Plant Control Parameter Search Range for Faster Auto-Tuning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing control parameter optimization methods for plant operations are inefficient, requiring excessive calculation time due to large search ranges, especially as the number of control parameters increases, and often involve lengthy simulations that can be computationally intensive.

Innovation Solution

A control parameter automatic-adjustment apparatus and method that employs a learning unit, simulator, and knowledge database to optimize control parameters by reducing the search range through reinforcement learning, genetic algorithms, and Bayesian learning, while utilizing a search range determination unit to focus on likely optimal parameters, thereby shortening calculation time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the search range for control parameter optimization is large to ensure comprehensive coverage, then the optimization completeness is improved, but the calculation time increases exponentially

Engineering Contradiction:
Improveoptimization completenessVSAvoidcalculation time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent applies preliminary action by using genetic algorithms to pre-determine the search range before reinforcement learning optimization. This preliminary step narrows down the extensive search range to a more manageable scope, reducing the calculation time for the subsequent reinforcement learning phase while ensuring that the optimal parameters are not missed.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The optimization process is segmented into two distinct phases: first, genetic algorithms are used to determine the search range, and second, reinforcement learning is applied to find the optimal parameters within that range. This segmentation allows each algorithm to focus on its strength, reducing overall computation time while maintaining optimization completeness.

Inventive Principle:
Principle #1Segmentation

2Manufacturing precision

If the number of control parameter divisions is increased to improve precision, then the manufacturing precision is improved, but the calculation time increases

Engineering Contradiction:
Improveparameter setting precisionVSAvoidcalculation time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The genetic algorithm performs preliminary action by establishing the search range that accommodates the required precision divisions. By pre-defining this range based on the number of parameter divisions needed, the system reduces the computational burden during the reinforcement learning phase, allowing high precision without exponential time increases.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent changes parameters dynamically by adjusting the search range based on the number of control parameter divisions. This allows the system to maintain high precision requirements while adapting the search space to minimize calculation time, effectively decoupling precision from computational cost.

Inventive Principle:
Principle #35Parameter changes

3Loss of time

If reinforcement learning is used to reduce the search range, then the calculation time is reduced, but the simulator calculation load increases

Engineering Contradiction:
Improveoptimization timeVSAvoidsimulator calculation load
Core Design Contradiction:
Loss of timeVSUse of energy by moving object

Solution Approach 1:

The calculation process is segmented into two phases with different computational characteristics. The genetic algorithm phase uses the simulator to establish the search range, while the reinforcement learning phase operates within this reduced range. This segmentation distributes the simulator calculation load, preventing any single phase from becoming excessively demanding while achieving overall time reduction.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The genetic algorithm performs preliminary action by determining the search range before reinforcement learning begins. This preliminary step reduces the subsequent reinforcement learning calculation load by limiting it to a focused search space, thereby reducing overall optimization time without creating excessive simulator burden in any single phase.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3214509B1Control parameter automatic-adjustment apparatus, control parameter automatic-adjustment method, and control parameter automatic-adjustment apparatus network
Publication Date: 2022.05.18 MITSUBISHI POWER LTD
  • EP3214509B1 patent drawingFigure 1
  • EP3214509B1 patent drawingFigure 2
  • EP3214509B1 patent drawingFigure 3

AI summary

Provided is a control parameter automatic-adjustment apparatus 400 that adjusts a control parameter 240 which is used when a control apparatus 200 of a plant 100 calculates an operational control signal 230, the apparatus 400 including: a simulator 600 that simulates operation of the plant 100; a learning unit 500 that searches for an optimal control parameter 532 using the simulator 600; and a knowledge database 700 that stores knowledge information 710 which associates an amount of change in the control parameter 541 with an amount of change in a state of the plant 100, in which the learning unit 500 includes a search range determination unit 550 that determines a control parameter search range based on the knowledge information 710 that is stored in the knowledge database 700.