Harris hawks optimization(HHO)algorithm is an efficient method of solving function optimization problems.However,it is still confronted with some limitations in terms of low precision,low convergence speed and stagnat...Harris hawks optimization(HHO)algorithm is an efficient method of solving function optimization problems.However,it is still confronted with some limitations in terms of low precision,low convergence speed and stagnation to local optimum.To this end,an improved HHO(IHHO)algorithm based on good point set and nonlinear convergence formula is proposed.First,a good point set is used to initialize the positions of the population uniformly and randomly in the whole search area.Second,a nonlinear exponential convergence formula is designed to balance exploration stage and exploitation stage of IHHO algorithm,aiming to find all the areas containing the solutions more comprehensively and accurately.The proposed IHHO algorithm tests 17 functions and uses Wilcoxon test to verify the effectiveness.The results indicate that IHHO algorithm not only has faster convergence speed than other comparative algorithms,but also improves the accuracy of solution effectively and enhances its robustness under low dimensional and high dimensional conditions.展开更多
联邦学习是一种新型的分布式机器学习方法,可以在不共享原始数据的前提下训练模型。当前,联邦学习方法存在针对模型准确率最优化、通信成本最优化、参与者性能分布均衡等多个目标同时优化难的问题,难以做到多目标的同步均衡。针对该问题...联邦学习是一种新型的分布式机器学习方法,可以在不共享原始数据的前提下训练模型。当前,联邦学习方法存在针对模型准确率最优化、通信成本最优化、参与者性能分布均衡等多个目标同时优化难的问题,难以做到多目标的同步均衡。针对该问题,提出联邦学习四目标优化模型及求解算法。将全局模型错误率、模型准确率分布方差、通信成本、数据成本作为优化目标,构建优化模型。同时,针对该模型的求解搜索空间大,传统NSGA-Ⅲ算法难以寻优的问题,提出基于佳点集初始化策略的改进NSGA-Ⅲ联邦学习多目标优化算法GPNSGA-Ⅲ(Good Point Set Initialization NSGA-Ⅲ),以求取Pareto最优解。该算法通过佳点集初始化策略将有限的初始化种群以均匀的方式分布在目标求解空间中,相较于原始算法,使第一代解最大限度地接近最优值,提升寻优能力。实验结果证明,GPNSGA-Ⅲ算法得到的Pareto解的超体积值相较于NSGA-Ⅲ算法平均提升107%;Spacing值相较于NSGA-Ⅲ算法平均下降32.3%;对比其他多目标优化算法,GPNSGA-Ⅲ算法能在保证模型准确率的情况下,更有效地实现模型分布方差、通信成本和数据成本的均衡。展开更多
锅炉燃烧优化在电厂锅炉经济稳定运行中起着重要作用,NO_(x)排放预测是其中的一个基本环节,因此提出了一种基于改进蜣螂优化算法优化卷积神经网络(convolutional neural network,CNN)与双向长短期记忆神经网络(long short term memory,L...锅炉燃烧优化在电厂锅炉经济稳定运行中起着重要作用,NO_(x)排放预测是其中的一个基本环节,因此提出了一种基于改进蜣螂优化算法优化卷积神经网络(convolutional neural network,CNN)与双向长短期记忆神经网络(long short term memory,LSTM)的组合模型超参数的超超临界锅炉NO_(x)排放预测的方法。首先通过Pearson相关性判定与NO_(x)排放相关的特征参数;其次建立CNN-LSTM预测模型,利用卷积神经网络CNN提取分层数据结构,长短期记忆网络挖掘长期依赖关系,然后结合佳点集、t分布变异策略对蜣螂算法进行改进,用改进后的算法对LSTM超参数进行优化得到最终预测模型;最后与其他神经网络模型进行对比验证。以某660 MW机组锅炉深度调峰实际数据进行预测,结果得到NO_(x)排放浓度实际值与预测值的平均绝对误差为3.3516,平均相对误差为2.4667,数据结果表明该预测模型具有更准确的预测效果。展开更多
基金supported by the National Natural Science Foundation of China(61872126)。
文摘Harris hawks optimization(HHO)algorithm is an efficient method of solving function optimization problems.However,it is still confronted with some limitations in terms of low precision,low convergence speed and stagnation to local optimum.To this end,an improved HHO(IHHO)algorithm based on good point set and nonlinear convergence formula is proposed.First,a good point set is used to initialize the positions of the population uniformly and randomly in the whole search area.Second,a nonlinear exponential convergence formula is designed to balance exploration stage and exploitation stage of IHHO algorithm,aiming to find all the areas containing the solutions more comprehensively and accurately.The proposed IHHO algorithm tests 17 functions and uses Wilcoxon test to verify the effectiveness.The results indicate that IHHO algorithm not only has faster convergence speed than other comparative algorithms,but also improves the accuracy of solution effectively and enhances its robustness under low dimensional and high dimensional conditions.
文摘联邦学习是一种新型的分布式机器学习方法,可以在不共享原始数据的前提下训练模型。当前,联邦学习方法存在针对模型准确率最优化、通信成本最优化、参与者性能分布均衡等多个目标同时优化难的问题,难以做到多目标的同步均衡。针对该问题,提出联邦学习四目标优化模型及求解算法。将全局模型错误率、模型准确率分布方差、通信成本、数据成本作为优化目标,构建优化模型。同时,针对该模型的求解搜索空间大,传统NSGA-Ⅲ算法难以寻优的问题,提出基于佳点集初始化策略的改进NSGA-Ⅲ联邦学习多目标优化算法GPNSGA-Ⅲ(Good Point Set Initialization NSGA-Ⅲ),以求取Pareto最优解。该算法通过佳点集初始化策略将有限的初始化种群以均匀的方式分布在目标求解空间中,相较于原始算法,使第一代解最大限度地接近最优值,提升寻优能力。实验结果证明,GPNSGA-Ⅲ算法得到的Pareto解的超体积值相较于NSGA-Ⅲ算法平均提升107%;Spacing值相较于NSGA-Ⅲ算法平均下降32.3%;对比其他多目标优化算法,GPNSGA-Ⅲ算法能在保证模型准确率的情况下,更有效地实现模型分布方差、通信成本和数据成本的均衡。
文摘锅炉燃烧优化在电厂锅炉经济稳定运行中起着重要作用,NO_(x)排放预测是其中的一个基本环节,因此提出了一种基于改进蜣螂优化算法优化卷积神经网络(convolutional neural network,CNN)与双向长短期记忆神经网络(long short term memory,LSTM)的组合模型超参数的超超临界锅炉NO_(x)排放预测的方法。首先通过Pearson相关性判定与NO_(x)排放相关的特征参数;其次建立CNN-LSTM预测模型,利用卷积神经网络CNN提取分层数据结构,长短期记忆网络挖掘长期依赖关系,然后结合佳点集、t分布变异策略对蜣螂算法进行改进,用改进后的算法对LSTM超参数进行优化得到最终预测模型;最后与其他神经网络模型进行对比验证。以某660 MW机组锅炉深度调峰实际数据进行预测,结果得到NO_(x)排放浓度实际值与预测值的平均绝对误差为3.3516,平均相对误差为2.4667,数据结果表明该预测模型具有更准确的预测效果。
