Multi-objective optimization of dual-fuel engine performance in PPCI mode based on preference decision

• Published in: Fuel (Guildford) Vol. 312; p. 122901
• Main Authors: Ma, Cheng, Song, En-Zhe, Yao, Chong, Long, Yun, Ding, Shun-Liang, Xu, Duo, Liu, Zhao-Lu
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 15.03.2022; Elsevier BV
• Subjects: Compression; Decision making; Dual fuel; Dual-fuel engine; Emissions; Energy efficiency; Euclidean geometry; Fuel consumption; Fuel economy; Injection; Long short-term memory; LSTM; Multi-objective optimization; Multiple objective analysis; Neural networks; Nitrogen oxides; Optimization; Parameters; PPCI; Preference; Multi-objective optimization; LSTM; PPCI; Dual-fuel engine; Preference
• ISSN: 0016-2361; 1873-7153
• DOI: 10.1016/j.fuel.2021.122901

•LSTM-NSGA-II strategies with preferences are beneficial for engine optimization.•The preference strategy is used to solve the dual-fuel engine optimization problem.•Achieved a trade-off between economy and emissions for PPCI engines.•Interactive decision making is an effective way to achieve preference optimization. In order to obtain the best emission performance and fuel economy of the dual-fuel engine in PPCI (partial premixed compression ignition) mode while responding to different decision requirements. LSTM (long short-term memory) neural network is used to build a response model that can accurately reflect the relationship between input parameters and output performance of the dual-fuel engine. A clever combination of LSTM and NSGA-II is used to construct a multi-objective performance optimization framework for the dual-fuel engine. The control parameters including MIT (main injection timing), PIT (pre-injection timing), PIQ (pre-injection quantity), EAC (excess air coefficient), SRL (substitution rate limit), and IP (injection pressure) are integrated and optimized under each load condition according to the propulsion conditions. The introduction of decision-maker (DM) preference information is creative, which allows the optimization process to include the intention of the DM, which can guide the evolutionary direction of the population. An organized combination of reference point coordinates R, weight vector w of weighted Euclidean distance, and preference strength δ is the critical one. The interactive optimization process effectively reduces the DM’s reliance on empirical knowledge in preference parameter setting. The experimental results show that all solutions at preference low emission meet the IMO Tier-III limit for NOx, and the average NOx emission at full operating conditions is 1.22 g/(kW·h), which is 78.9% lower than the original engine, along with a 4.76% reduction in BSFC compared to the original engine. When low BSFC is preferred, the average BSFC is 8.44% lower than the original engine and 3.68% lower than when low emissions are preferred. It can be found that the lower BSFC is obtained at the expense of the environment, and the pursuit of higher fuel economy worsens emissions dramatically.