Time Series Information Research Articles

RESEARCH ON HUMAN POSTURE RECOGNITION METHOD BASED ON DEEP LEARNING

The dynamic recognition of human posture has very broad application prospects in fields such as human–computer interaction and virtual reality. A new method for dynamic recognition of human posture is proposed within the theoretical framework of deep learning. In our method, historical image information of human posture, current image information of human posture, and association information between each image are included as inputs in the deep learning process. Afterwards, the input information is formed into time series information and feature series information, which are then fused by the attention mechanism module. Finally, the dynamic recognition results of human posture are obtained through convolution operation. Experimental research was conducted on the AMASS dataset, and the results showed that our method can achieve better results in dynamic recognition of human posture, with both indicators superior to the other three methods. At the same time, our method has a fast convergence speed and the loss function remains low and continuously decreases during the deep learning process.

Mining of heterogeneous time series information for predicting chlorophyll accumulation in oceans

Harmful algal blooms cause environmental harm, financial losses, and disease epidemics. It is also known that the algal blooms cannot be eradicated; hence the best option is to foresee their growth and regulate it. Machine learning algorithms can be used to forecast their presence and further classify the threat that each concentration level presents. In this research work, the dataset collected from Santa Monica, US region is analyzed and processed to predict algae concentration using machine learning algorithms. In this process, the machine learning models such as multiple linear regression, Regression Gradient Boosting Decision Tree (RGBDT), and Hidden Markov Model (HMM) are applied to predict the chlorophyll (Chl-a) content, which serves as a proxy for the presence of algae in the water. The obtained results show that for prediction, the Multilinear regression model outperforms the RGBDT (Regression Gradient Boosting Decision Tree) algorithm. Similarly, for modeling chlorophyll using HMM (Hidden Markov Model), parameter bbp555.00_sd is the best among parameters like aot443.00_sd, kd490.00_sd, poc_sd and pic_sd. The multiple linear regression model gave an adjusted R-squared error of 0.94 with the parameter pic_sd having the least VIF value of 1.78 followed by aot and bbp which have VIF<5 (2.28 and 4.95 respectively). The outcome of the HMM-based model represents the probability of the presence of chlorophyll given the presence of each of the variables individually. From the results, it is observed that bbp has the highest probability of 0.405 implying that there is a 40% chance of chlorophyll in the presence of bbp.

Multivariate Trend‐Cycle‐Seasonal Decompositions with Correlated Innovations*

AbstractMultivariate analysis can help to focus on important phenomena, including trend and cyclical movements, but any economic information in seasonality is typically ignored. The present paper aims to more fully exploit time series information through a multivariate unobserved component model for quarterly data that exhibits seasonality together with cross‐variable component correlations. We show that economic restrictions, including common trends, common cycles and common seasonals can aid identification. The approach is illustrated using Italian GDP and consumption data.

A New and Effective Classification Method for Complex Time Series Based on Information Measure

The growing importance of time series information measure raises questions about how to effectively cluster a large number of nonlinear complex time series data and accurately extract more hidden information from them. In this paper, a clustering measurement and classification method for complex time series, the symmetrical exponential Tsallis relative information (SETRI) measure, is proposed, which aims to address these problems. The intrinsic characteristics of different types of time series information could be validly identified by this method. The modified multidimensional scaling (MDS) method, based on the SETRI measure, has the ability to display the data in the form of graphs for intuitive exhibition and accomplish the process of dimension reduction. The introduction of weighted permutation patterns allows a higher-accuracy classification not only for the time series dissimilarity quantification, but also for avoiding dispensable errors. Besides, the feasibility of the modified MDS classification method is visually and quantitatively verified by the simulated and real-world data. Compared with other MDS methods, the proposed method has better performance, which is reflected in the validity and rationality of the clustering results, thus further verifying the feasibility of the proposed method. Therefore, the new results will be helpful to develop complex data clustering and dimensionality reduction methods.

Transformer enhanced by local perception self-attention for dynamic soft sensor modeling of industrial processes

The use of dynamic soft sensor modeling methods to mine the time-varying and dynamic characteristics of industrial process data is of great significance for improving production efficiency and quality, given the rapid development of industrial processes and the increasing prominence of dynamic changes in the production process. However, existing dynamic soft sensor methods have limited long-term memory capacity, making it difficult to capture the long dynamic dependence, which can severely affect the results of the soft sensor model. To address this issue, we propose a dynamic soft sensor model based on local perception transformer, where the transformer is applied to achieve global perception of the variables. Through the application of the self-attention mechanism in the transformer encoder, the dynamic tracking and prediction of parameters can be realized by assigning different weights to the process variables and quality variables at different time steps, thereby adapting to the time-varying nature of the process. Additionally, convolution is used to generate a Query and Key in the self-attention mechanism, thereby enhancing local information learning. The proposed dot product self-attention calculation method effectively utilizes local information, thereby reducing the potential impact of abnormal data at a certain moment. Furthermore, by utilizing LSTM to extract time series information, the final prediction result was obtained. In soft sensor modeling experiments of the sulfur recovery unit and debutanizer tower, our proposed model demonstrated higher prediction accuracy compared to other methods, such as SVR, MLP, LSTM, CNN + LSTM, and vanilla transformer.

Towards a practical threat assessment methodology for crop landraces.

Crop landraces (LR), the traditional varieties of crops that have been maintained for millennia by repeated cycles of planting, harvesting, and selection, are genetically diverse compared to more modern varieties and provide one of the key components for crop improvement due to the ease of trait transfer within the crop species. However, LR diversity is increasingly threatened with genetic erosion and extinction by replacement with improved cultivars, lack of incentives for farmers to maintain traditional agricultural systems, and rising threats from climate change. Their active conservation is necessary to maintain this critical resource. However, as there are hundreds of thousands of LR and millions of LR populations for crops globally, active conservation is complex and resource-intensive. To assist in implementation, it is useful to be able to prioritise LR for conservation action and an obvious means of prioritisation is based on relative threat assessment. There have been several attempts to propose LR threat assessment methods, but none thus far has been widely accepted or applied. The aim of this paper is to present a novel, practical, standardised, and objective methodology for LR threat assessment derived from the widely applied IUCN Red Listing for wild species, involving the collation of time series information for LR population range, LR population trend, market, and farmer characteristics and LR context information. The collated information is compared to a set of threat criteria and an appropriate threat category is assigned to the LR when a threshold level is reached. The proposed methodology can be applied at national, regional, or global levels and any crop group.

Financial Fraud Identification Model of Listed Companies based on Time-Series Information

The aim of this research is to establish a high-precision financial fraud identification model for listed companies, which is mainly based on the financial indicators of time series. Support vector machine and K-means clustering algorithm are especially used in the research process. Firstly, local linear embedding is used to reduce the dimensionality of the selected financial indicators to extract the low-dimensional characteristics. Then the samples are classified into financial fraud and non-fraud by support vector machine, and the recognition model is constructed. At the same time, the research also uses K-means clustering algorithm to analyze the pattern of financial fraud. The experiment of dimensionality reduction proves that the model has a high effect on the processing of financial data, and the error between the data after dimensionality reduction and the original data is small. In addition, the clustering effect of the model also shows a clear pattern of fraud. In practical application, the accuracy rate of this model is as high as 94.89%, showing high accuracy and recall rate, and its F1 value is 87.08%, showing its feasibility and effectiveness in practice. The results highly prove that the performance of the financial fraud identification model proposed in this study is excellent, and it has a wide application prospect in the future.

Intelligent fatigue detection based on hierarchical multi-scale ECG representations and HRV measures

Fatigue detection based on electrocardiogram (ECG) from wearable devices is a convenient approach during exercise. However, how to model the data variations of ECG time series and capture effective information for accurate fatigue detection is still challenging. In this study, we propose a novel fatigue assessment architecture designed to adaptively extract and integrate both the ECG and heart rate variability (HRV) features, so as to obtain rich contextual information from ECG sequences and improve exercise-induced fatigue prediction performance. Inspired by the ECG observation process of human experts, the hierarchical transformer framework is designed to model the long-range dependencies of ECG time series and obtain multi-scale ECG representations. To reduce the computation cost, we also devise a sampling-efficient multi-head attention (SMA) mechanism. Finally, a novel gated feature fusion layer is developed to adaptively learn the weights of both the ECG and HRV features for fatigue prediction. The proposed fatigue prediction system is validated with real fatigue datasets which were collected from healthy participants. The experimental results show that the proposed approach outperforms other recent state-of-the-art systems in terms of accuracy 94.0%, specificity 99.0%, recall 89.3%, F1-score 93.7%, G-mean score 91.7%, MCC 0.837, and Cohen’s Kappa 0.847, based on the ECG time series with duration 20s. This demonstrates that the proposed system can be used to accurately detect fatigue in real time during exercise, such that assisting coaches or athletes to monitor and adjust their exercise session for preventing overexertion and fatigue-induced injury.

An increment of diversity method for cell state trajectory inference of time-series scRNA-seq data

The increasing emergence of the time-series single-cell RNA sequencing (scRNA-seq) data, inferring developmental trajectory by connecting transcriptome similar cell states (i.e., cell types or clusters) has become a major challenge. Most existing computational methods are designed for individual cells and do not take into account the available time series information. We present IDTI based on the Increment of Diversity for Trajectory Inference, which combines time series information and the minimum increment of diversity method to infer cell state trajectory of time-series scRNA-seq data. We apply IDTI to simulated and three real diverse tissue development datasets, and compare it with six other commonly used trajectory inference methods in terms of topology similarity and branching accuracy. The results have shown that the IDTI method accurately constructs the cell state trajectory without the requirement of starting cells. In the performance test, we further demonstrate that IDTI has the advantages of high accuracy and strong robustness.

Fault prediction of unmanned aerial vehicles based on entropy weight fusion and temporal graph convolutional network with non-Euclidian structure

Predicting weak and hidden faults in unmanned aerial vehicles (UAVs) is challenging due to their variable operation conditions and complex mechanisms. Conventional neural network models process the multisensory data in the form of Euclidean structure, the intrinsic connections among the individual data points are easy to be disregarded. Additionally, multisensory data are always directly fed into the model without adequately considering the importance or contribution of each sensor. Hence, an UAV fault prediction method is proposed by combining entropy weight fusion with a temporal graph convolutional network (GCN) to address the above problems. Firstly, the importance of multisensory data of UAVs are evaluated by each entropy value, and the multisensory data fusion is further realized by multiplying corresponding signal and entropy weight. Secondly, the multisensory data combined with fusion data are transferred together into graph-structure by adjacent matrix based on the node connection between different sensor data. Finally, the graph-structure data with non-Euclidian distance properties are input into temporal GCN to both capture the spatial and temporal relationship of the data, achieving better fault prediction results of UAVs. It is demonstrated that the proposed method is both applicable and superior in characterizing and predicting fault time series information of UAVs through parameter analysis and comparison studies with various existing algorithms.

Research on Sequential Decision-Making of Major Accidents with Incomplete Information

In order to solve the problem of emergency decision-making with incomplete information and deal with the accident information in different time series at the scenes of major accidents, this paper proposes a method of sequential decision-making by utilizing the relevant knowledge of D-S evidence theory and game theory. Firstly, we took an oil tank fire accident as an example and sorted out historical cases and expert experiences to establish a logical relationship between key accident scenes and accident scene symptoms in the accident. Meanwhile, we applied the logistic regression analysis method to obtain the basic probability distribution of each key accident scene in the oil tank fire, and on this basis, we constructed an evidence set of the fire. Secondly, based on the D-S evidence theory, we effectively quantified the knowledge uncertainty and evidence uncertainty, with the incomplete and insufficient information taken as an evidence system of the development of key accident scenes to construct a situation prediction model of these accident scenes. Thirdly, based on the game theory, we viewed emergency decision-makers and major accidents as two sides of the game to compare and analyze accident states at different time points and solve the contradiction between loss costs of decision-making and information collection costs. Therefore, this paper has provided a solution for the optimization of accident schemes at different time stages, thus realizing the sequential decision-making at the scenes of major accidents. Furthermore, we combined the situation prediction model with sequential decision-making, with the basic steps described below: (1) We drew up an initial action plan in the case of an extreme lack of information; then, we (2) started to address the accident and constructed a framework of accident identification, (3) collected and dealt with the continuously added evidence information with the evolution of the accident, (4) calculated the confidence levels of key accident scenarios after evaluating different evidence and then predicted the accident state in the next stage, and (5) calculated the profit–loss ratio between the current decision-making scheme and the decision-making scheme of the next stage. Finally, we (6) repeated steps (3) to (5) until the accident completely vanished. We verified the feasibility of the proposed method with the explosion accident of the Zhangzhou P.X. project in Fujian on 6 April used as an example. Based on the D-S evidence theory, this method employs approximate reasoning on the incomplete and insufficient information obtained at the scenes of major accidents, thus realizing the situation prediction of key scenes of these accidents. Additionally, this method uses the game theory to solve the contradiction between decision-making loss costs and information collection costs, thus optimizing the decision-making schemes at different time stages of major accidents.

Estimation of SOH of Lithium-Ion Batteries Based on PSO-Bi GRU-Attention Network

The State of Health (SOH) of lithium-ion batteries is a critical parameter that characterizes their actual lifespan, and its accurate assessment ensures the safe and reliable operation of batteries. However, in practical applications, SOH cannot be directly measured. To further improve the accuracy of SOH estimation for lithium-ion batteries, this study employs the Particle Swarm Optimization (PSO) algorithm to search for the optimal hyperparameters of the Bidirectional Gated Recurrent Unit (Bi GRU) neural network, enabling the prediction of time series information. Additionally, Attention Mechanism (AM) is integrated to allocate weights to the prediction results, resulting in the SOH prediction for lithium-ion batteries. The propose model is validated using the B0005 battery from the NASA lithium battery dataset. Experimental results demonstrate that, compared to the Bi GRU-Attention and Bi GRU models, the propose model reduces the Root Mean Square Error (RMSE) by 52.34% and 66.88%, respectively.

Multi-scenario renewable energy absorption capacity assessment method based on the attention-enhanced time convolutional network

As the penetration rate of renewable energy in modern power grids continues to increase, the assessment of renewable energy absorption capacity plays an increasingly important role in the planning and operation of power and energy systems. However, traditional methods for assessing renewable energy absorption capacity rely on complex mathematical modeling, resulting in low assessment efficiency. Assessment in a single scenario determined by the source-load curve is difficult because it fails to reflect the random fluctuation characteristics of the source-load, resulting in inaccurate assessment results. To address and solve the above challenges, this paper proposes a multi-scenario renewable energy absorption capacity assessment method based on an attention-enhanced time convolutional network (ATCN). First, a source-load scene set is generated based on a generative adversarial network (GAN) to accurately characterize the uncertainty on both sides of the source and load. Then, the dependence of historical time series information in multiple scenarios is fully mined using the attention mechanism and temporal convolution network (TCN). Finally, simulation and experimental verification are carried out using a provincial power grid located in southwest China. The results show that the method proposed in this article has higher evaluation accuracy and speed than the traditional model.

Base on temporal convolution and spatial convolution transformer for fluid prediction through well logging data

Fluid prediction is important in exploration work, helping to determine the location of exploration targets and the reserve potential of the estimated area. Machine learning methods can better adapt to different data distributions and nonlinear relationships through model training, resulting in better learning of these complex relationships. We started by using the convolution operation to process the log data, which includes temporal convolution and spatial convolution. Temporal convolution is specifically designed to capture time series relationships in time series data. In well log data, time information is often critical for understanding fluid changes and other important details. Temporal convolution learns trends and cyclical changes in the data. The spatial convolution operation makes the model more sensitive to the local features in the logging data through the design of the local receptive field and improves the sensitivity to fluid changes. Spatial convolution helps capture spatial correlations at different depths or locations. This can help the model understand the change of fluid in the vertical direction and identify the spatial relationship between different fluids. Then, we use the transformer module to predict the fluid. The transformer module uses a self-attention mechanism that allows the model to focus on information with different weights at different locations in the sequence. In the well log data, this helps the model to better capture the formation characteristics at different depths or time points and improves the modeling ability of time series information. The fully connected structure in the transformer module enables each position to interact directly with other locations in the sequence. By applying it to the data of Tarim Oilfield, the experimental results show that the convolutional transformer model proposed in this paper has better results than other machine learning models. This study provides a new idea in the field of logging fluid prediction.

Using time-series imagery and 3DLSTM model to classify individual tree species

ABSTRACT Classification of individual tree species (ITS) is critical for fine-scale forest surveys. However, it is difficult to obtain the complete and high-precision data needed for ITS classification in large areas. Lower spatial resolution time-series imagery is more accessible than other types of imagery and contains rich phenological information. In this study, after delineating individual tree crowns using a 0.2-m unmanned aerial vehicle (UAV) image, we used 3-m time-series imagery and a new 3DLSTM model to identify ITS at the Gaofeng Forest Farm in Guangxi Province, China. The 3DLSTM ITS classification model combines three-dimensional convolutional neural network (3D CNN) and long short-term memory (LSTM) models; thus, spatial, multiband, and time-series information can be extracted simultaneously to identify ITS more accurately. In this study, when only 3-m Planet time-series imagery was used for classification, the 3DLSTM model offered an ITS classification accuracy of 92.68%, outperforming two ITS classifiers (DenseNet or AlexNet model) based on one individual image. Moreover, the 3DLSTM model was better at identifying broad-leaved tree species than other deep-learning models. The experimental results proved that ITS classification could be significantly improved using only 3DLSTM and time-series images, offering the possibility of classifying large-scale ITS at a low cost.

Spatio-temporal sequence prediction of CO2 flooding and sequestration potential under geological and engineering uncertainties

CO2 injection for subsurface hydrocarbon development not only enhances oil and gas recovery but also enables CO2 sequestration in the subsurface. It is essential to develop effective methods for evaluating the potential of CO2 flooding and sequestration. Despite the existence of methods for predicting the effectiveness of hydrocarbon development using historical production data, insufficient emphasis has been placed on adequately incorporating geological and engineering uncertainty information to enhance prediction accuracy. To address this issue, a novel spatial-temporal ResNet (ST-ResNet) model is proposed for predicting hydrocarbon production, CO2 sequestration volume and CO2 diffusion pattern in the subsurface, which represent the CO2 flooding and sequestration potential. First, high-dimensional reservoir property fields are parameterized using the combined method of principal component analysis and discrete cosine transform (PCA-DCT). Second, the spatial sequence information of various reservoir property fields is extracted with features based on residual neural network (ResNet). Then, the time series information such as dynamic well control parameters is encoded with stacked BiLSTM (SBiLSTM). Specifically, the ST-ResNet model incorporates the above modules to overcome the limitations of collaborative consideration of temporal and spatial information involved in subsurface hydrocarbon development. Comparison between simulation and prediction results on the 2D/3D reservoir model reveals a significant achievement in prediction accuracy by the ST-ResNet model (with R2 and SSIM scores of 0.947, 0.911 and 0.937, 0.922, respectively). In comparison to CNN, LSTM and their combined approach CNN-LSTM, the ST-ResNet model demonstrates an improvement of 3% to 5% in R2, along with reductions of 20% to 30% in MAE and 15% to 25% in RMSE, respectively. These results highlight the superior stability and generalization of the ST-ResNet model. The contribution of this work is to provide a more accurate and efficient prediction tool to guide the integrated development of CO2 flooding and sequestration in subsurface hydrocarbon reservoirs, which facilitates decision-making processes for engineers.

A Floating Small Target Identification Method Based on Doppler Time Series Information

Traditional radar detection methods heavily rely on the signal-to-clutter ratio (SCR); a variety of feature-based detection methods have been proposed, providing a new way for radar detection and the recognition of weak targets. Existing feature-based detection methods determine the presence or absence of a target based on whether the feature value is within the judgment region, generally focusing only on the distribution of features and making insufficient use of inter-feature chronological information. This paper uses the autoregressive (AR) model to model and predict the time sequence of radar echoes in the feature domain and takes the chronological information of historical frame features as the prior information to form new features for detection on this basis. A classification method for floating small targets based on the Doppler spectrum centroid sequence is proposed. By using the AR model to fit the Doppler spectrum centroid feature sequence of the target, the model coefficients are regarded as the secondary features for target identification. The measured data show that the correct classification and identification rate of this method for ship targets and floating small targets can reach over 92% by using 50 centroid features.

Air quality forecasting‐driven cloud resource allocation for sustainable energy consumption: An ensemble classifier approach

AbstractIn recent times, air quality prediction is turned out to be one of the important research topics among research communities to prevent lives from negative health impacts. Random fluctuations of PM2.5 level brought about by frequent variations in meteorological factors create difficulties air pollution management. Forecasting the quality of air using time series data serves as a defense mechanism against threatening hazards by providing immense support to take preventive measures. Besides, handling dynamic real time workloads, forecasted by the prediction model requires appropriate computing resources to distribute workloads based on demands. To achieve this goal, this paper proposes a new Air Quality Prediction‐enabled Resource Allocation scheme for cloud‐based software services, which offers dynamic adjustment of resources based on workload demands with high energy efficiency. The proposed system is a two phase system that executes both air quality prediction and resource allocation processes consecutively. A new weighted average ensemble classifier is designed by combining support vector machine (SVM), artificial neural network (ANN), and gradient boosting machine (GBM) techniques to measure PM2.5 level on time series information of Beijing PM2.5 dataset. The system then dynamically allocates appropriate computing resources using crossover particle swarm optimization (CPSO) algorithm based on the forecasted results of PM2.5 level. This system has the potential to contribute significantly to reducing energy consumption and improving air quality in cities worldwide. The experimental results conducted to determine the efficiency of the proposed system in terms of different metrics proves that it achieves greater performance with less error functions for PM2.5 level prediction as well as minimizes energy consumption for resource allocation when compared with existing methods.

A CO concentration prediction method for electronic nose based on TrellisNet with gated recurrent unit and dilated convolution

As one of the harmful substances in the atmosphere, carbon monoxide (CO) is harmful to human beings. With the wide application of gas sensors and machine learning algorithms, the accuracy of concentration predicting of various gas such as CO is constantly improving. Now we apply TrellisNet, a network utilizing both recurrent and convolutional techniques, to gas concentration prediction, which is also a time-series prediction task, with the aim of improving its performance. To enhance TrellisNet’s ability to retain long time series information, we replaced the activation units in each layer of the model with gated recurrent unit (GRU). Compared to using long short-term memory (LSTM) as the activation unit, our approach has lower computational complexity and offers a more stable model. Additionally, we introduced dilated convolutions in each layer, allowing the model to establish connections with as many past time steps as possible at a given time point, even with fewer layers. This further enhances the preservation of long time series information. We named our improved technique trellis convolutional dilated network (TrelliSense). Due to the injection of the same input values in each layer, TrelliSense also exhibits superior training stability. Experimental results demonstrate that TrelliSense outperforms other time prediction networks, including temporal convolutional network (TCN), LSTM, GRU, Gaussian-TCN and bidrectional lstm (Bi-LSTM) in terms of all error metrics (MAE, RMSE, SMAPE). Therefore, we argue that TrelliSense is a better method for predicting CO concentration.

Absorption function loss due to the history of previous ankle sprain explored by unsupervised machine learning

BackgroundAnkle sprains are common and cause persistent ankle function reduction. To biomechanically evaluate the ankle function after ankle sprains, the ground reaction force (GRF) measurement during the single-legged landing had been used. However, previous studies focused on discrete features of vertical GRF (vGRF), which largely ignored vGRF waveform features that could better identify the ankle function. PurposeTo identify how the history of ankle sprain affect the vGRF waveform during the single-legged landing with unsupervised machine learning considering the time-series information of vGRF. MethodsEighty-seven currently healthy basketball athletes (12 athletes without ankle sprain, 49 athletes with bilateral, and 26 athletes with unilateral ankle sprain more than 6 months before the test day) performed single-legged landings from a 20 centimeters (cm) high box onto the force platform. Totally 518 trials vGRF data were collected from 87 athletes of 174 ankles, including 259 ankle sprain trials (from previous sprain ankles) and 259 non-ankle sprain trials (from without sprain ankles). The first 100 milliseconds (ms) vGRF waveforms after landing were extracted. Principal component analysis (PCA) was applied to the vGRF data, selecting 8 principal components (PCs) representing 96% of the information. Based on these 8 PCs, k-means method (k = 3) clustered the 518 trials into three clusters. Chi-square test assessed significant differences (p < 0.01) in the distribution of ankle sprain and non-ankle sprain trials among clusters. FindingsThe ankle sprain trials accounted for a significantly larger percentage (63.9%) in Cluster 3, which exhibited rapidly increased impulse vGRF waveforms with larger peaks in a short time. SignificancePCA and k-means method for vGRF waveforms during single-legged landing identified that the history of previous ankle sprains caused a loss of ankle absorption ability lasting at least 6 months from an ankle sprain.