Actuator Control Parameter Tuning via Stationary Probability Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing actuator control systems face challenges in optimizing control strategies without a specified time horizon, particularly in adapting to changing conditions and ensuring reliable performance.
Innovation Solution
The method employs a Gaussian process model to determine a stationary probability distribution, which is adapted based on the actuator's manipulated and control variables, allowing for optimal parameter adjustment and iterative improvement of the control strategy using numerical quadrature and support point density optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a specified time horizon is used for control optimization, then computational complexity is reduced, but the control strategy cannot achieve optimal adjustment with unlimited time horizon
Solution Approach 1:
The patent segments the control optimization process into two distinct phases: an offline learning phase where the Gaussian process model is trained using historical data, and an online control phase where the pre-trained model is used for real-time parameter adjustment. This segmentation allows computationally intensive operations to be performed offline, reducing online computational complexity while maintaining optimal control performance.
Solution Approach 2:
The patent performs preliminary action by training the Gaussian process model offline before actual control operations. The model learns the system's behavior patterns and stationary probability distribution in advance, so that during online control, only lightweight inference is needed. This preliminary training enables the system to achieve optimal adjustment with unlimited time horizon without incurring high computational costs during real-time operation.
2Adaptability or versatility
If the model is continuously adapted based on manipulated and control variables, then the control strategy becomes more adaptive to changing conditions, but computational requirements increase
Solution Approach 1:
The patent implements periodic action by updating the Gaussian process model at discrete intervals rather than continuously. The model is adapted based on accumulated manipulated and control variables over time, performing batch updates that balance adaptability with computational efficiency. This periodic adaptation allows the system to respond to changing conditions while avoiding the excessive computational burden of continuous model retraining.
3Measurement precision
If numerical quadrature with optimized support point density is used, then integration accuracy over possible values of control variable is improved, but computational complexity increases
Solution Approach 1:
The patent applies local quality by optimizing the density of support points in the numerical quadrature based on the local characteristics of the probability distribution. Regions where the distribution has higher density or greater curvature receive more support points, while regions with lower density receive fewer points. This non-uniform distribution of support points maintains high integration accuracy in critical regions while reducing the total number of support points needed, thereby lowering computational complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for automatically setting at least one parameter (θ) of an actuator control system (45) which is designed to control a control variable (x) of an actuator (20) to a predefinable setpoint (xd). The actuator control system (45) is designed to generate a control variable (u) depending on the at least one parameter (θ), the setpoint (xd) and the control variable (x) and to actuate the actuator (20) depending on this control variable (u). A new value (θ*) of the at least one parameter (θ) is determined depending on a stationary probability distribution (p, θ) of the control variable (x), and the parameter (Θ) is then set to this new value (θ*).