<p> <span><span style="font-family:""><span style="font-family:Verdana;">Simulation (stochastic) methods are based on obtaining random samples </span><spa...<p> <span><span style="font-family:""><span style="font-family:Verdana;">Simulation (stochastic) methods are based on obtaining random samples </span><span style="color:#4F4F4F;font-family:Simsun;white-space:normal;background-color:#FFFFFF;"><span style="font-family:Verdana;">θ</span><sup><span style="font-family:Verdana;">5</span></sup></span><span style="font-family:Verdana;"></span><span style="font-family:Verdana;"> </span><span><span style="font-family:Verdana;"> </span><span><span style="font-family:Verdana;">from the desired distribution </span><em><span style="font-family:Verdana;">p</span></em><span style="font-family:Verdana;">(</span><span style="color:#4F4F4F;font-family:Verdana;white-space:normal;background-color:#FFFFFF;">θ</span><span style="font-family:Verdana;"></span><span style="font-family:Verdana;">)</span><span style="font-family:Verdana;"> </span><span style="font-family:Verdana;">and estimating the expectation of any </span></span><span><span style="font-family:Verdana;">function </span><em><span style="font-family:Verdana;">h</span></em><span style="font-family:Verdana;">(</span><span style="color:#4F4F4F;font-family:Verdana;white-space:normal;background-color:#FFFFFF;">θ</span><span style="font-family:Verdana;"></span><span style="font-family:Verdana;">)</span><span style="font-family:Verdana;">. Simulation methods can be used for high-dimensional dis</span></span><span style="font-family:Verdana;">tributions, and there are general algorithms which work for a wide variety of models. Markov chain Monte Carlo (MCMC) methods have been important </span><span style="font-family:Verdana;">in making Bayesian inference practical for generic hierarchical models in</span><span style="font-family:Verdana;"> small area estimation. Small area estimation is a method for producing reliable estimates for small areas. Model based Bayesian small area estimation methods are becoming popular for their ability to combine information from several sources as well as taking account of spatial prediction of spatial data. In this study, detailed simulation algorithm is given and the performance of a non-trivial extension of hierarchical Bayesian model for binary data under spatial misalignment is assessed. Both areal level and unit level latent processes were considered in modeling. The process models generated from the predictors were used to construct the basis so as to alleviate the problem of collinearity </span><span style="font-family:Verdana;">between the true predictor variables and the spatial random process. The</span><span style="font-family:Verdana;"> performance of the proposed model was assessed using MCMC simulation studies. The performance was evaluated with respect to root mean square error </span><span style="font-family:Verdana;">(RMSE), Mean absolute error (MAE) and coverage probability of corres</span><span style="font-family:Verdana;">ponding 95% CI of the estimate. The estimates from the proposed model perform better than the direct estimate.</span></span></span></span> </p> <p> <span></span> </p>展开更多
A modified Bayesian reliability assessment method of binomial components was proposed by fusing prior information of similar products.The traditional Bayesian method usually directly used all the prior data,ignoring t...A modified Bayesian reliability assessment method of binomial components was proposed by fusing prior information of similar products.The traditional Bayesian method usually directly used all the prior data,ignoring the differences between them,which might decrease the credibility level of reliability evaluation and result in data submergence.To solve the problem,a revised approach was derived to calculate groups of prior data's quantitative credibility,used for weighted data fusion.Then inheritance factor was introduced to build a mixed beta distribution to illustrate the innovation of new products.However,in many cases,inheritance factor was determined by Chi-square test that could not give out exact result with respect to rare failures.To make the model more precise,Barnard's exact test was suggested being used to calculate the inheritance factor.A numerical example is given to demonstrate that the modified method is successful and rational,while the classical method is too conservative and the traditional Bayesian method is too risky.展开更多
无人船环境感知是无人船智能航行的关键技术之一,目前主要依赖于可获取目标空间位置的激光雷达和提供目标类别信息的光学设备。为获得复杂海上环境下目标多维感知信息,提出一种无人船载激光雷达-相机的融合感知方法,融合PR-YOLOv8视觉...无人船环境感知是无人船智能航行的关键技术之一,目前主要依赖于可获取目标空间位置的激光雷达和提供目标类别信息的光学设备。为获得复杂海上环境下目标多维感知信息,提出一种无人船载激光雷达-相机的融合感知方法,融合PR-YOLOv8视觉检测结果和激光雷达三维点云,实现了海上目标高精度识别和空间定位。首先,利用标定板进行激光雷达和相机联合标定,构建了两传感器间的投影关系。随后,对于雷达分支,对目标点云聚类拟合,提取目标的特征信息并投影至图像;对于相机分支,基于YOLOv8提出PR-YOLOv8目标检测模型,获得高识别精度的目标检测边界框。最后,结合两分支检测结果,提出一种新的代价构建因子DSIoU(Distance-Scale Intersection over Union)关联目标,并结合贝叶斯理论,实现了多源感知信息的融合。采用青岛近海和内湖船只感知实验,验证了所提出方法的可行性和有效性。展开更多
为了提高短期风电功率预测的准确性,提出一种基于贝叶斯优化和特征融合的xLSTM(extended Long Short-Term Memory)-Transformer模型。该模型综合应用长短期记忆(LSTM)网络的时序处理能力和Transformer的自注意力机制的动态特征融合能力...为了提高短期风电功率预测的准确性,提出一种基于贝叶斯优化和特征融合的xLSTM(extended Long Short-Term Memory)-Transformer模型。该模型综合应用长短期记忆(LSTM)网络的时序处理能力和Transformer的自注意力机制的动态特征融合能力。借助贝叶斯优化方法,模型可在较少的迭代次数条件下优化超参数,显著降低模型对计算资源的依赖。实验结果表明,内蒙古某风电场数据集上,与单一的LSTM模型、Transformer模型、门控循环单元(GRU)模型以及未采用贝叶斯优化和特征融合的xLSTM-Transformer模型相比,当步长(LookBack)为4和8时,所提模型的决定系数R2较基准模型平均提升1.2%~11.3%;平均绝对误差(MAE)平均降低12.8%~38.4%;均方根误差(RMSE)平均降低8.6%~35.8%。结果表明,所提模型在短历史输入条件下具有更高的预测精度与稳定性。展开更多
文摘<p> <span><span style="font-family:""><span style="font-family:Verdana;">Simulation (stochastic) methods are based on obtaining random samples </span><span style="color:#4F4F4F;font-family:Simsun;white-space:normal;background-color:#FFFFFF;"><span style="font-family:Verdana;">θ</span><sup><span style="font-family:Verdana;">5</span></sup></span><span style="font-family:Verdana;"></span><span style="font-family:Verdana;"> </span><span><span style="font-family:Verdana;"> </span><span><span style="font-family:Verdana;">from the desired distribution </span><em><span style="font-family:Verdana;">p</span></em><span style="font-family:Verdana;">(</span><span style="color:#4F4F4F;font-family:Verdana;white-space:normal;background-color:#FFFFFF;">θ</span><span style="font-family:Verdana;"></span><span style="font-family:Verdana;">)</span><span style="font-family:Verdana;"> </span><span style="font-family:Verdana;">and estimating the expectation of any </span></span><span><span style="font-family:Verdana;">function </span><em><span style="font-family:Verdana;">h</span></em><span style="font-family:Verdana;">(</span><span style="color:#4F4F4F;font-family:Verdana;white-space:normal;background-color:#FFFFFF;">θ</span><span style="font-family:Verdana;"></span><span style="font-family:Verdana;">)</span><span style="font-family:Verdana;">. Simulation methods can be used for high-dimensional dis</span></span><span style="font-family:Verdana;">tributions, and there are general algorithms which work for a wide variety of models. Markov chain Monte Carlo (MCMC) methods have been important </span><span style="font-family:Verdana;">in making Bayesian inference practical for generic hierarchical models in</span><span style="font-family:Verdana;"> small area estimation. Small area estimation is a method for producing reliable estimates for small areas. Model based Bayesian small area estimation methods are becoming popular for their ability to combine information from several sources as well as taking account of spatial prediction of spatial data. In this study, detailed simulation algorithm is given and the performance of a non-trivial extension of hierarchical Bayesian model for binary data under spatial misalignment is assessed. Both areal level and unit level latent processes were considered in modeling. The process models generated from the predictors were used to construct the basis so as to alleviate the problem of collinearity </span><span style="font-family:Verdana;">between the true predictor variables and the spatial random process. The</span><span style="font-family:Verdana;"> performance of the proposed model was assessed using MCMC simulation studies. The performance was evaluated with respect to root mean square error </span><span style="font-family:Verdana;">(RMSE), Mean absolute error (MAE) and coverage probability of corres</span><span style="font-family:Verdana;">ponding 95% CI of the estimate. The estimates from the proposed model perform better than the direct estimate.</span></span></span></span> </p> <p> <span></span> </p>
基金National Natural Science Foundation of China(No.71371182)
文摘A modified Bayesian reliability assessment method of binomial components was proposed by fusing prior information of similar products.The traditional Bayesian method usually directly used all the prior data,ignoring the differences between them,which might decrease the credibility level of reliability evaluation and result in data submergence.To solve the problem,a revised approach was derived to calculate groups of prior data's quantitative credibility,used for weighted data fusion.Then inheritance factor was introduced to build a mixed beta distribution to illustrate the innovation of new products.However,in many cases,inheritance factor was determined by Chi-square test that could not give out exact result with respect to rare failures.To make the model more precise,Barnard's exact test was suggested being used to calculate the inheritance factor.A numerical example is given to demonstrate that the modified method is successful and rational,while the classical method is too conservative and the traditional Bayesian method is too risky.
文摘无人船环境感知是无人船智能航行的关键技术之一,目前主要依赖于可获取目标空间位置的激光雷达和提供目标类别信息的光学设备。为获得复杂海上环境下目标多维感知信息,提出一种无人船载激光雷达-相机的融合感知方法,融合PR-YOLOv8视觉检测结果和激光雷达三维点云,实现了海上目标高精度识别和空间定位。首先,利用标定板进行激光雷达和相机联合标定,构建了两传感器间的投影关系。随后,对于雷达分支,对目标点云聚类拟合,提取目标的特征信息并投影至图像;对于相机分支,基于YOLOv8提出PR-YOLOv8目标检测模型,获得高识别精度的目标检测边界框。最后,结合两分支检测结果,提出一种新的代价构建因子DSIoU(Distance-Scale Intersection over Union)关联目标,并结合贝叶斯理论,实现了多源感知信息的融合。采用青岛近海和内湖船只感知实验,验证了所提出方法的可行性和有效性。
文摘为了提高短期风电功率预测的准确性,提出一种基于贝叶斯优化和特征融合的xLSTM(extended Long Short-Term Memory)-Transformer模型。该模型综合应用长短期记忆(LSTM)网络的时序处理能力和Transformer的自注意力机制的动态特征融合能力。借助贝叶斯优化方法,模型可在较少的迭代次数条件下优化超参数,显著降低模型对计算资源的依赖。实验结果表明,内蒙古某风电场数据集上,与单一的LSTM模型、Transformer模型、门控循环单元(GRU)模型以及未采用贝叶斯优化和特征融合的xLSTM-Transformer模型相比,当步长(LookBack)为4和8时,所提模型的决定系数R2较基准模型平均提升1.2%~11.3%;平均绝对误差(MAE)平均降低12.8%~38.4%;均方根误差(RMSE)平均降低8.6%~35.8%。结果表明,所提模型在短历史输入条件下具有更高的预测精度与稳定性。
