Short-term Prediction Performance Research Articles

Enhancing traffic flow prediction: Dual-branch graph convolutional network with rough data inference and adaptive spatial dependencies

Traffic flow prediction is a significant application of deep learning in spatio-temporal forecasting analysis. Existing research faces challenges that hinder prediction accuracy. One challenge is the inadequate capturing of spatial dependencies in traffic flow due to fixed pre-defined graph structures. Moreover, manually designed prior graphs still have limitations in extracting spatial features. Another challenge is the instability in short-term prediction performance when pre-defined graphs are completely abandoned in favor of parameter training. Additionally, ordinary RNN sequence convolution methods also struggle to capture long-term sequential patterns in large historical traffic data, leading to gradient vanishing or exploding issues. To address these challenges, we proposes a graph convolutional network for traffic flow prediction. We combine an improved prior graph with an adaptive graph to form a dual-branch spatio-temporal neural network. In the first branch, we introduce a time graph based on rough data inference to complement the predefined static graph. In the second branch, we construct an adaptive learning framework that dynamically learns the adjacency matrix and captures global road information. By utilizing enhanced multi-scale gated convolution, we extract spatio-temporal dependencies. Our method surpasses most baseline approaches according to extensive experiments conducted on public datasets.

Evaluation of Bayesian spatiotemporal infectious disease models for prospective surveillance analysis

BackgroundCOVID-19 brought enormous challenges to public health surveillance and underscored the importance of developing and maintaining robust systems for accurate surveillance. As public health data collection efforts expand, there is a critical need for infectious disease modeling researchers to continue to develop prospective surveillance metrics and statistical models to accommodate the modeling of large disease counts and variability. This paper evaluated different likelihoods for the disease count model and various spatiotemporal mean models for prospective surveillance.MethodsWe evaluated Bayesian spatiotemporal models, which are the foundation for model-based infectious disease surveillance metrics. Bayesian spatiotemporal mean models based on the Poisson and the negative binomial likelihoods were evaluated with the different lengths of past data usage. We compared their goodness of fit and short-term prediction performance with both simulated epidemic data and real data from the COVID-19 pandemic.ResultsThe simulation results show that the negative binomial likelihood-based models show better goodness of fit results than Poisson likelihood-based models as deemed by smaller deviance information criteria (DIC) values. However, Poisson models yield smaller mean square error (MSE) and mean absolute one-step prediction error (MAOSPE) results when we use a shorter length of the past data such as 7 and 3 time periods. Real COVID-19 data analysis of New Jersey and South Carolina shows similar results for the goodness of fit and short-term prediction results. Negative binomial-based mean models showed better performance when we used the past data of 52 time periods. Poisson-based mean models showed comparable goodness of fit performance and smaller MSE and MAOSPE results when we used the past data of 7 and 3 time periods.ConclusionWe evaluate these models and provide future infectious disease outbreak modeling guidelines for Bayesian spatiotemporal analysis. Our choice of the likelihood and spatiotemporal mean models was influenced by both historical data length and variability. With a longer length of past data usage and more over-dispersed data, the negative binomial likelihood shows a better model fit than the Poisson likelihood. However, as we use a shorter length of the past data for our surveillance analysis, the difference between the Poisson and the negative binomial models becomes smaller. In this case, the Poisson likelihood shows robust posterior mean estimate and short-term prediction results.

Based on the Vasicek Model Error Analysis of the New Energy Power Generation and Energy Storage Allocation Decision Method

New energy unit output prediction error probability characteristics analysis and decision alternatives to the system are of great significance. The conventional analytical model based on a specific analytical form is not flexible enough to describe the distribution law of the actual output of new energy, and there are errors. In order to suppress the impact of uncertainty, centralized new energy usually needs to be equipped with energy storage applications with sufficient capacity, thus affecting economic benefits. Therefore, from the point of view of stochastic process, according to a special stochastic differential equation Vasicek model, this paper gives the parameter estimation method of the new energy output model, avoids the specific analytical expression of the probability distribution, and proposes an improved energy storage allocation method based on this model. The example analysis shows that the proposed algorithm has better short-term prediction performance, which makes the decision of energy storage capacity allocation more accurate and economical.

A Real-Time Linear Prediction Algorithm for Detecting Abnormal BDS-2/BDS-3 Satellite Clock Offsets

Due to space environment interference, imperfect data processing model, and the performance of atomic clocks, real-time satellite clock products often contain outliers or irregular biases. We propose a real-time linear moving short-term prediction algorithm to predict clock offsets and detect abnormalities. The proposed algorithm mainly includes phase/frequency anomaly detection and real-time prediction part. Both the phase and frequency domains are used to detect abnormal clock offsets with previous epochs for building the clock prediction model accurately. The real-time moving prediction module utilizes the high short-term prediction performance to check the clock abnormality. The performance of the algorithm is then evaluated for all satellites with real-time estimated satellite clock offsets. To verify the feasibility and effectiveness of the proposed linear moving model and algorithm, the results of the grey model GM(1,1) and the ARIMA model are also compared. The experimental results indicated that the algorithm can detect clock outliers, frequency modulation, and phase jumps, and the linear model has a better clock performance improvement. After the abnormalities are removed with the proposed algorithm, the average STD accuracy of the real-time clock offsets for all satellites is improved by 15.5%, compared to an improvement of 11.4% by the GM(1,1) model and 11.5% by the ARIMA model. The PPP results demonstrate that the proposed clock prediction algorithm improves the positioning accuracy by 8.1%, 13.3%, and 16.9% in the east, north, and up components, respectively.

A systematic data-driven approach for production forecasting of coalbed methane incorporating deep learning and ensemble learning adapted to complex production patterns

Coalbed methane(CBM) as an essential component of clean energy is of strategic significance to global sustainable development, with its production forecast being the basis for reservoirs development. However, due to complex mass transfer mechanism under different geological conditions, traditional physics-based methods are difficult to cope with the complex and various CBM production behavior. Considering the actual production characteristics, this study innovatively combines supervised learning and unsupervised learning to develop an autonomous data governance system based on Local Outlier Factor(LOF) and eXtreme Gradient Boosting(Xgboost), which can perform anomaly detection of dynamic CBM data as well as missing value supplement. Additionally, a data-driven production forecasting model is designed incorporating a Bi-directional Long-Short-Term-Memory(Bi-LSTM) to tackle the various production curve patterns of CBM wells. The production curves from 782 wells of real CBM reservoirs were classified into I, II, III, and IV patterns in terms of their characteristics, with percentages of 32%, 29%, 20%, and 19% respectively. The results reveal that the LOF-Xgboost governance system not only enhances the short-term prediction performance but also significantly improves the long-term prediction robustness. The improved Bi-LSTM-based production model of CBM shows more powerful adaptability and accuracy for both conventional curve patterns and irregular production behavior.

Incremental Capacity Curve Health-Indicator Extraction Based on Gaussian Filter and Improved Relevance Vector Machine for Lithium–Ion Battery Remaining Useful Life Estimation

Accurate prediction of the remaining useful life (RUL) of lithium–ion batteries is the focus of lithium–ion battery health management. To achieve high–precision RUL estimation of lithium–ion batteries, a novel RUL prediction model is proposed by combining the extraction of health indicators based on incremental capacity curve (IC) and the method of improved adaptive relevance vector machine (RVM). First, the IC curve is extracted based on the charging current and voltage data. To attenuate the noise effects on the IC curve, Gaussian filtering is used and the optimal filtering window is determined to remove the noise interference. Based on this, the peak characteristics of the IC curve are analyzed and four groups of health indicators are extracted, and the strong correlation between health indicators and capacity degradation is determined using Pearson correlation analysis. Then, to optimize the traditional fixed kernel parameter RVM model, an RVM regression model whose kernel parameters are optimized by the Bayesian algorithm is established. Finally, four sets of datasets under CS2 battery in the public dataset of the University of Maryland are carried out for experimental validation. The validation results show that the improved RVM model has better short–term prediction performance and long–term prediction stability, the RUL prediction error is less than 20 cycles, and the mean absolute error is less than 0.02. The performance of the improved RVM model is better than that of the traditional RVM model.

Temporal self-attention-based Conv-LSTM network for multivariate time series prediction

Time series play an important role in many fields, such as industrial control, automated monitoring, and weather forecasting. Because there is often more than one variable in reality problems and they are related to each other, the multivariable time series (MTS) introduced. Using historical observations to accurately predict MTS is still very challenging. Therefore, a new time series prediction model proposed based on the temporal self-attention mechanism, convolutional neural network and long short-term memory (Conv-LSTM). When the standard attention mechanism for time series is combined with recurrent neural network (RNN), it heavily depends on the hidden state of the RNN. Particularly in the first time step, the initial hidden state (typically 0) must be artificially introduced to calculate the attention weight of that step, which results in additional noise in the calculation of the attention weight. To address this problem and increase the flexibility of the attention layer, a new self-attention mechanism designed to extract the temporal dependence of the MTS, which called temporal self-attention. In this attention mechanism, long short-term memory (LSTM) adopted as a sequence encoder to calculate the query, key, and value to obtain a more complete temporal dependence than standard self-attention. Because of flexibility of this structure, the DA-Conv-LSTM model was improved, in which a SOTA attention-based method used for MTS prediction. Our improved model compared with six baseline models on multiple datasets (SML2010 and NASDAQ100), and applied to satellite state prediction (our private dataset). The effectiveness of our temporal self-attention was demonstrated by experiments. And the best short-term prediction performance was achieved by our improved model.

A Multi-Network based Hybrid LSTM model for ionospheric anomaly detection: A case study of the Mw 7.8 Nepal earthquake

We propose a Multi-Network-based Hybrid Long Short Term Memory (N-LSTM) model for ionospheric anomaly detection to forecast highly irregular data of the ionospheric Total Electron Content (TEC). Previously purposed models were suffering from the problems of vanishing gradient, exploding gradient, uncertainty, and parameter bias. LSTM is used to overcome the vanishing and exploding gradient problems and the remaining two (uncertainty and parameter bias) are subjugated by N-LSTM. We have also discussed the implementation of the model by using the Mw 7.8 Nepal Earthquake (EQ) that occurred on April 25, 2015. The proposed model detects a significant negative anomaly about 14 days before the impending EQ. Furthermore, we find gradual increments in the TEC time-series, where the TEC values gradually increased, in terms of positive anomalies, for four continuous days (April 21–24, 2015) till the mainshock. Further analysis revealed that the nighttime TEC enhancements are more dominant than the daytimes, which persisted for 10 h (22:00–08:00 LT) in the absence of the solar flux, we named them the absolute seismic precursors. These gradual enhancements establish a perfect correlation with several atmospheric anomalies, reported in the previous studies, where the variations in different atmospheric parameters like Outgoing Longwave Radiation (OLR), air temperature, Ozone, and sub-ionospheric Very-Low Frequency (VLF), were reported for 4 consistent days till the EQ day. The N-LSTM result was compatible with a 30-day running median, which is the classical method of ionospheric anomaly detection. We also examined the planetary K-index (Kp), disturbance storm-time (Dst), solar radio flux (F10.7), and solar wind speed (VSW) indices to check the possible attribution of space weather on the ionosphere during the analysis. The results showed that the anomalies of the ionospheric TEC were more dominantly caused by the EQ phenomenon than the space weather during the nighttime hours. Also, N-LSTM provided sufficient short-term prediction performance of the ionospheric TEC and anomaly detection.

A Novel Combined Model for Short-Term Emission Prediction of Airspace Flights Based on Machine Learning: A Case Study of China

In order to improve the capability of situational awareness and operational efficiency by considering environmental impact, a prediction model for short-term flight emissions within en route airspace is proposed in this paper. First, the measurement method of fuel consumption and flight emissions based on actual meteorological data is established, and the pattern of flight emissions is analyzed. Then, an adaptive weighting approach is proposed by considering prediction results obtained from a long–short term memory (LSTM) prediction model and extreme gradient boosting (XGBoost) prediction model, respectively. Taking the Guangzhou area control centre (ACC) AR05 sector in central and southern China as an example, the model is trained and tested on emission datasets with three statistical scales, 60 min, 30 min, and 15 min. The result shows that the combined variable–weight prediction model has the greatest prediction effect compared to six other models. In terms of time scale, the prediction performance is best on the 60 min statistical scale dataset; larger statistical unit magnitudes of emissions during the predicting process show better short-term prediction performance. In addition, the increase in data features when training the model plays an essential role in promoting model accuracy. The model established in this paper has high prediction accuracy and stability, which is capable of providing short-term prediction of airspace flight emissions.

Short-Term Infectious Diarrhea Prediction Using Weather and Search Data in Xiamen, China

Infectious diarrhea has high morbidity and mortality around the world. For this reason, diarrhea prediction has emerged as an important problem to prevent and control outbreaks. Numerous studies have built disease prediction models using large-scale data. However, these methods perform poorly on diarrhea data. To address this issue, this paper proposes a parsimonious model (PM), which takes historical outpatient visit counts, meteorological factors (MFs) and Baidu search indices (BSIs) as inputs to perform prediction. An experimental evaluation was done to compare the short-term prediction performance of ten algorithms for four groups of inputs, using data collected in Xiamen, China. Results show that the proposed method is effective in improving the prediction accuracy.

Comparative Study: Common ANN and LS‐SVM Exchange Rate Performance Prediction

Due to the mathematical modeling principle deficiency, the data-driven neural network and support vector machine methods have become the powerful basic methods for the exchange rate prediction. Based on the analysis of the characteristics of exchange rate time series data, the exchange rate prediction performance of Artificial neural network (ANN) and Least squares-Support vector machine (LS-SVM) is explored. The parameter optimization method of the two-times training is proposed. The fundamental principle of LS-SVM prediction is analysed in detail. By virtue of daily, monthly and quarterly data of three currency exchange rates, the prediction performance of LS-SVM is examined. The comparison is made with ANN prediction results based on the same data in relevant literature review. According to the experimental result, LS-SVM has better short-term prediction performance, and it is superior to ANN in most cases in terms of prediction precision.

Satellite lithium-ion battery remaining useful life estimation with an iterative updated RVM fused with the KF algorithm

Lithium-ion batteries have become the third-generation space batteries and are widely utilized in a series of spacecraft. Remaining Useful Life (RUL) estimation is essential to a spacecraft as the battery is a critical part and determines the lifetime and reliability. The Relevance Vector Machine (RVM) is a data-driven algorithm used to estimate a battery’s RUL due to its sparse feature and uncertainty management capability. Especially, some of the regressive cases indicate that the RVM can obtain a better short-term prediction performance rather than long-term prediction. As a nonlinear kernel learning algorithm, the coefficient matrix and relevance vectors are fixed once the RVM training is conducted. Moreover, the RVM can be simply influenced by the noise with the training data. Thus, this work proposes an iterative updated approach to improve the long-term prediction performance for a battery’s RUL prediction. Firstly, when a new estimator is output by the RVM, the Kalman filter is applied to optimize this estimator with a physical degradation model. Then, this optimized estimator is added into the training set as an on-line sample, the RVM model is re-trained, and the coefficient matrix and relevance vectors can be dynamically adjusted to make next iterative prediction. Experimental results with a commercial battery test data set and a satellite battery data set both indicate that the proposed method can achieve a better performance for RUL estimation.

Combined logarithmic and exponential function model for fitting postseismic GNSS time series after 2011 Tohoku-Oki earthquake

The time series of a postseismic deformation is commonly fitted by a logarithmic or exponential decay function. However, the high-quality postseismic Global Navigation Satellite System (GNSS) time series of the 2011 Mw 9 Tohoku-Oki earthquake indicates that a single decay function cannot be used to represent the postseismic behaviour. We therefore combined the logarithmic (log) and exponential (exp) decay functions and developed methods for obtaining global solutions using nonlinear least squares calculations for such complex functions. Our models significantly improved the fitting performance of the postseismic time series and the prediction performance of the evolution of postseismic deformation. The solutions obtained by the proposed models and methods enabled distinction between the contributions of the log and exp functions, and explanation of characteristic phenomena such as the subsidence that occurs immediately after an earthquake is reversed to an uplift. The analysis of the solutions may suggest that there has been a continuous increase in the contribution of viscoelastic relaxation to postseismic deformation in eastern Japan, whereas the contribution of afterslip has rapidly decreased. The short-term prediction performance and the universal applicability of the proposed models to the Tohoku-Oki earthquake have contributed to the detection of a slow-slip event in the Tokai region. Rather than the existence of a unique single relaxation time for each surface site, our results suggest a unique single relaxation time for each postseismic deformation mechanism at a given subsurface location. Although the predictions were highly dependent on the assigned steady velocities and the long-term relaxation time constants, they indicate that the coseismic subsidence of the Yamoto station in Miyagi prefecture will recover around the year 2020. The estimated relaxation time constants of the present models appeared to be uniform throughout eastern Japan.

Short-term prediction of photovoltaic energy generation by intelligent approach

Population growth and quickly depleting fossil fuel reserves are creating demand for the development and use of renewable energy resources such as solar energy. The evaluation and forecasting of energy demands have become concerns for facility managers, and predicting energy generation plays a critical role in power-system management, scheduling, and dispatch operations. A reliable energy supply forecast helps to prevent unexpected loads and provides vital information for decisions made on energy generation and purchase. However, study of energy generation prediction by the photovoltaic (PV) system has been limited over the years, especially concerning short-term predictions. This study will adopt the artificial neural network (ANN) to mimic the nonlinear correlation between the metrological parameters and energy generated by the PV system. It aims to find that short-term prediction performance is comparable with real-time prediction performance when ahead solar angles are applied to the predictions.