Shield tunneling attitude control method based on multi-objective optimization of deviation correction curve

Abstract

The present invention relates to the technical field of shield tunneling correction control, and specifically refers to a shield tunneling posture control method that performs multi-objective optimization based on a correction curve. Obtain historical data of the shield tunneling system and multiple prediction models for training; select a corresponding correction curve based on the deviation of the current shield posture, and determine the turning radius of the correction curve; plan the specific path of the correction curve to form an analytical expression of the correction curve; obtain the correction target posture based on the analytical expression of the correction curve, obtain the multi-step prediction posture based on the trained optimization prediction model, and determine the correction target function based on the multi-step prediction posture and the correction target posture; select decision variables and construct constraints; obtain recommended parameters for shield posture correction and control based on the correction target function and constraints; and perform correction control on the shield posture based on the recommended parameters. This application improves the level of shield tunneling posture control and plays a role in ensuring the shield tunneling posture and the quality of the formed tunnel.

Description

Technical Field

[0001] The invention relates to the technical field of shield tunneling deviation correction control, and in particular to a shield tunneling posture control method that performs multi-objective optimization according to a deviation correction curve. Background Art

[0002] Shield tunneling attitude control is the key to ensure that shield-type tunnels meet acceptance requirements. However, the traditional manual control of shield attitude can only respond passively after attitude deviation based on experience, which has serious subjectivity and lag, and cannot meet the requirements of fine control of shield attitude, resulting in frequent shield attitude misalignment events. Therefore, the demand for fine control of shield attitude is becoming more and more urgent.

[0003] The prior art entitled "A shield tunneling posture correction control method" proposes a shield tunneling posture correction method, which includes the following steps: Step 1, constructing a correction principle model of the shield machine; the correction principle model includes a correction trajectory curve; Step 2, determining the minimum correction radius of the correction principle model; the minimum correction radius of the correction principle model is the curvature radius of the correction trajectory curve; Step 3, using an artificial ant colony algorithm to process the preset shield machine parameters to obtain an optimal feature subset; Step 4, constructing a shield machine correction mathematical model based on the minimum correction radius and the optimal feature subset; Step 5, based on the correction mathematical model, optimizing the optimal feature subset through an artificial bee colony algorithm to obtain control parameters; and controlling the shield machine tunneling posture for correction according to the control parameters. This method is mainly based on the constructed shield machine correction principle model to perform correction control on the shield tunneling posture. The shield machine parameters are screened by artificial ant colony algorithm to obtain the optimal feature subset, and then the shield machine correction mathematical model is constructed according to the minimum correction radius and the optimal feature subset of the correction principle model. The optimal feature subset is optimized by bee colony algorithm to obtain control parameters, and the shield machine tunneling posture is controlled according to the control parameters to achieve correction. By introducing the ant colony algorithm, the optimal feature subset can be screened out, and the optimal features are used as the initial population of the artificial bee colony algorithm, which can improve the solution quality, improve the correction accuracy, and achieve better correction effect.

[0004] However, there are still some problems with this method: 1. The shield attitude prediction research is mainly based on single-step prediction, and the prediction results lack continuity and trend, and cannot provide sufficient decision-making information for operators; 2. When planning the correction curve, the serious situation of the shield attitude deviation is not distinguished, resulting in the correction curve being too slow or too fast, which is inconsistent with the on-site construction requirements; 3. The shield attitude correction control simply takes the design axis as the target attitude, which can easily cause drastic correction and serpentine correction of the shield attitude. Summary of the invention

[0005] The purpose of this application is to address the deficiencies of the above-mentioned background technology and to provide a shield tunneling posture control method that performs multi-objective optimization based on a correction curve.

[0006] The technical solution of the present application is: a shield tunneling posture control method for multi-objective optimization based on a correction curve, comprising the following steps:

[0007] S1, read the historical data exported by the shield tunneling system, and pre-process the historical data to select feature parameters to construct a data set;

[0008] S2. Construct a multi-step prediction model for shield posture based on multi-head attention mechanism and double-layer gated recurrent unit, and train the multi-step prediction model based on the dataset;

[0009] S3, judging the deviation of the current shield posture based on the displacement deviation of the shield head posture and the angle deviation of the shield, and selecting the corresponding correction curve based on the deviation;

[0010] S4, determining the turning radius of the correction curve based on the selected correction curve;

[0011] S5, planning the specific path of the correction curve based on the determined turning radius and the reverse tangent double circle curve, and forming an analytical formula of the correction curve;

[0012] S6, obtaining the correction target posture based on the correction curve analytical expression, obtaining the multi-step predicted posture based on the trained optimization prediction model, and determining the correction target function according to the multi-step predicted posture and the correction target posture;

[0013] S7, selecting the tunneling parameters that can be actively controlled during the shield tunneling process as decision variables, and constructing the constraint conditions of the decision variables;

[0014] S8. Based on the deviation correction objective function and constraint conditions, the third-generation non-dominated genetic algorithm is used to perform multi-objective optimization on the shield tunneling parameters to obtain the optimal solution of the decision variables, and the optimal solution of the decision variables is used as the recommended parameter for the deviation correction and control of the shield attitude;

[0015] S9. Perform correction control on the shield machine posture based on the recommended parameters and repeat the operation until smooth and orderly control of the shield machine posture is achieved.

[0016] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided by the present application, the method for preprocessing historical data and selecting feature parameters to construct a data set in step S1 includes: selecting data related to the shield posture in the historical data as feature parameters, using the shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation and shield tail vertical deviation of the shield tunneling system as label values ​​to construct a data set, dividing the data set into a training set and a test set, and performing minimum-maximum normalization on the training set and the test set, respectively.

[0017] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided in the present application, the method for training a multi-step prediction model based on a data set in step S2 includes: establishing a feature parameter time window and a label value time window sliding along the time dimension on the data set, processing the feature parameters and label values ​​based on the feature parameter time window and the label value time window to obtain feature sequence samples and label sequence samples; taking the feature sequence samples as input and the label sequence samples as output, and training the multi-step prediction model in a rolling manner.

[0018] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided by the present application, the method for selecting a corresponding correction curve based on the deviation situation in step S3 includes: setting a limit value for distinguishing the quality of the shield posture based on the shield head posture displacement deviation and the shield angle deviation; comparing the shield posture of the shield tunneling system at the current moment with the set limit value; if the shield head posture displacement deviation and the shield angle deviation do not exceed the corresponding limit values, it is judged that the current shield posture deviation situation is good, and a slow correction curve is selected, otherwise a fast correction curve is selected.

[0019] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided by the present application, the method for determining the turning radius of the correction curve based on the selected correction curve in step S4 includes: if a fast correction curve is selected, the maximum value of the turning radius required by the minimum turning radius of the shield tunneling system, the minimum turning radius of the pipe segment and the maximum change in the gap between the tail of each shield ring is used as the turning radius of the fast correction curve; if a slow correction curve is selected, the product of the turning radius of the fast correction curve and the turning coefficient is used as the turning radius of the slow correction curve; the turning coefficient is greater than 1.

[0020] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided by the present application, the method for planning the specific path of the selected correction curve based on the determined turning radius and the reverse tangent double circular curve in step S5 includes: taking the shield head posture displacement deviation of 0 and the shield angle deviation of 0 as the target posture, taking the turning radius obtained in step S4 as the curve radius, and planning the corresponding fast correction path or slow correction path based on the reverse tangent double circular curve.

[0021] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided by the present application, in step S5, the correction curve analytical expression is as follows:

[0022]

[0023] in: L 1 - one of the two circular curves with opposite tangency;

[0024] L 2 - the other curve in the double circular curve with opposite tangency;

[0025] x ——the excavation distance of the shield head along the design axis direction;

[0026] y ——deviation of shield head relative to design axis;

[0027] y ——deviation of shield head relative to design axis;

[0028] θ ——The angle of the current shield posture relative to the design axis;

[0029] R - turning radius;

[0030] α ——The angle corresponding to one of the two circular curves;

[0031] β ——The angle between the line connecting the centers of the double circle curve and the direction perpendicular to the design axis;

[0032] φ ——the angle corresponding to the other curve in the double circular curve;

[0033] Y 1 ——Current shield head deviation value;

[0034] Y 2 ——Current shield tail deviation value;

[0035] l ——The distance from the head to the tail of the shield.

[0036] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided by the present application, the method for determining the correction target function according to the multi-step predicted posture and the correction target posture in step S6 includes: taking the product of the predicted time interval of the multi-step prediction model and the current excavation speed as the shield tunneling distance of a correction cycle; substituting the obtained shield tunneling distance of a correction cycle into the analytical expression of the correction curve to obtain the shield posture value corresponding to the shield tunneling distance, and taking the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation and shield tail vertical deviation in the shield posture value as the correction target posture of this correction cycle; taking the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation and shield tail vertical deviation predicted by the multi-step prediction model as the multi-step predicted posture of this correction cycle; and taking the minimum value of the absolute value of the difference between the multi-step predicted posture and the correction target posture as the correction target function.

[0037] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided by the present application, in the step S7, the shield tunneling speed, cutter head speed, screw machine speed, upper partition pressure, lower partition pressure, left partition pressure and right partition pressure are selected as decision variables; the deviation between the sum of the upper and lower partition pressures in the decision variables and the sum of the upper and lower partition pressures collected by the current shield tunneling system is not greater than a first set value, the deviation between the sum of the left and right partition pressures in the decision variables and the sum of the left and right partition pressures collected by the current shield tunneling system is not greater than a second set value, and the absolute value of the difference between the decision variable and the corresponding value collected by the current shield tunneling system is not greater than a third set value, thereby constructing a constraint condition.

[0038] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided in this application, the formulas of the correction objective function and constraint conditions are as follows:

[0039]

[0040] in: f 1 ——Shield head horizontal deviation correction objective function;

[0041] f 2 ——Shield head vertical deviation correction objective function;

[0042] f 3 ——Target function of shield tail horizontal deviation correction;

[0043] f 4——Target function of shield tail vertical deviation correction;

[0044] x 1 ——Shield tunneling speed;

[0045] x 2 ——Cutter head speed;

[0046] x 3 ——Screw conveyor speed;

[0047] x 4 ——upper partition pressure;

[0048] x 5 ——lower partition pressure;

[0049] x 6 ——Left partition pressure;

[0050] x 7 ——right partition pressure;

[0051] y 1 ——Correct the horizontal deviation of the shield head in the target posture;

[0052] y 2 ——Correct the vertical deviation of the shield head in the target posture;

[0053] y 3 ——Correct the horizontal deviation of the shield tail in the target posture;

[0054] y 4 ——Correct the vertical deviation of the shield tail in the target posture.

[0055] x i初始 ——The current values ​​of the seven decision variables read by the shield tunneling system.

[0056] According to a shield tunneling posture control method for multi-objective optimization based on a correction curve provided by the present application, the method for correcting the shield posture based on recommended parameters in step S9 includes: correcting the shield posture according to the recommended parameters, and performing correction control for the next cycle after completing the correction control of this cycle; when starting the next cycle, if the shield posture is corrected from a state greater than a limit value to a state not greater than the limit value after the end of this cycle, the correction curve of the next cycle is automatically switched to a slow correction curve, and the corresponding correction path is replanned; conversely, if the shield posture is corrected from a state not greater than the limit value to a state greater than the limit value after the end of this cycle, the correction curve of the next cycle is automatically switched to a fast correction curve, and the corresponding correction path is replanned.

[0057] The advantages of this application are as follows: 1. This application has established an overall implementable path for intelligent control of shield tunneling posture from shield posture evaluation, deviation correction curve planning, posture multi-step prediction to posture auxiliary control, which can assist manual fine control of shield posture, provide a scientific basis for intelligent control of shield posture, accurately predict shield tunneling, and achieve a good deviation correction control effect. Through multi-objective optimization, shield tunneling can be accurately adjusted, solving the current problem of accurate prediction and control of shield posture that needs to be overcome, reducing the work intensity of shield operators, improving the control level of shield tunneling posture, and playing a role in ensuring the shield tunneling posture and the quality of the formed tunnel;

[0058] 2. This application constructs a data set by selecting historical data of the current shield tunneling system. The selected feature parameters are closely related to shield deviation correction. The preprocessed data set can be used to train the multi-prediction model, so that the trained multi-prediction model can accurately predict the future shield posture, thereby improving the reflection of the shield posture prediction results on the long-term trend;

[0059] 3. The training method of the multi-prediction model in this application is highly targeted. The data set is split into feature sequence samples and label sequence samples, and multiple rounds of rolling training can be performed. Multiple trainings can greatly improve the accuracy of the multi-prediction model. The trained multi-prediction model can accurately predict the shield posture of the shield tunneling system.

[0060] 4. This application proposes a classification and determination method based on the seriousness of the shield attitude deviation, evaluates the current shield attitude deviation, performs rapid deviation correction when the attitude deviation is serious, and performs slow deviation correction when the attitude deviation is good, which meets the actual construction needs on site. Compared with the traditional shield attitude control method, it can effectively avoid the problems of excessive shield attitude fluctuation, serpentine deviation correction and even attitude loss control caused by over-correction of the shield attitude;

[0061] 5. The method of obtaining the turning radius of the correction curve in this application is very simple. It is determined based on the minimum turning radius of the shield tunneling system, the minimum turning radius of the pipe segment, and the turning radius required for the maximum change in the gap between the tail of each ring of the shield. It fully considers the operation of the shield tunneling system and conforms to the actual shield construction situation.

[0062] 6. The method of constructing a fast deviation correction path or a slow deviation correction path in this application is very simple. The fast deviation correction path or the slow deviation correction path is the target posture of the shield correction. Correction operation based on the target posture can effectively improve the refinement of the shield posture control and avoid the consequences of over-correction or even loss of control of the shield posture caused by simply taking the design axis as the correction target when the shield posture deviation is large;

[0063] 7. This application constructs a special correction curve analytical formula to facilitate subsequent accurate calculations. The analytical formula is a mathematical explanation of the fast correction path or the slow correction path. Based on the analytical formula, accurate data of the target posture can be quickly obtained, which greatly reduces the difficulty of subsequent correction adjustment;

[0064] 8. This application constructs a deviation correction objective function, which is obtained based on the target posture and the multi-step predicted posture. The deviation correction objective function facilitates the subsequent precise control of the shield tunneling. On the one hand, the deviation correction objective function can ensure that the shield tunneling is on the deviation correction path. On the other hand, the future status of the shield tunneling is considered through prediction, ensuring the stability and accuracy of the deviation correction.

[0065] 9. This application constructs the decision variables and constraints for controlling the shield tunneling system, provides guidance measures for the deviation correction control and adjustment of the shield tunneling, facilitates the further accurate acquisition of the corresponding recommended parameters, and ensures the rationality and feasibility of the recommended construction parameters;

[0066] 10. The deviation correction control of this application is carried out in a cycle. After each cycle is completed, the completed cycle will be summarized and analyzed to prepare and adjust for the next cycle, ensuring that the shield deviation correction is always under reasonable operation, thereby improving the efficiency and accuracy of the shield posture adjustment.

[0067] The present application constructs an intelligent auxiliary control method for shield tunneling posture through a multi-objective optimization method, which can realize an intelligent prediction, planning and control method of the shield posture, solve the current urgent problem of accurate prediction and control of the shield posture, reduce the work intensity of shield operators, improve the control level of shield tunneling posture, and play a role in ensuring the shield tunneling posture and the quality of the formed tunnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 : Schematic diagram of the control method flow of this application;

[0069] Figure 2 : Schematic diagram of the comparison between the shield attitude offset and the limit value of this application;

[0070] Figure 3 : Schematic diagram of the relative relationship between the shield posture and the designed axial direction of this application;

[0071] Figure 4 : Schematic diagram of the correction curve of this application. DETAILED DESCRIPTION

[0072] The embodiments of the present application are described in detail below, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0073] In the description of the present application, it should be understood that the terms "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0074] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0075] The present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0076] The present application relates to a shield tunneling attitude control method that performs multi-objective optimization based on a correction curve. The control method of the present application is used to control the attitude during shield tunneling and to correct the shield tunneling. The control method of the present application constructs an intelligent auxiliary control method for shield tunneling attitude through multi-step prediction of shield attitude, evaluation of shield attitude deviation, classification planning of correction curves, and multi-objective optimization methods. The multi-step prediction model of shield attitude is used as the fitness function, the correction curve is used as the target attitude, the attitude prediction result is close to the target attitude as the goal, and the key parameters that can be actively controlled during construction are used as decision variables. A global optimization is performed through a multi-objective optimization algorithm to obtain the optimal decision variable value as a construction recommended parameter, thereby realizing an intelligent prediction, planning, and control method for shield attitude, solving the current problem of accurate prediction and control of shield attitude that needs to be overcome and solved, reducing the work intensity of shield operators, and improving the control level of shield tunneling attitude, which plays a role in ensuring the shield tunneling attitude and the quality of the formed tunnel.

[0077] Specifically, Figure 1 As shown, a shield tunneling attitude control method for multi-objective optimization based on a correction curve of the present application is performed in the following steps:

[0078] S1, read the historical data exported by the shield tunneling system, and pre-process the historical data to select feature parameters to construct a data set;

[0079] The shield tunneling data and shield tunneling results of different shield tunneling systems are different. By obtaining the historical data of the current shield tunneling system, the multi-step prediction model can be trained with these historical data, so that the multi-step prediction model can fully adapt to the current shield tunneling system and make the multi-step prediction model more accurate in predicting the shield tunneling of the shield tunneling system.

[0080] S2. Construct a multi-step prediction model for shield posture based on multi-head attention mechanism and double-layer gated recurrent unit, and train the multi-step prediction model based on the dataset;

[0081] Create a multi-step prediction model for shield posture based on the multi-head attention mechanism (Multihead-Attention) and the double-layer gated recurrent unit (GRU). The multi-prediction model is trained based on the data set obtained in step S1 to make it fully correspond to the current shield tunneling system;

[0082] S3, judging the deviation of the current shield posture based on the displacement deviation of the shield head posture and the angle deviation of the shield, and selecting the corresponding correction curve based on the deviation;

[0083] By collecting the deviation of the shield tunneling system, it is determined whether the current deviation of the shield tunneling system is serious. If it is serious, a faster adjustment method needs to be collected. If it is not serious, a slower adjustment method can be adopted. Based on this principle, the corresponding correction curve is constructed. The correction curve constructed here is actually the correction control mode, that is, whether to select fast correction or slow correction;

[0084] S4, determining the turning radius of the correction curve based on the selected correction curve;

[0085] The turning radius of the correction curve is determined based on the minimum turning radius of the shield tunneling system, the minimum turning radius of the segment, and the turning radius required for the maximum change in the gap between the tails of each shield ring. It is determined based on the parameters of the shield tunneling system and must meet the construction requirements of the shield tunneling system.

[0086] S5, planning the specific path of the correction curve based on the determined turning radius and the reverse tangent double circle curve, and forming an analytical formula of the correction curve;

[0087] The specific path can be determined by the turning radius and the curve form adopted by the deviation correction curve of the present application. The reverse tangent double circular curve adopted in the present application forms an ideal path, which is assumed to achieve the desired deviation correction effect by moving the shield tunneling system along the path.

[0088] The constructed analytical formula of the correction curve is to quantify the path, which is convenient for subsequent accurate calculations;

[0089] S6, obtaining the correction target posture based on the correction curve analytical expression, obtaining the multi-step predicted posture based on the trained optimization prediction model, and determining the correction target function according to the multi-step predicted posture and the correction target posture;

[0090] The correction curve analytical expression obtains the correction target posture, which is the ideal posture. The multi-step prediction posture is the predicted posture obtained based on the multi-step prediction model, that is, the posture in the future time according to the current shield posture without any adjustment. By comparing the multi-step prediction posture and the correction target posture, the correction objective function can be obtained.

[0091] S7, selecting the tunneling parameters that can be actively controlled during the shield tunneling process as decision variables, and constructing the constraint conditions of the decision variables;

[0092] The decision variables are the parameters that can be actively controlled in the shield tunneling system, because these parameters can be controlled, and controlling these parameters can correct the shield tunneling system. By determining the specific conditions of these decision variables, it is convenient to control and adjust the shield tunneling system in the future.

[0093] The construction of constraint conditions is to facilitate the subsequent acquisition of recommended parameters. At the same time, the constraint conditions take into account the rationality and feasibility of the shield tunneling system construction;

[0094] S8. Based on the deviation correction objective function and constraint conditions, the third-generation non-dominated genetic algorithm is used to perform multi-objective optimization on the shield tunneling parameters to obtain the optimal solution of the decision variables, and the optimal solution of the decision variables is used as the recommended parameter for the deviation correction and control of the shield attitude;

[0095] Substituting the constraints into the deviation correction objective function for optimization calculation, the optimal solution of the decision variable can be obtained. The optimal solution is used as the recommended parameter, which is the adjustment parameter required for deviation correction control.

[0096] S9. Correct the shield machine posture based on the recommended parameters and repeat the operation until the shield machine posture is smoothly and orderly controlled;

[0097] The deviation correction control of the present application is a cyclic operation, that is, the cyclic control is carried out according to the set cycle time. After each cycle is completed, the operation mode of the next cycle is determined according to the shield excavation situation after the completion of the cycle, and the operation is carried out in sequence until the deviation correction control operation of the shield excavation system is completed.

[0098] In some embodiments of the present application, this embodiment optimizes the above-mentioned steps S1 and S2. Specifically, the above-mentioned steps S1 and S2 can be said to be the precursor of the entire shield tunneling system correction control operation, and its purpose is to obtain a trained multi-step prediction model.

[0099] First, the historical data of the shield tunneling system is collected. There are a lot of historical data of the shield tunneling system, including the historical posture data of the shield, the tunneling system, the slag discharge system, the grouting system and other data automatically recorded during the shield tunneling construction process. The historical data is preprocessed, including missing value filling, tunneling section data extraction, abnormal data cleaning and filling; but in fact, the parameters that need to be considered in this application are not so many, only the parameters related to the shield posture are needed. The parameters with a high correlation with the shield posture selected in this embodiment are used as characteristic parameters, including the horizontal and vertical deviations of the shield cutout and the horizontal and vertical deviations of the tail, the propulsion speed, the total thrust of the cylinder, the cutter head torque, the cutter head speed, the propulsion cylinder speed and zone pressure, the pressure of each soil pressure sensor in the soil bin, the screw machine speed, the screw machine torque, the real-time grouting pressure and grouting amount of each grouting pipeline, and the horizontal deviation of the shield head. y 1 , shield head vertical deviation y 2 , Horizontal deviation of shield tail y 3 and shield tail vertical deviation y 4As the label value, a data set of the multi-step prediction model is constructed, and then the data set is divided into a training set and a test set, and the training set and the test set are respectively subjected to minimum-maximum normalization.

[0100] Then, the multi-step prediction model is trained based on the obtained data set. This embodiment builds a multi-step prediction model for shield posture based on multi-head attention mechanism (Multihead-Attention) and double-layer gated recurrent unit (GRU). A feature parameter time window sliding along the time dimension is established on the data set. T 1 and label value time window T 2 , based on the characteristic parameter time window T 1 and label value time window T 2 Process the feature parameters and label values ​​to obtain the feature sequence sample X T (t) and label sequence sample Y T (t), the specific formula is as follows:

[0101]

[0102] Where: X T (t)——feature sequence sample;

[0103] Y T (t)——Tag sequence sample;

[0104] t ——The time point at which the network model inputs the time window data;

[0105] m - the number of characteristic parameter data types;

[0106] The feature sequence sample X T (t) Feature sequence sample as input, label sequence sample Y T (t) is used as the output, and the multi-step prediction model is trained in a rolling manner. Each time the input is of known length T 1 The characteristic sequence of excavation parameters is used to predict the next period of time (length is T 2 )’s label value, thereby realizing multi-step prediction of shield posture, performing model training multiple times, and saving the trained model for subsequent use.

[0107] In other embodiments of the present application, this embodiment optimizes the above-mentioned steps S3, S4 and S5. Steps S3, S4 and S5 are for obtaining the correction target posture, and the correction target posture is determined based on the shield posture and the correction target axis of the current shield tunneling system. The correction target posture is a geometric data model constructed based on the current shield posture and the correction target axis.

[0108] First, it is necessary to determine which correction curve to choose, and set the limit value for distinguishing the good and bad shield postures based on the shield head posture displacement deviation and shield angle deviation. For example, the shield head posture displacement deviation limit value is set to 30mm, and the shield angle deviation limit value is set to 0.2°; compare the shield posture of the shield tunneling system at the current moment with the set limit value, such as Figure 2 As shown; if the shield head posture displacement deviation and the shield angle deviation do not exceed the corresponding limit values, that is, the current shield head posture displacement deviation ≤30mm and the current shield angle deviation ≤0.2°, it is judged that the current shield posture deviation is good, and the slow correction curve is selected; if the current shield head posture displacement deviation is greater than 30mm or the current shield angle deviation is greater than 0.2°, it is considered that the current shield posture deviation is large, and the fast correction curve needs to be selected.

[0109] In this embodiment, both the fast correction curve and the slow correction curve are planned based on the reverse tangent double circle curve. If the fast correction curve is selected, the minimum turning radius of the shield tunneling system is R 1 , Minimum turning radius of the segment R 2 The turning radius required by the maximum change in the gap between the shield and the tail of each ring R 3 The maximum value is used as the turning radius of the fast correction curve R， Right now R=max [ R 1 , R 2 , R 3 ]; If a slow correction curve is selected, the turning radius of the fast correction curve R and turning coefficient k The turning radius of the slow correction curve is the product of R’ ,Right now R'=kR ; The turning coefficient is greater than 1, and can be specifically designed to be 2.

[0110] Among them, the turning radius required for the maximum change in the gap between each ring and the tail is R 3 The calculation formula is as follows:

[0111]

[0112] in: Δs ——The maximum change of the gap between the tail and the shield of each ring;

[0113] β ——The corner of a shield tunneling ring;

[0114] L 盾尾 —— Length of shield tail;

[0115] L ——The length of one ring of shield tunneling;

[0116] R 3 ——The turning radius required for the maximum change in the gap between the tail and the shield of each ring;

[0117] D ——Installation diameter of shield thrust cylinder.

[0118] In practical applications, the relationship between the shield posture and the target deviation correction axis is as follows: Figure 3 The eight types shown take the shield head posture displacement deviation of 0 and the shield angle deviation of 0 as the target posture, take the turning radius obtained in step S4 as the curve radius, and plan the corresponding fast correction path or slow correction path based on the reverse tangent double circle curve.

[0119] After determining the fast correction path or the slow correction path, the correction curve analytical formula can be constructed. Figure 3 The first case shown in the figure constructs the analytical formula of the correction curve (see Figure 4 As shown), the formula is as follows:

[0120]

[0121] in: L 1 - one of the two circular curves with opposite tangency;

[0122] L 2 - the other curve in the double circular curve with opposite tangency;

[0123] x ——the excavation distance of the shield head along the design axis direction;

[0124] y ——deviation of shield head relative to design axis;

[0125] y ——deviation of shield head relative to design axis;

[0126] θ ——The angle of the current shield posture relative to the design axis;

[0127] R - turning radius;

[0128] α ——the central angle of the L2 curve in the double circular curve;

[0129] φ ——the central angle of the L1 curve in the double circular curve;

[0130] Y 1 ——Current shield head deviation value;

[0131] Y 2 ——Current shield tail deviation value;

[0132] l ——The distance from the head to the tail of the shield.

[0133] The correction target posture of the shield tunneling system can be obtained through the analytical expression of the correction curve.

[0134] In a preferred embodiment of the present application, this embodiment optimizes the above-mentioned step S6, which combines the multi-step prediction posture predicted by the multi-step prediction model with the correction target posture obtained in step S5 to obtain the correction target function of the current shield tunneling system.

[0135] Introduce the multi-step prediction model (Multihead-Attention-GRU) trained in step S2 as the optimized fitness function MG(x) The product of the prediction time interval of the multi-step prediction model (actually the time of a correction cycle) and the current excavation speed (obtained by the shield excavation system itself) is taken as the shield excavation distance of a correction cycle; the shield excavation distance of a correction cycle is substituted into the correction curve analytical formula to obtain the shield attitude value corresponding to the shield excavation distance, and the shield head horizontal deviation in the shield attitude value is replaced by y 1 , shield head vertical deviation y 2 , Horizontal deviation of shield tail y 3 and shield tail vertical deviation y 4 As the correction target posture of this correction cycle; the horizontal deviation of the shield head predicted by the multi-step prediction model MG 1 ( x ), Horizontal deviation of shield tail MG 2 ( x ), shield head vertical deviation MG3 ( x ) and the vertical deviation of the shield tail MG 4 ( x ) is used as the multi-step prediction posture of this correction cycle. In fact, the horizontal deviation of the shield head, the vertical deviation of the shield head, the horizontal deviation of the shield tail and the vertical deviation of the shield tail of the current shield tunneling system are input into the multi-step prediction model. The multi-step prediction model calculates the horizontal deviation of the shield head MG 1 ( x ), Horizontal deviation of shield tail MG 2 ( x ), shield head vertical deviation MG 3 ( x ) and the vertical deviation of the shield tail MG 4 ( x ); multi-step predicted pose [ MG 1 ( x ) ,MG 2 ( x ) , M.G. 3 ( x ) , M.G. 4 ( x )] and the correction target posture [ y 1, ,y 2 , y 3 , y 4 ] is taken as the correction objective function.

[0136] The deviation correction control operation of this embodiment is mainly to control the items that can be actively controlled in the shield tunneling system, that is, the decision variables mentioned in step S7 of this application. This embodiment optimizes the above step S7. Specifically, in the above step S7, the shield tunneling speed is selected. x 1 , Cutter speed x 2 , Screw machine speed x 3 , Upper zone pressure x 4 , Lower zone pressure x 5 , Left partition pressure x 6 and right partition pressure x 7As decision variables; the deviation between the sum of the upper and lower partition pressures in the decision variables and the sum of the upper and lower partition pressures collected by the current shield tunneling system is not greater than the first set value (the first set value of this embodiment is 10%), the deviation between the sum of the left and right partition pressures in the decision variables and the sum of the left and right partition pressures collected by the current shield tunneling system is not greater than the second set value (the second set value of this embodiment is 10%), and the absolute value of the difference between the decision variable and the corresponding value collected by the current shield tunneling system is not greater than the third set value (the third set value of this embodiment is 10%), to construct constraint conditions. For the specific example of constructing the deviation correction objective function and constraint conditions, see the following formula:

[0137]

[0138] in: f 1 ——Shield head horizontal deviation correction objective function;

[0139] f 2 ——Shield head vertical deviation correction objective function;

[0140] f 3 ——Target function of shield tail horizontal deviation correction;

[0141] f 4 ——Target function of shield tail vertical deviation correction;

[0142] x 1 ——Shield tunneling speed;

[0143] x 2 ——Cutter head speed;

[0144] x 3 ——Screw conveyor speed;

[0145] x 4 ——upper partition pressure;

[0146] x 5 ——lower partition pressure;

[0147] x 6 ——Left partition pressure;

[0148] x 7 ——right partition pressure;

[0149] y 1——Correct the horizontal deviation of the shield head in the target posture;

[0150] y 2 ——Correct the vertical deviation of the shield head in the target posture;

[0151] y 3 ——Correct the horizontal deviation of the shield tail in the target posture;

[0152] y 4 ——Correct the vertical deviation of the shield tail in the target posture.

[0153] x i初始 ——The current values ​​of the seven decision variables read by the shield tunneling system;

[0154] The shield tunneling speed in this embodiment x 1 , Cutter speed x 2 , Screw machine speed x 3 , Upper zone pressure x 4 , Lower zone pressure x 5 , Left partition pressure x 6 and right partition pressure x 7 The value range of each decision variable is constructed based on the constraints, and then the decision variables within the value range are substituted into the fitness function MG(x) Calculate in, and finally according to the above shield head horizontal deviation correction objective function f 1 , shield head vertical deviation correction objective function f 2 , shield tail horizontal deviation correction objective function f 3 , shield tail vertical deviation correction objective function f 4 Calculate in the formula.

[0155] Based on the correction objective function and constraints obtained in the above steps, the third-generation non-dominated genetic algorithm (NSGA-III) is used to perform multi-objective optimization on the shield tunneling parameters. The Pareto solution set is obtained by global optimization. Taking the posture in the next time period on the correction curve as the target, the optimal solutions of the above 7 decision variables are obtained as the recommended parameters for shield posture correction and control.

[0156] In some embodiments of the present application, this embodiment optimizes the above-mentioned step S9. Specifically, this embodiment is an actual operation method for correcting the shield posture. After obtaining the extension parameters, the shield posture is corrected according to the recommended parameters. After completing the correction control of this cycle, the correction control of the next cycle is performed; when the next cycle starts, if the shield posture is corrected from a state greater than the limit value to a state not greater than the limit value after the end of this cycle, the correction curve of the next cycle is automatically switched to a slow correction curve, and the corresponding correction path is replanned; conversely, if the shield posture is corrected from a state not greater than the limit value to a state greater than the limit value after the end of this cycle, the correction curve of the next cycle is automatically switched to a fast correction curve, and the corresponding correction path is replanned.

[0157] like Figure 1As shown, the shield tunneling posture control method for multi-objective optimization based on the correction curve of the present application is as follows: data related to the shield posture in the historical data are selected as feature parameters, and a data set is constructed with the shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation and shield tail vertical deviation of the shield tunneling system as label values, the data set is divided into a training set and a test set, and the training set and the test set are respectively subjected to minimum-maximum normalization; a feature parameter time window and a label value time window sliding along the time dimension are established on the data set, and the feature parameters and label values ​​are processed based on the feature parameter time window and the label value time window to obtain feature sequence samples and label sequence samples; the feature sequence samples are converted into the feature parameter time window and the label value time window. The label sequence samples are used as input and the label sequence samples are used as output, and the multi-step prediction model is trained in a rolling manner; based on the shield head posture displacement deviation and shield angle deviation, the boundary value for distinguishing the good and bad shield posture is set; the shield posture of the shield tunneling system at the current moment is compared with the set boundary value; if the shield head posture displacement deviation and the shield angle deviation do not exceed the corresponding boundary value, it is judged that the current shield posture deviation is good, and the slow correction curve is selected, otherwise the fast correction curve is selected; if the fast correction curve is selected, the maximum value of the turning radius of the minimum turning radius of the shield tunneling system, the minimum turning radius of the pipe segment, and the maximum change in the gap between the shield tail of each ring is used as the fast correction curve turning radius; if a slow correction curve is selected, the product of the turning radius of the fast correction curve and the turning coefficient is used as the turning radius of the slow correction curve; the method for planning the specific path of the selected correction curve based on the determined turning radius and the reverse tangent double circular curve includes: taking the shield head posture displacement deviation as 0 and the shield angle deviation as 0 as the target posture, and obtaining the turning radius as the curve radius, planning the corresponding fast correction path or slow correction path based on the reverse tangent double circular curve, and constructing the correction curve analytical expression of the correction path; taking the product of the predicted time interval of the multi-step prediction model and the current excavation speed as the shield excavation distance of a correction cycle; taking the obtained one Substitute the shield tunneling distance of a correction cycle into the analytical expression of the correction curve to obtain the shield attitude value corresponding to the shield tunneling distance, and use the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation and shield tail vertical deviation in the shield attitude value as the correction target attitude of this correction cycle; the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation and shield tail vertical deviation predicted by the multi-step prediction model are used as the multi-step prediction attitude of this correction cycle; the minimum value of the absolute value of the difference between the multi-step prediction attitude and the correction target attitude is used as the correction objective function; the shield tunneling speed, cutter head speed, screw machine speed, upper partition pressure, lower partition pressure, left partition pressure and right partition pressure are selected as decision variables;The constraint conditions are constructed based on the deviation between the sum of the upper and lower partition pressures in the decision variables and the sum of the upper and lower partition pressures collected by the current shield tunneling system being no greater than the first set value, the deviation between the sum of the left and right partition pressures in the decision variables and the sum of the left and right partition pressures collected by the current shield tunneling system being no greater than the second set value, and the absolute value of the difference between the decision variables and the corresponding values ​​collected by the current shield tunneling system being no greater than the third set value. Based on the correction objective function and the constraint conditions, the third generation non-dominated genetic algorithm (NSGA-III) is used to perform multi-objective optimization on the shield tunneling parameters, and the Pareto solution set is obtained by global optimization. The posture of the next time period on the correction curve is taken as the target, and the above 7 are obtained. The optimal solution of the decision variables is used as the recommended parameters for shield posture correction and control; the shield posture is corrected and controlled according to the recommended parameters. After the correction control of this cycle is completed, the correction control of the next cycle is carried out; when the next cycle starts, if the shield posture is corrected from a state greater than the limit value to a state not greater than the limit value after the end of this cycle, the correction curve of the next cycle is automatically switched to a slow correction curve, and the corresponding correction path is replanned; on the contrary, if the shield posture is corrected from a state not greater than the limit value to a state greater than the limit value after the end of this cycle, the correction curve of the next cycle is automatically switched to a fast correction curve, and the corresponding correction path is replanned. ;

[0158] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present application. The present application may have various changes and improvements without departing from the spirit and scope of the present application, and these changes and improvements fall within the scope of the present application to be protected. The scope of protection claimed in the present application is defined by the attached claims and their equivalents.

Claims

1. A shield tunneling attitude control method for multi-objective optimization based on a correction curve, characterized in that: The following steps are included: S1, read the historical data exported by the shield tunneling system, and pre-process the historical data to select feature parameters to construct a data set; S2. Construct a multi-step prediction model for shield posture based on multi-head attention mechanism and double-layer gated recurrent unit, and train the multi-step prediction model based on the dataset; S3, judging the deviation of the current shield posture based on the displacement deviation of the shield head posture and the angle deviation of the shield, and selecting the corresponding correction curve based on the deviation; S4, determining the turning radius of the correction curve based on the selected correction curve; S5, planning the specific path of the correction curve based on the determined turning radius and the reverse tangent double circle curve, and forming an analytical formula of the correction curve; S6, obtaining the correction target posture based on the correction curve analytical expression, obtaining the multi-step predicted posture based on the trained optimization prediction model, and determining the correction target function according to the multi-step predicted posture and the correction target posture; S7, selecting the tunneling parameters that can be actively controlled during the shield tunneling process as decision variables, and constructing the constraint conditions of the decision variables; S8. Based on the deviation correction objective function and constraint conditions, the third-generation non-dominated genetic algorithm is used to perform multi-objective optimization on the shield tunneling parameters to obtain the optimal solution of the decision variables, and the optimal solution of the decision variables is used as the recommended parameter for the deviation correction and control of the shield attitude; S9. Correct the shield machine posture based on the recommended parameters and repeat the operation until the shield machine posture is smoothly and orderly controlled; The method for selecting the corresponding correction curve based on the deviation in step S3 includes: setting a limit value for distinguishing the good and bad shield postures based on the shield head posture displacement deviation and the shield angle deviation; The shield posture of the shield tunneling system at the current moment is compared with the set limit value; if the shield head posture displacement deviation and the shield angle deviation do not exceed the corresponding limit value, it is judged that the current shield posture deviation is good and the slow correction curve is selected, otherwise the fast correction curve is selected.

2. A shield tunneling attitude control method for multi-objective optimization based on a correction curve as claimed in claim 1, characterized in that: The method for preprocessing historical data and selecting feature parameters to construct a data set in step S1 includes: selecting data related to the shield posture in the historical data as feature parameters, using the shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation and shield tail vertical deviation of the shield tunneling system as label values ​​to construct a data set, dividing the data set into a training set and a test set, and performing minimum-maximum normalization on the training set and the test set, respectively.

3. The shield tunneling attitude control method for multi-objective optimization based on the correction curve according to claim 1, characterized in that: The method for training the multi-step prediction model based on the data set in step S2 includes: establishing a feature parameter time window and a label value time window sliding along the time dimension on the data set, processing the feature parameters and label values ​​based on the feature parameter time window and the label value time window to obtain feature sequence samples and label sequence samples; taking the feature sequence samples as input and the label sequence samples as output, and training the multi-step prediction model in a rolling manner.

4. The shield tunneling attitude control method for multi-objective optimization based on the correction curve according to claim 1, characterized in that: The method for determining the turning radius of the correction curve based on the selected correction curve in step S4 includes: if a fast correction curve is selected, the maximum value of the turning radius of the minimum turning radius of the shield tunneling system, the minimum turning radius of the segment and the maximum change in the gap between the tail of each shield ring is used as the turning radius of the fast correction curve; if a slow correction curve is selected, the product of the turning radius of the fast correction curve and the turning coefficient is used as the turning radius of the slow correction curve; the turning coefficient is greater than 1.

5. The shield tunneling attitude control method for multi-objective optimization based on the correction curve according to claim 4, characterized in that: The method for planning the specific path of the selected correction curve based on the determined turning radius and the reverse tangent double circular curve in step S5 includes: taking the shield head posture displacement deviation of 0 and the shield angle deviation of 0 as the target posture, taking the turning radius obtained in step S4 as the curve radius, and planning the corresponding fast correction path or slow correction path based on the reverse tangent double circular curve.

6. A shield tunneling attitude control method for multi-objective optimization based on a correction curve as claimed in claim 5, characterized in that: The method for determining the correction target function according to the multi-step predicted posture and the correction target posture in step S6 includes: taking the product of the predicted time interval of the multi-step prediction model and the current excavation speed as the shield excavation distance of a correction cycle; substituting the obtained shield excavation distance of a correction cycle into the analytical expression of the correction curve to obtain the shield posture value corresponding to the shield excavation distance, and taking the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation and shield tail vertical deviation in the shield posture value as the correction target posture of this correction cycle; taking the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation and shield tail vertical deviation predicted by the multi-step prediction model as the multi-step predicted posture of this correction cycle; and taking the minimum value of the absolute value of the difference between the multi-step predicted posture and the correction target posture as the correction target function.

7. A shield tunneling attitude control method for multi-objective optimization based on a correction curve as claimed in claim 6, characterized in that: In step S7, shield tunneling speed, cutter head speed, screw machine speed, upper partition pressure, lower partition pressure, left partition pressure and right partition pressure are selected as decision variables; The constraint conditions are constructed based on the fact that the deviation between the sum of the upper and lower partition pressures in the decision variable and the sum of the upper and lower partition pressures collected by the current shield tunneling system is not greater than the first set value, the deviation between the sum of the left and right partition pressures in the decision variable and the sum of the left and right partition pressures collected by the current shield tunneling system is not greater than the second set value, and the absolute value of the difference between the decision variable and the corresponding value collected by the current shield tunneling system is not greater than the third set value.

8. The shield tunneling attitude control method for multi-objective optimization based on the correction curve according to claim 7, characterized in that: The formulas of the deviation correction objective function and constraint conditions are as follows: in: f 1 ——Shield head horizontal deviation correction objective function; f 2 ——Shield head vertical deviation correction objective function; f 3 ——Target function of shield tail horizontal deviation correction; f 4 ——Target function of shield tail vertical deviation correction; x 1 ——Shield tunneling speed; x 2 ——Cutter head speed; x 3 ——Screw conveyor speed; x 4 ——upper partition pressure; x 5 ——lower partition pressure; x 6 ——Left partition pressure; x 7 ——right partition pressure; y 1 ——Correct the horizontal deviation of the shield head in the target posture; y 2 ——Correct the vertical deviation of the shield head in the target posture; y 3 ——Correct the horizontal deviation of the shield tail in the target posture; y 4 ——Correct the vertical deviation of the shield tail in the target posture; x i初始 ——The current values ​​of the seven decision variables read by the shield tunneling system; MG 1 ( x )——Shield head horizontal deviation predicted by multi-step prediction model; MG 2 ( x )——horizontal deviation of the shield tail predicted based on the multi-step prediction model; MG 3 ( x )——the vertical deviation of the shield head predicted by the multi-step prediction model; MG 4 ( x )——The vertical deviation of the shield tail predicted based on the multi-step prediction model.

9. A shield tunneling attitude control method for multi-objective optimization based on a correction curve as claimed in claim 8, characterized in that: The method for correcting the shield posture based on the recommended parameters in step S9 includes: correcting the shield posture according to the recommended parameters, and performing correction control for the next cycle after completing the correction control of this cycle; when starting the next cycle, if the shield posture is corrected from a state greater than a limit value to a state not greater than the limit value after the end of this cycle, the correction curve of the next cycle is automatically switched to a slow correction curve, and the corresponding correction path is replanned; conversely, if the shield posture is corrected from a state not greater than the limit value to a state greater than the limit value after the end of this cycle, the correction curve of the next cycle is automatically switched to a fast correction curve, and the corresponding correction path is replanned.

Images