Information processing systems and programs

Abstract

To provide a system, method and program that make it possible to practically simulate a behavior of an excavating machine.SOLUTION: An information processing system 30 for reproducing a behavior of an excavating machine, includes: a prediction unit 40 that predicts operation results of the excavating machine when controlled by control information using control information output from a learned model used to control the excavating machine and adjustable parameters; an acquisition unit 41 that acquires the operation results of the excavating machine when actually controlled by the control information; a verification unit 42 that verifies the predicted operation results based on the acquired operation information; and an adjustment unit 43 that adjusts the parameters in accordance with verification results.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] This invention relates to an information processing system for reproducing the behavior of an excavating machine and a program for causing a computer to execute the reproduction process. [Background technology]

[0002] A technology is known for automating the control of excavation machinery such as shield machines (hereinafter simply referred to as shields) using a trained model that has been trained using machine learning (see, for example, Patent Document 1). When shield excavation control is performed using a trained model, if there are abnormal values ​​or missing values ​​in the data input to the trained model, the safety and quality of the construction will deteriorate. Therefore, in the above technology, as a criterion for predicting failure risk assessment, it is determined whether all of the measured values ​​are within the range of the training data of the measured values ​​in the training data used when training the learning model (i.e., whether they are in an interpolated state), and only if the evaluation is passed is the set value of the control data for the operation of the shield estimated using the trained model.

[0003] When training a learning model, if data where the measured values ​​deviate from the instructed values ​​is used as training data, the model may not be able to learn to make correct estimations, and the learning model may not estimate appropriate setting values ​​for operations corresponding to the measured values.

[0004] Therefore, a technique has been proposed to determine whether or not to use the data to be judged, which includes operational performance data and operational measurement data, as training data for a learning model that estimates the set values ​​for the operation of the excavating machine, based on the degree of discrepancy between the situation measurement data, which measures the excavation status, and the indicated value, which is the target value for excavation (see, for example, Patent Document 2). [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2019-143388 [Patent Document 2] Japanese Patent Publication No. 2019-143389 [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] By combining the two conventional technologies described above, the validity and accuracy of automating shield excavation control using a trained model can be improved.

[0007] In real-world excavation projects, conditions such as alignment and geological features often change significantly during construction, and the specifications of excavation machinery vary from project to project. Therefore, the training data that a single company can collect has diverse backgrounds. With such training data, it is difficult to improve the generalization performance of the learning model, resulting in frequent failures in prediction failure risk assessments.

[0008] To manage the risk of prediction failure, simulations are necessary when the output of a trained model is used as input to a real device. However, simulators that theoretically reproduce shield behavior to date have resulted in complex models that include many unknown parameters, making them too complex to be practical and thus not yet in practical use. [Means for solving the problem]

[0009] The present invention has been made in view of the above problems, and is an information processing system for reproducing the behavior of an excavating machine, A prediction means that predicts the operation result of an excavator when controlled by the control information, using control information output from a trained model used to control the excavator and adjustable parameters. An acquisition means for obtaining the operating results of an excavator when it is actually controlled by control information, A verification means for verifying the predicted operation result based on acquired operation information, A means for adjusting parameters according to the verification results. An information processing system including this is provided. [Effects of the Invention]

[0010] According to the present invention, it becomes possible to simulate the behavior of an excavation machine in practical use.

Brief Description of the Drawings

[0011] [Figure 1] The figure which showed the structural example of the shield as an example of an excavation machine. [Figure 2] The figure which illustrated how the shield turns on a sharp curve. [Figure 3] The figure explaining data assimilation. [Figure 4] The figure which showed an example of the hardware configuration of an information processing system. [Figure 5] The functional block diagram which showed an example of the functional configuration of an information processing system. [Figure 6] The flowchart which showed an example of the excavation control of the shield. [Figure 7] The flowchart which showed an example of the process of estimating the operation result of the shield. [Figure 8] The table which summarized the variables related to the simulation. [Figure 9] The figure which showed an example of a probability distribution.

Modes for Carrying Out the Invention

[0012] FIG. 1 is a diagram showing a configuration example of a shield as an example of an excavation machine. An excavation machine is a machine that excavates earth and sand, rock, the ground, etc. When the entire cross-section of a tunnel is the excavation target, there are a shield, a tunnel boring machine, a free cross-section excavation machine, etc. The excavation machine may be any machine as long as it can excavate the entire cross-section of the tunnel. Hereinafter, the excavation machine will be described as a shield.

[0013] The shield 10 is a tunnel excavation machine that excavates the earth and sand in front, holds down the excavation face that is about to collapse, and discharges the excavated earth and sand in an amount that balances the amount of earth and sand excavated to the outside of the pit and advances. Behind the shield, tunnel lining blocks called segments are assembled in a ring shape to complete the tunnel structure.

[0014] The shield 10 has a cutter head 11 at its tip for cutting the front earth and sand, and the cutter head 11 is provided with a plurality of cutting bits arranged in a cylindrical or radial pattern on a rotatable substantially circular face plate. The shield 10 also includes a shield jack 12, and advances by pressing the shield jack 12 against the assembled segment 13 and extending the shield jack 12. A plurality of shield jacks 12 are arranged at predetermined intervals along the inner circumference of a skin plate 14 which is a steel outer cylinder constituting the shield 10.

[0015] In the development of complex urban underground spaces, it is required to excavate three-dimensional sharp curves. For this reason, the shield 10 is provided with a folding mechanism such as a middle folding jack 15 that enables an intermediate fold so that the skin plate 14 is divided into two parts, a front body 14a and a rear body 14b, and the front body 14a and the rear body 14b are connected by a hinge to be able to cope with sharp curves. Similar to the shield jack 12, a plurality of middle folding jacks 15 are arranged at predetermined intervals along the inner circumference of the front body 14a of the skin plate 14.

[0016] The shield 10 includes a chamber 16 immediately behind the cutter head 11 as a stirring chamber for plastically fluidizing the excavated earth and sand while applying earth pressure or water pressure to hold the excavation face, so that the excavated earth and sand can be discharged. The excavated earth and sand in the chamber 16 are taken out by a screw conveyor 17, carried out of the pit by a belt conveyor 18, and disposed of as discharged soil.

[0017] The shield 10 includes an erector 19 for assembling the segment 13 inside the rear body 14b. The erector 19 grips the segment 13, conveys it to a predetermined position, and installs it. The shield 10 also includes a backfill injection device 20 for injecting an injection material into the back of the segment 13 (between the segment 13 and the excavated tunnel wall surface).

[0018] The shield 10 moves forward by extending the shield jacks 12. Furthermore, by changing the length (stroke) of each shield jack 12, which is arranged in multiple locations along the inner circumference of the shield 10 and skin plate 14, and, if necessary, by changing the stroke of each folding jack 15, which is arranged in multiple locations along the inner circumference of the skin plate 14, the shield moves forward by turning in a predetermined direction. The direction of forward movement is the direction in the horizontal direction and the direction in the vertical direction relative to the horizontal direction (pitching).

[0019] Figure 2 illustrates how a shield with a folding mechanism bends a sharp curve. The angle formed by the center line passing through the center of the cross-section of the front body 14a and the center line passing through the center of the cross-section of the rear body 14b when the shield is folded by the folding jack 15 is defined as the folding angle α. When the shield 10 bends a sharp curve using the folding jack 15, the orientation of the rear body 14b and the folding angle α are manipulated to adjust the posture of the shield 10 so that it moves along the excavation direction in which the deviation between the shield 10 and the planned line is minimized.

[0020] When navigating sharp curves, vehicles like buses and trucks require space greater than their width to accommodate the inner wheel difference and overhang from the center of rotation. Similarly, Shield 10 requires a similar space, expanded from the excavation diameter of the straight section, to curve. However, it cannot curve unless it creates this extra space by excavating it itself.

[0021] Therefore, a copy cutter 21 is provided on the side of the cutter head 11 located on the front shell 14a of the shield 10, as a projection that can protrude radially into the tunnel. The copy cutter 21 is a device for creating a space in which the shield 10 can curve.

[0022] When the shield 10 makes a sharp turn, the folding jack 15 on the inside of the sharp curve is retracted and the folding jack 15 on the outside is extended, causing the shield to curve in a roughly V-shaped bending position. To achieve this, it is necessary to anticipate the possible positions of the shield 10 and excavate (over-excavate) when the copy cutter 21 is in that position.

[0023] The space left over from the excavation is not left as a void, but is filled immediately after excavation with a soil conditioner that has appropriate plastic fluidity. Leaving it as a void poses a risk of the excavation wall collapsing, making it difficult for the shield 10 to receive appropriate reaction forces and friction from the excavation wall, and thus making the operation of the shield 10 difficult.

[0024] The movement of the copy cutter 21 is controlled continuously in accordance with the rotation angle of the cutter head 11. This control can be implemented, for example, using an equation that shows the relationship between the protrusion length of the copy cutter 21 and a predetermined circular angle.

[0025] While the excavation control of such shield 10 can be performed manually, in recent years it has been possible to automate it using a pre-trained machine learning model. The machine learning model takes measured values ​​(observed values) of monitoring items that monitor the behavior of shield 10 as input data and outputs control information used for controlling the excavation of shield 10. In order for the model to predict control information from such input data, it is trained by being given training data, which consists of pairs of input data and control information obtained through manual control.

[0026] The essence of machine learning is to make predictions (interpolation) within the range learned from the training data. Therefore, a trained model (pre-trained model) should not be used outside the range it was trained for (extrapolation). Unintended extrapolation increases the risk of prediction failure.

[0027] The training data collected by a single company has diverse backgrounds because the specifications of the Shield 10 vary depending on the ground being excavated, and conditions such as alignment and geology often change significantly during construction. For example, even when excavating the same mountain, conditions such as geology can vary greatly due to the presence of large rocks or groundwater.

[0028] A trained model identifies patterns and trends from input data and makes predictions for unknown data. The training data collected by artificially controlling Shield 10 is data where geological and other conditions fluctuate greatly. Therefore, even if a training model is trained using this training data, the examples of patterns and trends that should be covered are insufficient, and the model does not converge to sufficient training results, making it unable to handle unknown data. As a result, we face a situation where there is no prospect of improving the generalization ability, which is the ability to handle unknown data.

[0029] As a result, there is often a shortage of training data, leading to frequent failures in prediction failure risk assessments, and the learned model is not utilized for excavation control. To make the learned model useful, repeated additional training for correction is required, resulting in poor productivity.

[0030] To manage the risk of prediction failure, a simulation is needed to test what happens when the output of a trained model is actually input into Shield 10 and manipulated. For the simulation, a simulator is used as a device or program that describes natural laws and rules of thumb based on existing knowledge in mathematical formulas and returns a response that mimics the actual machine to virtual operations.

[0031] However, since available knowledge is limited, it is impossible to create a simulator that reproduces all real-world phenomena. Therefore, conventionally, a simulator is considered practical if it can realistically reproduce the phenomenon of interest. For example, a flight simulator that allows the pilot to raise or lower the nose of the aircraft and change its tilt by operating the control stick is known as a practical simulator.

[0032] In the case of Shield 10, there is a need for a simulator for pre-assessment when excavating unknown sharp curves. To simulate the behavior of Shield 10, it is necessary to create a situation similar to its behavior using mathematical models, etc. However, since the model is composed of ground with many unknown elements, where contributions from interactions such as friction and pressure with the surrounding ground are dominant, the model becomes complex and includes many unknown parameters. Because such a complex model including many unknown parameters is difficult to handle and reproduce, it has not yet been put into practical use.

[0033] Therefore, instead of attempting to reproduce a model that includes numerous unknown parameters, we adopt a method called assimilation simulation, which involves introducing tunable parameters and adjusting them to match actual observational results if the simulation fails to reproduce the desired outcome.

[0034] Here, we will explain the concept of assimilation with reference to Figure 3. The horizontal axis represents elapsed time, and the vertical axis represents values ​​such as predicted values ​​obtained from simulations or observed values ​​obtained by operating the actual machine, with the unknown true state shown by a dotted line. Suppose we prepare a simulator that roughly reproduces the true state and perform a simulation from the initial value (a) shown by the double circle, and obtain the prediction shown by the solid line, and the predicted value (b) shown by the white circle at elapsed time t deviates significantly from the observed value (c) shown by the black circle. Note that the observed value (c) also deviates from the true state shown by the dashed line, but this is due to observation error.

[0035] If we continue the simulation based on the predicted value (b) at elapsed time t, it is predicted that the result will deviate even more significantly from the observed value. Therefore, we adjust the parameters based on the observed value (c) at elapsed time t, modify the simulator model to reduce the deviation at elapsed time t, and obtain the modified model, i.e., perform data assimilation. After performing data assimilation, the predicted value (b) is corrected to the analyzed value (d) shown by the double circle, and the result from the analyzed value (d) to the predicted value (e) shown by the white circle is shown by the solid line, which is closer to the true state shown by the dashed line.

[0036] Since there is no prospect of improving the generalization performance of the learning model used for drilling control of Shield 10, once a certain level of generalization performance is achieved, the trained model will be put into drilling control, and the results of that application will be verified by a simulation incorporating the data assimilation described above. Depending on whether the simulation results are acceptable or not, it will be possible to decide whether to use automated drilling control with the trained model or manual control. This makes it possible to manage the risk of prediction failure under automated drilling control using the learning model.

[0037] Therefore, the present invention provides an information processing system capable of performing simulations incorporating the above-mentioned data assimilation. The information processing system may be the same as the one used for training a learning model used for excavation control of the shield 10, or for performing predictions using the trained model.

[0038] Figure 4 shows an example of the hardware configuration of the information processing system 30. The information processing system 30 can be configured using a general-purpose computer. Therefore, the information processing system 30 can adopt a hardware configuration similar to that of a general-purpose computer. The information processing system 30 includes, as hardware, a CPU (Central Processing Unit) 31, ROM (Read Only Memory) 32, RAM (Random Access Memory) 33, HDD (Hard Disk Drive) 34, external device I / F 35, input / output I / F 36, display device 37, and input device 38.

[0039] The CPU 31 controls the entire system and executes various applications. These applications may include simulators that perform the above-mentioned simulations. The ROM 32 stores the BIOS (Basic Input / Output System), which loads the OS (Operating System) during system startup and controls input / output to peripheral devices, as well as firmware that controls internal system circuits such as the HDD 34. The RAM 33 is used as main memory and provides a workspace for the CPU 31. The HDD 34 stores various applications executed by the CPU 31, the OS, various configuration information, and various data. Although an HDD 34 is used here, it is not limited to this, and other storage devices such as an SSD (Solid State Drive) may also be used.

[0040] External device I / F 35 connects this system to external devices such as control panels and various environmental measurement devices (sensors) and controls communication with these external devices. Display device 37 is a device for displaying information such as liquid crystal displays and organic EL (Electro-Luminescence) displays. Input device 38 is a device for inputting information from a keyboard or mouse, selecting applications, and issuing instructions to execute applications. Input / output I / F 36 controls the output of information to the display device 37 and the input of information from the input device 38. In this example, the display device 37 and the input device 38 are described as separate devices, but this is not the only option; a touch panel with the functions of both the display device 37 and the input device 38 may be used.

[0041] The information processing system 30 may also include other circuits, such as a communication interface that connects to a network such as the Internet and controls communication with communication devices on the network, or a short-range wireless communication circuit that enables short-range wireless communication using Bluetooth® or the like.

[0042] The information processing system 30 may consist of one device or program, or it may consist of two or more devices or programs.

[0043] Figure 5 is a functional block diagram showing an example of the functional configuration of the information processing system 30. Each function of the information processing system 30 is realized by the CPU 31 executing programs such as applications provided by the information processing system 30. Note that some or all of the functions of the information processing system 30 may be implemented as hardware such as circuits.

[0044] The information processing system 30 includes, as functional units, a prediction unit 40, an acquisition unit 41, a verification unit 42, and an adjustment unit 43. The prediction unit 40 uses control information output from a trained model used for excavation control of the shield 10 and adjustable parameters to predict the operation result of the shield 10 when excavation is controlled by the control information. A simulator is used to predict the operation result of the shield 10. The simulator is created as a mathematical model, such as a formula, which includes adjustable parameters.

[0045] The trained model is a model trained using training data. The training data is data that associates the manipulative quantities of the shield 10 actually operated by a person with the observed values ​​measured from the results of that operation. Examples of manipulative quantities include the bending angle α, the thrust force of the shield jack 12, the rotational torque of the cutter head 11, and the stroke of the copy cutter 21. Examples of observed values ​​include the orientation of the shield 10 and the pitching, which indicates the vertical rotation angle. Therefore, by measuring the orientation and pitching of the shield 10 using sensors and inputting this data into the trained model, the trained model can consider the planned tunnel line and output the manipulative quantities as control information. By setting this control information on the shield 10, the excavation control of the shield 10 can be automated.

[0046] While a trained model is constructed to best reproduce the training data through training, it can suffer from overfitting, where it only matches the training data. Overfitting occurs when there is insufficient training data or when the model is too complex. Therefore, test data separate from the training data is prepared, and it is checked whether the trained model can reproduce the prepared test data (the model's generalization performance).

[0047] The acquisition unit 41 acquires the operation results of the shield 10 when excavation is actually controlled according to the control information. The operation results of the shield 10 are observed values ​​measured using the sensors, etc.

[0048] The verification unit 42 verifies the operation result predicted by the prediction unit 40 based on the information acquired by the acquisition unit 41, i.e., the observed values. The prediction unit 40 takes the control information output by the trained model as input, performs a simulation, and predicts the direction and other values ​​measured by the sensors mentioned above. The verification unit 42 compares the predicted operation result with the observed values ​​and verifies the validity of the predicted operation result based on whether the difference is greater than or equal to a threshold.

[0049] The adjustment unit 43 adjusts the parameters according to the verification results. If the verification unit 42 determines that the difference between the predicted operating result and the observed value is greater than a threshold and that the predicted result is not valid, the adjustment unit 43 adjusts the parameters. The adjusted parameters are set in the mathematical model that the prediction unit 40 uses for simulation, and the mathematical model is modified.

[0050] The information processing system 30, including the acquisition unit 41, verification unit 42, and adjustment unit 43 described above, is a system that incorporates data assimilation techniques and, in principle, has the characteristic of being able to follow environmental changes and discontinuous situations, which is a feature of assimilation. However, in extreme cases where no prior knowledge is incorporated at all, some of the characteristics of assimilation will be lost. Nevertheless, since simulations can be performed using mathematical formulas that are simply linear combinations of the parameters of interest, it has the characteristic of being easier to handle compared to existing models.

[0051] The information processing system 30 is not limited to the prediction unit 40, acquisition unit 41, verification unit 42, and adjustment unit 43, but may also include other functional units. Other functional units of the information processing system 30 may include, for example, a learning unit that trains a learning model using training data, a determination unit that confirms generalization performance, and a prediction execution unit that performs predictions using the trained model.

[0052] Furthermore, the information processing system 30 may also include, as other functional units, an evaluation unit that evaluates the result of control based on the control information output from the trained model by determining whether or not the result is within an acceptable range.

[0053] Figure 6 is a flowchart showing the excavation control flow of Shield 10. Before performing excavation control, the learning model, parameters to be set in the simulation, etc. are prepared. Excavation control of Shield 10 starts from step 100. In step 101, manual control is performed to acquire training data. In step 102, training data is acquired, and in step 103, the learning model is trained using the acquired training data.

[0054] In step 104, the generalization performance of the learned model is checked using test data prepared separately from the training data. That is, it is checked whether the control information can be predicted with a predetermined accuracy and whether the predetermined generalization performance is achieved. The accuracy of the prediction can be any accuracy and can be determined arbitrarily. If the generalization performance is checked and the prediction is made with the predetermined accuracy, proceed to step 105; otherwise, return to step 102 and repeat the learning process.

[0055] In step 105, the system is manually controlled, and observed values ​​measured by sensors attached to the shield 10 are used as input to predict control information using a trained model. In step 106, the predictions from the trained model are used as input to perform a simulation, the predictions are corrected as needed, and the predicted or corrected predictions are output.

[0056] In step 107, based on the outputted prediction, automatic excavation control (predictive control) is performed using the trained model. Even if the predictive control proceeds in an impossible direction, the behavior of shield 10 is slow, and there is ample time to correct it. For this reason, predictive control is performed at this stage.

[0057] In step 108, the control results of the predictive control are evaluated. If the control results of the predictive control are within an acceptable range, indicating that the shield 10 is advancing at an acceptable angle relative to the planned line, the process proceeds to step 109. On the other hand, if it is outside the acceptable range, the process returns to step 101, where manual control is performed to correct the direction in which the shield 10 is advancing, etc. Then, in step 102, training data is acquired, and in step 103, the learning model is retrained.

[0058] In step 109, it is determined whether or not the drilling is complete. If it is complete, the process proceeds to step 110 to terminate the drilling control. If it is not complete, the process returns to step 105 to predict the control information for the next time step.

[0059] Figure 7 is a flowchart showing an example of the simulation in step 106 shown in Figure 6. Starting from step 200, in step 201, the simulation is performed using the predictions of the trained model as input. The simulation is performed by inputting the above predictions into a mathematical model that includes adjustable parameters. The simulation virtually controls and operates the shield 10 using the predicted control information, and the result of the operation is output. In step 202, the control information predicted by the trained model is actually set on the shield 10, and the result of the excavation control is obtained as an observed value. The obtained observed value is then compared with the simulation result. Based on the comparison, it is determined whether the prediction is valid or not based on whether the difference between the observed value and the simulation result is greater than or equal to a threshold.

[0060] If the above difference exceeds a threshold and the prediction is deemed invalid, proceed to step 203 to correct the prediction. That is, adjust the parameters used in the simulation to correct the prediction so that the simulation results are closer to the observed values.

[0061] In step 202, if the prediction is deemed valid, the valid prediction is output. If the prediction is deemed invalid, the corrected prediction is output in step 203, and the simulation is terminated in step 204.

[0062] Next, data assimilation will be explained in more detail with an example. As procedure (1), prepare the variables shown in Figure 8 as variables involved in the simulation. Note that the variables shown in Figure 8 are just an example, so you may prepare only some of these variables, or you may prepare other variables, or you may prepare other variables in addition to these.

[0063] The variables shown in Figure 8 can be broadly divided into three categories: variables indicating the manipulated amount of the shield 10, variables indicating the operating environment, and variables indicating the effects of operation. The variables used as control information are those indicating the manipulated amount and those indicating the operating environment, while the observed values ​​measured by sensors, etc., are variables indicating the effects of operation.

[0064] After preparing the variables involved in the simulation, step (2) involves preparing a mathematical formula that is assumed to hold true between the prepared variables. In the Shield 10 simulator, theoretical construction is difficult, so a hypothetical formula such as Equation 1 below can be used. Equation 1 below is a formula that outputs the effect of operation when a certain manipulated variable is applied in a certain operating environment.

[0065]

number

[0066] After preparing the mathematical formula, in step (3), the simulator is given degrees of freedom to track real-world observation results. These degrees include a time-varying parameter θ for the manipulated variable and the variable representing the operating environment, i.e., θ X1H ~θ X10H θ F1H ~θ F6H θ X1V ~θ X10V θ F1V ~θ F6V We introduce a total of 32 parameters as coefficients. Then, equation 1 above can be rewritten as equation 2 below.

[0067]

number

[0068] After introducing the parameters, as step (4), perform a simulation. When giving the operation effects ΔH(t) and ΔV(t) at a certain time t, ΔH and ΔV obtained in the same way at the previous times t - 1, t - 2, ···, 2, 1, and the initial value S H (0), S V (0) are used, the state after the operation at time t can be expressed as shown in Equation 3 below.

[0069]

Equation

[0070] Here, the output S(t) of the simulator is described as shown in Equation 4 below.

[0071]

Equation

[0072] In the parameter adjustment, by comparing the output of the simulator with the actually obtained observed values, the reality following property of the simulator is evaluated and corrected. Therefore, an observation error w expected in the actual observed values is added in advance.

[0073]

Equation

[0074] In Equation 5 above, Y(t) (with a ^ on the top of Y) is the predicted observed value given by the simulator. For w, for example, a sample from a normal distribution with an average of 0 and a variance of ν is given.

[0075] If the simulation results indicate that parameter θ needs adjustment at time t, then step (5) is performed to adjust θ (evaluate and correct the simulator's ability to track reality). A dataset is prepared containing the simulator's predicted observed value Y(t) (with a ^ symbol at the top of Y), the actual observed value Y(t), the manipulated variable F(t) considered to be the cause of these values, and the operating environment X(t). Here, the initial value of θ used when calculating Y(t) for the first time before any θ adjustment has been performed is a value sampled from an appropriate probability distribution.

[0076] A suitable probability distribution p(x) could be a uniform distribution, such as the one shown in Figure 9, where x is distributed with the same probability of 1 / 2 in the interval from -1 to 1. Note that this is just one example and is not the only possible distribution.

[0077] After preparing the dataset, θ is adjusted. This adjustment uses a simulator optimization technique called assimilation. There are many known algorithms for assimilation, and details of each algorithm are omitted here. In the Shield 10 simulator, for example, a method using a Gaussian kernel can be applied.

[0078] In assimilation, the value of θ in an equation that is unsatisfactory in its initial state is successively modified in a way that is satisfactory. Here, θ is not a fixed specific number, but rather a realized value sampled from a certain probability distribution π(θ). Therefore, the actual object to be modified in assimilation is the probability distribution π(θ). If the probability distribution before modification is called the "prior distribution," then the probability distribution after modification is called the "posterior distribution."

[0079] In parameter tuning, θ is sampled again from the posterior distribution, and the process returns to step (3) above to calculate the next time step t=t+1. This is repeated until a satisfactory simulator is obtained.

[0080] In this way, the parameter θ can be adjusted to correct the prediction, but there may be cases where the correction of the probability distribution π(θ), which is the actual target of the correction, does not converge well through the repetition of the above steps (3) to (5), leaving the user dissatisfied. In such cases, the problem can be identified by tracing the correction process and addressed by adjusting the hyperparameters, which are empirical tuning elements included in the assimilation procedure.

[0081] As explained above, even in cases where it is difficult to prepare sufficient training data, such as shield excavation control, using a trained model that has been trained to some extent allows for the validity of predictions to be verified through simulation, enabling the management of prediction failure risks and expanding the scope of application of excavation control using trained models. Furthermore, the application of assimilation methods makes it possible to practically simulate shields, which have been considered difficult due to their complexity and the inclusion of numerous unknown parameters.

[0082] This system can be used not only as a simulator for testing trained models, but also, for example, to search for the optimal operation when manually manipulating a shield, or as a tool for pre-evaluating the results of such operations.

[0083] While the simulation system, method, and program of the present invention have been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments described above. Other embodiments, additions, modifications, and deletions can be made within the scope of what a person skilled in the art can conceive, and any embodiment that achieves the function and effect of the present invention is included within the scope of the present invention. [Explanation of Symbols]

[0084] 10...Shield 11…Cutter head 12... Shield Jack 13... Segments 14…Skin plate 14a...front torso 14b…back torso 15... Folding jack 16... Chamber 17... Screw conveyor 18... Belt conveyor 19… Erector 20... Backfill injection device 21... Copy cutter 30… Information Processing Systems 31…CPU 32...ROM 33…RAM 34…HDD 35…External device I / F 36…Input / Output Interface 37…Display device 38…Input device 40…Prediction Department 41…Acquisition part 42…Verification Department 43...Adjustment section

Claims

1. An information processing system that reproduces the behavior of an excavating machine, A prediction means that predicts the operation result of the excavator when controlled by the control information, using control information output from a trained model used to control the excavator and adjustable parameters. An acquisition means for acquiring the operating result of the excavator when it is actually controlled by the control information, Verification means for verifying the predicted operation result based on the acquired operation result, An adjustment means for adjusting the parameters according to the verification results, Information processing systems, including those mentioned above.

2. The information processing system according to claim 1, wherein the verification means determines the validity of the prediction means based on whether the difference between the observed value, which is the acquired operation result, and the predicted value, which is the predicted operation result, is greater than or equal to a threshold.

3. The information processing system according to claim 2, wherein the prediction means calculates the predicted value using a mathematical formula in which the time-varying control information is a variable and the time-varying parameter is a coefficient.

4. The information processing system according to claim 2 or 3, wherein the adjustment means adjusts the parameters when the prediction means is determined to be invalid as a result of the verification.

5. The information processing system according to claim 4, wherein the adjustment means adjusts the parameters by modifying the probability distribution for sampling the parameters and sampling again from the modified probability distribution.

6. The information processing system according to claim 5, wherein the adjustment of the parameters by the adjustment means, the calculation of the predicted values ​​by the prediction means, and the acquisition of the operation results of the excavating machine by the acquisition means are repeated until the prediction means is determined to be valid by the verification means.

7. A learning means for learning a learning model used to control the excavation machine using training information, A determination means for determining whether the trained model has a predetermined generalization performance by applying test data to the trained model, which is a trained model, If it is determined that the predetermined generalization performance is present, a prediction execution means that performs the prediction of the control information using the trained model, An evaluation means for evaluating the result by determining whether the result of control using the control information output from the trained model is within an acceptable range. The information processing system according to claim 1, further comprising:

8. A program that causes a computer to perform a process that reproduces the behavior of an excavating machine, A step of predicting the operation result of the excavator when controlled by the control information, using control information output from a trained model used to control the excavator and adjustable parameters. The steps include: obtaining the operating result of the excavation machine when it is actually controlled using the control information; A step of verifying the predicted operation result based on the acquired operation result, The steps include adjusting the parameters according to the verification results, The steps include: predicting the operating result of the excavator again using the control information output from the trained model and the adjusted parameters; A program that executes something.

Images