Backtracking search optimization algorithm-based least square support vector machine and its applications

Based on statistical learning theory, least square support vector machine can effectively solve the learning problem of small samples. However, the parameters of the least square support vector machine model have a great influence on its performance. At the same time, there is no clear theoretical basis for how to choose these parameters. In order to cope with the parameters optimization of the least square support vector machine, a backtracking search optimization algorithm-based least square support vector machine model is proposed. In this model, backtracking search optimization algorithm is introduced to optimize the parameters of the least square support vector machine. Meanwhile, the least square support vector machine model is updated by the prediction error combined with the sliding window strategy to solve the problem of mis-match between the prediction model and the actual sample data in the time-varying system. The performance of the proposed model is verified by classification and regression problems. The classification performance of the model is verified by five Benchmark datasets, and the regression prediction performance is verified by the dynamic liquid level of the oil production process. Compared with genetic algorithm, particle swarm optimization algorithm, and improved free search algorithm optimized least square support vector machine, the simulation results show that the proposed model has higher classification accuracy with less computation time, and higher prediction accuracy and reliability for the dynamic liquid level. The proposed model is effective.