Shield tunneling parameter optimization method based on PSO algorithm

Abstract

The invention provides a shield tunneling parameter optimization method based on a PSO algorithm. The shield tunneling parameter optimization method comprises the steps of obtaining original tunneling data; data preprocessing: removing parameters of the shield tunneling machine in a non-tunneling state and a shutdown restart state, and removing data abnormal points; dividing geological conditions; building a prediction model: building a BP neural network model, building a corresponding mapping formula, building a relation between the total thrust and the cutterhead rotating speed and the tunneling speed and the cutterhead torque, and determining a fitness function; and a prediction stage: inputting the BP neural network model into a PSO particle swarm optimization algorithm, verifying the optimal solutions of the cutterhead rotating speed and the total thrust obtained by training, and outputting the shield active operation parameters, namely the cutterhead rotating speed and the total thrust, obtained by taking the lowest shield energy consumption and the shortest construction period as targets. According to the method, shield tunneling parameters in different stratums can be optimized, the purposes of reducing shield energy consumption and improving the tunneling speed are achieved, and therefore the construction cost is saved, and the construction period is shortened.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a method for optimizing shield tunneling parameters based on the PSO algorithm. Background Technique

[0002] With the rapid development of underground traffic engineering construction in China, the use of large-diameter slurry shield construction has become a common construction method. In response to the country's goal of "achieving carbon peak and carbon neutrality as soon as possible", reducing project costs and enabling the tunnel to be put into use earlier, reducing shield tunneling energy consumption and shortening the construction period of the shield have become the two most concerned goals of relevant parties. During the shield construction process, the setting of shield tunneling parameters will have a significant impact on the shield tunneling effect. Among them, the cutterhead rotation speed and total thrust of the shield machine are manually controlled by the shield machine driver and are the main research objects of shield tunneling parameter prediction.

[0003] In current prediction technologies, there are BP neural network prediction models that can output the total thrust of the shield, random forest prediction models, prediction models of the GA-BPNN algorithm, etc., and there are also prediction models that use convolutional neural networks and long short-term memory networks to predict the cutterhead rotation speed. However, most of the existing research and inventions only focus on a single tunnel, and their prediction results may not be applicable to other tunnels. There is less research based on a sample set formed by aggregating data from multiple tunnels. The general distribution rules and ranges of the shield cutterhead rotation speed and total thrust are not clear, and there is less research on providing corresponding operation suggestions for the shield machine driver based on the energy consumption and construction period of the shield machine, making it difficult to achieve such goals. Summary of the Invention

[0004] Inventive Purpose: To propose a method for optimizing shield tunneling parameters based on the PSO algorithm to solve the above problems existing in the above-mentioned prior art.

[0005] Technical Solution: A method for optimizing shield tunneling parameters based on the PSO algorithm includes:

[0006] Step S1, obtaining original tunneling data: collecting construction data of no less than 8 shield tunnels, and this data includes various stratum conditions as the original tunneling data;

[0007] Step S2, data preprocessing: removing the parameters of the shield machine in the non-tunneling state and the restart state after shutdown from the original tunneling data obtained in Step S1, selecting the first 10% of each ring of tunneling state as the rising section, and the data of the last 90% as the stable section, removing the parameters of the rising section, using the Mahalanobis distance method to process the data to remove data outliers, normalizing the training set data to reduce the influence of differences in different parameter dimensions and scales, and then performing statistical analysis on the tunneling parameters;

[0008] Step S3, Geological condition division: Establish corresponding grading standards according to different soil categories. For composite soil layers, take corresponding score values as parameter values to obtain the value ranges of the cutter head rotation speed and total thrust of the shield machine in different soils.

[0009] Step S4, Establish a prediction model: Establish the relationship between the total thrust, cutter head rotation speed and tunneling speed, cutter head torque. Take the total thrust and cutter head rotation speed as the initial parameters, input the initial parameters as the training set into the artificial neural network model, and train the neural network to obtain the fitness formulas for the optimization of shield energy consumption and construction period optimization.

[0010] Step S5, Prediction stage: Use the training set in Step S4, load the PSO particle swarm optimization algorithm into the BP artificial neural network model, and predict the optimal hyperparameterized artificial neural network model; Take the minimum shield energy consumption and the shortest construction period as the decision-making objectives, and output the predicted cutter head rotation speed and total thrust respectively, to obtain the shield energy consumption and tunneling speed under two cutter head rotation speeds and total thrusts; After comparison, select a set of data that meets the engineering requirements.

[0011] The specific process includes:

[0012] Step S51, Establish an optimal hyperparameterized artificial neural network model for predicting the cutter head rotation speed and total thrust. The specific process is as follows:

[0013] Establish a BP artificial neural network model, through K-fold cross-validation, predict the cutter head rotation speed and total thrust of the shield machine for the next 2 to 3 rings, and then construct the corresponding mapping formula to obtain the relationship between the total thrust, cutter head rotation speed of the shield active operation parameters and the cutter head rotation speed, tunneling speed and cutter head torque of the shield passive parameters:

[0014] f v (x n ,x F ) = v;

[0015] f T (x n ,x F ) = T;

[0016] Where: x n is the cutter head rotation speed, x F is the total thrust, v is the tunneling speed, and T is the cutter head torque;

[0017] Step S52, Establish a PSO optimization model. The PSO particle swarm optimization is used to make decision optimization with the minimum shield energy consumption and the shortest construction period as the goals to obtain the optimal cutter head rotation speed and total thrust.

[0018] Step S521: Initialize the positions and velocities of the particle swarm. Initialize the velocity to 0 or randomly generate it within a certain range.

[0019] Step S522: Load the relationship model between the total thrust, cutterhead rotation speed, tunneling speed, and cutterhead torque, and input it into the PSO optimization model.

[0020] Step S523: Calculate the fitness value of each particle using the derived fitness formula to measure the quality of the particle position. Update the historical best solution and global best solution of each particle, and update its velocity and position according to the historical best solution and global best solution of the particle.

[0021] Step S524: If the stopping condition is met, output the optimal solution; otherwise, continue to update the historical best solution and global best solution of each particle until the maximum number of iterations is reached.

[0022] Step S53: Establish a BP neural network. The specific process includes:

[0023] Step S531: Build the network structure of the BP neural network model with an input layer, an intermediate layer, a Dropout layer, and an output layer. Set the optimizer to Adam; the activation function of the input layer is the sigmod function.

[0024] Step S532: Set the number of iterations to n, and use the early stopping method to process the training process. When the error of the model on the validation set has not improved for 10 consecutive iterations, interrupt the training process.

[0025] Step S533: After determining the minimum mean square error of the validation set, the hyperparameters of the BP neural network model at this time can be determined.

[0026] According to the further improvement of the present invention, when constructing the BP neural network model, the sample set is divided into a training set and a test set in a ratio of 8:2. Select 1 / 10 of the data in the training set as the validation set to determine the relationship between pairwise data in the same group of parameters, determine the prediction model, and evaluate the error between the cutterhead speed and cutterhead torque of the prediction model and the actual values using the test set.

[0027] According to the further improvement of the present invention, step S2 specifically includes:

[0028] Step S21: Remove the data of the non-tunneling state of the shield machine from the original tunneling data obtained in step S1: Remove the data of the shield machine when it is stopped and assembling segments, and set a discriminant function: f i =F i ·v i ·T i ·n i ;

[0029] Wherein, F i is the total thrust at the i-th time step, with the unit of kN; v i is the tunneling speed at the i-th time step, with the unit of mm / min; T i is the cutterhead torque at the i-th time step, with the unit of MN·m; n i is the cutterhead rotation speed at the i-th time step, with the unit of rpm;

[0030] When any one of the total thrust, tunneling speed, cutterhead torque, and cutterhead rotation speed of the shield machine is 0, the discriminant function is 0 at this time, and it can be judged that the shield machine is in a non-tunneling state at this time, and the non-tunneling state data is removed;

[0031] Step S22, taking the stable stage, including:

[0032] Step S221, reordering the data with a duration of more than 30 minutes in the tunneling cycle section;

[0033] Step S222, arranging the data of each ring tunneling state in time, taking the first 10% of the data as the rising section, and the last 90% of the data as the stable section;

[0034] Step S23, removing abnormal data: using the Mahalanobis distance method to process abnormal data. The Mahalanobis distance method is a data similarity measurement method, which calculates the distance between data points based on the covariance matrix of the data. The specific calculation method is:

[0035]

[0036] Wherein, x i is a piece of data in the dataset, N is the total amount of data in the dataset, μ is the mean of the samples, and S is the covariance matrix of the data. Compared with other processing methods such as the box plot and the 3σ method;

[0037] Step 24, smoothing processing: using the moving average method to filter and denoise the digital signal, selecting an appropriate window size to move along the time axis of the signal, taking the average value of all samples within the window as the value of the sample at the center of the window, setting the window to 5, and performing smoothing processing on all shield tunneling parameters, while removing most of the noise and retaining the overall change trend during the entire shield tunnel tunneling process;

[0038] Step S25, normalization processing: before establishing the model, using the normalization method to process the training set data to reduce the influence of different parameter dimensions and scale differences, and improve the fitting accuracy and speed. The specific method is:

[0039]

[0040] In the formula, x norm is the normalized data, x is the original data, x max and x min are respectively the maximum and minimum values in the original data.

[0041] According to a further improvement of the present invention, the step S3 specifically includes:

[0042] Step S31: Score the soil layer according to the type of soil layer, plastic index and the proportion of particles with different particle sizes in the soil, and its value range is between 0-9 points to obtain a grading standard table;

[0043] Step S32: For the composite soil layer, select the weighted average value of the corresponding scores as the scoring value, and the calculation formula is as follows:

[0044]

[0045] In the formula, n is the total number of soils in each ring, S i is the proportion (%) of different types of soils in each ring of soil, d i is the score of each ring of soil;

[0046] Step S33: Divide different types of soils according to the data in the grading standard table in step S31, and obtain the value ranges of the cutter head rotation speed and total thrust of the shield machine in different types of soils.

[0047] According to a further improvement of the present invention, the step S4 specifically includes:

[0048] Step S41: Select the main parameters of shield construction and tunneling, calculate the Pearson correlation coefficient, Spearman correlation coefficient and feature importance based on random forest between the shield tunneling parameters pairwise, and draw the corresponding heat maps; for the feature selection based on the Pearson correlation coefficient and Spearman correlation coefficient, respectively find the parameters that have a certain correlation (|r|≥0.2) with the tunneling speed, cutter head rotation speed, total thrust and cutter head torque; for the feature selection based on the feature importance of random forest, select the shield tunneling parameters involved in the cumulative importance from high to low to 0.95; finally select the intersection of the features selected by the three methods;

[0049] Step S42: Use the BP neural network model to construct the relationship between the total thrust and cutter head rotation speed and the tunneling speed and cutter head torque through forward propagation, backward propagation and gradient descent;

[0050] Step S43: Use the grid search method to optimize the hyperparameters of the BP neural network model, specify the possible value ranges of the hyperparameters and the possible values of each parameter; traverse the possible combinations of hyperparameters; use K-fold cross-validation;

[0051] Among them, the process of K-fold cross-validation method includes:

[0052] Divide the data in the training set into K groups on average, randomly select 1 group of data as the validation set, and use the remaining K - 1 groups for training;

[0053] Repeat the above process K times, and the validation sets selected each time are different. Take the MSE of each validation set, and take the average value of K times as the final MSE.

[0054] According to the further improvement of the present invention, the shield parameter prediction process in step S5 is divided into two parts: the minimum energy consumption decision optimization of the shield and the shortest construction period decision optimization:

[0055] The minimum energy consumption decision optimization of the shield is specifically: the energy consumption of the shield refers to the work done per unit volume of soil excavation, and its calculation formula is:

[0056]

[0057] Among them, W T is the work done by the cutter head torque, n is the cutter head speed, T is the cutter head torque, t is the time, W F is the work done by the total thrust, F is the total thrust, and v is the excavation speed;

[0058] Substitute the above relational expressions into this formula to obtain the fitness function of the shield energy consumption:

[0059]

[0060] Apply the fitness function to the algorithm process to determine the historical optimal solution p best and the global optimal solution g best , set the relevant parameters of the particle swarm, where the inertia weight w = 0.9, the acceleration constant c 1 = c 2 = 0.5, the maximum particle speed V max = 1, the number of particles N = 100, the number of iterations G max = 100. After performing the optimization calculation, obtain the optimal cutter head speed n and the total thrust F;

[0061] The shortest construction period decision optimization is specifically: increasing the excavation speed is used to improve the excavation efficiency and shorten the construction period. The time consumed for shield excavation can be expressed, where in the formula, is the excavation distance, and v is the excavation speed; substituting the relational expression into it can obtain the fitness function as follows:

[0062]

[0063] In the formula, L is the shield excavation distance, set to 1m; t is the excavation time; f v(x i,n , x i,F ) is the shield tunneling speed mapping formula for the i-th ring;

[0064] Among them, the cutter head rotation speed x i,n ∈[n min , n max , and the total thrust x i,F ∈[F min , F max . When t takes a smaller value, a better fitness value is obtained. When the global optimal solution is obtained, that is, when the construction period is the shortest, the optimal parameter combination is obtained;

[0065] When using the PSO algorithm to optimize the shield operation, set each parameter. Among them, the inertia weight w = 1.5, the acceleration constant c 1 = c 2 = 0.5, the maximum particle speed V max = 1, the number of particles N = 200, and the number of iterations G max = 300. After performing the optimization calculation, the optimal cutter head rotation speed n and the total thrust F are obtained.

[0066] Beneficial effects: Combining the shield tunneling construction experience and the Pearson correlation coefficient, through data preprocessing, removing the rising section data and abnormal data in the steady section during the operation of the shield machine is beneficial to ensuring the rationality, scientificity, and accuracy of the data, reducing the data dimension, and improving the prediction efficiency of the model.

[0067] Adding a Dropout layer to the BP neural network to avoid problems such as model overfitting, gradient disappearance, and gradient explosion, so as to enhance the generalization ability of the model and improve the accuracy of the model.

[0068] Using the K-fold cross-validation method to perform cross-validation on the BP neural network model to ensure better evaluation index scores for the training and validation of the model in the case of less sample data.

[0069] Using the PSO optimization algorithm to optimize the BP neural network model, enabling the particles in the PSO optimization model to continuously update their own positions, moving the individual optimal solution and the historical optimal solution towards the target position. It requires fewer parameters, making the model relatively simple and easy to implement, helping the model to converge quickly and avoid falling into the local optimal solution.

[0070] The finally output parameters are fewer, which is beneficial for the shield machine driver to more intuitively adjust the driving parameters to achieve the goals of low shield energy consumption and short construction period. Brief Description of the Drawings

[0071] Figure 1 is the overall flowchart of the present invention.

[0072] Figure 2 It is a schematic diagram of the K-fold cross validation principle of the present invention.

[0073] Figure 3 It is a structural diagram of the BP neural network model of the present invention.

[0074] Figure 4 It is a heat map of the Pearson correlation coefficient of shield tunneling parameters of the present invention.

[0075] Figure 5 The Spearman correlation coefficient thermal diagram of shield tunneling parameters of the present invention

[0076] Figure 6 It is a flow chart of the PSO particle swarm optimization algorithm of the present invention.

[0077] Figure 7 The fitness curve of the PSO algorithm in the application case of the present invention. DETAILED DESCRIPTION

[0078] The technical solution of the present invention is further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0079] In order to make the technical problem of the present invention clearer, the following provides embodiments of the present invention and gives a more comprehensive description. However, the specific embodiments given here are only used to explain the present invention and are not used to limit the use environment and scope of the invention.

[0080] Embodiment 1:

[0081] See also Figure 1 This embodiment provides a shield key parameter optimization method based on the PSO algorithm, more specifically, a shield parameter optimization method based on the PSO particle swarm optimization algorithm with the goal of reducing shield energy consumption and shortening construction period, including the following steps:

[0082] Step S1, acquiring original excavation data: collecting construction data of no less than 8 shield tunnels, the data including various stratum conditions as original excavation data;

[0083] Step S2, data preprocessing: remove the parameters of the shield machine in the non-excavation state and the shutdown and restart state in the original excavation data obtained in step S1, select the first 10% of the excavation state of each ring as the rising section, and the last 90% of the data as the stable section, remove the parameters of the rising section, use the Mahalanobis distance method to process the data to remove data anomalies, normalize the training set data to reduce the difference in the dimension and scale of different parameters, and then perform statistical analysis on the excavation parameters;

[0084] Step S3, Geological condition division: Establish corresponding grading standards according to different soil categories. For composite soil layers, adopt corresponding score values as parameter values to obtain the value ranges of the cutter head rotation speed and total thrust of the shield machine in different soils.

[0085] Step S4, Establish a prediction model: Establish the relationship between the total thrust, cutter head rotation speed and tunneling speed, cutter head torque. Take the total thrust and cutter head rotation speed as initial parameters, input the initial parameters as a training set into the artificial neural network model, and train the neural network to obtain the fitness formulas for the optimization of shield energy consumption and construction period optimization.

[0086] Step S5, Prediction stage: Use the training set in Step S4, load the PSO particle swarm optimization algorithm into the BP artificial neural network model, and predict the optimized artificial neural network model with the optimal hyperparameters; Take the minimum shield energy consumption and the shortest construction period as the decision-making goals, and output the predicted cutter head rotation speed and total thrust respectively, and obtain the shield energy consumption and tunneling speed under two cutter head rotation speeds and total thrusts; After comparison, select a set of data that meets the engineering requirements.

[0087] The specific process includes:

[0088] Step S51, Establish an optimized artificial neural network model with optimal hyperparameters for predicting the cutter head rotation speed and total thrust. The specific process is as follows:

[0089] Establish a BP artificial neural network model, through K-fold cross-validation, predict the cutter head rotation speed and total thrust of the shield machine for the next 2 to 3 rings, and then construct the corresponding mapping formula to obtain the relationship between the total thrust and cutter head rotation speed of the shield active operation parameters and the tunneling speed and cutter head torque of the shield passive parameters:

[0090] f v (x n ,x F ) = v;

[0091] f T (x n ,x F ) = T;

[0092] In the formula, x n is the cutter head rotation speed; x F is the total thrust, v is the tunneling speed, and T is the cutter head torque;

[0093] Step S52, Establish a PSO optimization model. The PSO particle swarm optimization is used to make decision optimization with the goal of minimizing shield energy consumption and the shortest construction period to obtain the optimal cutter head rotation speed and total thrust.

[0094] Step S521: Initialize the positions and velocities of the particle swarm, initialize the velocity to 0 or randomly generate it within a certain range;

[0095] Step S522: Load the relationship model between the total thrust, cutterhead rotation speed, tunneling speed, and cutterhead torque, and input it into the PSO optimization model;

[0096] Step S523: Calculate the fitness value of each particle using the derived fitness formula to measure the quality of the particle position, update the historical best solution and global best solution of each particle, and update its velocity and position according to the historical best solution and global best solution of the particle;

[0097] Step S524: If the stop condition is met, output the optimal solution; otherwise, continue to update the historical best solution and global best solution of each particle until the maximum number of iterations is reached;

[0098] Step S53: Establish a BP neural network. The specific process includes:

[0099] Step S531: Build the network structure of the BP neural network model with an input layer, an intermediate layer, a Dropout layer, and an output layer, and set the optimizer to Adam; the activation function of the input layer is the sigmod function;

[0100] Step S532: Set the number of iterations to n, and use the early stopping method to process the training process. When the error of the model on the validation set has not improved for 10 consecutive iterations, interrupt the training process;

[0101] Step S533: After determining the minimum mean square error of the validation set, the hyperparameters of the BP neural network model at this time can be determined.

[0102] According to a further improvement of the present invention, when constructing the BP neural network model, the sample set is divided into a training set and a test set in a ratio of 8:2, and 1 / 10 of the data in the training set is selected as the validation set to determine the relationship between pairwise data in the same group of parameters, determine the prediction model, and evaluate the error between the cutterhead speed and cutterhead torque of the prediction model and the actual value using the test set.

[0103] According to a further improvement of the present invention, the specific steps of step S2 include:

[0104] Step S21: Remove the data of the shield machine in the non-tunneling state from the original tunneling data obtained in step S1: Remove the data when the shield machine is stopped and assembling segments, and set a discriminant function: f i =F i ·v i ·T i ·n i ;

[0105] In the formula, F i is the total thrust at the i-th time step, with the unit of kN; v i is the tunneling speed at the i-th time step, with the unit of mm / min; T i is the cutterhead torque at the i-th time step, with the unit of MN·m; n i is the cutterhead rotation speed at the i-th time step, with the unit of rpm;

[0106] When any one of the total thrust, tunneling speed, cutterhead torque and cutterhead rotation speed of the shield machine is 0, the discriminant function is 0 at this time, and it can be judged that the shield machine is in a non-tunneling state at this time, and the non-tunneling state data is removed;

[0107] Step S22, taking the steady stage, including:

[0108] Step S221, reordering the data with a duration of more than 30 minutes in the tunneling cycle section;

[0109] Step S222, arranging the data of each ring tunneling state in time, taking the first 10% of the data as the rising section, and the last 90% of the data as the steady section;

[0110] Step S23, removing abnormal data: using the Mahalanobis distance method to process abnormal data. The Mahalanobis distance method is a data similarity measurement method, which calculates the distance between data points based on the covariance matrix of the data. The specific calculation method is:

[0111]

[0112] In the formula, x i is a piece of data in the dataset, N is the total amount of data in the dataset, μ is the mean of the samples, and S is the covariance matrix of the data; compared with other processing methods such as the box plot and 3σ method, the data distribution after processing is basically the same as that of the box plot and 3σ method, but the amount of data after removing abnormal data can be larger, and the prediction effect is better;

[0113] Step S24, smoothing processing: using the moving average method to filter and denoise the digital signal, selecting an appropriate window size to move along the time axis of the signal, taking the average value of all samples in the window as the value of the sample at the center of the window, setting the window to 5, and performing smoothing processing on all shield tunneling parameters, while removing most of the noise and retaining the overall change trend during the entire shield tunnel tunneling process;

[0114] Step S25, normalization processing: Before establishing the model, the training set data is processed by the normalization method to reduce the influence of different parameter dimensions and scale differences, and improve the fitting accuracy and speed. The specific method is:

[0115]

[0116] Wherein, x norm is the normalized data, x is the original data, x max and x min are respectively the maximum and minimum values in the original data.

[0117] According to a further improvement of the present invention, the step S3 specifically includes:

[0118] Step S31: Assign scores to the soil layer according to the type of soil layer, plasticity index and the proportion of particles with different particle sizes in the soil, and the value range is between 0-9 points to obtain a grading standard table; the grading standard is as follows:

[0119] Table 1 Grading standard table

[0120]

[0121]

[0122] Step S32: For the composite soil layer, select the weighted average of the corresponding scores as the scoring value, and the calculation formula is as follows:

[0123]

[0124] Wherein, n is the total number of soils in each ring, S i is the proportion (%) of different types of soils in each ring of soil, d i is the score of each ring of soil;

[0125] Step S33: Divide different types of soils according to the data in the grading standard table in step S31, and obtain the value ranges of the cutter head speed and total thrust of the shield machine in different types of soils.

[0126] According to a further improvement of the present invention, the step S4 specifically includes:

[0127] Step S41: Calculate the Pearson correlation coefficient, Spearman correlation coefficient and feature importance based on random forest between the shield tunneling parameters pairwise, and draw the corresponding heat maps; for feature selection based on the Pearson correlation coefficient and Spearman correlation coefficient, respectively find the parameters that have a certain correlation (|r|≥0.2) with the tunneling speed, cutter head speed, total thrust and cutter head torque; for feature selection based on the feature importance of random forest, select the shield tunneling parameters involved when the importance accumulates from high to low to 0.95; finally, select the intersection of the features selected by the three methods.

[0128] The Pearson correlation coefficient is usually used to explore the linear relationship between two variables. Calculate the Pearson correlation coefficient pairwise for the shield tunneling parameters in tunnel engineering and draw the corresponding heat map. The darker the color of the heat map, the greater the correlation between the two data. For example, Figure 4 As shown, where r > 0 indicates a positive correlation between the data, r < 0 indicates a negative correlation between the data, r = 0 indicates that the two data are not correlated, r = 1 indicates that the two data are completely positively correlated, and r = -1 indicates that the two data are completely negatively correlated.

[0129] The calculation process of the Pearson correlation coefficient is as follows:

[0130] Calculate the covariance cov(X,Y) and standard deviation s of the two variables. The covariance cov(X,Y) is used to represent whether the change trends of the two variables are the same, and the standard deviation s is used to represent the degree of dispersion of a single variable. Substitute the covariance and standard deviation into the calculation formula of the correlation coefficient to obtain the calculation formula of the Pearson correlation coefficient:

[0131]

[0132] where, s X and s Y represent the standard deviations of X and Y respectively.

[0133] The Spearman correlation coefficient is a method to measure the correlation between two variables and is applicable to the situation where the data presents ordinal variables or non-linear relationships. Its calculation method is based on comparing the ranks (or grades) of two sets of data, thus avoiding the influence of the data distribution form and outliers on the correlation judgment.

[0134] The basic idea of the Spearman correlation coefficient is to convert the original data into rank data and calculate the correlation between the rank data. The rank refers to sorting the data from small to large and then replacing the original data with integers from 1 to n (n is the sample size), so that each data corresponds to a rank. By comparing the rank data, the correlation size between the two sets of data can be obtained.

[0135] The value range of the Spearman correlation coefficient is from -1 to 1. When the coefficient is 1, it means that the two sets of data are completely positively correlated; when it is -1, it means that the two sets of data are completely negatively correlated; and 0 means that the two sets of data have no correlation. For example, Figure 5As shown, when the absolute value of the Spearman correlation coefficient is ≥0.8, it indicates a very strong correlation between parameters; when its absolute value is ≥0.6 and <0.8, it indicates a strong correlation between parameters; when its absolute value is ≥0.4 and <0.6, it indicates a moderate correlation between parameters; when its absolute value is ≥0.2 and <0.4, it indicates a weak correlation between parameters; when its absolute value <0.2, it indicates a very weak correlation between parameters. In addition, the Spearman correlation coefficient also has invariance and monotonicity, that is, for the same set of data, no matter how the order is arranged, the calculated correlation coefficient is always the same; for two sets of data, if one set of data is greater (less) than the other set of data, then their correlation coefficients should also be greater (less) than the other set of data.

[0136] The calculation process of the Spearman correlation coefficient is as follows:

[0137] ① Sort the two sets of data respectively to obtain the ranks of the two sets of data;

[0138] ② Calculate the rank difference d of each data, that is, d = x - y;

[0139] In the formula: x and y are the ranks of the two sets of data respectively;

[0140] ③ Calculate the sum of squares of rank differences S, that is

[0141] ④ Calculate the Spearman correlation coefficient r according to the following formula s :

[0142] It should be noted that when there are cases where the ranks of the two sets of data are the same, their average rank needs to be used as the representative rank to calculate the rank difference.

[0143] The variable importance scoring method of random forest can be used for feature selection, feature engineering and model interpretation. Different from traditional feature selection methods, the method based on random forest can handle complex feature problems such as high dimensionality, non-linearity, and multicollinearity. At the same time, random forest can also explain the model through feature importance scoring to further understand the internal reasons for the model prediction results.

[0144] The variable importance scoring method based on random forest mainly evaluates the importance of features by calculating the average decrease in impurity of each feature on all decision trees in the random forest. In the random forest, each time a decision tree is split, an optimal feature is selected for splitting. Therefore, the importance of a feature can be measured by calculating the average value of the decrease in impurity brought by this feature when it is used for splitting on all decision trees.

[0145] After each decision tree in the random forest is constructed, the importance of variables can be evaluated using the splitting contribution of each variable in the decision tree. In classification problems, Gini Importance or MeanDecrease Impurity (MDI) can be used to measure the importance of variables; in regression problems, Mean DecreaseAccuracy (MDA) can be used to measure the importance of variables.

[0146] Gini importance measures the importance of a variable by calculating the number of times each variable is used as a splitting node. During the construction of each decision tree, the random forest algorithm selects the splitting method that maximizes the Gini coefficient based on different splitting variables and splitting points. Therefore, the variable that is used as a splitting node more frequently has a higher Gini importance. The calculation method of Gini importance is as follows:

[0147]

[0148] where M is the number of decision trees in the random forest, T m represents the m-th decision tree, v t is the variable used for splitting at the t-th node, j is the variable to be calculated, and Δi(t,m) is the decrease in impurity at node t. Therefore, the calculation method of Gini importance is mainly based on the number of times each variable is selected as a splitting node in the random forest for measurement.

[0149] And Mean Decrease in Impurity (MDI) measures the importance of a variable by calculating the average value of the decrease in node impurity caused by each variable in the random forest. The measurement method of node impurity can be the Gini coefficient, entropy, classification error rate, etc. The formula for calculating the mean decrease in impurity is as follows:

[0150]

[0151] where Δi(t,m) represents the decrease in impurity at node t in the m-th decision tree, and v t is the variable used for splitting node t. The MDI calculation method is mainly based on the average value of the decrease in node impurity caused by each variable in the random forest for measurement. Similar to Gini importance, the variable that is used more frequently to decrease node impurity has a higher MDI.

[0152] MDA is a method used to measure the importance of variables in regression problems. It is calculated by calculating the average impact of each feature on the model accuracy. Specifically, calculate the proportion of the decrease in model accuracy when a certain variable does not participate in modeling. The larger this proportion, the greater the contribution of the variable to the model accuracy and the higher its importance. The formula for calculating MDA is as follows:

[0153]

[0154] Among them, represents the model prediction result obtained by training on the j-th sample after the i-th feature is excluded, represents the average value of all model prediction results, and N represents the number of samples.

[0155] Based on the variable importance scoring method of random forest, explore the importance of shield tunneling parameters of Tunnel Projects 1-9 on tunneling speed, cutterhead rotation speed, total thrust, and cutterhead torque. The scores are shown in Table 2:

[0156] Table 2 Variable Importance Scoring Based on Random Forest

[0157]

[0158]

[0159] Finally, based on the feature selection of Pearson correlation coefficient and Spearman correlation coefficient, find the parameters that have a certain correlation (|r|≥0.2) with tunneling speed, cutterhead rotation speed, total thrust, and cutterhead torque respectively. For the feature selection based on the importance of random forest features, select the shield tunneling parameters involved in the cumulative importance from high to low to 0.95. Finally, select the intersection of the features selected by the three methods

[0160] Step S42: Use the BP neural network model to construct the relationship between total thrust and cutterhead rotation speed and tunneling speed and cutterhead torque through forward propagation, backward propagation, and gradient descent;

[0161] Step S43: Use the grid search method to optimize the hyperparameters of the BP neural network model, specify the possible value ranges of the hyperparameters and the possible values of each parameter; traverse the possible hyperparameter combinations; use K-fold cross-validation; in this embodiment, use the k-fold cross-validation method, where the value of k is 10, that is, the training set is evenly divided into 10 groups, randomly select 1 group of data as the validation set each time, use the remaining 9 groups of data for training, repeat the above training and validation process 10 times, and the validation sets used each time are different. Finally, take the average value of the MSE of the 10 validation sets as the final MSE. After determining the mean square error of the minimum validation set, the hyperparameters of the BP neural network model at this time can be determined, and further the relationship between total thrust and cutterhead rotation speed and tunneling speed and cutterhead torque can be determined.

[0162] Among them, the process of the K-fold cross-validation method includes:

[0163] Divide the data in the training set into K groups on average, randomly select 1 group of data as the validation set from them, and use the remaining K - 1 groups for training;

[0164] Repeat the above process K times, and each time the selected validation set is different. Take the MSE of each validation set, and take the average value of K times as the final MSE.

[0165] According to the further improvement of the present invention, the shield parameter prediction process in step S5 is divided into two parts: the minimum energy consumption decision optimization of the shield and the shortest construction period decision optimization: both parts use the BP neural network model loaded into the PSO algorithm, and use different fitness functions to make decisions and optimizations. In the structure of the BP neural network model, there is a total of 1 input layer (the activation function is sigmod), one hidden layer, one Dropout layer (Dropout = 0.5), and one output layer. Adding the Dropout layer aims to avoid problems such as overfitting, gradient disappearance, and gradient explosion of the model, enhance its generalization ability, the optimizer is set to adam, the learning rate is set to 0.001, Epoch is set to 5000, and the early stopping method is used to process the training process. If the error of the model on the validation set has not been improved for 20 consecutive epochs, the training process is interrupted to ensure its efficiency. After obtaining the mean square error of the validation set after the BP neural network is trained, the hyperparameters of the BP neural network model at this time can be determined. Load the BP neural network model into the PSO algorithm and perform PSO optimization with two different decision-making goals.

[0166] The minimum energy consumption decision optimization of the shield is specifically as follows: The energy consumption of the shield refers to the work done per unit volume of soil excavation, and its calculation formula is:

[0167]

[0168] Among them, W T is the work done by the cutterhead torque, n is the cutterhead rotation speed, T is the cutterhead torque, t is the time, W F is the work done by the total thrust, F is the total thrust, and v is the excavation speed;

[0169] Substitute the above relational expressions into this formula to obtain the fitness function of the shield energy consumption:

[0170]

[0171] In the formula, n is the rotation speed of the shield cutterhead, f T (x n , x F ) is the mapping formula of the shield cutterhead torque; F is the total thrust of the shield; R is the excavation radius of the shield cutterhead; f v (x n , x F ) is the mapping formula of the shield excavation speed;

[0172] Apply the fitness function to the algorithm process to determine the historical optimal solution p best and the global optimal solution g best , set the relevant parameters of the particle swarm, where the inertia weight w = 0.9, the acceleration constant c 1 = c 2 = 0.5, the maximum particle velocity V max = 1, the number of particles N = 100, the number of iterations G max = 100. After performing the optimization calculation, the optimal cutterhead rotation speed n and the total thrust F are obtained.

[0173] Generally speaking, the optimal solution can be basically found after 10 iterations. As Figure 7 shown, it should be noted that after optimizing the two tunneling parameters, the energy consumption of the shield should be minimized on the premise of ensuring that the tunneling speed is not too low. Otherwise, the tunneling efficiency will be low, resulting in a longer construction period, which is contrary to the original intention of optimizing the shield tunneling parameters. After the prediction, it should be compared with the average tunneling speed and the predicted value under different soil types in the corresponding situation. If there is no situation where the tunneling speed is too low, the predicted result is selected. In this embodiment, after PSO optimization, the predicted tunneling speed values are all not greater than the rated tunneling speed of the shield machine. Except for the sand layers of three tunnels, the output tunneling speeds are all greater than the average tunneling speed in the corresponding strata, meeting the requirements of shield tunneling efficiency and construction period.

[0174] The optimization of the shortest construction period decision is specifically as follows: increasing the tunneling speed is used to improve the tunneling efficiency and shorten the construction period. The tunneling time of the shield can be expressed by . In the formula, is the excavation distance, and v is the tunneling speed; substituting the relational expression into it, the following fitness function can be obtained:

[0175]

[0176] In the formula, L is the shield excavation distance, set to 1m; t is the tunneling time; f v (x i,n , x i,F ) is the shield tunneling speed mapping formula for the i-th ring;

[0177] Among them, the cutterhead rotation speed x i,n ∈[n min , n max , the total thrust x i,F ∈[F min , F max ; when t takes a smaller value, a better fitness value is obtained at this time. When the global optimal solution is obtained, that is, when the condition of the shortest construction period is met, the optimal parameter combination is obtained;

[0178] When using the PSO algorithm to optimize the shield operation, various parameters are set. Among them, the inertia weight w = 1.5, and the acceleration constants c 1 = c 2 = 0.5, the maximum particle velocity V max = 1, the number of particles N = 200, and the number of iterations G max = 300. After performing the optimization calculation, the optimal cutterhead rotation speed n and the total thrust F are obtained.

[0179] Generally speaking, in the same tunnel project, the larger the particle size, the smaller the cutterhead rotation speed and the greater the total thrust required. After obtaining the optimal parameters, a comparison is made with the mean value of the formation where the shield machine is located to determine whether the PSO optimization algorithm significantly increases the tunneling speed, thereby judging the effectiveness of the PSO algorithm. In this embodiment, after PSO optimization, except for the sand layer, the tunneling speed after optimization in the remaining formations is greater than the mean value of the tunneling speed before optimization. Although the minimum value of the tunneling speed in the sand layer is less than the mean value of the tunneling speed, the mean value of the tunneling speed is also relatively conservative within the range of the tunneling speed after optimization. Therefore, the method of using the PSO algorithm to optimize the operation parameters to shorten the construction period is reliable.

[0180] It should be noted that different from taking the shortest construction period as the decision-making goal, when taking the minimum energy consumption of the shield as the decision-making goal, it is not that the smaller the energy consumption, the better. The optimal cutterhead rotation speed and total thrust should be selected on the premise of ensuring its construction efficiency. If the cutterhead rotation speed or total thrust parameters do not conform to the engineering facts, they should be adjusted in time to ensure the normal progress of the construction. After optimizing the shield parameters with the two decision-making goals, the predicted values of the two sets of shield energy consumption and tunneling speed at this time are output, and the two given parameters are selected for the best, and the best cutterhead rotation speed and total thrust at this time are selected.

[0181] Example 2:

[0182] Using the PSO algorithm to optimize the operation parameters with the minimum energy consumption of the shield for Tunnel Projects 1 to 9 as the decision-making goal, the optimal solution is basically found after 10 iterations. There are significant differences in the fitness values of the shield energy consumption in the soil layer, sand layer, and gravel layer, and they increase with the increase of the particle size in the formation. For different shield machines and formations of Tunnel Projects 1 - 9, the optimal active parameters when the shield energy consumption is the smallest are shown in Table 3.

[0183] Table 3 Optimized operation parameter table for the minimum shield energy consumption in different formations

[0184]

[0185]

[0186] After optimizing the operation parameters, not only the lowest energy consumption of the shield needs to be considered, but also the tunneling speed cannot be too low. Otherwise, the tunneling efficiency will be low and the construction period will be too long, which will violate the original intention of optimizing the shield operation parameters. Through the BP neural network model, the output tunneling speed is obtained from the relational expressions of the total thrust, cutter head rotation speed and tunneling speed. After comparing with the average tunneling speed in the corresponding formation and optimizing through PSO, the tunneling speed is greater than the average tunneling speed in the corresponding formation. The comparison between the optimized tunneling speed and the average tunneling speed in its corresponding formation is shown in Table 4.

[0187] Table 4 Tunneling speed table after optimizing the operation parameters with the minimum shield energy consumption as the decision in different formations

[0188]

[0189] Example 3:

[0190] Based on the shortest construction period as the decision-making method, the PSO algorithm is used to optimize the shield operation parameters. The settings of each parameter of the PSO algorithm are as follows: the inertia weight ω is set to 1.5, and the acceleration constants c 1 and c 2 are set to 0.5, the maximum particle speed V max is set to 1, the number of particles N is 200, and the number of iterations G max is 300 times. In different formations, the optimal parameters for the shortest construction period are shown in Table 5.

[0191] Table 5 Table of operation parameters after optimizing the operation parameters with the shortest construction period as the decision in different formations

[0192]

[0193] After optimizing through the PSO algorithm, the tunneling speed increases significantly. After counting the average values of the tunneling speed before and after optimization of tunnels No. 1 to No. 9. Except for the sand layer, the lowest value of the optimized tunneling speed is greater than the average value of the tunneling speed before optimization. Although the lowest value of the tunneling speed in the sand layer after optimization is less than the average tunneling speed, the average tunneling speed is also conservative within the range of the optimized tunneling speed. Therefore, the PSO algorithm optimization model with the shortest construction period as the decision can significantly improve the tunneling speed and shorten the construction period. The comparison of the tunneling speed before and after optimization is shown in Table 6

[0194] Table 6 Comparison of tunneling speed before and after optimization in different formations

[0195]

[0196] Example 4:

[0197] It is an application case of an ongoing project for the optimization method of key parameters of shield tunneling based on the PSO algorithm.

[0198] In this embodiment, a shield tunnel project in a coastal area of China is taken as an example, and the shield tunnel data of rings 761-770 from February 27, 2023 to March 2, 2023 of the tunnel is selected for analysis. After preprocessing the shield data, there are a total of 12 tunneling segments. The formation where the shield is located is a gravel layer. The PSO algorithm is used to optimize the shield operation parameters with the minimum energy consumption of the shield and the shortest construction period as the decision-making objectives. The comparison table of operation parameters and tunneling speed before and after optimization is shown in Table 7.

[0199] Table 7 Comparison table of operation parameters and tunneling speed with the minimum energy consumption of the shield and the shortest construction period as the decision-making

[0200]

[0201] It can be seen from the comparison in Table 7 that after the operation parameters are optimized by the PSO algorithm, both the tunneling speed and the energy consumption of the shield are significantly improved. Among them, in the optimization with the minimum energy consumption of the shield as the decision-making objective, the tunneling speed after optimization is increased by 16.28% compared with the true mean value, and the energy consumption of the shield after optimization is reduced by 39.97% compared with the true mean value; in the optimization with the shortest construction period as the decision-making objective, the tunneling speed after optimization is increased by 24.58% compared with the true mean value, and the energy consumption of the shield after optimization is reduced by 21.91% compared with the true mean value, indicating that the method of optimizing the shield tunneling parameters based on the PSO algorithm is feasible.

[0202] In the comparison of the two sets of predicted values with the minimum energy consumption of the shield and the shortest construction period, it is found that there is not much difference in the tunneling speed between the two, but the predicted value of the energy consumption of the shield with the minimum energy consumption of the shield as the decision-making objective is much lower than the predicted value with the shortest construction period as the decision-making objective. If there are no strict requirements for the construction period, the operation parameters with the minimum energy consumption of the shield as the decision-making objective can be selected for tunneling. This approach can not only significantly reduce the energy consumption of the shield and save project costs, but also use a relatively high tunneling speed to shorten the overall construction period, which is an energy-efficient construction plan. It should be noted that the results obtained here are only for the above examples, and it does not mean that the optimization algorithm with the minimum energy consumption of the shield as the decision-making objective is superior to the optimization algorithm with the shortest construction period as the decision-making objective. In actual cases, it is required that users give full play to their subjective initiative and select the tunneling parameters that are more suitable for the actual requirements of the project after comparing the tunneling parameters obtained from the two different decision-making objectives.

[0203] All of the above are preferred examples of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. A shield tunneling parameter optimization method based on PSO algorithm, characterized in that: The steps include: Step S1, acquiring original excavation data: collecting construction data of no less than 8 shield tunnels, the data including various stratum conditions as original excavation data; Step S2, data preprocessing: remove the parameters of the shield machine in the non-excavation state and the shutdown and restart state in the original excavation data obtained in step S1, select the first 10% of the excavation state of each ring as the rising section, and the last 90% of the data as the stable section, remove the parameters of the rising section, use the Mahalanobis distance method to process the data to remove data anomalies, normalize the training set data to reduce the difference in the dimension and scale of different parameters, and then perform statistical analysis on the excavation parameters and organize them into time series data; Step S3, geological condition classification: establish corresponding classification standards according to different types of soil, take corresponding scores as parameter values ​​for composite soil layers, and obtain the value range of the cutter head speed and total thrust of the shield machine in different soils; Step S4, establishing a prediction model: establishing a relationship between total thrust and cutterhead speed and excavation speed and cutterhead torque, taking total thrust and cutterhead speed as initial parameters, inputting the initial parameters into an artificial neural network model as a training set, training the neural network, and obtaining a fitness formula for optimizing shield energy consumption and construction period; Step S5, prediction stage: using the training set in step S4, the PSO particle swarm optimization algorithm is loaded into the BP artificial neural network model, and the optimal hyperparameterized artificial neural network model is predicted; taking the minimum shield energy consumption and the shortest construction period as decision goals, the predicted cutterhead speed and total thrust are output respectively, and the shield energy consumption and excavation speed under two cutterhead speeds and total thrusts are obtained; after comparison, a set of data that meets the engineering requirements is selected; The specific process includes: Step S51: Establishing an optimal hyperparameterized artificial neural network model for predicting the cutter head rotation speed and total thrust. The specific process is as follows: A BP artificial neural network model was established, and the cutterhead speed and total thrust of the shield machine for the next 2 to 3 rings were predicted through K-fold cross validation. Then the corresponding mapping formula was constructed to obtain the relationship between the active operating parameters of the shield machine, total thrust and cutterhead speed, and the passive parameters of the shield machine, tunneling speed and cutterhead torque: f v (x n ,x F )=v; f T (x n ,x F )=T; Where: x n is the cutter head speed, x F is the total thrust, v is the excavation speed, and T is the cutter head torque; Step S52, establishing a PSO optimization model, wherein the PSO particle swarm optimization is used to optimize the decision making with the goal of minimizing shield energy consumption and shortening construction period, and obtain the optimal cutter head speed and total thrust; Step S521, initializing the position and speed of the particle group, initializing the speed to 0 or generating it randomly within a certain range; Step S522, loading the relationship model between total thrust and cutterhead rotation speed and tunneling speed and cutterhead torque, and inputting it into the PSO optimization model; Step S523: Calculate the fitness value of each particle using the derived fitness formula to measure the quality of the particle position, update the historical optimal solution and the global optimal solution of each particle, and update the speed and position of the particle according to the historical optimal solution and the global optimal solution; Step S524: if the stop condition is met, output the optimal solution; otherwise, continue to update the historical optimal solution and the global optimal solution of each particle until the maximum number of iterations is reached; Step S53: Establishing a BP neural network, the specific process includes: Step S531, using an input layer, an intermediate layer, a Dropout layer and an output layer to build a BP neural network model network structure, the optimizer is set to Adam; wherein the input layer activation function is the sigmoid function; Step S532, set the number of iterations to n, use the early stopping method to process the training process, and interrupt the training process when the error of the model on the validation set has not been improved for 10 consecutive iterations; Step S533: After the minimum validation set mean square error is determined, the hyperparameters of the BP neural network model at this time can be determined.

2. The shield tunneling parameter optimization method based on PSO algorithm according to claim 1 is characterized in that: When constructing the BP neural network model, the sample set is divided into a training set and a test set in a ratio of 8:

2. 1 / 10 of the data in the training set is selected as the validation set to determine the relationship between the data in the same group of parameters and to determine the prediction model. The test set is used to evaluate the error between the cutter head speed and cutter head torque of the prediction model and the actual value.

3. The shield tunneling parameter optimization method based on PSO algorithm according to claim 1 is characterized in that: The step S2 specifically includes: Step S21, removing the data of the shield machine in the non-drilling state from the original excavation data obtained in step S1: removing the data of the shield machine when it is stopped and assembled, and setting a discriminant function: f i =F i ·v i ·T i ·n i ; In the formula, F i is the total thrust at the i-th time step, in kN; v i is the tunneling speed at the i-th time step, in mm / min; T i is the cutter head torque at the i-th time step, in MN·m; n i is the cutter head speed at the i-th time step, in rpm; When any one of the total thrust, excavation speed, cutter head torque and cutter head speed of the shield machine is 0, the discriminant function is 0, and it can be judged that the shield machine is in a non-excavation state, and the non-excavation state data is removed; Step S22, taking a stable stage, includes: Step S221, select and re-sort the data with the duration of the excavation cycle segment being more than 30 minutes; Step S222, arrange the excavation status data of each ring by time, take the first 10% of the data as the rising section, and the last 90% of the data as the stable section; Step S23, removing abnormal data: using the Mahalanobis distance method to process the abnormal data. The Mahalanobis distance method is a data similarity measurement method, which is based on the covariance matrix of the data to calculate the distance between data points. The specific calculation method is: In the formula, x i is a piece of data in the data set, N is the total amount of data in the data set, μ is the mean of the sample, and S is the covariance matrix of the data. This method is compared with processing methods such as box plot and 3σ method; Step 24, smoothing: Use the moving average method to filter and denoise the digital signal, select an appropriate window size to move along the time axis of the signal, take the average value of all samples in the window as the value of the window center sample, set the window to 5, and smooth all shield tunneling parameters to remove most of the noise while retaining the overall change trend of the entire shield tunneling process; Step S25, normalization processing: Before building the model, the training set data is processed by normalization method to reduce the impact of different parameter dimensions and scale differences and improve the fitting accuracy and speed. The specific method is: In the formula, x norm is the normalized data, x is the original data, and x max and x min are the maximum and minimum values ​​in the original data respectively.

4. The shield tunneling parameter optimization method based on PSO algorithm according to claim 1 is characterized in that: The step S3 specifically includes: Step S31, assigning points to the soil layer according to the type of soil layer, plasticity index and proportion of particles of different sizes in the soil, with the value range being between 0 and 9 points, to obtain a grading standard table; Step S32: For the composite soil layer, the weighted average of the corresponding scores is selected as the scoring value, and the calculation formula is as follows: Where n is the total number of soils in each ring, S i is the proportion of different types of soil in each ring of soil, d i is the score of each ring of soil; Step S33, classify different types of soil according to the data in the classification standard table of step S31, and obtain the cutter head rotation speed and total thrust value range of the shield machine in different types of soil.

5. The shield tunneling parameter optimization method based on PSO algorithm according to claim 1 is characterized in that: The step S4 specifically includes: Step S41, respectively calculating the Pearson correlation coefficient, Spearman correlation coefficient and feature importance based on random forest between shield tunneling parameters, and drawing corresponding heat maps; for feature selection based on Pearson correlation coefficient and Spearman correlation coefficient, respectively find out parameters with certain correlation |r|≥0.2 with tunneling speed, cutterhead speed, total thrust and cutterhead torque; for feature selection based on random forest feature importance, select shield tunneling parameters involved in which importance is accumulated from high to low to 0.95; finally select the intersection of features selected by the three methods; Step S42, using the BP neural network model through forward propagation, back propagation and gradient descent to construct the relationship between the total thrust and the cutterhead speed and the tunneling speed and the cutterhead torque; Step S43, using a grid search method to optimize the hyperparameters of the BP neural network model, specifying the possible value range of the hyperparameters and the possible value of each parameter; traversing possible hyperparameter combinations; using K-fold cross validation; Among them, the process of K-fold cross validation method includes: Divide the data in the training set into K groups equally, randomly select one group of data as the validation set, and the remaining K-1 groups are used for training; The above process is repeated K times, and the validation set extracted each time is different. The MSE of each validation set is taken, and the average of K times is taken as the final MSE.

6. The shield tunneling parameter optimization method based on PSO algorithm according to claim 1 is characterized in that: The shield parameter prediction process in step S5 is divided into two parts: shield energy consumption minimum decision optimization and construction period shortest decision optimization: The shield energy consumption minimum decision optimization is as follows: Shield energy consumption refers to the work done by excavating a unit volume of soil, and its calculation formula is: Among them, W T is the work done by the cutter head torque, n is the cutter head speed, T is the cutter head torque, t is the time, W F is the work done by the total thrust, F is the total thrust, and v is the tunneling speed; Substituting the above relationship into this formula, we can get the fitness function of shield energy consumption: Where n is the shield cutter head speed, f T (x n ,x F ) is the shield cutter head torque mapping formula, F is the total thrust of the shield, R is the shield cutter head excavation radius, f v (x n ,x F ) is the shield tunneling speed mapping formula; Apply the fitness function to the algorithm process to determine the historical optimal solution p best and the global optimal solution g best , set the relevant parameters of the particle swarm, including inertia weight w = 0.9, acceleration constant c1 = c2 = 0.5, and maximum particle speed V max =1, number of particles N = 100, number of iterations G max =100, after optimization calculation, the optimal cutter head speed n and total thrust F are obtained; The optimization of the shortest construction period decision is as follows: increasing the excavation speed is used to improve the excavation efficiency, shortening the construction period, and the shield excavation time can be used Indicates that, is the excavation distance, v is the excavation speed; substituting the relationship into it, the fitness function can be obtained as follows: Where, L is the shield excavation distance, which is set to 1m; t is the excavation time; f v (x i,n ,x i,F ) is the shield tunneling speed mapping formula of the i-th ring; Among them, the cutter head speed x i,n ∈[n min ,n max ], total thrust x i,F ∈[F min ,F max ], when t takes a smaller value, a better fitness value is obtained, and when the global optimal solution is obtained, that is, when the condition of the shortest construction period is met, the optimal parameter combination is obtained; When using the PSO algorithm to optimize shield operation, various parameters are set, including the inertia weight w = 1.5, plus Speed ​​constant c1=c2=0.5, maximum particle speed V max =1, number of particles N = 200, number of iterations G max =300, After optimization calculation, the optimal cutter head speed n and total thrust F are obtained.

Images