Prediction Model Structure Learning for Unseen State Reproducibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing models constructed by supervised learning from system data have low reproducibility for states not experienced during training, leading to issues like overfitting and degraded prediction performance.
Innovation Solution
A learning apparatus and method that determines a prediction model structure from system function and structure knowledge, uses input value determination to refine learning, and updates model parameters to minimize output differences, employing neural ordinary differential equations and gradient methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a prediction model is constructed by supervised learning using data obtained from a control target system, then the model can be trained on actual system behavior, but the reproducibility of states away from training data is low
Solution Approach 1:
The patent applies preliminary action by incorporating domain knowledge about the control target system into the prediction model structure before training. The structure determination unit uses information about system functions and structures to pre-configure the prediction model architecture, ensuring it reflects actual system behavior patterns. This preliminary structuring enables the model to generalize better to unseen states while maintaining accuracy on training data.
2Reliability
If the prediction model structure is determined from system information, then the model structure aligns with actual system behavior, but the model complexity increases
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the prediction model structure based on input data characteristics. The structure determination unit modifies model parameters such as the number of hidden layers, neurons per layer, and connection patterns according to the specific control target system being modeled. This allows the model to achieve high reproducibility for unseen states while maintaining appropriate complexity levels for each application.
3Measurement precision
If supervised learning is performed using collected data, then the model learns from actual system data, but overfitting occurs and generalization performance degrades
Solution Approach 1:
The patent applies feedback by implementing a structure determination unit that continuously evaluates model performance and adjusts the prediction model structure accordingly. During training, the system monitors both training accuracy and generalization performance, using this feedback to refine the model architecture. This feedback mechanism prevents overfitting by ensuring the model maintains appropriate complexity and generalization capability while learning from training data.
Data Source
AI summary
An object of the present disclosure is to perform learning of a model relatively efficiently. A prediction model apparatus according to the present disclosure includes: prediction model structure determination means for determining a structure of a prediction model by using information about a structure of a system to be modeled; and model learning means for performing learning of the model so that a difference between an output value of the system to be modeled and an output value of the model becomes small.