Search

KR-20260067131-A - System for generating of virtual sensor data for simulating the physical world

KR20260067131AKR 20260067131 AKR20260067131 AKR 20260067131AKR-20260067131-A

Abstract

A method for generating virtual sensor data for simulating the physical world comprises: a step of acquiring real sensor data collected from a real sensor; a step of training a deep learning network model to generate virtual sensor data using the acquired real sensor data; a step of generating virtual sensor data using the trained deep learning network model; a step of performing a simulation within a digital twin environment; and a step of verifying the validity of the virtual sensor data based on the simulation results.

Inventors

  • 김지은

Assignees

  • 한국전자기술연구원

Dates

Publication Date
20260512
Application Date
20241105

Claims (12)

  1. The system acquires actual sensor data collected from an actual sensor; A step in which the system performs training of a deep learning network model for generating virtual sensor data using acquired actual sensor data; A step in which the system generates virtual sensor data by utilizing a trained deep learning network model; The system performs a simulation within a digital twin environment; and A method for generating virtual sensor data for simulating a physical world, comprising the step of verifying the validity of virtual sensor data based on simulation results.
  2. In claim 1, The step of performing a simulation within a digital twin environment is, A method for generating virtual sensor data for simulating the physical world, characterized by performing a simulation within a digital twin environment using actual sensor data and virtual sensor data.
  3. In claim 2, A method for generating virtual sensor data for simulating a physical world, characterized by further including the step of the system performing retraining of a deep learning network model according to the result of validation of validity.
  4. In claim 3, The system, A method for generating virtual sensor data for simulating a physical world, characterized by repeating the process of generating virtual sensor data using the retrained deep learning network model, performing a simulation in a digital twin environment reflecting the generated virtual sensor data, and verifying the validity of the virtual sensor data based on the simulation results, until the verification of the validity of the virtual sensor data is passed when retraining a deep learning network model.
  5. In claim 3, The step of verifying validity is, A method for generating virtual sensor data for simulating a physical world, characterized by verifying the validity of virtual sensor data by comparing the results of a simulation in a digital twin environment using only actual sensor data with the results of a simulation in a digital twin environment using both actual sensor data and virtual sensor data.
  6. In claim 5, The system, A method for generating virtual sensor data for simulating a physical world, characterized by utilizing the results of a simulation as correct data in a digital twin environment that uses only actual sensor data, comparing the results of the simulation with the results of a digital twin environment that uses actual sensor data and virtual sensor data, and obtaining the data values of the virtual sensor data based on the comparison results.
  7. In claim 3, The system, When performing a simulation, utilize a simulation model that performs the simulation within a digital twin environment, and The simulation model is, A method for generating virtual sensor data for simulating the physical world, characterized by a model that learns the probability of precipitation P(Rain) or precipitation amount (Rn) according to temperature (T) and humidity (H) through a deep learning network model.
  8. In claim 6, The above simulation model is, A method for generating virtual sensor data for simulating a physical world, characterized by calculating the probability of precipitation P(Rain) by referring to the following Equation 1, in the case of a model that learns the probability of precipitation P(Rain) according to temperature (T) and humidity (H) through a deep learning network model. (Formula 1) P(Rain)=1/1+e(β 0 +β 1 T+β 2 H)
  9. In claim 8, The system, A method for generating virtual sensor data for simulating a physical world, characterized in that, when a simulation model is a model that learns P(Rain), which is the probability of precipitation according to temperature (T) and humidity (H), if the values of the physical coefficients β0 , β1 , and β2 of Equation 1 finally converge to specific values, the verification of the validity of the virtual sensor data is considered passed, and further retraining of the deep learning network model is stopped.
  10. A communication unit for acquiring actual sensor data collected from an actual sensor; and A virtual sensor data generation system for simulating the physical world, comprising: a processor that uses acquired actual sensor data to train a deep learning network model for generating virtual sensor data, utilizes the trained deep learning network model to generate virtual sensor data, performs a simulation within a digital twin environment, and verifies the validity of the virtual sensor data based on the simulation results.
  11. A step in which the system performs training of a deep learning network model for generating virtual sensor data using actual sensor data collected through actual sensors; A step in which the system generates virtual sensor data by utilizing a trained deep learning network model; A step in which the system performs a simulation within a digital twin environment using actual sensor data and virtual sensor data; The system verifies the validity of virtual sensor data based on simulation results; and A method for generating virtual sensor data for simulating the physical world, comprising the step of the system performing retraining of a deep learning network model based on the result of validation of validity.
  12. A learning unit that performs training of a deep learning network model to generate virtual sensor data using actual sensor data collected through actual sensors, or performs retraining of the deep learning network model according to the results of validation; A virtual sensor data generator that generates virtual sensor data using a trained deep learning network model; A simulation execution unit that performs a simulation within a digital twin environment using actual sensor data and virtual sensor data; and A virtual sensor data generation system for simulating the physical world, comprising a validation unit that validates the validity of virtual sensor data based on simulation results.

Description

System for generating of virtual sensor data for simulating the physical world The present invention relates to a data generation system for a virtual sensor, and more specifically, to a system for generating virtual sensor data by performing training on a virtual sensor data generator that generates a virtual sensor for simulating the physical world, and utilizing the trained virtual sensor data generator to generate virtual sensor data. A virtual sensor is a technology that generates virtual data by learning from data acquired from physical sensors. It is utilized in cases where actual physical sensors cannot be installed in every desired location due to reasons such as cost or visibility. In the case of most of these virtual sensors, since data is generated using data-driven learning algorithms, sensor data can be generated by first acquiring data regarding the location to be detected and then learning from the measured data. In this approach, there is a disadvantage in that the significance of using virtual sensors is weakened because the validity of the data can be verified based on actual measured data values. In other words, virtual sensor algorithms of this type must first acquire data regarding the desired location, and since data-driven learning methods are utilized, there are limitations in simulating the real physical world. FIG. 1 is a drawing provided in the description of the configuration of a virtual sensor data generation system for simulating a physical world according to an embodiment of the present invention. FIG. 2 is a drawing provided for a more detailed configuration description of the processor illustrated in FIG. 1, and FIGS. 3 and 4 are drawings provided to describe a method for generating virtual sensor data for simulating a physical world according to an embodiment of the present invention. The present invention will be described in more detail below with reference to the drawings. To clearly explain the invention, parts unrelated to the description have been omitted from the drawings, and in the drawings, the width, length, thickness, etc., of the components may be exaggerated for convenience. FIG. 1 is a diagram provided for the configuration description of a virtual sensor data generation system for simulating the physical world according to one embodiment of the present invention. A virtual sensor data generation system for simulating a physical world according to the present embodiment (hereinafter collectively referred to as the 'system') is, By simultaneously applying a data-driven learning method and a learning method that reflects the real physical world to a virtual sensor data generator, the virtual sensor data generator is trained based on a data-driven approach to simulate the real world more effectively than in existing cases. Designed to accurately reflect physical objects, the virtual sensor data generator can be calibrated by reflecting simulation results within a digital twin environment that virtually represents objects. To this end, the system may include a communication unit (100), a processor (200), and a storage unit (300). The communication unit (100) is equipped with a communication module connected to a network, so that it can acquire actual sensor data collected from an actual sensor. The storage unit (300) is provided to store programs and data necessary for the operation of the processor (200). The processor (200) is designed to learn a virtual sensor data generator based on a data-driven approach and accurately reflect a physical object, and can process all matters for correcting the virtual sensor data generator by reflecting the simulation results within a digital twin environment that virtually represents the object. Specifically, the processor (200) selects a real physical model and, through the communication unit (100), acquires real sensor data of a real sensor connected to the selected real physical model, and can perform training of a virtual sensor data generator to generate virtual sensor data using the acquired real sensor data. A virtual sensor data generator can generate virtual sensor data using a deep learning network model. That is, the processor (200) can use actual sensor data as training data for a deep learning network model. And the processor (200) can generate virtual sensor data by utilizing the learned deep learning network model to correct the deep learning network model of the learned virtual sensor data generator, and can perform a simulation in a digital twin environment by utilizing the generated virtual sensor data and actual sensor data. Here, the simulation model that performs simulation within the digital twin environment refers to the model of the digital twin environment that virtually represents the actual physical model. For example, if the actual physical model is a model for the probability of precipitation P(Rain) or amount of precipitation (Rn) according to temperature (T) and humidity (H), the simulation mo