基于感知器的SVM自学习模型

提出了一种基于感知器的SVM分类模型（PSVM）。该模型在对分类器的训练中,引入感知器分类思想,其先利用SVM的核函数进行核计算,判断其分类性能,分类正确则不作任何修改,反之则转化成感知器分类问题。实验结果表明该模型不但能提高SVM的分类性能,而且还可以降低SVM分类性能对核函数及参数选择的依赖。     

Short-term electricity demand and gas price forecasts using wavelet transforms and adaptive models

This paper presents some forecasting techniques for energy demand and price prediction, one day ahead. These techniques combine wavelet transform (WT) with fixed and adaptive machine learning/time series models (multi-layer perceptron (MLP), radial basis functions, linear regression, or GARCH). To create an adaptive model, we use an extended Kalman filter or particle filter to update the parameters continuously on the test set. The adaptive GARCH model is a new contribution, broadening the applicability of GARCH methods. We empirically compared two approaches of combining the WT with prediction models: multicomponent forecasts and direct forecasts. These techniques are applied to large sets of real data (both stationary and non-stationary) from the UK energy markets, so as to provide comparative results that are statistically stronger than those previously reported. The results showed that the forecasting accuracy is significantly improved by using the WT and adaptive models. The best models on the electricity demand/gas price forecast are the adaptive MLP/GARCH with the multicomponent forecast; their NMSEs are 0.02314 and 0.15384 respectively.     

The potential of different artificial neural network (ANN) techniques in daily global solar radiation modeling based on meteorological data

The main objective of present study is to predict daily global solar radiation (GSR) on a horizontal surface, based on meteorological variables, using different artificial neural network (ANN) techniques. Daily mean air temperature, relative humidity, sunshine hours, evaporation, and wind speed values between 2002 and 2006 for Dezful city in Iran (32°16′N, 48°25′E), are used in this study. In order to consider the effect of each meteorological variable on daily GSR prediction, six following combinations of input variables are considered:

(I)

Day of the year, daily mean air temperature and relative humidity as inputs and daily GSR as output.

(II)

Day of the year, daily mean air temperature and sunshine hours as inputs and daily GSR as output.

(III)

Day of the year, daily mean air temperature, relative humidity and sunshine hours as inputs and daily GSR as output.

(IV)

Day of the year, daily mean air temperature, relative humidity, sunshine hours and evaporation as inputs and daily GSR as output.

(V)

Day of the year, daily mean air temperature, relative humidity, sunshine hours and wind speed as inputs and daily GSR as output.

(VI)

Day of the year, daily mean air temperature, relative humidity, sunshine hours, evaporation and wind speed as inputs and daily GSR as output.

Multi-layer perceptron (MLP) and radial basis function (RBF) neural networks are applied for daily GSR modeling based on six proposed combinations.The measured data between 2002 and 2005 are used to train the neural networks while the data for 214 days from 2006 are used as testing data.The comparison of obtained results from ANNs and different conventional GSR prediction (CGSRP) models shows very good improvements (i.e. the predicted values of best ANN model (MLP-V) has a mean absolute percentage error (MAPE) about 5.21% versus 10.02% for best CGSRP model (CGSRP 5)).     

多分类器区分性组合在二次解码中的应用

提出利用基于隐马尔可夫模型的谱特征模型、基于高斯混合模型的声调分类器以及基于多层感知器的音素分类器模型的组合来提高语音识别中二次解码中的识别率。在模型组合中,使用上下文相关的模型权重加权模型得分,并使用区分性训练来优化上下文相关权重来进一步改进识别结果。对人工选取各种上下文相关权重集合进行了性能评估,连续语音识别实验表明,使用局部分类器进行二次解码能够明显降低系统误识率。在模型组合中,使用当前音节类型及左上下文相结合的模型权重集合能够最大程度降低系统误识率。实验表明该方法得到的识别结果优于基于谱特征与基频特征和音素后验概率特征合并得到特征组合的识别系统。     

集成学习算法的研究与应用

集成学习算法的思想就是集成多个学习器,并组合它们的预测结果,以形成最终的结论。典型的学习模型组合方法有投票法,专家混合方法,堆叠泛化法与级联法,但这些方法的性能都有待进一步提高。提出了一种新颖的集成学习算法--增强的集成学习算法（ReinforcedEnsemble）。ReinforcedEnsemble集成算法由两大部分组成：ReinforcedEnsemble特征提取算法与ReinforcedEnsemble基分类器。通过实验,将ReinforcedEnsemble算法与其他集成学习算法进行了性能比较。实验结果表明,所提出的算法在多项指标上均达到最优。     

A new texture and shape based technique for improving meningioma classification

Over the past decade, computer‐aided diagnosis is rapidly growing due to the availability of patient data, sophisticated image acquisition tools and advancement in image processing and machine learning algorithms. Meningiomas are the tumors of brain and spinal cord. They account for 20% of all the brain tumors. Meningioma subtype classification involves the classification of benign meningioma into four major subtypes: meningothelial, fibroblastic, transitional, and psammomatous. Under the microscope, the histology images of these four subtypes show a variety of textural and structural characteristics. High intraclass and low interclass variabilities in meningioma subtypes make it an extremely complex classification problem. A number of techniques have been proposed for meningioma subtype classification with varying performances on different subtypes. Most of these techniques employed wavelet packet transforms for textural features extraction and analysis of meningioma histology images. In this article, a hybrid classification technique based on texture and shape characteristics is proposed for the classification of meningioma subtypes. Meningothelial and fibroblastic subtypes are classified on the basis of nuclei shapes while grey‐level co‐occurrence matrix textural features are used to train a multilayer perceptron for the classification of transitional and psammomatous subtypes. On the whole, average classification accuracy of 92.50% is achieved through the proposed hybrid classifier; which to the best of our knowledge is the highest. Microsc. Res. Tech. 77:862–873, 2014. © 2014 Wiley Periodicals, Inc.     

Prevention of transient instability employing rules based on back propagation based ANN for series compensation

Power system preventive control is dominated by generation rescheduling. Other preventive control actions are rarely mentioned in the literature. This paper presents a transient stability preventive control design that takes series compensation into account. The design methodology is aided with the ‘IF-THEN’ rules extracted from a trained multilayer perceptron (MLP) artificial neural network (ANN). The proposed method can add more degrees of freedom by incorporation of series compensation, and have additional flexibility in deploying preventive control actions. Two preventive control schemes are presented and applied to 39-bus power systems. Extensive computational studies have demonstrated the effectiveness of the proposal.     

The Simulation of Microbial Enhanced Oil Recovery by Using a Two-layer Perceptron Neural Network

The authors simulated a reservoir by using two-layer perceptron. Indeed a model was developed to simulate the increase in oil recovery caused by bacteria injection into an oil reservoir. This model was affected by reservoir temperature and amount of water injected into the reservoir for enhancing oil recovery. Comparing experimental and simulation results and also the erratic trend of data show that the neural networks have modeled this system properly. Considering the effects of nonlinear factors and their erratic and unknown impacts on recovered oil, the perceptron neural network can develop a proper model for oil recovery factor in various conditions. The neural networks have not been applied in modeling of microbial enhanced oil recovery since now. Finally, we are going to design a controller for the neural network. This controller is designed for the case where output of the network is oil recovery factor. For this purpose, the network is designed as a one-layer network in which just one output matches each time. In this case, a one-layer network will have acceptable results.     

基于变分贝叶斯算法和MLP网络的后非线性混合盲源分离方法研究

传统的后非线性模型往往要求其后非线性函数是可逆的,否则无法进行源信号的分离。然而在实际中,这一要求并不完全满足。针对此不足,结合变分贝叶斯推论和多层感知器网络,提出一种改进的多层感知器后非线性模型,它通过多层感知器来模拟后非线性函数,实现对不可逆后非线性函数混合的盲分离。仿真和实验结果表明该方法是有效的。     

基于模板匹配与人工神经网确认的人脸检测

本文提出了一种基于模板匹配与人工神经网确认的人脸检测算法.算法使用一组双眼-人脸模板对搜索侯选人脸,并在搜索过程中使用多层感知器进行确认,以减少错误报警.大量实验证明了该算法的有效性.