针对高维网络数据存在大量冗余和不相关的特征导致入侵检测准确率低的问题,提出了一种改进的多因子优化蝙蝠算法(IMFBA)用于数据特征选择,筛选出具有最大信息量的特征子集,提高网络入侵检测精度。首先,在多因子优化框架下设计全局特征...针对高维网络数据存在大量冗余和不相关的特征导致入侵检测准确率低的问题,提出了一种改进的多因子优化蝙蝠算法(IMFBA)用于数据特征选择,筛选出具有最大信息量的特征子集,提高网络入侵检测精度。首先,在多因子优化框架下设计全局特征选择任务和局部特征选择任务,并通过基于蝙蝠算法所设计的选型交配和垂直文化传播算子实现不同任务间的信息共享,从而帮助全局特征选择任务更快锁定最优解空间,提高算法收敛速度和稳定性。其次,通过将反向学习策略和差分进化引入蝙蝠算法,重新设计算法初始解选择阶段及个体更新过程,弥补其缺少突变机制的不足,增强解的多样性,帮助算法摆脱局部最优。最后,提出一种自适应参数调整策略,根据潜在最优解质量决定其指导个体更新的权重,避免在多任务特征选择过程中出现知识负迁移现象,实现全局搜索与局部开发之间的平衡。实验结果表明:IMFBA所选特征子集对网络入侵数据集KDD CUP 99和NSL-KDD分类结果的准确率分别为95.37%和85.14%,相较于完整特征集提升了3.01百分点和9.78百分点。IMFBA算法能选择更高质量特征子集并提升网络入侵检测准确率。展开更多
针对中证1000成分股的投资策略构建,提出了基于支持向量机的多因子融合量价数据选股策略,应用于实时或模拟市场数据的回测中,以评估策略的有效性。本文对基于支持向量机的多因子选股模型进行改进,将量价数据融合到选股模型中,在多因子...针对中证1000成分股的投资策略构建,提出了基于支持向量机的多因子融合量价数据选股策略,应用于实时或模拟市场数据的回测中,以评估策略的有效性。本文对基于支持向量机的多因子选股模型进行改进,将量价数据融合到选股模型中,在多因子选股模型筛选出的股票基础上,进一步融合量价数据再次筛选,以期望获得更优的收益。回测结果表明,加入量价数据与未加入量价数据的模型对比,策略收益率提高了4.83%,策略年化收益率提高了11.43%,策略累积收益率与夏普比率显著优于多因子选股策略,最大回撤比之减小或者略高,预测涨跌更为接近实际股票涨跌趋势。实验结果表明:基于支持向量机的多因子融合量价数据选股策略应用在量化投资上,是十分有效的。In response to the construction of investment strategy for the constituent stocks of China Securities 1000, a stock selection strategy with the fusion of data of multi-factor, volume and price based on support vector machine is proposed which is applied to real-time or simulated market data backtesting, in order to evaluate the effectiveness of the strategy. We improve the multi-factor model of stock selection based on support vector machine, integrate the volume and price data into the model of stock selection, and further fuse the volume and price data to screen the stocks selected by the multi-factor model of stock selection to obtain more optimal returns. According to the backtest results, compared with the model with or without the addition of volume and price data and the model, the strategy return rate increases by 4.83%, the strategy annual return rate increases by 11.43%, the strategy cumulative return to sharp ratio is significantly better than that of the multi-factor stock selection strategy, and the maximum retracement ratio decreases or is slightly higher. The improved forecast moves more closely to the actual trend of stock movements. The results show that the strategy of stock selection with the fusion of data of multi-factor volume and price based on support vector machine is very effective in quantitative investment.展开更多
文摘针对高维网络数据存在大量冗余和不相关的特征导致入侵检测准确率低的问题,提出了一种改进的多因子优化蝙蝠算法(IMFBA)用于数据特征选择,筛选出具有最大信息量的特征子集,提高网络入侵检测精度。首先,在多因子优化框架下设计全局特征选择任务和局部特征选择任务,并通过基于蝙蝠算法所设计的选型交配和垂直文化传播算子实现不同任务间的信息共享,从而帮助全局特征选择任务更快锁定最优解空间,提高算法收敛速度和稳定性。其次,通过将反向学习策略和差分进化引入蝙蝠算法,重新设计算法初始解选择阶段及个体更新过程,弥补其缺少突变机制的不足,增强解的多样性,帮助算法摆脱局部最优。最后,提出一种自适应参数调整策略,根据潜在最优解质量决定其指导个体更新的权重,避免在多任务特征选择过程中出现知识负迁移现象,实现全局搜索与局部开发之间的平衡。实验结果表明:IMFBA所选特征子集对网络入侵数据集KDD CUP 99和NSL-KDD分类结果的准确率分别为95.37%和85.14%,相较于完整特征集提升了3.01百分点和9.78百分点。IMFBA算法能选择更高质量特征子集并提升网络入侵检测准确率。
文摘针对中证1000成分股的投资策略构建,提出了基于支持向量机的多因子融合量价数据选股策略,应用于实时或模拟市场数据的回测中,以评估策略的有效性。本文对基于支持向量机的多因子选股模型进行改进,将量价数据融合到选股模型中,在多因子选股模型筛选出的股票基础上,进一步融合量价数据再次筛选,以期望获得更优的收益。回测结果表明,加入量价数据与未加入量价数据的模型对比,策略收益率提高了4.83%,策略年化收益率提高了11.43%,策略累积收益率与夏普比率显著优于多因子选股策略,最大回撤比之减小或者略高,预测涨跌更为接近实际股票涨跌趋势。实验结果表明:基于支持向量机的多因子融合量价数据选股策略应用在量化投资上,是十分有效的。In response to the construction of investment strategy for the constituent stocks of China Securities 1000, a stock selection strategy with the fusion of data of multi-factor, volume and price based on support vector machine is proposed which is applied to real-time or simulated market data backtesting, in order to evaluate the effectiveness of the strategy. We improve the multi-factor model of stock selection based on support vector machine, integrate the volume and price data into the model of stock selection, and further fuse the volume and price data to screen the stocks selected by the multi-factor model of stock selection to obtain more optimal returns. According to the backtest results, compared with the model with or without the addition of volume and price data and the model, the strategy return rate increases by 4.83%, the strategy annual return rate increases by 11.43%, the strategy cumulative return to sharp ratio is significantly better than that of the multi-factor stock selection strategy, and the maximum retracement ratio decreases or is slightly higher. The improved forecast moves more closely to the actual trend of stock movements. The results show that the strategy of stock selection with the fusion of data of multi-factor volume and price based on support vector machine is very effective in quantitative investment.
