 0; r is the Euclidean distance r between the fluorescent insect i and the fluorescent insect jijIs shown in formula (2):
Figure FDA0002787483490000021
in equation (2): n is the number of fireflies; x is the number ofi,kIs the kth component of firefly i, i and j are positive integers;
the attraction degree β (r) is expressed by the formula (3):
Figure FDA0002787483490000022
in equation (3): beta is a0The maximum attraction degree is the attraction degree of the light source under the condition that r is 0; m is a set parameter;
location update function xi(t +1) is represented by formula (4):
xi(t+1)＝xi(t)+β[xj(t)-xi(t)]+α(rand-1/2) (4)
in equation (4): x is the number ofi(t) is firefly xiThe spatial position after the t-th movement; t is a positive integer and represents the number of iterations; alpha is a step factor and is in [0,1 ]]Within the range ofA constant of (d); rand is a random factor, at [0,1 ]]The medicine is uniformly distributed within the scope.
6. The prediction method of the neural network-based surrounding rock deformation prediction system according to claim 3, wherein the training and predicting the NAR dynamic neural network comprises: the relation between the deformation value y (t) at the moment t and the deformation values y (t-1), y (t-2), … and y (t-d) at the previous d moments is set, and the model is shown in formula (5):
y(t)＝f[y(t-1),y(t-2),…,y(t-d)] (5)
in equation (5): f [ ] is a nonlinear function, d is the delay order;
then, the prediction results of all individuals are transmitted to a firefly algorithm, iterative updating is carried out, the optimal individual, namely a network parameter, is globally searched, wherein the initial position of the firefly is randomly set, and I is initiated0Equal to the objective function value, where the objective function uses mean square error MSE as the determination criterion, which is specifically shown in formula (6):
Figure FDA0002787483490000031
in equation (6): y isiIs the predicted value of the ith time step, yi' is the measured value of the ith time step, i is a positive integer;
respectively calculating I and beta according to a formula (1) and a formula (3), determining the moving direction of the firefly individual by I, calculating the position of the firefly and updating by the formula (5), randomly disturbing the position of the optimal individual, recalculating I and beta according to the updated spatial position of the firefly, outputting the optimal individual in the last iteration as the optimal network parameter serving as the optimal search result when the maximum iteration number is reached, and ending the algorithm; otherwise, adding 1 to the iteration times, and repeatedly executing the search command until the preset maximum iteration times is reached.
CN202011303458.6A 2020-05-28 2020-11-19 Neural network-based surrounding rock deformation prediction system and prediction method Pending CN112597694A (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010466859 2020-05-28
CN2020104668597 2020-05-28

Publications (1)

Publication Number Publication Date
CN112597694A true CN112597694A (en) 2021-04-02

Family

ID=75183626

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011303458.6A Pending CN112597694A (en) 2020-05-28 2020-11-19 Neural network-based surrounding rock deformation prediction system and prediction method

Country Status (1)

Country Link
CN (1) CN112597694A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113340211A (en) * 2021-08-03 2021-09-03 中国工程物理研究院激光聚变研究中心 Interference image phase demodulation method based on deep learning
CN113719283A (en) * 2021-09-07 2021-11-30 武汉理工大学 Method and device for predicting working hours of mine rock drilling equipment
CN117574781A (en) * 2024-01-15 2024-02-20 国网湖北省电力有限公司经济技术研究院 Intelligent prediction method and system for security risk of surrounding rock of underground factory building of pumped storage power station
CN117852402A (en) * 2024-01-08 2024-04-09 西南交通大学 Tunnel surrounding rock stability prediction method, device, equipment and storage medium
CN117852402B (en) * 2024-01-08 2024-06-25 西南交通大学 Tunnel surrounding rock stability prediction method, device, equipment and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101344389A (en) * 2008-08-20 2009-01-14 中国建筑第八工程局有限公司 Method for estimating tunnel surrounding rock displacement by neural network
US20200018164A1 (en) * 2018-07-12 2020-01-16 China Institute Of Water Resources And Hydropower Research Advanced monitoring device for whole-process deformation curve of surrounding rock of tunnel excavation and implementation method thereof

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101344389A (en) * 2008-08-20 2009-01-14 中国建筑第八工程局有限公司 Method for estimating tunnel surrounding rock displacement by neural network
US20200018164A1 (en) * 2018-07-12 2020-01-16 China Institute Of Water Resources And Hydropower Research Advanced monitoring device for whole-process deformation curve of surrounding rock of tunnel excavation and implementation method thereof

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YUE PAN ET AL: "Research on deformation prediction of tunnel surrounding rock using the model combining firefly algorithm and nonlinear auto‑regressive dynamic neural network", ENGINEERING WITH COMPUTERS, pages 1443 - 1453 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113340211A (en) * 2021-08-03 2021-09-03 中国工程物理研究院激光聚变研究中心 Interference image phase demodulation method based on deep learning
CN113719283A (en) * 2021-09-07 2021-11-30 武汉理工大学 Method and device for predicting working hours of mine rock drilling equipment
CN113719283B (en) * 2021-09-07 2023-01-17 武汉理工大学 Method and device for predicting working hours of mine rock drilling equipment
CN117852402A (en) * 2024-01-08 2024-04-09 西南交通大学 Tunnel surrounding rock stability prediction method, device, equipment and storage medium
CN117852402B (en) * 2024-01-08 2024-06-25 西南交通大学 Tunnel surrounding rock stability prediction method, device, equipment and storage medium
CN117574781A (en) * 2024-01-15 2024-02-20 国网湖北省电力有限公司经济技术研究院 Intelligent prediction method and system for security risk of surrounding rock of underground factory building of pumped storage power station
CN117574781B (en) * 2024-01-15 2024-04-16 国网湖北省电力有限公司经济技术研究院 Intelligent prediction method and system for security risk of surrounding rock of underground factory building of pumped storage power station

Similar Documents

Publication Publication Date Title
CN112597694A (en) Neural network-based surrounding rock deformation prediction system and prediction method
CN110046743B (en) Public building energy consumption prediction method and system based on GA-ANN
CN111310968A (en) LSTM neural network circulation hydrological forecasting method based on mutual information
CN111784070A (en) Intelligent landslide short-term early warning method based on XGboost algorithm
CN111666671A (en) Real-time inversion method for creep parameters of surrounding rock mass
CN112528365B (en) Method for predicting healthy evolution trend of underground infrastructure structure
CN112614021B (en) Tunnel surrounding rock geological information prediction method based on built tunnel information intelligent identification
CN116089870A (en) Industrial equipment fault prediction method and device based on meta-learning under small sample condition
CN116341272A (en) Construction safety risk management and control system for digital distribution network engineering
CN115654381A (en) Water supply pipeline leakage detection method based on graph neural network
CN113886989A (en) Petroleum drilling parameter optimization method and system based on machine learning
CN117391674A (en) Reliability-based preventive maintenance optimization method and device for electrical equipment
CN113988210A (en) Method and device for restoring distorted data of structure monitoring sensor network and storage medium
CN116957356B (en) Scenic spot carbon neutralization management method and system based on big data
CN117113644A (en) Slope temporary slip forecasting method, system and medium based on deep monitoring deformation sequence
CN114626115A (en) Building hourly thermal load prediction modeling method based on transfer learning
CN116960962A (en) Mid-long term area load prediction method for cross-area data fusion
CN116663126A (en) Bridge temperature effect prediction method based on channel attention BiLSTM model
CN116591768A (en) Tunnel monitoring method, system and device based on distributed network
CN115577856A (en) Method and system for predicting construction cost and controlling balance of power transformation project
CN115217152A (en) Method and device for predicting opening and closing deformation of immersed tunnel pipe joint
CN115422840B (en) Daily-scale runoff estimation method based on physical model hybrid deep learning model
CN117216846B (en) Reinforced concrete member hysteresis curve prediction method, system, equipment and medium
CN117830031B (en) Water supply network terminal water quality turbidity prediction method and related equipment
CN109886420A (en) A kind of adaptive coalcutter cutting height intelligent predicting system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination