Prediction Method of Sulfur Dioxide Emission
An accurate SO<sub>2</sub> prediction model of circulating fluidized bed (CFB) units can help operators make appropriate adjustments to unit operation. The SO<sub>2</sub> prediction accuracy of the mathematical model is limited due to the complexity of the combustion reaction...
Enregistré dans:
| Auteurs principaux: | , |
|---|---|
| Format: | article |
| Langue: | EN |
| Publié: |
IEEE
2021
|
| Sujets: | |
| Accès en ligne: | https://doaj.org/article/b56297d45b334998bd7b44510545b6e6 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| Résumé: | An accurate SO<sub>2</sub> prediction model of circulating fluidized bed (CFB) units can help operators make appropriate adjustments to unit operation. The SO<sub>2</sub> prediction accuracy of the mathematical model is limited due to the complexity of the combustion reaction in the circulating fluidized bed boiler. In order to obtain accurate predictions of SO<sub>2</sub>, a prediction model, which consists of the long short-term memory neural network (LSTM) using wide and deep structures, is proposed in this paper. Such structure improves the ability to extract linear relationships in the prediction model. The parameters of the wide structure are fixed using pre-training, which improves the prediction accuracy of the model. A differential prediction method is used for SO<sub>2</sub> prediction, which reduces the impact caused by the autocorrelation of the data. An improved mean impact value (MIV) algorithm is used to choose the best input variables combination scheme. Based on the original mean impact value algorithm (MIV), the temporal information is integrated, and repeated experiments are carried out to reduce the impact of model parameters initialization on the results. The improved MIV algorithm achieved higher prediction accuracy. The prediction model takes good prediction accuracy on the actual operating data of the 330MW CFB unit. The effectiveness of these changes is verified through comparative experiments. Compared to other existing prediction algorithms, the prediction model in this paper achieves the best prediction performance for several data sets. |
|---|