RBF Neural Network Initialization for Stable Incremental Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing closed-loop control systems, particularly those using PID controllers, face challenges in tuning and stability as load conditions change, and neural networks are difficult to optimize in real-time operating conditions, leading to unpredictable system behavior.
Innovation Solution
The method involves initializing a Radial Basis Function (RBF) neural network using maximum error, maximum first-order change in error, and maximum second-order change in error to calculate incremental changes in process variables, with an actor-critic reinforced learning model for progressive weight tuning, allowing for deterministic initialization and stable control from the outset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional PID controllers are used for closed-loop control, then the control system is simple to implement, but the system becomes difficult to tune and unstable when load conditions change
Solution Approach 1:
The patent transforms the fixed parameters of traditional PID controllers into adaptive parameters by using an RBF neural network. The network dynamically adjusts control parameters based on real-time error, first-order change in error, and second-order change in error, allowing the system to adapt to changing load conditions while maintaining stability.
Solution Approach 2:
The patent applies preliminary action by pre-initializing the RBF neural network with deterministic values derived from maximum error, maximum first-order change in error, and maximum second-order change in error. This pre-initialization ensures the network starts with reasonable parameter values, eliminating the need for extensive training periods and providing immediate stable control.
2Reliability
If neural networks are used for real-time control optimization, then control performance can be improved, but the system behavior becomes unpredictable and difficult to optimize
Solution Approach 1:
The patent eliminates the unpredictability and optimization difficulty of neural networks by applying preliminary deterministic initialization. The RBF network parameters are pre-calculated using maximum error, maximum first-order change in error, and maximum second-order change in error, ensuring predictable and stable behavior from the outset without requiring extensive training or tuning.
Solution Approach 2:
The patent incorporates feedback mechanisms by using error, first-order change in error, and second-order change in error as inputs to the RBF neural network. This feedback loop allows the system to continuously adjust control parameters based on actual system performance, maintaining predictable and optimized behavior in real-time operations.
3Reliability
If RBF neural network is initialized using maximum error values, then deterministic initialization and stable control are achieved, but the system requires additional parameters (maximum first-order change in error, maximum second-order change in error, maximum output increment)
Solution Approach 1:
The patent manages the complexity of multiple initialization parameters by applying preliminary action - all maximum values (error, first-order change in error, second-order change in error, output increment) are determined during system setup or first use. This pre-determination ensures deterministic initialization and stable control while the parameters remain fixed during normal operation, reducing ongoing complexity.
Solution Approach 2:
The patent achieves universality by using the same set of maximum parameters (error, first-order change in error, second-order change in error, output increment) for multiple purposes: initializing the RBF network, scaling inputs, and ensuring stable control across varying load conditions. This multi-functional use of parameters reduces overall system complexity despite the increased number of initialization values required.
Data Source
AI summary
According to one or more embodiments of the present invention, a computer-implemented method for adjusting a process variable using a closed loop system includes initializing a radial basis function neural network (RBF network) using a maximum error (emax), a maximum first order change in error (Δemax), a maximum second order change in error (Δ2emax), and a maximum output increment (Δomax), associated with the closed loop system being controlled. The method further includes inputting, to the RBF network, input values including an error, a first order change in error, and a second order change in error. The method further includes computing, by the RBF network, control parameters based on the input values, and computing, by the RBF network, an incremental change in the process variable based on the control parameters. The method further includes adjusting, by a controller, an output device to change the process variable by the incremental change.