Route prediction apparatus, method for route prediction, and route prediction program

Abstract

To control the excavation of a shield tunneling machine with higher accuracy.SOLUTION: The route prediction apparatus includes: a basic model unit having a basic model for predicting a route of a shield tunneling machine on the basis of parameters for driving the shield tunneling machine; a difference prediction model generation unit for generating a learned difference prediction model by causing artificial intelligence to perform machine learning, on the basis of the difference between a predicted result of the route of the shield tunneling machine obtained from the basic model and an actual reaching position of the shield tunneling machine; and a prediction unit for generating a composite model combining the basic model and the difference prediction model to predict the tip position and an orientation of the shield tunneling machine.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a course prediction device, a course prediction method, and a course prediction program. [Background technology]

[0002] In the above technical field, Non-Patent Document 1 discloses that a method for operating a shield machine for a planned alignment including curves and a method for allocating multiple segments of different shapes are simulated in advance to set these planned values. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Shimizu Corporation Develops AI-Based Shield Tunneling Planning Support System, Corporate Information, May 25, 2018, [Retrieved December 17, 2023], Internet (URL: https: / / www.shimz.co.jp / company / about / news-release / 2018 / 2018005 / html) Summary of the Invention [Problem to be solved by the invention]

[0004] However, the technique described in Non-Patent Document 1 above was unable to predict the path of the shield machine with higher accuracy. [Means for solving the problem]

[0005] In order to achieve the above object, a course prediction device according to the present invention comprises: a basic model unit that uses a basic model to predict the path of the shield machine based on parameters for excavating the shield machine; a differential prediction model generation unit that uses artificial intelligence to machine-learn the difference between the predicted path of the shield machine and the actual position reached by the shield machine based on the basic model, and generates a learned differential prediction model; a prediction unit that generates a composite model of the basic model and the differential prediction model to predict the tip position and orientation of the shield machine; Equipped with.

[0006] In order to achieve the above object, a course prediction method according to the present invention includes: a basic model generation step of predicting a path of the shield machine using a basic model, in order to predict the path of the shield machine based on parameters for excavating the shield machine; a differential prediction model generation step in which, based on the basic model, artificial intelligence is trained to learn the difference between the predicted path of the shield machine and the position actually reached by the shield machine, thereby generating a trained differential prediction model; a prediction step of generating a composite model of the basic model and the differential prediction model to predict the tip position and orientation of the shield machine; Includes:

[0007] Furthermore, in order to achieve the above object, the course prediction program according to the present invention comprises: a basic model generation step of predicting a path of the shield machine using a basic model, in order to predict the path of the shield machine based on parameters for excavating the shield machine; a differential prediction model generation step in which, based on the basic model, artificial intelligence is trained to learn the difference between the predicted path of the shield machine and the position actually reached by the shield machine, thereby generating a trained differential prediction model; a prediction step of generating a composite model of the basic model and the differential prediction model to predict the tip position and orientation of the shield machine; to be executed by the computer. [Effects of the Invention]

[0008] According to the present invention, the path of a shield tunneling machine can be predicted with greater accuracy. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a diagram for explaining an outline of the operation of a course prediction device according to a preferred embodiment of the present invention; [Figure 2] 1 is a block diagram illustrating the configuration of a course prediction device according to a preferred embodiment of the present invention. [Figure 3] 1 is a diagram illustrating a hardware configuration of a course prediction device according to a preferred embodiment of the present invention. [Figure 4] 3 is a flowchart illustrating a processing procedure of a course prediction device according to a preferred embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present invention will be described in detail by way of example with reference to the drawings. However, the configurations, numerical values, processing flows, functional elements, etc. described in the following embodiments are merely examples, and are open to modification and alteration, and are not intended to limit the technical scope of the present invention to the following description.

[0011] A path prediction device according to a preferred embodiment of the present invention will be described with reference to Figs. 1 to 5. Fig. 1 is a diagram for explaining an overview of a path prediction device 100 according to this embodiment. The path prediction device 100 predicts the excavation of a shield machine 1113 using a basic model 111 generated to predict the path according to the operation details of the operator of the shield machine 113, and a difference prediction model 112 that has learned the deviation (difference) in the path of the shield machine 113 predicted by the basic model 111. In other words, the path prediction device 100 causes the shield machine 113 to excavate using a composite model 114 of the basic model 111 and the difference prediction model 112. The basic model 111 used is one that has been generated in advance.

[0012] The differential prediction model 112 is generated by machine learning using artificial intelligence in the learning section 110. Then, in the construction section 120, the path prediction device 100 controls the excavation of the shield tunneling machine 113 using a combined model 114 of the basic model 111 and the differential prediction model 112. The path prediction device 100 may be implemented in the shield tunneling machine 113, or may be a server, PC, mobile terminal, or the like that can communicate with the shield tunneling machine 113 via wired or wireless communication. The basic model 111 is not trained even when it is moved to a different construction site, and the same basic model 111 is repeatedly used. The basic model 111 may be adjusted in advance before the learning section 110 by considering the conversion coefficients of input parameters and calculation offset values ​​in its internal calculation formulas.

[0013] Next, the configuration of the path prediction device 100 will be described with reference to Fig. 2. The path prediction device 100 has a basic model unit 201, a differential prediction model generation unit 202, a composite output generation unit 203, and a prediction unit 204.

[0014] Basic model unit 201 predicts the path of shield machine 113 using a basic model 111 created in advance to predict the path of shield machine 113 from parameters for excavating shield machine 113. The parameters refer to data from the shield machine (such as reaction forces acting on the jack and cutter plate) and the details of the operator's operations.

[0015] First, a basic model 111 is generated that predicts the path of the shield machine 113 from parameters input to cause the machine to excavate. Basic model 111 is a model for predicting the path of the shield machine using a calculation method devised based on conventional methods, with parameters such as the pressure of the shield jack input to the shield machine and the bending angle caused by the extension and contraction of the bending jack as input data. Basic model 111 can then be made into a model (basic model 111) that can predict the path of the shield machine 113 with a certain level of accuracy or higher by correcting the theoretical formula, for example by adding a coefficient to the calculation method, with reference to the operation of the shield machine 113 by a veteran engineer.

[0016] Based on the above calculation method, the basic model 111 predicts the path of the shield machine 113 based on parameters (rotational moment and articulation angle) that have a significant effect on the excavation distance and direction of the shield machine 113. Here, the rotational moment is calculated as a moment from the thrust direction and thrust force based on the arrangement and operating status of the thrust jacks (which jacks are in which positions and with what pressure they are pushing the segments). The articulation angle is calculated by accumulating the thrust direction due to the articulation angle between the front and rear bodies, starting from the point where the path changes, and the effective articulation angle is calculated from the articulation angle caused by the extension of the articulation jacks and the delay distance (the distance from when the articulation jacks are extended until the direction actually changes), and this is compared with the excavation vector (D n-1 ) is used as input data, and the current excavation vector (D n ) is calculated. The amount of overcutting (extension of the copy cutter) also has a large effect on the excavation direction, but since this is a process carried out before bending, it is difficult to determine its effect, and so it is not used as a parameter. However, if a method is established that can determine the effect of overcutting, this does not prevent it from being used as a parameter.

[0017] The differential prediction model generation unit 202 uses the basic model 111 to predict the path of the shield tunneling machine 113, and uses artificial intelligence to machine-learn the difference between the predicted position where the shield tunneling machine 113 will arrive and the actual position where the shield tunneling machine 113 arrives, thereby generating a learned differential prediction model 112.

[0018] In other words, the differential prediction model generation unit 202 trains the artificial intelligence to learn characteristics such as soil quality and other characteristics specific to the excavation site. Similarly, the differential prediction model generation unit 202 trains the artificial intelligence to learn characteristics such as machine characteristics specific to the shield machine 113 used for excavation, i.e., the behavior of the shield machine 113 that cannot be covered by the general-purpose basic model 111.

[0019] The differential prediction model 112 is generated in the learning section 110. The learning section 110 is a section for generating the differential prediction model 112, and is a section for having the artificial intelligence learn the above-mentioned difference by machine learning. The length (distance) of the learning section 110 is a length corresponding to several rings of connected segments, for example, a length corresponding to four to five connected rings, but is not limited to this. The differential prediction model generation unit 202 generates the differential prediction model 112 by having the artificial intelligence learn the difference while lining four to five rings of segments.

[0020] The differential prediction model generation unit 202 then uses artificial intelligence to machine-learn the weighting of the parameters to generate a trained differential prediction model 112. Parameters input to cause the shield tunneling machine 113 to excavate include, but are not limited to, the rotation moment and the bending angle, for example.

[0021] Here, the rotational moment is calculated as a moment from the direction of thrust and thrust force based on the arrangement and operating conditions of the thrust jacks (which jacks are in which positions and with what pressure they are pushing the segments). The bending angle is calculated by accumulating the direction of thrust due to the bending angle between the front and rear bodies, starting from the point where the course changes. The effective bending angle is calculated from the bending angle caused by the extension of the bending jacks and the delay distance (the distance from when the bending jacks are extended until the actual change in direction), and this is compared with the excavation vector (D n-1 ) as input data, and the current driving vector (D n ) is calculated. The amount of overcutting (extension of the copy cutter) also has a large effect on the excavation direction, but since this is a process carried out before bending, it is difficult to determine its effect, and so it is not used as a parameter. However, if a method is established that can determine the effect of overcutting, this does not prevent it from being used as a parameter.

[0022] The differential prediction model generation unit 202 causes the artificial intelligence to learn the weighting of these input parameters. By learning the weighting, it becomes possible to distinguish and learn between parameters that have a large and small impact on the path of the shield machine 113.

[0023] The composite output generation unit 203 generates a composite output by combining an output obtained by inputting parameters including at least the rotation moment and the bending angle into the basic model 111 and an output obtained by inputting the output from the basic model 111 into the differential prediction model 112.

[0024] That is, in the construction section 120 after learning in the learning section 110 described above is completed, the combined output generation unit 203 inputs the output from the basic model 111 as an explanatory variable into the differential prediction model 112. Then, the path prediction device 100 excavates while predicting the path of the shield tunneling machine 113, using a combined output obtained by combining the output from the differential prediction model 112 and the output from the basic model 111.

[0025] Parameters that have a significant effect on the excavation distance and excavation direction of the shield machine 113 are input to the basic model 111. The input parameters are, for example, the rotation moment and the bending angle. These parameters are input either alone or together.

[0026] The composite output generation unit 203 sets the weighting coefficients of the shield jack placement and operating status, the bending angle, delay distance, etc., which are input data for the rotational moment and bending angle model, to values ​​lower than those of the other major parameters (for example, 1 / 2 or less of the parameter with the highest weighting coefficient), and inputs these to the differential prediction model 112. In other words, by preventing these parameters, which have a significant impact on the excavation of the shield tunneling machine 113, from having too great an impact on the differential prediction model 112, it is possible to take into account the impact of parameters other than these.

[0027] Prediction unit 204 predicts the path of shield machine 113 using composite model 114 of basic model 111 and differential prediction model 112. That is, prediction unit 204 predicts the path of shield machine 113 using an output that combines the output of basic model 111 and the output of differential prediction model 112. In construction section 120 beyond learning section 110, prediction unit 204 uses basic model 111 to make a basic path prediction of shield machine 113, and uses differential prediction model 112 to correct the difference (error) from the prediction by basic model 111. Prediction unit 204 predicts the path of shield machine 113 by adding a correction value to the path predicted by basic model 111 to correct the difference from this path.

[0028] The hardware configuration of the trajectory prediction device 100 will be described with reference to FIG. 4. The CPU (Central Processing Unit) 410 is a processor for arithmetic control, and by executing programs, it realizes the various functional components of the trajectory prediction device 100 shown in FIG. 2. The CPU 410 may have multiple processors and execute different programs, modules, tasks, threads, etc. in parallel. The ROM (Read Only Memory) 420 stores fixed data such as initial data and programs, as well as other programs. The network interface 430 communicates with other devices via a network. The CPU 410 is not limited to a single CPU, and may include multiple CPUs or a GPU (Graphics Processing Unit) for image processing. The network interface 430 preferably has a CPU independent of the CPU 410 and writes and reads transmitted and received data to and from an area of ​​the RAM (Random Access Memory) 440. It is also preferable to provide a DMAC (Direct Memory Access Controller) (not shown) for transferring data between the RAM 440 and the storage 450. The CPU 410 recognizes that data has been received or transferred to the RAM 440 and processes the data accordingly. The CPU 410 also prepares the processing results in the RAM 440, and leaves the subsequent transmission or transfer to the network interface 430 or DMAC.

[0029] The RAM 440 is a random access memory used by the CPU 410 as a temporary storage work area. The RAM 440 has a storage area reserved for storing data necessary for implementing this embodiment. The input parameter values ​​441 are data necessary for predicting the path of the shield machine 113, such as the details of operations by the operator of the shield machine 113 input by the operator or various data transmitted from the shield machine 113. The actual arrival position data 442 is data on the desired position of the shield machine 113. The arrival position data 443 is data on the position predicted to be reached if the shield machine 113 is caused to excavate using the basic model 111. The difference data 444 is data related to the difference (error) between the actual arrival position and the position predicted to be reached using the basic model 111. The combined output data 445 is a combination of the output obtained by inputting parameters to the basic model 111 and the output obtained by inputting the output from the basic model 111 and various parameters to the difference prediction model 112.

[0030] The transmitted / received data 446 is data transmitted and received via the network interface 430. The RAM 440 also has an application execution area 447 for executing various application modules.

[0031] The storage 450 stores databases and various parameters, as well as the following data or programs required to implement this embodiment.

[0032] Storage 450 further stores basic model module 451, difference prediction model generation module 452, and prediction module 453. The basic model unit is a module that uses basic model 111 to predict the path of shield machine 113 based on parameters for excavating shield machine 113. Difference prediction model generation module 452 is a module that uses artificial intelligence to machine-learn the difference between the predicted path of shield machine 113 and the actual position reached by shield machine 113, based on basic model 111, to generate trained difference prediction model 112. Prediction module 453 is a module that generates a combined model of basic model 111 and difference prediction model 112 to predict the tip position and orientation of shield machine 113. These modules 451 to 453 are loaded into application execution area 447 of RAM 440 by CPU 410 and executed. Control program 454 is a program for controlling the entire path prediction device 100.

[0033] The input / output interface 460 interfaces input / output data with input / output devices. A display unit 461 and an operation unit 462 are connected to the input / output interface 460. A storage medium 464 may also be connected to the input / output interface 460. A speaker 463 serving as an audio output unit, a microphone (not shown) serving as an audio input unit, or a GPS position determination unit may also be connected. Note that the RAM 440 and storage 450 shown in FIG. 4 do not include programs or data relating to the general-purpose functions of the course prediction device 100 or other feasible functions.

[0034] Next, the processing procedure of the path prediction device 100 will be described with reference to the flowchart shown in Fig. 5. This flowchart is executed by the CPU 410 in Fig. 4 using the RAM 440, and realizes each functional configuration of the path prediction device 100 in Fig. 2.

[0035] In step S501, basic model unit 201 predicts the path of shield machine 113 using basic model 111 for predicting the path of shield machine 113. In step S503, differential prediction model generation unit 202 uses artificial intelligence to learn the difference between the predicted path of shield machine 113 and the position actually reached by shield machine 113, based on basic model 111, to generate trained differential prediction model 112.

[0036] In step S505, the prediction unit 204 creates a composite model 114 of the basic model 111 and the differential prediction model 112. In S507, the composite model 114 is used to predict the tip position and orientation of the shield machine 113.

[0037] According to this embodiment, the shield machine is caused to excavate using the basic model in the learning section, so it can excavate with a certain degree of accuracy even in the learning section, making it possible to generate a highly accurate differential prediction model. Furthermore, in the construction section, the composite model is used to predict the path and the shield machine is caused to excavate, so the shield machine can excavate along the excavation plan line. Furthermore, because the shield machine is caused to excavate using the basic model in the learning section, the distance of the learning section required to generate the differential prediction model can be shortened, making it possible to generate a highly accurate differential prediction model in a short period of time.

[0038] While the present invention has been described above with reference to the embodiments, it is not limited to the above-described embodiments and can be modified as appropriate. The configuration and details of the present invention can be modified in various ways that are understandable to those skilled in the art and are within the scope of the present invention. Furthermore, systems or devices that combine the individual features included in each embodiment in any way are also included in the scope of the present invention. The input parameters of the basic model may include, in addition to the rotation moment and the center bend angle, parameters that take into account the influence of the amount of overcutting, for example. Furthermore, the weighting coefficients for the rotation moment and the center bend angle in the differential prediction model only need to be lower than the other major parameters, and may be set to predetermined values, for example.

[0039] The present invention may also be applied to a system consisting of multiple devices or to a single device. Furthermore, the present invention may also be applied when an information processing program that realizes the functions of the embodiments is supplied to a system or device and executed by a built-in processor. Therefore, the technical scope of the present invention also includes a program installed on a computer to realize the functions of the present invention, a medium storing the program, a WWW (World Wide Web) server from which the program is downloaded, and a processor that executes the program. In particular, the technical scope of the present invention also includes a non-transitory computer-readable medium storing a program that causes a computer to execute at least the processing steps included in the above-described embodiments.

Claims

1. a basic model unit that uses a basic model to predict the path of the shield machine based on parameters for excavating the shield machine; a differential prediction model generation unit that uses artificial intelligence to machine-learn the difference between the predicted path of the shield machine and the position that the shield machine actually reaches, based on the basic model, to generate a learned differential prediction model; and a prediction unit that generates a composite model of the basic model and the differential prediction model to predict the tip position and orientation of the shield machine; A course prediction device equipped with the above.

2. a composite output generation unit that generates a composite output by combining an output obtained by inputting the parameters including at least the rotation moment and the bending angle into the basic model and an output obtained by inputting the output from the basic model into the differential prediction model; The path prediction device according to claim 1 , wherein the prediction unit predicts the position and orientation of the tip of the shield machine based on the generated composite output.

3. The differential prediction model generation unit further generates the trained differential prediction model by subjecting the weighting coefficients of the parameters to machine learning using artificial intelligence; The course prediction device according to claim 2 , wherein the composite output generation unit sets weighting coefficients for the rotation moment and the bending angle to values ​​lower than other parameters and inputs the weighting coefficients to the basic model.

4. a basic model step of predicting a path of the shield machine using a basic model, in order to predict the path of the shield machine based on parameters for excavating the shield machine; a differential prediction model generation step in which, based on the basic model, artificial intelligence is trained to learn the difference between the predicted path of the shield machine and the position actually reached by the shield machine, thereby generating a trained differential prediction model; a prediction step of generating a composite model of the basic model and the differential prediction model to predict the tip position and orientation of the shield machine; A career prediction method including:

5. a basic model step of predicting a path of the shield machine using a basic model, in order to predict the path of the shield machine based on parameters for excavating the shield machine; a differential prediction model generation step in which, based on the basic model, artificial intelligence is trained to learn the difference between the predicted path of the shield machine and the position actually reached by the shield machine, thereby generating a trained differential prediction model; a prediction step of generating a composite model of the basic model and the differential prediction model to predict the tip position and orientation of the shield machine; A career prediction program that runs on a computer.

Images