Graph Construction Method for GNN-Based Multivariate Time-Series Forecasting

Multivariate time-series forecasting(MTSF)plays an important role in diverse real-world applications.To achieve better accuracy in MTSF,time-series patterns in each variable and interrelationship patterns between variables should be considered together.Recently,graph neural networks(GNNs)has gained much attention as they can learn both patterns using a graph.For accurate forecasting through GNN,a well-defined graph is required.However,existing GNNs have limitations in reflecting the spectral similarity and time delay between nodes,and consider all nodes with the same weight when constructing graph.In this paper,we propose a novel graph construction method that solves aforementioned limitations.We first calculate the Fourier transform-based spectral similarity and then update this similarity to reflect the time delay.Then,we weight each node according to the number of edge connections to get the final graph and utilize it to train the GNN model.Through experiments on various datasets,we demonstrated that the proposed method enhanced the performance of GNN-based MTSF models,and the proposed forecasting model achieve of up to 18.1%predictive performance improvement over the state-of-the-art model.

A Hybrid Neural Network Model for Marine Dissolved Oxygen Concentrations Time-Series Forecasting Based on Multi-Factor Analysis and a Multi-Model Ensemble

Dissolved oxygen(DO)is an important indicator of aquaculture,and its accurate forecasting can effectively improve the quality of aquatic products.In this paper,a new DO hybrid forecasting model is proposed that includes three stages:multi-factor analysis,adaptive decomposition,and an optimizationbased ensemble.First,considering the complex factors affecting DO,the grey relational(GR)degree method is used to screen out the environmental factors most closely related to DO.The consideration of multiple factors makes model fusion more effective.Second,the series of DO,water temperature,salinity,and oxygen saturation are decomposed adaptively into sub-series by means of the empirical wavelet transform(EWT)method.Then,five benchmark models are utilized to forecast the sub-series of EWT decomposition.The ensemble weights of these five sub-forecasting models are calculated by particle swarm optimization and gravitational search algorithm(PSOGSA).Finally,a multi-factor ensemble model for DO is obtained by weighted allocation.The performance of the proposed model is verified by timeseries data collected by the pacific islands ocean observing system(PacIOOS)from the WQB04 station at Hilo.The evaluation indicators involved in the experiment include the Nash–Sutcliffe efficiency(NSE),Kling–Gupta efficiency(KGE),mean absolute percent error(MAPE),standard deviation of error(SDE),and coefficient of determination(R^(2)).Example analysis demonstrates that:①The proposed model can obtain excellent DO forecasting results;②the proposed model is superior to other comparison models;and③the forecasting model can be used to analyze the trend of DO and enable managers to make better management decisions.

Weighted Time-Variant Slide Fuzzy Time-Series Models for Short-Term Load Forecasting

Short-term load forecast plays an important role in the day-to-day operation and scheduling of generating units. Season and temperature are the most important factors that affect the load change, but random factors such as big sport events or popular TV shows can change demand consumption in particular hours, which will lead to sudden load changes. A weighted time-variant slide fuzzy time-series model (WTVS) for short-term load forecasting is proposed to improve forecasting accuracy. The WTVS model is divided into three parts, including the data preprocessing, the trend training and the load forecasting. In the data preprocessing phase, the impact of random factors will be weakened by smoothing the historical data. In the trend training and load forecasting phase, the seasonal factor and the weighted historical data are introduced into the Time-variant Slide Fuzzy Time-series Models (TVS) for short-term load forecasting. The WTVS model is tested on the load of the National Electric Power Company in Jordan. Results show that the proposed WTVS model achieves a significant improvement in load forecasting accuracy as compared to TVS models.

Medium-Term Electric Load Forecasting Using Multivariable Linear and Non-Linear Regression

Medium-term forecasting is an important category of electric load forecasting that covers a time span of up to one year ahead. It suits outage and maintenance planning, as well as load switching operation. We propose a new methodol-ogy that uses hourly daily loads to predict the next year hourly loads, and hence predict the peak loads expected to be reached in the next coming year. The technique is based on implementing multivariable regression on previous year's hourly loads. Three regression models are investigated in this research: the linear, the polynomial, and the exponential power. The proposed models are applied to real loads of the Jordanian power system. Results obtained using the pro-posed methods showed that their performance is close and they outperform results obtained using the widely used ex-ponential regression technique. Moreover, peak load prediction has about 90% accuracy using the proposed method-ology. The methods are generic and simple and can be implemented to hourly loads of any power system. No extra in-formation other than the hourly loads is required.

Deep learning for time series forecasting:The electric load case

Management and efficient operations in critical infrastructures such as smart grids take huge advantage of accurate power load forecasting,which,due to its non-linear nature,remains a challenging task.Recently,deep learning has emerged in the machine learning field achieving impressive performance in a vast range of tasks,from image classification to machine translation.Applications of deep learning models to the electric load forecasting problem are gaining interest among researchers as well as the industry,but a comprehensive and sound comparison among different-also traditional-architectures is not yet available in the literature.This work aims at filling the gap by reviewing and experimentally evaluating four real world datasets on the most recent trends in electric load forecasting,by contrasting deep learning architectures on short-term forecast(oneday-ahead prediction).Specifically,the focus is on feedforward and recurrent neural networks,sequence-to-sequence models and temporal convolutional neural networks along with architectural variants,which are known in the signal processing community but are novel to the load forecasting one.

A Dynamic Forecasting System with Applications in Production Logistics

Production logistics involve the co-ordination of ac tivities such as production and materials control (PMC), inventory management, p roduct life cycle management, etc. Those activities demand for an accurate forec asting model. However, the conventional methods of making sell and buy decision based on human forecast or conventional moving average and exponential smoothing methods is no longer be sufficient to meet the future need. Furthermore, the un derlying statistics of the market information change from time to time due to a number of reasons such as change of global economic environment, government poli cies and business risks. This demands for highly adaptive forecasting model which is robust enough to response and adapt well to the fast changes in the dat a characteristics, in other words, the trajectory of the "dynamic characteristic s" of the data. In this paper, an adaptive time-series modelling method was proposed for short -term dynamic forecasting. The method employs an autoregressive (AR) time-seri es model to carry out the forecasting process. A modified least mean square (MLM S) adaptive filter algorithm was established for adjusting the AR model coeffici ents so as to minimise the sum of squared of forecasting errors. A prototype dyn amic forecasting system was built based on the adaptive time-series modelling m ethod. Basically, the dynamic forecasting system can be divided into two phases, i.e. the Learning Phase and the Application Phase. The learning procedures star t with the determination of upper limit of the adaptation gain based on the conv ergence in the mean square criterion. Hence, the optimum ELMS filter parameters are determined using an iteration algorithm which changes each filter parameter i.e. the order, the adaptation gain andthe values initial coefficient vector on e by one inside a predetermined iteration range. The set of parameters which giv es the minimum value for sum of squared errors within the iteration range is sel ected as the optimum set of filter parameters. In the Application Phase, the sys tem is operated under a real-time environment. The sampled data is processed by the optimised ELMS filter and the forecasted data are calculated based on the a daptive time-series model. The error of forecasting is continuously monitored w ithin the predefined tolerance. When the system detects excessive forecasting er ror, a feedback alarm signal was issued for system re-calibration. Experimental results indicated that the convergence rate and sum of squared erro rs during initial adaptation could be significantly improved using the MLMS algorithm. The performance of the system was verified through a series of experi ments conducted on the forecast of materials demand and costing in productio n logistics. Satisfactory results were achieved with the forecast errors confini ng within in most instances. Further applications of the system can be found i n sales demand forecast, inventory management as well as collaborative planning, forecast and replenishment (CPFR) in logistics engineering.

A New Multi-Method Combination Forecasting Model for ESDD Predicting

Equal Salt Deposit Density (ESDD) is a main factor to classify contamination severity and draw pollution distribution map. The precise ESDD forecasting plays an important role in the safety, economy and reliability of power system. To cope with the problems existing in the ESDD predicting by multivariate linear regression (MLR), back propagation (BP) neural network and least squares support vector machines (LSSVM), a nonlinear combination forecasting model based on wavelet neural network (WNN) for ESDD is proposed. The model is a WNN with three layers, whose input layer has three neurons and output layer has one neuron, namely, regarding the ESDD forecasting results of MLR, BP and LSSVM as the inputs of the model and the observed value as the output. In the interest of better reflection of the influence of each single forecasting model on ESDD and increase of the accuracy of ESDD prediction, Morlet wavelet is used to con-struct WNN, error backpropagation algorithm is adopted to train the network and genetic algorithm is used to determine the initials of the parameters. Simulation results show that the accuracy of the proposed combina-tion ESDD forecasting model is higher than that of any single model and that of traditional linear combina-tion forecasting (LCF) model. The model provides a new feasible way to increase the accuracy of pollution distribution map of power network.

A Hybrid Neural Network-based Approach for Forecasting Water Demand

Water is a vital resource.It supports a multitude of industries,civilizations,and agriculture.However,climatic conditions impact water availability,particularly in desert areas where the temperature is high,and rain is scarce.Therefore,it is crucial to forecast water demand to provide it to sectors either on regular or emergency days.The study aims to develop an accurate model to forecast daily water demand under the impact of climatic conditions.This forecasting is known as a multivariate time series because it uses both the historical data of water demand and climatic conditions to forecast the future.Focusing on the collected data of Jeddah city,Saudi Arabia in the period between 2004 and 2018,we develop a hybrid approach that uses Artificial Neural Networks(ANN)for forecasting and Particle Swarm Optimization algorithm(PSO)for tuning ANNs’hyperparameters.Based on the Root Mean Square Error(RMSE)metric,results show that the(PSO-ANN)is an accurate model for multivariate time series forecasting.Also,the first day is the most difficult day for prediction(highest error rate),while the second day is the easiest to predict(lowest error rate).Finally,correlation analysis shows that the dew point is the most climatic factor affecting water demand.

Forecasting International Tourism Regional Expenditure

The vast majority of tourism forecasting studies have centered on tourist arrivals at an aggregated level.Little research has been done of forecasting tourist expenditure at a national level let alone at a regional level.This study uses expenditure data to assess the relative economic impact of tourism into regional areas.By comparing five time-series models(the Na?ve,Holt,ARMA and Basic Structural Model(BSM)with and without intervention),and three econometric models(the Vector Autoregressive(VAR)model and the Time Varying Parameter(TVP)with and without intervention),the study sought to find the most accurate model for forecasting tourism expenditure two years ahead for each of the 31 provinces of China's Mainland.The results show that TVP models outperform other time series and econometric models.The research also provides practical management outcomes by providing methods for forecasting tourist expenditure as an indicator of economic growth in China’s provinces.The research concludes with the findings on the most appropriate model for regional forecasting and potential new variables suitable at the regional level.

Multivariate Time Series Forecasting with Transfer Entropy Graph

Multivariate Time Series(MTS)forecasting is an essential problem in many fields.Accurate forecasting results can effectively help in making decisions.To date,many MTS forecasting methods have been proposed and widely applied.However,these methods assume that the predicted value of a single variable is affected by all other variables,ignoring the causal relationship among variables.To address the above issue,we propose a novel end-to-end deep learning model,termed graph neural network with neural Granger causality,namely CauGNN,in this paper.To characterize the causal information among variables,we introduce the neural Granger causality graph in our model.Each variable is regarded as a graph node,and each edge represents the casual relationship between variables.In addition,convolutional neural network filters with different perception scales are used for time series feature extraction,to generate the feature of each node.Finally,the graph neural network is adopted to tackle the forecasting problem of the graph structure generated by the MTS.Three benchmark datasets from the real world are used to evaluate the proposed CauGNN,and comprehensive experiments show that the proposed method achieves state-of-the-art results in the MTS forecasting task.

基于多变量相空间重构和径向基函数神经网络的综合能源系统电冷热超短期负荷预测

为解决能源危机问题,提高能源利用率,综合能源系统(integrated energy system,IES)成为发展创新型能源系统的重要方向。准确的多元负荷预测对IES的经济调度和优化运行有着重要的影响,而借助混沌理论能够进一步挖掘IES多元负荷潜在的耦合特性。提出了一种基于多变量相空间重构(multivariate phase space reconstruction,MPSR)和径向基函数神经网络(radial basis function neural network,RBFNN)相结合的IES超短期电冷热负荷预测模型。首先,分析了IES中能源子系统之间的耦合关系,运用Pearson相关性分析定量描述多元负荷和气象特征的相关性。然后,采用C-C法对时间序列进行MPSR以进一步挖掘电冷热负荷和气象特征在时间上的耦合特性。最后,利用RBFNN模型对电冷热负荷间耦合关系进行学习并预测。实验结果表明,所提方法有效挖掘并学习电冷热负荷在时间上的耦合特性,且在不同样本容量下具有良好且稳定的预测效果。

基于多元变分模态分解和混合深度神经网络的短期光伏功率预测

针对传统分解预测方法忽略太阳辐照度等多维气象因素与光伏功率在时域和频域上的耦合关系以及深度神经网络在训练中出现的特征学习效率低、训练速度慢、过拟合等问题,提出基于多元变分模态分解(MVMD)和混合深度神经网络的短期光伏功率预测方法。首先,采用MVMD对光伏功率及多维气象序列进行时频同步分析,将其分解为频率对齐的多元本征模态函数,从而降低序列中非线性和波动性的影响。其次,针对多元本征模态函数,分别建立基于混合深度神经网络的预测模型。该模型采用卷积神经网络和双向长短时记忆神经网络来分别提取光伏功率及气象序列的空间相关特征和时间相关特征,并采用注意力机制来增强对重要时间点特征的学习权重。此外,使用残差连接来加快网络的训练速度以及缓解过拟合问题。通过实际工程实验分析,验证了该文方法的优越性。

基于历史天气的区域电网负荷预测

随着社会经济的迅速发展,人们对电能的需要日益增加,但是在电网运行中,常常会出现电力产能过剩或者不足的情况,为保证电力系统安全稳定、经济运行,就必须掌握各种区域电网负荷的变化规律和发展趋势。论文对重庆市区供电分公司供电区域电网中长期负荷进行预测,提出一种预测区域电网中长期负荷的方法,即一种基于前12个月历史天气条件和区域电网负荷关联关系的多元非线性拟合的特征参数因子曲线的中长期负荷预测方法,建立基于不同算法的多种预测模型,通过归一化处理,得到的区域电网中长期负荷预测的精度高,与实际区域电网负荷之间的误差小,对于区域电网中长期负荷预测分析具有重要参考利用价值。

基于注意力机制的ADE-Bi-IndRNN模型的中国粮食产量预测

为更加准确地预测我国粮食总产量,基于自适应差分进化算法来智能地选择基于注意力机制的双向独立循环神经网络的超参数,并考虑了粮食作物单位产量、农业生产条件、科技因素、农业保险、市场及经济因素五大类影响因素,构建了基于注意力机制的ADE-Bi-IndRNN粮食产量预测模型。经过预测分析得出我国2020—2024的粮食产量分别为6.67亿吨、6.72亿吨、6.80亿吨、6.99亿吨、7.02亿吨,总体呈现震荡上涨趋势,平均年增长率为1.15%。同时,通过对多个变量进行的注意力权重的分析,发现现阶段对我国粮食总产量预测贡献最大的三个变量为:谷物单位面积产量,粮食作物总播种面积,耕地灌溉面积,且政府对农业保险的政策性补贴、粮食进口量、谷物生产价格指数、农业生产资料指数也有助于提升我国的粮食总产量,并据此对我国粮食行业发展提出了建议。

多元数据融合的智能配电网负荷分析预测管理系统

针对多元大数据在智能配电网中的应用问题,为实现配电网的精益管理、科学预测和合理规划,文章开发了多元数据融合的智能配电网负荷分析预测管理系统。对软件系统的总体框架进行了设计;对软件系统的各功能模块进行开发和介绍;给出了软件系统的一个应用实例。该系统充分利用海量的历史负荷数据进行负荷特性分析,建立负荷特征库以及业扩信息库,通过对新接入用户进行信息匹配实现负荷管理及最大负荷预测。此外,该系统建立负荷预测方法模型库,可提供不同维度的负荷预测功能,从传统的地区负荷预测转变为馈线负荷预测,结合馈线现状以及业扩信息优化用户接入决策。总的来说,该系统具有功能模块数据链路互通、不同功能之间能提供信息支持、整体采用模块化设计思想等特点,可满足电网企业的日常应用需求。

Diffusionmodels for time-series applications: a survey

Diffusion models, a family of generative models based on deep learning, have become increasinglyprominent in cutting-edge machine learning research. With distinguished performance in generating samples thatresemble the observed data, diffusion models are widely used in image, video, and text synthesis nowadays. Inrecent years, the concept of diffusion has been extended to time-series applications, and many powerful models havebeen developed. Considering the deficiency of a methodical summary and discourse on these models, we providethis survey as an elementary resource for new researchers in this area and to provide inspiration to motivate futureresearch. For better understanding, we include an introduction about the basics of diffusion models. Except forthis, we primarily focus on diffusion-based methods for time-series forecasting, imputation, and generation, andpresent them, separately, in three individual sections. We also compare different methods for the same applicationand highlight their connections if applicable. Finally, we conclude with the common limitation of diffusion-basedmethods and highlight potential future research directions.

基于多变量LSTM模型的黄河流域气象干旱预测研究

干旱是对人类社会发展影响最严重的自然灾害之一,气象干旱预测是干旱研究中的重要方向。为提高气象干旱的预测精度,将多变量方法应用到长短期记忆模型(Long short-term memory,LSTM)预测黄河流域标准化气象干旱指数(Standardized precipitation evapotranspiration index,SPEI)的过程中,并和单变量LSTM模型的结果进行对比。使用均方根误差、平均绝对误差、纳什效率指数作为评价指标。结果显示,在对黄河流域临夏站、陶乐站、铜川站各自5种时间尺度SPEI(1、3、6、9和12个月)的预测中,多变量LSTM预测结果的3种评价指标值均明显优于单变量LSTM预测结果;可视化结果也显示多变量LSTM方法的预测曲线更接近观测值曲线。研究证明了多变量LSTM模型对于提高黄河流域气象干旱指数预测精度的有效性与适用性。

基于MPSR和IRBM的电力系统中长期负荷预测

针对电力系统中长期负荷波动大及不确定因素导致负荷预测误差较大的问题,提出了一种基于多变量相空间重构(MPSR)和改进受限波尔兹曼机(IRBM)的电力系统中长期负荷预测方法。首先,利用多元线性回归分析方法分析天气因素与电负荷之间的相关性,并将其与电负荷序列组成多变量时间序列;然后,利用C-C法确定每一时间序列的最优嵌入维数和时间延迟,实现多变量相空间重构;最后,采用多变量相空间重构建立的数据集训练电力系统负荷预测模型,同时利用梯度优化法对参数进行优化,得到预测模型。结果表明:相比长短期记忆神经网络和粒子群优化BP神经网络,所提出的预测方法有较高的精准度。

Influence of vapor pressure deficit on vegetation growth in China

Vapor pressure deficit(VPD)plays a crucial role in determining plant physiological functions and exerts a substantial influence on vegetation,second only to carbon dioxide(CO_(2)).As a robust indicator of atmospheric water demand,VPD has implications for global water resources,and its significance extends to the structure and functioning of ecosystems.However,the influence of VPD on vegetation growth under climate change remains unclear in China.This study employed empirical equations to estimate the VPD in China from 2000 to 2020 based on meteorological reanalysis data of the Climatic Research Unit(CRU)Time-Series version 4.06(TS4.06)and European Centre for Medium-Range Weather Forecasts(ECMWF)Reanalysis 5(ERA-5).Vegetation growth status was characterized using three vegetation indices,namely gross primary productivity(GPP),leaf area index(LAI),and near-infrared reflectance of vegetation(NIRv).The spatiotemporal dynamics of VPD and vegetation indices were analyzed using the Theil-Sen median trend analysis and Mann-Kendall test.Furthermore,the influence of VPD on vegetation growth and its relative contribution were assessed using a multiple linear regression model.The results indicated an overall negative correlation between VPD and vegetation indices.Three VPD intervals for the correlations between VPD and vegetation indices were identified:a significant positive correlation at VPD below 4.820 hPa,a significant negative correlation at VPD within 4.820–9.000 hPa,and a notable weakening of negative correlation at VPD above 9.000 hPa.VPD exhibited a pronounced negative impact on vegetation growth,surpassing those of temperature,precipitation,and solar radiation in absolute magnitude.CO_(2) contributed most positively to vegetation growth,with VPD offsetting approximately 30.00%of the positive effect of CO_(2).As the rise of VPD decelerated,its relative contribution to vegetation growth diminished.Additionally,the intensification of spatial variations in temperature and precipitation accentuated the spatial heterogeneity in the impact of VPD on vegetation growth in China.This research provides a theoretical foundation for addressing climate change in China,especially regarding the challenges posed by increasing VPD.

基于机器学习的电力物资需求预测

准确、合理的电力物资需求预测,可以为物资采购打下良好的基础,为企业提前统筹资源创造有利条件。考虑历史需求数据、项目投资及施工进度,提出一种基于机器学习的需求预测模型。从外部因素、内部因素、历史数据3个方面全量梳理物资需求相关的影响因子,先采用引入时间序列算法对数据进行处理,再利用多元神经网络算法构建需求预测模型。以某公司10 kV电力电缆为例进行需求预测,结果表明该模型能有效预测短期物资需求。