A hybrid switching predictive controller with proportional integral derivative gains and GMDH neural representation of automotive engines for coldstart emission reductions. (February 2016)

• Record Type: Journal Article
• Authors: Mozaffari, Ahmad; Azad, Nasser L.; Hedrick, J. Karl; Taghavipour, Amir
• Is Part Of: Engineering applications of artificial intelligence. Volume 48(2015:Dec.)
• Page Start: 72
• Page End: 94
• Publication Date: 2016-02
• Subjects: Coldstart operation -- Hybrid switching predictive controller -- GMDH -- Optimal control; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2015.10.013
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: In this research, a high-performance predictive controller is developed for automotive coldstart emission reductions. The proposed control scheme combines a hybrid switching predictive controller (HSPC) with proportional integral derivative (PID) gains to simultaneously minimize the cumulative hydrocarbon emissions ( HC cum ) and the control input variations for a given engine during the coldstart operation. It is essential to use a sufficiently accurate surrogate meta-representation of the real engine within this model-based controller to predict the states of the plant and impart proper control commands to the system. The existing studies in the research literature have clearly demonstrated that automotive engines have a highly transient nonlinear behavior during coldstart periods and different disturbances can affect their operations. To cope with the mentioned difficulties, several coldstart experiments are performed to capture a comprehensive database for the considered engine. Thereafter, a powerful knowledge-based black-box meta-modeling tool, known as group method data handling (GMDH), is adopted to have a neural representation of the engine׳s coldstart behavior. As a real-time controller, the proposed PID-based HSPC requires a fast and robust solver to calculate the gains of PID in a computationally efficient manner. Here, a multivariate quadratic fit-sectioning algorithm (MQFSA) is proposed to deterministically determine the control commands. Other than the considered online optimizer, a powerful chaos-enhanced evolutionary algorithm (CEA) is used to heuristically optimize the prediction horizon ( H P ) and control commands horizon ( H U ) to achieve the best results. It is demonstrated that using such an optimizer, instead of trial-and-errors, to heuristically set the control and plant prediction horizon lengths is an effective strategy. Finally, several comparative studies are conducted to further indicate the efficacy of the proposed PID-based HSPC for the automotive coldstart control problem.