Time Series Analysis Algorithm Research Articles

Streamflow in the United States: Characteristics, trends, regime shifts, and extremes

Long-term streamflow observations contain essential information for understanding hydrological changes and managing water resources. A continental-scale dataset or analysis of temporal streamflow change is still lacking across hydrologic gauges in the Conterminous United States (CONUS). Here, we compiled 70 years of streamflow records from 1951 to 2021 at ~ 8000 hydrologic stations across the CONUS and characterized temporal trends, regime shifts, and extreme events using a Bayesian time series analysis algorithm. We found that the occurrences of sudden streamflow changes (e.g., regime shifts and extreme events) have been increasing with time across the CONUS. In addition, we derived 181 streamflow indicators that are valuable for hydrological and biological applications, such as the duration and frequency of high or low streamflow events. The Mississippi River Basin, especially the middle and lower parts, was a hot spot of high-frequency high-flow events. Overall, we anticipate the dataset generated here offers valuable information for understanding and quantifying changes in water resources across the CONUS.

Machine learning-based understanding of aquatic animal behaviour in high-turbidity waters

Inspired by the ambitions envisioned in the Fourth Industrial Revolution for aquaculture, also known as Aquaculture 4.0, the aquaculture (marine animal farming) industry is seeking to adopt data-driven Artificial Intelligence (AI) to help significantly improve business operations. One of the major barriers is the manual annotation of animal behaviour data, which is a time-consuming task that demands high levels of concentration from biologists. To address this challenge, this paper proposes novel automatic animal behaviour monitoring tailored for industrial scenarios. Our approach introduces a real-time machine-learning-based instance segmentation system that is specialised for underwater environments, where large groups of shrimp are farmed. The implemented system achieves an accuracy rate of 89% at 30 frames per second (fps) and can accurately detect shrimp in high-density areas under poor lighting conditions and high turbidity waters, despite the challenges of occlusion and overlapping. A key innovation of our method is the implementation of a new density cluster algorithm for time series and video analysis. This approach provides a more efficient and accurate way of monitoring animal behaviour, significantly saving time and effort for biologists and advancing the capabilities of automated aquaculture systems.

Global Potato Production Forecasting Based on Time Series Analysis and Advanced Waterwheel Plant Optimization Algorithm

Global Potato Production Forecasting Based on Time Series Analysis and Advanced Waterwheel Plant Optimization Algorithm

Forecasting Student Employment Trends in Colleges and Universities Education Using Time Series Analysis Algorithms

Student employment trends in colleges have been evolving, with a notable shift towards internships, co-op programs, and other experiential learning opportunities. Colleges are increasingly emphasizing the importance of gaining practical work experience alongside academic studies to enhance students' employability. This trend reflects the growing demand from employers for graduates who possess not only theoretical knowledge but also relevant skills and hands-on experience. This study investigates student employment trends within Colleges and Universities education through the application of time series analysis algorithms. Utilizing historical data spanning five years from 2019 to 2023, the study examines the dynamics of student employment rates and explores the effectiveness of forecasting methodologies, including ARIMA and Ranking with ARIMA (R-ARIMA). The findings reveal a consistent upward trend in student employment rates over the study period, indicating a favorable environment for student career opportunities within higher education institutions. Moreover, the accuracy of forecasting models is demonstrated through the close alignment between predicted and observed employment rates. Optimal ARIMA configurations are identified, providing insights for strategic planning and resource allocation to support student career development and success. The findings reveal a consistent upward trend in student employment rates over the study period, with the percentage of employed students increasing annually. For instance, the student employment rate rises from 65% in 2019 to 77% in 2023. This indicates a favourable environment for student career opportunities within higher education institutions. Moreover, the accuracy of forecasting models is demonstrated through the close alignment between predicted and observed employment rates. Optimal ARIMA configurations are identified, with the best-performing model achieving an AIC score of 95.3 and a BIC score of 99.8. These configurations provide insights for strategic planning and resource allocation to support student career development and success.

Application of Time Series Analysis and Regression Algorithm in Assessing the Effectiveness of Monetary Policy in the Era of Digital Economy

Monetary policy is the series of actions to manage country's money supply and to attain economic growth. An objection to using interest rates in monetary policy is that the use of interest rates has the problem of time uncertainty. Therefore, this research introduces Conditional Autoregressive Value-at-Risk and Henry Gas Solubility Optimization (Caviar-HGSO) with Deep Long Short Term Memory (DeepLSTM) to monetary policy forecasting to demonstrate its effectiveness. Here, Caviar-HGSO is a hybrid of Caviar and HGSO used to tune DeepLSTM weights. The HGSO algorithm follows the wind state to balance exploration and exploitation in search space while avoiding local optimization. A Caviar model estimates the parameter over time by autoregressive process and determine a parameter by regression analysis. The advantages of these two optimization algorithms lead to better monetary policy forecasts. The experimental results show that Caviar-HGSO_DeepLSTM performs better regarding MAE, MSE, and RMSE, namely 0.438, 0.103, and 0.351.

Analysis of Economic Impact Based on Time Series Model -- Taking the GDP of Two Provinces as an Example

Time series analysis is an important branch of statistics, which is widely used in various fields, as well as in the economic field. In the economic field, a reasonable GDP forecast helps a country to set credible, substantive, and strategic goals. In recent years, Northeast China has seen an obvious decline in the national economic ranking, and the relevant GDP data of Jilin and Heilongjiang provinces are more obvious. This paper is based on time series analysis and support vector regression (SVR) algorithm to build a series model. The model was established by using 60 GDP data of Jilin Province and Heilongjiang Province from 1959 to 2022 with similar economic facts. After calculation and verification, the optimal model of Jilin Province is ARIMA (1, 1,0), the SVR with parameter C of 10 and 2 of 0.007, and the combined model with it as the main model is the optimal model of Heilongjiang Province. The GDP of the two provinces will continue to rise in the next five years, and the GDP of Jilin Province will rise faster, and the GDP gap between the two provinces will gradually narrow.

Mapping Forest Abrupt Disturbance Events in Southeastern China—Comparisons and Tradeoffs of Landsat Time Series Analysis Algorithms

Forest change monitoring is a fundamental and routine task for forest survey and planning departments, and the resulting forest change information acts as an underlying asset for sustainable forest management strategy development, ecological quality assessment, and carbon cycle research. The traditional ground-based manual monitoring of forest change has the disadvantages of high time and labor costs, low accessibility, and poor timeliness over wide regions. Remote sensing technology has become a popular approach for multi-scale forest change monitoring due to its multiple available spatial, spectral, temporal, and radiometric resolutions and wide coverage. Particularly, the free access policy of long time series archive data of Landsat (around 50 years) has triggered many automated analysis algorithms for landscape-scale forest change analysis, such as VCT, LandTrendr, BFAST, and CCDC. These automated algorithms differ in their principles, parameter settings, execution complexity, and disturbance types to be detected. Thus, selecting a suitable algorithm to satisfy the particular forest management demands is an urgent and challenging task for forest managers in a given forested area. In this study, Lishui City, Zhejiang Province, a typical plantation forest region in Southern China where forest disturbance widely and frequently exists, was selected as the study area. Based on the GEE platform, the algorithmic adaptability of VCT, LandTrendr, and CCDC in monitoring abrupt forest disturbance events was compared and assessed. The results showed that the kappa coefficients of the abrupt disturbance events detected by the three algorithms were at 0.704 (LandTrendr), 0.660 (VCT), and 0.727 (CCDC), and the corresponding overall accuracies were at 0.852, 0.830, and 0.862, respectively. The validated disturbance occurrence time consistency reached nearly 80% for the three algorithms. In light of the characteristics of forest disturbance occurrence in southeastern China as well as the algorithmic complexity, efficiency, and adaptability, LandTrendr was recommended as the most suitable one in this region or other similar regions. Overall, the forest change monitoring process based on GEE is becoming more simplified and easily implemented, and the comparisons and tradeoffs in this study provide a reference for the choice of long time series forest monitoring algorithms.

Ensemble Improved Permutation Entropy: A New Approach for Time Series Analysis.

Entropy quantification approaches have gained considerable attention in engineering applications. However, certain limitations persist, including the strong dependence on parameter selection, limited discriminating power, and low robustness to noise. To alleviate these issues, this paper introduces two novel algorithms for time series analysis: the ensemble improved permutation entropy (EIPE) and multiscale EIPE (MEIPE). Our approaches employ a new symbolization process that considers both permutation relations and amplitude information. Additionally, the ensemble technique is utilized to reduce the dependence on parameter selection. We performed a comprehensive evaluation of the proposed methods using various synthetic and experimental signals. The results illustrate that EIPE is capable of distinguishing white, pink, and brown noise with a smaller number of samples compared to traditional entropy algorithms. Furthermore, EIPE displays the potential to discriminate between regular and non-regular dynamics. Notably, when compared to permutation entropy, weighted permutation entropy, and dispersion entropy, EIPE exhibits superior robustness against noise. In practical applications, such as RR interval data classification, bearing fault diagnosis, marine vessel identification, and electroencephalographic (EEG) signal classification, the proposed methods demonstrate better discriminating power compared to conventional entropy measures. These promising findings validate the effectiveness and potential of the algorithms proposed in this paper.

Forecasting Stock Market Prices Using Machine Learning and Deep Learning Models: A Systematic Review, Performance Analysis and Discussion of Implications

The financial sector has greatly impacted the monetary well-being of consumers, traders, and financial institutions. In the current era, artificial intelligence is redefining the limits of the financial markets based on state-of-the-art machine learning and deep learning algorithms. There is extensive use of these techniques in financial instrument price prediction, market trend analysis, establishing investment opportunities, portfolio optimization, etc. Investors and traders are using machine learning and deep learning models for forecasting financial instrument movements. With the widespread adoption of AI in finance, it is imperative to summarize the recent machine learning and deep learning models, which motivated us to present this comprehensive review of the practical applications of machine learning in the financial industry. This article examines algorithms such as supervised and unsupervised machine learning algorithms, ensemble algorithms, time series analysis algorithms, and deep learning algorithms for stock price prediction and solving classification problems. The contributions of this review article are as follows: (a) it provides a description of machine learning and deep learning models used in the financial sector; (b) it provides a generic framework for stock price prediction and classification; and (c) it implements an ensemble model—“Random Forest + XG-Boost + LSTM”—for forecasting TAINIWALCHM and AGROPHOS stock prices and performs a comparative analysis with popular machine learning and deep learning models.

System dynamics monitoring using PIC micro-controller-based PLSE

The permutation largest slope entropy (PLSE) algorithm has been shown to be effective to distinguish between regular and non-regular dynamics from time series analysis. However, as it is the case for many non-linear time series analysis algorithms, such a characterization is locally made and does not allow one to capture some micro-phenomena, such as intermittency, that may occur in the system behavior. This paper presents a PIC micro-controller based implementation of the PLSE for a real-time monitoring of system dynamics. The PLSE algorithm is optimized to fit the program and data memory of low-end processors using the XC8 compiler and the MPLAB X IDE. The resulting algorithm is implemented on the PIC16F18446 and deployed on the Explorer 8 development board. The effectiveness of the developed tool is validated by considering an electrical circuit of the Duffing oscillator that can generate both periodic and chaotic dynamics. By comparing the PLSE values with the phase portraits and previous results on the Duffing oscillator circuit, the developed tool efficiently allows one to monitor the behavior of dynamical systems.

Forecast of differential settlement of structures using data of monitoring system

The article analyzes the results of observations of uneven settlement of various buildings, obtained in the course of long-term observations with the help of automated systems of deformation monitoring. The main purpose of the work is to evaluate the possibilities of predicting sedimentary processes using statistical forecasting models. Emphasis is placed on approach using time series analysis algorithms. The results of simulation using the integrated autoregressive moving average model are demonstrated. To build a forecast, data on the foundation settlement over a 5-year period were used. Predictive values obtained for 1-year and 2-year intervals are compared with the data of real observations. The comparison showed that at the chosen forecasting horizon, the deviation of the predicted settlement value from the actually observed one is comparable to the accuracy of sediment measurement by hydraulic leveling sensors. The study showed the promise of using this approach to predict the integrity of structures within the framework of an automated system of deformation monitoring.

Quantum Dynamic Mode Decomposition Algorithm for High-Dimensional Time Series Analysis

The dynamic mode decomposition (DMD) algorithm is a widely used factorization and dimensionality reduction technique in time series analysis. When analyzing high-dimensional time series, the DMD algorithm requires extremely large amounts of computational power. To accelerate the DMD algorithm, we propose a quantum-classical hybrid algorithm that we call the quantum dynamic mode decomposition (QDMD) algorithm. Given a time series X ∈ R n × ( m + 1) with n ≫ m , the QDMD algorithm first executes quantum singular value decomposition on a matrix related to X and obtains a quantum state containing the main singular values and singular vectors of the decomposed matrix, then performs a low-sampling-frequency process on the obtained quantum state and computes the low-dimensional projection of the DMD operator through the sampling results. Finally, the algorithm computes the DMD eigenvalues and prepares the amplitude-encoding states of the DMD modes using the obtained classical information and X . Considering the main variables, the complexity of the QDMD algorithm is O ~ M m polylog n / ϵ , where M = O ~ m 3 / ϵ 2 denotes the number of samples. Compared with the classical DMD algorithm, which has complexity O ~ n m 2 log 1 / ϵ , the QDMD algorithm provides an exponential acceleration of n , at the cost of greater dependence on M and ϵ . We test the effects of M on the QDMD algorithm in the specific application scenarios of data denoising, scene background extraction, and fluid dynamics analysis. We determined that the QDMD algorithm requires only a small number of samples M in specific applications, further demonstrating the quantum advantage of the QDMD algorithm in high-dimensional data analysis.

Integrating Predictive Big Data Analytics with Behavioral Machine Learning Models for Proactive Threat Intelligence in Industrial IoT Cybersecurity

This paper introduces a comprehensive framework for industrial Internet of Things (IoT) cybersecurity, integrating multiple algorithms to enhance threat intelligence. The proposed framework encompasses five key algorithms, each addressing specific aspects of data preprocessing, time series analysis, predictive analytics, and behavioral machine learning. The Data Preprocessing and Integration algorithm refines raw IoT data through a meticulous 20-step process, ensuring high-quality input for subsequent analyses. The Time Series Analysis algorithm delves into temporal patterns, while the Random Forest algorithm focuses on predictive analytics for proactive threat detection. The LSTM Ensemble algorithm extends the analysis into behavioral machine learning, capturing temporal dependencies and detecting anomalies. The Weighted Average Ensemble combines outputs from predictive analytics and behavioral models, leveraging their correlation for enhanced threat intelligence. An ablation study dissects the individual contributions of each algorithmic component, shedding light on their specific impacts. The results highlight the significance of each step, guiding optimizations for improved performance. The proposed framework outperforms existing methods in various performance metrics, showcasing its potential as a robust solution for proactive threat intelligence in complex industrial environments. This framework stands at the forefront of industrial IoT cybersecurity, offering a holistic and adaptive approach to address evolving threats. The ablation study enhances the transparency and understanding of the framework, contributing to its continuous refinement and effectiveness in safeguarding critical industrial systems.

Assessing the impact of Sentinel-1 derived planting dates on rice crop yield modeling

• Sentinel-1A SAR-based dynamic planting dates (PD) for farm management activities. • Difference between SAR-based PD and the Crop Calendar based PD can be <=75 days. • SAR-based PD significantly improves rice yield estimates using crop models. Planting date (PD) is one of the critical component in crop modeling that has a large-scale impact on seasonal yield estimation. Here, we aimed to estimate the various paddy/rice attributes (particularly, acreage and PD) using the Sentinel-1A observations, and assessing the impact of such key parameters on end-of-season yields. The method comprises: (1) a multi-temporal time-series analysis algorithm of Sentinel-1A observations for rice mapping, (2) a PD retrieval algorithm, and (3) rice yield estimation using the Synthetic Aperture Radar (SAR)-derived PD in a modeling system. Our study was implemented over the rainfed paddy fields in Cambodia between 2017 thru 2020. We demonstrated that the backscatter ratio ( σ VH 0 σ VV 0 ) from Sentinel-1A can improve the accuracy of rice mapping, while σ VH 0 backscatter can provide PD estimations. Our results showcased reasonable/good performance in mapping the paddy fields and estimating PD in the study area. PD found in March and April were indicative of irrigation application in some parts of Cambodia, specifically in the southeastern part (also part of the Mekong delta) where irrigation in paddy fields is a common practice. Differences between the crop-calendar based PD (June 1) and the SAR-derived PD can range as much as 75 days. In the rice-dominated provinces such as Banteay-Meanchey and Svay-Rieng of Cambodia, when SAR-derived PD are used, rice yield normalized bias improved compared to the fixed PD data by a mean of 7–12 % and 30–48 % respectively within the study period. In addition, the PD derived from SAR also significantly reduced the uncertainty in yield estimation. Overall, this study demonstrated the potential of using the SAR-derived PD in improving the rice yield estimation over rainfed paddy fields.

Data augmentation for disruption prediction via robust surrogate models

The goal of this work is to generate large statistically representative data sets to train machine learning models for disruption prediction provided by data from few existing discharges. Such a comprehensive training database is important to achieve satisfying and reliable prediction results in artificial neural network classifiers. Here, we aim for a robust augmentation of the training database for multivariate time series data using Student $t$ process regression. We apply Student $t$ process regression in a state space formulation via Bayesian filtering to tackle challenges imposed by outliers and noise in the training data set and to reduce the computational complexity. Thus, the method can also be used if the time resolution is high. We use an uncorrelated model for each dimension and impose correlations afterwards via colouring transformations. We demonstrate the efficacy of our approach on plasma diagnostics data of three different disruption classes from the DIII-D tokamak. To evaluate if the distribution of the generated data is similar to the training data, we additionally perform statistical analyses using methods from time series analysis, descriptive statistics and classic machine learning clustering algorithms.

Prediction of Salinity of Nakdong River Estuary Using Deep Learning Algorithm (LSTM) for Time Series Analysis

Nakdong river estuary is being operated with the goal of expanding the period of seawater inflow from this year to 2022 every month and creating a brackish water area within 15 km of the upstream of the river bank. In this study, the deep learning algorithm Long Short-Term Memory (LSTM) was applied to predict the salinity of the Nakdong Bridge (about 5 km upstream of the river bank) for the purpose of rapid decision making for the target brackish water zone and prevention of salt water damage. Input data were constructed to reflect the temporal and spatial characteristics of the Nakdong River estuary, such as the amount of discharge from Changnyeong and Hamanbo, and an optimal model was constructed in consideration of the hydraulic characteristics of the Nakdong River Estuary by changing the degree according to the sequence length. For prediction accuracy, statistical analysis was performed using the coefficient of determination (R-squred) and RMSE (root mean square error). When the sequence length was 12, the R-squred 0.997 and RMSE 0.122 were the highest, and the prior prediction time showed a high degree of R-squred 0.93 or more until the 12-hour interval.

A MVMD–MMFE algorithm and its application in the flow patterns identification of horizontal oil–water two-phase flow

Abstract Aiming to extract the main information features of fluid multivariate conductance signals and identify the flow patterns under different flow velocities, we present a multichannel time series analysis algorithm based on the multivariate variational mode decomposition (MVMD) and multivariate multiscale fuzzy entropy (MMFE). Firstly, by simulating a multichannel complex signal and performing a series of sensitivity experiments within various noise intensities, we prove the feasibility of the MVMD in chaotic time series. Then, we employ the MVMD to decompose multivariate conductance signals into the intrinsic mode function (IMF) and calculate the MMFE of the IMFs for different flow patterns. Meanwhile, the multivariate empirical mode decomposition (MEMD) is also applied on the comparison of signal decomposition. Finally, we discuss the classification consequence under different mode values k to realize the optimal decomposition. The experimental results show that the MVMD–MMFE algorithm can extract the main information of fluid multichannel signals and distinguish three horizontal oil–water flow patterns effectively, which provides an idea for studying the nonlinear characteristics of the chaotic system.

CUDA-based parallelization of time-weighted dynamic time warping algorithm for time series analysis of remote sensing data

The time weighted dynamic time warping (TWDTW) algorithm is an important algorithm for the time series analysis of remote sensing images. However, due to the limitation of computational complexity, it is difficult to apply this algorithm to large datasets. Therefore, combined with the characteristics of the TWDTW algorithm. This paper proposes a time weighted dynamic warping time parallel algorithm based on CUDA. First, a GPU many-core stream processor is used to process independent pixel data in parallel, and the cumulative cost matrix is established through a multithreading architecture to reduce the dependence between data. Then, the memory model is optimized to reduce memory access transactions in order to improve the memory access efficiency of the algorithm. Taking the Sentinel-2 remote sensing image as experimental data, the algorithm based on different time series lengths, data scales and thread organizations is experimentally verified, and the execution performance of the algorithm under different conditions is explored and analyzed. The experimental results show that this method can significantly improve the computational efficiency of the algorithm.

Sequence-to-Sequence Learning for Prediction of Soil Temperature and Moisture

Soil temperature and moisture play a significant influence on vegetation and climate. Measurements of soil temperature and moisture can now be obtained through soil temperature and moisture sensors, but the measurements can only be done for a specified period of time in the past. Predicting soil temperature and moisture for the future is of significant importance because it can provide guidance on making plant plans. With deep learning algorithms, the characteristics of soil temperature and moisture in a certain upcoming period can be predicted from data measured in the past. In this letter, we develop a sequence-to-sequence learning model for multistep ahead prediction, which features a long short-term memory (LSTM)-based encoder&#x2013;decoder structure for modeling long-term temporal correlations as well as an autoencoder for modeling spatial correlations (i.e., considering data from neighboring locations). For performance evaluation, we use real data collected by the temperature and moisture sensors on soils at the Heihe River Basin in the west of China. Our experimental results show that: 1) the proposed model significantly outperforms two widely used time series analysis algorithms, autoregressive integrated moving average (ARIMA) and support vector regression (SVR), and 2) the proposed spatial encoder&#x2013;decoder LSTM modeling method is indeed effective in that the root-mean-squared error (RMSE) and mean absolute error (MAE) are only 0.22 &#x00B0;C and 0.17 &#x00B0;C for 24-h (144 steps ahead) prediction of soil temperature and the values are only 0.28&#x0025; and 0.23&#x0025; for soil moisture.

A versatile computational algorithm for time-series data analysis and machine-learning models

Here we introduce Local Topological Recurrence Analysis (LoTRA), a simple computational approach for analyzing time-series data. Its versatility is elucidated using simulated data, Parkinsonian gait, and in vivo brain dynamics. We also show that this algorithm can be used to build a remarkably simple machine-learning model capable of outperforming deep-learning models in detecting Parkinson’s disease from a single digital handwriting test.