The BP model based on PSO-GA hybrid algorithm optimization for the calculation of n-alcohol densities

Alcohol has become an additive and alternative to traditional fuels due to its sustainability, cleanliness, and compatibility. The experimental data of 12 n-alcohols (C1 to C12) were collected for density model study. The dataset comprises 1434 experimental data points, with temperatures ranging from 283.15K to 573.15K and pressures ranging from 0.1 MPa to 472.8 MPa. Four machine learning model frameworks (RBF, SVM, ELM, BP) are constructed and evaluated using data sets. Among them, the BP neural network model shows superior performance. To optimize the weight and bias of BP, particle swarm optimization (PSO), genetic algorithm (GA), and hybrid PSO-GA are introduced. Compared to the PSO and GA algorithms, the PSO-GA hybrid algorithm optimizes the weights and biases, and constructs the optimal network structure. The performance of the three models was compared, and the PSO-GA-BP model showed the best results. It achieved the highest R2 value of 0.9902, the lowest RMSE of 0.33, and the lowest AARD of 0.19 %. In addition, an independent data set consisting of eleven additional alcohols was used to test the generalization ability of the model. The relative deviation calculated by is mostly within 3 %, and AARD is 1.97 %. The proposed PSO-GA-BP model is compared with the classical Tait equation, and the density calculation is performed using an independent data set. The results show that the PSO-GA-BP model can give satisfactory calculation results, and its AARD is 1.58 %.