Positioning Control Using Probabilistic Manipulated Variables
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing control systems that restrict upper and lower limits of manipulated variables generated by neural networks can lead to the generation of unnecessary variables, hindering the improvement of control characteristics.
Innovation Solution
A control apparatus that includes a measuring device and a controller with a compensator and converter, where the converter outputs manipulated variables based on a target probability distribution to minimize unnecessary variable generation, improving control characteristics by optimizing the probability of each manipulated variable within predetermined limits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If the upper and lower limits of the manipulated variable are simply restricted, then the control system is easy to implement, but the neural network generates unnecessary manipulated variables which prevents improvement of control characteristics
Solution Approach 1:
The patent changes the parameter representation from direct manipulated variable values to probability distributions. The neural network outputs probabilities for each possible manipulated variable within the restricted range, and a probability distribution generator converts these into actual manipulated variables. This parameter transformation allows the system to maintain simplicity while improving control characteristics by enabling probabilistic optimization.
Solution Approach 2:
The patent adds a probabilistic dimension to the control system. Instead of directly outputting manipulated variables, the neural network outputs probability distributions across multiple dimensions (different possible manipulated variable values). This dimensional expansion allows for more nuanced control decisions while maintaining the simplicity of bounded output ranges.
2Device complexity
If the neural network outputs manipulated variables directly with restricted ranges, then the device complexity is low, but waste calculations occur and control characteristics cannot be improved
Solution Approach 1:
The patent introduces a probability distribution generator as an intermediary component between the neural network and the manipulated variable output. This intermediary converts the neural network's probability outputs into actual manipulated variables, eliminating waste calculations by ensuring that only necessary variables are generated. The intermediary adds minimal complexity while significantly improving control efficiency.
3Device complexity
If manipulated variables are generated without probability optimization, then the control system is simple, but positioning accuracy and control deviations are not improved
Solution Approach 1:
The patent implements feedback through the probability distribution mechanism. The neural network learns from control deviations and adjusts the probability distributions of manipulated variables accordingly. This feedback loop continuously optimizes positioning accuracy by adjusting the likelihood of selecting manipulated variables that have proven effective in reducing control deviations, while maintaining system simplicity through the bounded probability output range.
Data Source
AI summary
A control apparatus for controlling a controlled object includes a measuring device configured to measure a state of the controlled object, and a controller configured to generate a manipulated variable corresponding to an output of the measuring device and a target value. The controller includes a compensator configured to output an index corresponding to the output of the measuring device and the target value, and a converter configured to convert the index into the manipulated variable such that a probability at which a predetermined manipulated variable is generated is a target probability.