Bayesian Control Parameter Search With Learned Acquisition Functions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing Bayesian optimization methods for technical systems rely on manually defined acquisition functions, which can lead to inefficient searches, particularly in high-dimensional problems, and require extensive computing time, limiting their effectiveness in finding functional optima with minimal tests.

Innovation Solution

Automatically determining acquisition functions using reinforcement learning or imitation learning within a simulation environment, allowing for iterative optimization and adaptation based on process quality, resulting in more efficient and faster evaluation of test points.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If manually defined acquisition functions are used in Bayesian optimization, then the method can be implemented with simple manual configuration, but the search efficiency is poor and computing time is extensive

Engineering Contradiction:
ImproveEase of implementationVSAvoidSearch efficiency
Core Design Contradiction:
Ease of manufactureVSProductivity

Solution Approach 1:

The system automatically determines acquisition functions using reinforcement learning or imitation learning, enabling the optimization process to self-improve without manual intervention. The learning method iteratively optimizes the acquisition function based on feedback from simulation results, allowing the system to adapt to specific problem classes autonomously.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The invention changes the acquisition function parameters dynamically through learning processes. Instead of using fixed manually-defined acquisition functions, the system learns and adapts the acquisition function parameters based on problem-specific characteristics, thereby improving search efficiency for different problem types.

Inventive Principle:
Principle #35Parameter changes

2Ease of operation

If manually defined acquisition functions are used in Bayesian optimization, then the implementation is straightforward, but the computing time is extensive

Engineering Contradiction:
ImproveEase of implementationVSAvoidComputing time
Core Design Contradiction:
Ease of operationVSLoss of time

Solution Approach 1:

The system performs preliminary learning in a simulation environment before actual optimization. The acquisition function is trained offline using reinforcement learning or imitation learning on simulated data, so that when deployed on real technical systems, it requires minimal computing time during the actual optimization process.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The invention uses simulation copies of the technical system to train the acquisition function. By learning from simulated environments that replicate real system behavior, the system avoids extensive computing time on actual hardware while still achieving effective optimization strategies.

Inventive Principle:
Principle #26Copying

3Device complexity

If conventional acquisition functions are used, then the method is simple to implement, but fewer test points cannot be achieved

Engineering Contradiction:
ImproveMethod complexityVSAvoidOptimization precision
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The learning method automatically adapts the acquisition function to the specific problem class through self-learning in simulation environments. This self-service capability allows the system to achieve higher optimization precision without requiring complex manual configuration or extensive test points.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The invention dynamically adjusts acquisition function parameters based on learned problem characteristics. By changing the acquisition function parameters according to specific problem types, the system achieves better optimization precision with fewer test points compared to conventional fixed acquisition functions.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3650964B1Method for controlling or regulating a technical system
Publication Date: 2022.08.10 ROBERT BOSCH GMBH
  • EP3650964B1 patent drawingFigure 1

AI summary

A method for controlling or regulating a technical system is presented, in which a parameter of the control or regulation is determined by Bayesian optimization, and a next test point is determined by an acquisition function within the Bayesian optimization. The acquisition function is automatically determined by a learning algorithm within the Bayesian optimization, the parameter of the control or regulation is determined using this acquisition function, and the technical system is controlled or regulated based on the determined parameter.