神经网络模型用于多变量综合预测

本文研究神经网络用于多变量时间序列预测的原理与方法，提出组合多种信息的综合预测方法。以股票交易为例，用神经网络组合各类信息，运用信心股价理论对中国股市的发展进行跟踪预测。在此基础上进一步从信息利用的角度说明了神经网络预测方法的特点。结果表明，神经网络模型用于多变王时间序列预测，其精度和趋势均较统计方法有所提高；神经网络综合预测，对中短期股票价格的预测，有实用价值。     

A GENERAL APPROACH BASED ON AUTOCORRELATION TO DETERMINE INPUT VARIABLES OF NEURAL NETWORKS FOR TIME SERIES FORECASTING

Input selection is probably one of the most critical decision issues in neural network designing, because it has a great impact on forecasting performance. Among the many applications of artificial neural networks to finance, time series forecasting is perhaps one of the most challenging issues. Considering the features of neural networks, we propose a general approach called Autocorrelation Criterion (AC) to determine the inputs variables for a neural network. The purpose is to seek optimal lag periods, which are more predictive and less correlated. AC is a data-driven approach in that there is no prior assumption about the models for time series under study. So it has extensive applications and avoids a lengthy experimentation and tinkering in input selection. We apply the approach to the determination of input variables for foreign exchange rate forecasting and conduct comparisons between AC and information-based in-sample model selection criterion. The experiment results show that AC outperforms inf     

基于神经网络矫正的非线性短时负荷预测模型

为了解决传统神经网络负荷预测模型中,当预测日天气出现快速变化时预测误差随之增加的问题,提出了一种改进的未来一小时实时负荷预测模型。在该模型中,预测负荷通过对预测日的类似日负荷数据加一个矫正值来获得,矫正值从神经网络产生,网络结构得到简化。由于采用在线实时学习方式,该模型可以学习快速的天气变化和预测误差之间的关系,减小预测误差。仿真结果验证了该模型的有效性。     

BP神经网络模型的改进

本文在深入研究ＢＰ神经网络模型的基础上，提出了对该模型的若干改进技术处理方法，并通过仿真试验．实践表明，这些改进和技术处理方法是有效的．     

A multiscale modeling approach incorporating ARIMA and anns for financial market volatility forecasting

The financial market volatility forecasting is regarded as a challenging task because of irreg ularity, high fluctuation, and noise. In this study, a multiscale ensemble forecasting model is proposed. The original financial series are decomposed firstly different scale components （i.e., approximation and details） using the maximum overlap discrete wavelet transform （MODWT）. The approximation is pre- dicted by a hybrid forecasting model that combines autoregressive integrated moving average （ARIMA） with feedforward neural network （FNN）. ARIMA model is used to generate a linear forecast, and then FNN is developed as a tool for nonlinear pattern recognition to correct the estimation error in ARIMA forecast. Moreover, details are predicted by Elman neural networks. Three weekly exchange rates data are collected to establish and validate the forecasting model. Empirical results demonstrate consistent better performance of the proposed approach.     

电梯群控系统交通流的预测方法

交通流预测是智能电梯群控系统的重要组成部分,对交通流进行预测可使群控系统跟随交通流的变化调节控制策略.将基于神经网络的时间序列预测理论应用到电梯群控系统的交通分析中,构造了一种交通流时间序列预测模型,并提出了调整预测神经网络结构以提高预测精度的方法.仿真实验表明了这种交通流智能预测方法是有效的.     

神经元网络在股价预测中的应用

利用时延神经元网络模型（ＴｉｍｅＤｅｌａｙＮｅｕｒａｌＮｅｔｗｏｒｋ）对四川长虹的股价作了预测。股价的涨跌预报可视作高维空间的非线性分类问题,本文使用增益可调的反向传播算法,对其走势作了预报。借助前馈神经网络对非线性函数的逼近能力,本文对四川长虹股价这个时间序列作了连续若干天的一步预测。最后,我们采用不同形式的误差函数对预测结果作了比较。     

An Ensemble Model of Arima and Ann with Restricted Boltzmann Machine Based on Decomposition of Discrete Wavelet Transform for Time Series Forecasting

Time series forecasting research area mainly focuses on developing effective forecasting models to improve prediction accuracy. An ensemble model composed of autoregressive integrated moving average (ARIMA), artificial neural network (ANN), restricted Boltzmann machines (RBM), and discrete wavelet transform (DWT) is presented in this paper. In the proposed model, DWT first decomposes time series into approximation and detail. Then Khashei and Bijari’s model, which is an ensemble model of ARIMA and ANN, is applied to the approximation and detail to extract their both linear and nonlinear components and fit the relationship between the components as a function instead of additive relationship. Furthermore, RBM is used to perform pre-training for generating initial weights and biases based on inputs feature for ANN. Finally, the forecasted approximation and detail are combined to obtain final forecasting. The forecasting capability of the proposed model is tested with three well-known time series: sunspot, Canadian lynx, exchange rate time series. The prediction performance is compared to the other six forecasting models. The results indicate that the proposed model gives the best performance in all three data sets and all three measures (i.e. MSE, MAE and MAPE).     

Time Series Neural Network Forecasting Methods

ＴｉｍｅＳｅｒｉｅｓＮｅｕｒａｌＮｅｔｗｏｒｋＦｏｒｅｃａｓｔｉｎｇＭｅｔｈｏｄｓＷＥＮＸｉｎｈｕｉ；ＣＨＥＮＫｅｉｚｈｏｕ（ＴｈｅＣｅｎｔｌａｌｏｆＮｅｕｒａｌＮｅｔｗｏｌｋ，Ｘｉ＇ｄｉａｎＵｎｉｖｅｒｓｉｔｙ，Ｘｉａｎ７１００７１，Ｃｈｉｎａ）Ａ...     

A Practical Radial Basic Function Network and Its Applications

1 .INTRODUCTIONRecently ,artificialneuralnetworks (ANNs)havebe comeimportanttoolsinmanyfields,suchasstatisti caldataanalysis,patternrecognition ,signalprocess ing,automaticcontrol,forecastingandartificialin telligence.Thebasicproblemintheabovefieldsisfunctionapproximation .Atpresent,therearetwomainmethodsforfunctionestimation ,oneisthetra ditionalmethodnamed“parametricapproach”or“model basedapproach”andtheotherisANNmethodnamed“nonparametricapproach”foritsmanygeneralproperties.With…     

神经网络在预测中的一些应用研究

预测是一个很难的研究课题,近来利用新的科学理论探索新的预测方法是预测界人士进行预测课题研究的一个重要方面．其中人工神经网络在预测领域中的应用研究进展最快,它包括研究神经网络预测方法和神经网络在预测过程中作为辅助工具．本文介绍几种主要的神经网络预测方法及其应用和利用神经网络确定ＡＲＭＡ 模型的结构．     

基于季节调整和BP神经网络的月度负荷预测

月度电力负荷序列中离群值及节假日因素会影响月度负荷预测的准确性.为此,提出了基于季节调整方法和BP神经网络的月度电力负荷组合预测模型.首先,利用季节调整方法对原始负荷序列进行预处理,消除离群值和春节假日的影响;然后用BP神经网络对回归残差序列建模预测得到预测结果或对季节调整后序列和季节成分序列分别建模预测,并对分量预测结果重构后得到最终预测结果的方法.通过实例对预测效果进行检验,结果表明提出的预测方法的预测表现要优于BP神经网络,SARIMA,支持向量机等模型,可以获得更高的预测精度.     

基于支持向量机的时间序列预测模型分析与应用

阐述了支持向量机在时间序列预测中应用的理论基础,给出了时间序列预测分析的基本框架。将支持向量机预测模型应用于某型航空发动机的滑油金属含量监测中,并与递归神经网络预测器进行了比较。得出支持向量机由于采用了新型的结构风险最小化准则表现出优秀的推广能力,可预测区间较长且具有较高的准确度,而递归神经网络模型在中、短期预测中与支持向量机相差不大,在较长区间预测中效果较差的结论。     

FORECASTING AUTOMOBILE WARRANTY PERFORMANCE IN PRESENCE OF ‘MATURING DATA＇ PHENOMENA USING MULTILAYER PERCEPTRON NEURAL NETWORK

Automobile companies that spend billions of dollars annually towards warranty cost, give high priority to warranty reduction programs. Forecasting of automobile warranty performance plays an important role towards these efforts. The forecasting process involves prediction of not only the specific months-in-service (MIS) warranty performance at certain future time, but also at future MIS values. However, ‘maturing data‘ (also called warranty growth) phenomena that causes warranty performance at specific MIS values to change with time, makes such a forecasting task challenging. Although warranty forecasting methods such as log-log plots and dynamic linear models appear in literature, there is a need for applications addressing the well recognized issue of ‘maturing data‘. In this paper we use an artificial neural network for the forecasting of warranty performance in presence of ‘maturing data‘ phenomena. The network parameters are optimized by minimizing the training and testing errors using response surface methodology. This application shows the effectiveness of neural networks in the forecasting of automobile warranty performance in the presence of the ‘maturing data‘ phenomena.     

前馈网络泛化性能改进的目的规划方法研究

以目的规划模型为基础，将前馈网络准则函数改进、网络灵敏度降低、先验知识运用有机地结合在一起，提出了前馈网络泛化性能改进的目的规划方法．文中给出了该方法的数学模型、求解方法以及算例     

基于小波网络的非线性组合预测方法研究

提出了一种基于小波网络的非线性组合预测新方法，以克服线性组合预测方法在解决非平衡时间序列组合建模问题所遇到的困难和存在的不足，并给出了相应的学习算法求解小波函数线性组合的尺度和时延参数以及神经网络权值。理论分析和大量的应用实例表明：本方法具有很强的泛化能力与自适应数据和函数变化的能力，在处理诸如经济时间序列这种具有一定程度不确定性的非线性系统的组合建模和预测方面有较高的应用价值。     

参数自适应的长期IFTS预测算法

针对模糊时间序列预测理论多局限于短期时间范围预测以及对不确定数据集模糊变化趋势描述和论域区间划分研究不足的问题,构建了参数自适应的长期直觉模糊时间序列预测模型。新模型通过引入滑动窗口机制和参数自适应的直觉模糊C均值聚类算法优化论域区间划分,利用矢量预测技术解决时间序列长期范围预测误差积累的问题,有效地提高了复杂环境下时间序列长期趋势预测的精度,扩展了直觉模糊时间序列预测理论的应用范围。最后,通过典型实例验证了该方法的有效性和优越性。     

前馈网络目的规划算法及其应用研究

提出了前馈网络目的规划算法。与通常ＢＰ算法相比,该方法进行了三个方面的改进：（１）准则函数的改进;（２）网络灵敏度的降低;（３）领域先验知识的运用。理论分析及大气中ＳＯ２浓度预测应用研究表明该方法有效地改善了前馈网络泛化性能,提高了预报精度。     

精英遗传改进的非线性灰色神经网络算子与军费开支多目标组合预测应用

军费开支属于复杂经济系统下具有宏观经济特征的一类非线性时间序列。在多目标组合下的军费开支预测问题研究背景下,提出了一种基于精英遗传算法(elite genetic algorithm,EGA)改进的非线性灰色神经网络计量组合预测模型,给出了总体建模思路与非线性灰色神经网络算子分系统和EGA分系统设计方法,解决了多准则目标优化的NP完全问题,并对模型的预测效果进行比较分析。采集美国27年间(1990-2016年)军费开支时间序列进行实证检验,分析结论认为非线性灰色神经网络算子能够有效提高模型精度,EGA算法在收敛速度与精度上优于标准遗传算法,采用所建立的预测模型进行军费开支预测精度更高,效果更好。     

复杂系统态势评估模型及其本体论实现方法

针对复杂系统的在线态势评估问题，依据信息融合理论，提出了一种新的网络在线态势评估模型。它采用定性与定量、局部与综合相结合的评估策略，即首先进行快速定性评估，一旦系统异常，立即启动集成神经网络组对来自系统多侧面的故障特征信息进行定量分析和分类，最后两级D-S证据推理模型在各自的融合中心实现对各子网络的融合，提高了评估的精度和可靠性。基于本体论成功地实现了该模型，在丰满水电数字仿真系统的成功应用验证了模型的有效性和实际应用价值。