Method and system for predicting permanent strain of unbonded aggregates based on machine learning and sequential backward selection

Abstract

The present invention provides a method for predicting permanent strain, comprising: specifying a machine learning model that is pre-trained to predict permanent strain for non-bonded aggregates using a dataset related to non-bonded aggregates; extracting a feature set for predicting permanent strain for non-bonded aggregates from the dataset using a pre-equipped sequential reverse selection algorithm based on the machine learning model; and predicting permanent strain for non-bonded aggregates using the feature set based on the machine learning model.

Description

Technology Field

[0001] The present invention relates to a method and system for predicting the permanent strain of unbonded aggregates based on machine learning and sequential reverse selection. Background Technology

[0002] Recently, due to the increased availability of data and computational resources, research on various topics in machine learning (ML) is being conducted across a wide range of fields. In particular, in the fields of civil engineering and construction, research primarily focuses on supervised learning that predicts target variables using carefully selected independent features.

[0003] To this end, machine learning models implemented based on decision trees, Support Vector Machines (SVM), Random Forests, Multivariate Adaptive Regression Splines (MARS), and k-Nearest Neighbors (k-NN) are being utilized.

[0004] Meanwhile, in the fields of civil engineering and construction, machine learning models are being implemented to solve various problems, such as slope stability analysis, Tunnel Boring Machine (TBM) performance, foundation design, and pavement condition detection; through this, it has been possible to resolve various issues previously faced in these fields.

[0005] However, in most research and inventions, machine learning models are primarily developed based on open source packages, and consequently, limitations in interpretability of machine learning models or problems of overfitting may occur.

[0006] Accordingly, there is a need for a framework that selects optimal features to efficiently utilize machine learning models. The problem to be solved

[0007] The present invention relates to a method and system for predicting the permanent strain of non-bonded aggregates based on machine learning and sequential reverse selection, which can effectively select the most suitable machine learning model to predict the permanent strain of non-bonded aggregates.

[0008] Furthermore, the present invention relates to a method and system for predicting the permanent strain of non-bonded aggregates based on machine learning and sequential reverse selection, which can identify the factor having the greatest influence in the process of predicting the permanent strain of non-bonded aggregates through a machine learning model and predict the permanent strain of non-bonded aggregates with high accuracy based on this. means of solving the problem

[0009] To solve the problem described above, the permanent strain prediction method according to the present invention may include, in a permanent strain prediction method using a permanent strain prediction system, the steps of: receiving a data set related to non-bonded aggregate; specifying a machine learning model that is pre-trained to predict permanent strain for said non-bonded aggregate using said data set; extracting a feature set for predicting permanent strain for said non-bonded aggregate from said data set using a pre-prepared sequential reverse selection algorithm based on said machine learning model; and predicting permanent strain for said non-bonded aggregate using said feature set based on said machine learning model.

[0010] In addition, the permanent strain prediction system according to the present invention includes an input unit that receives a data set related to non-bonded aggregate; and a control unit that predicts the permanent strain of the non-bonded aggregate based on the data set. The control unit specifies a machine learning model that is pre-trained to predict the permanent strain of the non-bonded aggregate using the data set, extracts a feature set for predicting the permanent strain of the non-bonded aggregate from the data set using a pre-prepared sequential reverse selection algorithm based on the machine learning model, and can predict the permanent strain of the non-bonded aggregate using the feature set based on the machine learning model.

[0011] Additionally, a program stored on a computer-readable recording medium according to the present invention is executed by one or more processes in an electronic device and is a program stored on a computer-readable recording medium, wherein the program may include instructions for performing the steps of: receiving a data set related to non-bonded aggregate; specifying a machine learning model that is pre-trained to predict permanent strain for the non-bonded aggregate using the data set; extracting a feature set for predicting permanent strain for the non-bonded aggregate from the data set using a pre-prepared sequential reverse selection algorithm based on the machine learning model; and predicting permanent strain for the non-bonded aggregate using the feature set based on the machine learning model. Effects of the invention

[0012] According to various embodiments of the present invention, a method and system for predicting permanent strain of non-bonded aggregate based on machine learning and sequential reverse selection can effectively select the most suitable machine learning model for predicting permanent strain of non-bonded aggregate by comparing a plurality of different machine learning models and identifying the machine learning model with the smallest error based on the comparison result.

[0013] In addition, according to various embodiments of the present invention, a method and system for predicting the permanent strain of non-bonded aggregates based on machine learning and sequential reverse selection can identify the factor having the greatest influence in the process of predicting the permanent strain of non-bonded aggregates through a machine learning model by performing sequential reverse selection on a dataset of non-bonded aggregates using a previously selected machine learning model, and based on this, can predict the permanent strain of non-bonded aggregates with high accuracy. Brief explanation of the drawing

[0014] FIGS. 1 and FIGS. 2 illustrate an embodiment of a permanent strain prediction system according to the present invention. FIG. 3 is a flowchart illustrating a permanent strain prediction method according to the present invention. FIG. 4 illustrates an example of specifying a machine learning model. FIGS. 5 and FIGS. 7 illustrate an example of extracting a set of features. FIG. 8 illustrates an example of predicting permanent strain. FIG. 9 is a block diagram illustrating the structure of a computing device that performs the permanent strain prediction method of the present invention. Specific details for implementing the invention

[0015] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Identical or similar components are assigned the same reference number regardless of the drawing symbols, and redundant descriptions thereof will be omitted. The suffixes "module" and "part" used for components in the following description are assigned or used interchangeably solely for the ease of drafting the specification and do not have distinct meanings or roles in themselves. Furthermore, in describing the embodiments disclosed in this specification, if it is determined that a detailed description of related prior art could obscure the essence of the embodiments disclosed in this specification, such detailed description will be omitted. Additionally, the attached drawings are intended only to facilitate understanding of the embodiments disclosed in this specification; the technical concept disclosed in this specification is not limited by the attached drawings, and it should be understood that they include all modifications, equivalents, and substitutions that fall within the spirit and technical scope of the present invention.

[0016] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but said components are not limited by said terms. These terms are used solely for the purpose of distinguishing one component from another.

[0017] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

[0018] A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0019] In this application, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0020] FIGS. 1 and FIGS. 2 illustrate an embodiment of a permanent strain prediction system according to the present invention.

[0021] Referring to FIG. 1, a permanent strain prediction system (100) according to the present invention specifies one of a plurality of machine learning models (21) that are pre-trained to predict permanent strain (50) from a data set (10) for non-bonded aggregate, extracts a feature set (40) for predicting permanent strain (50) for non-bonded aggregate from the data set (10), and inputs the previously extracted feature set (40) to the previously specified machine learning model (20) to predict permanent strain (50) for non-bonded aggregate.

[0022] Here, non-bonded aggregate may refer to a mixture aggregated by physical properties, and such non-bonded aggregate may include gravel, sand, aggregate, etc., as well as mixtures such as cement, asphalt, etc., before hardening.

[0023] Additionally, the permanent strain (50) may indicate the degree of permanent deformation when a predetermined load or load is applied to the non-bonded aggregate. That is, the permanent strain (50) may indicate the degree of permanent deformation even after the load or load applied to the non-bonded aggregate is removed.

[0024] Accordingly, the data set (10) may include factors that affect the permanent strain (50) among the factors related to the non-bonded aggregate. In this case, depending on the embodiment, the data set (10) may randomly include various factors related to the non-bonded aggregate.

[0025] For example, the data set (10) may include the number of cycles for composing non-bonded aggregate, deviation stress, coefficient of uniformity (Cu), coefficient of curvature (Cc), particle size (e.g., D10, D30, D50 and D60, etc.), optimal moisture content (OMC), maximum dry unit weight (MDUW), degree of compaction, etc., and, depending on the embodiment, may further include maximum shear strength, cohesion, friction angle, angularity index (AI), surface texture index (STI), flat and elongated ratio (FER), etc., derived from a forged triaxial test, etc.

[0026] Meanwhile, the data set (10) may include various factors identified as affecting permanent strain (50) in addition to the factors described above.

[0027] Additionally, the feature set (40) may be a set of major factors that affect the permanent strain (50) among the multiple factors (11) included in the data set (10). To this end, the feature set (40) may be a set of factors that have a relatively small effect on the accuracy of the permanent strain (50) or one or more factors (41) that have a relatively large effect on the accuracy of the permanent strain (50) may be extracted by removing one or more factors that have a relatively small effect on the accuracy of the permanent strain (50) based on a sequential reverse selection algorithm (30) during the process of predicting the permanent strain (50) corresponding to the data set (10) using a machine learning model (20).

[0028] At this time, the sequential reverse selection algorithm (30) can be implemented by repeating the process of generating a plurality of feature sets from which any of the plurality of factors (11) included in the data set (10) have been removed, and identifying one feature set among the plurality of feature sets that has the smallest error in the permanent strain (50) according to the machine learning model (20).

[0029] Accordingly, the permanent strain prediction system (100) can extract one or more factors (41) among the multiple factors (11) included in the data set (10) as a feature set (40) that are specified to minimize the error of the permanent strain (50) through the machine learning model (20).

[0030] Meanwhile, the machine learning model (20) may be trained to predict the permanent strain (50) corresponding to the non-bonded aggregate when a data set (10) related to the non-bonded aggregate is input.

[0031] At this time, the machine learning model (20) may be one machine learning model (20) selected from among a plurality of machine learning models (21) implemented in different ways. That is, the permanent strain prediction system (100) can select one machine learning model (20) with the smallest error among the plurality of different machine learning models (21).

[0032] At this time, the plurality of machine learning models (21) may include various machine learning models based on XGBoost (Extreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine), Gradient Boosting Regressor, AdaBoost Regressor, Extra Trees (or, Extremely Randomized Trees), Random Forest Regressor, k-Nearest Neighbors Regression, Ridge Regression and Bayesian Ridge Regression, etc.

[0033] At this time, the machine learning model (20) may be trained by a separate external server (or device), and, depending on the embodiment, the permanent strain prediction system (100) may train it.

[0034] That is, the permanent strain prediction system (100) can train a machine learning model (20) using training data and correct data, which is label data for the training data. At this time, the permanent strain prediction system (100) can train the machine learning model (20) in different ways depending on the implementation method of the machine learning model (20).

[0035] In this regard, the training data and the correct answer data may each correspond to a plurality of factors related to the non-bonded aggregate and a permanent strain (50) corresponding to the plurality of factors, and such training data and correct answer data may be included in the data set (10), and may be composed of a data set different from the data set (10) according to the embodiment.

[0036] Accordingly, the permanent strain prediction system (100) can specify a machine learning model (20) using a dataset configured for learning and extract a feature set (40) for non-bonded aggregate, and can be implemented to predict the permanent strain (50) for non-bonded aggregate based on the machine learning model (20) and feature set (40) previously specified using a dataset other than the dataset configured for learning.

[0037] In this case, the data set configured for learning may include multiple training data and multiple correct data corresponding to each of the multiple training data, and the data set configured to predict permanent strain (50) for non-bonded aggregate may include data corresponding to each of the multiple factors corresponding to the previously extracted feature set (40).

[0038] At this time, the permanent strain prediction system (100) may be implemented as a separate device (or system) comprising a module for specifying a machine learning model (20) and extracting a feature set (40) for non-bonded aggregate and a module for predicting permanent strain (50) for non-bonded aggregate, and, according to an embodiment, may be implemented as a single device (or system) in which each module is integrated.

[0039] Referring to FIG. 2, for this purpose, the permanent strain prediction system (100) may include an input unit (110), a storage unit (120), a control unit (130), and an output unit (140).

[0040] The input unit (110) may receive information necessary for the operation of the permanent strain prediction system (100) according to the present invention. To this end, the input unit (110) may be connected to a separate input device, server, or external storage device via a wireless or wired network.

[0041] Accordingly, the input unit (110) can receive a data set (10) from a separate input device, server, or external storage device, etc.

[0042] Additionally, the storage unit (120) may store commands and information necessary for the operation of the permanent strain prediction system (100) according to the present invention. For example, the storage unit (120) may store a data set (10) input through the input unit (110).

[0043] Additionally, the storage unit (120) may store a plurality of machine learning models or any one machine learning model (20) specified based on the plurality of machine learning models, and may store information about one or more factors specified as feature sets among the plurality of factors included in the data set (10).

[0044] Additionally, the storage unit (120) may store a sequential reverse selection algorithm (30) configured to extract a feature set from a data set (10), and a principal component analysis algorithm to be described below.

[0045] Additionally, the storage unit (120) can store the permanent strain (50) predicted based on the data set (10).

[0046] The control unit (130) can control the overall operation of the permanent strain prediction system (100) according to the present invention. That is, the control unit (130) can specify one of a plurality of machine learning models based on a data set (10), extract a feature set from the data set (10), and predict the permanent strain (50) for non-bonded aggregate using the previously specified machine learning model (20) and the feature set.

[0047] Specifically, the control unit (130) can receive, through the input unit (110), a data set (10) that includes a plurality of factors used to form a non-bonded aggregate and a permanent strain of the non-bonded aggregate formed by the plurality of factors.

[0048] At this time, according to the embodiment, the control unit (130) may identify outliers from the data set (10) and perform scaling on the data set (10) so that the identified outliers fall within a predetermined numerical range.

[0049] Furthermore, the control unit (130) can specify a pre-trained machine learning model (20) to predict permanent strain (50) for non-bonded aggregate using the data set (10).

[0050] To this end, the control unit (130) can train multiple different machine learning models using a data set (10) and, based on the training results, can identify one machine learning model (20) among the multiple machine learning models that has the highest accuracy.

[0051] At this time, the control unit (130) can train each of the plurality of machine learning models and specify one of the machine learning models (20) according to the training result, and according to the embodiment, one of the machine learning models (20) can be specified based on the error in the permanent strain (50) output by inputting a data set (10) to each of the plurality of machine learning models that were previously trained.

[0052] Furthermore, the control unit (130) can extract a feature set for predicting permanent strain (50) for non-bonded aggregate from a data set (10) by using a sequential reverse selection algorithm (30) that is pre-equipped based on a machine learning model (20).

[0053] At this time, the control unit (130) can calculate the degree to which each of the multiple factors related to non-bonded aggregate appearing from the data set (10) influence the prediction of permanent strain (50) based on the previously specified machine learning model (20), and extract a feature set based on the calculated degree.

[0054] For example, the control unit (130) can generate multiple feature sets having one or more factors by using multiple factors related to non-bonded aggregates appearing from the data set (10), and input each of the multiple feature sets into a machine learning model (20) to calculate an error (e.g., coefficient of determination, root mean square error, etc.) for the permanent strain (50) corresponding to each of the multiple feature sets.

[0055] Accordingly, the control unit (130) can extract one set of features with the smallest error for the previously calculated permanent strain (50) as a feature set.

[0056] At this time, the control unit (130) can repeat the process of selecting one feature set based on the error while reducing the number of one or more factors belonging to the feature set until the number of one or more factors belonging to the feature set corresponds to a predetermined number, or until a predetermined condition regarding the error of the permanent strain (50) according to the feature set is satisfied.

[0057] Here, a predetermined condition for the error may be a condition in which the rate of change of the error corresponding to the previously selected feature set becomes lower than a predetermined threshold value, or a condition in which the process of selecting the feature set is determined based on whether the error increases or decreases.

[0058] Furthermore, the control unit (130) may predict the permanent strain (50) for the non-bonded aggregate using a feature set based on a machine learning model (20). To this end, the control unit (130) may predict the permanent strain (50) for the non-bonded aggregate by inputting a feature set extracted from the data set (10) based on the sequential reverse selection algorithm (30) to one of the machine learning models (20) selected from among the previously different machine learning models.

[0059] The output unit (140) may output information generated by the operation of the permanent strain prediction system (100) according to the present invention. To this end, the output unit (140) may be connected to a separate visual output device, server, or external storage device via a wireless or wired network.

[0060] Accordingly, the output unit (140) can output a data set (10), permanent strain (50), etc., so that a user can visually check it through a separate output device, server, or external storage device, and, depending on the embodiment, can also transmit the data set (10), permanent strain (50), etc., to another device.

[0061] Additionally, the output unit (140) may output information generated during the process of specifying one of the plurality of machine learning models, extracting a feature set, or output information generated during the process of predicting the permanent strain (50) for non-bonded aggregate.

[0062] Based on the configuration of the permanent strain prediction system (100) described above, the permanent strain prediction method will be explained in more detail below.

[0063] FIG. 3 is a flowchart illustrating a method for predicting permanent strain according to the present invention. FIG. 4 illustrates an embodiment specifying a machine learning model. FIG. 5 and FIG. 7 illustrate an embodiment extracting a feature set. FIG. 8 illustrates an embodiment predicting permanent strain.

[0064] Referring to FIG. 3, the permanent strain prediction system (100) according to the present invention can receive a data set related to non-bonded aggregate (S100).

[0065] Specifically, the permanent strain prediction system (100) can receive a data set including multiple factors used to form the non-bonded aggregate and the permanent strain for the non-bonded aggregate formed by the multiple factors.

[0066] For example, a permanent strain prediction system (100) can receive a data set including the number of cycles for forming non-bonded aggregate, deviation stress, coefficient of uniformity (Cu), coefficient of curvature (Cc), particle size (e.g., D10, D30, D50 and D60, etc.), optimal moisture content (OMC), maximum dry unit weight (MDUW), and degree of compaction.

[0067] Additionally, the permanent strain prediction system (100) may receive a data set that further includes maximum shear strength, cohesion, friction angle, angularity index (AI), surface texture index (STI), flat and elongated ratio (FER), etc., derived from a forging triaxial test.

[0068] Furthermore, the permanent strain prediction system (100) can identify outliers from a data set and perform scaling on the data set so that the identified outliers fall within a predetermined numerical range.

[0069] For example, a permanent strain prediction system (100) can divide the range of data appearing from a data set into quartiles and determine an interquartile range based on the first quartile and the third quartile.

[0070] Here, the first quartile may include data belonging to the bottom 25 percent of the data set according to the quartile, and the third quartile may include data belonging to the top 25 percent of the data set according to the quartile.

[0071] Accordingly, the permanent strain prediction system (100) can identify data that falls outside a predetermined range based on the interquartile range as outliers, and can scale outliers based on the interquartile range.

[0072] At this time, the permanent strain prediction system (100) may also scale all data included in the data set based on the interquartile range.

[0073] In one embodiment, the permanent strain prediction system (100) can perform scaling according to the interquartile range according to the following mathematical formula 1.

[0074]

[0075] Here, is the scaled data, x is the data included in the dataset before scaling is applied, median(x) represents the median of the data included in the dataset, Q1(x) represents the first quartile, and Q3(x) represents the third quartile.

[0076] The permanent strain prediction system (100) according to the present invention may specify a machine learning model that has been pre-trained to predict permanent strain for non-bonded aggregates using a data set (S200).

[0077] Specifically, the permanent strain prediction system (100) can train multiple different machine learning models using a data set and, based on the training results, can identify one machine learning model among the multiple machine learning models that has the highest accuracy (or the lowest error).

[0078] For example, a permanent strain prediction system (100) can train various machine learning models based on XGBoost, LightGBM, Gradient Boosting Regressor, AdaBoost Regressor, Extra Trees (or Extremely Randomized Trees), Random Forest Regressor, k-Nearest Neighbors Regression, Ridge Regression and Bayesian Ridge Regression, etc. using a dataset.

[0079] Here, XGBoost, LightGBM, Gradient Boosting Regressor, and AdaBoost Regressor are boosting-based machine learning models that may be implemented to improve model performance by iteratively improving the accuracy of weak learners.

[0080] In addition, Extra Trees and Random Forest Regressor are types of ensemble models that may be implemented to improve accuracy by combining outputs predicted from multiple models.

[0081] In addition, k-Nearest Neighbors Regression may be implemented to predict the data to be predicted based on the average of the k nearest neighbor data.

[0082] In addition, Ridge Regression and Bayesian Ridge Regression are models based on regularization regression, and may be implemented to minimize model complexity and prevent overfitting.

[0083] In this way, the permanent strain prediction system (100) can train multiple machine learning models using a data set and identify the machine learning model with the highest accuracy based on the training results.

[0084] To this end, the permanent strain prediction system (100) can divide the dataset into training data and test data, train a machine learning model using the training data, and verify the performance of the machine learning model using the test data.

[0085] In one embodiment, the permanent strain prediction system (100) may divide a data set into a predetermined number of K (e.g., 10) folds, designate one of the K folds as test data, and designate the remaining K-1 folds as training data.

[0086] At this time, the permanent strain prediction system (100) can repeat the training of the machine learning model K times so that each of the previously divided K folds trains the machine learning model as test data.

[0087] For example, a permanent strain prediction system (100) can train a machine learning model by specifying a first fold among K folds as test data and the remaining folds excluding the first fold as training data in the first training, and train a machine learning model by specifying a second fold among K folds as test data and the remaining folds excluding the second fold as training data in the second training.

[0088] In this way, the permanent strain prediction system (100) can repeat K learning cycles, and according to the embodiment, such K learning cycles can be repeated for a predetermined number of epochs.

[0089] Accordingly, as shown in FIG. 4, the permanent strain prediction system (100) trains each of a plurality of machine learning models (21) using training data according to the data set (10), and among the plurality of machine learning models (21), can identify one machine learning model that has the smallest error (23) (e.g., coefficient of determination, root mean square error, etc.) regarding the permanent strain (51) predicted by each machine learning model (21) based on test data.

[0090] In one embodiment, the permanent strain prediction system (100) may have one machine learning model among a plurality of machine learning models (21) that is based on XGBoost.

[0091] Referring again to FIG. 3, the permanent strain prediction system (100) according to the present invention can extract a feature set for predicting permanent strain for non-bonded aggregate from a data set using a sequential reverse selection algorithm that is pre-equipped based on a machine learning model (S300).

[0092] Specifically, the permanent strain prediction system (100) can determine the degree to which each of the multiple factors related to non-bonded aggregate appearing from the data set influences the permanent strain prediction based on the machine learning model specified earlier, and can extract a feature set based on the determined degree.

[0093] For example, a permanent strain prediction system (100) can generate multiple feature sets having one or more factors by using multiple factors related to non-bonded aggregates appearing from a data set, and input each of the multiple feature sets into a machine learning model to calculate an error (e.g., coefficient of determination, root mean square error, etc.) for the permanent strain corresponding to each of the multiple feature sets.

[0094] Accordingly, the permanent strain prediction system (100) can extract a feature set with the smallest error in the previously calculated permanent strain as a feature set. That is, the error in the permanent strain calculated above may be calculated as the degree to which it affects the permanent strain prediction.

[0095] In this case, in one embodiment, the permanent strain prediction system (100) can generate multiple feature sets by subtracting the number of factors appearing from the data set one by one when the number of one or more factors included in the feature set is predetermined, and can repeat the process of selecting one of the previously generated multiple feature sets until the number of factors included in the feature set satisfies the previously predetermined number.

[0096] That is, the permanent strain prediction system (100) can, in the case where N factors related to non-bonded aggregate are included in the feature set and the feature set is determined to include m factors, generate N feature sets with N-1 factors in the first selection and select one feature set with the smallest error among the N feature sets, and in the second selection, generate N-1 feature sets with N-2 factors based on the N-1 factors included in the previously selected feature set and select one feature set with the smallest error among the generated N-1 feature sets.

[0097] Accordingly, the permanent strain prediction system (100) repeats the above selection process until the number of factors included in the feature set satisfies a predetermined m, and when any one feature set having m factors is selected, one or more factors included in the feature set can be extracted as a feature set.

[0098] In another embodiment, the permanent strain prediction system (100) compares an error (e.g., coefficient of determination, root mean square error, etc.) for any one feature set selected in the previous selection step with an error for any one feature set selected in the current selection step, and if the error of any one feature set selected in the previous selection step is smaller according to the comparison result, one or more factors included in any one feature set selected in the previous selection step can be extracted as a feature set.

[0099] At this time, the permanent strain prediction system (100) can repeat the process of regenerating and selecting a feature set based on one or more factors included in the feature set selected in the current selection stage if the error of any one feature set selected in the current selection stage is smaller according to the comparison result.

[0100] That is, as shown in FIG. 5, the permanent strain prediction system (100) can, in the first selection, generate N feature sets (13, 14) having N-1 factors based on N factors related to non-bonded aggregates included in the data set (11), and select one first feature set among the N feature sets (13, 14) that has the smallest error according to the machine learning model (20), and as shown in FIG. 6, in the second selection, generate N-1 feature sets (17, 18) having N-2 factors based on N-1 factors included in the previously selected first feature set (15), and select one second feature set among the generated N-1 feature sets (17, 18) that has the smallest error according to the machine learning model (20).

[0101] At this time, the permanent strain prediction system (100) compares the error (25) for the first feature set (15) and the error (26) for the second feature set (19) as shown in FIG. 7, and if the error (25) for the first feature set (15) is smaller according to the comparison result, N-1 factors included in the first feature set (15) can be extracted as a feature set.

[0102] Additionally, the permanent strain prediction system (100) can continue to perform a third selection step if the error (26) for the second feature set (19) is smaller according to the comparison result.

[0103] At this time, the permanent strain prediction system (100) generates N-2 feature sets having N-3 factors based on N-2 factors included in the previously selected second feature set (19) in the third selection, selects one of the third feature sets with the smallest error among the generated N-2 feature sets, compares the error (26) for the second feature set (19) with the error for the third feature set, and repeats the selection step according to the comparison result, or extracts the N-2 factors included in the second feature set (19) as a feature set.

[0104] Furthermore, the permanent strain prediction system (100) may also extract a set of principal component features based on a principal component analysis algorithm for a data set related to non-bonded aggregates.

[0105] For example, the permanent strain prediction system (100) can calculate a covariance matrix representing the correlation between multiple factors included in a data set. Accordingly, the permanent strain prediction system (100) calculates eigenvalues ​​and eigenvectors for the covariance matrix and can extract a predetermined number of factors as a principal component feature set in order of increasing magnitude of the eigenvalues. Here, the predetermined number can be determined based on the magnitude of the eigenvalues.

[0106] Furthermore, the permanent strain prediction system (100) may compare a feature set extracted from a data set based on a sequential reverse selection algorithm with a principal component feature set extracted from a data set based on a principal component analysis algorithm, and determine a new feature set based on the comparison result.

[0107] For example, a permanent strain prediction system (100) can compare the error in permanent strain according to a feature set extracted based on a sequential reverse selection algorithm with the error in permanent strain according to a principal component feature set extracted based on a principal component analysis algorithm, and based on the comparison result, identify the feature set (or principal component feature set) with a smaller error.

[0108] As another example, the permanent strain prediction system (100) may compare multiple factors included in a feature set extracted based on a sequential reverse selection algorithm with multiple factors included in a principal component feature set extracted based on a principal component analysis algorithm, and specify a new feature set based on the comparison result.

[0109] At this time, according to an embodiment, the permanent strain prediction system (100) may specify a new feature set that includes all of the multiple factors extracted according to each algorithm by merging multiple factors according to a sequential reverse selection algorithm and multiple factors according to a principal component analysis algorithm, or may generate a new feature set that includes one or more factors that overlap among the multiple factors according to the sequential reverse selection algorithm and the multiple factors according to the principal component analysis algorithm.

[0110] Referring again to FIG. 3, the permanent strain prediction system (100) according to the present invention can predict the permanent strain of non-bonded aggregate using a feature set based on a machine learning model (S400).

[0111] Referring to FIG. 8, specifically, the permanent strain prediction system (100) can predict the permanent strain (50) for non-bonded aggregate by inputting a feature set (40) extracted from a data set (10) based on a sequential reverse selection algorithm into one of the previously selected machine learning models (20) among the previously different machine learning models (21).

[0112] In one embodiment, the permanent strain prediction system (100) can predict permanent strain by inputting a set of features including the number of cycles, deviation stress, optimal moisture content, maximum dry unit weight, internal friction angle and FER, etc., into a machine learning model based on XGBoost.

[0113] Through the above configurations, the permanent strain prediction system (100) according to the present invention can effectively select the most suitable machine learning model to predict the permanent strain of non-bonded aggregate by comparing a plurality of different machine learning models and specifying the machine learning model with the smallest error based on the comparison result.

[0114] In addition, the permanent strain prediction system (100) according to the present invention can identify the factor that has the greatest influence in the process of predicting the permanent strain of the non-bonded aggregate through the machine learning model by performing sequential reverse selection on the data set of non-bonded aggregate using the previously selected machine learning model, and can predict the permanent strain of the non-bonded aggregate with high accuracy based on this.

[0115] Furthermore, the permanent strain prediction system (100) according to the present invention is composed of a computing device and can perform at least one function related to the aforementioned permanent strain prediction method.

[0116] FIG. 9 is a block diagram illustrating the structure of a computing device that performs the permanent strain prediction method of the present invention.

[0117] The computing device (1000) may include a user interface module (1001), a network communication module (1002), one or more processors (1003), data storage (1004), one or more camera(s) (1018), one or more sensors (1020) and a power system (1022), all of which may be connected to each other via a system bus, a network or other connection mechanism (1005).

[0118] The user interface module (1001) may be operable to transmit data to an external user input / output device and / or receive data from an external user input / output device.

[0119] For example, in the present invention, the permanent strain prediction system (100) receiving a data set may be performed by external input using a user interface module. At this time, the user interface module (1001) may include a touch screen, a computer mouse, a keyboard, a keypad, a touchpad, a trackball, a joystick, a voice recognition module, or other similar devices.

[0120] Additionally, the user interface module (1001) may also be configured to provide output to one or more user display devices, such as a cathode ray tube (CRT), liquid crystal display, light-emitting diode (LED), display using digital light processing (DLP) technology, or printer.

[0121] The user interface module (1001) may also be configured to produce an audible output using a device such as a speaker, a speaker jack, an audio output port, an audio output device, an earphone and / or other similar device.

[0122] The user interface module (1001) may additionally be configured with one or more tactile devices capable of generating tactile outputs, such as vibrations and / or other outputs, detectable by touch and / or physical contact with the computing device (1000).

[0123] The network communication module (1002) may include one or more devices that provide one or more wireless interface(s) (1007) and / or one or more wired interface(s) (1608) that are configurable to communicate through a network.

[0124] Additionally, the network communication module (1002) may be configured to provide reliable security and / or authenticated communication.

[0125] One or more processors (1003) may include one or more general-purpose processors and / or one or more special-purpose processors (e.g., digital signal processor, tensor processing unit (TPU), graphics processing unit (GPU), neural network processing unit (NPU), application integrated circuit, application semiconductor (ASIC), etc.). One or more processors (1003) may be configured to execute computer-readable instructions (1006) contained in data storage (1004) and / or other instructions described herein.

[0126] As an example of this, the learning and inference described in this specification are performed in a neural network processing unit (NPU), and efficiency can be increased by processing data operations at high speed with low power.

[0127] The data storage (1004) may include one or more non-transient computer-readable storage media that can be read and / or accessed by at least one of one or more processors (1003).

[0128] One or more computer-readable storage media may include volatile and / or non-volatile storage components, such as optical, magnetic, organic, or other memory or disk storage devices. In some examples, data storage (1004) may be implemented using a single physical device (e.g., one optical, magnetic, organic, or other memory or disk storage device), whereas in other examples, data storage (1004) may be implemented using two or more physical devices.

[0129] The data storage (1004) may include computer-readable instructions (1006) and additional data. The data storage (1004) may include a storage necessary to perform at least some of the methods, scenarios, and techniques described herein and / or at least some of the functions of the device and network.

[0130] In some examples, the data storage (1004) may include a storage for the learned neural network model (1010) (e.g., a machine learning model) described in the present invention.

[0131] Meanwhile, the computing device (1000) may include one or more camera(s) (1018), one or more sensors (1020) and / or a power system (1022).

[0132] The camera(s) (1018) can capture light and / or electromagnetic radiation emitted as visible light, infrared radiation, ultraviolet radiation and / or light of one or more other frequencies. The sensor (1020) may be configured to measure conditions within the computing device (1000) and / or conditions of the computing device (1000) environment and to provide data regarding these conditions. The power system (1022) may include one or more batteries (1024) and / or one or more external power interfaces (1026) for providing power to the computing device (1000).

[0133] Meanwhile, although the permanent strain prediction system (100) of the present invention has been described above as being implemented as a computing device, the present invention is not limited thereto. For example, the functions of the neural network and / or computing device may be distributed among a plurality of computing clusters.

[0134] Furthermore, the present invention described above can be implemented as a program that is executed by one or more processes in an electronic device and stored on a computer-readable recording medium.

[0135] Accordingly, the present invention can be implemented as computer-readable code or instructions on a medium on which a program is recorded. That is, various control methods according to the present invention can be provided in the form of integrated or individual programs.

[0136] Meanwhile, computer-readable media include all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable media include HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0137] Furthermore, the computer-readable medium may be a server or cloud storage that includes a storage and is accessible to an electronic device via communication. In this case, the computer may download the program according to the present invention from the server or cloud storage via wired or wireless communication.

[0138] Furthermore, in the present invention, the computer described above is an electronic device equipped with a processor, namely a CPU (Central Processing Unit), and no special limitations are placed on its type.

[0139] Meanwhile, the above detailed description should not be interpreted restrictively in all respects but should be considered exemplary. The scope of the invention should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention.

Claims

Claim 1 A method for predicting permanent strain using a permanent strain prediction system, comprising: receiving a dataset related to non-bonded aggregate; training a plurality of different machine learning models to predict permanent strain for the non-bonded aggregate using the dataset, calculating a permanent strain prediction error using the dataset for each of the plurality of machine learning models, and identifying one machine learning model with the lowest error based on the permanent strain prediction error; generating a plurality of feature sets by removing at least one factor among a plurality of factors included in the dataset using a pre-prepared sequential reverse selection algorithm based on the machine learning model, and identifying at least one feature set as a feature set based on the result calculated by inputting each of the plurality of feature sets into the machine learning model, thereby extracting the feature set for predicting the permanent strain for the non-bonded aggregate from the dataset; and predicting the permanent strain for the non-bonded aggregate using the feature set based on the machine learning model. Claim 2 A permanent strain prediction method according to claim 1, wherein the data set includes the number of cycles for composing non-bonded aggregate, deviation stress, coefficient of uniformity (Cu), coefficient of curvature (Cc), particle size, optimal moisture content (OMC), maximum dry unit weight (MDUW), degree of compaction, maximum shear strength, cohesion, friction angle, angularity index (AI), surface texture index (STI), and flat and elongated ratio (FER). Claim 3 A permanent strain prediction method according to claim 1, wherein the step of receiving the data set comprises: a step of identifying outliers from the data set; and a step of performing scaling on the data set so that the identified outliers fall within a predetermined numerical range. Claim 4 A permanent strain prediction method according to claim 3, wherein the step of identifying the outlier comprises: a step of dividing the range of data appearing from the data set into quartiles; a step of specifying the quartile range based on the first quartile and the third quartile among the divided quartiles; and a step of specifying the outlier based on the quartile range. Claim 5 delete Claim 6 In claim 1, the specified machine learning model is a permanent strain prediction method, which is a machine learning model based on XGBoost. Claim 7 A method for predicting permanent strain according to claim 1, wherein the step of extracting the feature set comprises: a step of determining the degree to which each of the plurality of factors related to non-bonded aggregate appearing from the data set based on the specified machine learning model based on the sequential reverse selection algorithm influences the prediction of permanent strain; and a step of extracting the feature set based on the determined degree. Claim 8 A method for predicting permanent strain according to claim 7, wherein the step of determining the degree of influence on the prediction of permanent strain comprises: a step of generating a plurality of feature sets having one or more factors using the plurality of factors; and a step of inputting each of the plurality of feature sets into a machine learning model to calculate an error for the permanent strain corresponding to each of the plurality of feature sets as the degree of influence on the prediction of permanent strain. Claim 9 A permanent strain prediction system comprising: an input unit receiving a data set related to non-bonded aggregate; and a control unit predicting a permanent strain for the non-bonded aggregate based on the data set, wherein the control unit trains a plurality of different machine learning models to predict the permanent strain for the non-bonded aggregate using the data set, calculates a permanent strain prediction error using the data set for each of the plurality of machine learning models, identifies one machine learning model with the lowest error based on the permanent strain prediction error, generates a plurality of feature sets by removing at least one factor among a plurality of factors included in the data set using a pre-prepared sequential reverse selection algorithm based on the machine learning model, identifies at least one feature set as a feature set based on the result calculated by inputting each of the plurality of feature sets into the machine learning model, extracts the feature set for predicting the permanent strain for the non-bonded aggregate from the data set, and predicts the permanent strain for the non-bonded aggregate using the feature set based on the machine learning model. Claim 10 A program stored on a computer-readable recording medium that is executed by one or more processes in an electronic device, the program comprising: a step of receiving a data set related to non-bonded aggregate; a step of training a plurality of different machine learning models to predict permanent strain for the non-bonded aggregate using the data set, calculating a permanent strain prediction error using the data set for each of the plurality of machine learning models, and identifying one machine learning model with the lowest error based on the permanent strain prediction error; a step of generating a plurality of feature sets by removing at least one factor among a plurality of factors included in the data set using a pre-prepared sequential reverse selection algorithm based on the machine learning model, and identifying at least one feature set as a feature set based on the result calculated by inputting each of the plurality of feature sets into the machine learning model, thereby extracting the feature set for predicting the permanent strain for the non-bonded aggregate from the data set; and a step of performing the step of predicting the permanent strain for the non-bonded aggregate using the feature set based on the machine learning model.

Images