Disturbance Rejection Model for Combustion Emission Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for optimizing combustion processes in fossil fuel-fired power plants lack the ability to accurately predict changes in outputs due to input variations while estimating uncertainty, limiting their effectiveness in optimizing NOx and CO emissions.
Innovation Solution
A disturbance rejection model with a neural network is developed, which calculates a confidence metric for input-output pairings to predict future system outputs and recommends modifications, using Bayesian techniques to improve model accuracy and optimize system performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional optimization systems are used to control combustion parameters, then NOx emissions can be reduced, but the system cannot accurately predict output changes due to input variations and cannot estimate uncertainty
Solution Approach 1:
The system uses measured output values to update the neural network model through Bayesian inference, creating a feedback loop that continuously improves prediction accuracy and provides uncertainty estimates. The measured outputs are compared with predicted outputs, and the difference (error) is used to update the model parameters, ensuring the system adapts to actual system behavior while quantifying confidence in predictions.
Solution Approach 2:
The system changes the parameters of the neural network model dynamically by updating the mean and covariance of the Gaussian process based on measured data. This allows the model to adapt its predictions and uncertainty estimates in real-time, improving both accuracy and reliability without requiring a complete model redesign.
2Measurement precision
If a neural network model is used to predict system outputs, then prediction capability is improved, but the system lacks the ability to estimate uncertainty in predictions
Solution Approach 1:
The Bayesian framework acts as an intermediary between the neural network predictions and the measured outputs, providing a mathematically rigorous way to quantify uncertainty. The Gaussian process prior over the neural network weights allows the system to propagate uncertainty from the model through to the predictions, preserving uncertainty information that would otherwise be lost in deterministic predictions.
3Object-generated harmful factors
If combustion optimization is performed to reduce emissions, then environmental performance is improved, but system performance and efficiency may be compromised
Solution Approach 1:
The system dynamically adjusts combustion parameters based on real-time predictions and measured outputs, allowing the system to optimize for emissions reduction while adapting to changing operating conditions that affect performance. The Bayesian update mechanism enables continuous adaptation, finding the optimal balance between emissions and performance as conditions evolve.
Data Source
AI summary
A method for modeling a system that includes a disturbance rejection model configured for modeling an operation of the system so to generate a predicted value for a system output. The disturbance rejection model having a network for mapping system inputs to the system output, and input-output pairings, each representing a unique pairing of one of the system inputs with the system output. The method may include the steps of: calculating a confidence metric for a selected input-output pairing of the disturbance rejection model; and recommending a modification be made to the disturbance rejection model based upon the confidence metric calculated for the selected one of the input-output pairing. The confidence metric may indicate a probability that a predicted sign of a gain in the system output made by the disturbance rejection model is correct when the system input of the selected input-output pairing is varied.