基于主成分分析的人工智能台风路径预报模型

利用主成分分析可以从具有随机噪声干扰的气象场提取主要信号特征,排除随机干扰的能力,论文以1980～2010年共31年6～9月西行进入南海海域的台风样本为基础,综合考虑台风移动路径的气候持续因子和数值预报产品动力预报因子,采用主成分分析的特征提取与逐步回归计算相结合的预报因子信息数据挖掘技术,以进化计算的遗传算法,生成期望输出相同的多个神经网络个体,建立了一种新的非线性人工智能集合预报模型,进行了分月台风路径预报模型的预报建模研究。在预报建模样本、独立预报样本相同的情况下,分别采用人工智能集合预报方法和气候持续法进行了预报试验,试验对比结果表明,前者较后者在6、7、8和9月份台风路径预报中,平均绝对误差分别下降了7.4%、4.8%、12.4%、17.0%。另外,论文进一步在初选预报因子和样本个例相同的情况下,通过比较新模型与直接采用主成分分析方法选因子并分别运用逐步回归和遗传—神经网络集合预报模型进行计算的预报精度差异表明,前者具有更高的预报精度,其原因是该方法挖掘利用了全部备选预报因子的有用预报信息,而且遗传—神经网络集合预报模型的是由多个神经网络个体预报结果合成,集合模型的各个神经网络个体的网络结构,是通过遗传算法的优化计算确定的,因此,该集合预报模型的泛化能力显著提高,在实际天气预报中具有较好的实用性和推广价值。     

DOWNSCALING FORECAST OF MONTHLY PRECIPITATION OVER GUANGXI BASED ON BP NEURAL NETWORK MODEL

Variables fields such as enstrophy, meridional-wind and zonal-wind variables are derived from monthly 500 hPa geopotential height anomalous fields. In this work, we select original predictors from monthly 500-hPa geopotential height anomalous fields and their variables in June of 1958 - 2001, and determine comprehensive predictors by conducting empirical orthogonal function (EOF) respectively with the original predictors. A downscaling forecast model based on the back propagation (BP) neural network is built by use of the comprehensive predictors to predict the monthly precipitation in June over Guangxi with the monthly dynamic extended range forecast products. For comparison, we also build another BP neural network model with the same predictands by using the former comprehensive predictors selected from 500-hPa geopotential height anomalous fields in May to December of 1957 - 2000 and January to April of 1958 - 2001. The two models are tested and results show that the precision of superposition of the downscaling model is better than that of the one based on former comprehensive predictors, but the prediction accuracy of the downscaling model depends on the output of monthly dynamic extended range forecast.     

STUDY ON MIXED MODEL OF NEURAL NETWORK FOR FARMLAND FLOOD/DROUGHT PREDICTION

The paper concerns a flood/drought prediction model involving the continuation of time seriesof a predictand and the physical factors influencing the change of predictand.Attempt is made toconstruct the model by the neural network scheme for the nonlinear mapping relation based onmulti-input and single output.The model is found of steadily higher predictive accuracy by testingthe output from one and multiple stepwise predictions against observations and comparing theresults to those from a traditional statistical model.     

STUDY ON MIXED MODEL OF NEURAL NETWORK FOR FARMLAND FLOOD/DROUGHT PREDICTION*

The paper concerns a flood/drought prediction model involving the continuation of time series of a predictand and the physical factors influencing the change of predictand.Attempt is made to construct the model by the neural network scheme for the nonlinear mapping relation based on multi-input and single output.The model is found of steadily higher predictive accuracy by testing the output from one and multiple stepwise predictions against observations and comparing the results to those from a traditional statistical model.     

短期气候可预报期限的时空分布

在非线性误差增长理论的基础上,研究了位势高度场与温度场月和季节时间尺度可预报期限的时空分布特征,结果表明:(1)在500 hPa位势高度场上,年平均月和季节尺度可预报期限的空间分布都存在明显的南北经向性差异,其中在热带地区月和季节尺度可预报期限都为最大,月尺度可预报期限都在6个月以上, 其中最高值超过了9个月,而季节尺度可预报期限基本上都在8个月以上,其中最高值超过了11个月;从热带地区到南北半球中纬度地区,随着纬度的升高,月和季节尺度可预报期限也迅速减少。(2)在500 hPa位势高度场上,月和季节尺度可预报期限的空间分布都有明显的季节变化。冬季月和季节尺度可预报期限除了在热带地区较大外,在北太平洋和邻近的北美西北部地区、北大西洋地区以及南极地区,冬季月和季节尺度可预报期限也相对周围地区较高。夏季除了北非和西亚地区月和季节尺度可预报期明显大于冬季以外,大部分地区月和季节尺度可预报期限比冬季明显减少。(3)500 hPa温度场月和季节尺度可预报期限的空间分布以及随季节的变化特征基本上与高度场相同,只是在热带大部分地区,高度场相对温度场来说月和季节尺度可预报性更高,更适合用来作长期预报。     

A prediction scheme with genetic neural network and Isomap algorithm for tropical cyclone intensity change over western North Pacific

A western North Pacific tropical cyclone (TC) intensity prediction scheme has been developed based on climatology and persistence (CLIPER) factors as potential predictors and using genetic neural network (GNN) model. TC samples during June–October spanning 2001–2010 are used for model development. The GNN model input is constructed from potential predictors by employing both a stepwise regression method and an Isometric Mapping (Isomap) algorithm. The Isomap algorithm is capable of finding meaningful low-dimensional architectures hidden in their nonlinear high-dimensional data space and separating the underlying factors. In this scheme, the new developed model, which is termed the GNN-Isomap model, is used for monthly TC intensity prediction at 24- and 48-h lead times. Using identical modeling samples and independent samples, predictions of the GNN-Isomap model are compared with the widely used CLIPER method. By adopting different numbers of nearest neighbors, results of sensitivity experiments show that the mean absolute prediction errors of the independent samples using GNN-Isomap model at 24- and 48-h forecasts are smaller than those using CLIPER method. Positive skills are obtained as compared to the CLIPER method with being above 12 % at 24 h and above 14 % at 48 h. Analyses of the new scheme suggest that the useful linear and nonlinear prediction information of the full pool of potential predictors is excavated in terms of the stepwise regression method and the Isomap algorithm. Moreover, the GNN is built by integrating multiple individual neural networks with the same expected output and network architecture is optimized by an evolutionary genetic algorithm, so the generalization capacity of the GNN-Isomap model is significantly enhanced, indicating a potentially better operational weather prediction.     

基于遗传算法的神经网络短期气候预测模型

用遗传算法优化神经网络的连接权和网络结构,并在遗传进化过程中采取保留最佳个体的方法,进行短期气候预测建模研究。该方法克服了由于神经网络初始权值的随机性和网络结构确定过程中所带来的网络振荡,以及网络极易陷入局部解问题。作为应用实例,以广西全区4月份平均降水作为预报量及前期500hPa月平均高度场,海温场高相关区作为预报因子,建立基于遗传算法的神经网络短期气候预测模型。将这种方法与传统的逐步回归方法作对比分析,结果表明,该方法具有预报精度高,稳定性好的特点。     

基于均生函数的风电场风速短临预报模型

风电场风速短临预报是风电预报业务的重要环节之一.选取2011年1、4、7和10月河北及内蒙古两地的两座测风塔观测资料,综合采用均生函数、灰色关联度和多元回归等多种统计预报方法,建立了一套分站点的、15 min动态滚动的未来0～4 h风速短临预报模型.实际对比分析表明,所建模型比传统ARMA模型的月平均预报误差减少了2.2％～10.8％,在实际中具有重要的应用价值.     

一种新的线性回归模型及其应用示例

回归分析是统计分析中常用的方法之一。传统的回归模型不具备全域分析能力,而变量场之间的关系多采用SVD（Singular Value Decomposition）进行分析,与传统的回归分析有所脱节。更为广义的线性回归模型是传统线性回归模型的延拓,在标量情况下,该模型可转化为传统线性回归模型。该模型的基本特征包含乘法不可互易性、等价于传统线性回归（因子项为标量时）、可分析性、延拓性、降维特征及容错性等。该模型解决了传统的线性回归模型不具备全域分析能力及模型表达能力受限于模型维数的现实问题。本文采用了NCEP（National Centers for Environmental Prediction）降水、高度场、风场月平均资料及国家气候中心西太平洋副热带高压指数资料,利用该模型和传统回归方案进行对比分析,分析结果表明,该模型具有一定的实用参考价值。     

定性和定量长期预报模型的综合分析

根据相同的 50 0 h Pa和海温场预报因子 ,利用神经网络灵活可变的拓朴结构 ,分别构造了定性和定量的降水长期预报模型。并在同等条件下 ,建立了逐步回归预报方程。通过对比分析表明 ,这种定性和定量相结合的神经网络综合预报分析方法 ,是增强预报结果可靠性和稳定性的一种有效途径。该预报建模方法具有比较合理的分析依据 ,值得进一步探索、应用。     

人工神经网络在月降水量预测业务中的研究和应用综述

月降水量的年际变化具有显著的非线性变化特征,预测难度大,历来是重大气象灾害预测的重点难点问题。BP(back propagation)神经网络在月降水量预测业务中的研究和应用中,取得了较好的成果,其中应用较广泛的是PCA-BP神经网络模型、遗传算法优化神经网络、RBF神经网络预测模型、小波神经网络模型、粒子群-神经网络模型等,这些方法也在广西月降水量预测业务中得到很好的应用,对提高月降水量预测能力有较大帮助。因此,有必要对目前神经网络在月降水量预测中的优势和不足进行综述,提出未来研究需要关注的重点关键问题。     

All India summer monsoon rainfall prediction using an artificial neural network

The prediction of Indian summer monsoon rainfall (ISMR) on a seasonal time scales has been attempted by various research groups using different techniques including artificial neural networks. The prediction of ISMR on monthly and seasonal time scales is not only scientifically challenging but is also important for planning and devising agricultural strategies. This article describes the artificial neural network (ANN) technique with error- back-propagation algorithm to provide prediction (hindcast) of ISMR on monthly and seasonal time scales. The ANN technique is applied to the five time series of June, July, August, September monthly means and seasonal mean (June + July + August + September) rainfall from 1871 to 1994 based on Parthasarathy data set. The previous five years values from all the five time-series were used to train the ANN to predict for the next year. The details of the models used are discussed. Various statistics are calculated to examine the performance of the models and it is found that the models could be used as a forecasting tool on seasonal and monthly time scales. It is observed by various researchers that with the passage of time the relationships between various predictors and Indian monsoon are changing, leading to changes in monsoon predictability. This issue is discussed and it is found that the monsoon system inherently has a decadal scale variation in predictability. Received: 13 March 1999 / Accepted: 31 August 1999     

Nonlinear relation of the Arctic oscillation with the quasi-biennial oscillation

A nonlinear principal component analysis (NLPCA) is applied to a set of monthly mean time series from January 1956 to December 2007 consisting of the Arctic oscillation (AO) index derived from 1,000-hPa geopotential height anomalies poleward of 20°N latitude and the zonal winds observed at seven pressure levels between 10 and 70?hPa in the equatorial stratosphere to investigate the relation of the AO with the quasi-biennial oscillation (QBO). The NLPCA is conducted using a new, compact neural network model. The NLPCA modeling of the dataset of the AO index and QBO winds offers a clear picture of the relation between the two oscillations. In particular, the phase of covariation of the oscillations defined by the two nonlinear principal components of the dataset progresses with a predominant 28.4-month periodicity. This predominant cycle is modulated by an 11-year cycle. The variation of the AO index with the QBO phase also shows that the average AO index is positive when the westerly QBO phase descends past 30?hPa and, conversely, the average AO index is negative when the easterly QBO phase descends past 30?hPa. This relationship is evident during the boreal cold season from November to April but non-existent during the boreal warm season from May to October.     

Monthly prediction of rainfall over India and its homogeneous zones during monsoon season: a supervised principal component regression approach on general circulation model products

A supervised principal component regression (SPCR) technique has been employed on general circulation model (GCM) products for developing a monthly scale deterministic forecast of summer monsoon rainfall (June–July–August–September) for different homogeneous zones and India as a whole. The time series of the monthly observed rainfall as the predictand variable has been used from India Meteorological Department gridded (1°?×?1°) rainfall data. Lead 0 (forecast initialized in the same month) monthly products from GCMs are used as predictors. The sources of these GCMs are International Research Institute for Climate and Society, Columbia University, National Center for Environmental Prediction, and Japan Agency for Marine Earth Science and Technology. The performance of SPCR technique is judged against simple ensemble mean of GCMs (EM) and it is found that over almost all the zones the SPCR model gives better skill than EM in June, August, and September months of monsoon. The SPCR technique is able to capture the year to year observed rainfall variability in terms of sign as well as the magnitude. The independent forecasts of 2007 and 2008 are also analyzed for different monsoon months (Jun–Sep) in homogeneous zones and country. Here, 1982–2006 have been considered as development year or training period. Results of the study suggest that the SPCR model is able to catch the observational rainfall over India as a whole in June, August, and September in 2007 and June, July, and August in 2008.     

基于SSA-MGF的BP神经网络多步预测模型

采用奇异谱分析（Singular Spectrum Analysis，SSA）方法对标准化样本序列进行准周期信号分量重建，将重建序列构造均值生成函数（Mean Generating Function．MGF）延拓矩阵作为输入因子，原样本序列作为输出因子，构建BP神经网络多步预测模型。通过实际建模并与逐步回归等方法进行对比预测试验，结果表明，基于SSA-MGF的BP神经网络多步预测模型预测效果优于其他3种模型，说明SSA的去噪及BP神经网络预报模型对于提高预测准确率是相对有效的，是一种具有较高应用价值的多步预测方法。     

基于智能计算的广西冷湿极端天气定性和定量组合预报方法研究

论文以逐日气温和降水量数据、NCEP/NCAR再分析资料以及预报场资料为基础,将表征冬季低温冷害的冷湿指数作为预报量,先利用随机森林方法进行冬季逐日冷湿极端天气定性判别预报分析,再进一步以粒子群算法为基础的模糊神经网络集成个体生成技术方法,建立一种新的非线性智能计算定量集成预报模型（PSO-FNN）,进行了广西冷湿极端天气定量预报模型的预报建模研究。结果表明,论文提出的这种以不同的智能计算方法构建的定性、定量综合预报分析方法,比较符合极端天气小概率事件的预报特点,其中随机森林算法构建的定性预报模型,对广西冷湿极端天气事件的预报TS评分（Threat Score）为0.77,空报率为0.23,漏报率为0,ETS评分（Equitable Threat Score）为0.41,TSS评分（True Skill Statistic）为0.53。而采用粒子群—模糊神经网络方法构建的极端冷湿指数定量集成预报模型比其他线性和非线性预报模型具有更好的预报精度。其中PSO-FNN集成预报模型在预报建模样本和独立预报样本个例相同的情况下,比回归方法的预报平均绝对误差下降了25%以上,比一般的普通模糊神经网络预报平均绝对误差下降了14.37%。主要原因是因为PSO-FNN集成预报模型通过改进集成个体的预报能力和增强集成个体的种群差异性,提高了集成预报模型的预报精度。因此,该智能计算集成预报模型的泛化能力显著提高,预报结果稳定可靠,为冷湿极端天气客观预报提供了新的预报工具和预报建模方法。     

一种分离时间尺度的统计降尺度模型的建立和应用——以华北汛期降水为例

针对预报量变化中存在受不同物理因子控制的不同时间尺度变率特征, 本文提出了分离时间尺度的统计降尺度模型。应用滤波方法, 将不同尺度的变率分量分开, 在各自对应的时间尺度上利用不同的大尺度气候因子分别建立降尺度模型。华北汛期 (7～8月) 降水具有年际变率和年代际变率, 本文以华北汛期降水为例利用分离时间尺度的统计降尺度模型进行预测研究。采用的预报因子来自海平面气压场、 500 hPa位势高度场、 850 hPa经向风场和海表温度场以及一些已知的大尺度气候指数。利用基于交叉检验的逐步回归法建立模型。结果表明, 年际尺度上, 华北汛期降水与前期6月赤道中东太平洋海温以及同期中国东部的低层经向风密切相关; 年代际尺度上, 在东印度洋—西太平洋暖池海温的作用下, 华北降水与前期6月西南印度洋海平面气压有同步变化关系。年际模型和年代际模型的结果相加得到对总降水量的降尺度结果。1991～2008年的独立检验中, 模型估计的降水和观测降水的相关系数是0.82, 平均均方根误差是14.8%。结合模式的回报资料, 利用降尺度模型对1991～2001年的华北汛期降水进行回报试验。相比于模式直接预测的降水, 降尺度模型预测的结果有明显改进。改进了模式预测中年际变率过小的问题, 与观测降水的相关系数由0.12提高到0.45。     

PREDICTION OF FLOOD SEASON PRECIPITATION IN SOUTHWEST CHINA BASED ON IMPROVED PSO-PLS

In order to achieve the best predictive effect of the Partial Least Squares (PLS) regression model, Particle Swarm Optimization (PSO) algorithm is applied to automatically filter the optimal subset of a set of candidate factors of PLS regression model in this study. An improved version of the Particle Swarm Optimization-Partial Least Squares (PSO-PLS) regression model is applied to the station data of precipitation in Southwest China during flood season. Using the PSO-PLS regression method, the prediction of flood season precipitation in Southwest China has been studied. By introducing the precipitation period series of the mean generating function (MGF) extension as an alternative factor, the MGF improved PSO-PLS regression model was also build up to improve the prediction results. Randomly selected 10%, 20%, 30% of the modeling samples were used as a test trial; random cross validation was conducted on the MGF improved PSO-PLS regression model. The results show that the accuracy of PSO-PLS regression model and the MGF improved PSO-PLS regression model are better than that of the traditional PLS regression model. The training results of the three prediction models with regard to the regional and single station precipitation are considerable, whereas the forecast results indicate that the PSO-PLS regression method and the MGF improved PSO-PLS regression method are much better than the traditional PLS regression method. The MGF improved PSO-PLS regression model has the best forecast performance on precipitation anomaly during the flood season in the southwest of China among three models. The average precipitation (PS score) of 36 stations is 74.7. With the increase of the number of modeling samples, the PS score remained stable. This shows that the PSO algorithm is objective and stable. The MGF improved PSO-PLS regression prediction model is also showed to have good prediction stability and ability.     

A PRELIMINARY STUDY ON THUNDERSTORM FORECAST WITH LS-SVM METHOD

The LS-SVM(Least squares support vector machine) method is presented to set up a model to forecast the occurrence of thunderstorms in the Nanjing area by combining NCEP FNL Operational Global Analysis data on 1.0°×1.0° grids and cloud-to-ground lightning data observed with a lightning location system in Jiangsu province during 2007-2008.A dataset with 642 samples,including 195 thunderstorm samples and 447 non-thunderstorm samples,are randomly divided into two groups,one(having 386 samples) for modeling and the rest for independent verification.The predictors are atmospheric instability parameters which can be obtained from the NCEP data and the predictand is the occurrence of thunderstorms observed by the lightning location system.Preliminary applications to the independent samples for a 6-hour forecast of thunderstorm events show that the prediction correction rate of this model is 78.26%,false alarm rate is 21.74%,and forecasting technical score is 0.61,all better than those from either linear regression or artificial neural network.     

最优化因子处理及加权多重回归模型

因子的优劣是回归分析的关键。故在建立回归方程时,一般除对预报因子进行直线相关普查外,还要进行非线性相关普查。如将原因子x用x~(-1)、x~(1/2)、x~2 、e~x、In x等函数形式进行变换。但由于这些函数形式有限,故不一定能找到最优的表达形式;为此冯耀煌等仅给出了x~a和e~(ax)两种通式,其中a为待