Hybrid Neural ODE State Quantity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing state quantity prediction methods, particularly those using physical models, face challenges in accurately predicting dynamic characteristics, leading to increased prediction errors when assumptions are not met, and machine learning models fail to consider dynamic changes in equilibrium points, resulting in insufficient prediction accuracy.
Innovation Solution
A state quantity prediction device and method that combines a static physical model with a neural network to handle nonlinear dynamic characteristics by separating the prediction function into nonlinear and linear components, using a neural ODE to learn dynamic changes and integrating differential predictive values to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a physical model is used for prediction, then the model can be based on static equilibrium equations or state equations, but prediction error increases when assumed preconditions are not met
Solution Approach 1:
The patent combines a physical model with a neural ODE model to create a hybrid prediction system. The physical model provides static equilibrium predictions while the neural ODE learns dynamic characteristics of prediction errors, merging the strengths of both approaches to achieve accurate predictions under varying conditions
2Adaptability or versatility
If a machine learning model is used for prediction, then dynamic characteristics can be learned, but prediction error increases in extrapolation areas of learning data
Solution Approach 1:
The neural ODE acts as an intermediary that learns the dynamic characteristics of prediction errors between the physical model and actual measurements. It bridges the gap by modeling how errors evolve over time, enabling accurate predictions in extrapolation areas while capturing dynamic behavior
3Adaptability or versatility
If the scale of the neural network is increased to learn nonlinear behavior, then dynamic characteristics can be captured, but generalization performance decreases due to overlearning
Solution Approach 1:
The patent uses a neural ODE with a fixed-scale neural network that dynamically adapts to nonlinear behaviors through continuous time representation. The differential equation structure allows the model to capture complex nonlinear dynamics without increasing network scale, preventing overlearning while maintaining generalization performance
Data Source
AI summary
A state quantity prediction device includes: a first differential predictive value calculation unit configured to deal with a nonlinear component of a function whose variables are dynamic characteristics of the state quantity with respect to the input parameter and a difference value between a past predictive value of the state quantity and a predictive value of the physical model, input the input parameter and the past predictive value of the state quantity, and include a learned neutral network for outputting a first differential predictive value; a second differential predictive value calculation unit configured to deal with a linear component of the function, input the input parameter and the past predictive value, and output a second differential predictive value; and a state quantity predictive value calculation unit configured to calculate a predictive value of the state quantity.