基于核主成分的神经网络集合预报建模研究

研究台风预报建模,对提高准确性有重大意义.针对台风强度的非线性、时变性特点,采用神经网络与遗传算法相结合的方法建立一种新的台风强度客观预报模型.在预报模型的输入计算上,考虑台风强度前期预报因子数较多以及因子的非线性性质,先用逐步回归筛选出预报因子,再采用核主成分分析方法在被剔除的因子中提取包含了原数据较多信息的两个核主成分与用逐步回归选入的因子一起作为预报模型的输入数据.以南中国海海域1980-2008年6、7、8、9月各个月的台风强度为预报研究的对象,分别进行仿真试验,试验结果表明,神经网络集合预报模型的预报结果符合实际应用的要求,且预报效果较好,其预报平均绝对误差明显小于同等条件下的逐步回归预报方法的结果.     

面向卫星云图及深度学习的台风等级分类

台风是最严重的自然灾害之一,做好台风等级分类预测是防灾减灾的关键。针对现有卫星云图特征提取复杂、识别率低等问题,基于卷积神经网络框架,以北太平洋1978—2016年近1 000多个台风过程的卫星云图作为样本,提出改进深度学习模型Typhoon-CNNs。该框架采用循环卷积策略增强模型表征力,使用十折交叉验证引入信息熵、Dropout置零率以优化模型灵敏度及防止过拟合。通过800样本测试集对模型Typhoon-CNNs进行验证,实验结果表明,本文算法的分类精度达到92.5%,台风和超强台风2个等级的预测正确率达到99%,优于传统分类方法。最后对模型提取的特征图进一步分析,模型能够准确识别台风眼和螺旋云带,从而证明Typhoon-CNNs对台风等级分类的可行性。     

深圳地区日极值气温的降尺度研究

气象与人类日常生活的关系十分密切,气象预报一直是人类社会高度关注的问题。随着经济的发展和社会的进步,人类对天气预报的准确性提出了越来越高的要求,迫切希望实现气象要素精细化预报。获取详细准确的区域气象资料是实现气象精细化预报的首要条件,全球大气环流模式是目前预估大尺度未来全球气候变化最重要的模式,能较好地模拟出大尺度的平均特征。但是模式预报输出的空间分辨率较低,无法获取精细的区域气象资料,很难对区域天气情景变化做出详细的预测,而降尺度方法可用于弥补这方面的缺陷。文章的研究工作主要是利用统计降尺度的多元线性回归方法和BP神经网络方法对深圳地区近十年的日最低温度和最高温度进行降尺度分析研究。采用的数据是美国国家环境预报中心/美国国家大气研究中心提供的FNL全球分析资料和深圳国家基本观测站——竹子林站的实际观测数据,重点研究了基于BP神经网络方法和多元线性回归方法的统计降尺度模型的设计与实现过程,并对两种方法的结果进行了比较,为区域站点的统计降尺度应用提供了设计方法和参考。     

模型输入对模糊神经网络预报模型的影响研究

为了探索模型输入对模糊神经网络预报模型预测性能的影响,研究了通过减少预报模型自变量组合的复共线性影响,并结合相似系数计算分析方法建立了一种新的模糊神经网络预报模型。以气象学科的逐日降水预报作为研究对象,利用这种新的模糊神经网络预报模型进行了实际预报试验,并与常规的模糊神经网络预报方法,中国气象局T213数值预报模式以及逐步回归预报方法的预报结果进行了对比分析。结果表明,这种基于条件数和相似系数计算的模糊神经网络预报新方法对49天降水的独立样本预报平均绝对误差为7.33 mm,预报误差比模糊神经网络预报模型下降了5.9%,比传统的逐步回归方法下降了14.9%,比中国气象局T213数值预报模式的预报结果下降了13.4%。显示了很好的应用前景。     

波浪数值预报可视化系统的设计与实现

针对全球及中国近海海域,采用第三代海浪数值模式WAVEWATCHIII建立波浪数值预报模型,由此计算得到研究区域的波要素。基于预报模型计算结果建立波浪数值预报可视化系统,采用MSchart控件实现了全球及中国近海海面风场输入口功能和单点波要素静态图显示功能,同时依据提供的全球及中国近海海面风场预报7天全球及中国近海海浪场,采用GRADS软件实现全球及中国近海预报波浪场动态可视化的功能。     

基于Seq2Seq和Attention的时序卫星云图台风等级预测

台风预测可为台风预警预报提供先验信息,辅助相关部门进行科学决策,以减少灾害损失。利用时间序列台风卫星云图,提出一种新的台风等级预测模型SeqTyphoon,将注意力机制和序列到序列引入模型预测未来时刻台风图像,然后利用卷积神经网络对预测的台风图像进行台风等级预测。通过日本气象厅发布的1981—2017年3万多张时序台风卫星云图,构建了训练集、验证集和测试集,分别对应29 519、3 804、1 995张台风图像。针对SeqTyphoon模型,分别进行了台风云图的不同时间间隔、不同预测时长及不同空间分辨率对台风图像预测精度影响的对比实验。实验结果表明,台风云图均为32像素×32像素,时间间隔为6h比时间间隔为12h的训练集和验证集的均方根误差分别降低5.41%、5.72%,前者训练集的均方根误差达到0.092 2,验证集为0.095 4,前者台风等级预测准确率为后者的2倍;台风云图为32像素×32像素,时间间隔为6h时,预测未来6～48h的台风图像,训练集和验证集的均方根误差均递增,台风等级预测准确率递减;时间间隔为6h,图像为64像素×64像素的训练集的均方根误差为0.089 6,验证集为0.091 1,台风等级预测总体准确率为83.2%。综上,影响台风图像的最主要因素是相邻台风云图的时间间隔,其次是预测时长与空间分辨率大小。     

基于混合卷积神经网络算法的风场预测研究

在农业生产中,准确的风速预报对农作物安全防范有着至关重要的作用。针对云南地区的高海拔和多山,基于卷积神经网络框架,提出了卷积长短时序分析神经网络-卷积门控循环单元神经网络(ConvLSTM-ConvGRU)混合风速预测模型。通过神经网络框架的改进,有效的提高了模型对风场空间特征的提取。利用美国国家环境预报中心(NCEP)提供的再分析风速数据集,使用ConvLSTM、ConvGRU、ConvLSTM-ConvGRU混合模型分别对云南地区的风速进行。实验结果表明：ConvLSTM-ConvGRU混合风速预测模型能够有效对云南地区风场进行预测,相较于另外两个模型提高了预测准确度。     

BP神经网络在地震预测中的应用研究

为了对未来地震趋势进行预测,提出一种BP神经网络预测方法。利用Matlab建立BP神经网络模型,并以滇西南地震数据为学习样本对网络模型进行了训练和仿真测试研究。研究结果表明,利用BP神经网络模型预报的地震震级与实际震级误差在8%以下,说明所建模型具有较好的适应性和预报精度。该方法对地震震级的预测具有一定的指导作用和参考价值。     

基于Jena的台风灾害领域本体模型推理

针对传统台风灾害预测以统计学方法为主,而缺乏语义驱动和智能推理的问题,提出一种基于Jena的台风灾害领域本体模型推理机制。首先,在分析台风灾害的影响因素和演变历史的基础上,采用网络本体语言(OWL)建立台风灾害领域本体模型; 其次,利用Jena推理引擎和自定义规则对台风灾害本体模型进行推理,挖掘被隐藏的台风灾害影响因素或灾害链信息; 最后,构建了本体驱动的台风灾害专家系统(Onto-TDES)。实验结果证明,该方案能初步解决传统台风灾害预测缺乏语义驱动和智能推理的问题,提高台风灾害管理与预测的智能化水平。     

同化FY-3C/MWHTS观测资料反演的海面气压场对台风数值预报的影响

由中国风云三号C星（FY-3C）搭载的微波温湿探测仪（MWHTS）的亮温观测资料能够实时反演得到高分辨率、高精度的海面气压场。基于三维变分同化方法将FY-3C/MWHTS观测资料反演的海面气压场同化进入中尺度天气研究与预报（Weather Research and Forecasting, WRF）模式,以台风“Maria”和“Noru”为例,通过控制实验和同化试验的对比分析,探讨了同化反演的海面气压场对台风数值预报的影响。初始化敏感性试验结果表明,同化海面气压场使初始时刻台风中心气压与位置更接近实况,并且调整了台风初始温度场和风场的结构和分布。台风的数值预报结果表明:同化反演的海面气压场能够改进台风的路径和强度预报精度。     

Forecasting the output of integrated circuit industry using genetic algorithm based multivariable grey optimization models

Accurate and immediate forecast in the short product life cycle of semiconductor market is difficult, but important. This paper proposed a grey model with factor analysis techniques to deal with the multi-factor forecasting problems. In the grey modeling, the use of genetic algorithm has the ability to search global optimum solution to construct two improved multivariable grey forecasting models that are AGAGM(1,N) and GAGM(1,N). These two models are applied for forecasting Taiwanese integrated circuit output. The results of the factor analysis show that the major factors of Taiwan’s integrated circuit output comprise R&D intensity, foreign investment, index of industrial production, trade specialization coefficient, and intra-industry trade coefficient. The improved multivariable grey forecasting models are found to be feasible and effective.     

Weather analysis using ensemble of connectionist learning paradigms

This paper presents a comparative analysis of different connectionist and statistical models for forecasting the weather of Vancouver, Canada. For developing the models, one year's data comprising of daily temperature and wind speed were used. A multi-layered perceptron network (MLPN) and an Elman recurrent neural network (ERNN) were trained using the one-step-secant and Levenberg–Marquardt algorithm. Radial basis function network (RBFN) was employed as an alternative to examine its applicability for weather forecasting. To ensure the effectiveness of neurocomputing techniques, the connectionist models were trained and tested using different datasets. Moreover, ensembles of the neural networks were generated by combining the MLPN, ERNN and RBFN using arithmetic mean and weighted average methods. Subsequently, performance of the connectionist models and their ensembles were compared with a well-established statistical technique. Experimental results obtained have shown RBFN produced the most accurate forecast model compared to ERNN and MLPN. Overall, the proposed ensemble approach produced the most accurate forecast, while the statistical model was relatively less accurate for the weather forecasting problem considered.     

Combining high frequency data with non-linear models for forecasting energy market volatility

The popularity of realized measures and various linear models for volatility forecasting has been the focus of attention in the literature addressing energy markets’ price variability over the past decade. However, there are no studies to help practitioners achieve optimal forecasting accuracy by guiding them to a specific estimator and model. This paper contributes to this literature in two ways. First, to capture the complex patterns hidden in linear models commonly used to forecast realized volatility, we propose a novel framework that couples realized measures with generalized regression based on artificial neural networks. Our second contribution is to comprehensively evaluate multiple-step-ahead volatility forecasts of energy markets using several popular high frequency measures and forecasting models. We compare forecasting performance across models and across realized measures of crude oil, heating oil, and natural gas volatility during three qualitatively distinct periods: the pre-crisis period, the 2008 global financial crisis, and the post-crisis period. We conclude that the newly proposed approach yields both statistical and economic gains, while reducing the tendency to over-predict volatility uniformly during all the tested periods. In addition, the proposed methodology is robust to a substantial structural break induced by the recent financial crisis. Our analysis favors median realized volatility because it delivers the best performance and is a computationally simple alternative for practitioners.     

神经网络在卷烟销售量预测的应用研究

研究卷烟销售预测准确问题,卷烟销售量具有季节性和周期性动态变化规律,并受经济、人口等因素的影响,使系统存在明显的非线性特征,波动范围比较大,传统线性预测模型难以准确预测。为了提高卷销售预测精度,提出一种能够反映卷烟销售量变化规律的Elman神经网络的卷烟销售预测模型。首先采用逐步拓阶方法确定卷烟销售量的最佳滞后阶数,然后利用最佳滞后阶数最对卷烟销售数据进行重组,并输入Elman神经网络学习,利用Elman神经网络的动态和反馈特点对卷烟销售量进行预测。将建立的模型应用于云南某烟草公司某种卷烟销售的预测,结果表明,Elman神经网络模型有效提高了卷烟销售预测精度,降低了预测误差,为烟草行业销售管理预测提供科学依据。     

Nonlinearity in Forecasting of High-Frequency Stock Returns

Using high-frequency S&P 500 data, we examined intraday efficiency by comparing the ability of several nonlinear models to forecast returns for horizons of 5, 10, 30 and 60?min. Taking into account fat tails and volatility dynamics, we compared the forecasting performance of simple random walk and autoregressive models with Markov switching, artificial neural network and support vector machine regression models in terms of both statistical and economic criteria. Our empirical results for out-of-sample forecasts for high and low volatility samples at different time periods provide weak evidence of intraday predictability in terms of statistical criteria, but corroborate the superiority of nonlinear model predictability using economic criteria such as trading rule profitability and value-at-risk calculations.     

A hybrid ANFIS model based on empirical mode decomposition for stock time series forecasting

Time series forecasting is an important and widely popular topic in the research of system modeling, and stock index forecasting is an important issue in time series forecasting. Accurate stock price forecasting is a challenging task in predicting financial time series. Time series methods have been applied successfully to forecasting models in many domains, including the stock market. Unfortunately, there are 3 major drawbacks of using time series methods for the stock market: (1) some models can not be applied to datasets that do not follow statistical assumptions; (2) most time series models that use stock data with a significant amount of noise involutedly (caused by changes in market conditions and environments) have worse forecasting performance; and (3) the rules that are mined from artificial neural networks (ANNs) are not easily understandable.To address these problems and improve the forecasting performance of time series models, this paper proposes a hybrid time series adaptive network-based fuzzy inference system (ANFIS) model that is centered around empirical mode decomposition (EMD) to forecast stock prices in the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) and Hang Seng Stock Index (HSI). To measure its forecasting performance, the proposed model is compared with Chen's model, Yu's model, the autoregressive (AR) model, the ANFIS model, and the support vector regression (SVR) model. The results show that our model is superior to the other models, based on root mean squared error (RMSE) values.     

基于SVM的先分类再回归方法及其在产量预测中的应用

针对非线性问题,提出了基于支持向量机分类基础的先分类、再回归的预测方法。根据实际需要和专业知识先将样本集进行分类,判别测试样本的类别后,再利用回归算法预测测试样本的值。利用这一算法进行粮食产量预测,并与其他模型预测结果相比,准确度远优于其他产量预测方法。实验说明：先分类、再回归得到的拟合值比直接利用回归得到的拟合值要精确。     

沿海风暴潮预警预报与灾情评估系统研究

台风是一种严重的自然灾害,引起的风暴增水会对沿海城市产生较大危害,快速准确的风暴潮预警预报及灾情评估是当前研究的热点和难点问题。结合实际业务需求,借鉴类似系统的开发模式,设计提出适用于沿海风暴潮预警预报与灾情评估系统的总体架构,开发预报与评估系统前端主要功能和服务器端相关功能组件,并在广东省进行应用部署。应用结果表明,预报和评估系统实用性和可操作性强,可为沿海防汛部门在台风期的防台会商决策提供强有力的技术支撑。     

遗传优化支持向量机在电力负荷预测中的应用

研究电力负荷准确预测问题,电力负荷与影响因子之间呈现复杂非线性关系,传统预测方法无法刻画其变化规律,预测精度低。为提高电力负荷预测精度,提出一种采用遗传优化支持向量机的电力负荷预测模型。采用最小二乘支持向量机的非线性逼近能力去描述电力负荷与影响因子间的复杂非线性关系,并采用自适应遗传算法优化最小二乘支持向量机的参数。采用某省1990～2008年电力负荷数据仿真测试,结果表明,遗传优化支持向量机提高了电力负荷的预测精度,预测平均误差低于其它对比模型,电力负荷预测提供了一种新的研究思路和途径。     

Forecasting output of integrated circuit industry by support vector regression models with marriage honey-bees optimization algorithms

Integrated circuit (IC) is a vital component of most electronic commodity. IC manufacturing in Taiwan is booming, with revenues from the ICs industry having grown significantly in the recent years. Given the nature of technology, capital intensity and high value-added, accurate forecasting of IC the industry output can improve the competitivity of IC cooperation. Support vector regression (SVR) is an emerging forecasting scheme that has been successfully adopted in many time-series forecasting areas. Additionally, the data preprocessing procedure and the determination of SVR parameters significantly impact the forecasting accuracy of SVR models. Thus, this work develops a support vector regression model with scaling preprocessing and marriage in honey-bee optimization (SVRSMBO) model to accurately forecast IC industry output. The scaling preprocessing procedure is utilized to lower the fluctuation of input data, and the marriage in honey-bees optimization (MBO) algorithm is adopted to determine the three parameters of the SVR model. Numerical data collected from the previous literature are used to demonstrate the performance of the proposed SVRSMBO model. Simulation results indicate that the SVRSMBO model outperforms other forecasting models. Hence, the SVRSMBO model is a promising means of forecasting IC industry output.