Hybrid Neural Process Modeling for Closed-Loop Dynamics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data-driven modeling techniques for industrial processes, such as those using LSTM and ResNet, struggle with high memory usage and inefficiencies in capturing dynamics, while conventional static neural networks fail to represent both transient and steady-state behaviors.
Innovation Solution
A hybrid neural network model incorporating a first-principle model with ordinary differential equations and memoryless nonlinear blocks, specifically a Hammerstein-Wiener model structure, is used to predict outputs in closed-loop industrial processes, integrating controller parameters and dynamic models for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If LSTM or ResNet networks are used to capture dynamics, then dynamic behavior can be represented, but memory usage becomes excessively high
Solution Approach 1:
The patent segments the dynamic modeling task into two distinct components: a neural network that outputs derivative values and a separate ODE solver that integrates these derivatives to produce predictions. This segmentation allows each component to be optimized independently, with the neural network using minimal memory to only compute derivatives rather than maintaining complex temporal state structures.
Solution Approach 2:
The patent replaces the mechanical memory structures inherent in LSTM (gates, cell states) and ResNet (layered transformations) with a mathematical substitution approach. Instead of using neural network architecture to inherently capture temporal dynamics, the system substitutes a simple derivative-computing network with an ODE solver that mathematically performs the temporal integration, eliminating the need for heavy memory structures.
2Device complexity
If static neural networks are used for modeling, then model simplicity is maintained, but transient and steady-state behaviors cannot be represented
Solution Approach 1:
The patent transforms the static neural network into a dynamic modeling system by having the network output time-derivative values rather than direct predictions. The ODE solver then integrates these derivatives over time to produce predictions that naturally capture both transient (time-dependent) and steady-state behaviors, while the neural network itself remains relatively simple in structure.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a method and system for modelling industrial processes, including closed loop feedback processes. The system comprises at least one sensor for measuring at least one input for the industrial process; and a processor which is configured to receive a measurement of the at least one input from the at least one sensor at an input time. The processor is also configured to implement a hybrid neural network model to output a derivative of the at least one output at the input time based on the received measurement, wherein the neural network model incorporates at least one neural network block and a first-principle block incorporating dynamic model comprising an ordinary differential equation defining the rate of change over time of the at least one output as a function of the or each associated input; input the derivative to an ordinary differential equation solver to predict the at least one output at a subsequent time; and output the prediction of at least one output at the subsequent time using the at least one measured input.