Time Series Prediction Research Articles

Comparative Analysis of LSTM Variants for Stock Price Forecasting on NSE India: GRU's Dominance and Enhancements

The intricate realm of time series prediction using stock market datasets from the NSE India is delved into by this research. The supremacy of LSTM architecture for forecasting in time series is initially affirmed, only for a paradigm shift to be encountered when exploring various LSTM variants across distinct sectors on the NSE (National Stock Exchange) of India. Prices of various stocks in five different sectors have been predicted using multiple LSTM model variants. Contrary to the assumption that a specific variant would excel in a particular sector, the Gated Recurrent Unit (GRU) emerged as the top performer, prompting a closer examination of its limitations and subsequent enhancement using technical indicators. The ultimate objective is to unveil the most effective model for predicting stock prices in the dynamic landscape of NSE India.

Application Analysis of Big Data-Driven Algorithm Music Creation Generation System

In the field of music, music generation systems based on big data-driven algorithms have become an important research direction. This paper aims to explore the application of big data-driven algorithmic music generation systems in music creation and music education. Through the statistical analysis of the user’s operation record of the music, and the prediction of the artist’s music listening volume in the next stage, it can be judged that the artist’s music listening volume is the highest in a period of time, these artists, as music ambassadors of their respective eras, not only represent a sound, but are also leaders and promoters of music trends. In order to more accurately grasp and predict the dynamics of the music market, especially the popularity of specific artists, this paper proposes an innovative data analysis method that combines quadratic exponential smoothing, autoregressive moving average (ARIMA) model, and BP (backpropagation) neural network model. Quadratic exponential smoothing is a time series prediction technique that predicts future trends by smoothing historical data. This method is particularly effective in handling data with obvious trend and seasonal characteristics, and is also applicable for predicting the number of artist auditions in the music industry. It can help us capture long-term trends in the popularity of artists and provide a basis for prediction. This paper discusses the application prospect of a music generation system driven by big data algorithms in music creation and music education.

Maximizing Forecasting Precision: Empowering Multivariate Time Series Prediction with QPCA-LSTM

Maximizing Forecasting Precision: Empowering Multivariate Time Series Prediction with QPCA-LSTM

FPGA-based Block Minifloat Training Accelerator for a Time Series Prediction Network

Time series forecasting is the problem of predicting future data samples from historical information and recent deep neural networks (DNNs) based techniques have achieved excellent results compared with conventional statistical approaches. Many applications at the edge can utilise this technology and most implementations have focused on inference, an ability to train at the edge would enable the deep neural network (DNN) to adapt to changing conditions. Unfortunately, training requires approximately three times more memory and computation than inference. Moreover, edge applications are often constrained by energy efficiency. In this work, we implement a block minifloat (BM) training accelerator for a time series prediction network, N-BEATS. Our architecture involves a mixed precision GEMM accelerator that utilizes block minifloat (BM) arithmetic. We use a 4-bit DSP packing scheme to optimize the implementation further, achieving a throughput of 779 Gops. The resulting power efficiency is 42.4 Gops/W, 3.1x better than a graphics processing unit in a similar technology.

Short-Term Traffic Flow Prediction: Techniques and Methods

With the development of the economy, the expansion of the city scale, the increase in the population of urban residents and the increase in the per capita car ownership, the traffic congestion problem has an increasing impact on urban operation. Especially during the peak period of commuting traffic and holidays in large cities, most cities are facing the serious problem of road congestion. In order to facilitate the travel of residents and reduce air pollution, it is very important to keep the main road of the city smooth and rapid passage of vehicles. Therefore, cities should vigorously develop short-time traffic flow prediction, and accurate prediction of short-time traffic flow is the key to realizing intelligent urban traffic operation and efficient traffic flow management improvement. This paper will talk about the time series prediction model of short-time traffic flow prediction methods and advantages and disadvantages to provide auxiliary support for solving the traffic congestion problem, provide efficient travel path decision-making guidance for urban residents, facilitate the saving of urban people's travel costs, and also provide effective police deployment decision-making information for the traffic management department.

Recurrent Neural Network for Quantitative Time Series Predictions of Bridge Condition Ratings

Traditional forecasting models for bridge conditions, such as ARIMA and Markov chains, often fail to adequately capture nonlinear and dynamic relationships among critical variables like age, traffic patterns, and environmental factors, leading to suboptimal maintenance decisions, increased long-term maintenance costs, and heightened safety risks. This study addresses these limitations by developing recurrent neural network (RNN) models utilizing Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures with a TimeDistributed output layer. This novel approach enables accurate forecasting of the Bridge Health Index (BHI) and condition ratings for key components—deck, superstructure, and substructure—while effectively modeling temporal dependencies. Applied to bridge data from Georgia, USA, the regression models (BHI) achieved R2 values exceeding 0.84, while the classification models (components condition ratings) demonstrated accuracy between 84.78% and 87.54%. By modeling complex temporal trends in bridge deterioration, our method processes time-dependent data from multiple bridges simultaneously, revealing intricate relationships that influence bridge performance within a state’s inventory. These results provide actionable insights for maintenance planning, optimized resource allocation, and reduced risks of unexpected failures. This research establishes a robust framework for bridge performance prediction, ensuring improved infrastructure safety and resilience amid aging assets and constrained maintenance budgets.

DCFA-iTimeNet: Dynamic cross-fusion attention network for interpretable time series prediction

DCFA-iTimeNet: Dynamic cross-fusion attention network for interpretable time series prediction

Chaotic time series prediction of motor system based on WOA-MFRFNN

Abstract The complexity of the operation of permanent magnet synchronous motor (PMSM) has heightened interest in its stability and reliability. Evidence suggests that chaotic oscillatory behavior can occur during PMSM operation, making the prediction of such chaotic motion crucial for maintaining stable motor performance. In this paper, a WOA-MFRFNN (Whale Optimization Algorithm-Multi Functional Recurrent Fuzzy Neural Network) model is proposed to predict chaotic time series of PMSM, where WOA is used to train the network weights and the MFRFNN contains two TSK fuzzy rules for fuzzy neural networks, which are used for the output and the state of network, respectively. The experimental results of WOA-MFRFNN are compared with those of SSA-BP, ROA-RF and iTransformer. The results imply that the WOA-MFRFNN outperforms other methods and the RMSE can reach 5.45×10-4. It is also shown that the WOA-MFRFNN model exhibits good predictive effect on the real-world PMSM dataset.


Spatial–Temporal Reconstruction of Trajectories in Free Space Using Automatic Target Position Detection Data

The monitoring technology for targets such as aircraft and vehicles has rapidly developed in recent years and is widely used in national airspace security supervision, urban traffic supervision, and the tracking of special targets. However, the sparse trajectories of targets, primarily caused by the insufficient density of monitoring points, significantly reduce their usability. Therefore, it is important to reconstruct the target trajectories. Existing methods for the reconstruction of target trajectories often rely on topological data and convert trajectory reconstruction into a trajectory matching problem. Such methods heavily rely on topological data and cannot reconstruct trajectories in free space. To address this issue, we proposed a trajectory reconstruction method, named Prob-Attn, which does not rely on topological data and can accurately reconstruct target trajectories in free space. This method can be divided into two steps: first, a spatial trajectory construction module is proposed to determine the spatial trajectories of targets. Then, based on the reconstructed spatial trajectory of the target, this paper proposes a time series prediction model based on historical trajectories and an attention mechanism, which considers the impact of the target’s activity cycle and the surrounding status to predict the time series inside the trajectory. Finally, the proposed method is evaluated on real automatic vehicle detection datasets collected in Chongqing, China. The experimental results show that, compared with traditional methods, the proposed method can reconstruct the spatiotemporal trajectory of the target more accurately. The reconstructed trajectory data can be used for critical applications such as the intent and behavior analysis of key targets in national airspace and ground areas, providing valuable insights into security and safety.

Generating synthetic turbulence with vector autoregression of proper orthogonal decomposition time coefficients

This study introduces vector autoregression (VAR) as a linear procedure that can be used for synthesizing turbulence time series over an entire plane, allowing them to be imposed as an efficient turbulent inflow condition in simulations requiring stationary and cross-correlated turbulence time series. VAR is a statistical tool for modelling and prediction of multivariate time series through capturing linear correlations between multiple time series. A Fourier-based proper orthogonal decomposition (POD) is performed on the two-dimensional (2-D) velocity slices from a precursor simulation of a turbulent boundary layer at a momentum thickness-based Reynolds number, $Re_{\theta }=790$ . A subset of the most energetic structures in space are then extracted, followed by applying a VAR model to their complex time coefficients. It is observed that VAR models constructed using time coefficients of 5 and 30 most energetic POD modes per wavenumber (corresponding to $66\,\%$ and $97\,\%$ of turbulent kinetic energy, respectively) are able to make accurate predictions of the evolution of the velocity field at $Re_{\theta }=790$ for infinite time. Moreover, the 2-D velocity fields from the POD–VAR when used as a turbulent inflow condition, gave a short development distance when compared with other common inflow methods. Since the VAR model can produce an infinite number of velocity planes in time, this enables reaching statistical stationarity without having to run an extremely long precursor simulation or applying ad hoc methods such as periodic time series.

A neuromorphic radial-basis-function net using magnetic bits for time series prediction

Magnetic tunnel junctions (MTJs) are considered strong candidates for constructing neuromorphic systems owing to their low power consumption and high integrability. However, research on MTJ-based local approximation network is still lacking. In this work, we propose an MTJ-based radial basis function (RBF) network and numerically investigate its time-series prediction capability. The results demonstrate that the MTJ-based RBF network can enhance its prediction performance by utilizing increased environmental temperatures, achieving performance better than traditional software artificial neural networks.

Time Series Modelling and Prediction of WTI Crude Oil Prices Trend using Machine Learning Techniques

It is challenging to forecast crude oil prices. Given that the international crude oil price is time-varying and nonlinear, suggest a new mixed method for oil price forecasting. First, the traditional time series technique ARIMA method for predictions is incorporated. Then, The time-varying and nonlinear components of crude oil prices are forecasted using machine learning approaches such as boosted trees, SVM, ANN, LSTM, bagged trees and linear regression, etc. The ultimate predicted outcomes of crude oil prices are then calculated by adding the predicted prices of each component. The results showed that the new proposed hybrid method was capable of a high degree of oil price prediction based on its exceptional ability to adapt sample selection, data frequencies and structural breaks. In addition, the comparisons show that the new method proves to be more accurate than these well-known methods for crude oil price forecasts. The WTI daily data from 1986 to 2018 has been utilised for forecasting of crude oil trend for 2018 in next 50 days. Forecasts were made using the time series forecasting approach with the ARIMA model and the machine learning approach. Validation metrics have been used to evaluate the performance of these models and it was noted that they performed well. The effects of prediction methods on crude oil data are discussed with the help of graphs for the predictions for next 50 days, actual and predicted plots, model performance plot, and validation performance plot for all the methods. Compared to the previous published work, the accuracy of the problem has been verified by comparing it to the quantitative outputs exhibited in this, and the agreement between the results is excellent, which has led to confidence in the quantitative results projected in this study.. KEYWORDS :Crude oil price, ARIMA model, Time series forecasting, Machine learning techniques, Error metrics.

Improved fractional-order gradient descent method based on multilayer perceptron

The fractional-order gradient descent (FOGD) method has been employed by numerous scholars in Artificial Neural Networks (ANN), with its superior performance validated both theoretically and experimentally. However, current FOGD methods only apply fractional-order differentiation to the loss function. The application of FOGD based on Autograd to hidden layers leverages the characteristics of fractional-order differentiation, significantly enhancing its flexibility. Moreover, the implementation of FOGD in the hidden layers serves as a necessary foundation for establishing a family of fractional-order deep learning optimizers, facilitating the widespread application of FOGD in deep learning. This paper proposes an improved fractional-order gradient descent (IFOGD) method based on Multilayer Perceptron (MLP). Firstly, a fractional matrix differentiation algorithm and its fractional matrix differentiation solver is proposed based on MLP, ensuring that IFOGD can be applied within the hidden layers. Subsequently, we overcome the issue of incorrect backpropagation direction caused by the absolute value symbol, ensuring that the IFOGD method does not cause divergence in the value of the loss function. Thirdly, fractional-order Autograd (FOAutograd) is proposed based on PyTorch by reconstructing Linear layer and Mean Squared Error Loss module. By combining FOAutograd with first-order adaptive deep learning optimizers, parameter matrices in each layer of ANN can be updated using fractional-order gradients. Finally, we compare and analyze the performance of IFOGD with other methods in simulation experiments and time series prediction tasks. The experimental results demonstrate that the IFOGD method exhibits performances.

A systematic study of Echo State Networks topologies for chaotic time series prediction

A systematic study of Echo State Networks topologies for chaotic time series prediction

An Improved Parallel Heterogeneous Long Short-Term Memory Model with Bayesian Optimization for Time Series Prediction

Currently, Deep Learning (DL) with the Recurrent Neural Networks (RNN) variants is being applied successfully in many domains of Engineering for prediction. In view of the demand for precise forecasting and the aid of Artificial Intelligence Tools, time series prediction reveals a vital task in decision-making and risk assessment. However, the application of novel Recurrent DL models for obtaining an accurate prediction of time series is yet to be explored. Recent trends reveal that Hybrid Neural Networks and DL models are appropriate for time series forecasts. At the same time, the model's selection and the hyperparameter's tuning can greatly impact its performance. To address this problem, a parallel long-term memory (PLSTM) model integrated with Bayesian hyperparameter optimization (PLSTM-BO) is proposed for time series prediction. The model is tuned in terms of key parameters, including the number of neurons, dropout, learning rate, and optimization technique. The model's performance is assessed using the SARS-COVID-19 cumulative cases, deaths, recovery cases, and NIFTY 50 stock closing price time series dataset. The obtained results convey that the current model exhibits remarkable performance compared to existing models.

Intelligent hybrid approaches utilizing time series forecasting error for enhanced structural health monitoring

Over the past decade, the growing importance of machine learning-based structural health monitoring (SHM) for early-stage damage detection has become evident. Time series forecasting, using deep learning, has emerged as a key focus, significantly contributing to improving damage detection, localization, and quantification processes. Researchers in SHM have conducted numerous studies utilizing neural networks based on time series forecasting, grounded in traditional methods. This study diverges from existing research by directly incorporating neural network prediction errors in time series for detecting, localizing, and quantifying damage. The proposed methods are well-suited for online structural monitoring. They eliminate the need for data classification methods and damage-sensitive feature extraction techniques by relying solely on training the neural network with data from structurally sound conditions. However, the testing process does require data from damaged conditions. To address the non-linear and non-stationary characteristics of the signals, the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) method is applied for signal processing. This method processes response signals (i.e., time series) from three well-known benchmark structures: the University of Central Florida structure, the Qatar University Grandstand Simulator, and the Z24 Bridge. Subsequently, the first intrinsic mode function (IMF) obtained from signal decomposition is independently input into Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) neural networks for time series prediction. Optimal parameter values for the LSTM and GRU neural networks are chosen using the Bayesian Optimization (BO) algorithm before the prediction process. By introducing three indices—Statistical Distance Function (SDF), error index, and accuracy index—the evaluation not only emphasizes the accuracy of the methods but also explores the localization and quantification of damage. The results demonstrate that both ICEEMDAN-BO-LSTM-SDF and ICEEMDAN-BO-GRU-SDF methods have successfully achieved accurate detection, localization, and quantification without the need for data classification and damage-sensitive feature extraction methods, and merely by utilizing data from healthy states for neural network training.

Walking Back the Data Quantity Assumption to Improve Time Series Prediction in Deep Learning

Walking Back the Data Quantity Assumption to Improve Time Series Prediction in Deep Learning

Experimental realization of a performance-enhanced reservoir computer based on a photonic-filter feedback laser

Reservoir computing (RC), especially time-delayed RC, as a lightweight, high-speed machine learning paradigm, shows excellent performance in time-series prediction and recognition tasks. Within this framework, time delays play a vital role in dynamic systems, i.e., significantly affecting the transient behavior and the dimensionality of reservoirs. In this work, we explore a multidelay system as the core computational element of RC, which is constructed using a semiconductor laser with photonic-filter feedback. We demonstrate experimentally that the photonic-filter feedback scheme can improve the mapping of scalar inputs into higher-dimensional dynamics, and thus enhance the prediction and classification ability in time series and nonlinear channel equalization tasks. In particular, the rich neural dynamics in turn boosts its memory capacity, which offers great potential for short-term prediction of time series. The numerical results show good qualitative agreement with the experiment. We show that improved RC performance can be achieved by utilizing a small coupling coefficient and eschewing feedback at integer multiples, which can induce detrimental resonance. This work provides an alternative photonic platform to achieve high-performance neural networks based on high-dimensional dynamic systems.

Enhanced photonic reservoir computing using an optically injected VCSEL with random polarized optical feedback.

We propose and numerically demonstrate a photonic time-delay reservoir computing (TDRC) system exhibiting enhanced parallel task processing performance, where an optically injected vertical-cavity surface-emitting laser (VCSEL) under random distributed optical feedback acts as the reservoir computer. To assess its effectiveness, we perform two benchmark tasks including chaotic time-series prediction and waveform recognition task, where the TDRC is associated with two different random feedback structures, i.e., orthogonally polarized optical feedback (OPOF) and parallelly polarized optical feedback (PPOF). Benefiting from the enhanced nonlinearity offered by the random distributed optical feedback, the proposed TDRC excels at parallel task processing with the PPOF structure, whereas the performance of the OPOF structure may be deteriorated. Additionally, we reveal the effect of the injection strength, feedback strength, pump current, and number of virtual nodes on the proposed TDRC. Our work paves the way for the performance enhancement of parallel task processing based on polarization multiplexing in a VCSEL-based TDRC.

Smart Water Management with Digital Twins and Multimodal Transformers: A Predictive Approach to Usage and Leakage Detection

Effective water management is crucial in urban and rural settings, requiring efficient usage and timely detection of issues like leakages for sustainability. This paper introduces an integrated framework that combines Digital Twin technology with a multimodal transformer-based model for accurate water usage prediction and leakage detection. The system synchronizes real-time data from various sensors including flow meters, pressure sensors, and thermal imaging devices with a Digital Twin of the water network. Advanced transformer models, specifically the Informer model for long-term time-series prediction and a Water Multimodal Transformer for anomaly detection, process these data to capture complex patterns and dependencies. Experimental results demonstrate the framework’s effectiveness: the Informer model achieved an R2 score of 0.9995 and a Mean Squared Error (MSE) of 2.2, outperforming traditional models. For leakage detection, the model attained 98.4% accuracy and precision, an F1 score of 97.6%, a low False Positive Rate of 0.0019, and an Area Under the Curve (AUC) of 0.984. By fusing diverse sensor data and utilizing advanced transformer architectures, the framework provides a comprehensive view of the water network, enabling real-time decision-making, enhancing forecasting accuracy, and reducing water waste. This scalable solution supports sustainable water management practices in both urban and industrial contexts.