Learning device, learning method, and program

Abstract

One of the objectives is to provide a learning device, learning method, and program that can provide a machine learning model capable of accurately estimating the flight state of a flying object. [Solution] The learning device of the embodiment includes a processing unit that learns a machine learning model to output the flight state in response to input pressure at multiple locations on the surface of a flying object, based on multiple training datasets which combine pressure at multiple locations on the surface of the flying object with the flight state which is the state of the flying object while it is in flight. The processing unit adjusts the weights of the training datasets to be reflected in the learning of the machine learning model based on the probability of the flight state which the flying object can take.

Claims

[Claim 1] The system includes a processing unit that trains a machine learning model to output the flight state in response to input pressures at multiple locations on the surface of a flying object, based on multiple training datasets that combine pressures at multiple locations on the surface of the flying object with the flight state, which is the state of the flying object while it is in flight. The processing unit adjusts the weights of the training dataset to be reflected in the learning of the machine learning model based on the probability of the flight state that the flying object can take. Learning device. [Claim 2] The aforementioned flying object is a flying object, The aforementioned flight conditions include at least the Mach number, dynamic pressure, pitch angle, and yaw angle of the projectile. The learning device according to claim 1. [Claim 3] The aforementioned processing unit, Using a trajectory calculation method, the flight states that the flying object can take are calculated as the trajectory of the flying object. Based on the frequency of the trajectory of the aforementioned flying object, the probability is calculated. The learning device according to claim 1 or 2. [Claim 4] When the processing unit aims to accurately determine the flight state with a high probability, it increases the weight as the probability increases and decreases the weight as the probability decreases. The learning device according to claim 1 or 2. [Claim 5] When the processing unit aims to accurately determine the flight state with a low probability, it reduces the weight as the probability increases and increases the weight as the probability decreases. The learning device according to claim 1 or 2. [Claim 6] The aforementioned processing unit, The pressures at multiple locations included in the training dataset are input to the machine learning model. The error between the estimated flight state, which is the flight state output by the machine learning model in response to the input pressure at the multiple locations, and the ground truth flight state, which is the flight state included in the training dataset, is calculated. The machine learning model is trained so that the error weighted by the aforementioned probabilities becomes smaller. The learning device according to claim 1 or 2. [Claim 7] A learning method using computers, This includes training a machine learning model to output the flight state in response to input pressures at multiple locations on the surface of a flying object, based on multiple training datasets that combine pressures at multiple locations on the surface of the flying object with the flight state, which is the state of the flying object while it is in flight. Training the machine learning model includes adjusting the weights of the training dataset to be reflected in the training of the machine learning model based on the probabilities of the flight states that the flying object can take. Learning methods. [Claim 8] A program to be executed by a computer, This includes training a machine learning model to output the flight state in response to input pressures at multiple locations on the surface of a flying object, based on multiple training datasets that combine pressures at multiple locations on the surface of the flying object with the flight state, which is the state of the flying object while it is in flight. Training the machine learning model includes adjusting the weights of the training dataset to be reflected in the training of the machine learning model based on the probabilities of the flight states that the flying object can take. program.

Images