Time Series Decomposition Research Articles

Gated Spatial–Temporal Merged Transformer Inspired by Multimask and Dual Branch for Traffic Forecasting

As an essential part of intelligent transportation system (ITS), traffic forecasting has provided crucial role for traffic management and risk assessment. However, complex spatial–temporal dependencies, heterogeneity, dynamicity, and periodicity of traffic data influence the traffic forecasting performance. Consequently, we propose a novel effective gated spatial–temporal merged transformer (GSTMT) inspired by multimask and dual branch for accurate traffic forecasting in this paper. Specifically, we first conduct a concatenation of gated spatial static mask transformer (GSSMT) and gated spatial dynamic mask transformer (GSDMT) with residual network. The GSSMT and GSDMT evolve from the traditional transformer by making preferable modifications that include gated linear unit (GLU), multimask mechanism including static mask matrix (SMM) and dynamic mask matrix (DMM), and spatial attention (SA). Among them, GLU is to promote the performance of capturing spatial dependency, dynamicity, and heterogeneity due to advanced performance for controlling information flow through layers. Additionally, by developing multimask mechanism including two novel SMM and DMM, the proposed GSTMT can precisely model the static and dynamic spatial structure for effectively highlighting static dependency and dynamicity. And SA is injected for enhancing the ability of capturing spatial dependency of GSSMT and GSDMT. Secondly, we develop a dual‐branch gated temporal transformer (DBGTT) for capturing temporal dependency, heterogeneity, dynamicity, and periodicity via incorporating the GLU and mixed time series decomposition (MTD) into traditional transformer. Similarly, we also introduce the GLU for empowering DBGTT with capability of capturing temporal dependency, dynamicity, and heterogeneity. In addition, MTD, which brings dual‐branch mechanism, can enhance the DBGTT for capturing more detailed temporal information via exploiting global and periodic profile of traffic data. At last, some experiments, which are performed on several real‐world traffic datasets, demonstrate the better results over classic traffic forecasting methods.

A Novel Sequence-to-Sequence-Based Deep Learning Model for Multistep Load Forecasting.

Load forecasting is critical to the task of energy management in power systems, for example, balancing supply and demand and minimizing energy transaction costs. There are many approaches used for load forecasting such as the support vector regression (SVR), the autoregressive integrated moving average (ARIMA), and neural networks, but most of these methods focus on single-step load forecasting, whereas multistep load forecasting can provide better insights for optimizing the energy resource allocation and assisting the decision-making process. In this work, a novel sequence-to-sequence (Seq2Seq)-based deep learning model based on a time series decomposition strategy for multistep load forecasting is proposed. The model consists of a series of basic blocks, each of which includes one encoder and two decoders; and all basic blocks are connected by residuals. In the inner of each basic block, the encoder is realized by temporal convolution network (TCN) for its benefit of parallel computing, and the decoder is implemented by long short-term memory (LSTM) neural network to predict and estimate time series. During the forecasting process, each basic block is forecasted individually. The final forecasted result is the aggregation of the predicted results in all basic blocks. Several cases within multiple real-world datasets are conducted to evaluate the performance of the proposed model. The results demonstrate that the proposed model achieves the best accuracy compared with several benchmark models.

Urbanization and malaria have a contextual relationship in endemic areas: A temporal and spatial study in Ghana.

In West Africa, malaria is one of the leading causes of disease-induced deaths. Existing studies indicate that as urbanization increases, there is corresponding decrease in malaria prevalence. However, in malaria-endemic areas, the prevalence in some rural areas is sometimes lower than in some peri-urban and urban areas. Therefore, the relationship between the degree of urbanization, the impact of living in urban areas, and the prevalence of malaria remains unclear. This study explores this association in Ghana, using epidemiological data at the district level (2015-2018) and data on health, hygiene, and education. We applied a multilevel model and time series decomposition to understand the epidemiological pattern of malaria in Ghana. Then we classified the districts of Ghana into rural, peri-urban, and urban areas using administratively defined urbanization, total built areas, and built intensity. We converted the prevalence time series into cross-sectional data for each district by extracting features from the data. To predict the determinant most impacting according to the degree of urbanization, we used a cluster-specific random forest. We find that prevalence is impacted by seasonality, but the trend of the seasonal signature is not noticeable in urban and peri-urban areas. While urban districts have a slightly lower prevalence, there are still pockets with higher rates within these regions. These areas of high prevalence are linked to proximity to water bodies and waterways, but the rise in these same variables is not associated with the increase of prevalence in peri-urban areas. The increase in nightlight reflectance in rural areas is associated with an increased prevalence. We conclude that urbanization is not the main factor driving the decline in malaria. However, the data indicate that understanding and managing malaria prevalence in urbanization will necessitate a focus on these contextual factors. Finally, we design an interactive tool, 'malDecision' that allows data-supported decision-making.

Practices and Innovations in Analyzing Digital Enabling High-Quality Development of Vocational Education Based on Time-Series Data Analytics

Abstract With the rapid development of the social economy and the continuous progress of science and technology, vocational education plays a crucial role in cultivating high-quality talents and promoting economic development. In this paper, based on the requirements of practice and innovation of digitization-enabled high-quality development of vocational education, a time-series decomposition model based on Transformer is proposed, which is defined as the DMR former model. The original data are decomposed using the temporal decomposition model, combined with the multi-scale fusion residual attention mechanism to capture and process the temporal feature information of vocational education in multiple time scales, and finally, the obtained results are analyzed. The results show that vocational education performs poorly in the classroom effect, most of the attention of vocational education students in the classroom is not concentrated, and the degree of liking for the classroom is lower than 0.5. After the digital empowerment of classroom effect, the student’s performance can be stabilized at more than 85 compared to the average of about 60 in the previous period, which is a good effect of improvement. After improving the curriculum of the College of Vocational Education, the employment rate of students increased to more than 95%. The high-quality development of vocational education and meeting the social demand for talent can be promoted through time series data analysis and digital empowerment.

Linear Transformation and its Properties with Application in Time Series Filtering

The primary objective of this article is to justify the concept of a linear transformation that derives its algebraic properties by means of variation matrices represented so far. The mathematical formulas will be used to elaborate and justify the argument of linear transformation. By displaying the significance of the t-transformation for estimation of latent variables in the time series decomposition, the paper obtains a general expression for smoothing matrices characterized by symmetric and asymmetric weighting system. The article further elaborates the concept of sub-matrix of the symmetric weights that is t-invariant, whereas the sub matrices of the asymmetric weights are the t-transformation of each other. By virtue of this relation, the properties of the t-transformation project useful imperative information on the smoothing of time series data, which may be required to clarify the concept of the topic raised so far. Eventually, the paper illustrates the role of the t-transformation on the weighting systems of several smoothers often applied for trend cycle estimation such as the locally weighted regression smoother, the cube smoothing spine, the Gaussian Kernel and 13-term trend cycle Henderson filter. By so doing, the article will pose the interrelated facts of linear transformation, from which the prospective researchers can benefit.

Spatiotemporal consistency and inconsistency of meteorological and agricultural drought identification: A case study of India

Drought identification is challenging because the severity, intensity and frequency of extreme drought events are heterogenous. Incorporating multiple datasets is to generate standardized drought indices for effective drought assessment, monitoring, and develop early warning systems. This study aims to explore the spatio–temporal relationship among drought indices of precipitation, soil moisture, vegetation condition and health based on standardizing multiple drought indices with trend detection. The indices were derived through standardized drought analysis from 2001 to 2020 over India and regional level. Moreover, time series decomposition was utilized to remove seasonal and random components for trend detection. This study enhanced the comprehension of meteorological and agricultural drought conditions in India by standardizing and analyzing trends in multiple drought indices. The results revealed that the most severe drought periods for subregions were consistent for all indices with trend analysis. All indices also reported a significant decline in trend for the 2009 drought event across the subregions in India. These findings contributed to a better understanding of drought dynamics in the subregion. However, vegetation-based indices were not always consistent to the indices of precipitation and soil moisture due to ecosystem resilience or heat stress. This study offered policymakers diverse information, including correlations and variations of multiple drought indices to assist in detecting the most severe periods of drought at a regional scale.

Modelling biochar long-term carbon storage in soil with harmonized analysis of decomposition data

The climate change mitigation benefits of biochar systems arise largely from carbon storage in biochar. However, while biochar is increasingly recognized as a carbon dioxide removal technology, there are on-going scientific discussions on how to estimate the persistence of biochar carbon when biochar is used in soils. Estimates vary from decades to millennia, building on different modelling approaches and evidence. Here, we revisited the persistence estimates derived from extrapolation of biochar incubation experiments, with the aims of making incubation data available, modelling choices transparent, and results reproducible. An extensive dataset of biochar incubations, including 129 biochar decomposition time series, was compiled and is made available alongside code for its analysis. Biochar persistence correlations were sensitive to data selection procedures and to the curve fitting modelling step, while soil temperature adjustments methods had less impact. Biochar H/C ratio remained the main predictor of biochar persistence, in line with previous research, regardless of the extrapolation assumptions (multi-pool exponential functions or power function) used in curve fitting. The relation between H/C and percentage of biochar carbon remaining after 100 years (BC100) was better explained by a power model than a linear model, with R2 values between 0.5 and 0.9. Using multi-pool exponential functions, estimated BC100 varied between 90 % and 60 % for H/C from 0 to 0.7. However, using power functions, BC100 was constrained between 90 % and 80 % for the same H/C range. Additional information about the biochar, the pyrolysis conditions or the environmental incubation conditions did not significantly increase explained variance. Notably, the dataset lacks observations at H/C ratios below 0.2, of biochar made from manure and biosolids, biochar from processes other than slow pyrolysis, field studies, and incubation temperatures below 10 °C, which should guide future experimental work. The detailed analysis performed in this study does not cast doubts on the longevity of biochar carbon storage; rather, it confirms previous knowledge by critically examining the modelling, elucidating the assumptions and limitations, and making the analysis fully reproducible. There is a need for further interdisciplinary work on integration of various theories and approaches to biochar persistence, ultimately leading to the formulation of policy-relevant conclusions.

Anomaly Detection Method for Lithium-Ion Battery Cells Based on Time Series Decomposition and Improved Manhattan Distance Algorithm.

Abnormalities in individual lithium-ion batteries can cause the entire battery pack to fail, thereby the operation of electric vehicles is affected and safety accidents even occur in severe cases. Therefore, timely and accurate detection of abnormal monomers can prevent safety accidents and reduce property losses. In this paper, a battery cell anomaly detection method is proposed based on time series decomposition and an improved Manhattan distance algorithm for actual operating data of electric vehicles. First, time series decomposition is performed on the voltage data of all battery cells in the battery pack to obtain the voltage trend component of each cell. Then, the improved Manhattan distance algorithm is utilized to calculate and compare the Manhattan distance values between adjacent cell trend components, to determine the abnormal cells inside the battery pack. Furthermore, the Manhattan distance values at the same sampling moment are calculated within the data sequence to detect the specific time when the abnormal cells malfunction. The data analysis and experimental verification results based on actual vehicle operating conditions indicate that this method can accurately identify an abnormal cell within the battery pack and diagnose the specific moment of abnormality in the battery cell at an early stage of failure, with good robustness.

Modeling cryptocurrency market dynamics using machine learning tools

The article analyzes the dynamics of the cryptocurrency market (Bitcoin) using econometric estimation tools based on machine learning models. The forecasting method is improved based on time series decomposition and lagged shifts of financial indicators. An ensemble of short-term forecast models for the Bitcoin exchange rate is built, and its accuracy is analyzed and compared to individual component models. Time series models are used along with calculated financial indicators (ADODS, NATR, TRANGE, ATR, OBV, RSI, ADTV). The absolute deviation of the short-term forecast amounted to $9.5, which is 0.06% of the absolute value.

Gaussian-Linearized Transformer with Tranquilized Time-Series Decomposition Methods for Fault Diagnosis and Forecasting of Methane Gas Sensor Arrays

Methane is considered as a clean energy that is widely used in places with high environmental requirements. The increasing demand for methane exploration in polar and deep sea extreme environments has a positive role in carbon neutrality policies. As a result, there will be a gradual increase in exploration activities for deep sea methane resources. Methane sensors require high reliability but are prone to faults, so fault diagnosis and forecasting of gas sensors are of vital practical significance. In this work, a Gaussian-linearized transformer model with a tranquilized time-series decomposition method is proposed for fault diagnosis and forecasting tasks. Since the traditional transformer model requires more computational expense with time complexity of O (N2) and is not applicable to continuous-sequence prediction tasks, two blocks of the transformer are improved. First, a Gaussian-linearized attention block is modified for fault-diagnosis tasks so that its time complexity can be changed to O (N), which can reduce computational resources. Second, a model with proposed attention for fault forecasting replaces the traditional embedding block with a decomposed block, which can input the continuous sequence data to the model completely and preserve the continuity of the methane data. Results show that the Gaussian-linearized transformer improves the accuracy of fault diagnosis to 99% and forecasting with low computational cost, which is superior to that of traditional methods. Moreover, the least mean-square-error loss of fault forecasting is 0.04, which is lower compared with the traditional time series prediction models and other deep learning models, highlighting the great potential of the proposed transformer for fault diagnosis and fault forecasting of gas sensor arrays.

Predictive Maintenance in IoT Devices using Time Series Analysis and Deep Learning

The pervasive integration of Internet of Things (IoT) devices across industries has ushered in a new era of data-driven operational efficiency. However, the reliability and uninterrupted functionality of these interconnected devices necessitate innovative approaches to maintenance. This research focuses on the development and implementation of a predictive maintenance framework for IoT devices, leveraging the synergies between Time Series Analysis (TSA) and Deep Learning (DL) techniques. The primary objective of this study is to enhance the accuracy and efficiency of predictive maintenance processes, ultimately minimizing downtime and optimizing resource utilization. The research methodology involves the collection of diverse data types from IoT devices, encompassing sensor readings, error logs, and historical maintenance records. A meticulous data preprocessing stage follows, involving cleaning, normalization, and feature extraction to prepare the dataset for analysis. The core analytical components of the proposed framework include Time Series Analysis for uncovering temporal patterns in the IoT data. Statistical methods and time series decomposition are applied to identify trends and seasonality, providing valuable insights into the device's performance over time. Concurrently, Deep Learning models, specifically recurrent neural networks (RNNs) and long short-term memory networks (LSTMs), are employed to predict maintenance needs based on historical patterns. Results obtained from the application of the predictive maintenance framework to real-world IoT datasets demonstrate promising accuracy and efficiency in anticipating maintenance requirements. The paper identifies existing challenges in predictive maintenance for IoT devices and suggests future research directions. These include the exploration of edge computing, federated learning, and the integration of explainable AI to enhance model interpretability. In conclusion, the study underscores the significance of predictive maintenance in ensuring the reliability of IoT devices, offering a roadmap for industries seeking to harness the full potential of data analytics and artificial intelligence for operational excellence.

The influence of optimization algorithm on the signal prediction accuracy of VMD-LSTM for the pumped storage hydropower unit

Pumped storage hydropower units, as a crucial type of energy conversion equipment, play a pivotal role in energy storage and balance. Predicting pressure pulsation signals in pump-turbine units can help anticipate potential unit malfunctions within the pump-turbine system. However, conventional prediction methods may be constrained by their ability to extract signal features and perform pattern recognition. With the continuous advancement in the realm of artificial intelligence, there has been a rise in the evolution of prediction involving variational mode decomposition (VMD) for time series decomposition and the utilization of genetic algorithms (GAs) for parameter optimization of long short-term memory (LSTM) networks has emerged. To address the fitness function issue in the context of genetic algorithms, this study employs three optimization algorithms to combine with the VMD-LSTM model. Three different prediction models are formed and compared with both the conventional LSTM and VMD-LSTM prediction methods. The results indicate that, in the prediction of pressure pulsation data at the three monitoring points of the pump-turbine, the predictive performance of the VMD-whale optimization algorithm (WOA)-LSTM model surpasses that of the conventional LSTM model and the VMD-LSTM prediction approach. Furthermore, the magnitude of prediction errors is reduced in comparison to the VMD-sparrow search algorithm (SSA)-LSTM and VMD-improved grey wolf optimization (IGWO)-LSTM prediction models.

Spillover Effects of Land Port Economy: A Case Study of Huaihua International Land Port in Hunan, China

Previous studies have shown that the construction and development of international land ports have a significant economic spillover effect on the neighboring regions. The construction and development of Huaihua International Land Port inevitably exerts economic spillover effects on the adjacent regions. It is necessary to analyze and predict the trend of economic data before and after the construction of the land port in the study area to evaluate accurately the role and effect of the land port in driving the economic development of the local and neighboring regions. Then, the impact of land ports as a variable is quantified by analyzing the difference between the predicted and actual results, thus providing a basis for formulating scientific policies. Local governments’ general public budget revenue (GPBR) is an important indicator to measure the economic development of a region. The paper uses the local governments’ GPBR of Huaihua International Land Port Economic Development Zone, Hecheng District, and Zhijiang Dong Autonomous County from January 2019 to June 2023. Based on this, time series decomposition and SARIMA modeling are carried out. It analyzes the impact of Huaihua International Land Port on the region and its economic spillover effect on the surrounding regions. The results found that the impact presented by the model is consistent with the actual situation. Therefore, the evaluation results have high accuracy and reliability.

A Landslide Displacement Prediction Model Based on the ICEEMDAN Method and the TCN–BiLSTM Combined Neural Network

Influenced by autochthonous geological conditions and external environmental changes, the evolution of landslides is mostly nonlinear. This article proposes a combined neural network prediction model that combines a temporal convolutional neural network (TCN) and a bidirectional long short-term memory neural network (BiLSTM) to address the shortcomings of traditional recurrent neural networks in predicting displacement-fluctuation-type landslides. Based on the idea of time series decomposition, the improved complete ensemble empirical mode decomposition with an adaptive noise method (ICEEMDAN) was used to decompose displacement time series data into trend and fluctuation terms. Trend displacement is mainly influenced by the internal geological conditions of a landslide, and polynomial fitting is used to determine the future trend displacement; The displacement of the fluctuation term is mainly influenced by the external environment of landslides. This article selects three types of landslide-influencing factors: rainfall, groundwater level elevation, and the historical displacement of landslides. It uses a combination of gray correlation (GRG) and mutual information (MIC) correlation modules for feature screening. Then, TCN is used to extract landslide characteristic factors, and BiLSTM captures the relationship between features and displacement to achieve the prediction of wave term displacement. Finally, the trend term and fluctuation term displacement prediction values are reconstructed to obtain the total displacement prediction value. The results indicate that the ICEEMDAN–TCN–BiLSTM model proposed in this article can accurately predict landslide displacement and has high engineering application value, which is helpful for planning and constructing landslide disaster prevention projects.

Inferring “high-frequent” mixed urban functions from telecom traffic

Precise distinction of mixed functions on urban land is essential for urban studies and planning, while existing methods are limited by high sampling bias, low observation frequency, and lack of semantic information in common data sources. In this paper, we introduce a new proxy for human behavior, the telecom traffic data as a remedy to the above limitations, and present an analytical framework which utilizes anonymized and aggregated telecom traffic data to infer mixed urban functions at spatiotemporal granularities as fine as buildings and hours. A time-series decomposition method is designed to map the mixture of urban functions, which is further refined by a hierarchical agglomerative clustering method taking urban textures as an additional source of information. In a case study in Shenzhen, China, we find the function of urban buildings can be decomposed into the mixture of three basic functions, namely dwelling, work, and recreation. We further find that the introduction of urban texture information helps identify particular forms of functional combination, which indicate special-function buildings such as urban villages and roadside shops. This study implies ways to improve urban management through methodological contributions in mixed urban function identification alongside the introduction of the telecom traffic, a kind of “high-frequency” urban data, and also helps inspire a rethinking of the form/function dichotomy in the era of “High-frequent” cities.

A Dual-View Approach for Multistation Short-Term Passenger Flow Prediction in Bus Transit Systems

Timely and accurate prediction of bus passenger flow plays a crucial role in uncovering real-time traffic demand, presenting an essential and formidable challenge in the realm of bus scheduling and management. The extensive application of deep learning methods in transit passenger flow prediction can be attributed to their exceptional ability to effectively capture spatiotemporal features, resulting in superior performance. However, prevailing deep learning models in transit passenger flow prediction tend to ignore the data enhancement. Additionally, the predominant focus on a single station in the prediction task presents challenges in effectively capturing the spatiotemporal features of the entire network. A model named TSD-ST is proposed to better accomplish the task of predicting short-term transit passenger flow at multistation. The TSD-ST model leverages time series decomposition for data enhancement. Simultaneously, in addition to considering the adjacency graph, the similarity of all the stations of the entire transit network is also considered and uses multigraph convolution and graph fusion modules. This approach enables the TSD-ST model to effectively capture spatiotemporal dependencies. Experiments based on real-world bus transit datasets confirm that the TSD-ST model shows better performance in prediction tasks at 30-min, 60-min, and 90-min time scales, with an average improvement of 21.87%. The effectiveness of each component has been verified through ablation experiments.

Multi-step prediction of chlorophyll concentration based on adaptive graph-temporal convolutional network with series decomposition

Chlorophyll concentration can well reflect the nutritional status and algal blooms of water bodies, and is an important indicator for evaluating water quality. The prediction of chlorophyll concentration change trend is of great significance to environmental protection and aquaculture. However, there is a complex and indistinguishable nonlinear relationship between many factors affecting chlorophyll concentration. In order to effectively mine the nonlinear features contained in the data. This paper proposes a time-series decomposition adaptive graph-time convolutional network prediction model. Firstly, the original sequence is decomposed into trend component and periodic component by moving average method. Secondly, based on the graph convolutional neural network, the water quality parameter data is modeled, and a parameter embedding matrix is defined. The idea of matrix decomposition is used to assign weight parameters to each node. The adaptive graph convolution learns the relationship between different water quality parameters, updates the state information of each parameter, and improves the learning ability of the update relationship between nodes. Finally, time dependence is captured by time convolution to achieve multi-step prediction of chlorophyll concentration. The validity of the model is verified by the water quality data of the coastal city Beihai. The results show that the prediction effect of this method is better than other methods. It can be used as a scientific resource for environmental management decision-making.

Automatic Detection of Maintenance Scenarios for Equipment and Control Systems in Industry

The well-known methods of scene extraction on video are focused on analyzing the similarity between frames. However, they do not all analyze the composition of the image scene, which may remain the same during maintenance. Therefore, this paper proposes an algorithm for equipment maintenance scene detection based on human hand tracking. It is based on the assumption that, when servicing technological equipment, it is possible to determine the change in repair action by the position of the service engineer’s hands. Thus, certain information and the algorithm that processes these changes allow us to segment the video into actions performed during the service. We process the time series obtained by moving the hand position using spectral singular value decomposition for multivariate time series. To verify the algorithm, we performed maintenance on the control cabinet of a mining conveyor and recorded the work on a first-person video, which was processed using the developed method. As a result, we obtained some scenes corresponding to opening the control cabinet, de-energizing the unit, and checking the contacts with a multimeter buzzer test. A third-person video of motor service was similarly processed. The algorithm demonstrated the results in separate scenes of removing screws, working with a multimeter, and disconnecting and replacing motor parts.

Short-term Wind Power Prediction based on CEEMDAN-TCN

In order to improve the accuracy of wind power prediction, a wind power prediction method based on time series decomposition and error correction is proposed in this paper. Firstly, the maximum information coefficient (MIC) method is used to select the features with strong correlation with wind power, so as to reduce the complexity of the original data; Then, according to the non-stationary characteristics of wind power, the wind power is decomposed into several stationary subsequences by using adaptive noise complete set empirical mode decomposition (CEEMDAN); Finally, the time convolution network (TCN) is used to dynamically model the multivariable time series of wind power; In order to further improve the prediction accuracy, a light quantization hoist (LightGBM) is introduced to correct the error of the prediction value. The simulation results show that the proposed method has higher short-term wind power prediction accuracy than other prediction models.

Short-Term Fine-Grained Regional MUF Prediction for HF Communication Based on Time-Series Decomposition

Short-Term Fine-Grained Regional MUF Prediction for HF Communication Based on Time-Series Decomposition