Neural Network Process Controller Training Without Expert Programming
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Programming a controller for industrial processes is a long and expensive task due to the complexity of connections between manipulated and disturbance variables and controlled variables, requiring substantial human expert involvement.
Innovation Solution
A method involving the training of neural networks to predict and control industrial processes using off-line historical data, where a first neural network serves as a predictor and a second neural network is trained as a controller, allowing for offline data usage without relying on existing controller policies, and employing reward functions to adjust the controller based on target variables.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional controller programming methods are used, then controller functionality is achieved, but the task becomes long and expensive requiring substantial human expert involvement
Solution Approach 1:
The system performs preliminary action by collecting and storing historical process data (controlled variables, manipulated variables, disturbance variables) in advance. This pre-collected data is then used to train neural networks offline, eliminating the need for time-consuming manual programming when the controller is actually needed. The neural networks are trained using historical data before deployment, allowing rapid controller generation.
Solution Approach 2:
The patent replaces traditional manual controller programming (mechanical/expert system) with automated neural network-based control. Instead of human experts manually programming controller logic, the system uses neural networks trained on historical data to automatically generate control policies, substituting human expertise with an automated learning system.
2Ease of manufacture
If traditional controller programming methods are used, then controller functionality is achieved, but costs increase due to substantial human expert involvement
Solution Approach 1:
The patent replaces expensive human expert involvement with automated neural network training. The system uses historical data to train neural networks that automatically learn optimal control strategies, eliminating the need for costly human expert programming while maintaining or improving controller performance.
Solution Approach 2:
The system enables self-service by allowing the controller to learn optimal control policies autonomously from historical data. The neural networks automatically adjust their parameters and structures based on the historical process data, without requiring continuous human intervention or expertise, thereby reducing development and maintenance costs.
3Extent of automation
If neural networks are trained using historical data without existing controller policies, then automation increases, but training accuracy must be maintained without policy guidance
Solution Approach 1:
The system implements feedback by using the predictor neural network to generate predictions of controlled variables, which are then compared against actual historical values. The error between predictions and actual values is used to adjust and refine the predictor and controller neural networks, ensuring high accuracy without requiring policy guidance during training.
Solution Approach 2:
The system performs preliminary action by pre-training the predictor neural network using historical data before training the controller. This preliminary training establishes an accurate baseline model of the process, which then guides the controller training process, ensuring that the automated training maintains high prediction accuracy.
Data Source
AI summary
A method of generating a controller for a continuous process. The method includes receiving from a storage memory, off-line stored values of one or more controlled variables and one or more manipulated variables of the continuous process over a plurality of time points. The off-line stored values are used to train a first neural network to operate as a predictor of the controlled variables. Then, the method includes training a second neural network to operate as a controller of the continuous process using the first neural network after it was trained to operate as the predictor for the continuous process and employing the second neural network as a controller of the continuous process.