针对传统卷积神经网络故障诊断方法提取特征不丰富,容易丢失故障敏感信息,且在单一尺度处理方法限制实际复杂工况下故障特性的深度挖掘问题,提出了注意力机制的多尺度卷积神经网络和双向长短期记忆(bi-directional long short-term memo...针对传统卷积神经网络故障诊断方法提取特征不丰富,容易丢失故障敏感信息,且在单一尺度处理方法限制实际复杂工况下故障特性的深度挖掘问题,提出了注意力机制的多尺度卷积神经网络和双向长短期记忆(bi-directional long short-term memory,BiLSTM)网络融合的迁移学习故障诊断方法。该方法首先应用不同尺寸池化层和卷积核捕获振动信号的多尺度特征;然后引入多头自注意力机制自动地给予特征序列中的不同部分不同的权重,进一步加强特征表示的能力;其次利用BiLSTM结构引入双向性质提取特征前后之间的内部关系实现信息的逐层传递;最后利用多核最大均值差异减小源域和目标域在预训练模型中各层上的概率分布差异并利用少量标记的目标域数据再对模型进行训练。试验结果表明,所提方法在江南大学(JNU)、德国帕德博恩大学(PU)公开轴承数据集上平均准确率分别为98.43%和97.66%,该方法在重庆长江轴承股份有限公司自制的轴承故障数据集上也表现出了极高的准确率和较快的收敛速度,为有效诊断振动旋转部件故障提供了实际依据。展开更多
随着电力现货市场的开展,短期电价预测对于各市场主体的决策有着重要意义,而高比例清洁能源与储能的不断接入给短期电价预测带来很大挑战。提出一种基于最大信息系数法(maximum information coefficient,MIC)、集成经验模态分解(ensembl...随着电力现货市场的开展,短期电价预测对于各市场主体的决策有着重要意义,而高比例清洁能源与储能的不断接入给短期电价预测带来很大挑战。提出一种基于最大信息系数法(maximum information coefficient,MIC)、集成经验模态分解(ensemble empirical mode decomposition,EEMD)和改进Informer的短期电价多步预测模型。首先,采用MIC分析出与电价相关性较高的几类因素作为模型原始输入序列;然后,将上述原始序列进行EEMD分解后得到多条本征模函数(intrinsic mode function,IMF)和一个残余项后输入改进Informer分别得到翌日24点多步预测结果,再对预测结果进行滤波;最后,将滤波后序列的预测结果叠加得到最终的预测值。以西班牙电力市场数据进行验证,实验结果证明该模型可以有效提高电力市场短期电价多步预测精度。展开更多
A grey model with periodic term for sea-level analysis is presented.The present model keeps some advantages of tea GM (1, 1 )model, which well reflects the trend of sea-level changes and gives out the change rate as ...A grey model with periodic term for sea-level analysis is presented.The present model keeps some advantages of tea GM (1, 1 )model, which well reflects the trend of sea-level changes and gives out the change rate as well as the acceleration of sea level conveniently.level conveniently.In addition, the present model can reproduce the periodic phenomena of sealevel, hence, it overcomes the shortcomings of the GM(1,1) model that is unsuitable for forecasting monthly mean sealevel with apparent periodicity, and its prediction accuracy is improved.The present model is used to analyse Guangxi coast sea level,the results show that the rise rates of relative sea level at Beihai, Weizhou and Bailongwei are 1 .67,2 .51 and 0.89 mm/a respectively, the relative sea level at Shitoubu has a falling trend with a rate of 0. 5- 1 .0 mm/a, the rise rate of eustatic sea level along the Guangxi coast is 2 .0 mm/a. In comparison with the model with a lineartrend term plus a periodic term, the simulation accuracies of both models are about the same.展开更多
A term structure model bearing features of stochastic volatility and stochastic mean drift with jump （SVJ-SD model for short） is built in the paper to describe the stochastic behavior of interest rates.Based on samp...A term structure model bearing features of stochastic volatility and stochastic mean drift with jump （SVJ-SD model for short） is built in the paper to describe the stochastic behavior of interest rates.Based on sample data of an interest rate of national bond repurchase,maximum likelihood （ML）,linear Kalman filter and efficient method of moments （EMM） are used to estimate the model.While ML works well for simple models,it may lead to considerable deviation in parameter estimation when dynamic risks of interest rates are considered in them.Linear Kalman filter is a tractable and reasonably accurate technique for estimation cases where ML was not feasible.Moreover,when compared with the first two approaches,using EMM can obtain better parameter estimates for complex models with non-affine structures.展开更多
依托规模庞大的常规水电增建混合式抽水蓄能是加快抽蓄发展的重要途径。相比传统抽蓄电站,融合改造的混合式抽水蓄能电站具有水力联系更紧密、抽-发工况转换更复杂、“量调并重”角色更鲜明等显著特点,为探索其运行模式,本文提出一种融...依托规模庞大的常规水电增建混合式抽水蓄能是加快抽蓄发展的重要途径。相比传统抽蓄电站,融合改造的混合式抽水蓄能电站具有水力联系更紧密、抽-发工况转换更复杂、“量调并重”角色更鲜明等显著特点,为探索其运行模式,本文提出一种融合改造的混合式抽水蓄能与风电联合运行短期调度模型。该模型以联合体整体收益最大为目标,以机组为最小调度单元,针对常规机组和抽蓄机组的差异化运行特性分别精细化建模,并引入状态变量实现运行状态的解耦与关联切换。在模型求解方面,通过线性化方法及建模技巧将原模型转换为混合整数线性规划(Mixed Integer Linear Programming,MILP)模型,最后在JAVA环境中采用CPLEX工具进行求解。以西南某流域电站为参考构建的应用示例验证了本文模型和求解方法的有效性,可为推进常规水电站的融合改造提供借鉴。展开更多
文摘针对传统卷积神经网络故障诊断方法提取特征不丰富,容易丢失故障敏感信息,且在单一尺度处理方法限制实际复杂工况下故障特性的深度挖掘问题,提出了注意力机制的多尺度卷积神经网络和双向长短期记忆(bi-directional long short-term memory,BiLSTM)网络融合的迁移学习故障诊断方法。该方法首先应用不同尺寸池化层和卷积核捕获振动信号的多尺度特征;然后引入多头自注意力机制自动地给予特征序列中的不同部分不同的权重,进一步加强特征表示的能力;其次利用BiLSTM结构引入双向性质提取特征前后之间的内部关系实现信息的逐层传递;最后利用多核最大均值差异减小源域和目标域在预训练模型中各层上的概率分布差异并利用少量标记的目标域数据再对模型进行训练。试验结果表明,所提方法在江南大学(JNU)、德国帕德博恩大学(PU)公开轴承数据集上平均准确率分别为98.43%和97.66%,该方法在重庆长江轴承股份有限公司自制的轴承故障数据集上也表现出了极高的准确率和较快的收敛速度,为有效诊断振动旋转部件故障提供了实际依据。
文摘A grey model with periodic term for sea-level analysis is presented.The present model keeps some advantages of tea GM (1, 1 )model, which well reflects the trend of sea-level changes and gives out the change rate as well as the acceleration of sea level conveniently.level conveniently.In addition, the present model can reproduce the periodic phenomena of sealevel, hence, it overcomes the shortcomings of the GM(1,1) model that is unsuitable for forecasting monthly mean sealevel with apparent periodicity, and its prediction accuracy is improved.The present model is used to analyse Guangxi coast sea level,the results show that the rise rates of relative sea level at Beihai, Weizhou and Bailongwei are 1 .67,2 .51 and 0.89 mm/a respectively, the relative sea level at Shitoubu has a falling trend with a rate of 0. 5- 1 .0 mm/a, the rise rate of eustatic sea level along the Guangxi coast is 2 .0 mm/a. In comparison with the model with a lineartrend term plus a periodic term, the simulation accuracies of both models are about the same.
基金Sponsored by the National Natural Science Foundation of China(60979010)
文摘A term structure model bearing features of stochastic volatility and stochastic mean drift with jump （SVJ-SD model for short） is built in the paper to describe the stochastic behavior of interest rates.Based on sample data of an interest rate of national bond repurchase,maximum likelihood （ML）,linear Kalman filter and efficient method of moments （EMM） are used to estimate the model.While ML works well for simple models,it may lead to considerable deviation in parameter estimation when dynamic risks of interest rates are considered in them.Linear Kalman filter is a tractable and reasonably accurate technique for estimation cases where ML was not feasible.Moreover,when compared with the first two approaches,using EMM can obtain better parameter estimates for complex models with non-affine structures.
文摘依托规模庞大的常规水电增建混合式抽水蓄能是加快抽蓄发展的重要途径。相比传统抽蓄电站,融合改造的混合式抽水蓄能电站具有水力联系更紧密、抽-发工况转换更复杂、“量调并重”角色更鲜明等显著特点,为探索其运行模式,本文提出一种融合改造的混合式抽水蓄能与风电联合运行短期调度模型。该模型以联合体整体收益最大为目标,以机组为最小调度单元,针对常规机组和抽蓄机组的差异化运行特性分别精细化建模,并引入状态变量实现运行状态的解耦与关联切换。在模型求解方面,通过线性化方法及建模技巧将原模型转换为混合整数线性规划(Mixed Integer Linear Programming,MILP)模型,最后在JAVA环境中采用CPLEX工具进行求解。以西南某流域电站为参考构建的应用示例验证了本文模型和求解方法的有效性,可为推进常规水电站的融合改造提供借鉴。
