Hybrid algorithm based on Levenberg-Marquardt Bayesian Regularization Algorithm and Genetic Algorithm

In order to overcome the insufficiencies of the convergence of the low speed, a low precision of the forecast and easy convergence to a local minimum point of error function on BP Neural Networks (BPNN), a new hybrid algorithm-LMBRGA, which uses both the Levenberg-Marquardt(LM) Bayesian Regularization Algorithm(LMBRA) and Genetic Algorithm(GA) to optimize BPNN, is proposed. The specific process was as follows. Firstly, the GA optimized the best weights and thresholds as the training initial values of BPNN. Then, the BPNN after initialization was trained by the LMBRA until the network has converged. Finally, the network model, which met the requirements after being examined by the test samples, was applied to predict the resident consumption level of Chengdu. By Simulation Experiments analysis, the LMBRGA hybrid algorithm has faster convergence rate than the LMBRA. From the average relative forecasting error (ARFE)´s comparison of the predictive results, it clearly indicates that the forecast precision of the LMBRGA hybrid algorithm is higher than another five optimization algorithms.