摘要
提出了一种基于分子动力学模拟与云模型理论的改进混合蛙跳算法(MD-CM-SFLA).该算法将青蛙个体等效成分子,仅考虑最差个体和全局最优个体之间的吸引力,采用一种新的分子间作用力来代替两体间经典的Lennard-Jones作用力,并利用Velocity-Verlet算法和正态云发生器代替混合蛙跳算法的更新策略,有效平衡了种群的多样性和搜索的高效性.然后,将MD-CM-SFLA算法与BP神经网络相结合,设计出一种MD-CM-SFLA神经网络,并将其应用于耳语音情感识别中.耳语音情感识别结果表明,MD-CM-SFLA神经网络相对于BP神经网络具有明显的优势,在相同的测试条件下,其平均识别率较BP神经网络提高5.2%.由此表明,利用MD-CM-SFLA算法优化BP神经网络的参数,可以实现BP神经网络的快速收敛,获得较好的学习能力,从而为耳语音情感识别提供一种新思路.
A molecular dynamics simulation and cloud model theory based shuffled frog leaping al- gorithm （MD-CM-SFLA） is proposed. In this algorithm, an individual frog is equivalent to a mo- lecular and only the attractive force between the worst individual and the global best individual is considered. A new intermolecular force instead of the classic two-body Lennard-Jones force is adopt- ed, and the Velocity-Verlet algorithm and a normal cloud generator are substitued for the update strategy of the shuffled frog leaping algorithm （ SFLA）. The population diversity and the search effi- ciency are effectively balanced. Then, a MD-CM-SFLA neural network is proposed through combi- ning the MD-CM-SFLA with back propagation （BP） neural network, and it is applied to the whis- pered speech emotion recognition. The experimental results of whispered speech emotion recognition indicate that compared with BP neural network, the MD-CM-SFLA neural network has obvious ad- vantages. Under the same test conditions, the average recognition rate of the MD-CM-SFLA neural network is 5.2% higher than that of BP neural network. Therefore, utilizing MD-CM-SFLA algo- rithm to optimize the parameters of BP neural network can obtain fast convergence velocity and good learning ability, thus providing a new idea for whispered speech emotion recognition.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2012年第5期848-853,共6页
Journal of Southeast University：Natural Science Edition
基金
国家自然科学基金资助项目(60872073
60975017
51075068)
教育部博士点基金资助项目(20110092130004)
江苏省高校自然科学基金资助项目(10KJB510005)
