Meta-Control Combustion Efficiency with a Data-Mining Approach

Grant# 04-06
Principal Investigator: Andrew Kusiak
Organization: The University of Iowa
Technical Area: Energy Efficiency

Control of the combustion process of a power plant boiler involves many variables that impact its efficiency. Some of these variables are measured, while the values of others are unknown often due to lack of sensors. Determining robust and optimal control settings is a key factor to boiler efficiency optimization.

In this research, a data-mining approach is proposed to extract knowledge and models from a wide range of quantitative and qualitative data. Data mining is a natural extension of more traditional tools such as neural networks, multivariable control algorithms, or even traditional analog controllers. The computer extracted knowledge will be applied to improve combustion efficiency. The parameter settings leading to optimal efficiency in the form of control signatures that will be derived by the data-mining approach proposed in this project. The proposed approach is applicable to boilers that are fired with coal, biomass, and possibly other types of fuels. The effectiveness of the models developed in this project will be demonstrated at Boiler 11 located at The University of Iowa Power Plant.

The basic project deliverable is the demonstration of models and control signatures created to optimize combustion efficiency at The University of Iowa Power Plant. The proposed approach optimizes coal burners for various criteria, e.g., efficiency of the combustion process and emissions, and it considers tradeoffs among various criteria. This computational tool optimizes combustion efficiency by modifying combustion control settings for a spectrum of control systems ranging from analog to digital.

Some of the data-mining algorithms to be considered in the project have been developed in the Intelligent Systems Laboratory at The University of Iowa and tested with a great success at the coal-fired boiler at the UI Power Plant.

The main outcome of this project will be improved combustion efficiency by at least 1 percent over the project duration. Additional benefits will be obtained by increasing boiler availability.

Work To Date: (Technical Report – January 2007)
An online boiler performance optimization framework based on an online clustering technique was presented. The recommendation system, DACOMO (Data Mining Combustion Optimization), which was built based on the framework, was successfully applied to an industrial boiler, and significant savings were observed using the data from a live demonstration. The confidence index, which was developed based on the concept of a virtual age, was used to help the operators determine when to follow the system recommendations.

The boiler used in the experiment was relatively small (20MW). The savings from the proposed methodology will multiply when the system is deployed for large commercial boilers. The system in its current implementation computes a recommendation every minute. The operators have implanted the recommendations every several minutes.

A demo can be viewed at