Research on prediction of wax deposition rate based on improved Support Vector Machines model

Accurately predict the wax deposition rate can ensure the safe operation of waxy crude oil pipelines. This paper proposes a combination model for wax deposition rate prediction based on Least Absolute Shrinkage and Selection Operator (LASSO), Improved Simulated Annealing Particle Swarm Optimization (ISAPSO) and Support Vector Machines (SVM). Taking 200 sets of operating data in Jinghui Pipeline as the research samples, which are simplified by using LASSO algorithm to extract key factors affected the wax deposition rate. Adopting ISAPSO which is obtained by optimizing the particle initialization, inertia weight and learning factors of SAPSO algorithm to optimize the input weight and the spread of SVM. Using MATLAB software to build the prediction model and make the prediction, and the prediction results of LASSO-ISAPSO-SVM model are compared with those of Back Propagation Neural Network (BPNN) model and LASSO-ISAPSO-RBF model. The results are as follows: the key factors affected the wax deposition rate are extracted from 8 to 5 by LASSO; ISAPSO converges 54 iterations earlier than SAPSO and has higher optimization accuracy; The predicted value of LASSO-ISAPSO-SVM model is closer to the actual value compared with BPNN model and LASSO-ISAPSO-RBF model, with the average absolute error within
$+0.25$
and the average relative error is less than 8.1%.