Short-term Time Series Research Articles

Neural dynamical operator: Continuous spatial-temporal model with gradient-based and derivative-free optimization methods

Data-driven modeling techniques have been explored in the spatial-temporal modeling of complex dynamical systems for many engineering applications. However, a systematic approach is still lacking to leverage the information from different types of data, e.g., with different spatial and temporal resolutions, and the combined use of short-term trajectories and long-term statistics. In this work, we build on the recent progress of neural operator and present a data-driven modeling framework called neural dynamical operator that is continuous in both space and time. A key feature of the neural dynamical operator is the resolution-invariance with respect to both spatial and temporal discretizations, without demanding abundant training data in different temporal resolutions. To improve the long-term performance of the calibrated model, we further propose a hybrid optimization scheme that leverages both gradient-based and derivative-free optimization methods and efficiently trains on both short-term time series and long-term statistics. We investigate the performance of the neural dynamical operator with three numerical examples, including the viscous Burgers' equation, the Navier–Stokes equations, and the Kuramoto–Sivashinsky equation. The results confirm the resolution-invariance of the proposed modeling framework and also demonstrate stable long-term simulations with only short-term time series data. In addition, we show that the proposed model can better predict long-term statistics via the hybrid optimization scheme with a combined use of short-term and long-term data.

Will it get there? A deep learning model for predicting next-trip state of charge in Urban Green Freight Delivery with electric vehicles

To enhance urban freight efficiency and green development, China has implemented the Urban Green Freight Delivery (UGFD) project, which includes optimizing vehicle traffic control policies and increasing the number of new energy vehicles (NEV). However, range anxiety is a significant challenge for freight drivers performing delivery tasks with electric vehicles (a major component of NEV). We constructed a prediction model for the state of charge (SOC), or battery remaining energy percentage when UGFD vehicles reach the next trip point, aiming to alleviate this issue. The model consists of three modules: (1) a vehicle SOC context prediction module, (2) a vehicle energy consumption prediction module, and (3) a multi-perspective SOC prediction value fusion module. Specifically, in the SOC context prediction module, historical SOC sequences, vehicle status (loading/unloading, charging), and time intervals between SOC points are used to accurately describe context change trends, and directly predict the vehicle SOC at the next trip point. The energy consumption prediction module combines community-level and grid-level geographical location information for the vehicle stops using weather, vehicle parameters, etc., to model the spatial dynamic correlation of energy consumption. The vehicle SOC at the next trip point is the difference between the current vehicle SOC and the predicted energy consumption. The multi-perspective SOC prediction value fusion module is a combination of the predicted values from the context and energy consumption perspectives, resulting in the final vehicle SOC prediction value. Taking Suzhou, China as an example, the results show that the mean absolute error, root mean square error, and symmetric mean absolute percentage error for the constructed model are 23.67%, 10.39%, and 20.03% less, respectively, than for the baseline models focusing on SOC short-term time series prediction. The research results can provide scientific evidence for formulating refined energy management, charging station layout, and freight delivery optimization approaches.

A VMD-LSTNet-Attention model for concentration prediction of mixed gases

Gases typically exist as mixed states, which normally contain more information and exhibit more intricate features than single gas. Hence, the task of predicting mixed gas concentrations poses a challenging endeavor in the field of gas detection. In this study, a Variational Mode Decomposition method was combined with Long and Short-term Time-series Network-Attention to build a VMD-LSTNet-Attention model, based on which the response signals of electronic nose (E-nose) is processed, achieving a high-precision concentrations prediction of carbon monoxide (CO) and ethylene mixed gases. Firstly, a dataset containing the raw response signals of CO and ethylene was collected using an gas data collection system. Then, VMD algorithm was used to extract and decompose the essential features from the E-nose signals into multiple components. Further, in pursuit of optimal decomposition results, Particle Swarm Optimization (PSO) algorithm was employed to optimize the penalty factor α and determine the most effective decomposition level K. Subsequently, the LSTNet-Attention model was employed to analyze the concentrations of mixed gases, and the predictive R-square values for CO and ethylene reach 0.993 and 0.975, respectively. Finally, the comparison with the traditional deep learning models and the popular models demonstrates the effectiveness of the proposed VMD-LSTNet-Attention model for concentration prediction of mixed gases.

Feasibility analysis of using short-term rainfall time series to evaluate rainwater harvesting systems considering climate change

Employing recent short-term historical rainfall data may enhance the performance of rainwater harvesting systems (RWHs) in response to climate change. However, this assumption lacks extensive research, and the evaluation of RWHs currently relies on long-term historical rainfall time series. This study evaluates the feasibility of this assumption and aims to identify the optimal rainfall time series for evaluating RWH performance under climate change. We evaluated RWHs in residential buildings across 16 Japanese cities utilizing historical rainfall time series of varying lengths and 30-year predicted rainfall time series. The minimum rainfall time series length was obtained based on the similarity index between the evaluation results for historical and future periods. The corresponding optimal series can be determined from the distribution of similarity indices in the minimum length. Finally, we introduce supply pressure indices (SPIs) to summarize the rainfall characteristics of these optimal rainfall time series. Our findings highlight that the minimum rainfall time series length increased from 1 year to 30 years as building non-potable water demand rose and city locations varied. Utilizing rainfall time series incorporating recent rainfall data yielded more dependable evaluation results for RWHs under climate change. These optimal rainfall time series share common characteristics with SPIs ranging from 5.37 to 17.87 mm/d, contingent on the local rainfall patterns. Our study concludes that utilizing recent short-term historical rainfall data is feasible to evaluate and design RWHs under climate change.

Use of Socio-economic, Climatic, and Land use Land Cover Patterns in Solid Waste Forecasting with Integrated Gradient LSTNet Based Model in Lomé, Togo

ABSTRACT Few studies have explored recurrent neural network models in municipal solid waste (MSW) generation forecasting regarding the inefficiency of conventional and probabilistic tools. This study aimed to develop a new approach of Integrated Gradient Long- and Short-term Time series Network (LSTNet+IG)-based models, assess the baseline neural network models, and provide features influences explanation on MSW generation by using socio-economic, climatic, and Land Use Land Cover (LULC) patterns. Hence, the Random Forest Regressor and Gradient Boosting Regressor methods were used for feature selection, and the IG component was incorporated to LSTNet model for feature influence explanation alongside baseline models implementation. Moreover, the metrics such as root relative square error (RSE) and relative absolute error (RAE) were used to evaluate models’ performance. As a result, the LSTNet+IG outperforms others with the lowest RSE (0.9727) and RAE (0.9108). Moreover, five key influencing features were found, namely the number of households and institutions, mean relative humidity, electricity consumption, built area, and mean temperature. Hence, the socio-economic, climatic, and LULC features in MSW generation forecasting have shown to be important in decision-making for effective MSW management. Therefore, the LSTNet+IG model adoption by stakeholders could help to anticipate and mitigate MSW mismanagement through better planning.

Deep learning for Bitcoin price direction prediction: models and trading strategies empirically compared

This paper applies deep learning models to predict Bitcoin price directions and the subsequent profitability of trading strategies based on these predictions. The study compares the performance of the convolutional neural network–long short-term memory (CNN–LSTM), long- and short-term time-series network, temporal convolutional network, and ARIMA (benchmark) models for predicting Bitcoin prices using on-chain data. Feature-selection methods—i.e., Boruta, genetic algorithm, and light gradient boosting machine—are applied to address the curse of dimensionality that could result from a large feature set. Results indicate that combining Boruta feature selection with the CNN–LSTM model consistently outperforms other combinations, achieving an accuracy of 82.44%. Three trading strategies and three investment positions are examined through backtesting. The long-and-short buy-and-sell investment approach generated an extraordinary annual return of 6654% when informed by higher-accuracy price-direction predictions. This study provides evidence of the potential profitability of predictive models in Bitcoin trading.

Deep learning-based quantitative analyses of feedback in the land-atmosphere interactions over the Vietnamese Mekong Delta

During the past several decades, the Vietnamese Mekong Delta (VMD) has experienced many severe droughts, resulting in significant impacts on both agriculture and aquaculture. In the evolution and intensification of droughts, local feedbacks in the Land-Atmosphere (LA) interactions were considered to play a crucial role. It is critical to quantify the impact of LA variables on drought processes and severity with the feedback loop of water and energy balances (e.g., soil moisture-latent and sensible heat-precipitation). In this study, a deep learning model, named Long- and Short-term Time-series Network (LSTNet), was applied to simulate the LA interactions over the VMD. With the ERA5 data as modelling inputs, the role of each key variable (e.g., soil moisture, sensible and latent heat) in the LA interactions over the period of 2011–2020 was quantified, and the variations of their inter-relationships in the future period (2015–2099) were also investigated based on the CMIP6 data. The LSTNet model has demonstrated that the deep learning algorithm can effectively capture the relative importance of key variables in the LA interactions. We found that it is crucial to evaluate the effects of soil moisture and sensible heat on the LA interactions, particularly in the dry periods when negative anomalies in soil moisture and sensible heat would significantly reduce the amount of precipitation. In addition, the decline in soil moisture and the rise in sensible heat are anticipated to further diminish precipitation in the future under the changing climate.

Short-term time-series observations of phytoplankton light-absorption and productivity in Prydz Bay, coastal Antarctica

The optical characteristics of coastal Antarctic waters exhibit complexity due to the dynamic hydrography influenced by meltwater intrusion, which alters nutrient levels, thermohaline structure, and optically active substances (OAS) regimes. Studies on bio-optical variability and its implications on phytoplankton productivity (PP) are scanty in coastal polar regions. On this backdrop, time-series measurements (72 h at 6 h intervals) of bio-optical properties such as phytoplankton biomass (chlorophyll-a), absorption (aph), and total suspended matter (TSM) concurrently with PP were measured to understand their interplay and variability in relation to the ambient physicochemical settings in the under-sampled Prydz Bay, coastal Antarctica. Our findings revealed thermohaline stratification within the bay, likely attributed to the inflow of less saline meltwater from nearby glaciers and minimal wind activity. The consistent presence of sub-surface chlorophyll maximum (SCM) beneath the stratified layer underscored the light-acclimatization response of shade-adapted phytoplankton. Surface waters exhibited higher TSM compared to deeper layers, indicating glacial melt influence, while the depth of the sunlit layer remained relatively stable, suggesting limited water mass movement and/or variability in OAS at the study site. An inverse relation between chlorophyll-a and chlorophyll-a-specific phytoplankton light absorption (a*ph(λ)) manifested ‘pigment package effect’ within the prevailing phytoplankton community, implying reduced light-absorption efficiency and consequent lower PP. Compared to chlorophyll-a, the phytoplankton light absorption (aph(λ)) emerged as a better proxy for explaining PP variability. Nutrient availability was not limiting, which was conducive to micro (large) phytoplankton growth. Classification of phytoplankton size classes (micro, nano, and pico) based on the B/R ratio (aph at Blue (443 nm)/Red (676 nm) region) confirmed the dominance of larger (micro) phytoplankton that are more susceptible to package effect, thus have implications on reduced PP potential of this polar marine ecosystem.

Ultra-short-term wind power forecasting method based on multi-variable joint extraction of spatial-temporal features

Accurate and reliable wind power forecasting is imperative for wind power stations' stable and efficient operation. Information such as wind speed and wind direction in the same wind field has spatial-temporal differences. Considering the spatial-temporal changes in wind fields can improve model prediction accuracy. However, existing methods suffer from limited ability to capture correlation features among variables, information loss in spatial-temporal feature extraction, and neglect short-term temporal features. This paper introduces a novel ultra-short-term wind power forecasting method based on the combination of a deep separable convolutional neural network (DSCNN) and long- and short-term time-series network (LSTNet), incorporating maximum information coefficient (MIC) to realize multi-variable joint extraction of spatial-temporal features. The method utilizes MIC to jointly analyze and process the multi-variate variables before spatial-temporal feature extraction to avoid information redundancy. The spatial features between input variables and wind power are extracted by deep convolution and pointwise convolution in DSCNN. Then, a convolutional neural network and gated recurrent unit in LSTNet are combined to capture long-term and short-term temporal features. In addition, an autoregressive module is employed to accept features extracted by MIC to enhance the model's learning of temporal features. Based on real datasets, the performance of models is validated through comprehensive evaluation experiments such as comparison experiments, ablation experiments, and interval prediction methods. The results show that the proposed method reduces mean absolute error by up to 4.66% and provides more accurate prediction intervals, verifying the accuracy and effectiveness of the proposed method.

Recalibration of missing low-frequency variability and trends in the North Atlantic Oscillation

Multi-decadal trends in the wintertime North Atlantic Oscillation (NAO) are under-represented by coupled general circulation models (CGCMs), consistent with a lack of autocorrelation in their NAO index series. This study proposes and tests two simple “reddening” approaches for correcting this problem in simulated indices based on simple one parameter short-term (AR; Auto-Regressive order 1) and long-term (FD; Fractional-Difference) time series filters. Using CGCMs from the Coupled Model Intercomparison Project Phase 6 (CMIP6), the FD filter successfully improves the autocorrelation structure of the NAO, and in turn the simulation of extreme trends, while the AR filter is less successful. The 1963–1993 NAO trend is the maximum 31-year trend in the historical period. Raw CGCMs underestimate the likelihood of this trend by a factor of ten but this discrepancy is corrected after reddening. CMIP6 future projections show that long-term (2024–2094) NAO ensemble mean trends systematically increase with the magnitude of radiative forcing: -2.4 to 3.5 hPa/century for low-to-high forcing after reddening (more than double the range using raw output). The related likelihood of future maximum 31year trends comparable to 1963–1993 ranges from 3 to 7% whereas none of these CMIP6 projections simulate this without reddening. Near-term projections of the next 31 years (2024–2054) are less sensitive than long term trends to the future scenario, showing weak-to-no forced trend. However, reddening increases the ensemble range by 74% (to +/-1 standard deviation/decade), which could increase/decrease regional climate change signals in the Northern Hemisphere by magnitudes that are underestimated when using raw CGCM output.

팜유 가격 예측을 위한 딥러닝 기반 단기 시계열 예측 모델링

팜유 가격 예측을 위한 딥러닝 기반 단기 시계열 예측 모델링

Explaining empirical dynamic modelling using verbal, graphical and mathematical approaches.

Empirical dynamic modelling (EDM) is becoming an increasingly popular method for understanding the dynamics of ecosystems. It has been applied to laboratory, terrestrial, freshwater and marine systems, used to forecast natural populations and has addressed fundamental ecological questions. Despite its increasing use, we have not found full explanations of EDM in the ecological literature, limiting understanding and reproducibility. Here we expand upon existing work by providing a detailed introduction to EDM. We use three progressively more complex approaches. A short verbal explanation of EDM is then explicitly demonstrated by graphically working through a simple example. We then introduce a full mathematical description of the steps involved. Conceptually, EDM translates a time series of data into a path through a multi-dimensional space, whose axes are lagged values of the time series. A time step is chosen from which to make a prediction. The state of the system at that time step corresponds to a 'focal point' in the multi-dimensional space. The set (called the library) of candidate nearest neighbours to the focal point is constructed, to determine the nearest neighbours that are then used to make the prediction. Our mathematical explanation explicitly documents which points in the multi-dimensional space should not be considered as focal points. We suggest a new option for excluding points from the library that may be useful for short-term time series that are often found in ecology. We focus on the core simplex and S-map algorithms of EDM. Our new R package, pbsEDM, enhances understanding (by outputting intermediate calculations), reproduces our results and can be applied to new data. Our work improves the clarity of the inner workings of EDM, a prerequisite for EDM to reach its full potential in ecology and have wide uptake in the provision of advice to managers of natural resources.

Use of one-dimensional CNN for input data size reduction in LSTM for improved computational efficiency and accuracy in hourly rainfall-runoff modeling

A deep learning architecture, denoted as CNNsLSTM, is proposed for hourly rainfall–runoff modeling in this study. The architecture involves a serial coupling of the one-dimensional convolutional neural network (1D-CNN) and the long short-term memory (LSTM) network. In the proposed framework, multiple layers of the CNN component process long-term hourly meteorological time series data, while the LSTM component handles short-term meteorological time series data and utilizes the extracted features from the 1D-CNN. In order to demonstrate the effectiveness of the proposed approach, it was implemented for hourly rainfall–runoff modeling in the Ishikari River watershed, Japan. A meteorological dataset, including precipitation, air temperature, evapotranspiration, longwave radiation, and shortwave radiation, was utilized as input. The results of the proposed approach (CNNsLSTM) were compared with those of previously proposed deep learning approaches used in hydrologic modeling, such as 1D-CNN, LSTM with only hourly inputs (LSTMwHour), a parallel architecture of 1D-CNN and LSTM (CNNpLSTM), and the LSTM architecture, which uses both daily and hourly input data (LSTMwDpH). Meteorological and runoff datasets were separated into training, validation, and test periods to train the deep learning model without overfitting, and evaluate the model with an independent dataset. The proposed approach clearly improved estimation accuracy compared to previously utilized deep learning approaches in rainfall = runoff modeling. In comparison with the observed flows, the median values of the Nash–Sutcliffe efficiency for the test period were 0.455–0.469 for 1D-CNN, 0.639–0.656 for CNNpLSTM, 0.745 for LSTMwHour, 0.831 for LSTMwDpH, and 0.865–0.873 for the proposed CNNsLSTM. Furthermore, the proposed CNNsLSTM reduced the median root mean square error (RMSE) of 1D-CNN by 50.2%–51.4%, CNNpLSTM by 37.4%–40.8%, LSTMwHour by 27.3%–29.5%, and LSTMwDpH by 10.6%–13.4%. Particularly, the proposed CNNsLSTM improved the estimations for high flows (≧75th percentile) and peak flows (≧95th percentile). The computational speed of LSTMwDpH is the fastest among the five architectures. Although the computation speed of CNNsLSTM is slower than LSTMwDpH's, it is still 6.9–7.9 times faster than that of LSTMwHour. Therefore, the proposed CNNsLSTM would be an effective approach for flood management and hydraulic structure design, mainly under climate change conditions that require estimating hourly river flows using meteorological datasets.

PV Power Forecasting in the Hexi Region of Gansu Province Based on AP Clustering and LSTNet

Accurate PV power forecasting is becoming a mandatory task to integrate the PV system into the power grid, schedule it, and ensure the safety of the power grid. In this paper, a novel model for PV power prediction using AP-LSTNet has been proposed. It consists of a combination of affinity propagation clustering and long-term and short-term time series network models. First, the affinity propagation algorithm is used to divide the regionally distributed photovoltaic station clusters into different seasons. The Pearson correlation coefficient is used to determine the strong correlation between meteorological factors of photovoltaic power, and the bilinear interpolation method is used to encrypt the meteorological data of the corresponding photovoltaic station cluster. Furthermore, LSTNet is used to mine the long-term and short-term temporal and spatial dependence of photovoltaic power, and meteorological factor series and linear components of auto-regression are superimposed to realize the simultaneous prediction of multiple photovoltaic stations in the group. Finally, PV power plants in five cities, Wuwei, Jinchang, Zhangye, Jiuquan, and Jiayuguan in the Hexi region of Gansu Province, China, will be selected to test the proposed model. The experimental comparison shows that the prediction model achieves high prediction accuracy and robustness.

Double-level optimal fuzzy association rules prediction model for time series based on DTW-i[formula omitted] fuzzy C-means

Information granulation theory has been widely used in short-term time-series forecasting research and holds significant weight. However, the error accumulation due to the lack of granular accuracy, along with information redundancy or deficiency in predictions, significantly affects short-term prediction accuracy. To compensate for these shortcomings, this paper proposes a double-level optimal fuzzy association rules prediction model for short-term time-series forecasting, which can strengthen the performance of information granulation in prediction. Firstly, this paper proposes a concept of breakpoints, which can accurately segment complex linear trends in time series and thus obtain a granular time series with highly accurate linear fuzzy information granules (LFIGs). Secondly, a improved distance is proposed to more accurately reflect the similarity between LFIGs by addressing counter-intuitive problems in the original distance. Theoretical analysis shows that the improved distance can effectively reduce errors in granular calculation. Then, a granule-suited fuzzy c-means algorithm is proposed for clustering LFIGs. Finally, this paper proposes a double-level optimal fuzzy association rules prediction model, which establishes the optimal rules for each cluster and selects the optimal two rules for prediction by the contribution of the clusters. The experimental results show that the prediction method effectively avoids the problems of information redundancy and information deficiency, and increases forecast accuracy. The model’s exceptional performance is demonstrated through comparative analysis with existing models in experimental investigations.

An Integrated Reservoir Predictor Based on Spatiotemporal Information Transformation

Multistep prediction of high-dimensional time series is an essential and challenging task. In this study, we propose an integrated reservoir predictor for making accurate and robust multistep-ahead forecasts based on short-term high-dimensional time series. Initially, a conjugated pair of Spatiotemporal Information (STI) equations is derived using Takens’ embedding theory to transform the spatial information of high-dimensional variables into one-dimensional temporal information of the target variable and vice versa. Next, by exploiting reservoir networks, reservoir-based STI equations are established to efficiently capture nonlinear dynamics of the target system with only linear optimization. Then, through an integration phase, the integrated reservoir predictor can output precise and robust predictions of the multistep-ahead states of any target variable. The integrated reservoir predictor outperforms some other prediction methods (including reservoir computing, long-short-term-memory network, convolutional neural network and support vector regression), when applied to classical dynamic systems (e.g. 60D double scroll model, 40D Lorenz 96 model, and 60D Rössler model) and real-world datasets (solar generation data and PM2.5 concentration records), as indicated by evaluation metrics such as Pearson correlation coefficients exceeding 0.9 and root-mean-square errors below 0.3, even in the presence of noise in training data.

A new LSTNet-based temperature prediction model for permanent magnet

Permanent magnet synchronous motors (PMSMs) can effectively protect against demagnetization by accurate permanent magnet (PM) temperature prediction; nevertheless, due to the nonlinear properties and intricate internal structure of PMSMs, accurate PM temperature prediction methods still encounter difficulties. This paper proposes a new PM temperature prediction model (LSTNet-Improved) based on long- and short-term time series network (LSTNet) to increase the prediction accuracy of PM temperature. By adding a multi-scale convolutional (MSC) layer to the convolutional neural network (CNN) layer of LSTNet, the short-term detail-dependent information between PMSM variables can be obtained by the model, and adds a CNN skip (CNN-skip) layer in place of the gate recurrent unite skip (GRU-skip) layer to find the long- and short-term local repetitive patterns between the variables, which yields additional feature information. Furthermore, a nonlinear bias term (MNB) is added to the original multi-head attention layer in order to solve the complex nonlinear relationship between multivariate time series. These layers are applied after the CNN-skip layer and gated recurrent unite (GRU) layers, respectively. This allows the model to become more robust and automatically learn to the complex nonlinear relationship between the sequences, focus on the important feature information, and lessen the impact of redundant information on the PMs’ temperature prediction task. The experimental results show that the goodness of fit (R2), root mean square error (RMSE) and mean absolute error (MAE) of the proposed model are 99.93%, 14.55% and 9.45%, respectively, and compared with LSTNet, the R2 is improved by 0.09%, the RMSE and MAE are reduced by 7.03% and 6.51%. These results confirm that the model can correctly predict the PM temperature, efficiently extract both long- and short-term patterns among the variables of the PMSM, and effectively focus on the key feature information in the intricate nonlinear series.

Informer Model based Wind Power Forecast with Tropical Storms Present

When severe tropical storms pass through, regional wind speeds fluctuate greatly, and the volatility of wind farm output also increases significantly. At the same time, the duration of tropical storms is long, and it is difficult for short-term time series data prediction models to be effective, in which case the unstable output of wind turbines will have a greater impact on power system dispatching. This paper first examines the characteristics of tropical storm movement, namely the change in wind speed, and then uses the Informer long time series data prediction model to predict the total change in wind turbine output in the next 10 days after the storm has passed. The actual case proves that the Informer model is ideal in predicting the long time series of wind power output during a tropical storm.

Heart rate dynamics and asymmetry during sympathetic activity stimulation and post-stimulation recovery in ski mountaineers-a pilot exploratory study.

There is a lack of studies on non-linear heart rate (HR) variability in athletes. We aimed to assess the usefulness of short-term HR dynamics and asymmetry parameters to evaluate the neural modulation of cardiac activity based on non-stationary RR interval series by studying their changes during sympathetic nervous system activity stimulation (isometric handgrip test) and post-stimulation recovery in professional ski mountaineers. The correlation between the changes in the parameters and the respiratory rate (RespRate) and also the duration of the career was analyzed. Short-term (5 min) and ultra-short-term (1 min) rates of patterns with no variations (0V), number of acceleration runs of length 1 (AR1), and short-term Porta's Index were greater, whereas Guzik's Index (GI) was smaller during sympathetic stimulation compared to rest. GI increased and the number of AR1 decreased during recovery. Greater increases in GI and RMSSD were associated with greater decreases in RespRate during recovery. Greater increases in RespRate from rest to short-term sympathetic stimulation were associated with greater increases in 0V (Max-min method) and AR1 but also with greater decreases in decelerations of short-term variance and accelerations and decelerations of long-term variance. Greater increases in 0V (Max-min method) and number of AR1 during sympathetic stimulation were associated with a shorter career duration. Greater decreases in these parameters during recovery were associated with a longer career duration. Changes in measures of HR dynamics and asymmetry, calculated based on short-term non-stationary RRi time series induced by sympathetic stimulation and post-stimulation recovery, reflected sympathovagal shift and were associated with condition-related alterations in RespRate and career duration in athletes who practice ski mountaineering.

Intra-hour solar irradiance forecasting using topology data analysis and physics-driven deep learning

Real-time Direct Normal Irradiance (DNI) prediction is crucial for reliable and economic operation of Concentrated photothermal Solar Power (CSP) system in arid desert areas. However, the stochastic characteristics of short-term multidimensional meteorological time series make intra-hour DNI prediction a challenging task. In this study, we have proposed a deep learning model called TLD, which is combined with topological features captured by Topology Data Analysis (TDA) and temporal features captured by LSTM to address this challenge. Experimental results demonstrated that TLD outperformed the five latest models (Ridge, RF, C_GRU, BiLSTM, and GBRT) on seven solar radiation datasets in arid desert areas. Further analysis revealed that the proportion of cloudy days is a key factor affecting the model's performance. To enhance the forecast ability of TLD, we developed a physics-informed hybrid model named TLDP based on TLD and a smart persistence model, which fully combines the DNI prediction ability of TLD under cloudy conditions and that of the smart persistence model under sunny conditions. Experimental results of eight datasets collected from real-world solar photothermal power stations indicated that TLDP outperformed existing models, which may lay a foundation for more economical and stable operation of CSP plants in arid desert areas.