Moth Flame Optimization(MFO)is a nature-inspired optimization algorithm,based on the principle of navigation technique of moth toward moon.Due to less parameter and easy implementation,MFO is used in various field to ...Moth Flame Optimization(MFO)is a nature-inspired optimization algorithm,based on the principle of navigation technique of moth toward moon.Due to less parameter and easy implementation,MFO is used in various field to solve optimization problems.Further,for the complex higher dimensional problems,MFO is unable to make a good trade-off between global and local search.To overcome these drawbacks of MFO,in this work,an enhanced MFO,namely WF-MFO,is introduced to solve higher dimensional optimization problems.For a more optimal balance between global and local search,the original MFO’s exploration ability is improved by an exploration operator,namely,Weibull flight distribution.In addition,the local optimal solutions have been avoided and the convergence speed has been increased using a Fibonacci search process-based technique that improves the quality of the solutions found.Twenty-nine benchmark functions of varying complexity with 1000 and 2000 dimensions have been utilized to verify the projected WF-MFO.Numerous popular algorithms and MFO versions have been compared to the achieved results.In addition,the robustness of the proposed WF-MFO method has been evaluated using the Friedman rank test,the Wilcoxon rank test,and convergence analysis.Compared to other methods,the proposed WF-MFO algorithm provides higher quality solutions and converges more quickly,as shown by the experiments.Furthermore,the proposed WF-MFO has been used to the solution of two engineering design issues,with striking success.The improved performance of the proposed WF-MFO algorithm for addressing larger dimensional optimization problems is guaranteed by analyses of numerical data,statistical tests,and convergence performance.展开更多
针对当前网络入侵检测中的数据量较大、数据维度较高的特点,将飞蛾扑火优化(MFO)算法应用于网络入侵检测的特征选择中。鉴于MFO算法收敛过快、易陷入局部最优的问题,提出一种融合粒子群优化(PSO)的二进制飞蛾扑火优化(BPMFO)算法。该算...针对当前网络入侵检测中的数据量较大、数据维度较高的特点,将飞蛾扑火优化(MFO)算法应用于网络入侵检测的特征选择中。鉴于MFO算法收敛过快、易陷入局部最优的问题,提出一种融合粒子群优化(PSO)的二进制飞蛾扑火优化(BPMFO)算法。该算法引入MFO螺旋飞行公式,具有较强的局部搜索能力;结合了粒子群优化(PSO)算法的速度更新方法,让种群个体随着全局最优解和历史最优解的方向移动,增强算法的全局收敛性,从而避免易陷入局部最优。仿真实验以KDD CUP 99数据集为实验基础,分别采用支持向量机(SVM)、K最近邻(KNN)算法和朴素贝叶斯(NBC)3种分类器,与二进制飞蛾扑火优化(BMFO)算法、二进制粒子群优化(BPSO)算法、二进制遗传算法(BGA)、二进制灰狼优化(BGWO)算法和二进制布谷鸟搜索(BCS)算法进行了实验对比。实验结果表明,BPMFO算法应用于网络入侵检测的特征选择时,在算法精度、运行效率、稳定性、收敛速度以及跳出局部最优的综合性能上具有明显优势。展开更多
及时准确识别母猪的发情行为可以有效增加受胎率和产仔量,对提高养殖企业的繁育水平和经济效益具有重要意义。该研究针对生猪养殖过程中母猪发情行为识别存在主观性强、智能化水平低、假警报和错误率高、识别不及时等问题,提出了一种基...及时准确识别母猪的发情行为可以有效增加受胎率和产仔量,对提高养殖企业的繁育水平和经济效益具有重要意义。该研究针对生猪养殖过程中母猪发情行为识别存在主观性强、智能化水平低、假警报和错误率高、识别不及时等问题,提出了一种基于飞蛾扑火算法(Moth-Flame Optimization,MFO)优化长短时记忆网络(Long Short Term Memory,LSTM)的母猪发情行为识别方法。利用安装在母猪颈部的姿态传感器获得母猪姿态数据,然后使用姿态数据训练MFO-LSTM姿态分类模型,将母猪姿态分为立姿、卧姿和爬跨3类。通过对姿态分类结果进行分析,确定以爬跨行为和活动量2个特征作为发情行为识别依据,使用MFO-LSTM分类算法判断母猪是否发情。以山西省太原市杏花岭区五丰养殖场的试验数据对该方法进行验证,结果表明,该方法在以30 min为发情行为识别时间时的识别效果最好,发情行为识别的错误率为13.43%,召回率为90.63%,特效性为81.63%,与已有的母猪发情行为识别方法相比错误率降低了80%以上。该方法在保证识别准确率的情况下有效降低了错误率,可满足母猪养殖生产过程中发情行为自动识别要求。展开更多
文摘针对无人机长期跟踪过程中尺度变换导致目标丢失和跟踪精度低的问题,提出了一种基于飞蛾扑火优化(moth-flame optimization,MFO)的尺度比例感知空间长期跟踪器。首先,设计了高斯初始化以代替飞蛾扑火优化算法的随机初始化策略,降低优化算法在跟踪过程中的计算复杂度,减少算力浪费;其次,结合快速梯度直方图特征,构建了改进的飞蛾扑火优化跟踪器;然后,为了解决无人机航拍长期跟踪中目标尺度变化的问题,设计了一种自适应尺度变换的判别尺度空间跟踪(discriminative scale space tracking,DSST)算法,进一步提出了一种尺度比例感知空间跟踪器,解决了尺度滤波器中因长宽比固定而导致的跟踪漂移;同时,分析了滤波器响应峰值在各背景下的变化情况,提出了一种能反映环境变化下跟踪置信度的指标,并通过置信度将MFO优化跟踪框架与尺度比例感知空间跟踪器相结合,解决了尺度变化与长期跟踪目标丢失的问题;最后,在无人机长期跟踪数据集上开展了性能验证。结果表明:提出的算法可有效防止漂移现象的发生,提升跟踪效率;与目前跟踪领域中12种同类文献算法进行对比可知,提出的算法精度较高,满足实时性,能够有效解决无人机长期跟踪下的尺度变化及目标丢失等问题。
文摘Moth Flame Optimization(MFO)is a nature-inspired optimization algorithm,based on the principle of navigation technique of moth toward moon.Due to less parameter and easy implementation,MFO is used in various field to solve optimization problems.Further,for the complex higher dimensional problems,MFO is unable to make a good trade-off between global and local search.To overcome these drawbacks of MFO,in this work,an enhanced MFO,namely WF-MFO,is introduced to solve higher dimensional optimization problems.For a more optimal balance between global and local search,the original MFO’s exploration ability is improved by an exploration operator,namely,Weibull flight distribution.In addition,the local optimal solutions have been avoided and the convergence speed has been increased using a Fibonacci search process-based technique that improves the quality of the solutions found.Twenty-nine benchmark functions of varying complexity with 1000 and 2000 dimensions have been utilized to verify the projected WF-MFO.Numerous popular algorithms and MFO versions have been compared to the achieved results.In addition,the robustness of the proposed WF-MFO method has been evaluated using the Friedman rank test,the Wilcoxon rank test,and convergence analysis.Compared to other methods,the proposed WF-MFO algorithm provides higher quality solutions and converges more quickly,as shown by the experiments.Furthermore,the proposed WF-MFO has been used to the solution of two engineering design issues,with striking success.The improved performance of the proposed WF-MFO algorithm for addressing larger dimensional optimization problems is guaranteed by analyses of numerical data,statistical tests,and convergence performance.
文摘针对当前网络入侵检测中的数据量较大、数据维度较高的特点,将飞蛾扑火优化(MFO)算法应用于网络入侵检测的特征选择中。鉴于MFO算法收敛过快、易陷入局部最优的问题,提出一种融合粒子群优化(PSO)的二进制飞蛾扑火优化(BPMFO)算法。该算法引入MFO螺旋飞行公式,具有较强的局部搜索能力;结合了粒子群优化(PSO)算法的速度更新方法,让种群个体随着全局最优解和历史最优解的方向移动,增强算法的全局收敛性,从而避免易陷入局部最优。仿真实验以KDD CUP 99数据集为实验基础,分别采用支持向量机(SVM)、K最近邻(KNN)算法和朴素贝叶斯(NBC)3种分类器,与二进制飞蛾扑火优化(BMFO)算法、二进制粒子群优化(BPSO)算法、二进制遗传算法(BGA)、二进制灰狼优化(BGWO)算法和二进制布谷鸟搜索(BCS)算法进行了实验对比。实验结果表明,BPMFO算法应用于网络入侵检测的特征选择时,在算法精度、运行效率、稳定性、收敛速度以及跳出局部最优的综合性能上具有明显优势。
文摘及时准确识别母猪的发情行为可以有效增加受胎率和产仔量,对提高养殖企业的繁育水平和经济效益具有重要意义。该研究针对生猪养殖过程中母猪发情行为识别存在主观性强、智能化水平低、假警报和错误率高、识别不及时等问题,提出了一种基于飞蛾扑火算法(Moth-Flame Optimization,MFO)优化长短时记忆网络(Long Short Term Memory,LSTM)的母猪发情行为识别方法。利用安装在母猪颈部的姿态传感器获得母猪姿态数据,然后使用姿态数据训练MFO-LSTM姿态分类模型,将母猪姿态分为立姿、卧姿和爬跨3类。通过对姿态分类结果进行分析,确定以爬跨行为和活动量2个特征作为发情行为识别依据,使用MFO-LSTM分类算法判断母猪是否发情。以山西省太原市杏花岭区五丰养殖场的试验数据对该方法进行验证,结果表明,该方法在以30 min为发情行为识别时间时的识别效果最好,发情行为识别的错误率为13.43%,召回率为90.63%,特效性为81.63%,与已有的母猪发情行为识别方法相比错误率降低了80%以上。该方法在保证识别准确率的情况下有效降低了错误率,可满足母猪养殖生产过程中发情行为自动识别要求。