基于深度学习的ELM实时识别研究

基于深度学习的方法，在HL-2A装置上开发出了一套边缘局域模(ELM)实时识别算法。算法使用5200次放电数据(约24.19万数据切片)进行学习，得到一个深度为22层的卷积神经网络。为衡量算法的识别能力，识别了HL-2A装置自2009年实现稳定ELMy H模放电以来所有历史数据(约26000次放电数据)，共识别出1665次H模放电，其中误识别35次，误报率为2.10%。在实际的1634次H模放电中，漏识别4次，漏识别率为0.24%。该误报率和漏报率可以满足ELM实时识别的精度要求。识别算法在实时控制环境下，对单个时间点的平均计算时间为0.46ms，可以满足实时控制的计算速度要求。     

多时滞随机神经网络的稳定性

通过构建李雅普偌夫函数的方法和利用半鞅收敛定理对一类随机时滞神经网络的全局指数稳定进行了分析,提出了易于判定随机时滞神经网络几乎必然指数稳定性新的代数判据,推广了[1]中的主要结论.     

Characterization of spherical particles using high-order neural networks and scanning flow cytometry

We retrieve the radius R, real n and imaginary k parts of the refractive index of homogeneous spherical particles using angular distribution of the light-scattering intensity. To solve the inverse light-scattering problem we use a high-order neural-network technique. The effect of network parameters on optimization is examined. The technique is evaluated for noise-corrupted input data at 0.6 μm<R<10.6 μm, 1.02<n<1.38, and 0<k<0.03. The errors of retrieval for nonabsorbing particles do not exceed 0.05 μm for radius and 0.015 for refractive index. The experimental verification is fulfilled by experimental data retrieved by means of a scanning flow cytometer. The light-scattering profiles of polystyrene beads and spherized red blood cells are processed with the high-order neural networks and a non-linear regression at Mie theory. The parameters retrieved by the high-order neural networks correlate well with the parameters retrieved by the least-square method.     

神经网络在工程爆破应力波规律探讨中的应用

采用BP（Back Propagation)前馈人工神经网络模型，对工程爆破中柱状震源的自由场应力波传播规律进行了探讨。结果表明：利用人神经网络模型的非线性映射功能，可以较好地给出工程爆破引起的近区自由场力学规律，对于同类型问题的研究，也有着很重要的意义。     

A multi-target approach for isolated word recognition using artificial neural network

I.TntroductionStatisticalandneuralnetworkmcthodsforpatternclassiflcationusesignificantlydifTerentapproachesintrainingaclassificr.Inthestatistica1approach,thcformationofaclassifier1arge1ydependsonthestatisticsofthetrainingpatternsand,insomecases,theassumptionsaboutthedistributionofthepopulation.Theneuralnetworkmethodisnon-parametricandcanbeadaptivcinthetrainingprocessl'l.Becauseofitssimplicityandflexibility,theneuralnetworkhasbecnincreasing1yusedforpatternclassiflcation.Itisnowwe11knownthatan…     

Establishing structure-property correlations and classification of base oils using statistical techniques and artificial neural networks

The present paper describes various classification techniques like cluster analysis, principal component (PC)/factor analysis to classify different types of base stocks. The API classification of base oils (Group I-III) has been compared to a more detailed NMR derived chemical compositional and molecular structural parameters based classification in order to point out the similarities of the base oils in the same group and the differences between the oils placed in different groups. The detailed compositional parameters have been generated using and nuclear magnetic resonance (NMR) spectroscopic methods. Further, oxidation stability, measured in terms of rotating bomb oxidation test (RBOT) life, of non-conventional base stocks and their blends with conventional base stocks, has been quantitatively correlated with their NMR and elemental (sulphur and nitrogen) data with the help of multiple linear regression (MLR) and artificial neural networks (ANN) techniques. The MLR based model developed using NMR and elemental data showed a high correlation between the ‘measured’ and ‘estimated’ RBOT values for both training (R=0.859) and validation (R=0.880) data sets. The ANN based model, developed using fewer number of input variables (only NMR data) also showed high correlation between the ‘measured’ and ‘estimated’ RBOT values for training (R=0.881), validation (R=0.860) and test (R=0.955) data sets.     

Retention modeling and simultaneous optimization of pH value and gradient steepness in RP‐HPLC using feed‐forward neural networks

A novel approach is proposed for the simultaneous optimization of mobile phase pH and gradient steepness in RP‐HPLC using artificial neural networks. By presetting the initial and final concentration of the organic solvent, a limited number of experiments with different gradient time and pH value of mobile phase are arranged in the two‐dimensional space of mobile phase parameters. The retention behavior of each solute is modeled using an individual artificial neural network. An “early stopping” strategy is adopted to ensure the predicting capability of neural networks. The trained neural networks can be used to predict the retention time of solutes under arbitrary mobile phase conditions in the optimization region. Finally, the optimal separation conditions can be found according to a global resolution function. The effectiveness of this method is validated by optimization of separation conditions for amino acids derivatised by a new fluorescent reagent.     

应用反向传播神经网络预测化合物脑血分配系数

选用27种三维结构性质描述符对脑血分配系数预测建立神经网络模型.网络模型选用典型的适合函数逼近的两层结构神经网络对脑血分配系数(lgBB,BB为脑血浓度比)进行预测,计算中采用的模型具有一个双曲正切型激活函数的隐含层和一个线性激活函数的输出层.计算表明,使用小心选择的反向传播神经网络模型对化合物脑血分配系数具有较好的预测能力.     

Rigorous results on the thermodynamics of the dilute Hopfield model

We study the Hopfield model of an autoassociative memory on a random graph onN vertices where the probability of two vertices being joined by a link isp(N). Assuming thatp(N) goes to zero more slowly thanO(1/N), we prove the following results: (1) If the number of stored patternsm(N) is small enough such thatm(N)/Np(N) 0, asN, then the free energy of this model converges, upon proper rescaling, to that of the standard Curie-Weiss model, for almost all choices of the random graph and the random patterns. (2) If in additionm(N) < ln N/ln 2, we prove that there exists, forT< 1, a Gibbs measure associated to each original pattern, whereas for higher temperatures the Gibbs measure is unique. The basic technical result in the proofs is a uniform bound on the difference between the Hamiltonian on a random graph and its mean value.