ML Controller Training Using Control-Action Time Windows
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Contemporary control devices for complex technical systems, such as gas turbines and wind turbines, often require large volumes of representative training data to optimize their operation, but insufficient data coverage can hinder successful training, especially when operating conditions are not well-represented.
Innovation Solution
A method and control device that extract a temporal sequence of training data focusing on time windows with changes in control actions, allowing for more efficient training by isolating and prioritizing data within these windows, which contain significant information about control interventions and their effects, using machine learning methods like reinforcement learning or neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large volume of training data is used for machine learning training, then the training can cover more operating conditions, but the training efficiency decreases and the training time increases
Solution Approach 1:
The patent segments the training data by identifying and extracting time windows that contain control action changes from the continuous training data stream. This segmentation approach divides the large volume of training data into meaningful, discrete segments (time windows) that are more efficient for training, thereby reducing overall training time while maintaining coverage of relevant operating conditions.
Solution Approach 2:
The patent extracts only the relevant portions of training data - specifically the time windows containing control action changes - from the complete training dataset. By taking out and using only these essential segments for training, the system achieves effective training with less data, improving training efficiency without sacrificing adaptability.
2Productivity
If training data is selectively extracted from time windows with control action changes, then training efficiency improves, but the quantity of training data decreases
Solution Approach 1:
The patent applies local quality by assigning different importance or selection criteria to different portions of the training data. Specifically, time windows containing control action changes are identified and extracted as high-quality training segments, while other portions are excluded. This selective approach ensures that the training data has high information density and relevance, improving training efficiency despite reduced overall volume.
3Reliability
If all training data is used for training, then comprehensive coverage is achieved, but redundant or less relevant data reduces training success
Solution Approach 1:
The patent extracts and removes redundant or less relevant training data by identifying and excluding time periods without control action changes. This extraction process filters out noisy or uninformative segments, leaving only the essential training data that contains meaningful control-intervention relationships, thereby improving training success and information quality.
Data Source
AI summary
A technical system controller is trained using a machine learning method. For this purpose, a chronological sequence of training data is detected for the machine learning method. The training data includes state data, which specifies states of the technical system, and control action data, which specifies control actions of the technical system. A chronological sequence of control action data is extracted specifically from the training data and is checked for a change over time. If a change over time is ascertained, a time window including the change is ascertained, and training data which can be found within the time window is extracted in a manner which is specific to the time window. The controller is then trained by the machine learning method using the extracted training data and is thus configured for controlling the technical system.