Novel estimation method for floating amount of shield tunnel segment

Abstract

The invention discloses a novel estimation method for the floating amount of a shield tunnel segment, and belongs to the technical field of shield tunnel construction, and the method comprises the steps: S100, monitoring the whole floating process of the shield tunnel segment in real time based on a segment floating process monitoring system, recording and storing floating process monitoring data and two-dimensional image data at different moments, and carrying out the calculation of the floating amount of the shield tunnel segment; acquiring parameter data of a plurality of influence factors; s200, a data set is formed based on the collected final floating amount and the influence factor parameters, a segment floating amount prediction model is trained and constructed based on the data set, and the floating amount of the output segment reaching a stable state is predicted based on the currently collected influence factor parameter data; and S300, introducing a reinforcement learning algorithm, and based on the combination of the floating amount prediction value, the floating process monitoring data and the influence factor parameter data, automatically adjusting shield tunneling parameters in the influence factor parameters so as to minimize the segment floating amount. By introducing machine learning and reinforcement learning algorithms, accurate prediction and intelligent control of the floating amount of the duct piece are realized.

Description

Technical Field

[0001] The invention relates to the technical field of shield tunnel construction, and more specifically to a novel method for estimating the floating amount of a shield tunnel segment. Background Art

[0002] In soft soil areas, the segments always face the problem of floating after they are separated from the shield tail during shield tunnel construction. The uneven floating between the segments will lead to engineering problems such as segment damage, cracking, and water leakage. The objectively existing floating space of the segments is an external factor that aggravates the floating amount of the segments. The real internal factor that causes the segments to float is that the segments are in an unbalanced force state after they are separated from the shield tail. The upward external force makes the segments tend to float. Among them, the complex interaction between the grouting slurry, the tunnel segments, and the surrounding soil affects the stress state of the segments. The shield machine excavation speed, the soil disturbance range, the distribution pattern of the slurry, and the solidification time after mixing with the surrounding water and soil affect the floating stability time of the segments.

[0003] The existing patent document CN113255037A discloses a new method for estimating the floating amount of the dual-mode shield tunnel under soft upper and hard lower strata. The floating amount of the segment structure under the special condition can be obtained by superimposing the floating amount of the segment under the soft upper and hard lower strata and the floating amount of the segment structure caused by the slurry buoyancy. The former uses finite element software to establish a stratum-structure model, and analyzes the ratio of the elastic modulus of the upper and lower strata as a variable to obtain the relationship between the floating amount and the ratio of the elastic modulus of the upper and lower strata; the latter uses finite element software to establish a load-structure model, and analyzes the ratio of the upper and lower slurry pressures of the segment structure as a variable to obtain the relationship between the floating amount and the upper and lower slurry pressure ratio. However, in actual engineering, the floating amount of the shield tunnel segment is affected by multiple factors such as tunnel geometry parameters, shield excavation parameters, geological condition parameters, etc. Different geological conditions and construction conditions will lead to different floating amounts of the segment. Therefore, how to accurately predict the position of the shield tunnel segment after the floating stability during the construction period and then judge the initial linear shape of the tunnel when it is completed is a challenging task.

[0004] Therefore, it is necessary to provide a new estimation method for the buoyancy of shield tunnel segments to overcome the shortcomings of existing estimation methods for the buoyancy of segments, and to adaptively adjust the working parameters of the shield machine according to the predicted value of the buoyancy of the segments, so as to minimize the buoyancy of the segments. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] The purpose of the present application is to provide a new method for estimating the buoyancy of shield tunnel segments, which solves the problem that the existing method for estimating the buoyancy of segments proposed in the above background technology uses the buoyancy of segments under the conditions of upper soft and lower hard strata and the buoyancy of the segment structure caused by slurry buoyancy to obtain the buoyancy by superimposing the two parts. The two parts determine the influencing factors of the buoyancy and the application scenarios have great limitations, and it is difficult to promote and apply to the technical problem of calculating the buoyancy of shield tunnel segments affected by multiple factors. By introducing machine learning and reinforcement learning algorithms, the geological characteristics, construction parameters and real-time monitoring data are organically combined to achieve accurate prediction and intelligent control of the buoyancy of the segments.

[0007] 2. Technical solution

[0008] Based on the above purpose, the present application provides a new method for estimating the floating amount of shield tunnel segments, comprising the following steps:

[0009] S100, based on the segment floating process monitoring system, real-time monitoring of the whole process of shield tunnel segment floating, recording and storing the floating process monitoring data and two-dimensional image data at different times, and obtaining parameter data of influencing factors including but not limited to shield tunneling parameters, tunnel geometric parameters and geological condition parameters;

[0010] S200, forming a data set based on the collected final buoyancy and influencing factor parameters, constructing a segment buoyancy prediction model based on the data set training, inputting the currently collected influencing factor parameter data into the trained segment buoyancy prediction model, and outputting a predicted value of the buoyancy when the segment floats to a stable state;

[0011] S300, introduce reinforcement learning algorithm, based on the combination of floating amount prediction value, floating process monitoring data and influencing factor parameter data, automatically adjust the shield tunneling parameters in the influencing factor parameters to minimize the floating amount of the pipe segment.

[0012] Preferably, the segment floating process monitoring system in step S100 includes a floating monitoring sensor module, a data acquisition and transmission module, and a data storage management module, wherein the floating monitoring sensor module includes a pressure difference static level, a stress sensor, and an image sensor, which respectively obtain floating process monitoring data including the floating amount of the segment floating process, stress changes, and a two-dimensional image of the tunnel cross-section.

[0013] Preferably, in step S200, the XGBoost algorithm in the nonlinear regression model is used to build a segment floating volume prediction model, the data set is divided into a training set and a validation set, the segment floating volume prediction model is trained based on the training set, the XGBoost regressor is adjusted according to the performance of the model on the validation set, and the model is debugged using cross-validation.

[0014] On the basis of the above technical solution, a new method for estimating the floating amount of shield tunnel segments is enriched and designed, specifically: after completing the predicted value of the floating amount of the segment in step S200, the following steps are performed:

[0015] S210, integrating the floating process monitoring data and the two-dimensional image collected by the segment floating process monitoring system to form a spatiotemporal data set;

[0016] S220. A prediction model for the pipe segment floating process is built based on a neural network. The neural network structure includes a CNN network layer, an LSTM layer and a deconvolution layer connected in sequence to predict the pipe segment floating process and obtain a time series of predicted values ​​of the pipe segment floating process.

[0017] Furthermore, on the basis of step S300, the predicted value of the floating process is added to the combination of the predicted value of the floating amount, the monitoring data of the floating process and the parameter data of the influencing factors, and the final floating amount and the floating process are comprehensively considered to automatically adjust the shield excavation parameters to minimize the floating amount of the pipe segment.

[0018] Preferably, the CNN-LSTM structure in the pipe segment floating process prediction model is used to extract the spatiotemporal features of the two-dimensional image, and the short-term prediction of the two-dimensional image is achieved through the deconvolution layer.

[0019] Further preferably, a convolutional neural network is used to extract the spatial features of the two-dimensional image, and the output results of the convolutional neural network are input into a long short-term memory network to capture time series features.

[0020] 3. Beneficial effects

[0021] The novel method for estimating the floating amount of shield tunnel segments in the present application has at least the following technical effects or advantages:

[0022] The shield tunnel segment buoyancy estimation method of the present invention organically combines geological characteristics, construction parameters and real-time monitoring data by introducing machine learning and reinforcement learning algorithms, so as to realize accurate prediction of segment buoyancy. In combination with the reinforcement learning algorithm, through continuous learning and optimization of construction parameters, it can automatically adjust the operation strategy of the shield machine under different geological conditions, realize intelligent control of segment buoyancy, thereby effectively reducing the risk of buoyancy and improving construction safety and efficiency.

[0023] The shield tunnel segment floating amount estimation method of the present invention provides an integrated intelligent solution, which can dynamically adapt to complex environmental changes during the construction process, reduce unnecessary construction interference and resource consumption, and ultimately improve the overall controllability and economy of the project. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic diagram of the floating phenomenon of shield tunnel segments;

[0025] Figure 2 It is a schematic diagram showing the development of the floating volume of the shield tunnel segment over time after it is separated from the shield tail;

[0026] Figure 3 It is a schematic flow chart of a novel method for estimating the floating amount of shield tunnel segments in Example 1 of the present application;

[0027] Figure 4 It is a flow chart of a new method for estimating the floating amount of shield tunnel segments in Example 2 of the present application. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0029] Example 1

[0030] For most large-diameter shield tunnel projects, especially those crossing ground water systems, the difficulty and risk of shield construction are greatly increased due to the large excavation section and complex terrain. The most prominent problem is the floating of the shield tail segment. Figure 1 and Figure 2 As shown in the figure, the floating of the tail segment refers to the vertical movement of the segment caused by the combined action of the external space and its own force after the segment is separated from the tail. From the analysis of the segment floating mechanism, the physical quantity that directly causes the segment to float should be the external forces on the segment such as buoyancy, gravity, and thrust component. From the analysis of the segment floating process, the process from the completion of assembly to the stable floating of the segment can be roughly divided into the following four stages: 1) Initial floating stage, corresponding to the time from the completion of assembly to the separation from the tail; 2) Rapid floating stage, corresponding to 5 to 10 hours after separation from the tail, the floating amount accounts for about 70% to 90% of the final floating amount; 3) Slow floating stage, generally lasting 10 to 20 hours, the floating displacement accounts for 10% to 30% of the final floating amount; 4) The stable floating stage, the segment position remains stable or has a small fluctuation within 1 to 5 mm. It is not difficult to see that the floating process of the segment is a nonlinear process, and its floating amount first increases rapidly and then tends to be stable after a period of time.

[0031] The above content is the general law of the pipe segment floating process. The floating amount of the pipe segment under different environmental conditions and construction conditions is different. Therefore, it is necessary to accurately predict the position of the shield tunnel pipe segment after it floats and stabilizes during the construction period, and further adjust the operation strategy of the shield machine according to the predicted value of the pipe segment floating amount to effectively control the pipe segment floating amount. Based on this purpose, the embodiment of the present invention provides a new method for estimating the floating amount of the shield tunnel pipe segment, see Figure 3, including the following steps:

[0032] S100, based on the segment floating process monitoring system, real-time monitoring of the whole process of shield tunnel segment floating, recording and storing the floating process monitoring data and two-dimensional image data at different times, and obtaining parameter data of influencing factors including but not limited to shield tunneling parameters, tunnel geometric parameters and geological condition parameters;

[0033] S200, forming a data set based on the collected final buoyancy and influencing factor parameters, constructing a segment buoyancy prediction model based on the data set training, inputting the currently collected influencing factor parameter data into the trained segment buoyancy prediction model, and outputting a predicted value of the buoyancy when the segment floats to a stable state;

[0034] S300, introduce reinforcement learning algorithm, based on the combination of floating amount prediction value, floating process monitoring data and influencing factor parameter data, automatically adjust the shield tunneling parameters in the influencing factor parameters to minimize the floating amount of the pipe segment.

[0035] In this embodiment, the segment floating process monitoring system in step S100 includes a floating monitoring sensor module, a data acquisition and transmission module, and a data storage management module, wherein the floating monitoring sensor module includes a pressure difference static level, a stress sensor, and an image sensor, which respectively obtain floating process monitoring data including the floating amount of the segment floating process, stress changes, and a two-dimensional image of the tunnel cross-section.

[0036] In step S200, the XGBoost algorithm in the nonlinear regression model is used to build a segment floating amount prediction model. In shield tunnel construction, the XGBoost model can effectively capture the key factors affecting the floating of the segment and provide accurate floating amount prediction by analyzing multi-dimensional feature data. The input features of the XGBoost model include multiple influencing factor parameters collected during the construction process, including shield tunneling parameters, tunnel geometry parameters and geological condition parameters. Among them, the shield tunneling parameters are the tunneling parameters directly derived from the internal system of the shield machine, and the tunnel geometry parameters and geological condition parameters are obtained from the geological engineering report of the project issued by a third party. The shield tunneling parameters can be selected from jack thrust / kN, cutterhead torque / MN·m, cutterhead support pressure / kN, grouting pressure / bar, grouting volume / m 3, excavation speed / mm / min, for the same shield tunnel project, the tunnel geometry parameters only use the tunnel burial depth / m, and the geological condition parameters use cohesion / kPa, internal friction angle / °, compression modulus / MPa, groundwater level / m and formation permeability coefficient. The final buoyancy collected is combined with the above-mentioned influencing factor parameters to form a data set, which is divided into a training set and a validation set. The segment buoyancy prediction model is trained based on the training set, and it is continuously iterated to reduce the prediction error. The parameter settings for XGBoost model construction and training are as follows:

[0037] Learning rate (eta): controls the contribution of each tree, usually between [0.01, 0.3];

[0038] The number of trees (n_estimators): the number of weak learners, usually set between [100, 1000].

[0039] Maximum depth (max_depth): The maximum depth of each tree, which controls the model complexity.

[0040] Subsample ratio: To prevent overfitting, randomly select some samples for training.

[0041] Regularization parameters (alpha, lambda): used for L1 and L2 regularization.

[0042] Use the trained model to predict the floating amount of the test data set, adjust the parameters of the XGBoost regressor according to the performance of the model on the validation set, and use cross-validation to debug the model. And use the Bayesian optimization method to further optimize the model's hyperparameters to improve the prediction accuracy.

[0043] After completing the construction and training of the segment buoyancy prediction model using the XGBoost algorithm, the currently collected influencing factor parameter data is input into the trained segment buoyancy prediction model, and the predicted value of the buoyancy when the segment floats to a stable state is output.

[0044] After obtaining the predicted value of the floating amount in the future, step S300 is executed to introduce the reinforcement learning algorithm. In the reinforcement learning framework, the state space S is a combination of construction states used to describe the construction environment. For the construction parameter optimization of the shield tunnel, the state space can be defined as a multidimensional vector, covering the following information:

[0045] S={soil_cohesion,water_level,stresses_variation,permeability_coefficient,current_thrust_force,Support_pressure,predicted_uplift}

[0046] Among them, soil_cohesion is the soil cohesion, water_level is the groundwater level, stresses_variation is the segment stress variation, permeability_coefficient is the formation permeability coefficient, current_thrust_force is the current thrust of the shield machine, support_pressure is the cutterhead support pressure, and predicted_uplift is the predicted value of the final uplift.

[0047] Action space A is a combination of shield construction operations. For shield construction, the action space includes adjusting the following parameters:

[0048] A={thrust_adjustment, speed_adjustment, grouting_adjustment}

[0049] Among them, thrust_adjustment is to increase or decrease the thrust of the shield machine, speed_adjustment is to speed up or slow down the tunneling speed, and grouting_adjustment is to adjust the grouting volume and pressure;

[0050] The action space allows the model to flexibly adjust the parameter configuration of shield construction under different geological conditions to achieve the best construction effect.

[0051] In shield construction optimization, the factors comprehensively considered by the reward function include but are not limited to the control of floating volume, construction efficiency and safety.

[0052] The reward function is:

[0053] R(s,a)=-α×uplift_amount-β×energy_consumption+γ×stability_factor

[0054] Among them, uplift_amount is the uplift amount of the pipe segment measured in the current construction step, energy_consumption is the energy consumption of the shield machine operation, stability_factor is the safety stability coefficient during the construction process, α, β, γ are the reward coefficients of each part. By setting different reward coefficients (α, β, γ), the algorithm can adjust its emphasis on uplift control, energy consumption and stability to ensure the flexibility of the model in practical applications.

[0055] The deep Q network algorithm in deep reinforcement learning is selected for construction parameter optimization. DQN can combine deep learning technology to process high-dimensional state space and effectively learn the optimal strategy in the construction process. The state-action value, i.e., Q value, is estimated through the convolutional neural network. The Q value represents the state S t Next select Action A t The expected reward after the next state is calculated by using a fixed-size experience pool to store the four-tuple of past state, action, reward, and next state (S t ,A t ,R t ,S t+1 ), by randomly sampling from the experience pool, the stability and convergence of the algorithm are improved. Specifically, in each construction step, the current state S is observed t , including soil cohesion, groundwater level, stratum permeability, current thrust of the shield machine, cutterhead support pressure, and predicted value of the final floating amount. The ε-greedy strategy is used to select actions, that is, randomly select actions with ε probability, explore new strategies, and select the optimal action A estimated by the current Q network with 1-ε probability. t , perform strategy optimization and execute the selected action A t , get the next state S t+1 And the corresponding reward R t , and the quaternion (S t ,A t ,R t ,S t+1 ) is stored in the experience pool, a batch of experience is randomly sampled from the experience pool, and the objective function of DQN is used to update the Q network. The objective function is:

[0056] Q(S t ,A t )=R t +γ×max a′ Q(S t+1 ,a′).

[0057] In the above technical scheme, reinforcement learning can automatically adjust the shield tunneling parameters in the influencing factor parameters based on a combination of the floating amount prediction value, the floating process monitoring data and the influencing factor parameter data to minimize the floating amount of the pipe segment.

[0058] Example 2

[0059] Based on Example 1, this Example 2 provides a new method for estimating the floating amount of shield tunnel segments, referring to Figure 4 After the predicted value of the floating amount of the pipe segment is completed in step S200, the following steps are performed:

[0060] S210, integrating the floating process monitoring data and the two-dimensional image collected by the segment floating process monitoring system to form a spatiotemporal data set;

[0061] S220. A prediction model for the pipe segment floating process is built based on a neural network. The neural network structure includes a CNN network layer, an LSTM layer and a deconvolution layer connected in sequence to predict the pipe segment floating process and obtain a time series of predicted values ​​of the pipe segment floating process.

[0062] Then, step S300 is executed to introduce a reinforcement learning algorithm to automatically adjust the shield tunneling parameters in the influencing factor parameters based on a combination of the floating amount prediction value, the floating process prediction data, the floating process monitoring data and the influencing factor parameter data to minimize the floating amount of the pipe segment.

[0063] In this embodiment, after completing the prediction of the final buoyancy amount based on the XGBoost model, the monitoring data of the buoyancy amount and stress change of the segment buoyancy process obtained by the pressure difference static level, stress sensor and image sensor respectively are integrated with the two-dimensional image of the tunnel section after cleaning and preprocessing to obtain a spatiotemporal data set. A prediction model for the floating process of pipe segments is constructed using a neural network. The neural network structure includes a CNN network layer, an LSTM layer, and a deconvolution layer connected in sequence. The CNN network layer is used to extract the spatial features of a two-dimensional image. The processed two-dimensional image data is used as input. The convolution layer uses a convolution kernel to extract spatial features from the input data. The pooling layer downsamples the output of the convolution layer to reduce the dimension of the feature map. The output of the pooling layer is flattened and input to the fully connected layer to extract high-level features. Secondly, the high-level features extracted by the CNN network layer are used as input to the LSTM layer to capture time series features. Then, the hidden state output by the LSTM layer is input to the deconvolution layer. The deconvolution layer upsamples the input data to restore it to the original two-dimensional image size, and outputs a predicted two-dimensional image for a period of time in the future, representing the predicted floating process data. The prediction in this embodiment is a rolling prediction based on dynamic online monitoring data, that is, the short-term evolution of future pipe segment floating process data is predicted based on real-time monitoring data over a recent period of time.

[0064] Then, step S300 is executed. Step S300 in this embodiment 2 is based on step S300, adding the predicted value of the floating process of the pipe segment to the combination of the floating amount predicted value, the floating process monitoring data and the influencing factor parameter data, comprehensively considering the final floating amount and the floating process, and automatically adjusting the shield excavation parameters to minimize the floating amount of the pipe segment. The corresponding state space S in the reinforcement learning framework is:

[0065] S={soil_cohesion,water_level,stresses_variation,permeability_coefficient,current_thrust_force,Support_pressure,predicted_uplift,predicted_uplift_process}

[0066] Among them, soil_cohesion is soil cohesion, water_level is groundwater level, stresses_variation is segment stress variation, permeability_coefficient is formation permeability coefficient, current_thrust_force is the current thrust of the shield machine, support_pressure is the cutterhead support pressure, predicted_uplift is the predicted value of the final uplift, and predicted_uplift_process is the predicted value of the current uplift.

[0067] Action space A:

[0068] A={thrust_adjustment, speed_adjustment, grouting_adjustment}

[0069] Among them, thrust_adjustment is to increase or decrease the thrust of the shield machine, speed_adjustment is to speed up or slow down the excavation speed, and grouting_adjustment is to adjust the grouting volume and pressure.

[0070] Reward function R:

[0071] R(s,a)=-α×uplift_amount-β×energy_consumption+γ×stability_factor

[0072] Among them, uplift_amount is the uplift amount of the segment measured in the current construction step, energy_consumption is the energy consumption of the shield machine operation, stability_factor is the safety stability factor during the construction process, and α, β, and γ are the reward coefficients of each part.

[0073] In the technical solution of this embodiment, the CNN-LSTM structure in the pipe segment floating process prediction model is used to extract the spatiotemporal characteristics of the two-dimensional image, and the short-term prediction of the two-dimensional image is realized through the deconvolution layer. In addition, reinforcement learning is used to comprehensively consider the final floating amount and the floating process, and the shield excavation parameters are automatically adjusted to minimize the pipe segment floating amount.

[0074] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

Claims

1. A new method for estimating the floating volume of shield tunnel segments, characterized in that: The following steps are involved: S100, based on the segment floating process monitoring system, real-time monitoring of the whole process of shield tunnel segment floating, recording and storing the floating process monitoring data and two-dimensional image data at different times, and obtaining parameter data of influencing factors including but not limited to shield tunneling parameters, tunnel geometric parameters and geological condition parameters; S200, forming a data set based on the collected final buoyancy and influencing factor parameters, constructing a segment buoyancy prediction model based on the data set training, inputting the currently collected influencing factor parameter data into the trained segment buoyancy prediction model, and outputting a predicted value of the buoyancy when the segment floats to a stable state; S300, introduce reinforcement learning algorithm, based on the combination of floating amount prediction value, floating process monitoring data and influencing factor parameter data, automatically adjust the shield tunneling parameters in the influencing factor parameters to minimize the floating amount of the pipe segment.

2. The novel method for estimating the floating amount of shield tunnel segments according to claim 1 is characterized in that: The pipe segment floating process monitoring system in step S100 includes a floating monitoring sensor module, a data acquisition and transmission module, and a data storage management module, wherein the floating monitoring sensor module includes a pressure difference static level, a stress sensor, and an image sensor, which respectively obtain floating process monitoring data including the floating amount of the pipe segment floating process, stress changes, and a two-dimensional image of the tunnel cross-section.

3. The novel method for estimating the floating amount of shield tunnel segments according to claim 1 is characterized in that: In step S200, the XGBoost algorithm in the nonlinear regression model is used to build a segment floating volume prediction model, the data set is divided into a training set and a validation set, the segment floating volume prediction model is trained based on the training set, the XGBoost regressor is adjusted according to the performance of the model on the validation set, and the model is debugged using cross-validation.

4. The novel method for estimating the floating amount of shield tunnel segments according to claim 1 is characterized in that: The reinforcement learning algorithm in step S300 combines the construction states used to describe the construction environment into a state space S: S={soil_cohesion,water_level,stresses_variation,permeability_coefficient,current_thrust_force,Support_pressure,predicted_uplift} Among them, soil_cohesion is soil cohesion, water_level is groundwater level, stresses_variation is segment stress variation, permeability_coefficient is formation permeability coefficient, current_thrust_force is current thrust of shield machine, support_pressure is cutterhead support pressure, and predicted_uplift is the predicted value of final uplift; Action space A is the shield construction operation combination: A={thrust_adjustment, speed_adjustment, grouting_adjustment} Among them, thrust_adjustment is to increase or decrease the thrust of the shield machine, speed_adjustment is to speed up or slow down the tunneling speed, and grouting_adjustment is to adjust the grouting volume and pressure; The factors comprehensively considered in the reward function include but are not limited to the control of floating volume, construction efficiency and safety.

5. The novel method for estimating the floating amount of shield tunnel segments according to claim 4 is characterized in that: The reward function is: R(s,a)=-α×uplift_amount-β×energy_consumption+γ×stability_factor Among them, uplift_amount is the uplift amount of the pipe segment measured in the current construction step, energy_consumption is the energy consumption of the shield machine operation, stability_factor is the safety stability coefficient during the construction process, α, β, and γ are the reward coefficients of each part, which are used to adjust the emphasis on uplift control, energy consumption, and stability.

6. The novel method for estimating the floating amount of shield tunnel segments according to claim 4 is characterized in that: After the predicted value of the floating amount of the pipe segment is completed in step S200, the following steps are performed: S210, integrating the floating process monitoring data and the two-dimensional image collected by the segment floating process monitoring system to form a spatiotemporal data set; S220. A prediction model for the pipe segment floating process is built based on a neural network. The neural network structure includes a CNN network layer, an LSTM layer and a deconvolution layer connected in sequence to predict the pipe segment floating process and obtain a time series of predicted values ​​of the pipe segment floating process.

7. The novel method for estimating the floating amount of shield tunnel segments according to claim 6 is characterized in that: On the basis of step S300, the predicted value of the floating process is added to the combination of the predicted value of the floating amount, the monitoring data of the floating process and the parameter data of the influencing factors, and the final floating amount and the floating process are comprehensively considered to automatically adjust the shield excavation parameters to minimize the floating amount of the pipe segment.

8. The novel method for estimating the floating amount of shield tunnel segments according to claim 7 is characterized in that: The state space S in the reinforcement learning framework: S={soil_cohesion,water_level,stresses_variation,permeability_coefficient,current_thrust_force,Support_pressure,predicted_uplift,predicted_uplift_process} Among them, predicted_uplift_process is the predicted value of the current uplift amount.

9. The novel method for estimating the floating amount of shield tunnel segments according to claim 6 is characterized in that: The CNN-LSTM structure in the segment floating process prediction model is used to extract the spatiotemporal features of the two-dimensional image, and the short-term prediction of the two-dimensional image is achieved through the deconvolution layer.

10. The novel method for estimating the floating amount of shield tunnel segments according to claim 9 is characterized in that: Convolutional neural networks are used to extract the spatial features of two-dimensional images, and the output results of the convolutional neural networks are input into long short-term memory networks to capture time series features.

Images