基于Tri-training-SSAE半监督学习算法的电力系统暂态稳定评估

基于机器学习的暂态稳定评估方法主要采用监督学习方法,为了解决监督学习方法所需的有标签样本难以获取的问题,提出基于三体训练-稀疏堆叠自动编码器(Tri-training-SSAE)半监督学习算法的电力系统暂态稳定评估方法。构建基于堆叠稀疏自动编码器的暂态稳定评估模型;在传统的三体训练过程中加入伪标签样本置信度判断,以减小噪声数据对模型训练的影响;以堆叠稀疏自动编码器为基分类器构建三体训练-稀疏堆叠自动编码器模型,利用大量的无标签样本提高模型的泛化能力。通过IEEE 39节点系统与华东某省级电网进行分析验证,结果表明,所提方法在有标签样本数较少时具有更高的评估准确度。

融合光谱度量标记迁移和Tri-training的高光谱遥感图像半监督分类算法

针对海量的高光谱遥感图像光谱和丰富的空间信息中可用于分类的有标记样本远少于无标记样本的数据特性,提出了一种融合光谱度量标记迁移和Tri-training的高光谱遥感图像半监督光谱-空间分类算法。该算法提出了一种基于光谱度量的标记迁移方法,通过结合迁移标记和Tri-training预测标记进行扩充样本标记预测,提高了扩充样本标记的准确性。同时,该算法基于空间相关性选择扩充样本,综合运用光谱和空间特征提升图像分类的精度。在两个公开的高光谱遥感图像数据集上进行了实验,结果表明该算法优于基于Tri-training算法的高光谱遥感图像的分类性能。

基于Tri-training算法的多分类信用评级方法

随着经济的快速发展,信用贷款在企业资金周转中的作用越来越重要.信用评级是信用贷款发放的基本依据之一.本文针对实际信用评级中有标签样本数量不足的问题,提出一种基于Tri-training算法的多分类信用评级方法,该方法选择支持向量机、决策树和最大熵模型作为基分类器组合.最后,本文使用真实的信用数据集验证了该方法的实际效果.

Adaptive Fault Estimation for Dynamics of High Speed Train Based on Robust UKF Algorithm

This paper proposes an adaptive unscented Kalman filter algorithm(ARUKF)to implement fault estimation for the dynamics of high⁃speed train(HST)with measurement uncertainty and time⁃varying noise with unknown statistics.Firstly,regarding the actuator and sensor fault as the auxiliary variables of the dynamics of HST,an augmented system is established,and the fault estimation problem for dynamics of HST is formulated as the state estimation of the augmented system.Then,considering the measurement uncertainties,a robust lower bound is proposed to modify the update of the UKF to decrease the influence of measurement uncertainty on the filtering accuracy.Further,considering the unknown time⁃varying noise of the dynamics of HST,an adaptive UKF algorithm based on moving window is proposed to estimate the time⁃varying noise so that accurate concurrent actuator and sensor fault estimations of dynamics of HST is implemented.Finally,a five-car model of HST is given to show the effectiveness of this method.

A Simulated Annealing Algorithm for Training Empirical Potential Functions of Protein Folding

In this paper are reported the local minimum problem by means of current greedy algorithm for training the empirical potential function of protein folding on 8623 non-native structures of 31 globular proteins and a solution of the problem based upon the simulated annealing algorithm. This simulated annealing algorithm is indispensable for developing and testing highly refined empirical potential functions.

基于改进Tri-training算法投票机制的中文问句分类

原始的Tri-training算法在三个分类器给出的分类结果均不同时,默认第一个分类器给出的分类结果为分类器模型的最终结果,这在一定程度上有可能会降低分类器在这种情况下的分类精度。本文提出一种基于平时优秀思想的投票机制算法,该算法避免了默认将第一个分类器给出的结果作为分类器模型的分类结果这种片面的情况,并利用其对哈工大中文问句集和本文扩展问句集进行分类实验。结果表明,本文算法有良好的适应性,且分类正确率明显提高;适当增大训练集和未标记样本数据,可以增强分类器的泛化能力,从而使分类正确率提高。

Automatic Driving Operation Strategy of Urban Rail Train Based on Improved DQN Algorithm

To realize a better automatic train driving operation control strategy for urban rail trains,an automatic train driving method with improved DQN algorithm(classical deep reinforcement learning algorithm)is proposed as a research object.Firstly,the train control model is established by considering the train operation requirements.Secondly,the dueling network and DDQN ideas are introduced to prevent the value function overestimation problem.Finally,the priority experience playback and“restricted speed arrival time”are used to reduce the useless experience utilization.The experiments are carried out to verify the train operation strategy method by simulating the actual line conditions.From the experimental results,the train operation meets the ATO requirements,the energy consumption is 15.75%more energy-efficient than the actual operation,and the algorithm convergence speed is improved by about 37%.The improved DQN method not only enhances the efficiency of the algorithm but also forms a more effective operation strategy than the actual operation,thereby contributing meaningfully to the advancement of automatic train operation intelligence.

A Second Order Training Algorithm for Multilayer Feedforward Neural Networks

ＡＳｅｃｏｎｄＯｒｄｅｒＴｒａｉｎｉｎｇＡｌｇｏｒｉｔｈｍｆｏｒＭｕｌｔｉｌａｙｅｒＦｅｅｄｆｏｒｗａｒｄＮｅｕｒａｌＮｅｔｗｏｒｋｓＴａｎＹｉｎｇ（谭营）ＨｅＺｈｅｎｙａ（何振亚）（ＤｅｐａｒｔｍｅｎｔｏｆＲａｄｉｏＥｎｇｉｎｅｅｒｉｎｇ，Ｓｏｕ...

Research of Genetic Training Algorithm for Identifying Mechanical Failure Modes within the Framework of Case-Based Reasoning

The combination of case-based reasoning （CBR） and genetic algorithm （GA） is considered in the problem of failure mode identification in aeronautical component failure analysis. Several imple- mentation issues such as matching attributes selection, similarity measure calculation, weights learning and training evaluation policies are carefully studied. The testing applications illustrate that an accuracy of 74.67 % can be achieved with 75 balanced-distributed failure cases covering 3 failure modes, and that the resulting learning weight vector can be well applied to the other 2 failure modes, achieving 73.3 % of recognition accuracy. It is also proved that its popularizing capability is good to the recognition of even more mixed failure modes.

DOBD Algorithm for Training Neural Network: Part I. Method

Overfitting is one of the important problems that restrain the application of neural network. The traditional OBD (Optimal Brain Damage) algorithm can avoid overfitting effectively. But it needs to train the network repeatedly with low calculational efficiency. In this paper, the Marquardt algorithm is incorporated into the OBD algorithm and a new method for pruning network-the Dynamic Optimal Brain Damage (DOBD) is introduced. This algorithm simplifies a network and obtains good generalization through dynamically deleting weight parameters with low sensitivity that is defined as the change of error function value with respect to the change of weights. Also a simplified method is presented through which sensitivities can be calculated during training with a little computation. A rule to determine the lower limit of sensitivity for deleting the unnecessary weights and other control methods during pruning and training are introduced. The training course is analyzed theoretically and the reason why DOBD algorithm can obtain a much faster training speed than the OBD algorithm and avoid overfitting effectively is given.

DOBD Algorithm for Training Neural Network: Part II. Application

In the first part of the article, a new algorithm for pruning networkDynamic Optimal Brain Damage(DOBD) is introduced. In this part, two cases and an industrial application are worked out to test the new algorithm. It is verified that the algorithm can obtain good generalization through deleting weight parameters with low sensitivities dynamically and get better result than the Marquardt algorithm or the cross-validation method. Although the initial construction of network may be different, the finial number of free weights pruned by the DOBD algorithm is similar and the number is just close to the optimal number of free weights. The algorithm is also helpful to design the optimal structure of network.

基于Tri-training的主动学习算法

半监督学习和主动学习都是利用未标记数据,在少量标记数据代价下同时提高监督学习识别性能的有效方法。为此,结合主动学习方法与半监督学习的Tri-training算法,提出一种新的分类算法,通过熵优先采样算法选择主动学习的样本。针对UCI数据集和遥感数据,在不同标记训练样本比例下进行实验,结果表明,该算法在标记样本数较少的情况下能取得较好的效果。将主动学习与Tri-training算法相结合,是提高分类性能和泛化性的有效途径。

基于Tri-Training半监督分类算法的研究

在实际应用中,容易获取大量的未标记样本数据,而样本数据是有限的,因此,半监督分类算法成为研究者关注的热点。文中在协同训练Tri-Training算法的基础上,提出了采用两个不同的训练分类器的Simple-Tri-Training方法和对标记数据进行编辑的Edit-Tri-Training方法,给出了这三种分类方法与监督分类SVM的分类实验结果的比较和分析。实验表明,无标记数据的引入,在一定程度上提高了分类的性能;初始训练集和分类器的选取以及标记过程中数据编辑技术,都是影响半监督分类稳定性和性能的关键点。

结合Tri-training半监督学习和凸壳向量的SVM主动学习算法

为解决监督学习过程中难以获得大量带有类标记样本且样本数据标记代价较高的问题,结合主动学习和半监督学习方法,提出基于Tri-training半监督学习和凸壳向量的SVM主动学习算法.通过计算样本集的壳向量,选择最有可能成为支持向量的壳向量进行标记.为解决以往主动学习算法在选择最富有信息量的样本标记后,不再进一步利用未标记样本的问题,将Tri-training半监督学习方法引入SVM主动学习过程,选择类标记置信度高的未标记样本加入训练样本集,利用未标记样本集中有利于学习器的信息.在UCI数据集上的实验表明,文中算法在标记样本较少时获得分类准确率较高和泛化性能较好的SVM分类器,降低SVM训练学习的样本标记代价.

Tri-training算法中分类器组合的改进

Tri-training算法是半监督协同算法里的经典算法,该文针对算法中分类器的使用做了一些改进,由原先单一的分类器换成两个不同分类器的组合。使用SVM分类器和最大熵分类器的不同组合作为Tri-training算法里的三个分类器构成分类器模型,然后分别对稀疏型数据、密集型数据与原始Tri-training算法进行实验比较,从而验证改进的有效性。

一种自适应的Tri-Training半监督算法

Tri-Training算法是半监督算法的一种,在学习过程中容易错误标注无标记样本,从而降低分类性能,为此提出一种ADP-Tri-Training(Adaptive Tri-Training)算法,改进协同工作方式,根据几何中心设置分类器组成,然后应用模糊数学理论将多个独立的分类器组合,使得算法可以在多因素下综合评价样本,并在此基础上引入遗传算法动态设置组合权重以适应于具体的样本集,从而尽可能降低样本标注的错误率,多个实验结果表明ADP-Tri-Training算法具有更好的分类性能.

基于交叉熵的安全Tri-training算法

半监督学习方法通过少量标记数据和大量未标记数据来提升学习性能.Tri-training是一种经典的基于分歧的半监督学习方法,但在学习过程中可能产生标记噪声问题.为了减少Tri-training中的标记噪声对未标记数据的预测偏差,学习到更好的半监督分类模型,用交叉熵代替错误率以更好地反映模型预估结果和真实分布之间的差距,并结合凸优化方法来达到降低标记噪声的目的,保证模型效果.在此基础上,分别提出了一种基于交叉熵的Tri-training算法、一个安全的Tri-training算法,以及一种基于交叉熵的安全Tri-training算法.在UCI(University of California Irvine)机器学习库等基准数据集上验证了所提方法的有效性,并利用显著性检验从统计学的角度进一步验证了方法的性能.实验结果表明,提出的半监督学习方法在分类性能方面优于传统的Tri-training算法,其中基于交叉熵的安全Tri-training算法拥有更高的分类性能和泛化能力.

基于Tri-training算法的构造性学习方法

构造性机器学习(CML)算法在训练分类器时需要大量有标记样本,而获取这些有标记样本十分困难。为此,提出一种基于Tri-training算法的构造性学习方法。根据已标记的样本,采用不同策略构造3个差异较大的初始覆盖分类网络,用于对未标记数据进行标记,再将已标记数据加入到训练样本中,调整各分类网络参数,反复进行上述过程,直至获得稳定的分类器。实验结果证明,与CML算法和基于NB分类器的半监督学习算法相比,该方法的分类准确率更高。

改进M-training算法的高光谱图像分类

为了解决高光谱数据有标签样本数量有限的分类问题,提出将M-training算法应用于高光谱图像分类。采用两个SVM、一个K近邻(KNN)以及一个随机森林(RF)进行分类器组合,对传统M-training算法进行改进,增强分类器的多样性和差异性。为了充分考虑大量无标签样本的影响,采用有标签样本与无标签样本错误率加权作为有标签样本集更新的限制条件,从而有效地扩大了有标签样本集。实验结果表明:改进算法和传统的M-training算法相比较,在总体分类精度与Kappa系数上分别提高1. 85%～12. 10%与0. 021 5～0. 141 3,从而验证了该算法的有效性。

基于Co-training的用户属性预测研究

针对当前基于第三方应用数据进行用户属性预测算法研究,其较少考虑应用前台实际使用时长问题,由此,本文在应用的使用频率及使用时长的基础上,构造了应用前台均使用时长特征,该特征能进一步刻画用户对应用的兴趣度;同时,为充分利用大量未标注数据,从多角度特征对用户属性进行预测,由此本文采用了Co-training框架,该框架包含两个均由栈式自编码器与神经网络相结合的网络结构。实验过程中,对于栈式自编码算法,先利用未标注的数据对网络进行参数初始化,使得网络参数处于一个较优的位置,再利用有标注的数据,采用基于准确率的梯度下降算法,对网络参数进行更新,最终达到收敛。实验结果表明,本文算法在准确率、召回率、F1值上均有所提高。