KR102428840B1 - Computing system implementing and operating models describing transition between multiple resolution-based models - Google Patents

Abstract

A computing system according to an embodiment of the present invention is a computing system that implements and operates a model that represents a spatial transition based on a cell and describes a target system to which a multi-resolution modeling technique is applied, wherein the computing system is the target system. a communication interface capable of receiving or calling acquisition data obtained from; Obtaining a second resolution-based cell model based on a second resolution of the target cell, which is higher than the first resolution, based on a first resolution of the target cell and a first resolution-based cell model obtained based on the acquired data for the transition model; and at least one or more processors. The transition model is a relation between a first resolution-based data set in which the target cell is described based on the first resolution and a second resolution-based data set in which the target cell is described based on the second resolution among the acquired data. It may be a model trained based on .

Description

{COMPUTING SYSTEM IMPLEMENTING AND OPERATING MODELS DESCRIBING TRANSITION BETWEEN MULTIPLE RESOLUTION-BASED MODELS}

The present invention relates to a technique for modeling and simulating a physical complex system, and more particularly, to a technique for implementing and operating a digital model describing a target system using a computing system. Also, it relates to a technology for upgrading a digital model by using both a conventional theory-based model and a data-based model based on data analysis in the process of implementing and operating a digital model. The present invention is derived from research conducted as part of the regional enterprise innovative growth support (R&D) project of the Ministry of SMEs and Startups and the Jeju Regional Project Evaluation Group. [Task management number: P0010175, task name: BAS-based digital twin platform core SW and application model development].

Demands to improve productivity, economy, safety, etc. in industrial fields are spreading. Recently, technologies such as the Internet of Things (IoT), big data, artificial intelligence, and cyber physical systems (CPS) have been widely used. are receiving

A digital twin is a digital replica (twin) of physical objects (assets, processes and systems, etc.) can be considered as a model of

The digital twin implemented in the computing system is used for various purposes in the industrial field, such as predicting situations that may occur in reality or informing the conditions for optimizing operations while reflecting the real situation by interworking with the target object (physical asset). It is recognized as a means of enhancing competitiveness.

Internet of Things (IoT) technology and digital twin technology are closely related. The advancement of IoT platform technology enables smart services such as machine learning/AI-based prediction, fault diagnosis, optimization, and predictive diagnosis after collecting sensor data of the operating system in real time. Digital twin technology was started on the premise that it is used in industrial fields, but recently, efforts are being made to expand its application areas, such as cyber cities, to more diverse smart services. It is virtually impossible to mount all sensors for collecting all data required for various smart services in a system in operation due to physical constraints (location and number of sensors) and/or economic constraints (number of sensors).

As a means to partially solve this problem, in US Patent Publication No. 2017/0286572 "Digital twin of twinned physical system", the operation of the real sensor is simulated using a digital twin of the real sensor, and the real sensor does not work. When not, a technique to replace the operation of the real sensor by collecting data from the virtual sensor corresponding to the real sensor in the digital twin model was proposed.

In Korea Patent Publication No. 10-2019-0013610 "System, method, and control unit for controlling the operation of a technical system," virtual sensor data is generated using a digital twin of a real sensor, and when an abnormality of the real sensor is detected A technique for replacing anomaly sensors with virtual sensor data has been proposed.

In general, in order to perform motion/performance analysis or prediction on a system in the real world, an abstract model for the system is created and executed, and measures of interest such as motion/performance are measured/observed. In order to secure the reliability of the analysis/prediction results of the target system through these models, how well (accurately) the system is modeled is an important key.

As a modeling method, there is a method of creating an abstract model by using knowledge such as physical laws or behavior rules included in the target system, which is modeling and simulation that can express the causal relationship between the controlled input and the corresponding output as a model. (M&S) based method. This method has a limitation that detailed information about the system to be modeled must be available. In addition, when a model is built in a modeling and simulation (M&S)-based method, it is essential to verify the model with real data about how well it reflects the real system to ensure the validity of the model. When it is difficult to obtain data from the real world system, the validity of the model cannot be verified, so there may be a problem in that the reliability of the analysis/prediction results based on the model cannot be guaranteed.

As another modeling method for system prediction and analysis, there is a data modeling method of deriving rules/patterns/functions contained in the system by analyzing a lot of data acquired through operation and observation of the target system. A machine learning-based model, which can be said to be a representative method of data modeling, is a method that shows the correlation between one set of data and another data set. . However, the data model built through machine learning can be predicted on the premise that the system will operate without any changes in the future, but if the configuration or operating laws of the system change, it is impossible to predict the future with the previously learned model. There are limitations.

The term 'big data' is widely used as a means of predicting a diversified society. Big data refers to datasets of sizes beyond the ability of typical software tools to collect, manage, and process within an acceptable elapsed time. Since such large-capacity data provides more insight than existing limited data, it is receiving a lot of attention in research in various fields such as science, engineering, national defense, management, medicine, and politics. For this reason, modeling using big data is becoming an essential and important issue in the era of big data.

A model generated by a physical law or an action rule is called a physics-driven model, and a model that analyzes data and derives a rule hidden in the data is called a data-driven model. Although the physical-based model has the advantage of being not limited to the actual data and extending the scope of application, it has the disadvantage that verification by the actual data is required. There is a limit in that it cannot be applied to a range outside of the measured data. Recently, due to the development of artificial neural networks and deep learning techniques, it has been reported that a data-based model has better performance than a physical-based model. still exists In addition, the data-based model can only analyze the correlation between data and has a limitation in that it cannot explain the causal relationship between which data is a cause and which data is an effect.

There are attempts to complement the advantages and disadvantages of these data-based models and physically-based models. For example, US Patent Publication US 2018/0357343 "OPTIMIZATION METHODS FOR PHSICAL MODELS" proposes a hybrid model that combines the advantages and disadvantages of a physics-driven model with high fidelity and a data-driven model with low fidelity. In US 2018/0357343, the physics-driven model is executed when the calculated competence area of the data-driven model is out of the critical area, and the physics-driven model is developed by expressing the discrepancy between the data-driven model and the physics-driven model in the form of a function. After calibration, a hybrid model combining this relationship is used. At this time, the competence of the data-driven model is calculated based on the accuracy of the sample data of the data-driven model.

In the case of a single system in which input data and output data of a target system are clearly defined, a hybrid model combining the advantages of a data-driven model and a physics-driven model can be used by application of US 2018/0357343. However, when the target system becomes complex and there are many variables to be dealt with, more advanced data application is required to digitize the target system and implement it as a digital twin model.

Meanwhile, in the process of digitizing and modeling a target system, a multiple resolution model technique is used as a means to increase the accuracy of the target system and the efficiency of modeling. As an example of using a multi-resolution model for traffic flow simulation, Korean Patent Registration Patent KR 10-0969481 "Traffic flow simulation device and traffic flow simulation system including the same", KR 10-1815511 "Framework for traffic simulation and simulation using the same" method", and as an example of applying a multi-resolution model to fluid particle simulation, Korean Patent Registration KR 10-0872434 "Multi-resolution fluid particle simulation system and method" is mentioned. Multi-resolution models are generally simulated by applying different resolutions for each region, and are used to achieve high-resolution simulations using limited computational resources. In traffic flow simulation, a microscopic/microscopic model, a mesoscopic/mesoscopic model, and a macro/macroscopic model are often called, and a hybrid resolution model by combining them is also used.

The recent development of technologies such as the Internet of Things (IoT) provides an infrastructure that makes it easy to collect various data. There is a growing demand to improve the performance of modeling and simulation (M&S) by effectively utilizing big data.

US Patent Publication No. 2017/0286572 "Digital twin of twinned physical system" (October 5, 2017) Korean Patent Application Laid-Open No. 10-2019-0013610 "System, method and control unit for controlling the operation of a technical system" (February 11, 2019) US Patent Publication No. 2019/0033850 "Controlling operation of a technical system" (January 31, 2019) US Patent Publication No. 2019/0068618 "Using virtual sensors to accommodate industrial asset control systems during cyber attacks" (February 28, 2019) US Patent Publication US 2018/0357343 "OPTIMIZATION METHODS FOR PHSICAL MODELS" (December 13, 2018) Korean Patent Registration KR 10-0969481 "Traffic flow simulation device and traffic flow simulation system including the same" (July 2, 2010) KR 10-1815511 "Framework for traffic simulation and simulation method using the same" (December 29, 2017) Korean Patent Registration KR 10-0872434 "Multi-resolution fluid particle simulation system and method" (December 1, 2008) US Patent Publication US 2009/0112526 "SYSTEM AND METHOD FOR SIMULATING FLUID PARTICLE HAVING MULTI-RESOLUTION" (April 30, 2009)

In the prior art, the multi-resolution modeling technique is treated only as a concept of applying different resolutions for each spatial region. However, adjacent spatial regions are not completely separate objects, but have a close relationship with each other and influence each other. Accordingly, when different resolutions are applied to adjacent spatial regions, a problem arises as to how to convert different resolutions between adjacent spatial regions. In particular, with the recent development of the Internet of Things (IoT) and sensor technology, precise data is collected for each spatial region, and it is difficult to accurately model and simulate a target system without considering the correlation between these data.

The present invention is an invention derived to solve the problems of the prior art, and describes a resolution conversion between adjacent spatial regions using a cell automata technique, and in this process, each cell is each other due to the influence of the adjacent cells. An object of the present invention is to provide a method for solving a transition between different resolution models by applying a data-based model or by applying a hybrid model of a data-based model and a theory-based model.

Meanwhile, in the prior art and the above prior art, an operation or state of the real world is simulated by simplifying it into a single model. At this time, the use of a hybrid model that combines a physics-based model capable of deductive reasoning and a data-based model based on data learning has been disclosed by US 2018/0357343, which is a prior document, but the target system in the real world is very complex and operates with only a single model or It is not easy to determine the condition.

In addition, as the target system to be handled becomes larger and more complex, the limitation of attempts to describe the target system by a single model is revealed.

The present invention is an invention to solve the problems of the prior art and the prior art, by modeling a causal relationship between a plurality of causes and effects describing a target system, and combining sub-models that can partially describe the target system. We propose a system model that can comprehensively describe the entire target system. In this case, a coupling relationship and/or a data connection relationship between sub-models in the system model is defined based on structural information, and the structural information may be acquired using knowledge of the target system.

The present invention aims to propose a hybrid model that operates by combining the advantages of a physics-based model and a data-based model, and it is possible to apply the hybrid model to the system model described by the composite model by expanding the scope of application of the hybrid model. The purpose of this study is to propose a system modeling technique.

The present invention provides a mutually cooperative alternative that can overcome the limitations of each approach by complementarily utilizing the strengths of the two modeling methods of a physics-based model and a data-based model to enable robust analysis/prediction support. aim to do In this case, the purpose of this is to propose a technique that can implement a complex model by combining a physically-based model and a data-based model more strongly, unlike the prior art or prior art in which a physically-based model or a data-based model is selectively applied to the issue. do it with

It is an object of the present invention to propose a system model that makes use of the accuracy of the data-based model to the maximum, but can make inferences even on data outside the data range, which is a limit of the data-based model.

The present invention proposes a system model that can describe the operation or state of a target system due to complex causes using a data-based model, and can explain the correlation between data within the data-based model based on structural information. It aims to generate descriptive information or descriptive information.

An object of the present invention is to propose a system model that can apply Normative Analysis, which is difficult to implement in a data-based model while using a data-based model.

An object of the present invention is to implement a hybrid model that combines the advantages of a data-based model and a physical-based model as a complex system model, and to apply the complex system model to a digital twin to describe a very complex target system. do it with

The present invention is a configuration derived to achieve the above object, and the computing system according to an embodiment of the present invention represents at least one or more of a spatial shift and a temporal shift based on a cell, and among two or more resolutions for each cell. A model describing a target system to which a multi-resolution modeling technique described using at least one is applied may be implemented and operated.

The computing system may include: a communication interface capable of receiving or calling acquisition data obtained from the target system; Obtaining a second resolution-based cell model based on a second resolution of the target cell, which is higher than the first resolution, based on a first resolution of the target cell and a first resolution-based cell model obtained based on the acquired data for the transition model; and at least one or more processors.

The transition model is a relation between a first resolution-based data set in which the target cell is described based on the first resolution and a second resolution-based data set in which the target cell is described based on the second resolution among the acquired data. It may be a model trained based on .

The transition model may be a data-based model in which a correlation between the first resolution-based data set and the second resolution-based data set is learned.

The transition model includes a plurality of sub-models, and each of the plurality of sub-models is a model capable of receiving a portion of the acquired data and inferring or predicting at least a portion of the remaining data as an output, and the plurality of sub-models within the transition model Input data and output data of each of the sub-models may be defined based on structural information of the target system that may be acquired using knowledge of the target system.

The structural information includes information on a data connection relationship between a first sub-model and a second sub-model among the plurality of sub-models, and the data connection relationship indicates that output data of the first sub-model is the second sub-model. may include information on whether it can be used as input data of

At least one of the spatial shift and the temporal shift may be represented on a cell-by-cell basis using a cell automata technique.

The at least one processor may select first data generated corresponding to the first resolution from among new input data for the target cell and provide it as an input to the transition model, and the transition model may be applied to the first data. Based on this, the transition model may be controlled to infer the second resolution-based cell model from the first resolution-based cell model.

The at least one processor may provide the first resolution-based data set as an input to the transition model, and the transition model may learn a relation between the first resolution-based data set and the second resolution-based data set. The transition model can be controlled so as to

According to the present invention, resolution conversion between adjacent spatial regions in multi-resolution-based modeling and simulation is described using a cell automata technique, and in this process, the resolution between different resolution models of each cell due to the influence of adjacent cells is described. The transition can be solved by applying a data-based model or by applying a hybrid model of a data-based model and a theory-based model.

According to the present invention, a system model capable of integrally describing the entire target system by modeling a causal relationship between a plurality of causes and effects describing the target system and combining sub-models that can partially describe the target system is provided. can provide In this case, a coupling relationship and/or a data connection relationship between sub-models in the system model is defined based on structural information, and the structural information may be acquired using knowledge of the target system.

According to the present invention, it is possible to provide a hybrid model that operates by combining the advantages of a physical-based model and a data-based model. At this time, by extending the scope of application of the hybrid model, it is possible to implement a system model that can apply the hybrid model even to the system model described by the composite model.

According to the present invention, while maximizing the accuracy of the data-based model, it is possible to implement a system model capable of inferring data outside the data range, which is a limit of the data-based model.

According to the present invention, it is possible to provide a system model capable of describing the operation or state of a target system due to a complex cause using a data-based model. In addition, it is possible to generate descriptive information (explainable information or descriptive information) that can explain the interrelationship between data in the data-based model based on structural information.

According to the present invention, while using a data-based model, it is possible to implement a system model to which normative analysis, which is difficult to implement in a data-based model, can be applied.

According to the present invention, a hybrid model combining the advantages of a data-based model and a physically-based model can be implemented as a complex system model, and the complex system model can be applied to a digital twin for describing a very complex target system. have.

According to the present invention, the system model with the built-in data-based model embeds machine learning contents using the acquired data acquired through the operation and observation of the target system, so that the model itself becomes a verified model in the domain of the acquired data, and the data It is possible to overcome limitations that may be encountered when analyzing or predicting a target system by using only a base model or only a physics-based model.

By embedding machine learning into the system model, it reduces the degree of prior knowledge required about the target system and achieves the effect of model demonstration (verification with real data), while changing the structure/rule within the target system that machine learning alone could not do. It is possible to analyze and predict the behavior of the target system according to

According to the present invention, it can have the following advantages over the existing data-based model. In general, data-based models show very high accuracy for static data, but it is known that it is not easy to apply them to dynamic analysis with a variety of variables affecting motion. According to the present invention, a data-based model can also be applied to a target system having spatial variant, time variant, and stochastic characteristics.

Also, according to the present invention, non-linear characteristics of a target system may be reflected. Since many systems in the real world have non-linear characteristics without the theorem of overlap between input and output, it is a great advantage that the state transition function can consider non-linear characteristics. In addition, additional situational information such as weather conditions can be expressed through data obtained from real systems, so it can be applied as a robust modeling technique to the implementation of complex digital models such as digital twin models that mimic the real world.

1 is a diagram illustrating a computing system according to an embodiment of the present invention.
2 is a diagram illustrating an embodiment of a system model included in a computing system according to an embodiment of the present invention.
3 is a diagram illustrating an example of a process of deriving structural information used in a computing system according to an embodiment of the present invention.
4 is a diagram illustrating a domain of acquired data used by a computing system according to an embodiment of the present invention and an example of a theoretically expandable data domain.
5 is a diagram illustrating an embodiment of a sub-model included in a system model according to an embodiment of the present invention.
6 is a diagram illustrating a computing system according to an embodiment of the present invention.
7 is a diagram illustrating an example of an operation process of a system model according to an embodiment of the present invention.
8 is a diagram illustrating an embodiment of a theory-based primitive model on which a system model is based according to an embodiment of the present invention.
9 is a diagram illustrating an embodiment of a system model generated based on the theory-based primitive model of FIG. 8 .
10 is an operation flowchart illustrating a method of learning a system model executed by a computing system according to an embodiment of the present invention.
11 is an operational flow diagram illustrating a method executed by a computing system and operating a system model such that the system model infers a target system according to an embodiment of the present invention.
12 is a diagram illustrating an example of a manner in which a computing system describes a state change of a node in a theory-based model according to an embodiment of the present invention.
13 is a diagram illustrating an example of a target system to which a multi-resolution model is applied as a target system to be described by a system model according to an embodiment of the present invention.
14 is a diagram illustrating an example of a case where a system model according to an embodiment of the present invention intends to describe a traffic flow by applying a multi-resolution model as a target system to be described.
15 is a diagram illustrating disaggregation transition of data in a transition model and a multi-resolution model to be described by the transition model according to an embodiment of the present invention.

In addition to the above objects, other objects and features of the present invention will become apparent through the description of the embodiments with reference to the accompanying drawings.

A preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings. In describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted.

1 is a diagram illustrating a computing system 100 according to an embodiment of the present invention.

The computing system 100 includes a system model 120 that describes a target system, and may implement and operate the system model 120 . Specifically, the computing system 100 includes at least one processor 110 , and a communication interface 130 capable of receiving or calling acquisition data obtained from a target system.

The target system means a system in the real world to be analyzed, reasoned, and predicted, and may be a system operating according to a physical law or a system operating according to an operation rule. Laws on which the target system depends are not limited to physical laws, and according to embodiments, the target system may be described by social, psychological, economic rules or theories. The target system is closely related to the problem to be solved. When analyzing and predicting the spread of forest fires, the distribution of objects such as topography, trees, and rocks in the forest where wildfires can occur can constitute the target system. In the case of analyzing and predicting traffic flow, the target system may include a road network, a signal system, the capacity of each road, and a vehicle located on the road. In the case of analyzing the flow of traffic conditions in the event of an unexpected situation in this process, behavior changes due to human psychological behavior patterns may be added as elements to be identified in the target system. When analyzing and predicting the occurrence, spread, or healing process of a disease, cells, blood vessels, structures of a skeleton, organs, or blood flow may constitute a target system.

The target system may be a large-scale and complex system, such as a digital twin of a smart city. When there are very many variables that affect the target system, they can be combined to implement the target system and abstract the elements to be analyzed. For example, in the case of a smart city, national public data (demography, map, etc.) can be collected or utilized, and the collected data can be linked using an IoT infrastructure in the city. These data can be used to simulate changes in the behavior of citizens according to policy changes in a city, or to simulate economic activities and floating population activities according to disasters or geographic environments.

The processor 110 controls the system model 120 so that the system model 120 can learn the acquired data obtained from the target system for the implementation of the system model 120 . When a new input to the target system is given after the learning of the system model 120 is finished, the processor 110 generates the system model 120 based on the new input in order to predict an action to be taken by the target system in response to the new input. The system model 120 may be controlled to infer or predict the operation of the target system.

The system model 120 may include a theory-driven model and a data-driven model. The data-driven model may include an artificial neural network (ANN). The system model 120 is generally a collection of data stored in a memory or storage, and the analyzed parameters for the association between the acquired data as the learning object constitute the system model 120 . In the process of the processor 110 controlling the system model 120 , the processor 110 accesses the storage or memory in which the system model 120 is stored, and the processor 110 reads the system model 120 to perform other operations. Alternatively, it may be made in the form of rewriting a part of the parameters of the system model 120 using the result of the operation.

2 is a diagram illustrating an embodiment of a system model 120 included in the computing system 100 according to an embodiment of the present invention.

The system model 120 includes first sub-models 122a and 122b, and a second sub-model 124 . The first sub-models 122a and 122b may have the first data X1, X2, X3, and X4 as inputs and the second data Y1 and Y2 as outputs. The second sub-model 124 may have the second data Y1 and Y2 as inputs and the third data Z as outputs. The system model 120 may use the third data Z as a final output, but the spirit of the present invention is not limited to the embodiment of FIG. 2 .

Each of the plurality of sub-models 122a, 122b, and 124 included in the system model 120 is a model capable of receiving some of the acquired data and inferring or predicting at least some of the remaining data as outputs. Input data and output data of each of the plurality of sub-models 122a, 122b, and 124 in the system model 120 are to be defined based on structural information of the target system that can be obtained using knowledge of the target system. can The structural information includes information on a data connection relationship between the first sub-models 122a, 122b and the second sub-model 124 among the plurality of sub-models 122a, 122b, and 124, and the data connection relationship is the first Information on whether output data of the sub-models 122a and 122b can be used as input data of the second sub-model 124 may be included.

The first data X1 , X2 , X3 , X4 , the second data Y1 , Y2 , and the third data Z are data that can be obtained by actually measuring or observing the target system. The processor 110 is based on the structural information among the actual acquired data of the target system, the first sub-model (122a, 122b), and the first data (X1, X2, X3, X4) suitable for the second sub-model 124, The second data Y1 and Y2 and the third data Z may be selected.

The processor 110 provides the first data X1, X2, X3, and X4 as inputs of the first sub-models 122a and 122b based on the structural information of the acquired data, and provides the first data of the acquired data based on the structural information The two data Y1 and Y2 may be provided as outputs of the first sub-models Y1 and Y2 for learning the first sub-models 122a and 122b. The processor 110 may provide the second data Y1 and Y2 as an input of the second sub-model 124 based on the structural information. The processor 110 controls the first sub-model so that the first sub-models 122a and 122b can learn the relationship between the first data (X1, X2, X3, X4) and the second data (Y1, Y2). can

In addition, the processor 110 may provide the third data Z among the acquired data as an output of the second sub-model 124 for learning the second sub-model 124 based on the structural information. The processor 110 may control the second sub-model 124 so that the second sub-model 124 learns the relationship between the second data Y1 and Y2 and the third data Z.

In a state in which the first sub-models 122a and 122b and/or the second sub-model 124 are trained, the first sub-models 122a and 122b and/or the second sub-model 124 are generated based on new input data. It is possible to infer or predict the behavior of the target system.

In this case, the processor 110 may select new input data other than the measured data as the first data X1 , X2 , X3 , and X4 and provide it as an input to the first sub-models 122a and 122b . In this case, the data output by the inference or prediction of the first sub-models 122a and 122b is designated as the second data Y1 and Y2, and the second data Y1 and Y2 are the second data Y1 and Y2 of the second sub-model 124. It can be provided as input. The process in which the second data Y1 and Y2 output by the inference or prediction of the first sub-models 122a and 122b by the processor 110 are provided as an input of the second sub-model 124 is based on structural information can be executed. In addition, a process of selecting the first data X1 , X2 , X3 , and X4 from among the new input data by the processor 110 may also be executed based on the structural information.

The processor 110 allows the first sub-models 122a and 122b to infer or predict the operation, state, or state change of the target system based on the first data X1, X2, X3, and X4. The second sub-model 124 controls 122a and 122b, and allows the second sub-model 124 to infer or predict the operation, state, or state change of the target system based on the second data Y1 and Y2. ) can be controlled. At this time, FIG. 2 shows an embodiment in which the output data of the second sub-model 124 is provided to the user as the final output of the system model 120 as the third data Z. However, according to an embodiment of the present invention, the second Additional calculations may be performed based on the output data of the sub-model 124 , and thus data generated based on the output data of the second sub-model 124 may be provided to the user as the final output of the system model 120 . have.

At this time, each of the plurality of sub-models 122a, 122b, 124 is defined based on domain knowledge, experience, and theory that can be acquired in relation to the target system, and data that is learned based on a theory-based model or acquired data that can be deductively inferred It may be any one of the base models, or a combined model in which the theory-based model and the data-based model are complementary to each other. In this case, the data-based model may be a model using an artificial neural network (ANN). The theory-based model may be defined by physical laws, rules of operation, and the like, and obtainable domain knowledge, experience, and theories may include theories based on physics, psychology, economics, and sociology.

Although the general data-based model has recently improved the accuracy of data analysis and prediction through rapid development, the inside thereof is treated as a black box, and there is a problem in that it is difficult for humans to understand the internal operation. Therefore, accurate prediction is only possible within the actually obtained data domain, and it is difficult to accurately predict an input outside the data domain. On the other hand, a general theory-based model is capable of descriptive/deductive reasoning and inferences about inputs that are out of a set domain, but it is necessary to verify whether it predicts what will happen in a real system by applying real data.

The system model 120 according to an embodiment of the present invention can complementarily combine a data-based model and a theory-based model, and a large frame is configured like a theory-based model using structural information, and each sub-model The fields 122a, 122b, and 124 may be configured as a data-based model. In this case, since the entire structure of the system model 120 is obtained by knowledge, theory, or experience, descriptive/deductive reasoning is possible, and inference is also possible with respect to an input outside a predetermined domain. On the other hand, when each sub-model (122a, 122b, 124) is a data-based model, since it is a model already obtained based on data in the acquired data domain, additional verification by actual data is not required.

In addition, in the embodiment of the present invention, all sub-models 122a, 122b, 124 may be composed of a theory-based model, some of the sub-models 122a, 122b, 124 are theory-based models, and the rest are data-based models. may be configured. When all sub-models 122a, 122b, and 124 are composed of a theory-based model, consider a case where each sub-model 122a, 122b, and 124 is verified by actual data and exhibits the same accuracy as the data-based model. will be able

For example, in the system model 120 of FIG. 2 , only a specific sub-model 122a is a data-based model, and the other sub-models 122b and 124 may predict the actual target system with an accuracy comparable to that of the data-based model. It can be assumed that In this case, the specific sub-model 122a is valid only within the domain of the given acquisition data, but other sub-models 122b and 124 are operable in an extended range outside the domain of the acquisition data. The system model 120 can provide a solution that sufficiently satisfies the user's needs when the user wants to predict the operation of the target system when the user inputs X3 and X4 of the input data out of the existing range.

That is, when the target system can be modeled as a theory-based model in large part, if the performance of the data-based model is superior to that of the theory-based model only for a specific sub-model 122a, the data-based model is used as a specific sub-model 122a And, the remaining part of the system model 120 may use a theory-based model.

In general, data-driven models are known to have strength in predictive analysis that analyzes “what will happen?” based on descriptive analysis that analyzes “what happened?”, but “how did the event happen?” or "How do I get the results I want?" It is known that it is difficult to support a Normative Analysis that analyzes etc.

The system model 120 according to an embodiment of the present invention complementarily combines a data-based model and a theory-based model to support normative analysis using structural information based on theory while using a data-based model. .

In the above example, since the specific sub-model 122a is a black box, the operation rule of the specific sub-model 122a cannot be changed, but the operation rules of the other sub-models 122b and 124 can be changed. Also, it is not possible to change the internal operation rule of the specific sub-model 122a, but by changing the operation rules of other sub-models 122b and 124 in the vicinity connected to the specific sub-model 122a, the entire system model 120 ) may be treated as if the relative operation rules of a specific sub-model 122a are changed. Since it is possible to change the relative behavioral rules within the system model 120 , the system model 120 can perform normative analysis (including control policies, etc.) from predictive analysis, and cognitive analysis from normative analysis. (including planning, etc.);

If a single data-based model is treated as a black box, the theory-based model can be thought of as a white box, and the model combined with the data-based model and the theory-based model can be thought of as a gray box. At this time, when only the input and output of the data-based model are observed in connection, the corresponding sub-model block can be considered as a black box, but from the point of view of the entire system model 120, the input and output of the corresponding data-based model within the data domain Since the system operation can be completely reproduced by other theory-based models in the vicinity including data, from the perspective of the entire system model 120, even if the data-based model is included as a sub-block, it goes beyond the gray box and goes beyond the white box. It will be free to recognize as

In addition, according to an embodiment of the present invention, a theory-based model and a data-based model may be used together in one sub-model. For example, there may be a case where a theory-based model presents a mathematical relation, but does not accurately inform the coefficient or degree of the relation. In this case, the coefficients or orders of the theory-based model can be obtained by data-based learning.

For convenience of explanation, as a simple example, it is assumed that the first sub-models 122a and 122b are data-based models, and the second sub-models 124 are theory-based models. It may be noted that the second sub-model 124, which is a theory-based model, can be described by a simple linear relational expression of Z = mY1 + nY2. However, it is assumed that the values of the coefficients m and n cannot be known by theory alone. At this time, in order to obtain m and n, the coefficients m and n can be obtained by data-based learning between Y1, Y2, and Z.

That is, one sub-model may be modeled as a data-based model or, although it is a theory-based model, its coefficients or orders may be specified by data-based learning. Also, when the theoretical basis is clear, the theory-based model can be implemented as a highly definitive mathematical model.

There may be various sub-models such as these, and the system model 120 of the present invention is a feature that is different from the prior art in that the structure of the entire system model 120 is determined based on structural information, and such structural information Each sub-model is connected based on The system model 120 may provide a result such as a white box to the user based on the structural information. The characteristic of the system model 120 operating like a white box is explanable information or descriptive information about the correlation between data inside the system model 120 to the user even when the data-based model is used as a sub-model. It provides a platform that can provide

3 is a diagram illustrating an example of a process of deriving structural information used in a computing system according to an embodiment of the present invention.

Structural information defining the internal structure of the system model 120 may be obtained based on a theory-based primitive model that may be obtained using knowledge of the target system. The theory-based primitive model may be derived based on data connection relationships and causal relationships between parameters in the target system. Whether each of the various parameters included in the theory-based primitive model can be included in the acquired data, that is, whether it is a parameter actually measured or observed for the target system, is identified, and the identified information of each parameter is used as structural information for the target system is derived

Structural information can be derived through the process of deriving a theory-based primitive model from knowledge, experience, experiment, observation, theory, etc., and identifying data that can be obtained from the target system. This process may be executed by the computing system 100 or a separate computing system, or may be executed by human intervention in some processes and interaction with the computing system 100 .

For example, whether or not each of the parameters constituting each node of the theory-based primitive model corresponds to data that can be actually obtained from the target system in the state that setting the theory-based primitive model is executed by human behavior may be identified by the computing system.

In this case, the structural information may be selected as input data or output data of each of the plurality of sub-models among parameters in the target system constituting a theory-based raw model that can be obtained using knowledge about the target system. Information on related parameters may be included, and each of the related parameters may include sub-model data structure information defined as input data and output data of each of the plurality of sub-models based on the theory-based original model.

4 is a diagram illustrating a domain of acquired data used by a computing system according to an embodiment of the present invention, and an example of a theoretically expandable data domain.

When the first data (X1, X2, X3, X4) is obtained from the actual acquired data, the second data (Y1, Y2) appearing in the same temporal or spatial shift as the first data (X1, X2, X3, X4) The distribution of is shown in the acquisition data domain 410 . Meanwhile, the distribution of the third data Z appearing in the same temporal or spatial disparity as the second data Y1 and Y2 obtained from the actual acquired data is also shown in the acquired data domain 430 .

Data beyond the acquired data domains 410 and 430 may be provided as input to the system model 120 for the purpose of designing or planning the system model 120 . For example, if the first data (X1, X2, X3, X4) is out of the acquisition data domain 410, the system model 120 is the first data (X1, X2, X3, X1, X2, X3, out of the acquisition data domain 410) The operation of the target system is inferred based on X4), and the distribution of the second data Y1 and Y2 appearing at this time may constitute the extended data domain 420 .

Similarly, the system model 120 infers the operation of the target system based on the second data Y1 and Y2 that are out of the acquisition data domain 430, and the distribution of the third data Z that appears at this time is the extended data domain ( 440) can be configured.

The system model 120 of the present invention utilizes the accuracy of the data-based model to the maximum, but can infer or predict the operation of the target system even for data that is outside the acquired data domains 410 and 430, which are limits of the data-based model.

5 is a diagram illustrating an embodiment of a sub-model 122a included in the system model 120 according to an embodiment of the present invention.

The sub-model 122a of FIG. 5 includes a data-based model 510 and a theory-based model 520 therein, and the logic module 530 is located between the data-based model 510 and the theory-based model 520 . One output is selected or the output of the sub-model 122a is generated by combining the outputs of the data-based model 510 and the theory-based model 520 .

According to an embodiment of the present invention, the data-based model 510 and the theory-based model 520 may be competitively combined. For example, given the input data within the range of actual acquired data, the logic module 530 performs the data-based model 510 and the theory-based model based on the prediction accuracy of each of the data-based model 510 and the theory-based model 520 . It can be designed to select any one among models 520 .

According to an embodiment of the present invention, the data-based model 510 and the theory-based model 520 may be complementary to each other. Based on whether the input data (X1, X2) given in FIG. 5 is outside the acquisition data domain 410 , the logic module 530 applies either the data-based model 510 or the theory-based model 520 . you can choose For example, the data-based model 510 in the acquisition data domain 410 may be applied, and the theory-based model 520 may be applied to input data outside the acquisition data domain 410 .

In order to apply the theory-based model 520 in the extended data domain 420, a process for increasing the association/coherence between the theory-based model 520 and the data-based model 510 in the acquisition data domain 410 may be required. have. Such a process may be, for example, in the form of curve fitting for coefficients or orders of the theory-based model 520 . In order to improve the consistency between the theory-based model 520 and the data-based model 510 , a separate hybrid model having a form similar to that of the system model 120 of FIG. 2 may be applied.

4 and 5 together, attempts to extend data to extended data domains 420 and 440 are particularly difficult when, for example, the distribution of data within acquisition data domains 410 and 430 is qualitatively asymmetric. It makes sense. In the case of a system model implemented using a factory facility as a target system, normal operation data will be significantly more than abnormal operation data. Therefore, in the case of learning using data in an asymmetric state, there is a possibility that the system model is trained while being overfitted to the normal operation data. In addition, since it is not easy to acquire the abnormal operation data in an actual situation, the asymmetrical data distribution can be compensated for a little by expanding the data assuming a virtual situation.

Expansion of such data is necessary in order to secure a balanced distribution of data when predicting a future operation from a current operation of a target system. In addition, analysis and prediction of the system model for the extended data domains 420 and 440 not only in the normal operation/abnormal operation, but also in the case of design changes and planning for improvement of reliability, stability, quality, or performance of the target system. is needed

6 is a diagram illustrating a computing system 100 according to an embodiment of the present invention. Referring to FIG. 6 , the computing system 100 may further include a user interface 140 to easily implement a service model interacting with a user.

The computing system 100 further includes a user interface 140 for receiving and forwarding a user query regarding the target system to at least one processor 110 , wherein the at least one processor 110 interprets the user query to the computing system (Instructions can be converted to instructions that can be executed within 1000.

In response to the user's query, the processor 110 may interpret and express the operation and state of the target system so that the user can easily understand it based on the structural information.

7 is a diagram illustrating an example of an operation process of the system model 120 according to an embodiment of the present invention.

Referring to FIG. 7 , the sub-model 122a has a condition variable C1, the sub-model 122b has a condition variable C2, and the sub-model 124 has a condition variable C3. C1, C2, and C3 may be a condition variable, a control variable, or a design variable. Even if each of the sub-models 122a, 122b, and 124 is a data-based model, the condition variable reflected in the process of combining the sub-models 122a, 122b, and 124 with other sub-models in the system model 120 is It may be determined based on structural information.

When there is a user query, the processor 110 performs normative analysis, planning, and design optimization of the target system using structural information and condition variables/control variables/design variables C1, C2, and C3 for the user query. You can create a proposal.

The at least one or more processors 110 may include a user query regarding the association of at least two of the first data X1, X2, X3, X4, the second data Y1, Y2, and the third data Z. , the first data (X1, X2, X3, X4), the second data (Y1, Y2), and the third data ( Z) generates a corresponding search result by searching for at least one of a condition variable, a control variable, or a design variable of the target system corresponding to the association of at least two or more of Z), and generates a response to a user query based on the corresponding search result and a response to the user's query may be transmitted to the user interface 140 .

For example, when all of the sub-models 122a, 122b, and 124 are data-based models, the system model 120 predicts the third data Z from the first data X1, X2, X3, and X4. When the user query includes a process of deriving a result, the processor 110 connects between the second data Y1 and Y2, which are intermediate results that are not directly revealed in the system model 120, and the sub-models 122a, 122b, and 124 It is possible to provide descriptive information or descriptive information about the system model 120 describing the target system based on the relationship. In addition, the processor 110 may verify the validity, reliability, and stability of the descriptive information based on actual measurement data of the second data Y1 and Y2 that are intermediate results.

The user's query relates to the operation of the target system when the first data (X1, X2, X3, X4) is outside the acquisition data domain 410 covered by the acquisition data (intended harsh environment or unknown operating conditions). When a prediction for In response to the first data (X1, X2, X3, X4) may be controlled to predict the operation of the target system. At least one processor 110 applies one of a theory-based model capable of deductive reasoning as the first sub-models 122a and 122b, or a data-based model that is learned based on acquired data, or a theory-based model and data It can be applied to the first sub-models 122a and 122b by combining the base models to complement each other.

When the user query includes a condition in which the distribution of the third data Z is adjusted (it may include an intention to improve the reliability, stability, performance, and quality of the target system), at least one processor 110 . is a first distribution of data applicable to the first data (X1, X2, X3, X4) among the acquired data based on the structural information and the system model 120, and the second data (Y1, Y2) related to the first distribution ) and a third distribution of the third data Z related to the second distribution. In this case, it is assumed that the characteristic information indicated by the distribution of the third data Z in the acquired data domain 430 can be evaluated as an index such as stability, reliability, quality, and performance of the target system. This process, for example, varies the set of condition variables {C1, C2, C3} and repeats inference/prediction repeatedly to obtain simulation results corresponding to the set of condition variables {C1, C2, C3}. can be understood Alternatively, the processor 110 analyzes the relationship between the real distribution of each data in the acquired data and the target distribution of data representing a desired quality or performance, and a set of condition variables {C1, C2 for deriving a target distribution from the real distribution. , C3} may generate a proposal for a change. As such, at least one processor 110 provides a limiting condition of the range of the first distribution in response to a user query based on the first distribution, the second distribution, the third distribution, and the structural information, or A possibility of changing at least one of a condition variable, a control variable, or a design variable of the target system may be provided to the user.

For example, in FIG. 4, data is uniformly distributed in the real distribution in the acquisition data domains 410 and 430. In order to increase the stability and reliability of the target system or to improve the quality and performance of the final output, If data is more concentrated at the center of 410 and 430 , a target distribution may be set according to these needs. In this case, the user query may include complex needs. For example, the complex needs for improving the quality and performance of the final output without degrading the stability and reliability of the target system may be converted into a target distribution and set. The processor 110 may adjust the structural information and condition variables C1, C2, C3 in the system model 120 in order to obtain a target distribution in a given real distribution, or may generate a proposal to limit the range of input data.

8 is a diagram illustrating an embodiment of a theory-based primitive model 800 on which a system model is based, according to an embodiment of the present invention.

FIG. 9 is a diagram illustrating an embodiment of a system model 900 generated based on the theory-based primitive model 800 of FIG. 8 .

For system modeling, obtainable domain knowledge/experience and theory related to the target system are identified first, and a hypothetical model for the target system can be defined in the form of a gray box. Since it is difficult to simulate a hypothetical model by itself, it is necessary to construct a complete model by securing information such as operation functions and parameters necessary for model completion. The information necessary for model completion acquires big data by actually operating/observing the target system, and using the hypothetical model through machine learning for the acquired big data, the information necessary to generate the theory-based raw model 800 is provided. acquire By learning big data acquired through actual operation and observation of the target system using a machine learning algorithm such as an artificial neural network, information necessary for the theory-based primitive model 800 can be learned. The theory-based primitive model 800 for the target system is generated by applying the learned and verified information based on the actual data to the hypothetical model.

The theory-based primitive model 800 is implemented to have structural information by a number of parameters that can be represented in the target system. At this time, on the premise that all parameters are secured, the theory-based primitive model 800 corresponds to a white box. However, since not all parameters can actually be measured or observed from the target system, editing is required to implement them as the system model 900 .

It is assumed that y1' among the parameters in FIG. 8 is a parameter that cannot be actually measured or observed. The sub-models 810 and 820 related to the parameter y1' cannot be applied to the system model 900 as they are, and transformation is required to be applied to the system model 900 .

As shown in FIG. 9 , the sub-models 810 and 820 related to the parameter y1' may be merged into one sub-model 910 and included in the system model 900 . In this case, it is assumed that all of x1', x2', and w1' can be actually measured or observed from the target system.

Since only parameters that can be actually measured or observed from the target system are set as the main node of the system model 900, sub-models of the system model 900 may be implemented as data-based models or as theory-based models. . In this case, since the structural information includes information on a connection relationship and a coupling relationship between sub-models in the system model 900 , it may be involved in the operation of each sub-model as the semantics of the system model 900 .

In an embodiment of the present invention, the sub-models may be partially data-based models, and all sub-models may be configured as data-based models. However, since the structural information provides a connection relationship between each sub-model, that is, the internal structure of the system model 900, the system model 900 enables normative analysis while using a data-based model and provides descriptive information. and can support design changes and planning. If all of the parameters displayed in FIG. 9 are data that cannot be actually obtained in the target system, the system model 900 of FIG. 9 will be treated as a single black box, so it should be interpreted that the present invention is not applied in this case. something to do.

Structural information of the present invention is set based on observation data required for each sub-model to learn by treating the behavior of the sub-models in the system model 900 as a black box. If there is no observation data of the parameters inside the system model 900 , learning is impossible, and if there is no training data, the internal behavior of the system model 900 may not be detected.

The collected data (x1', x2', x3', x4', x5', x6', y2', y3', w1', w2', z') can be input variable values, output variable values, or states It does not matter whether it is a variable value, and it has an important meaning whether it is a value that can be actually measured or observed by a sensor or the like.

10 is an operation flowchart illustrating a learning method of the system model 120 executed by the computing system 100 according to an embodiment of the present invention.

The operating method of the system model 120 and the learning method of the system model 120 executed in the computing system 100 according to an embodiment of the present invention include a plurality of sub-models (122a, 122b, 124) is executed in a computing system 100 including a system model 120 , and at least one processor 110 .

According to an embodiment of the present invention, the learning method of the system model 120 executed by the computing system 100 implementing and operating the system model 120 describing the target system is obtained by the communication interface 140 receiving or calling data (S1010); providing, by the at least one processor 110, the first data (X1, X2, X3, X4) of the acquired data as input to the first sub-models (122a, 122b) (S1020); Providing, by the at least one processor 110 , the second data Y1 and Y2 defined as outputs of the first sub-models 122a and 122b as inputs of the second sub-model 124 based on the structural information (S1040); And at least one or more processors 110, the first sub-models 122a, 122b to learn the relationship between the first data (X1, X2, X3, X4) and the second data (Y1, Y2) Control the sub-models 122a and 122b ( S1030 ), or the second sub-model 124 is defined as an output of the second sub-model 124 based on the second data Y1 and Y2 and structural information. and controlling the second sub-model 124 to learn the relationship between the data Z ( S1050 ). When the first sub-models 122a and 122b do not require learning as a theory-based model, step S1030 may be omitted, and when the second sub-model 124 does not require learning as a theory-based model, Step S1050 may be omitted. However, at least one of the first sub-models 122a and 122b and the second sub-model 1224 is a data-based model, so that at least one of steps S1030 and S1050 is performed on the processor 110 is executed by

11 is an operational flow diagram illustrating a method executed by the computing system 100 and operating the system model 120 so that the system model 120 infers a target system in accordance with an embodiment of the present invention.

According to an embodiment of the present invention, the method of operating the system model 120 executed by the computing system 100 that implements and operates the system model 120 describing the target system includes the communication interface 140 being the target system. receiving or calling new input data to the system; at least one processor 110, selecting first data (X1, X2, X3, X4) based on structural information among new input data and providing the selected first data (X1, X2, X3, X4) as input to the first sub-models (122a, 122b); At least one processor 110, the first sub-model (122a, 122b) to infer the operation of the target system based on the first data (X1, X2, X3, X4), the first sub-model (122a, controlling 122b); providing, by the at least one processor 110 , the output data of the first sub-models 122a and 122b as second data Y1 and Y2 as inputs of the second sub-model 124 based on structural information; controlling, by the at least one processor 110 , the second sub-model 124 so that the second sub-model 124 can infer the operation of the target system based on the second data Y1 and Y2; and at least one processor 110 outputs the third data Z that is generated based on the output data of the second sub-model 124 or output data of the second sub-model 124 to the system model 120 and providing it to the user.

At this time, the operating method of the system model 120 of the present invention includes the step of at least one processor 110 interpreting a user query regarding the target system and converting it into an instruction command that can be executed in the computing system 100 (shown in the figure) not included) may be further included.

According to the present invention, it can have the following advantages over the existing data-based model. In general, data-based models show very high accuracy for static data, but it is known that it is not easy to apply them to dynamic analysis with a variety of variables affecting motion. According to the present invention, a data-based model can also be applied to a target system having spatial variant, time variant, and stochastic characteristics.

As an example of the spatial transition, when the entire system model is expressed on a cell-based basis, the state transition function may not be expressed for all cells as a whole, but may be expressed differently according to individual cells. Accordingly, the characteristics of the topography that vary depending on the location may be reflected for each cell. As an example of the time shift, the state transition function used to describe the system model may be set to be different depending on the time of morning, afternoon, or the like. For example, if there is a morning, afternoon, weekend, holiday, weekday, or specific event in the traffic flow, a system model can be implemented by reflecting seasonal factors. Such variation can be further specified and optimized through training of a data-based model.

As an example of the probabilistic feature, in the case of a traffic simulation, a point that the driver's behavior is not always determined, but may appear probabilistically may be reflected. Since the data-based model learns by reflecting only the acquired data, an event may occur with a specific probability under the conditions learned by the data-based model. At this time, when the situation of the road network is changed or a disaster occurs, the change of probability cannot be simulated in the pure data-based model, but in the system model of the present invention, it is probabilistic based on the data connection relationship within the model and structural information between the models. It can be simulated by adjusting or changing the properties.

Also, according to the present invention, non-linear characteristics of a target system may be reflected. Since many systems in the real world have non-linear characteristics without the theorem of overlap between input and output, it is a great advantage that the state transition function can consider non-linear characteristics. In addition, additional situational information such as weather conditions can be expressed through data obtained from the real system, so it is robust in the implementation of complex digital models such as digital twin models that mimic the real world.

12 is a diagram illustrating an example of a manner in which a computing system according to an embodiment of the present invention describes a state change of a node in a theory-based model.

A theory-based model is a model that describes the phenomenon of a target system by deductive reasoning based at least in part on physical laws or rules of operation. Parts that can be described as physical laws or behavior rules are executed by deductive reasoning, and if necessary, machine learning can be partially applied to update the state of a specific node.

12 shows an example of a model in which the target system is divided into cells according to spatial distribution and the state information of each cell is updated over time in order to represent a space variant and a time variant of the target system shown as hostile. The state variable s(i, j) of cell (i, j) may be updated over time. In this case, if the time t is included, the state variable can be expressed as s(t, i, j). Each cell may be configured to correspond to one node.

12 is an embodiment in which a target system is described using a cell automata technique, and a method of defining a cell automata and updating a state is shown. In addition, the cell automata model may learn a state transition function of each cell through data-based learning, and the learned state transition function may be provided to the cell automata model.

Using the big data collected from the target system, the transition function of each cell can be learned using machine learning such as artificial neural network modeling. The system model 120 described in FIGS. 1 to 11 may be implemented by combining the learned state transition functions and the cell automata model. When the system model 120 described in FIGS. 11 to 11 is applied, the empirical problem of the system model 120 as a theory-based model or simulation model is solved by a cooperative approach, and through simulation modeling based on structural information Through the implementation of the system model 120, improved modeling in which the structure and rules of the target system can be changed is possible.

The cell automata model is a discrete model dealt with in modeling mathematics, physics, complex systems, biology, microstructure, etc., and is defined in cells arranged in a regular grid. Each cell can have a finite number of states, and the grid is defined by a finite number of dimensions. For each cell, the cells called neighbors are defined in relation to that cell, for example, a neighbor can be defined as cells that are spaced apart from each other by one cell in all directions.

For example, when time t=0, the state of each cell is designated and this is called the initial state. A new generation is created from the previous generation by a state transition function. The conventional state transition function specifies a new state of each cell according to the state of each cell and its neighbors. It is a function.

In general, the above behavioral rules are the same for each cell, do not change over time and are applied simultaneously to all cells of each generation. The state transition refers to the process of transitioning from s(t, i, j) to s(t', i, j), and the state transition at this time may be represented by a state transition function T. The state of each cell can be defined according to the type of problem such as traffic, water pollution, and fire spread.

In general, the ideal state transition function reflects the state of the neighboring cells and the state transition occurs, but in the embodiment of FIG. 12, not only the state of the neighboring cells but also topographic information and external factors (weather conditions such as weather, temperature, etc.) are input as inputs. state transition can be made. That is, more accurate prediction of the next state of the cell can be made by reflecting the characteristics of the terrain in the ideal cell automata model and furthermore, real-time weather information.

At this time, the state transition function according to the embodiment of the present invention is not the same for each cell or does not change with time as in the conventional method, but varies according to the change in the location of the cell and the change in time, and the state transition rule is not deterministic. may have uncertainty. If the state transition function is obtained using domain knowledge as in the past, validation is required, but if the state transition function is obtained through machine learning of big data, the problem of validation can be solved because it is based on real data.

For example, the state variable s(t', i, j), in which the state variable s(t, i, j) is updated after a certain time elapses, is the cell/node (i, j) and adjacent cells/nodes (i, j). -1), (i-1, j), (i+1, j), (i, j+1) state variables s(t, i, j-1), s(t, i-1, j), s(t, i+1, j), s(t, i, j+1) and state variables s(t, i, j).

In this case, the state variables s(t', i, j) and s(t, i, j) and state variables s(t, i, j-1), s of adjacent cells/nodes are updated after a certain period of time has elapsed. Set the correlation parameters between (t, i-1, j), s(t, i+1, j), s(t, i, j+1) as parameters of the theory-based model describing the target system of FIG. can Meanwhile, s(t'', i, j), which is the next state of the state variable s(t', i, j), is the state variable s(t', i, j) and the state variables s(t) of adjacent cells/nodes. ', i, j-1), s(t', i-1, j), s(t', i+1, j), s(t', i, j+1) and parameters of the theory-based model can be obtained through In this case, the state transition function may be modeled by combining the theory-based model and the data-based model as described above.

In FIG. 12, only cells/nodes directly adjacent in space have been exemplified as affecting the next state, but this is only an example, and according to the laws of physics, the state change of each node is affected by spatial distribution and the passage of time according to the laws of physics. A variety of patterns can exist. Such a pattern may be defined by a physical law or may be searched for by a process such as machine learning.

In addition, although each node is defined according to the spatial distribution in FIG. 12 and an embodiment in which the state information of the node is updated over time is illustrated, the spirit of the present invention is not limited thereto, and in the theory-based model, each node has a spatial distribution and It may be distinguished and defined by at least one or more of / or the passage of time, and the state change of each node may be updated and tracked by at least one of spatial distribution and / or passage of time.

As an example of the state information indicated by each node according to the spatial distribution in FIG. 12, the spread of wildfire, the spread of fire in a building, the flooding status of a specific area, the ground/underground facilities (e.g., subway stations, tunnels, etc.) The flooding status by area can be mentioned. In addition, in an application such as a cyber city, information on the traffic flow status or whether or not a traffic jam occurs in an entire area or a partial area of the city may be displayed. Based on such state information, various service models such as intelligent evacuation route guidance by applying the theory-based model, digital twin model, and virtual sensor model of the present invention, and guidance of a movement route optimized in the current situation can be derived.

In addition, when used in an industrial field, a node may be each production facility in the production line or each part in the production facility, and the state information may be data such as vibration, heat generation, noise, etc., and information that can be determined through the production It may include information such as whether an abnormality has occurred in the facility, the location of the abnormal production facility or its parts, and the degree of abnormality.

When the theory-based model of the present invention is applied to the medical or biological field, a target system for analyzing and predicting the occurrence, spread, or healing process of a disease may be configured. In this case, the target system that can be modeled by the theory-based model may include a cell, a blood vessel, a structure of a skeleton, a structure of an organ, or blood flow. For example, when simulating blood flow in a blood vessel, the parameters included in the target system (parameters to be described) are fractional flow reserve (FFR), wall stress shear (WSS), and blood flow velocity (Velocity). ), and may include at least one of vascular pressure (Wall stress). In this case, the fractional blood flow reserve (FFR) refers to the ratio of the maximum blood flow of the distal and proximal normal blood vessels of the coronary artery stenosis site.

13 is a diagram illustrating an example of a target system to which a multi-resolution model is applied as a target system to be described by a system model according to an embodiment of the present invention.

In this case, although it has been described as multi-resolution for convenience of description, a case in which not only resolution but also fidelity is different may be included in the equivalent range. Resolution is defined according to SISO/SIW as "The degree of detail and precision used in the representation of real world aspects in a model or simulation" and fidelity is defined as "The degree to which the representation within a simulation is similar to a real world" feature, or condition in a measurable or perceivable manner".

Referring to FIG. 13 , a low-resolution model 1310 and a high-resolution model 1320 are illustrated. The low-resolution model 1310 may include regionally divided regional low-resolution models 1311, 1312, 1313, 1314, 1315, and 1316, and the high-resolution model 1320 includes regionally divided regional high-resolution models 1321, 1322, 1323, 1324, 1325, 1326).

A simulation model of macroscopic, macroscopic or microscopic resolution can be generated according to the purpose required in traffic simulation. Depending on the performance or effect index (scale) to be obtained through simulation, the type of object included in the model or the level of description may also vary. In order to properly respond to various analysis needs for traffic-related problems, it is possible to construct a simulation system suitable for the purpose by flexibly synthesizing unit regional models describing component objects. At this time, a technique for determining the resolution applied to each region according to a given condition is a known technique, and is known by Korean Patent Registration KR 10-1815511 and the like. Since applying a high resolution model when a low resolution is sufficient may lead to waste of memory and computing power, it is necessary to determine an appropriate resolution for each region for efficient operation.

For convenience of description, it is assumed that the simulation starts from the local low-resolution model 1311 . In this case, the first and second region low-resolution models 1311 and 1312 may be applied to the first region and the second region. Thereafter, it may be determined that the third area high-resolution model 1323 is applied to the third area by using conventional techniques and various known techniques. At this time, the simulation that has already been performed may activate the third regional low-resolution model 1313 from the first and second regional low-resolution models 1311 and 1312. From this dataset, the third regional high-resolution model 1323 for the third region ) needs to be obtained. Since this process is a process of subdividing the resolution, it can be called disaggregation.

It may be determined that the fourth region high-resolution model 1324 is also applied to the fourth region by the aforementioned conventional techniques and various well-known techniques. Thereafter, it may be determined to apply the fifth regional low-resolution model 1315 to the fifth region. In this case, the simulation that has already been performed can activate the fifth regional high-resolution model 1325 for the fifth region from the third and fourth regional high-resolution models 1323 and 1324, and from this dataset, the fifth region for the fifth region can be activated. It is necessary to obtain a five-region low-resolution model 1315 . Since this process is a process of lowering the resolution, it can be called aggregation.

Thereafter, the simulation may be ended by activating the sixth regional low-resolution model 1316 from the fifth regional low-resolution model 1315 .

In FIG. 13, an embodiment of a case in which regional low-resolution models (1311, 1312, 1313, 1314, 1315, 1316) and regional high-resolution models (1321,1322, 1323, 1324, 1325, 1326) are predefined for each region is shown, The spirit of the present invention is not limited to this embodiment, and it will be understood by those skilled in the art that a model having a required resolution for each region may be adaptively generated.

14 is a diagram illustrating an example of a case in which a system model according to an embodiment of the present invention intends to describe a traffic flow by applying a multi-resolution model as a target system to be described.

It is assumed that a macroscopic/Mesoscopic model is applied as a low-resolution model and a micro/Microscopic model is applied as a high-resolution model. For example, a micro model may be applied to road areas around an intersection, and a macroscopic model may be applied to other road areas.

In this case, according to the flow of simulation, the intensive model is applied to an area far from the intersection, and in this case, the number of vehicles on the entire road traveling in the same direction may be considered. n, which is an element of the set N, may be obtained as data in the corresponding region. When entering the area around the intersection, a microscopic model is applied. In this case, the number of vehicles per lane traveling in the same direction may be considered. Assuming that there are four lanes, n1, n2, n3, and n4 elements of the sets N1, N2, N3, and N4 for each lane may be acquired as data in the corresponding area. According to the simulation flow, disaggregation means the process of acquiring data n1, n2, n3, and n4 based on data n, and aggregation can mean the process of acquiring data n based on data n1, n2, n3, n4. have.

At this time, data n1, n2, n3, and n4 may correspond to the number and speed of vehicles in each lane traveling in the same direction, and data n may be the number and average speed of vehicles in all lanes traveling in the same direction. Each area can be meshed or segmented per unit length, for example 10 meters, 20 meters, or 100 meters, and the speed, average speed, maximum/minimum speed of the vehicle within that section can be given as data and parameters. have.

In the case of aggregation, for example, the number of vehicles can be given by n = n1 + n2 + n3 + n4, so the required calculation is not complicated. Since the given data set is simplified to N, there is no difficulty in operation. In this case, it is assumed that information loss occurs as the data set or parameters are simplified, but the lost information is not a problem if there is no major problem in achieving the purpose of the simulation.

In disaggregation, since it is necessary to obtain N1 x N2 x N3 x N4 from the simplified data and parameter set N, additional information is required in addition to the given low-resolution data and parameter set N. For this, various context information can be given as additional information. For example, location (may be classified according to the direction of lane travel (straight / left / right turn)), time zone (commuting time, normal time zone, late night), date (weekdays, weekends, holidays, holidays), season, weather ( Information on sunny, snow/rain, typhoon, yellow sand, fine dust) and regulatory information (traffic control or accident occurrence) may be given.

15 is a diagram illustrating disaggregation transition of data in a transition model and a multi-resolution model to be described by the transition model according to an embodiment of the present invention.

Referring to FIG. 15 , a data transition for disaggregation is defined as a target 1510 to be described, and a transition model 1520 for identifying the target 1510 is proposed. The transition model 1520 may model a data transition process for disaggregation by learning based on the big data set {n, (n1, n2, n3, n4) }.

At this time, the transition model 1520 may be implemented as a data-based model that is one single black box, and is a hybrid system model in which the data-based model and the theory-based model according to the embodiments of FIGS. 1 to 11 described above are combined. may be implemented by For example, there are physical laws or operating rules that are naturally applied, such as a conservation law in which vehicles entering and exiting a cell-separated area must be the same. The minimum structural information of the transition model 1520 may be defined by these physical laws or operation rules, and the transition model 1520 may have sub-models internally based on the structural information, and the sub-models are data-based models. can be implemented as When a complex data-based model based on structural information is applied to the transition model 1520, it has the same effect as if the actual data verification that the theory-based model must go through has already been performed while providing explanatory predictability. reliability can be increased.

14 and 15 , an embodiment in which a transition is made with respect to a data set is illustrated, but the spirit of the present invention is not limited thereto. For example, as various parameters obtained from the data set and the data set, a regional model including all parameters capable of describing a traffic flow may be defined based on a resolution given for each region. By applying the data-based model and the hybrid model illustrated in FIGS. 14 and 15 , a transition between resolution-based models for each region defined for each region may be modeled as a transition model.

12 to 15 , although the third regional high-resolution model 1323 is desired to be applied to the third region of FIG. 13, the second regional low-resolution model 1312 of the second region adjacent in the simulation flow is also When the cell automata technique of 12 is applied, the third local low-resolution model 1313 can be easily activated. In this case, a process of obtaining the third regional high resolution model 1323 based on the third regional low resolution model 1313 may be described using the transition model 1520 illustrated in FIGS. 14 and 15 . A third regional low-resolution model 1313, which is a low first resolution-based model, is obtained in the third region, and a third regional high-resolution model 1323, which is a high second resolution-based model, is obtained based on the third regional low-resolution model 1313. When acquiring, the transition model 1520 of FIGS. 14 and 15 may be applied.

The operating method and the learning method of the system model and/or the transition model executed by the computing system according to an embodiment of the present invention are implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. can be The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

However, the present invention is not limited or limited by the examples. Like reference numerals in each figure indicate like elements. Length, height, size, width, etc. introduced in the embodiments and drawings of the present invention may be exaggerated to help understanding.

As described above, the present invention has been described with specific matters such as specific components and limited embodiments and drawings, but these are only provided to help a more general understanding of the present invention, and the present invention is not limited to the above embodiments. , various modifications and variations are possible from these descriptions by those of ordinary skill in the art to which the present invention pertains.

Therefore, the spirit of the present invention should not be limited to the described embodiments, and not only the claims to be described later, but also all those with equivalent or equivalent modifications to the claims will be said to belong to the scope of the spirit of the present invention. .

100: computing system
110: processor
120: system model
130: communication interface
140: user interface
122a, 122b: first sub-model
124: second sub-model
1310: low-resolution model
1320: high-resolution model
1520: transition model

Claims (6)

A computing system that implements and operates a model describing a target system to which a multi-resolution modeling technique of representing at least one of a spatial disparity and a temporal disparity based on a cell, and describing at least one of two or more resolutions for each cell is applied In
the computing system
a communication interface capable of receiving or calling acquisition data obtained from the target system;
Based on the first resolution-based cell model obtained based on the first resolution of the target cell and the acquired data and including a state variable representing at least one of a spatial variation and a temporal variation of the target cell, the first resolution-based cell model a transition model for obtaining a second resolution-based cell model obtained based on a second resolution, which is higher than the first resolution, and including a state variable of the target cell; and
At least one processor;
The transition model is a relation between a first resolution-based data set in which the target cell is described based on the first resolution and a second resolution-based data set in which the target cell is described based on the second resolution among the acquired data. A computing system that is a model trained on the basis of According to claim 1,
The transition model is
A computing system which is a data-based model that learns the association between the first resolution-based data set and the second resolution-based data set. According to claim 1,
The transition model includes a plurality of sub-models,
Each of the plurality of sub-models is a model that can receive a part of the acquired data as an input and infer or predict at least a part of the rest as an output,
Input data and output data of each of the plurality of sub-models in the transition model are defined based on structural information of the target system that can be obtained using knowledge of the target system,
The structural information includes information on a data connection relationship between a first sub-model and a second sub-model among the plurality of sub-models, and the data connection relationship indicates that the output data of the first sub-model is the second sub-model. Computing system including information on whether it can be used as input data of According to claim 1,
At least one of the spatial shift and the temporal shift is represented based on a cell by a cell automata technique. According to claim 1,
the at least one processor
Selecting first data generated to correspond to the first resolution among new input data for the target cell and providing it as an input of the transition model;
and controlling the transition model so that the transition model can infer the second resolution-based cell model from the first resolution-based cell model based on the first data. The method of claim 1,
the at least one processor
providing the first resolution-based data set as an input of the transition model;
A computing system for controlling the transition model so that the transition model learns a relationship between the first resolution-based data set and the second resolution-based data set.

Images