For the target threat evaluation of warships formation air defense, the sample data are frequently insufficient and even incomplete. The existing evaluation methods rely too much on expertise and are difficult to carr...For the target threat evaluation of warships formation air defense, the sample data are frequently insufficient and even incomplete. The existing evaluation methods rely too much on expertise and are difficult to carry out for the dynamic evaluation on time series. In order to solve these problems, a threat evaluation method based on the AR(p)(auto regressive(AR))-dynamic improved technique for order preference by similarity to ideal solution(DITOPSIS) method is proposed. The AR(p) model is adopted to predict the missing data on the time series. Then, the entropy weight method is applied to solve each index weight at the objective point. Kullback-Leibler divergence(KLD) is used to improve the traditional TOPSIS, and to carry out the target threat evaluation. The Poisson distribution is used to assign the weight value.Simulation results show that the improved AR(p)-DITOPSIS threat evaluation method can synthetically take into account the target threat degree in time series and is more suitable for the threat evaluation under the condition of missing the target data than the traditional TOPSIS method.展开更多
Sustainable development of power and energy systems(PES)can effectively handle challenges of fuel shortage,environmental pollution,climate change,energy security,etc.Data of PES presents distinctive characteristics in...Sustainable development of power and energy systems(PES)can effectively handle challenges of fuel shortage,environmental pollution,climate change,energy security,etc.Data of PES presents distinctive characteristics including large collection,wide coverage,diverse temporal and spatial scales,inconsistent sparsity,multiple structures and low value density,putting forward higher requirements for real-time and accuracy of data analysis,and bringing great challenges to operation analysis and coordinated control of PES.In order to realize data quality improvement and further support flexible choice of operating mode,safe and efficient coordinated control,dynamic and orderly fault recovery of sustainable PES,this paper proposes an unscented particle filter algorithm,adopting unscented Kalman filter to construct importance density functions and KLD resampling to dynamically adjust the particle number.Simulation results obtained by taking an 85-node system as a benchmark for simulation verification show that compared with traditional PF algorithm and UKF algorithm,UPF algorithm has higher estimation accuracy.展开更多
To relax the target aspect sensitivity and use more statistical information of the High Range Resolution Profiles (HRRPs), in this paper, the average range profile and the variance range profile are extracted together...To relax the target aspect sensitivity and use more statistical information of the High Range Resolution Profiles (HRRPs), in this paper, the average range profile and the variance range profile are extracted together as the feature vectors for both training data and test data representa-tion. And a decision rule is established for Automatic Target Recognition (ATR) based on the mini-mum Kullback-Leibler Distance (KLD) criterion. The recognition performance of the proposed method is comparable with that of Adaptive Gaussian Classifier (AGC) with multiple test HRRPs, but the proposed method is much more computational efficient. Experimental results based on the measured data show that the minimum KLD classifier is effective.展开更多
In speech recognition, acoustic modeling always requires tremendous transcribed samples, and the transcription becomes intensively time-consuming and costly. In order to aid this labor-intensive process, Active Learni...In speech recognition, acoustic modeling always requires tremendous transcribed samples, and the transcription becomes intensively time-consuming and costly. In order to aid this labor-intensive process, Active Learning （AL） is adopted for speech recognition, where only the most informative training samples are selected for manual annotation. In this paper, we propose a novel active learning method for Chinese acoustic modeling, the methods for initial training set selection based on Kullback-Leibler Divergence （KLD） and sample evaluation based on multi-level confusion networks are proposed and adopted in our active learning system, respectively. Our experiments show that our proposed method can achieve satisfying performances.展开更多
贝叶斯信息判据(Bayesian Information Criterion,BIC)是一种传统的在线说话人分割(online speaker seg-mentation)算法,但是它对于不同的语料需要设置不同的阈值,无法达到普适性,而且在时延较低时性能较差。提出了一种基于相对熵(Kullb...贝叶斯信息判据(Bayesian Information Criterion,BIC)是一种传统的在线说话人分割(online speaker seg-mentation)算法,但是它对于不同的语料需要设置不同的阈值,无法达到普适性,而且在时延较低时性能较差。提出了一种基于相对熵(Kullback-Leibler Divergence,KLD)和贝叶斯信息判据(Bayesian Information Criterion,BIC)的在线说话人分割算法,相对熵能度量两个模型之间的距离,再根据距离的变化来确定分割点出现的范围,而贝叶斯信息判据算法能够得到疑似分割点的位置,将两者结合起来,假如疑似分割点的位置在预先判定的范围内,则分割点有效,反之无效。实验表明相比于传统的分割算法,基于相对熵和贝叶斯信息判据的在线说话人分割算法,无需根据不同的语料事先设置阈值,在保证8 s以内时延的情况下,相对于传统的方法错误率减少约12%。展开更多
基金supported by the Postdoctoral Science Foundation of China(2013T60923)
文摘For the target threat evaluation of warships formation air defense, the sample data are frequently insufficient and even incomplete. The existing evaluation methods rely too much on expertise and are difficult to carry out for the dynamic evaluation on time series. In order to solve these problems, a threat evaluation method based on the AR(p)(auto regressive(AR))-dynamic improved technique for order preference by similarity to ideal solution(DITOPSIS) method is proposed. The AR(p) model is adopted to predict the missing data on the time series. Then, the entropy weight method is applied to solve each index weight at the objective point. Kullback-Leibler divergence(KLD) is used to improve the traditional TOPSIS, and to carry out the target threat evaluation. The Poisson distribution is used to assign the weight value.Simulation results show that the improved AR(p)-DITOPSIS threat evaluation method can synthetically take into account the target threat degree in time series and is more suitable for the threat evaluation under the condition of missing the target data than the traditional TOPSIS method.
基金supported by China Electric Power Research Institute Innovation Fund Program:Research on inexact data correction and association method for D-IoT(5242001900DS)。
文摘Sustainable development of power and energy systems(PES)can effectively handle challenges of fuel shortage,environmental pollution,climate change,energy security,etc.Data of PES presents distinctive characteristics including large collection,wide coverage,diverse temporal and spatial scales,inconsistent sparsity,multiple structures and low value density,putting forward higher requirements for real-time and accuracy of data analysis,and bringing great challenges to operation analysis and coordinated control of PES.In order to realize data quality improvement and further support flexible choice of operating mode,safe and efficient coordinated control,dynamic and orderly fault recovery of sustainable PES,this paper proposes an unscented particle filter algorithm,adopting unscented Kalman filter to construct importance density functions and KLD resampling to dynamically adjust the particle number.Simulation results obtained by taking an 85-node system as a benchmark for simulation verification show that compared with traditional PF algorithm and UKF algorithm,UPF algorithm has higher estimation accuracy.
基金Partially supported by the National Natural Science Foundation of China (No.60302009).
文摘To relax the target aspect sensitivity and use more statistical information of the High Range Resolution Profiles (HRRPs), in this paper, the average range profile and the variance range profile are extracted together as the feature vectors for both training data and test data representa-tion. And a decision rule is established for Automatic Target Recognition (ATR) based on the mini-mum Kullback-Leibler Distance (KLD) criterion. The recognition performance of the proposed method is comparable with that of Adaptive Gaussian Classifier (AGC) with multiple test HRRPs, but the proposed method is much more computational efficient. Experimental results based on the measured data show that the minimum KLD classifier is effective.
基金Acknowledgements This study is supported by the National Natural Science Foundation of China （60705019）, the National High-Tech Research and Development Plan of China （ 2006AA010102 and 2007AA01Z417）, the NOKIA project, and the 111 Project of China under Grant No. 1308004.
文摘In speech recognition, acoustic modeling always requires tremendous transcribed samples, and the transcription becomes intensively time-consuming and costly. In order to aid this labor-intensive process, Active Learning （AL） is adopted for speech recognition, where only the most informative training samples are selected for manual annotation. In this paper, we propose a novel active learning method for Chinese acoustic modeling, the methods for initial training set selection based on Kullback-Leibler Divergence （KLD） and sample evaluation based on multi-level confusion networks are proposed and adopted in our active learning system, respectively. Our experiments show that our proposed method can achieve satisfying performances.
文摘贝叶斯信息判据(Bayesian Information Criterion,BIC)是一种传统的在线说话人分割(online speaker seg-mentation)算法,但是它对于不同的语料需要设置不同的阈值,无法达到普适性,而且在时延较低时性能较差。提出了一种基于相对熵(Kullback-Leibler Divergence,KLD)和贝叶斯信息判据(Bayesian Information Criterion,BIC)的在线说话人分割算法,相对熵能度量两个模型之间的距离,再根据距离的变化来确定分割点出现的范围,而贝叶斯信息判据算法能够得到疑似分割点的位置,将两者结合起来,假如疑似分割点的位置在预先判定的范围内,则分割点有效,反之无效。实验表明相比于传统的分割算法,基于相对熵和贝叶斯信息判据的在线说话人分割算法,无需根据不同的语料事先设置阈值,在保证8 s以内时延的情况下,相对于传统的方法错误率减少约12%。
