Direction control method for a shield tunneling machine, direction control system, and method for calculating predicted force points.

Abstract

To provide a direction control method of a shield machine capable of predicting a target force point to be determined by an operator precisely, a direction control system and a calculation method of a prediction force point.SOLUTION: In calculating a prediction force point, a step S202 of generating a teacher data obtained by combining a target force point which is input by an operator actually as a correct answer data to information for learning, a step S 203 of generating a plurality of prediction models by executing a machine learning per pattern obtained by sorting the teacher data on the basis of a model generation condition, a step S204 of selecting a prediction model having a smallest difference between an acting force point obtained by inputting information for assessment to each of the plurality of prediction models and the target force point input by the operator actually corresponding to the information for assessment and a process S205 of calculating a prediction force point for a section to be excavated using an optimal prediction model are executed.SELECTED DRAWING: Figure 10

Description

Technical Field

[0001] The present invention relates to a direction control method for a shield tunneling machine, a direction control system, and a method for calculating a predicted force point.

Background Art

[0002] Automated control of the tunneling direction of a shield tunneling machine has been under consideration. For example, in Patent Document 1, a recommended force point recommended as the acting force point of the jack thrust is calculated in order to cause the shield tunneling machine to tunnel along a planned line. The recommended horizontal force point, which is the horizontal component of the recommended force point, is calculated based on the results of regression analysis of the actual stroke difference and the actual horizontal force point. Further, the recommended vertical force point, which is the vertical component of the recommended force point, is calculated based on the regression analysis of the actual pitching angle difference and the actual vertical force point. Thereby, an operator can determine a target force point by referring to the recommended force point.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] By the way, the above-described recommended force point is merely an acting force point recommended based on the results of regression analysis, and it is not uncommon for it to be different from the target force point determined by the operator by referring to the recommended force point. Therefore, in the automated control of the tunneling direction, it is preferable that the target force point that the operator will determine can be predicted with high accuracy according to the situation at each time.

Means for Solving the Problems

[0005] A shield tunneling machine direction control method that solves the above problems controls the tunneling direction of the shield tunneling machine based on predicted points of force, which are the points of force applied by jack thrusts applied to the shield tunneling machine and are predicted to be target points of force input by the operator when tunneling the planned tunneling section. The direction control system of the shield tunneling machine performs the following steps: creating training data by combining training information with target points of force determined by the operator that correspond to the training information as ground truth data; classifying the training data into a plurality of patterns based on model generation conditions and generating a plurality of prediction models by performing machine learning for each of the plurality of patterns; inputting evaluation information into each of the plurality of prediction models and selecting the prediction model with the smallest difference between the points of force applied and the target points of force determined by the operator that correspond to the evaluation information as the optimal prediction model; and calculating the predicted points of force for the planned tunneling section using the optimal prediction model. The learning information includes measurement information regarding the state of the shield tunneling machine when it has excavated a predetermined section, information indicating the direction of excavation, information indicating the direction of rotation of the cutter head, and a recommended point of force application for the jack thrust. The evaluation information includes the measurement information, the information indicating, the direction of rotation information, and the recommended point of force application corresponding to the latest target point of force application. [Effects of the Invention]

[0006] According to the present invention, the target point of force that the operator will determine can be predicted with high accuracy. [Brief explanation of the drawing]

[0007] [Figure 1] This figure shows a schematic configuration of one embodiment of a shield tunneling machine and direction control system. [Figure 2] This is a schematic front view of a shield tunneling machine. [Figure 3] This diagram shows the planned excavation area, excavation direction section, and measurement points for the shield tunneling machine. [Figure 4] This diagram shows the point of force application on the control panel of the shield control device. [Figure 5] (a) is a diagram showing the horizontal deviation of the shield tunneling machine from the planned alignment, and (b) is a diagram showing the vertical deviation of the shield tunneling machine from the planned alignment. [Figure 6] This is a functional block diagram showing the configuration of the direction control device. [Figure 7] This is a functional block diagram showing the configuration of the target point of emphasis recommendation section. [Figure 8] This is a flowchart showing the process for calculating the recommended point of force. [Figure 9] This is a functional block diagram showing the configuration of the target force point prediction unit. [Figure 10] This flowchart shows the process leading up to the calculation of the prediction stress points. [Figure 11] This diagram shows the items that make up learning information. [Figure 12] This diagram illustrates the "stroke difference (ideal value)" and "stroke difference (ideal value) - stroke difference". [Figure 13] This is an example of a graph showing the target horizontal point of force, the average value of the target horizontal point of force, the average value of the predicted horizontal point of force, and the average value of the recommended horizontal point of force. [Figure 14] This figure shows an example of the results of comparing the points earned by the predicted emphasis points with the points earned by the recommended emphasis points. [Modes for carrying out the invention]

[0008] Referring to Figures 1 to 14, one embodiment of a method for controlling the direction of a shield tunneling machine, a direction control system, and a method for calculating the predicted point of force will be described. First, referring to Figures 1 to 3, we will briefly explain the general outlines of the shield tunneling machine 20, the point of force application of the jack thrust, the actual point of force E, the recommended point of force R, the target point of force G, the drilling instruction value (drilling instruction sheet), and the predicted point of force N.

[0009] (Shield drilling machine) As shown in FIGS. 1 and 2, the shield tunneling machine 20 is used when forming the tunnel 10 by the shield method. In the shield method, after the tunnel 10 is formed by excavation with the shield tunneling machine 20, a lining is formed by a plurality of segments 11 so as to cover the inner wall surface of the tunnel 10.

[0010] The shield tunneling machine 20 includes an outer casing 21, shield jacks 22, and a shield control device 23. The outer casing 21 is cylindrical. The above-mentioned lining is formed at the rear side inside the outer casing 21. At the tip of the outer casing 21, a cutter head 24 that can rotate around the central axis of the outer casing 21 is provided. The shield jacks 22 are provided at predetermined intervals in the circumferential direction of the outer casing 21 and along the circumferential surface of the outer casing 21. The shield jacks 22 obtain the reaction force for excavation from the above-mentioned lining. The shield control device 23 controls the driving of the shield tunneling machine 20, such as the rotation of the cutter head 24 and the extension of each shield jack 22.

[0011] The shield tunneling machine 20 excavates the ground by extending the shield jacks 22 while rotating the cutter head 24. The excavation direction of the shield tunneling machine 20 is controlled by the position of the acting point of the jack thrust of the plurality of shield jacks 22.

[0012] (Acting point of jack thrust) As shown in FIG. 2, the position of the acting point is determined by the arrangement pattern (jack pattern) of the jack pressure set for each shield jack 22. Specifically, the position of the acting point is determined by the arrangement of the shield jacks 22 where the jack pressure contributing to the excavation of the shield tunneling machine 20 is set and the arrangement of the shield jacks 22 where the jack pressure not contributing to the excavation is set. The shield tunneling machine 20 excavates in the excavation direction corresponding to the acting point by changing the position of the acting point according to the jack pattern.

[0013] (Excavation instruction value) As shown in Fig. 3, when constructing a tunnel using the shield tunneling machine 20, the tunneling supervisor creates a tunneling instruction sheet with the tunneling instruction values set so that the shield tunneling machine 20 tunnels along the planned alignment L in the planned tunneling range L1. The tunneling instruction values are set for each of a plurality of tunneling direction sections L2 obtained by dividing the planned tunneling range L1. In Fig. 3, an example is shown in which the width of one ring of the segment 11 is taken as the tunneling direction section L2, and the tunneling direction section L2 of three sections is taken as the planned tunneling range L1.

[0014] The tunneling instruction values include the instruction value of the stroke difference and the instruction value of the pitching angle difference. Further, the tunneling instruction values include the instruction values for the gyro azimuth, the water level value, and the position and extension amount of the cutter head.

[0015] The stroke difference is the difference in the extension and contraction amounts (shield jack strokes) of the shield jacks 22 located at the left and right ends in the shield tunneling machine 20 as shown in Fig. 2. The pitching angle difference is the difference in the pitching angles before and after tunneling the tunneling direction section L2, specifically, the difference in the inclination angles of the shield tunneling machine 20 in the vertical direction before and after tunneling the tunneling direction section L2. Further, in the tunneling instruction sheet, in addition to the tunneling instruction values, the instruction value of the mid-fold angle of the shield jack 22 is set for each tunneling direction section L2. Note that the distance of the tunneling direction section L2 for which each instruction value is set and the number of sections of the planned tunneling range L1 are not limited in any way.

[0016] The operator of the shield tunneling machine 20 controls the tunneling direction of the shield tunneling machine 20 using the direction control system 30 so as to satisfy the tunneling instruction sheet, thereby tunneling the shield tunneling machine 20 along the planned alignment L.

[0017] (Recommended force point R, predicted force point N, target force point G, and actual force point E) As shown in Figure 4, the direction control system 30 calculates a recommended point of force R and a predicted point of force N as points of force application for advancing the shield tunneling machine 20 along the planned alignment L. The recommended point of force R is the point of force application recommended as the target point of force G based on past performance data. The predicted point of force N is a prediction of the position of the target point of force G that the operator will actually input in the next excavation direction section L2, based on past performance data. The operator of the shield tunneling machine 20 determines the target point of force G for jack thrust by referring to the calculated recommended point of force R and predicted point of force N, while also taking into account the performance of the shield tunneling machine 20, the ground conditions, and past experience.

[0018] The point of application of force consists of a horizontal point of application and a vertical point of application. The horizontal and vertical points of application are calculated based on the resultant force of the jack thrusts at each of the multiple shield jacks 22. The horizontal point of application is the point of application of the horizontal component and lies on a horizontal line passing through the central axis A of the shield tunneling machine 20. The vertical point of application is the point of application of the vertical component and lies on a vertical line perpendicular to the horizontal line and passing through the central axis A of the shield tunneling machine 20.

[0019] The direction control system 30 calculates the recommended point of force R based on the actual stroke difference, actual pitching angle difference, actual horizontal point of force, and actual vertical point of force obtained each time a section of the excavation direction L2 is excavated.

[0020] The actual stroke difference and actual horizontal force point are the stroke difference and horizontal force point after excavating one section of the excavation direction section L2. The actual pitching angle difference and actual vertical force point are the pitching angle difference and vertical force point after excavating one section of the excavation direction section L2.

[0021] The direction control system 30 generates multiple prediction models M (=M1, M2, ..., Mn: where n is the number of prediction models to be generated) and calculates the prediction emphasis point N using the optimal prediction model Mc selected from among the multiple prediction models M. The method for creating the prediction models M and the method for selecting the optimal prediction model Mc will be described later.

[0022] The operator of the shield tunneling machine 20 inputs the target point of force G, determined by referring to the recommended point of force R and the predicted point of force N, to the shield control device 23 via the direction control system 30. The shield control device 23 controls the jack pressure of each shield jack 22 so that the actual point of force E of the jack thrust follows the target point of force G.

[0023] (Directional control system for shield tunneling machines) The direction control system 30 will be described with reference to Figures 1 and 5-14. As shown in Figure 1, the direction control system 30 is configured to communicate with the shield control device 23 installed on the shield tunneling machine 20. The direction control system 30 includes a tunneling management device 40 and a direction control device 50, which are configured to communicate with each other.

[0024] The shield control device 23, the excavation management device 40, and the direction control device 50 are all centered around an information processing device. This information processing device can be implemented, for example, by one or more dedicated hardware circuits such as ASICs, one or more processing circuits that operate according to a computer program (software), or a combination of both. The processing circuit includes a CPU and memory (ROM and RAM, etc.) that stores the program executed by the CPU. Memory, or computer-readable media, includes any available media accessible by a general-purpose or dedicated computer.

[0025] (Excavation management device) The excavation management device 40 collects, calculates, records, and stores various information during excavation work based on the measured values ​​of various measuring instruments 25 mounted on the shield tunneling machine 20, and also monitors the operating status of the shield tunneling machine 20.

[0026] Examples of various measuring instruments 25 include, for example, laser oscillators, optical distance meters, and laser targets as measuring instruments for the horizontal and vertical directions. Other examples of measuring instruments for the horizontal direction include gyroscopes and stroke meters, and examples of measuring instruments for the vertical direction include water level meters and pitching meters.

[0027] As shown in Figures 5(a) and 5(b), the tunneling management device 40 grasps the position and orientation of the shield tunneling machine 20, as well as the amount of horizontal deviation Dh and vertical deviation Dv relative to the planned alignment L based on the construction plan.

[0028] Furthermore, the tunneling management device 40 calculates the horizontal and vertical points of force application for point E based on the jack thrust of the shield jack 22. Based on the measured values ​​of the measuring instrument 25 described above, the tunneling management device 40 acquires data related to the operating status of the shield tunneling machine 20, such as the current values ​​of the stroke difference and pitching angle difference of the shield tunneling machine 20.

[0029] As data regarding the operating status of the shield tunneling machine 20, the tunneling management device 40 acquires, for example, data for calculating the recommended point of force, which is used to calculate the recommended point of force R. In acquiring data for calculating the recommended point of force, the tunneling management device 40 measures the position of the shield tunneling machine 20 based on the measurements of various measuring instruments 25. The tunneling management device 40 may acquire data for calculating the recommended point of force after each section of the tunneling direction section L2 shown in Figure 3 has been excavated, or it may acquire data for calculating the recommended point of force at each measurement point P. Based on these measurement results, the tunneling management device 40 calculates the stroke difference and pitching angle difference of the shield tunneling machine 20, and the points of force application (horizontal point of force and vertical point of force) corresponding to these stroke difference and pitching angle difference. The tunneling management device 40 also acquires the rotation direction (right or left) of the cutter head 24 during measurement.

[0030] The excavation management device 40 then records the calculation results and data indicating the direction of rotation as recommended force point calculation data (actual stroke difference, actual pitching angle difference, actual horizontal force point, actual vertical force point, actual direction of rotation). The excavation management device 40 transmits the recommended force point calculation data to the direction control device 50 each time it acquires the recommended force point calculation data.

[0031] (Direction control device) As shown in Figure 6, the direction control device 50 includes an input device 51, an output device 52, a processing device 53, a file device 54, and a main memory 55.

[0032] The input device 51 is a device that can be operated by the operator, such as a keyboard, mouse, scanner, or switch. The output device 52 is a device that allows the operator to check various information, such as a display or printer.

[0033] The processing unit 53 performs various processes based on various programs and data stored in the main memory 55. The processing unit 53 includes a target force point recommendation unit 61 that calculates a recommended force point R, a target force point prediction unit 62 that calculates a predicted force point N, and a pattern extraction unit 63 that extracts a jack pattern, as functional units that operate through program execution.

[0034] The file device 54 is a storage device consisting of semiconductor memory or a hard disk drive, etc. The file device 54 stores an instruction data file 66. The instruction data file 66 stores the drilling instruction values ​​set in the drilling instruction sheet. The drilling instruction values ​​are input to the direction control device 50 via the input device 51 and saved in the instruction data file 66 each time a drilling instruction value is set.

[0035] The file device 54 stores a data file 67, a regression analysis information file 68, and a recommended effort point setting file 69 as storage sections related to the recommended effort point R. The file device 54 stores the training data file 70, the teacher data file 71, and the prediction model file 72 as memory sections related to the prediction effort point N.

[0036] The file device 54 stores jack pattern files 73 and the like. The jack pattern files store data that associates points of force application with jack patterns that embody those points of force application.

[0037] (Recommended emphasis points for target areas) Referring to Figures 7 and 8, the target effort point recommendation unit 61 of the direction control device 50 will be described. The target effort point recommendation unit 61 stores the recommended effort point calculation data transmitted by the excavation management device 40 in the data file 67 of the file device 54. The target effort point recommendation unit 61 calculates the recommended effort point R through regression analysis using the recommended effort point calculation data. The recommended effort point R consists of a recommended horizontal effort point Rh, which is the horizontal component, and a recommended vertical effort point Rv, which is the vertical component. When the amount of recommended effort point calculation data necessary for regression analysis is accumulated in the data file 67, the target effort point recommendation unit 61 executes the recommended effort point calculation process each time one section of the excavation direction section L2 is excavated.

[0038] As shown in Figure 7, the target point recommendation unit 61 has, as functional units that operate through program execution, an actual target value calculation unit 75, a regression analysis unit 76, a rotation information acquisition unit 77, a regression equation selection unit 78, and a recommended point calculation unit 79.

[0039] As shown in Figure 8, the recommended force point calculation process first performs the actual target value setting process (step S101). The actual target value setting process sets the actual target value for the stroke difference and the actual target value for the pitching angle difference.

[0040] In this process, the actual target value calculation unit 75 obtains the current value of the stroke difference (the latest value of the actual stroke difference) stored in the data file 67 of the file device 54. Similarly, it obtains the current value of the pitching angle difference (the latest value of the actual pitching angle difference). The actual target value calculation unit 75 also obtains the drilling instruction value for the next drilling direction section L2 from the instruction data file 66.

[0041] The actual target value calculation unit 75 calculates the difference between the acquired current value of the stroke difference and the drilling instruction value of the stroke difference as the actual target value of the stroke difference. The actual target value calculation unit 75 also calculates the difference between the current value of the pitching angle difference and the drilling instruction value of the pitching angle difference as the actual target value of the pitching angle difference. The actual target value calculation unit 75 stores the calculated actual target values ​​of the stroke difference and the actual target values ​​of the pitching angle difference in the recommended force point setting file 69 of the file device 54.

[0042] Next, a regression analysis process (step S102) is performed. The regression analysis process derives a regression equation capable of calculating the recommended horizontal point of force Rh and a regression equation capable of calculating the recommended vertical point of force Rv by performing a regression analysis using the data for calculating the recommended point of force.

[0043] In this process, the regression analysis unit 76 extracts the necessary number of recommended force point calculation data from the recommended force point calculation data stored in the data file 67, which represent the actual stroke difference when the cutter head 24 rotates to the right and the actual horizontal force point corresponding to that actual stroke difference. The regression analysis unit 76 performs regression analysis on the extracted recommended force point calculation data and obtains a regression equation that shows the relationship between the actual stroke difference and the actual horizontal force point when rotating to the right. In the same procedure, the regression analysis unit 76 also obtains a regression equation using the recommended force point calculation data when the cutter head 24 rotates to the left, and a regression equation that does not take rotation direction into account using the recommended force point calculation data that includes both right and left rotations. The regression analysis unit 76 stores the obtained regression equations in the regression analysis information file 68 as regression equations capable of calculating the recommended horizontal force point Rh.

[0044] Furthermore, the regression analysis unit 76 obtains a regression equation capable of calculating the recommended vertical point of force Rv using the same method as the regression equation used to calculate the recommended horizontal point of force Rh, based on the data for calculating the recommended point of force. The regression analysis unit 76 stores the obtained regression equation in the regression analysis information file 68 as a regression equation capable of calculating the recommended vertical point of force Rv.

[0045] Next, the regression equation selection process (step S103) is performed. The regression equation selection process is the process of selecting a regression equation. In this process, the rotation direction information of the cutter head 24 planned for the next excavation direction section L2 is input to the direction control device 50. When the rotation direction information is input, the rotation information acquisition unit 77 extracts two types of regression equations from the regression analysis information file 68: a regression equation corresponding to the input rotation direction (for right rotation or left rotation) and a regression equation that does not take the rotation direction into account.

[0046] Once the regression equations are extracted, the regression equation selection unit 78 selects a suitable regression equation from among the two types of regression equations for each of the recommended horizontal point of force Rh and the recommended vertical point of force Rv by any means. The regression equation selection unit 78 stores the selected regression equations in the recommended point of force setting file 69.

[0047] Next, a calculation process (step S104) is performed. The calculation process calculates the recommended point of force R using the actual target values ​​and regression equations stored in the recommended point of force setting file 69. In this process, the recommended point of force calculation unit 79 calculates the recommended horizontal point of force Rh by inputting the actual target value of the stroke difference as an explanatory variable into the regression equation selected for the recommended horizontal point of force Rh. The recommended point of force calculation unit 79 also calculates the recommended vertical point of force Rv by inputting the actual target value of the pitching angle difference as an explanatory variable into the regression equation selected for the recommended vertical point of force Rv. The recommended point of force calculation unit 79 stores the calculated recommended point of force R in the recommended point of force setting file 69 of the file device 54. As shown in Figure 4, the recommended point of force R may be transmitted from the direction control device 50 to the shield control device 23 and displayed on the selection screen 26 on the control panel of the shield control device 23.

[0048] (Target emphasis prediction section) Referring to Figures 9 and 10, the target point of force prediction unit 62 of the direction control device 50 will be described. Based on the model generation conditions, the target point of force prediction unit 62 generates multiple prediction models M that predict the target point of force G determined by the operator. The target point of force prediction unit 62 also selects the optimal prediction model Mc from among the multiple prediction models M based on the model selection conditions. Then, the target point of force prediction unit 62 calculates the predicted point of force N based on the optimal prediction model Mc. The predicted point of force N is composed of a predicted horizontal point of force Nh, which is the horizontal component, and a predicted vertical point of force Nv, which is the vertical component.

[0049] As shown in Figure 9, the target emphasis prediction unit 62 includes a training data creation unit 80, a teacher data creation unit 81, a machine learning unit 82, a model selection unit 83, and a predicted emphasis calculation unit 84, as functional units that operate through program execution.

[0050] As shown in Figure 10, in calculating the predicted emphasis point N, the target emphasis point prediction unit 62 performs the following steps: creation of training data (step S201), creation of teacher data (step S202), generation of a prediction model (step S203), selection of the optimal prediction model (step S204), and calculation of the predicted emphasis point (step S205).

[0051] In the creation of training data (step S201), the training data creation unit 80 acquires learning information and target emphasis information, and creates training data based on the acquired learning information and target emphasis information. The training data creation unit 80 stores the created training data in the training data file 70.

[0052] In the creation of training data (step S202), the training data creation unit 81 creates training data T corresponding to the training data and stores the created training data T in the training data file 71 of the file device 54.

[0053] In the generation of prediction models (step S203), the machine learning unit 82 generates multiple prediction models M by performing machine learning using training data T based on the model generation conditions. The machine learning unit 82 stores the generated multiple prediction models M in the prediction model file 72 of the file device 54.

[0054] In the selection of the optimal prediction model (step S204), the model selection unit 83 selects the optimal prediction model Mc from among the multiple prediction models M stored in the prediction model file 72 based on the model selection conditions.

[0055] In the calculation of prediction emphasis points (step S205), the prediction emphasis point calculation unit 84 calculates prediction emphasis points N based on the optimal prediction model Mc. (Learning information) Refer to Figure 11 to explain the learning information. The learning information is based on past excavations by the shield tunneling machine 20. The learning information includes measurement information, indication information, rotation direction information of the cutter head 24, and recommended effort point information regarding the recommended effort point R.

[0056] In the learning data, the measurement information is the data obtained when excavating a predetermined excavation direction section L2. The instruction information, rotation direction information, and recommended force point information are the data used when excavating the excavation direction section L2 immediately following the predetermined excavation direction section L2.

[0057] (Measurement information) The measurement information consists of measured values ​​taken by various measuring instruments 25 mounted on the shield tunneling machine 20, and calculated values ​​calculated using those measured values.

[0058] Among the measurement information, the explanatory variables that correspond to the measured values ​​and relate to both the horizontal and vertical directions are tail clearance (left, right, up, down), shield jack stroke (left, right), shield jack speed (left, right), cutter torque, and rolling.

[0059] Among the measurement information, the explanatory variables that correspond to the measured values ​​and relate only to the horizontal direction are the actual stroke difference, stroke difference, gyro direction, and folding angle (horizontal). Among the measurement information, the explanatory variables that correspond to the calculated values ​​calculated using the measured values ​​and relate only to the horizontal direction are the stroke difference (ideal value), stroke difference (ideal value) - stroke difference, gyro direction (ideal value), and gyro direction (ideal value) - gyro direction.

[0060] Among the measurement information, the explanatory variables that correspond to the measured values ​​and relate only to the vertical direction are the actual pitching difference, pitching, water level value, and mid-bend angle (vertical). Among the measurement information, the explanatory variables that correspond to the calculated values ​​calculated using the measured values ​​and relate only to the vertical direction are pitching (ideal value), pitching (ideal value) - pitching, water level value (ideal value), and water level value (ideal value) - water level value.

[0061] (Explanatory variables using ideal values) Refer to Figure 12 to explain explanatory variables using ideal values. Here, we will explain using the stroke difference (ideal value) and "stroke difference (ideal value) - stroke difference" as examples.

[0062] Figure 12 is a graph showing the stroke difference, stroke difference (excavation instruction value), and stroke difference (ideal value), with the horizontal axis representing the distance in the excavation direction and the vertical axis representing the magnitude of the stroke difference. The stroke difference (excavation instruction value) is the value set in the excavation instruction sheet and is set for each excavation direction section L2.

[0063] Here, the stroke difference when excavating in the excavation direction section L2 should ideally approach the stroke difference (excavation instruction value) step by step each time measurement point P is passed, and satisfy the stroke difference (excavation instruction value) when excavation in the excavation direction section L2 is completed. Therefore, if the stroke difference immediately before excavation is plotted at the starting point of the excavation direction section L2 and the stroke difference (excavation instruction value) is plotted at the ending point, the value on the ideal straight line connecting these points is the stroke difference (ideal value). Also, "stroke difference (ideal value) - stroke difference" is the difference between the stroke difference and the stroke difference (ideal value) on the ideal straight line.

[0064] Although not shown in the diagram, the explanatory variables "gyro direction (ideal value)" and "gyro direction (ideal value) - gyro direction", "pitching (ideal value)" and "pitching (ideal value) - pitching", "water level value (ideal value)", and "water level value (ideal value) - water level value" are also calculated based on the same concept.

[0065] (Instruction information) The instruction information consists of the drilling instruction value set in the drilling instruction sheet, the calculated value calculated using the drilling instruction value, and the instruction value for the bending angle of the shield jack 22 set in the drilling instruction sheet.

[0066] Among the instruction information, the explanatory variables that relate to both the horizontal and vertical directions are the copy stroke (excavation instruction value), the copy start range (excavation instruction value), and the copy end range (excavation instruction value).

[0067] Of the indicated information, the explanatory variables that relate only to the horizontal direction are the stroke difference (excavation indicated value), gyro direction (excavation indicated value), and the indicated value of the bending angle (horizontal). Of the indicated information, the explanatory variables that are calculated using the excavation indicated value and relate only to the horizontal direction are the stroke difference (excavation indicated value) - stroke difference and the gyro direction (excavation indicated value) - gyro direction.

[0068] Among the indicator information, the explanatory variables that correspond to the drilling indicator value and relate only to the vertical direction are the pitching difference (drilling indicator value), the water level value (drilling indicator value), and the indicated value of the midpoint angle (vertical). Among the indicator information, the explanatory variables that are calculated using the drilling indicator value and relate only to the vertical direction are the pitching difference (drilling indicator value) - pitching and the water level value (drilling indicator value) - water level value.

[0069] (Recommended emphasis information) The recommended point of force information is information about the recommended point of force R. Of the recommended point of force information, the explanatory variables related to the horizontal direction are recommended point of force (recommended horizontal point of force (X coordinate)) and recommended point of force (gyro) X. Of the recommended point of force information related to recommended point of force R, the explanatory variables related to the vertical direction are recommended point of force (recommended vertical point of force (Y coordinate)) and recommended point of force (level) Y.

[0070] (How to obtain training data) Training data includes learning information and target force point information. Training data can be acquired either by acquiring data for each excavation direction section L2, or by acquiring data for each measurement point P.

[0071] (When acquiring training data for each section L2 in the excavation direction) In this case, each time the shield tunneling machine 20 excavates in the excavation direction section L2, the excavation work is stopped, and then measurements are taken using the measuring instrument 25. The training data creation unit 80 acquires measurement information based on these measurements. The training data creation unit 80 also acquires rotation direction information of the cutter head 24 immediately before measurement. Next, the training data creation unit 80 acquires instruction information for the next excavation direction section L2. The training data creation unit 80 also acquires the recommended force point R calculated by the target force point recommendation unit 61 for the next excavation direction section L2 as recommended force point information. Through this series of processes, the training data creation unit 80 acquires learning information including measurement information, rotation direction information, instruction information, and recommended force point information. The training data creation unit 80 may acquire learning information based on data input from the excavation management device 40, or it may acquire learning information using other equipment.

[0072] Next, the training data creation unit 80 acquires target point of force information. The training data creation unit 80 may acquire the final target point of force G in the excavation direction section L2 where excavation has been completed as target point of force information, or it may acquire the target point of force G entered by the operator for the planned excavation section, which is the next excavation direction section L2, as target point of force information. Then, the training data creation unit 80 saves the acquired learning information and data indicating the target point of force information as training data in the training data file 70.

[0073] In this embodiment, various variables were categorized as described above, but the categorization of these variables is not limited in any way. For example, variables related to various "ideal values" may be categorized as instruction information, and "stroke difference (excavation instruction value) - stroke difference" may be categorized as measurement information. Also, for example, various "instruction values ​​for bending angles" may be categorized as recommended force point information.

[0074] The shield tunneling machine 20 uses the target force point G entered by the operator to excavate the next excavation direction section L2. The acquisition of training data and the excavation work by the shield tunneling machine 20 are repeated until the shield tunneling machine 20 reaches the target point. As a result, the training data creation unit 80 stores a quantity of training data in the training data file 70 that corresponds to the number of sections of the excavation direction section L2 that have been excavated.

[0075] <When acquiring training data for each measurement point> In this case, the training data creation unit 80 acquires training data each time the machine reaches one of the multiple measurement points P located along the excavation direction section L2. In this case, the training data creation unit 80 acquires measurement information, instruction information, rotation direction information of the cutter head 24, recommended point of force R, and target point of force information each time the shield tunneling machine 20 passes through a measurement point P.

[0076] During excavation in the excavation direction section L2, if there are no changes in the indicated information, the rotation direction of the cutter head 24, the recommended point of force application R, and the target point of force application G (target point of force application information), this information will be the same at all measurement points P in the predetermined excavation direction section L2.

[0077] On the other hand, if the operator determines that there is a problem with the shield tunneling machine 20's ability to follow the drilling instruction value, they may change the target force point G or the rotation direction of the cutter head 24 at that point. In such cases, the training data creation unit 80 stores the changed information as training data for each measurement point P, along with the corresponding target force point information. The training data creation unit 80 creates training data in quantities equivalent to the product of the number of measurement points P set in the drilling direction section L2 and the number of sections (rings) in the drilling direction section L2. Any method may be used to confirm the ability to follow the drilling instruction value and to adjust the drilling instruction value, but for example, the method for adjusting the drilling direction of a shield tunneling machine disclosed in Japanese Patent Application No. 2019-165405 can be used.

[0078] In this way, the training data creation unit 80 creates training data in conjunction with the excavation work of the shield tunneling machine 20. The training data creation unit 80 then stores the training data in the training data file 70 as a series of continuous data that is continuous in the direction of excavation of the shield tunneling machine 20. Therefore, when performing data cleansing on the training data, if there are gaps, for example, in the tail clearance of the learning information, it is possible to linearly interpolate the data in the missing portion by using the training data before and after the missing portion.

[0079] (Training Data Creation Department) The training data creation unit 81 generates training data T based on the training data stored in the training data file 70. The training data creation unit 81 creates training data T from training data that has undergone the necessary data cleansing in advance.

[0080] The training data T is data used in machine learning. The training data T consists of training data Th and training data Tv. The training data creation unit 81 creates training data Th by combining the explanatory variables (horizontal direction) of the learning information and the corresponding target horizontal point of force Gh as ground truth data. The training data creation unit 81 also creates training data Tv by combining the explanatory variables (vertical direction) of the learning information and the corresponding target vertical point of force Gv as ground truth data. The training data creation unit 81 stores the created training data Th and Tv in the training data file 71 as a series of continuous data that is continuous in the tunneling direction of the shield tunneling machine 20.

[0081] If training data is acquired for each section L2 in the tunneling direction, the teacher data creation unit 81 creates a quantity of teacher data T equivalent to the number of sections (rings) in the tunneling direction section L2. On the other hand, if training data is acquired for each measurement point P, the teacher data creation unit 81 creates a quantity of teacher data T equivalent to the product of the number of measurement points P set up in the tunneling direction section L2 and the number of sections (rings) in the tunneling direction section L2. If the tunneling direction section L2 is treated as a segment ring and one ring is considered to be one unit of segment data, it is advisable to accumulate at least 100 rings worth of teacher data T.

[0082] (Machine Learning Department) The machine learning unit 82 performs machine learning using training data T by employing one of the machine learning methods (algorithms). Examples of algorithms include Random Forest, Deep Learning, Neural Network, and Support Vector Machine (SVM). In this embodiment, Random Forest is adopted as the machine learning method.

[0083] The machine learning unit 82 generates a prediction model Mh used to calculate the predicted horizontal point of force Nh by performing machine learning using the training data Th. Similarly, the machine learning unit 82 generates a prediction model Mv used to calculate the predicted vertical point of force Nv by performing machine learning using the training data Tv.

[0084] Furthermore, the machine learning unit 82 generates multiple prediction models Mh,Mv based on the model generation conditions. The model generation conditions specify which sections of training data T to use to generate the prediction models Mh,Mv, with the excavation direction section L2, from which the latest target point information is obtained, as the reference section. The model generation conditions specify the number of sections to be referenced based on the reference section. For example, the model generation conditions specify that the training data T be classified into five patterns: the most recent 10 sections, 30 sections, 50 sections, 70 sections, and 100 sections of the reference section, and prediction models Mh,Mv are generated. In this case, the machine learning unit 82 generates prediction models Mh,Mv using training data from the most recent 10 sections of the reference section, prediction models Mh,Mv using training data from the most recent 30 sections of the reference section, and so on. The machine learning unit 82 stores each of the generated prediction models Mh,Mv in the prediction model file 72 of the file device 54.

[0085] (Model selection section) The model selection unit 83 selects the optimal prediction models Mch and Mcv for each direction from among the prediction models Mh and Mv generated by the machine learning unit 82. Specifically, the model selection unit 83 performs accuracy verification based on the model selection conditions and selects the prediction model Mh and Mv with the highest accuracy as the optimal prediction model Mch and Mcv. The model selection unit 83 uses the training data for the excavation direction section L2, for which the latest target force point information has been obtained, as evaluation data.

[0086] In verifying accuracy in the horizontal direction, the model selection unit 83 calculates the average value of the target horizontal force point Gh for the evaluation data. When training data is acquired for each excavation direction section L2, the average value of the target horizontal force point Gh will be the value representing the target horizontal force point Gh of the latest target force point G entered by the operator.

[0087] Next, the model selection unit 83 inputs the training information contained in the evaluation data as evaluation information to the prediction model M generated by the machine learning unit 82, thereby calculating the predicted horizontal force point Nh based on each prediction model M. The model selection unit 83 then calculates the average value of the calculated predicted horizontal force point Nh for each prediction model M.

[0088] Figure 13 is an example of a graph showing the target horizontal point of force, the average value of the target horizontal point of force, and the average value of the predicted horizontal point of force for evaluation data acquired at each measurement point P, with the vertical axis representing the horizontal point of force and the horizontal axis representing the distance in the direction of excavation.

[0089] The model selection unit 83 then calculates the area S1 enclosed by the average value of the point of force 0 and the target horizontal point of force Gh, and the area S2 enclosed by the average value of the point of force 0 and each predicted horizontal point of force Nh. Having calculated various areas, the model selection unit 83 selects the prediction model Mh that calculates the area S2 with the smallest difference from area S1 as the optimal prediction model Mch. The model selection unit 83 then performs accuracy verification in the vertical direction using a similar method and selects the optimal prediction model Mcv from among multiple prediction models Mv.

[0090] (Predictive strength calculation unit) The prediction force point calculation unit 84 calculates the predicted horizontal force point Nh by inputting the explanatory variables related to the horizontal direction in the latest training data to the optimal prediction model Mch. The prediction force point calculation unit 84 also calculates the predicted vertical force point Nv by inputting the explanatory variables related to the vertical direction in the latest training data to the optimal prediction model Mcv. As shown in Figure 4, the predicted force point N may be transmitted from the direction control device 50 to the shield control device 23 and displayed on the selection screen 26 on the control panel of the shield control device 23.

[0091] Figure 14 shows the results of comparing the predicted point of force N of the optimal prediction model Mc with the recommended point of force R. In this comparison, for the horizontal direction, the area enclosed by the average value of the recommended horizontal point of force Rh in the evaluation data and the point of force position 0 was calculated. Then, the area of ​​the predicted horizontal point of force Nh (S2) and the area of ​​the recommended horizontal point of force Rh were compared, and the one with the smaller difference from the area of ​​the target horizontal point of force Gh was awarded a "score of 1". Similarly, for the vertical direction, the one with the smaller difference from the area of ​​the target vertical point of force Gv was awarded a "score of 1". Then, for both the horizontal and vertical directions, the score was calculated using hundreds of different patterns.

[0092] As shown in Figure 14, in the horizontal direction, the predicted horizontal point of force Nh scored approximately three times higher than the recommended horizontal point of force Rh, and in the vertical direction, the predicted vertical point of force Nv scored approximately twice as high as the recommended vertical point of force Rv. In other words, the predicted point of force N by the optimal prediction model Mc is closer to the target point of force G determined by the operator than the recommended point of force R.

[0093] (Pattern extraction unit) The pattern extraction unit 63 extracts jack patterns from the jack pattern file 73 that approximate the points of force application for each of the recommended point of force R, predicted point of force N, and target point of force G. These extracted jack patterns may be transmitted from the direction control device 50 to the shield control device 23 and displayed on the selection screen 26 on the control panel of the shield control device 23.

[0094] (action) When the shield tunneling machine 20 excavates the planned excavation section, the operator determines the target point of force G for the planned excavation section, referring to the recommended point of force R and predicted point of force N calculated by the direction control system 30. The operator inputs the determined target point of force G to the shield control device 23 via the direction control system 30. The shield control device 23 controls the jack pressure of each shield jack 22 so that the actual point of force E of the jack thrust follows the target point of force G. As a result, the shield tunneling machine 20 excavates the planned excavation section so that the actual point of force E follows the target point of force G.

[0095] The effects of this embodiment will now be explained. (1) According to the direction control system described above, when excavating in a section with many changes in excavation conditions, such as a curved section, the predicted point of force N is calculated based on training data of the excavation direction section L2 that is close to the planned excavation section, i.e., training data with similar excavation conditions. Also, when there are consecutive sections with few changes in excavation conditions, such as a straight section, the predicted point of force N can be calculated based on training data with similar excavation conditions and a larger amount of training data. In other words, according to the direction control system 30, even if the excavation conditions change, the predicted point of force N can be calculated based on training data obtained under similar excavation conditions. As a result, the position of the predicted point of force N can be set to a position close to the target point of force G determined by the operator. Furthermore, since the optimal prediction model Mc can be updated each time the planned excavation section is excavated, the predicted point of force N can be calculated according to the excavation conditions at that time.

[0096] (2) The learning information includes numerous items that may affect directional control. Therefore, by continuously performing excavation using the directional control system 30, a prediction model M can be generated that ensures sufficient accuracy for the predicted point of force N. This makes it possible to use the predicted point of force N instead of the target point of force that was previously determined by a skilled operator while considering the performance of the shield tunneling machine and the ground conditions. As a result, this can greatly contribute to the realization of automated operation of the shield tunneling machine.

[0097] This embodiment can be implemented with the following modifications. This embodiment and the following modifications can be combined with each other to the extent that they do not contradict each other technically. In the above embodiment, the number of sections referenced based on the reference section was used as the model generation condition. However, the model generation condition may be configured to classify training data according to the curvature of the planned excavation section, for example. In this case, the training data can be classified as linear elements of the planned excavation section, such as the stroke difference (ideal value) and pitching (ideal value) in the training information. With such a configuration, the predicted effort point can be calculated with even higher accuracy.

[0098] In the above embodiment, the shield control device 23 controls the jack pressure of each shield jack 22 with a jack pattern based on the target point of force G input by the operator. However, the shield control device 23 may be configured to control the jack pressure of each shield jack 22 based on a jack pattern manually instructed by the operator so that the target point of force G determined by the operator becomes the actual point of force E.

[0099] In the above embodiment, the training data creation unit 80 acquired the target point of force G input by the operator as target point of force information. However, the training data creation unit 80 may also acquire the actual point of force E as target point of force information indicating the target point of force determined by the operator, for example, when the operator manually instructs the jack pattern. In this case, the teacher data creation unit 81 creates teacher data using the actual point of force E as the correct answer data. [Explanation of Symbols]

[0100] 10...Tunnel, 11...Segment, 20...Shield tunneling machine, 21...Outer shell, 22...Shield jack, 23...Shield control device, 24...Cutter head, 25...Measuring instrument, 30...Direction control system, 40...Tunneling management device, 50...Direction control device, 51...Input device, 52...Output device, 53...Processing device, 54...File device, 55...Main memory, 61...Target point recommendation unit, 62...Target point prediction unit, 63...Pattern extraction unit, 66...Instruction data file, 67...Data file, 68...Regression analysis information file, 69...Recommended emphasis point setting file, 70...Training data file, 71...Teacher data file, 72...Predictive model file, 73...Jack pattern file, 75...Actual target value calculation unit, 76...Regression analysis unit, 77...Rotation information acquisition unit, 78...Regression equation selection unit, 79...Recommended emphasis point calculation unit, 80...Training data creation unit, 81...Teacher data creation unit, 82...Machine learning unit, 83...Model selection unit, 84...Predictive emphasis point calculation unit.

Claims

1. A method for controlling the direction of a shield tunneling machine, which controls the tunneling direction of the shield tunneling machine based on a predicted point of force, which is the point of force application of the jack thrust applied to the shield tunneling machine and is a target point of force input by the operator when tunneling in the planned tunneling section, The direction control system of the shield tunneling machine, A process of creating training data by combining training information with target effort points corresponding to said training information, which are determined by the operator, as correct answer data; The process of generating multiple predictive models by classifying the training data into multiple patterns based on model generation conditions and performing machine learning for each of the multiple patterns, The process of selecting the optimal prediction model from among the multiple prediction models, which has the smallest difference between the point of force applied obtained by inputting evaluation information into each of the aforementioned prediction models and the target point of force applied corresponding to the evaluation information, which is the target point of force applied determined by the operator. The process of calculating the predicted force point for the planned excavation section using the aforementioned optimal prediction model is performed. The aforementioned learning information is, This includes measurement information regarding the state of the shield tunneling machine when it has excavated a predetermined section, information indicating the direction of excavation, information regarding the rotation direction of the cutter head, and a recommended point of force application for the jack thrust. The aforementioned evaluation information is, The measurement information, instruction information, rotation direction information, and recommended point of force corresponding to the latest target point of force are included. The model generation conditions specify the number of intervals to be referenced based on the interval in which the latest target point was obtained. A method for controlling the direction of a shield tunneling machine.

2. A method for controlling the direction of a shield tunneling machine, which controls the tunneling direction of the shield tunneling machine based on a predicted point of force, which is the point of force application of the jack thrust applied to the shield tunneling machine and is a target point of force input by the operator when tunneling in the planned tunneling section, The direction control system of the shield tunneling machine, A process of creating training data by combining training information with target effort points corresponding to said training information, which are determined by the operator, as correct answer data; The process of generating multiple predictive models by classifying the training data into multiple patterns based on model generation conditions and performing machine learning for each of the multiple patterns, The process of selecting the optimal prediction model from among the multiple prediction models, which has the smallest difference between the point of force applied obtained by inputting evaluation information into each of the aforementioned prediction models and the target point of force applied corresponding to the evaluation information, which is the target point of force applied determined by the operator. The process of calculating the predicted force point for the planned excavation section using the aforementioned optimal prediction model is performed. The aforementioned learning information is, This includes measurement information regarding the state of the shield tunneling machine when it has excavated a predetermined section, information indicating the direction of excavation, information regarding the rotation direction of the cutter head, and a recommended point of force application for the jack thrust. The aforementioned evaluation information is, The measurement information, instruction information, rotation direction information, and recommended point of force corresponding to the latest target point of force are included. The aforementioned model generation condition is a change in the excavation conditions. A method for controlling the direction of a shield tunneling machine.

3. A shield tunneling machine direction control system that controls the tunneling direction of the shield tunneling machine based on a predicted point of force, which is the point of force application of the jack thrust applied to the shield tunneling machine and is a target point of force input by the operator when tunneling in the planned tunneling section, A training data creation unit creates training data by combining training information with target points determined by the operator that correspond to the training information, as correct answer data. A machine learning unit that generates multiple predictive models by classifying the training data into multiple patterns based on model generation conditions and performing machine learning for each of the multiple patterns, A model selection unit selects the optimal prediction model from among the multiple prediction models, which has the smallest difference between the point of force applied obtained by inputting evaluation information into each of the aforementioned prediction models and the target point of force applied corresponding to the evaluation information, which is the target point of force applied determined by the operator. The system includes a prediction force point calculation unit that calculates the prediction force point for the planned excavation section using the aforementioned optimal prediction model, The aforementioned learning information is, This includes measurement information regarding the state of the shield tunneling machine when it has excavated a predetermined section, information indicating the direction of excavation, information regarding the rotation direction of the cutter head, and a recommended point of force application for the jack thrust. The aforementioned evaluation information is, The measurement information, instruction information, rotation direction information, and recommended point of force corresponding to the latest target point of force are included. The model generation conditions specify the number of intervals to be referenced based on the interval in which the latest target point was obtained. Direction control system for a shield tunneling machine.

4. A shield tunneling machine direction control system that controls the tunneling direction of the shield tunneling machine based on a predicted point of force, which is the point of force application of the jack thrust applied to the shield tunneling machine and is a target point of force input by the operator when tunneling in the planned tunneling section, A training data creation unit creates training data by combining training information with target points determined by the operator that correspond to the training information, as correct answer data. A machine learning unit that generates multiple predictive models by classifying the training data into multiple patterns based on model generation conditions and performing machine learning for each of the multiple patterns, A model selection unit selects the optimal prediction model from among the multiple prediction models, which has the smallest difference between the point of force applied obtained by inputting evaluation information into each of the aforementioned prediction models and the target point of force applied corresponding to the evaluation information, which is the target point of force applied determined by the operator. The system includes a prediction force point calculation unit that calculates the prediction force point for the planned excavation section using the aforementioned optimal prediction model, The aforementioned learning information is, This includes measurement information regarding the state of the shield tunneling machine when it has excavated a predetermined section, information indicating the direction of excavation, information regarding the rotation direction of the cutter head, and a recommended point of force application for the jack thrust. The aforementioned evaluation information is, The measurement information, instruction information, rotation direction information, and recommended point of force corresponding to the latest target point of force are included. The aforementioned model generation condition is a change in the excavation conditions. Direction control system for a shield tunneling machine.

5. A method for calculating a predicted force point, which is the point of force application of the jack thrust applied to a shield tunneling machine, and predicts the target force point input by the operator during the excavation of the planned excavation section. The device that calculates the predicted force point, A process of creating training data by combining training information with target effort points corresponding to said training information, which are determined by the operator, as correct answer data; The process of generating multiple predictive models by classifying the training data into multiple patterns based on model generation conditions and performing machine learning for each of the multiple patterns, The process of selecting the optimal prediction model from among the multiple prediction models, which has the smallest difference between the point of force applied obtained by inputting evaluation information into each of the aforementioned prediction models and the target point of force applied corresponding to the evaluation information, which is the target point of force applied determined by the operator. The process of calculating the predicted force point for the planned excavation section using the aforementioned optimal prediction model is performed. The aforementioned learning information is, This includes measurement information regarding the state of the shield tunneling machine when it has excavated a predetermined section, information indicating the direction of excavation, information regarding the rotation direction of the cutter head, and a recommended point of force application for the jack thrust. The aforementioned evaluation information is, The measurement information, instruction information, rotation direction information, and recommended point of force corresponding to the latest target point of force are included. The model generation conditions specify the number of intervals to be referenced based on the interval in which the latest target point was obtained. Method for calculating prediction emphasis points.

6. A method for calculating a predicted force point, which is the point of force application of the jack thrust applied to a shield tunneling machine, and predicts the target force point input by the operator during the excavation of the planned excavation section. The device that calculates the predicted force point, A process of creating training data by combining training information with target effort points corresponding to said training information, which are determined by the operator, as correct answer data; The process of generating multiple predictive models by classifying the training data into multiple patterns based on model generation conditions and performing machine learning for each of the multiple patterns, The process of selecting the optimal prediction model from among the multiple prediction models, which has the smallest difference between the point of force applied obtained by inputting evaluation information into each of the aforementioned prediction models and the target point of force applied corresponding to the evaluation information, which is the target point of force applied determined by the operator. The process of calculating the predicted force point for the planned excavation section using the aforementioned optimal prediction model is performed. The aforementioned learning information is, This includes measurement information regarding the state of the shield tunneling machine when it has excavated a predetermined section, information indicating the direction of excavation, information regarding the rotation direction of the cutter head, and a recommended point of force application for the jack thrust. The aforementioned evaluation information is, The measurement information, instruction information, rotation direction information, and recommended point of force corresponding to the latest target point of force are included. The aforementioned model generation condition is a change in the excavation conditions. Method for calculating prediction emphasis points.

Images