Data-driven NOx Stability Evaluation

Abstract

The stability of NOx concentration in a coal-fired unit affects the operation performance of denitrification system. By constructing the stability evaluation model, the quantitative evaluation of the big data distribution of NOx concentration was realized. The original big data cleaning method combining K-means clustering and 3σ criterion was constructed, and the data of different working conditions were divided by the sliding window method. A NOx concentration reference value model was established based on the least-squares regression method. Based on the deviation degree, the deviation degree function and the stability coefficient function were constructed, and the stability scoring function was obtained after normalization, so as to realize the quantitative evaluation of the stability of full load NOx. The model was used to evaluate NOx data from a 660 MW coal-fired unit. The results show that the overall concentration of NOx concentration distribution is high and positively correlated with load rate. The stability score of variable load condition is about 2.35% lower than that of stable condition. When the load rate is increased from 50% to 100%, the stability score is increased by 21.28%. It shows that the model can effectively evaluate the stability of NOx distribution and provide a basis for benchmarking coal-fired units.

Keywords

• coal-fired unit
• nox stability
• data cleaning
• least-squares regression
• steady state detection