13
Recent status-of-the-art methods for multivariate time series forecasting can be categorized into graph-based approach and global-local approach. The former approach uses graphs to represent the dependencies among variables and apply graph neural networks to the forecasting problem. The latter approach decomposes the matrix of multivariate time series into global components and local components to capture the shared information across variables. However, both approaches cannot capture the propagation delay of the dependencies among individual variables of a multivariate time series, for example, the congestion at intersection A has delayed effects on the neighboring intersection B. In addition, graph-based forecasting methods cannot capture the shared global tendency across the variables of a multivariate time series; and global-local forecasting methods cannot reflect the nonlinear inter-dependencies among variables of a multivariate time series. In this paper, we propose to combine the advantages of both approaches by integrating Adaptive Global-Local Graph Structure Learning with Gated Recurrent Units (AGLG-GRU). We learn a global graph to represent the shared information across variables. And we learn dynamic local graphs to capture the local randomness and nonlinear dependencies among variables. We apply diffusion convolution and graph convolution operations to global and dynamic local graphs to integrate the information of graphs and update gated recurrent unit for multivariate time series forecasting. The experimental results on seven representative real-world datasets demonstrate that our approach outperforms various existing methods.
53Semi-supervised multi-label feature selection with adaptive structure learning and manifold learning
Semi-supervised multi-label feature selection with adaptive structure learning and manifold learning
The Kramers–Moyal Coefficients of Non-stationary Time Series and in the Presence of Microstructure (Measurement) Noise
Most real world time series have transient behaviours and are non-stationary. They exhibit different type of non-stationarities, such as trends, cycles, random-walking, and generally exhibit strong intermittency. Therefore local stochastic characteristics of time series, such as the drift and diffusion coefficients, as well as the jump rate and jump amplitude, will provide very important information for understanding and quantifying “real time” variability of time series. For diffusive processes the systems have a longer memory and a higher correlation time scale and, therefore, one expects the stochastic features of dynamics to change slowly. In contrast, a rapid change of dynamics with jumps will cause strong ramp events (abrupt changes) in small time scales. Beside nonstationarity, in real world data, such as stock market index, pattern of cosmic background radiation, genomic data, etc., there is unique trajectory. Therefore, we require analysing techniques to estimate local stochastic behaviour of time series, which will be applicable to stationary and nonstationary time series and those with unique trajectory. This chapter contains technical aspects of the approach for real-time estimating of the Kramers–Moyal (KM) coefficients, and the drift and diffusion coefficients, as well as jump contributions for time series. We present the Kernel method (Nadaraya-Watson estimator) to estimate the time-dependent KM coefficients, which can be used in analysing stationary and nonstationary time series. We will also provide the details of estimating the KM coefficients in the presence of microstructure (measurement) noise, and show how the statistical properties of the noise can be determined from vanishing \(\tau \) limit behaviour of the KM conditional moments.
Read more17Integrating Health Services into an MCH-FP Program in Matlab, Bangladesh: An Analytical Update
This is a follow-up to a 1984 study that analyzed the relationship between areal variation in the contraceptive prevalence time series and in the intensity of maternal and child health (MCH) services of the Matlab Family Planning Health Services Project. Results based on 69 months of observation suggested that the addition of MCH components to a program with basic MCH and comprehensive family planning services had no incremental impact on family planning efficacy. However, basic MCH services, involving clinical back-up to family planning and child care since the beginning of the program in Matlab, may have contributed to clients' faith in the clinic staff and the overall efficacy of the Matlab program. In the 18 months that followed the cut-off date for this analysis, contraceptive prevalence increased markedly in the study areas. The present analysis repeats the earlier one for the extended time series to determine if the intensification of health services in Matlab contributed to this secondary increase in prevalence, and to ascertain whether the MCH service regimes have long-run differential impacts. Results for the 87-month time series are similar to those of the previous analysis, suggesting that the introduction of additional MCH inputs in the Matlab service area over the 1982-83 period had no incremental impact on the contraceptive prevalence rate time trend.
Read more4Watershed-scale-based forecast method for leaf area index data based on the integration of time series MODIS products and meteorological data
The leaf area index (LAI) is an important parameter used to characterize land vegetation growth patterns, and LAI forecasts are highly beneficial to natural resource management decisions. However, the LAI can be affected by biophysical, climate, and anthropogenic factors, thus, its future watershed-scale patterns remain difficult to assess. Herein, an object-oriented watershed-scale LAI forecasting (OWLF) method is proposed to solve the forecasting difficulties encountered when using only past meteorological data. The model was constructed with filtered multi-year MODIS LAI time series and climate index variables. Then, the dependence of the LAI on four mean monthly climate index variables was evaluated, namely, latent evapotranspiration, rainfall capacity, average air temperature, and sunlight hours, in the Poyang Lake basin. By combining the climate indexes derived from the past 1–3 months, the model forecasts the subsequent months LAI for different land cover types. The LAI forecasting results derived with a 13-year time series (2000–2012) suggest that the OWLF method can effectively recognize the expected spatial patterns, and the data agreed reasonably well with LAI dynamics and phenological periods. This work offers a promising way to exploit combined satellite and climate index data in novel and more accurate watershed-scale forecasting studies.
Read more85TodyNet: Temporal dynamic graph neural network for multivariate time series classification
TodyNet: Temporal dynamic graph neural network for multivariate time series classification
Multi-Objective Prediction of Integrated Energy System Using Generative Tractive Network
Accurate load forecasting can bring economic benefits and scheduling optimization. The complexity and uncertainty arising from the coupling of different energy sources in integrated energy systems pose challenges for simultaneously predicting multiple target load sequences. Existing data-driven methods for load forecasting in integrated energy systems use multi-task learning to address these challenges. When determining the input data for multi-task learning, existing research primarily relies on data correlation analysis and considers the influence of external environmental factors in terms of feature engineering. However, such feature engineering methods lack the utilization of the characteristics of multi-target sequences. In leveraging the characteristics of multi-target sequences, language generation models trained on textual logic structures and other sequence features can generate synthetic data that can even be applied to self-training to improve model performance. This provides an idea for feature engineering in data-driven time-series forecasting models. However, because time-series data are different from textual data, existing transformer-based language generation models cannot be directly applied to generating time-series data. In order to consider the characteristics of multi-target load sequences in integrated energy system load forecasting, this paper proposed a generative tractive network (GTN) model. By selectively utilizing appropriate autoregressive feature data for temporal data, this model facilitates feature mining from time-series data. This model is capable of analyzing temporal data variations, generating novel synthetic time-series data that align with the intrinsic temporal patterns of the original sequences. Moreover, the model can generate synthetic samples that closely mimic the variations in the original time series. Subsequently, through the integration of the GTN and autoregressive feature data, various prediction models are employed in case studies to affirm the effectiveness of the proposed methodology.
Read more1Application of Data Science for the reconstruction of time series of meteorological variables in the Islas del Rosario (Colombian Caribbean), between the years 2013-2021
This study reviews two time series of meteorological variables measured by an automatic station located in Islas del Rosario (Colombian Caribbean), belonging to the Network for Measurement of Oceanographic Parameters and Marine Meteorology (RedMpomm) of the General Maritime Directorate (Dimar). The time series correspond to data of air temperature and wind magnitude in the period 2013-2021, which present some missing values. The objective of the study was to develop a model that would automatically reconstruct missing values in the time series, using the advantages of data science to complete information with estimated values. The importance of obtaining reconstructed series lies in having more solid databases to be used in the research and academic work carried out by Dimar. The methodology developed consisted of the use of imputation of medians from existing data on dates and times associated with missing values, all this through the use of data lags and complementary information such as periodicity relationships on the data set. The results showed that it was possible to implement a reliable methodology capable of estimating the most appropriate value to complete the different time series, which constitutes a first approximation for reconstruction of meteorological data.
Read moreAnalysis of Trend and Variability in Time Series of Extreme Daily Temperature of Abu Dhabi City (UAE)
Trend identification plays a significant role in climate change studies to make future predictions about possible consequences on the urban environment, agriculture, water availability, etc. The Variation of Extreme Temperature intensity is observed in many regions all over the globe as an outcome of abrupt changes in climate. Keeping in mind the non-stationary climate conditions, it is essential to incorporate potential future changes in climatological studies. In the present study, the statistical parameters and the variability in time series of extreme daily temperature were analyzed using Generalized Extreme Value (GEV) distribution and its suitability has been examined by superimposing onto Gringorten plotting positions. The trend analysis of extreme daily temperatures was carried out using an innovative trend template proposed by Sen [7] and non-parametric Mann-Kendall (MK) trend test. The trend detection and variability of extreme daily temperature of Abu Dhabi City station, UAE was also carried out in this study. 34 years of temperature data were used for the analysis. Also, the extreme daily temperatures time series were checked for their serial correlation.
Read moreInteraction of mean and variance of time series: a Markov switching analysis
Describes the analysis of interaction of the mean and the variance of time series using a Markov switching model. In particular, we include Markov switching heteroscedasticity in the standard time series model. The proposed Markov switching model is applied to investigate the link between the mean and variance for the Singapore private property price index series. We assume that the price data series consists of a stochastic-trend random-walk component and a stationary autoregressive component. A four-state model of the price data series is then specified. By incorporating regime shifts in both mean and variance structures, we then analyse the interaction of mean and variance over long and short horizons.
Read more3Формування ціни на молоко в Україні в контексті світових тенденцій
Мета. Метою цієї статті є оцінка тенденцій у зміні цін на молоко в Україні порівняно з іншими країнами світу та виявлення закономірностей у їх зміні, а також апробація власного методичного підходу до оцінки рівня варіації в динамічних рядах. Методологія / методика / підхід. У процесі дослідження використано діалектичний метод пізнання, системний підхід до вивчення економічних явищ і процесів, монографічний метод (аналіз наукових здобутків вітчизняних і зарубіжних учених із проблем ціноутворення, функціонування ринку молока, впливу рівня цін на ефективність виробництва в молочному скотарстві). Зі спеціальних методів дослідження використано абстрактно-логічний (для теоретичних узагальнень і формулювання висновків), економіко-статистичні (при оцінці найбільш виражених циклів у динамічних рядах, рівня їх взаємозв’язку, варіації), графічний (при побудові графічних зображень динамічних процесів). Також запропоновано та здійснено апробацію коефіцієнта варіації прогнозних значень. Результати. У результаті проведеного дослідження встановлено, що в Україні на ринку молока найбільш чітко вираженим є цикл тривалістю 12 місяців, що є свідченням сезонного характеру зміни цін. Подібних закономірностей не було виявлено в інших країнах, які ми досліджували. Переважаючою тенденцією там були кон’юнктурні зміни цін під впливом попиту та пропозиції на молочну продукцію. Також визначено характер і рівень взаємозв’язку між різними ринками молока з допомогою регресійного аналізу. Установлено, що в Україні ціни на молоко, виражені в доларах США, мали тісну залежність із їх величиною в Новій Зеландії, країнах Європейського союзу. Отримані результати дають можливість для прогнозування зміни цін на молоко в Україні. Оригінальність / наукова новизна. Дістала подальшого розвитку оцінка рівня варіації в динамічних рядах через апробацію запропонованого коефіцієнта варіації прогнозних значень, що дозволило об’єктивно оцінити ситуацію в різних країнах. Практична цінність / значущість. Практичне значення одержаних результатів полягає в тому, що виявлення закономірностей у динаміці зміни цін на молоко в Україні ставить перед державними органами питання розробки відповідної політики, спрямованої на мінімізацію рівня варіації цін на молоко, з метою підвищення їх прогнозованості.
Read more5Intervention analysis of total suspended particulate data from Chicago
Intervention analysis of total suspended particulate data from Chicago
1Processing of Missing Observations and Outliers in Time Series
Thanks to rapid and remarkable development of electronic devices such as computers and measurement instruments, a variety of time series has been available automatically and continuously, resulting in minute-by-minute accumulation of a large amount of data. However when time series is observed for a long period, part of the time series is sometimes unable to be observed owing to some accidental causes such as malfunction of instrument or some physical restrictions of the observation system. In such a case, unobserved data is called missing observations. Even with only several percent of the missing observations, if they are studded with the data, the length of actually available data may become very short if only a part observed continuously is taken out.
Read more2Multivariate Time Series Retrieval with Binary Coding from Transformer
Deep learning to binary coding improves multivariate time series retrieval performance by end-to-end representation learning and binary codes from training data. However, it is fair to say that exist deep learning retrieval methods, e.g., Encoder-Decoder based on recurrent or Convolutional neural network, failed to capture the latent dependencies between pairs of variables in multivariate time series, which results in substantial loss of retrieval quality. Furthermore, supervised deep learning to binary coding failed to meet the requirements in practice, due to the lack of labeled multivariate time series datasets. To address the above issues, this paper presents Unsupervised Transformer-based Binary Coding Networks (UTBCNs), a novel architecture for deep learning to binary coding, which consists of four key components, Transformer-based encoder (T-encoder), Transformer-based decoder (T-decoder), an adversarial loss, and a hashing loss. We employ the Transformer encoder-decoder to encode temporal dependencies, and inter-sensor correlations within multivariate time series by attention mechanisms solely. Meanwhile, to enhance the generalization capability of deep network, we add an adversarial loss based upon improved Wasserstein GAN (WGAN-GP) for real multivariate time series segments. To further improve of quality of binary code, a hashing loss based upon Convolutional encoder (C-encoder) is designed for the output of T-encoder. Extensive empirical experiments demonstrate that the proposed UTBCNs can generate high-quality binary codes and outperform state-of-the-art binary coding retrieval models on Twitter dataset, Air quality dataset, and Quick Access Recorder (QAR) dataset.
Read more2Weighted Cross-Validation Evolving Artificial Neural Networks to Forecast Time Series
Accurate time series forecasting is a key tool to support decision making and for planning our day to-day activities. In recent years, several works in the literature have adopted evolving artificial neural networks (EANN) for forecasting applications. EANNs are particularly appealing due to their ability to model an unspecified non-linear relationship between time series variables. In this work, a novel approach for EANN forecasting systems is proposed, where a weighted cross-validation is used to build an ensemble of neural networks. Several experiments were held, using a set of six real-world time series (from different domains) and comparing both the weighted and standard cross-validation variants. Overall, the weighted cross-validation provided the best forecasting results.
Read moreSCALABLE HUMAN KNOWLEDGE ABOUT NUMERIC TIME SERIES VARIATION AND ITS ROLE IN IMPROVING FORECASTING RESULTS
Instead of handling fuzzy sets associated with linguistic (L-) labels based on the developers’ intuition immediately, the study follows the hedge algebras (HA-) approach to the time series forecasting problems, in which the linguistic time series forecasting model was, for the first time, proposed and examined in 2020. It can handle the declared forecasting L-variable word-set directly and, hence, the terminology linguistic time-series (LTS) is used instead of the fuzzy time-series (FTS). Instead of utilizing a limited number of fuzzy sets, this study views the L-variable under consideration as to the numeric forecasting variable's human linguistic counterpart. Hence, its word-domain becomes potentially infinite to positively utilize the HA-approach formalism for increasing the LTS forecasting result exactness. Because the forecasting model proposed in this study can directly handle L-words, the LTS, constructed from the numeric time series and its L-relationship groups, considered human knowledges of the given time-series variation helpful for the human-machine interface. The study shows that the proposed formalism can more easily handle the LTS forecasting models and increase their performance compared to the FTS forecasting models when the words’ number grows.
Read more
Copyright 2026 Cactus Communications. All rights reserved.
