Short-term Time Series Research Articles

Minutely multi-step irradiance forecasting based on all-sky images using LSTM-InformerStack hybrid model with dual feature enhancement

In recent years, the government has attached great importance to promoting innovation in new energy technologies, particularly in the area of solar power generation. Photovoltaic (PV) power generation has become an essential factor in the promotion of renewable energy due to its inherent safety, environmental sustainability and abundant solar energy resources. The efficiency and stability of PV generation systems are closely related to the accurate prediction of real-time global horizontal irradiance (GHI). However, some existing research methods have made some progress in hourly resolution, but the multi-step prediction accuracy and model generalization ability in high time resolution are still unsatisfactory. Therefore, this paper proposes a novel minutely multi-step irradiance forecasting based on all-sky images using LSTM-InformerStack hybrid model with dual feature enhancement scheme. In this scheme, the advantages of both LSTM and InformerStack models are combined to further enhance the global and local features, which can better capture the short-term time series patterns and long-term dependencies and improve the accuracy of the time series. The experimental validation results show that the performance of this prediction scheme is improved by 11.2%, and 26.0% for nRMSE and R2, respectively, and the 24-step prediction can be improved up to 26.1% vs. the persistent model. Meanwhile, the prediction scheme of this paper is also expected to provide a reference in future multi-step time series accurate prediction with higher resolution.

Research on a novel photovoltaic power forecasting model based on parallel long and short-term time series network

Under the background of the global pursuit of carbon neutrality, the trend of photovoltaic power generation replacing traditional thermal power generation is increasingly apparent. To improve the performance of the model in photovoltaic power forecasting, this study proposed a novel deep learning-based model named PLSTNet for ultra-short-term prediction of photovoltaic power over a 5 min time span. This model is a novel dual-path prediction. On one hand, it effectively captures short-term fluctuations in time series data by combining CNN and RNN. On the other hand, it further captures and analyzes long-term trends in fluctuations through the use of a smoothing layer and RNN's recurrent skip layer. In one-step and multi-step forecasting experiments on annual and seasonal datasets, we compared the performance of the PLSTNet model with LSTNet, PHILNet, TCN_GRU, and ResCNN to assess its performance. In one-step and multi-step forecasting using the annual dataset, the MAE of the PLSTNet model is at least 15.5% lower than that of other models. Similarly, for seasonal datasets, the MAE of the PLSTNet model is at least 13.2% lower than other models. The experimental results demonstrate that in various photovoltaic power forecasting scenarios, the PLSTNet model has achieved higher accuracy in ultra-short-term predictions.

An ensemble neural network model for predicting the energy utility in individual houses

With the rapid increase in the world's human population and the advancement of technology, energy utilization has substantially increased. To sustain a stable stream of energy, it is vital to anticipate power utilization beforehand. This issue makes predicting short-term electrical time series use difficult. In this research, the user proposed a neural network with a deep learning prototype that performs energy prediction by combining. Long Short-Term Memory (LSTM) and Feed Forward Neural Networks (FFNN). In order to enhance the performance, the ensemble FFNN-LSTM network and the Stationary Wavelet Transform (SWT) method are combined to create a hybrid deep learning algorithm. The SWT's reduction of instability and expansion of data dimensionality may improve the FFNN-LSTM forecasting accuracy. The suggested technique's predicting effectiveness is even further improved by the ensemble Long short-term neural network. On the basis of a real-world household energy usage database gathered by the "UK-DALE" project, validation studies were carried out. Researchers determined that the LSTM-FFNN based SWT produced a better result in terms of accuracy and energy utilization than other existing results after contrasting the experimental findings to the baseline, which is the conventional LSTM and Moving Average (MA) based upon the Root Mean Squared Error (RMSE) score.

MPM: Multi Patterns Memory Model for Short-Term Time Series Forecasting

MPM: Multi Patterns Memory Model for Short-Term Time Series Forecasting

Distribution and composition of Decapod larvae assemblages on the Berlengas archipelago and Peniche coast (western coast of Portugal)

For the first time, the decapod larval community in the Berlenga and coastal Peniche region have been investigated in detail, from a short-term time series (2015 – 2020). Sampling was conducted in two locations, at Berlengas archipelago, and near the coast, at Peniche. The communities found at both stations were revealed to be 80% different in terms of species composition, but species of Brachyura, Caridea, and Anomura were the most abundant in both locations. A marked seasonality, with abundance peaks during summer and autumn, was found for the assemblages near the coast, but absent in the assemblages of the Berlenga region. This suggests that factors other than the environmental variables examined in this study may play a crucial role in shaping the assemblages more offshore. Water temperature, upwelling index, salinity, and chlorophyll-a were the main factors dictating species abundance, especially at Peniche. This emphasizes the significant role of the Iberian upwelling productivity, which in turn influences decapod communities. Also, the progressive warming of coastal water will most likely have a negative impact on the decapod meroplanktonic life forms, with further consequences on recruitment and adult life stages.

Short-term multivariate time series load data forecasting at low-voltage level using optimised deep-ensemble learning-based models

Increasing the renewable energy penetration, especially photovoltaic systems, requires accurate and short-term load forecasting for every individual electricity customer. This can significantly help distributed network service providers (DNSP) plan and operate electrical networks and provide better quality and more reliable electricity services to their customers. Due to the strong volatility of the load data, an optimised deep-ensemble learning methodology is proposed to develop several deep learning models to forecast the load data in individual and aggregate load scenarios. The adaptive wind-driven optimisation (AWDO) algorithm is used to tune the hyperparameters of four of the developed deep learning models, significantly improving their performance. Tuning the deep learning models’ hyperparameters shows significant improvements in the accuracy, as well as training, and testing times. In addition, a hybrid ensemble strategy that contains bagging, random Subspace, and boosting (BRSB) with ensemble pruning is adapted in the optimised deep learning-based models to extract deep features from multivariate data. In this context, the bidirectional long-short-term memory optimised by the AWDO algorithm (Bi-LSTM-AWDO) based hybrid ensemble learning forecasting model is compared exhaustively with several benchmark models, including the recently developed models in the state-of-the-art. The average root mean square error (RMSE), and the average mean absolute percentage error (MAPE) of the Bi-LSTM-AWDO model for individual loads are 0.121 kW, and 7.55%, respectively, while they are 0.025 kW, and 1.51% for aggregate loads, respectively. Accordingly, the Bi-LSTM-AWDO model outperforms other models in forecasting short-term load data for individual and aggregate households using actual smart meters’ measurements.

The Cost Realization on Time Frame Shift of Securities

Investors use a variety of investing methods depending on the time scales () of their short-term and long-term investment time horizons (ITH). The nature of the market varies greatly depending on the type of investment. Hurst exponents (H) and normalised variance (NV) techniques based on empirical mode decomposition (EMD) have been used to identify key features of the market over various time horizons. For the stock price's decomposed intrinsic mode functions (IMF), the values of H and NV have been estimated. We found H1 0.5 0.04 and H1 0.75 for the IMFs with time periods ranging from a few days to 3 months and 5 months, respectively. Two time series from the IMFs have been reconstructed based on the value of H1: a) Short-term time series [XST (t)] with H1 = 0.5.b) long-term time series [XLT (t)] with H1 0.75 and a time span of 5 months. The XST (t) and XLT (t) indicators reveal that market dynamics in short-term ITH are random and connected in long-term ITH. We also discovered that the NV is very low in short-term ITH and gradually increases in long-term ITH. The results also reveal that in the long run, stock prices are connected with the company's fundamental factors. The discovery may aid investors in developing investment and trading strategies for both short-term and long-term investment goals.

Data-driven time-variant reliability assessment of bridge girders based on deep learning

This article presents a new data-driven reliability analysis framework through combining the sample convolution and interaction network (SCINet) with Bayesian probability recursion to predict the time-variant reliability of bridge girders. The structural response predicted by SCINet and the state parameter (i.e., the variance of normal distribution) estimated by Bayesian dynamic linear model (BDLM) are used to form the limit state function to predict time-variant reliability. The results with a practical bridge show that the proposed method can predict future structural responses and time-variant reliability more accurately than BDLM, long short-term memory (LSTM), and long- and short-term time-series network (LSTNet) methods.

Exploring Deep Learning Approaches for Short-Term Passenger Demand Prediction

An accurate short-term passenger demand forecast makes a contribution to the coordination of traffic supply and demand. Forecasting the short-term passenger demand for the on-demand transportation service platform is of utmost significance since it might incentivize empty cars to relocate from over-supply regions to over-demand regions. Yet, because spatial, temporal, and exogenous dependencies need to be evaluated concurrently, short-term passenger demand forecasting may be rather difficult. This article aims to investigate several methods that can be utilized to forecast short-term traffic demand, with a primary emphasis on deep learning approaches. We examine varying degrees of temporal aggregation and how these levels affect various architectural configurations. In addition, by analyzing 22 models representing 5 distinct architectural configurations, we illustrate the influence of varying layer configurations within each architecture. The findings indicate that the long-term short memory (LSTM) structures perform the best for short-term time series forecasting, but more complex architectures do not significantly enhance the outcomes. Moreover, considering the spatiotemporal aspects results in an improvement in the prediction of more than fifty percent. In addition, we investigate the vectorization of time, also known as Time2Vec, as a way of embedding to make it possible for a selected algorithm to recognize periodic characteristics in time series, and we show that the outcome is improved by fifteen percent.

A Short-Term Time-Series Data Analysis Algorithm for Flavor Release during the Start of Eating.

"Retronasal aroma" refers to the aroma released from food during consumption and traveling through the nose after leaving the mouth. It is closely related to the behavior of odor compounds released from food into the mouth and plays a crucial role in our overall perception of flavor. As a result, research focusing on measuring the behavior of retronasal aroma has gained attention for exploring the relationship between sensory perception and flavor. We attempted to develop a data analysis method that specifically targets a time span of a few seconds to tens of seconds, starting from when food is placed in the mouth during eating and extending to just after swallowing. In this study, we observed a strong correlation between the periodic waveform data derived from performing the third derivative (jerk) on the detection intensity data obtained using a mass spectrometer and the behavior of the detection intensity. Furthermore, by performing a frequency analysis using a fast Fourier transform on the jerk data, it was possible to extract the frequencies that contribute to sensory perception during eating. Furthermore, the reconstructed jerk data derived from the extracted data using the inverse fast Fourier transform provided a clearer explanation of sensory perception during eating. Our algorithm suggests new short-term time-series data applications.

Day-Ahead Electricity Price Probabilistic Forecasting Based on SHAP Feature Selection and LSTNet Quantile Regression

Electricity prices are a central element of the electricity market, and accurate electricity price forecasting is critical for market participants. However, in the context of increasingly integrated economic markets, the complexity of the electricity system has increased. As a result, the number of factors required to consider in electricity price forecasting is growing. In addition, the high percentage of renewable energy penetration has increased the volatility of electricity generation, making it more challenging to predict prices accurately. In this paper, we propose a probabilistic forecasting method based on SHAP (SHapley Additive exPlanation) feature selection and LSTNet (long- and short-term time-series network) quantile regression. First, to reduce feature redundancy and overfitting, we use the SHAP method to perform feature selection in a high-dimensional input feature set, and specifically analyze the magnitude and manner in which features affect electricity prices. Second, we apply the LSTNet quantile regression model to predict the electricity value under different quantiles. Finally, the probability density function and the prediction interval of the predicted electricity prices are obtained by kernel density estimation. The case of the Danish electricity market validates the effectiveness and accuracy of our proposed method. The accuracy of the proposed method is better than that of other methods, and we assess the importance and direction of the impact of features on electricity prices.

A New Framework for Multivariate Time Series Forecasting in Energy Management System

This paper introduces a new integrated framework for short-term time series forecasting used as decision logic support in Energy Management Systems (EMS). The proposed framework defines data preprocessing, forecast model training, and adaptation techniques. For data preprocessing, Gaussian and Kalman filtering were used in offline learning on historical data and online (real-time) forecasting. Optimal parametrisation for both filters was selected based on the forecast accuracy. Next, a multiple-input, single-output forecast model based on the Long Short-Term Memory (LSTM) neural network was tuned for each device. The forecast horizon is single-step, and chaining of predictions was used for longer horizons. Finally, the best model was adapted to the changes in the time series data. A significant drop was observed in the forecast accuracy on the past data after adapting the model to the new data using a naive approach. Due to this drop, a novel sample rehearsal selection strategy for adaptation of the model is proposed and compared to other different strategies. All the techniques are demonstrated based on the examples of forecasting multivariate time series for different household devices: a water heater, a refrigerator and an HVAC device. The use of data preprocessing improved the forecast accuracy by up to 56% compared to no preprocessing. Moreover, by introducing a sample rehearsal, knowledge retention was improved by up to 76.6% compared to the naive approach, and by up to 62% compared to the random selection of samples.

Temperature Prediction for Expressway Pavement Icing in Winter Based on XGBoost–LSTNet Variable Weight Combination Model

Ice cover on pavement may reduce the road adhesion coefficient and increase the crash risks, which might result in more traffic crashes. The primary factor utilized to assess whether the wet pavement is icy or not is the pavement temperature. Therefore, forecasting pavement temperature is an effective method to judge road conditions and improve traffic safety. This paper proposes a combination model based on the extreme gradient boosting (XGBoost) model and long- and short-term time-series network (LSTNet) model to predict pavement temperature. Pavement temperature and meteorological data were collected for the cities along the Shandong part of the Beijing-Taipei Expressway (G3). In this study, nine meteorological variables were used. Subsequently, after correlation analysis, five variables, including air temperature, dew point temperature, relative humidity, evaporation, and potential evaporation, were selected for prediction. The method proposed in this study comprises the following steps. First, the XGBoost and the LSTNet models are respectively formulated based on the time-varying characteristics of pavement temperatures. Then, using the preset weight of the variable, the XGBoost model is used for preliminary prediction to add features. Finally, the experimental analysis is performed on the Qihe data set after the two models have been integrated using the inverse variance method. As revealed by the experimental results, the mean absolute error (MAE) and root-mean-square error (RMSE) of the proposed XGBoost-LSTNet model are 0.8235 and 1.2412, respectively. Compared with the long short-term memory (LSTM) model, random forest (RF) model, XGBoost model, and LSTNet model, the XGBoost-LSTNet model proposed in this paper has higher accuracy. The study’s findings can successfully increase wintertime expressway traffic safety and serve as a guide for managing maintenance and preventing icing-related accidents.

Unsupervised anomaly detection for long-span bridges combining response forecasting by deep learning with Td-MPCA

This paper proposes an unsupervised anomaly detection (AD) scheme combining response forecasting by deep learning (DL) and temperature-driven moving principal component analysis (Td-MPCA), by only utilizing the monitoring strain data of intact or baseline structures for training. The major contributions and novelties of the proposed method are as follows. First, the bidirectional long- and short-term time-series network with attention mechanism (BiLSTNet-A) model is carefully designed for response forecasting. It is proved to have better capability to predict the structural response more accurately than several state-of-the-art DL models using the monitoring data of a long-span bridge, even in the presence of drastic temperature changes and incomplete data, because BiLSTNet-A can better capture long- and short-term patterns of time series data simultaneously than these models. Second, an BiLSTNet-A-based unsupervised AD approach is developed as compared with the Td-MPCA method using both simulation and monitoring datasets. The results illustrate that BiLSTNet-A is more robust to temperature loading while Td-MPCA is less influenced by damage location. Finally, an unsupervised structural AD scheme is further developed by combining the proposed BiLSTNet-A-based AD and Td-MPCA methods. The results demonstrate that the proposed scheme can provide more reliable and robust AD results than the two methods used separately, and has potential for application in practical engineering.

Remaining Useful Life Prediction for Lithium-Ion Batteries Based on Improved Mode Decomposition and Time Series

Accurately predicting the remaining useful life (RUL) of lithium-ion batteries holds significant importance for their health management. Due to the capacity regeneration phenomenon and random interference during the operation of lithium-ion batteries, a single model may exhibit poor prediction accuracy and generalization performance under a single scale signal. This paper proposes a method for predicting the RUL of lithium-ion batteries. The method is based on the improved sparrow search algorithm (ISSA), which optimizes the variational mode decomposition (VMD) and long- and short-term time-series network (LSTNet). First, this study utilized the ISSA-optimized VMD method to decompose the capacity degradation sequence of lithium-ion batteries, acquiring global degradation trend components and local capacity recovery components, then the ISSA–LSTNet–Attention model and ISSA–LSTNet–Skip model were employed to predict the trend component and capacity recovery component, respectively. Finally, the prediction results of these different models were integrated to accurately estimate the RUL of lithium-ion batteries. The proposed model was tested on two public lithium-ion battery datasets; the results indicate a root mean square error (RMSE) under 2%, a mean absolute error (MAE) under 1.5%, and an absolute correlation coefficient (R2) and Nash–Sutcliffe efficiency index (NSE) both above 92.9%, implying high prediction accuracy and superior performance compared to other models. Moreover, the model significantly reduces the complexity of the series.

Smart solutions for urban health risk assessment: A PM2.5 monitoring system incorporating spatiotemporal long-short term graph convolutional network

Current spatial-temporal early warning systems aim to predict outdoor air quality in urban areas either at short or long temporal horizons. These systems implemented architectures without considering the geographical distribution of each air quality monitoring station, increasing the uncertainty of the forecasting framework. This study developed an integrated spatiotemporal forecasting architecture incorporating an extensive air quality PM2.5 monitoring network and simultaneously forecasts PM2.5 concentrations at all locations, allowing the monitoring of the health risk associated with exposure to these levels. First, this study uses a graph convolutional layer to incorporate the spatial relationship of the neighboring stations at their current state with real-time measurements. Then, it is coupled to a deep learning temporal model to form the long- and short-term time-series graph convolutional network (LSTGraphNet) model, anticipating high pollutant concentration events. This work tested the proposed model with a case study of an existing ambient air quality monitoring network in South Korea. LSTGraphNet model showed prediction performances of PM2.5 at multiple monitoring stations with a mean absolute error (MAE) of 1.82 μg/m3, 4.46 μg/m3, and 4.87 μg/m3 for forecasting horizons of one, three, and 6 h ahead, respectively. Compared to conventional sequential models, this architecture was superior among the state-of-the-art baselines, where the MAE decreased to 41%, respectively. The results of the study showed that the proposed architecture was superior to conventional sequential models and could be used as a tool for decision-making in smart cities by revealing hotspots of higher and lower PM2.5 concentrations in the long term.

Pyodine: an open, flexible reduction software for iodine-calibrated precise radial velocities

Context. Many telescopes use an iodine (I2) absorption cell to measure precise radial velocities (RVs), but their data reduction pipelines are all tailored to their respective instrumental characteristics and not openly accessible. Aims. For existing and future projects dedicated to measuring precise RVs, we have created an open-source, flexible data reduction software to extract RVs from échelle spectra via the I2 cell method. The software, called pyodine, is completely written in Python and has been built in a modular structure to allow for easy adaptation to different instruments. Methods. We present the fundamental concepts employed by pyodine, which build on existing I2 reduction codes, and give an overview of the software’s structure. We adapted pyodine to two instruments, Hertzsprung SONG located at Teide Observatory (SONG hereafter) and the Hamilton spectrograph at Lick Observatory (Lick hereafter), and demonstrate the code’s flexibility and its performance on spectra from these facilities. Results. Both for SONG and Lick data, the pyodine results generally match the RV precision achieved by the dedicated instrument pipelines. Notably, our code reaches a precision of roughly 0.69 m s−1 on a short-term solar time series of SONG spectra, and confirms the planet-induced RV variations of the star HIP 36616 on spectra from SONG and Lick. Using the solar spectra, we also demonstrate the capabilities of our software in extracting velocity time series from single absorption lines. A probable instrumental effect of SONG is still visible in the pyodine RVs, despite being a bit damped as compared to the original results. Conclusions. With pyodine we prove the feasibility of a highly precise, yet instrument-flexible I2 reduction software, and in the future the code will be part of the dedicated data reduction pipelines for the SONG network and the Waltz telescope project in Heidelberg.

Memory long and short term time series network for ultra-short-term photovoltaic power forecasting

Photovoltaic (PV) power is stochastic, intermittent and volatile, which has brought huge challenges to the safe and stable operation of the power grid. Accurate PV power forecasting is becoming a significant task for PV plant grid connecting, scheduling and guaranteeing the safety of the power grid. To improve the accuracy of PV power forecasting, a memory long and short term time series network (MLSTNet) model is proposed to perform ultra-short-term PV power forecasting with a time horizon varying from 15 min to 4 h. Firstly, the input variables are screened by calculating Spearman correlation coefficients between weather data. To cluster different types of data, the classification coefficient is set. Secondly, different type data are automatically identified in the MLSTNet model. The appropriate model parameters and network structures are adjusted for the higher forecasting speed and accuracy. By using the single-step rolling input method and the temporal attention convolutional neural network, the ability of temporal and spatial feature extraction has been significantly enhanced, allowing real-time forecasting and dynamic optimization of the proposed model. Thirdly, long-term and short-term historical data are fed into the model, and different periods dependencies are captured while incorporating multiple variable dimensions, leading to more memorable in the network. Finally, the experimental data is selected from the actual operation data of a PV plant in northern China, which represents the universal applicability and practical application of the model. The performance of the proposed model is demonstrated with actual PV plant dataset, the RMSE of proposed model decreased by 32.73%, 5.44%, 12.44%, and 31.22%. Results indicated that it is feasible to use the proposed MLSTNet model for time series forecasting. The core idea of the MLSTNet model is to achieve higher accuracy in ultra-short-term forecasting by enhancing the ability to extract temporal features and linearity in the model, providing a new idea for deep learning-based methods for ultra-short-term PV power forecasting.

An Excess Entropy Approach to Classify Long-Term and Short-Term Memory Stationary Time Series

Long-term memory behavior is one of the most important phenomena that has appeared in the time series analysis. Different from most definitions of second-order properties, an excess entropy approach is developed for stationary time series to classify long-term and short-term memory. A stationary sequence with finite block entropy is long-term memory if its excess entropy is infinite. The simulation results are graphically demonstrated after some theoretical results are simply presented by various stochastic sequences. Such an approach has advantages over the traditional ways that the excess entropy of stationary sequence with finite block entropy is invariant under instantaneous one-to-one transformation, and that it only requires very weak moment conditions rather than second-order moment conditions and thus can be applied to distinguish the LTM behavior of stationary sequences with unbounded second moment (e.g., heavy tail distribution). Finally, several applications on real data are exhibited.

A multi-scale video surveillance based information aggregation model for crime prediction

When studying the relationship and potential association between various data, researchers face multiple challenges: first, researchers need to collect data from multiple knowledge systems and channels, which challenges the breadth of expertise of researchers; second, it is challenging to perceive the changes of application scenarios and environment through these data such as image unstructured data that may contain useful information; third, short term behavior prediction with more frequent changes is challenging, especially when the sensitivity of the prediction model is insufficient and there is a lack of support from other feature data. This article designs a multi-scale information aggregation prediction model based on video surveillance for predicting and analyzing short-term time series with less data, such as criminal behavior. Firstly, a multi-scale human traffic perception and quantification model was established based on FairMOT to obtain human traffic time series data, which is an important association to strengthen the main features of subsequent short-term time series. Then, an improved prediction model, the Anti Saturation Gate Control Model (ASGCM), is designed to introduce entry-level techniques and add anti-saturation conversion modules, making ASGCM more sensitive, helping to reduce dependence on long-term features, and suppressing gradient disappearance and exploration problems. Finally, the application of urban crime prediction was used as an example for validation, and two sets of data were tested using a single sequence and a fused sequence, respectively. The results showed that the ASGCM method proposed in this paper can improve prediction accuracy while reducing computational complexity.