| With the advent of the data era,the amount of data in daily life is increasing,and the application of artificial neural networks has been greatly developed.As the most widely used model in artificial neural network,BP network is used in many fields,but it still has many shortcomings.How to optimize the convergence performance of BP network has become one of the current hotspots.People hope to find an effective optimization method to improve the convergence performance of the network by optimizing the network structure or using different optimization algorithms.Therefore,there is an adaptive learning algorithm.This paper proposes an improved self-adaptive learning algorithm,M-adam algorithm,which is applied in BP network,and has made some improvements to the problems of slow convergence speed of standard BP network and easy to fall into local minimum.The main research contents are as follows:This paper first introduces some theoretical foundations of neural networks and adaptive learning algorithms.By analyzing the application of adaptive learning Adam algorithm in BP network structure and parameter optimization,the update rules of Adam algorithm are improved,and M-adam algorithm is proposed..The M-adam algorithm changes from the original random second-order momentum to a gradually increasing second-order momentum,which ensures that the second-order momentum can reach the maximum value every time,and keeps the learning rate of the algorithm monotonically decreasing,thereby improving the later stage of the algorithm.Convergence ability and accuracy.Then,the M-adam algorithm is combined with the BP network to optimize the multi-layer feedforward topology and weight correction of the BP network.The BP network changes from a single learning rate to a learning rate that changes with the network operation.And have its own learning rate for each parameter.At the same time,the learning rate gradually decays with the network iteration,which improves the convergence efficiency of the network and solves the problem of the network falling into a local minimum.The convergence efficiency and accuracy of the network model are improved.The effectiveness of the improved algorithm is verified by comparative experiments.Finally,the BP network optimized by the M-adam adaptive learning algorithm is applied to the prediction of the domestic population.The data samples used for prediction come from the China Statistical Yearbook and are standardized.The forecast results are in line with actual population growth.At the same time,comparing the prediction results of the improved BP model with the standard BP model,the prediction accuracy is improved compared with the standard BP model. |
