CHINA·77779193永利(集团)有限公司-Official website

统计预测与统计计算系列报告

时间：2021年6月17日（周四）上午9：00-12：00

地点：教二楼913

报告人姓名：康雁飞

单位：北京航空航天大学经济与管理学院副教授

报告题目：Feature-based time series forecasting

报告摘要：The explosion of time series data in recent years has brought a flourish of new time series forecasting methods. Given that the statistical features of time series have an impact on their forecasting accuracies, we aim to examine feature-based time series forecast combination from a number of perspectives: 1) Generation of a diverse training dataset. we propose GRATIS to generate sets of diverse time series using MAR (mixture autoregressive) models, which is used as an evaluation tool for both feature-based time series model selection and model averaging; 2) Automation of feature extraction. We automate the extraction of features by time series imaging, and investigate forecast model averaging based on the image features; and 3) Forecast with forecasts.  We suggest an alternative is to change our focus from the historical data to the produced forecasts to extract features. We calculate the diversity of a pool of models based on the corresponding forecasts as a decisive feature and use meta-learning to construct diversity-based forecast combination. We evaluate the proposed approaches using a rich collection of real data and show that they result in competitive forecasting accuracy.

报告人姓名：王晓倩

单位：北京航空航天大学经济与管理学院在读博士生

报告题目：The uncertainty estimation of feature-based forecast combinations

报告摘要：Forecasting is an indispensable element of operational research (OR) and an important aid to planning. The accurate estimation of the forecast uncertainty facilitates several operations management activities, predominantly in supporting decisions in inventory and supply chain management and effectively setting safety stocks. In this paper, we introduce a feature-based framework, which links the relationship between time series features and the interval forecasting performance into providing reliable interval forecasts. We propose an optimal threshold ratio searching algorithm and a new weight determination mechanism for selecting an appropriate subset of models and assigning combination weights for each time series tailored to the observed features. We evaluate our approach using a large set of time series from the M4 competition. Our experiments show that our approach significantly outperforms a wide range of benchmark models, both in terms of point forecasts as well as prediction intervals.

报告人姓名：李莉

单位: 北京航空航天大学经济与管理学院在读博士生

报告题目：Feature-based Bayesian forecasting model averaging

报告摘要：In this work, we propose a novel framework for density forecast combination by constructing time-varying weights based on time-series features, which is called FEature-based BAyesian forecasting Model Averaging (FEBAMA). To this end, our approach has better interpretability compared to other black-box forecasting combination schemes. We apply our framework to stock market data and M3 competition data. Based on our structure, a simple maximum-a-posteriori scheme outperforms benchmark methods, and Bayesian variable selection can further enhance the accuracy for both point and density forecasts.

报告人姓名：李丰

单位：中央财经大学统计与数学学院副教授

报告题目：Highly-scalable distributed modelling and forecasting with dependent data

报告摘要：We propose a highly scalable framework and its computational implementation for distributed statistical computing of dependent data. We aim to address the overwhelmed demand for statistical models in a distributed environment with massive real-time data. Our solution is applicable for the new normal of distributed computing. Although different statistical computing techniques have their solutions for different types of limited amounts of data in the traditional manner, they are not readily applicable in the distributed system. Efficient distributed statistical computing implies real-time statistical modeling, estimation, validation, and predictions, as well as data-centroid decisions. It motivates the move-model-to-data philosophy and presents a unified framework for distributed statistical computing of time series models and realizes the concept of modern distributed statistical computing by studying the following four aspects: (1) an unified framework for distributed statistical computing of time series data (2) the distributed forecasting for ultra-long time series (3) distributed Bayesian sampling, and (4) model selection and decision-making for real-time distributed statistical models.

姓名：邱明悦

单位：首都师范大学在读硕士生

报告题目：区间删失数据比例风险模型下的贝叶斯变量选择

报告摘要：变量选择是模型构建中的一个重要问题，在生存分析的文献中受到了广泛的关注。然而，这一领域的可用方法主要集中在具有右删失结构的生存时间数据上。本文研究了区间删失数据背景下，比例风险模型的贝叶斯变量选择问题，将Balasso（Bayesian Adaptive Least Absolute Shrinkage and Selection Operator）应用在该问题上，使得变量选择和参数估计可以同时进行。构造了有效的MCMC（Markov chain Monte Carlo）方法以实现后验分布的抽样和推理。数据模拟结果验证了该方法的有效性。最后将提出的方法应用于尼日利亚人口与健康调查中儿童死亡率数据，根据结果对影响儿童死亡率的变量进行选择。

姓名：赵美言

单位：首都师范大学在读硕士生

报告题目： 海河流域北部干旱预测研究

报告摘要：本文基于海河流域北部4种时间尺度的（3、6、9、12个月）标准降水指数（Standard Precipitation Index，SPI）序列进行了干旱点、区间预测。点预测模型为随机森林（Random Forest，RF）、支持向量回归（Support Vector Regression，SVR）、弹性反向传播神经网（Resilient BackPropagation Artificial Neural Network，Rprop-ANN）。点预测评价指标为：平均相对误差（Mean Relative Error，MRE）、Kendalll秩相关系数（Kendall）。区间预测方法有三种。第一种方法基于分位数回归思想，分别为分位数回归森林（Quantile RegressionForest，QRF）、支持向量分位数回归（Support Vector Quantile Regression，SVQR）、神经网络分位数回归（Quantile Regression Neural Network，QRNN）。其余两种方法为本文扩展、改进的袋外-核密度估计（Out Of Bag-Kernel Density Estimation，OOB-KDE）、刀切法一致性-核密度估计（Split Conform-Kernel Density Estimation，SC-KDE）。区间预测评价指标为预测区间覆盖率（Prediction Interval Coverage Probability，PICP）、预测区间归一化宽度（Prediction Interval Normalized Average Width，PINAW）。研究得到如下结论：三种点预测模型中，RF预测性能最差，SVR、Rprop-ANN表现优异；本文改进的SC-KDE方法有效提高了预测区间可靠性；本文改善、扩展的OOB-KDE、SC-KDE方法性能优于基于分位数回归思想的区间预测方法。

主办单位： 77779193永利官网 交叉科学研究院 北京应用统计学会

联系人：胡涛