摘要
通过对微粒群PSO优化算法惯性因子和加速因子的动态调整,保证PSO算法迭代过程中全局与局部寻优能力的动态平衡,构造了一种更加稳定准确的动态微粒群DPSO优化算法。进而将动态微粒群DPSO优化算法与传统BP神经网络相结合,分别采用动态微粒群DPSO优化算法和自适应BP算法对神经网络权值进行全局优化和局部二次优化,建立基于动态微粒群优化算法的神经网络DPSO-NN预测模型。利用所建立的动态微粒群神经网络模型对渤海某海域年极值冰厚进行训练预测,并将训练预测结果与BP-NN、GA-NN、PSO-NN模型的训练预测结果、以及实际数据进行对比分析,验证DPSO-NN预测模型具有更优的训练稳定性和预测准确性,为冰区海洋平台安全评估提供了更为可靠的环境载荷参量。
Through adjusting inertia factor and accelerating factors dynamically, the dynamic particle swarm optimization (DPSO) algorithm was designed, so as to keep the dynamic balance between the global and local optimization. Combining DPSO algorithm and the BP neural network, the paper established the dynamic particle swarm optimization neural network (DPSO -NN) model to optimize the network weights for global optimization and locally quadratic optimization respectively. The ice thickness data of Bohai sea was utilized for DPSO - NN model training and predicting. The training and predicting results of the DPSO - NN model were contrasted with the results of BP - NN, GA - NN and PSO - NN models, so as to verify the stability and accuracy of DPSO - NN model. The result can provide more credible environment toad parameter for the safety evaluation of ocean platform.
出处
《水资源与水工程学报》
2013年第5期23-27,32,共6页
Journal of Water Resources and Water Engineering
基金
国家自然科学基金项目(51209218)
中央高校基本科研业务费专项资金项目(12CX04069A)
关键词
动态微粒群算法
神经网络
自适应BP算法
全局优化
海冰厚度
dynamic particle swarm optimization (DPSO) algorithm
neural network
self adaptive BPalgorithm
global optimization
thickness of sea ice