Optimization Efficiency of Swarm Intelligence Search in Base Oil Performance Prediction Model

Changes in the composition of lubricant base oils have important effects on their performance. While traditional lubricant development needs to be carried out by repeatedly testing and profiling a large number of oil samples for performance, machine learning analysis and processing techniques for data can not only improve efficiency, save costs and eliminate human factor interference, but also make it possible to develop corresponding oil products in a precise and adjustable manner. This paper investigated the effects of different ratios of three component contents of mineral oil (KN4010), polyalpha-olefin synthetic oil (PAO40) and polyol ester (PriEco 3000) in the composite base oil system on the changes of lubricant kinematic viscosity, viscosity index and rotational oxygen bomb performance index. Firstly, a least squares support vector machine (LSSVM) based base performance prediction model was developed and compared with four algorithms, namely, random forest (RF), back propagation neural network (BPNN), extreme learning machine (ELM) and multiple linear regression (MLR), to examine the predictive ability of the base model on three physicochemical performance indicators, namely, kinematic viscosity, viscosity index and rotational oxygen bomb, for model selection. Then the classical Particle Swarm Optimization (PSO) algorithm and the emerging Dragonfly Algorithm (DA) and Whale Optimization Algorithm (WOA) in recent years are used to construct the hybrid prediction model, and the parameters of the preferred LSSVM base prediction model are optimized, i.e., the kernel function width (σ²) and the regularization parameter (γ) are selected optimally. And examine the effects of different parameter optimization methods on the convergence speed, stability and prediction accuracy of the model under different physical and chemical properties with examples. Finally, the classical, base and hybrid models were evaluated and analyzed by comparing the prediction results of the models on the three performance metrics and the leave-one-out cross-validation (LOOCV) test, and their prediction ability in the case of limited data samples was examined, especially the expression of model applicability for sample generalization performance. The results showed that machine learning techniques had good predictive capability for oil performance, and the LSSVM base model could obtain relatively good prediction results under small sample conditions, while WOA-LSSVM could significantly reduce the prediction error of the model, with a mean absolute percentage error of 0.95%, a root mean square error of 1.16 and a coefficient of determination R² of 0.998 9 for kinematic viscosity, for example. And WOA-LSSVM enabled the model to search for better results, and its convergence accuracy and stability were better than the two optimization models, PSO-LSSVM and DA-LSSVM, and it was less likely to fall into local optimal solutions and converges faster. Through model testing and analysis by the leave-one-out cross-validation method, the prediction results of WOA-LSSVM were significantly better than the rest of the models, with excellent performance in all aspects, good prediction accuracy and generalization ability, and suitable for studying the problem of lubricant physicochemical performance prediction in small sample situations. The results of the above study provided an effective way for future information studies on the correlation between the composition components and performance indexes of lubricant products.