Neural Network Controller With Case-Based Loss Function For Closed-Loop Stability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional learning devices and AI systems, such as those described in Patent Literature 1, do not adequately consider closed-loop stability when controlling systems like robots or unmanned aircraft, which is crucial for maintaining system performance and safety.
Innovation Solution
A multilayer neural network controller with a weight matrix updated using a loss function divided into cases based on the gain of the closed loop, incorporating a penalty term to ensure stability, employs spectral normalization and Linear Matrix Inequality (LMI) solutions to maintain closed-loop stability and design a region of attraction (ROA).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional learning devices use simple loss functions to train neural networks, then training simplicity is improved, but closed-loop stability deteriorates
Solution Approach 1:
The patent changes the parameter structure of the loss function by introducing case-based division according to closed-loop gain values and incorporating penalty terms. This transforms the simple scalar loss function into a multi-parameter function that adapts to different operating conditions, thereby achieving both training effectiveness and closed-loop stability.
Solution Approach 2:
The patent implements feedback mechanisms by using the closed-loop gain as a criterion for case division and by incorporating penalty terms that provide feedback on stability violations. This allows the training process to adaptively adjust based on the actual closed-loop performance, ensuring stability while maintaining training feasibility.
2Measurement precision
If neural networks are trained to imitate expert techniques, then imitation accuracy is improved, but closed-loop stability deteriorates
Solution Approach 1:
The patent applies local quality by differentiating the loss function into different cases based on closed-loop gain values. Each case has its own loss function characteristics tailored to specific operating conditions, allowing the network to achieve high imitation accuracy in each local region while maintaining overall closed-loop stability through the structured case-based approach.
3Stability of the object's composition
If penalty terms are added to loss functions to ensure stability, then closed-loop stability is improved, but computational complexity deteriorates
Solution Approach 1:
The patent segments the loss function into distinct cases based on closed-loop gain thresholds. Each segment has a simplified loss function form with specific penalty terms appropriate for that gain range. This segmentation allows the complex stability-constrained optimization to be broken down into multiple simpler sub-problems, reducing overall computational complexity while maintaining stability guarantees.
Data Source
AI summary
A neural network controller according to the present disclosed technology is a multilayer neural network controller having a weight matrix. The weight matrix of the neural network controller is updated on the basis of a loss function that is divided into cases by the gain of the closed loop and that is switched in a mode of presence or absence of a penalty term.