Diesel selective catalytic reduction emission prediction based on physical model data-driven and variational autoencoder-fully connected neural network-improved Bayesian algorithm (VAE-FCNN-IBO)

In order to accurately predict NOx and NH3 concentrations downstream of the diesel engine selective catalytic reduction (SCR) system and to improve computational efficiency, this paper constructs a diesel engine SCR model and combines a data-driven approach with the design of a fully connected neura...

Celý popis

Uložené v:

Podrobná bibliografia
Vydané v:	Energy (Oxford) Ročník 337; s. 138611
Hlavní autori:	Wenlong liu, Gao, Ying, Zhu, Qi, You, Yuelin, Xia, Bocong
Médium:	Journal Article
Jazyk:	English
Vydavateľské údaje:	Elsevier Ltd 15.11.2025
Predmet:	Diesel engine Fully connected neural network Improved Bayesian algorithm Selective catalytic reduction Variational autoencoder Variational autoencoder Diesel engine Improved Bayesian algorithm Fully connected neural network Selective catalytic reduction
ISSN:	0360-5442
On-line prístup:	Získať plný text
Tagy:	Pridať tag Žiadne tagy, Buďte prvý, kto otaguje tento záznam!

Popis
Shrnutí:	In order to accurately predict NOx and NH3 concentrations downstream of the diesel engine selective catalytic reduction (SCR) system and to improve computational efficiency, this paper constructs a diesel engine SCR model and combines a data-driven approach with the design of a fully connected neural network (FCNN). Based on the inputs and outputs of the SCR model, the data were extended by a variational autoencoder (VAE) and the FCNN hyperparameters were adjusted by an improved Bayesian optimization (IBO). The data were collected at different operating conditions (280 °C, 330 °C, 380 °C and 480 °C) and the model predicted NOx concentrations with MAE below 32 ppm and NH3 concentrations with MAE below 7 ppm.Based on the full operating condition data, the NOx and NH3 concentrations downstream of the SCR were calculated and the data dimensions were expanded using the VAE and the FCNN hyperparameters were optimized by IBO to make the FCNN fully equivalent to the SCR model. Compared with the SCR model, the calculation time of the algorithm in this paper has improved by a factor of 479. Furthermore, the MAE of the algorithm in this paper is the most accurate, at 140.5412 ppm for NOx prediction and 3.5195 ppm for NH3. •The VAE algorithm expands the input data dimension of the diesel engine FCNN to improve learning.•An improved Bayesian algorithm is used to optimize the hyperparameters of the FCNN.•VAE-FCNN-IBO outperforms other algorithms in learning efficiency and generalization ability.•The trained VAE-FCNN-IBO network can replace the SCR model and speed up prediction.
ISSN:	0360-5442
DOI:	10.1016/j.energy.2025.138611