Machine Learned Control Heuristics for Complex Dynamical Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex dynamical systems, such as vehicles, require increasingly complex controls that demand high computational resources and longer computation times, posing challenges in efficient programming and configuration.
Innovation Solution
A method using machine learned, scenario-based control heuristics that employs a Model Predictive Control algorithm and machine learning algorithms like diffusion maps or support vector machines to generate optimal control values, which are then used to control the system efficiently, reducing computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional Model Predictive Control algorithms are used to control complex dynamical systems, then control accuracy is maintained, but computational requirements and computation time increase significantly
Solution Approach 1:
The patent pre-computes optimal control values for a comprehensive set of scenario parameter vectors and initial system state vectors before actual system operation. This offline pre-computation creates a lookup table of optimal controls that can be quickly retrieved during real-time operation, eliminating the need for complex real-time optimization calculations while maintaining control accuracy
Solution Approach 2:
The patent creates simplified representations (copies) of the complex control problem by discretizing scenario parameters and system states into finite sets. These discrete copies allow the control problem to be solved offline and stored in lookup tables, replacing complex real-time computations with simple table lookups that maintain accuracy while reducing computational burden
2Measurement precision
If conventional Model Predictive Control algorithms are used to control complex dynamical systems, then control accuracy is maintained, but computation time increases
Solution Approach 1:
The patent performs all computationally intensive optimization calculations in advance during an offline phase, generating lookup tables containing optimal control values for various scenarios. During real-time operation, the system only needs to perform simple table lookups based on current system state, reducing computation time from potentially seconds to milliseconds while preserving control accuracy
Solution Approach 2:
The patent implements a two-phase control approach: an offline phase where complex computations are performed with full computational resources available, and an online phase where pre-computed solutions are rapidly retrieved. This dynamic separation of computation phases allows accurate control without real-time computational delays
3Adaptability or versatility
If the system handles increasingly complex dynamical systems, then system capability is improved, but programming and configuration requirements increase
Solution Approach 1:
The patent creates a universal control framework using lookup tables that can handle various complex dynamical systems through a standardized interface. The same basic architecture and algorithm structure can be applied to different systems by simply changing the pre-computed lookup tables, eliminating the need to rewrite control software for each new system while maintaining the ability to handle increasing system complexity
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for performing an optimized control of a complex dynamical system (sys) using machine learned, scenario based control heuristics, the method comprising the steps of: providing (S1) a simulation model (f) for predicting a system state vector (x) of said dynamical system (sys) in time based on a current scenario parameter vector (p) and a control vector (u); using (S2) a Model Predictive Control, MPC, algorithm to provide the control vector (u) at every time during a simulation of said dynamical system (sys) using said simulation model (f) for different scenario parameter vectors (p0, p1, p2, ..) and initial system state vectors (x00, x01, x02, ..); calculating (S3) for every simulated combination of a scenario parameter vector (p) and initial system state vector (x0) a resulting optimal control value (u* (p, x0)) by the MPC algorithm and saving the resulting optimal control value; generating (S4) machine learned control heuristics (u a(p, x0)) approximating the relationship between the corresponding scenario parameter vector (p) and the initial system state vector (x0) for the saved resulting optimal control value (u* (p, x0)) using a machine learning algorithm; and using the generated machine learned control heuristics to control (S5) the complex dynamical system (sys) modelled by said simulation model (f).