About the project
Continuous growth and increase of variance in electricity
consumption can lead to frequent load changes. This calls
for novel control concepts in order to minimize emissions
and to sustain high efficiency during load changes.
From combustion point of view the main challenges for the
existing boilers are caused by a wide fuel selection, increasing
share of low quality and bio fuels, and co-combustion.
In steady operation, combustion is affected by the disturbances
in the feed rate of the fuel and by the incomplete
mixing of the fuel in the bed. It may cause changes in the
burning rate, oxygen level and increase CO2 emissions. This
is especially relevant for the new biomass fuels, which have
increasingly been used to replace coal. The bio-fuels are
rather inhomogeneous and very reactive in comparison with
Traditionally, mathematical models of CFB boiler operation, have been developed, incorporating operational parameters in the models. More recently, data mining approaches were considered for developing better understanding of the underlying processes in CFB boilers, or learning a model to optimize its efficiency.
In this project we are working on a few directions. One of the tiny but important problem we have been focusing in the recent past is a data driven approach for online mass flow estimation. Accurate and reliable (timely) estimates of the mass flow are required for efficient control of the boiler. Online estimation of fuel consumption in mechanical devices is a challenging task due to noise, presence of outliers and non-stationarity of the signal. Mechanical devices typically comprise of a number of moving parts. The movements of these parts cause interference in the observed sensor signal. The challenge is to filter out the true signal from the measured noise. In this study we investigate the online estimation of the fuel mass inside a CFB boiler. The main results of our work in this direction have been summarized in the DS'09, ICDM'09, SensorKDD'09, and SIGKDD Explorations publications.
Another change detection problem relates to the pressure fluctuation signal. See our SensorKDD'12 publication, in which we propose Quantile index method to detect both sudden and gradual changes in the online settings, Other directions include data fusion from sensor data, finding and quantifying relationships between input and output parameters of the boiler, and applying discovered knowledge in the control system of the boiler and thus validating our work.
- Alexandr Maslov, Mykola Pechenizkiy, Tommi Kärkkäinen, Matti Tähtinen. Quantile Index for Gradual and Abrupt Change Detection from CFB Boiler Sensor Data in Online Settings, Proceedings of the Sixth International Workshop on Knowledge Discovery from Sensor Data, Beijing, 2012, [PDF]
- Mykola Pechenizkiy, Jorn Bakker, Indre Zliobaite, Andriy Ivannikov, Tommi Kärkkäinen. 2009. Online Mass Flow Prediction in CFB Boilers with Explicit Detection of Sudden Concept Drift, SIGKDD Explorations, 11(2), pp. 109-116. [PDF]
- Indre Zliobaite, Jorn Bakker, Mykola Pechenizkiy. 2009. OMFP: An Approach for Online Mass Flow Prediction in CFB Boilers. Discovery Science 2009: 272-286. [PDF] [BIB]
- Andriy Ivannikov, Mykola Pechenizkiy, Jorn Bakker, Timo Leino, Mikko Jegoroff, Tommi Kärkkäinen, Sami Äyrämö. 2009. Online Mass Flow Prediction in CFB Boilers. ICDM 2009: 206-219. [PDF] [BIB]
- Jorn Bakker, Mykola Pechenizkiy, Indre Zliobaite, Andriy Ivannikov, Tommi Kärkkäinen. 2009. Handling outliers and concept drift in online mass flow prediction in CFB boilers. KDD Workshop on Knowledge Discovery from Sensor Data 2009: 13-22 (Best paper award) [PDF] [BIB]
- Pechenizkiy M., Tourunen A., Kärkkäinen T., Ivannikov A., Nevalainen H. 2006.Towards Better Understanding of Circulating Fluidized Bed Boilers: a Data Mining Approach. In: Proc. of 2nd Workshop on Practical Data Mining: Applications, Experiences and ChallengesDMBiz'06 (ECML/PKDD'06), Berlin, Germany, pp. 80-83. [PDF] [BIB]
Presentations & Posters
- Quantile Index for Gradual and Abrupt Change Detection from CFB Boiler Sensor Data in Online Settings. KDD Workshop on Knowledge Discovery from Sensor Data 2012, Beijing, China.
- Handling outliers and concept drift in online mass flow prediction in CFB boilers. KDD Workshop on Knowledge Discovery from Sensor Data 2009, Paris, France. (Best paper award)
- Towards Better Understanding of Circulating Fluidized Bed Boilers: a Data Mining Approach. Poster presented at 2nd Workshop on Practical Data Mining: Applications, Experiences and Challenges (DMBiz'06) @ ECML/PKDD 2006 Conference, Berlin, Germany.
Code & Datasets
We are working on making the source code (in Matlab) and time series datasets used in this project available for the research community. Please visit this page later.