CN117113842A - Tunnel surrounding rock deformation prediction method based on PSO-LSTM model - Google Patents

Abstract

The invention discloses a tunnel surrounding rock deformation prediction method based on a PSO-LSTM model, and the scheme of the invention provides a tunnel surrounding rock deformation prediction method based on a PSO-LSTM hybrid model, which is characterized in that a long-short-term neural network (LSTM) is established to process and train actual monitoring data, and a particle swarm algorithm is combined to optimize the learning rate and the number of hidden layer neurons, so that the aims of reducing sample training time and improving surrounding rock deformation prediction accuracy can be realized; the method can provide theoretical support for optimization of tunnel supporting scheme, tunnel safety construction and operation. The surrounding rock deformation result predicted by the method has high accuracy and high efficiency, and has wide application prospect in the construction, operation and maintenance of a large number of tunnel engineering in the future.

Description

Tunnel surrounding rock deformation prediction method based on PSO-LSTM model

Technical Field

The invention relates to the technical field of tunnel engineering, in particular to a tunnel surrounding rock deformation prediction method based on a PSO-LSTM model.

Background

In recent years, along with the rapid development of transportation industry in China, especially along with the enhancement of the national high-grade railway and highway supporting force, the scale of tunnel construction is also larger and larger, tunnel engineering is more and more under poor geology and environmental conditions, and complicated tunnel surrounding rock deformation is accompanied. The deformation characteristics and the development trend of the surrounding rock are important information for revealing the steady state of the surrounding rock. The deformation of the surrounding rock is abnormal or abrupt, the balance state of the surrounding rock is broken due to the fact that the stress of the tunnel is deteriorated, the normal use of the tunnel function is affected when the surrounding rock is light, and disaster accidents such as large deformation and collapse can occur when the surrounding rock is heavy. Due to the combined action of various factors such as construction load, supporting resistance and natural boundary external force, and factors such as heterogeneity and discontinuity of surrounding rocks, the deformation characteristics of the surrounding rocks of the tunnel are extremely complex, the deformation of the surrounding rocks often shows the characteristics of mutability, discontinuity and the like, and the method for acquiring the deformation data of the surrounding rocks and judging the stable state of the tunnel is particularly important. The tunnel deformation characteristics under the complex geological topography condition are complex, the monitoring difficulty is high, and due to the problems of various external factors, the stability of the instrument and the like, the monitoring data have the reality problems of discontinuity or loss and the like, so that a theoretical model is constructed, surrounding rock deformation is predicted based on the existing monitoring data, and the tunnel stability is evaluated as a current research hot spot.

At present, a great number of students at home and abroad conduct a great deal of research on tunnel surrounding rock deformation prediction, and common prediction methods mainly comprise a test method, a numerical simulation method, a gray theory method, a neural network and other prediction methods, but the methods have certain limitations, and have low prediction precision because of low processing efficiency, weak data learning capability, long-term dependence of data incapability of processing and the like of massive discontinuous data.

Although significant progress is made in the field of tunnel engineering construction, complex and variable geology and sensitive environmental conditions remain an insurmountable obstacle, which makes tunnel engineering construction extremely complex and challenging. The management and supervision of tunnel construction are optimized, the safety level of the tunnel is improved, a powerful guarantee is provided for sustainable development of tunnel quality, and the method is also a necessary requirement for development of tunnel engineering technology. Because tunnel engineering is a complex underground structure engineering, complex supporting and surrounding rock interaction exists, and various factors such as construction load, multiple external operating forces, poor geological environment conditions and the like influence, so that the deformation characteristics of the surrounding rock of the tunnel are extremely complex, and therefore, the tunnel deformation monitoring has the problems of data loss, discontinuity, high long-term monitoring cost and the like, and the tunnel deformation monitoring based on the deformation prediction of limited data has very important theoretical and practical significance for tunnel stability evaluation, safe construction and operation.

In summary, the scheme provides a deep learning model (Particle Swarm Optimization-Long Short-Term Memory, PSO-LSTM model) based on particle swarm optimization and Long-Short-Term Memory network, which is applied to prediction analysis of tunnel surrounding rock deformation for the first time.

Disclosure of Invention

Therefore, the invention aims to provide the tunnel surrounding rock deformation prediction method based on the PSO-LSTM model, which is high in efficiency and good in result accuracy.

In order to achieve the technical purpose, the invention adopts the following technical scheme:

a tunnel surrounding rock deformation prediction method based on a PSO-LSTM model comprises the following steps:

s01, obtaining deformation data of tunnel vault settlement and horizontal convergence and time points obtained by the deformation data to form time series data;

s02, preprocessing time sequence data;

s03, dividing the preprocessed time series data into a training set and a testing set according to a preset proportion;

s04, constructing an LSTM model for arch top settlement and horizontal convergence prediction in tunnel surrounding rock deformation, and carrying out parameter optimization on the LSTM model through a PSO algorithm;

s05, setting parameters of an LSTM model, training a vault subsidence and horizontal convergence prediction model based on the LSTM by utilizing a training set, and repeatedly training the LSTM by adopting a particle swarm parameter optimization model to ensure that the prediction accuracy of the model meets the preset requirement and obtain a PSO-LSTM model;

s06, inputting the test set into the PSO-LSTM model for prediction, and outputting a predicted value.

As a possible implementation manner, further, the scheme S01 further includes: and drawing a vault settlement-time and horizontal convergence-time deformation curve related to time according to the deformation data of the tunnel vault settlement and horizontal convergence, and performing preliminary analysis and evaluation according to the change rule of the deformation curve and the deformation rule of the tunnel surrounding rock.

As a possible implementation manner, further, in the scheme S02, preprocessing the time series data includes one or more of data cleaning, missing value filling, and normalization processing;

the data cleaning comprises abnormal data elimination of time series data;

the missing value filling comprises the steps of fitting a curve formula according to the time series data, and then obtaining a corresponding time series data value according to the curve formula so as to complete the supplement of the deformation data.

As a possible implementation manner, in the present embodiment S03, 80% of the data amount is extracted from the preprocessed time-series data as the training set, and the remaining 20% is used as the test set.

As a possible implementation manner, in the present solution S04, parameter optimization is performed on the LSTM model by using a PSO algorithm, which includes setting one or more parameters of a population size, an iteration number, and an inertia weight, so as to maximize prediction accuracy of the model.

Further, in the present embodiment S05, setting parameters of the LSTM model includes setting one or more of a hidden layer size and a learning rate parameter.

As a possible implementation manner, further, in the scheme S06, the test set is input into the PSO-LSTM model to make a prediction, and for the input data at each time, the PSO-LSTM model generates a predicted value according to the previous state and input, and uses the predicted value as the input value at the next time.

As a possible implementation manner, further, the performing parameter optimization on the LSTM model through the PSO algorithm in the solution S04 includes:

assuming that in a D-dimensional optimization space there are N particles that make up a population, the particles have a position vector and a velocity vector, comprising:

the position coordinates of the i-th particle are noted as: x is X _i ＝(x _i1 ,x _i2 ,...x _iD ),i＝1,2,...N；

The velocity vector of the i-th particle is noted as: v (V) _i ＝(v _i1 ,v _i2 ,...v _iD ),i＝1,2,...N；

Iterating the formula of the m-th generation to the m+1th generation particle update:

v _ij (m+1)＝ωv _ij (m)+c ₁ r ₁ (m)[p _ij (m)-x _ij (m)]+c ₂ r ₂ (m)[p _gj (m)-x _gj (m)]

x _ij (m+1)＝x _ij (m)+v _ij (m+1)

wherein ω is inertial weight, c ₁ ，c ₂ To learn factors, p _i For the individual extremum, p _g Is globally optimal.

As a possible implementation manner, further, the training of the dome subsidence and horizontal convergence prediction model based on LSTM in step S05 of the present solution includes:

constructing a forgetting door control unit f _n ：f _n ＝σ(W _f h _n-1 +U _f x _n +b _f )；

Building an update door control unit: i.e _n ＝σ(W _i h _n-1 +U _i x _n +b _i )；

C′ _n ＝tanh(W _c h _n-1 +U _c x _n +b _c )；

Building an output door control unit: o (O) _n ＝σ(W _o h _n-1 +U _o x _n +b _o )；

Output state results: h is a _n ＝tanh×O _n ；

C _n ＝f _n ×C _n-1 +i _n ×C′ _n

Wherein W is _f Weight of short-term memory state of forgetting gate, U _f B, inputting the weight of the vector for the forgetting gate _f The bias parameters are the bias parameters of the forgetting door; w (W) _i Is i _n Weights of short-term memory states of function, W _c Is C' _n Weights of short-term memory states of function, U _i Is i _n Weights of input vectors, U _c Is C' _n Weights of input vector, b _i To update gate i _n Bias parameters of the function, b _c To update door C' _n Bias parameters of the function.

As a possible implementation manner, the solution further includes:

s07, carrying out prediction result evaluation on the PSO-LSTM model according to a preset evaluation index, and optimizing the PSO-LSTM model according to the evaluation result so as to further improve the prediction accuracy of the PSO-LSTM model.

As a preferred embodiment, preferably, the evaluation index includes: the root mean square error RMSE, the mean absolute error MAE, the mean square error MSE and the mean absolute percentage error MAPE have the following calculation formulas:

wherein T is _n Days s representing prediction of surrounding rock deformation _i Representing the actual value of the deformation of the surrounding rock,the predicted values representing the deformation of the surrounding rock are MAE, MSE, RMSE, MAPE representing the average absolute error, the mean square error, the root mean square error and the average absolute percentage error of the prediction model respectively.

By adopting the technical scheme, compared with the prior art, the invention has the beneficial effects that: according to the tunnel surrounding rock deformation prediction method based on the PSO-LSTM hybrid model, the long-short-term neural network (LSTM) is established to process and train actual monitoring data, and the particle swarm algorithm is combined to optimize the learning rate and the number of hidden layer neurons, so that the aims of reducing sample training time and improving surrounding rock deformation prediction accuracy can be achieved. The method can provide theoretical support for optimization of tunnel supporting scheme, tunnel safety construction and operation. The surrounding rock deformation result predicted by the method has high accuracy and high efficiency, and has wide application prospect in the construction, operation and maintenance of a large number of tunnel engineering in the future.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of the unit structure of the LSTM model according to the scheme of the present invention;

FIG. 2 is a schematic diagram of a training flow of a PSO-LSTM tunnel surrounding rock deformation prediction model according to the scheme of the invention;

FIG. 3 is a schematic view of an example field construction of an embodiment of the present invention, showing shallow bias four-hole small clear-distance tunnel portal engineering;

FIG. 4 is a graph of predicted results for different proportions of data in an example embodiment of the present invention;

fig. 5 is a graph of predicted surrounding rock deformation results in an example of the scheme of the invention.

Detailed Description

The invention is described in further detail below with reference to the drawings and examples. It is specifically noted that the following examples are only for illustrating the present invention, but do not limit the scope of the present invention. Likewise, the following examples are only some, but not all, of the examples of the present invention, and all other examples, which a person of ordinary skill in the art would obtain without making any inventive effort, are within the scope of the present invention.

Referring to fig. 2, the tunnel surrounding rock deformation prediction method based on the PSO-LSTM model according to this embodiment includes:

s01, obtaining deformation data of tunnel vault settlement and horizontal convergence and time points obtained by the deformation data to form time series data;

s02, preprocessing time sequence data;

s03, dividing the preprocessed time series data into a training set and a testing set according to a preset proportion;

s04, constructing an LSTM model (shown in figure 1) for arch top settlement and horizontal convergence prediction in tunnel surrounding rock deformation, and carrying out parameter optimization on the LSTM model through a PSO algorithm;

s05, setting parameters of an LSTM model, training a vault subsidence and horizontal convergence prediction model based on the LSTM by utilizing a training set, and repeatedly training the LSTM by adopting a particle swarm parameter optimization model to ensure that the prediction accuracy of the model meets the preset requirement and obtain a PSO-LSTM model;

s06, inputting the test set into the PSO-LSTM model for prediction, and outputting a predicted value.

In the present solution S01, the manner of acquiring deformation data of settlement and horizontal convergence of the tunnel vault and the time point acquired by the deformation data may be in-situ monitoring acquisition or numerical simulation acquisition.

In addition, the scheme S01 further includes: and drawing a vault settlement-time and horizontal convergence-time deformation curve related to time according to the deformation data of the tunnel vault settlement and horizontal convergence, and performing preliminary analysis and evaluation according to the change rule of the deformation curve and the deformation rule of the tunnel surrounding rock.

In the scheme S02, preprocessing the time series data comprises more than one of data cleaning, missing value filling and normalization processing; after normalization processing is carried out on the data, when the predicted value is obtained in the following S06, the predicted value is timely and inversely normalized according to the requirement.

The data cleaning comprises abnormal data elimination of time series data;

the missing value filling comprises the steps of fitting a curve formula according to the time series data, and then obtaining a corresponding time series data value according to the curve formula so as to complete the supplement of the deformation data.

In this scheme S03, 80% of the data amount is extracted from the preprocessed time-series data as a training set, and the remaining 20% is used as a test set, and the former 80% is used as the training set and the latter 20% is used as the test set.

In the scheme S04, parameter optimization is performed on the LSTM model by using a PSO algorithm, which includes setting one or more parameters of population size, iteration number, and inertial weight, so as to maximize the prediction accuracy of the model.

In this scheme S05, setting parameters of the LSTM model includes setting one or more of a hidden layer size and a learning rate parameter.

In the scheme S06, the test set is input into the PSO-LSTM model for prediction, and for the input data at each time, the PSO-LSTM model generates a predicted value according to the previous state and input, and uses the predicted value as the input value at the next time.

The scheme S04 carries out parameter optimization on the LSTM model through a PSO algorithm, and comprises the following steps:

assuming that in a D-dimensional optimization space there are N particles that make up a population, the particles have a position vector and a velocity vector, comprising:

the position coordinates of the i-th particle are noted as: x is X _i ＝(x _i1 ,x _i2 ,...x _iD ),i＝1,2,...N；

The velocity vector of the i-th particle is noted as: v (V) _i ＝(v _i1 ,v _i2 ,...v _iD ),i＝1,2,...N；

Iterating the formula of the m-th generation to the m+1th generation particle update:

v _ij (m+1)＝ωv _ij (m)+c ₁ r ₁ (m)[p _ij (m)-x _ij (m)]+c ₂ r ₂ (m)[p _gj (m)-x _gj (m)]

x _ij (m+1)＝x _ij (m)+v _ij (m+1)

wherein ω is inertial weight, c ₁ ，c ₂ To learn factors, p _i For the individual extremum, p _g Is globally optimal.

In the scheme, step S05 training a vault subsidence and horizontal convergence prediction model based on LSTM comprises the following steps:

constructing a forgetting door control unit f _n ：f _n ＝σ(W _f h _n-1 +U _f x _n +b _f )；

Building an update door control unit: i.e _n ＝σ(W _i h _n-1 +U _i x _n +b _i )；

C′ _n ＝tanh(W _c h _n-1 +U _c x _n +b _c )；

Building an output door control unit: o (O) _n ＝σ(W _o h _n-1 +U _o x _n +b _o )；

Output state results: h is a _n ＝tanh×O _n ；

C _n ＝f _n ×C _n-1 +i _n ×C′ _n

Wherein W is _f Weight of short-term memory state of forgetting gate, U _f B, inputting the weight of the vector for the forgetting gate _f The bias parameters are the bias parameters of the forgetting door; w (W) _i Is i _n Weights of short-term memory states of function, W _c Is C' _n Weights of short-term memory states of function, U _i Is i _n Weights of input vectors, U _c Is C' _n Weights of input vector, b _i To update gate i _n Bias parameters of the function, b _c To update door C' _n Bias parameters of the function.

In order to improve the prediction reliability of the model, the scheme further comprises:

s07, carrying out prediction result evaluation on the PSO-LSTM model according to a preset evaluation index, and optimizing the PSO-LSTM model according to the evaluation result so as to further improve the prediction accuracy of the PSO-LSTM model.

Preferably, as a preferred embodiment, the evaluation index includes: the root mean square error RMSE, the mean absolute error MAE, the mean square error MSE and the mean absolute percentage error MAPE have the following calculation formulas:

wherein T is _n Days s representing prediction of surrounding rock deformation _i Representing the actual value of the deformation of the surrounding rock,the predicted values representing the deformation of the surrounding rock are MAE, MSE, RMSE, MAPE representing the average absolute error, the mean square error, the root mean square error and the average absolute percentage error of the prediction model respectively.

Examples of the embodiments

To further illustrate the methods disclosed in the above embodiments, an example is provided below for illustration:

the shallow buried bias tunnel of a certain municipal road is a four-hole small-clear-distance large-span tunnel engineering (shown in figure 3). The tunnel consists of a main line tunnel with two bidirectional eight lanes in the middle and two pedestrian tunnels on two sides, and the total length of the tunnel is about 1000m. The topography of the tunnel portal section is greatly fluctuated, the slope surface of the tunnel portal section is mostly strong weathered granite, joint cracks develop, surrounding rock stability is poor, and excavation disturbance is easy to generate large deformation and collapse. The tunnel has the characteristics of complex tunnel engineering structure, shallow buried bias voltage, large span, complex geological conditions and the like, and is an all-line control engineering. The method comprehensively acquires the deformation data of the surrounding rock of the tunnel, analyzes the deformation characteristics, and provides technical support for construction safety, support scheme optimization and long-term stability evaluation of the structure, which is an important research subject of the engineering. And obtaining on-site monitoring data such as vault settlement, horizontal convergence and the like of typical sections of four tunnel portal sections, wherein part of the data are shown in tables 1 and 2.

TABLE 1 vault settlement actual measurement (section)

Table 2 horizontal convergence monitoring data (section)

And carrying out normalization processing on arch top settlement and horizontal convergence by using MATLAB, and then distributing surrounding rock time-deformation data through the proportion of a training set and a testing set. Then, the preprocessed data are substituted into a PSO algorithm, the data are initialized, the fitness of the individual is calculated, and the optimization aim is to select the individual with high fitness. If the fitness is not the highest, returning to the second step to continue updating iteration until the fitness is the highest and unchanged, and the function value reaches the optimal value. And substituting the optimized result into the LSTM for learning and memorizing, outputting a predicted result when the set target times are reached, returning to continue learning and calculating when the set target times are not reached, and finally obtaining a predicted result of surrounding rock deformation. Dome changes were predicted for the following time days using 35%, 50%, 70% and 90% of the time data as training sets, respectively (as shown in fig. 4).

Taking an LSTM model as a control model, further calculating the prediction results of the two models (LSTM and PSO-LSTM) to evaluate the evaluation indexes of the prediction results, and evaluating the prediction effects as shown in the following table 3:

table 3 predictive evaluation under different ratio data

In summary, by combining fig. 5, it can be obtained that the errors of the vault subsidence, the horizontal convergence predicted value and the actual value of each tunnel are all within the allowable error range, which indicates that the prediction model has a reliable prediction effect. By training and optimizing the PSO-LSTM model, a relatively accurate surrounding rock deformation prediction result is obtained, and reliable support is provided for construction safety. The method can judge the stability trend of the tunnel in the early scheme selection, and can also judge the change trend of the next moment by utilizing the field monitoring data, thereby protecting the construction safety and the operation safety of the tunnel.

The foregoing description is only a partial embodiment of the present invention, and is not intended to limit the scope of the present invention, and all equivalent devices or equivalent processes using the descriptions and the drawings of the present invention or directly or indirectly applied to other related technical fields are included in the scope of the present invention.

Claims (10)

1. The tunnel surrounding rock deformation prediction method based on the PSO-LSTM model is characterized by comprising the following steps of:

s01, obtaining deformation data of tunnel vault settlement and horizontal convergence and time points obtained by the deformation data to form time series data;

s02, preprocessing time sequence data;

s03, dividing the preprocessed time series data into a training set and a testing set according to a preset proportion;

s04, constructing an LSTM model for arch top settlement and horizontal convergence prediction in tunnel surrounding rock deformation, and carrying out parameter optimization on the LSTM model through a PSO algorithm;

s05, setting parameters of an LSTM model, training a vault subsidence and horizontal convergence prediction model based on the LSTM by utilizing a training set, and repeatedly training the LSTM by adopting a particle swarm parameter optimization model to ensure that the prediction accuracy of the model meets the preset requirement and obtain a PSO-LSTM model;

s06, inputting the test set into the PSO-LSTM model for prediction, and outputting a predicted value.

2. The method for predicting tunnel surrounding rock deformation based on the PSO-LSTM model as set forth in claim 1, wherein S01 further includes: and drawing a vault settlement-time and horizontal convergence-time deformation curve related to time according to the deformation data of the tunnel vault settlement and horizontal convergence, and performing preliminary analysis and evaluation according to the change rule of the deformation curve and the deformation rule of the tunnel surrounding rock.

3. The tunnel surrounding rock deformation prediction method based on the PSO-LSTM model as claimed in claim 1, wherein in S02, preprocessing the time series data comprises more than one of data cleaning, missing value filling and normalization processing;

the data cleaning comprises abnormal data elimination of time series data;

the missing value filling comprises the steps of fitting a curve formula according to the time series data, and then obtaining a corresponding time series data value according to the curve formula so as to complete the supplement of the deformation data.

4. The method for predicting tunnel surrounding rock deformation based on the PSO-LSTM model as claimed in claim 1, wherein in S03, 80% of data amount is extracted from the preprocessed time series data as a training set, and the remaining 20% is used as a test set.

5. The tunnel surrounding rock deformation prediction method based on the PSO-LSTM model as claimed in claim 1, wherein in S04, parameter optimization is performed on the LSTM model by a PSO algorithm, and the method comprises the steps of setting more than one parameter of population size, iteration number and inertia weight so as to maximize the prediction accuracy of the model;

in S05, setting parameters of the LSTM model includes setting one or more of a hidden layer size and a learning rate parameter.

6. The method for predicting tunnel surrounding rock deformation based on the PSO-LSTM model of claim 1, wherein in S06, the test set is input into the PSO-LSTM model to perform prediction, and for the input data at each moment, the PSO-LSTM model generates a predicted value according to the previous state and input, and uses the predicted value as the input value at the next moment.

7. The method for predicting tunnel surrounding rock deformation based on the PSO-LSTM model as set forth in claim 1, wherein the step S04 of performing parameter optimization on the LSTM model by the PSO algorithm comprises the steps of:

assuming that in a D-dimensional optimization space there are N particles that make up a population, the particles have a position vector and a velocity vector, comprising:

the position coordinates of the i-th particle are noted as: x is X _i ＝(x _i1 ,x _i2 ,...x _iD ),i＝1,2,...N；

The velocity vector of the i-th particle is noted as: v (V) _i ＝(v _i1 ,v _i2 ,...v _iD ),i＝1,2,...N；

Iterating the formula of the m-th generation to the m+1th generation particle update:

v _ij (m+1)＝ωv _ij (m)+c ₁ r ₁ (m)[p _ij (m)-x _ij (m)]+c ₂ r ₂ (m)[p _gj (m)-x _gj (m)]

x _ij (m+1)＝x _ij (m)+v _ij (m+1)

wherein ω is inertial weight, c ₁ ，c ₂ To learn factors, p _i For the individual extremum, p _g Is globally optimal.

8. The method for predicting tunnel surrounding rock deformation based on the PSO-LSTM model as claimed in claim 1, wherein training the LSTM based dome subsidence and horizontal convergence prediction model in step S05 includes:

constructing a forgetting door control unit f _n ：f _n ＝σ(W _f h _n-1 +U _f x _n +b _f )；

Building an update door control unit: i.e _n ＝a(W _i h _n-1 +U _i x _n +b _i )；

C′ _n ＝tanh(W _c h _n-1 +U _c x _n +b _c )；

Building an output door control unit: o (O) _n ＝σ(W _o h _n-1 +U _o x _n +b _o )；

Output state results: h is a _n ＝tanh×O _n ；

C _n ＝f _n ×C _n-1 +i _n ×C′ _n

Wherein W is _f Weight of short-term memory state of forgetting gate, U _f B, inputting the weight of the vector for the forgetting gate _f The bias parameters are the bias parameters of the forgetting door; w (W) _i Is i _n Weights of short-term memory states of function, W _c Is C' _n Weights of short-term memory states of function, U _i Is i _n Weights of input vectors, U _c Is C' _n Weights of input vector, b _i To update gate i _n Bias parameters of the function, b _c To update door C' _n Bias parameters of the function.

9. Tunnel surrounding rock deformation prediction method based on PSO-LSTM model according to one of claims 1 to 8, characterized in that it further comprises:

s07, carrying out prediction result evaluation on the PSO-LSTM model according to a preset evaluation index, and optimizing the PSO-LSTM model according to the evaluation result so as to further improve the prediction accuracy of the PSO-LSTM model.

10. The method for predicting tunnel surrounding rock deformation based on the PSO-LSTM model of claim 9, wherein the evaluation index includes: the root mean square error RMSE, the mean absolute error MAE, the mean square error MSE and the mean absolute percentage error MAPE have the following calculation formulas:

wherein T is _n Days s representing prediction of surrounding rock deformation _i Representing the actual value of the deformation of the surrounding rock,the predicted values representing the deformation of the surrounding rock are MAE, MSE, RMSE, MAPE representing the average absolute error, the mean square error, the root mean square error and the average absolute percentage error of the prediction model respectively.