Recursive Neural Network–Particle Swarm Versus Nonlinear Multivariate Rational Function Algorithms for Optimization of Biodiesel Derived from Hevea brasiliensis

• Published in: Arabian journal for science and engineering (2011) Vol. 48; no. 12; pp. 15979 - 15998
• Main Authors: Esonye, Chizoo, Onukwuli, Okechukwu Donminic, Ubaka, Okolie Charles, Etim, Okon Anietie, Ume, Cyril Sunday, Agu, Chinedu Mathew
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.12.2023; Springer Nature B.V
• Subjects: Algorithms; Biodiesel fuels; Catalysts; Engineering; Flame ionization detectors; Fourier transforms; Fuel production; Functional groups; Gas chromatography; Guidance (motion); Heuristic; Heuristic methods; Humanities and Social Sciences; Independent variables; Materials testing; Methanol; multidisciplinary; Multivariate analysis; Neural networks; Optimization; Particle swarm optimization; Rational functions; Recurrent neural networks; Research Article-Chemical Engineering; Science; Sensitivity analysis; Multivariate rational functions; Recursive neural networks; Particle swarm optimization; Biodiesel
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-023-07947-x

This research reports the application of recursive neural network–particle swarm (RNN–PS) and nonlinear multivariate rational function (NLMRF) algorithms for optimization of biodiesel derived from non-edible feedstock of Hevea brasiliensis. Nonlinear auto-regressive with external input (NARX) model was applied with four input variables: time (45–65 min), process temperature (45–65 °C), methanol/oil molar proportion (4:1–12:1) and catalyst dosage (0.5–1.5 wt%), one target (biodiesel yield) and two delays. The RNN showed better correlation (SSE = 21.14 and R 2  = 0.98) than the three NLMRFs (SSE > 300 and R 2  < 0.63). Optimum conditions obtained with RNN–PS hybrid heuristic model were 60.55 min, 70 °C, methanol/oil molar ratio of 6.89 and catalyst concentration of 1.5wt% with a maximum biodiesel yield of 92.77wt% and experimental validation of 92.52 wt%. Sensitivity analysis result shows that the level of the impact of the input variables on the independent responses follows the order: catalyst concentration (43.77%) > reaction time (24.22%) > methanol/HBSO molar ratio (21.89%) > reaction temperature (10.12%). RNN–PS exhibited high capability in excellent capturing of mapping, great diversity trajectory search, rapid convergence and intrinsic guidance strategy. Functional groups, fatty acid compositions and physicochemical characteristics of the biodiesel obtained using Fourier transform infrared (FTIR), gas chromatography flame ionization detector (GC-FID) and American Standard for Material Testing (ASTMD) methods show that the quality of the biodiesel agreed with international standards. RNN–PSO is therefore proposed as a meta-heuristic and reliable optimization tool for developing a viable and sustainable route for biodiesel fuel production.