Temporal Prediction Research Articles

Optimising bearing degradation prediction: a bayesian vector autoregressive-assisted DeepAR model approach

Abstract Accurate prediction of the remaining useful life of bearings is crucial for the maintenance of rotating machinery. Current research often involves extracting multidimensional features from vibration signals to construct health degradation indicator, and health indicator is directly used for degradation prediction. However, constructed health indicator exhibit complex nonlinearity and fail to utilize the correlations with the original multidimensional features. To address this limitation, we propose an integrated prediction framework, known as the bayesian vector autoregression-assisted DeepAR method. The proposed two step prediction framework combines the advantages of covariance prediction and temporal correlation prediction, and realizes the full utilization of multidimensional features correlation information. Additionally, the proposed approach utilizes the intrinsic properties of statistical models to constrain improve the prediction results of deep learning methods, achieving nonlinear modeling of the degradation indicator while maintaining the stability of the prediction results. This process of prediction, re-estimation and adaptive adjustment achieves a balance in model performance. Finally, experimental studies using the PRONOSTIA and XJTU-SY datasets validate the effectiveness of the proposed method in improving the prediction accuracy of bearing health degradation indicators and enhancing the stability of multistep prediction in deep model.

SAM-Net: Spatio-Temporal Sequence Typhoon Cloud Image Prediction Net with Self-Attention Memory

Cloud image prediction is a spatio-temporal sequence prediction task, similar to video prediction. Spatio-temporal sequence prediction involves learning from historical data and using the learned features to generate future images. In this process, the changes in time and space are crucial for spatio-temporal sequence prediction models. However, most models now rely on stacking convolutional layers to obtain local spatial features. In response to the complex changes in cloud position and shape in cloud images, the prediction module of the model needs to be able to extract both global and local spatial features from the cloud images. In addition, for irregular cloud motion, more attention should be paid to the spatio-temporal sequence features between input cloud image frames in the temporal sequence prediction module, considering the extraction of temporal features with long temporal dependencies, so that the spatio-temporal sequence prediction network can learn cloud motion trends more accurately. To address these issues, we have introduced an innovative model called SAM-Net. The self-attention module of this model aims to extract an inner image frame’s spatial features of global and local dependencies. In addition, a memory mechanism has been added to the self-attention module to extract interframe features with long temporal and spatial dependencies. Our method shows better performance than the PredRNN-v2 model on publicly available datasets such as MovingMNIST and KTH. We achieved the best performance in both the 4-time-step and 10-time-step typhoon cloud image predictions. On a cloud dataset consisting of 10 time steps, we observed a decrease in MSE of 180.58, a decrease in LPIPS of 0.064, an increase in SSIM of 0.351, and a significant improvement in PSNR of 5.56 compared to PredRNN-v2.

Perturbation context in paced finger tapping tunes the error-correction mechanism

Sensorimotor synchronization (SMS) is the mainly specifically human ability to move in sync with a periodic external stimulus, as in keeping pace with music. The most common experimental paradigm to study its largely unknown underlying mechanism is the paced finger-tapping task, where a participant taps to a periodic sequence of brief stimuli. Contrary to reaction time, this task involves temporal prediction because the participant needs to trigger the motor action in advance for the tap and the stimulus to occur simultaneously, then an error-correction mechanism takes past performance as input to adjust the following prediction. In a different, simpler task, it has been shown that exposure to a distribution of individual temporal intervals creates a “temporal context” that can bias the estimation/production of a single target interval. As temporal estimation and production are also involved in SMS, we asked whether a paced finger-tapping task with period perturbations would show any time-related context effect. In this work we show that a perturbation context can indeed be generated by exposure to period perturbations during paced finger tapping, affecting the shape and size of the resynchronization curve. Response asymmetry is also affected, thus evidencing an interplay between context and intrinsic nonlinearities of the correction mechanism. We conclude that perturbation context calibrates the underlying error-correction mechanism in SMS.

SEGODE: a structure-enhanced graph neural ordinary differential equation network model for temporal link prediction

SEGODE: a structure-enhanced graph neural ordinary differential equation network model for temporal link prediction

Spatial and temporal predictions of mosquito potential breeding sites and densities: integration of satellite imagery, in-situ data, and process-based modeling

Abstract. Dengue fever, primarily transmitted worldwide by the mosquito Aedes aegypti, poses significant public health challenges in tropical and subtropical regions. While effective vector control is crucial in the absence of reliable dengue vaccines, traditional control methods face obstacles like mosquito resistance to insecticides and a very high cost. By combining geospatial data, including satellite imagery, as descriptors, and entomological surveys as target variables in a Random Forest model, we predicted the number of potential mosquito breeding sites, derived the associated environmental carrying capacity for larvae, and used the Arbocarto process-based model to predict Ae.aegypti population densities in an urban region of French Guiana, South America. Our findings highlight that remote sensing data may help predict the number of potential breeding sites over urban areas. Our simulations indicate higher mosquito densities in urban residential areas and a strong spatial and temporal heterogeneity. These densities fluctuate according to intra-annual variations in temperature and precipitation, with higher densities associated with intermediate housing. A comparison with the conventional estimation of environmental carrying capacity for larvae in the current Arbocarto procedure highlights the advantages of our approach. Our study demonstrates the utility of integrating remote sensing with predictive modeling to enhance vector surveillance and control strategies, and provides a replicable approach for monitoring a dengue vector mosquito population in dynamic urban landscapes.

Long-term (2000–2020) global 0.05° continuous atmospheric carbon dioxide mapping combining OCO-2 observations and model simulations

We reconstructed a global continuous 8-day XCO2 (column-averaged CO2 dry air mole fraction) product (GCXCO2) at a spatial resolution of 0.05° from 2000 to 2020, combining terrestrial/marine remote sensing data and model simulations based on developed and tested stacking machine learning method. The GCXCO2 product has the similar spatial pattern with OCO-2 satellite observations but with global seamless coverage, showing a higher spatial resolution and accuracy than CarbonTracker and CAMS model simulation data. A novel dynamic normalization strategy was developed to handle the great temporal variation issue and ensure the temporal expansion of the prediction model. The sampled based 10-fold cross-validation shows an overall satisfactory result at a global scale, with R2 = 0.974 and root-mean-square error (RMSE) = 0.551 ppm (parts per million). Further spatial extension and temporal prediction experiments also proved that dependable results could be obtained in the regions and time periods without valid OCO-2 satellite observations (R2 = 0.958 and R2 = 0.886, respectively). Compared with Total Carbon Column Observing Network (TCCON) ground station observations, the GCXCO2 demonstrates a better accuracy and a higher spatial resolution than the model simulation data. Based on the GCXCO2 product, an upward annual trend of approximately 2.105 ppm/year can be found for global XCO2 between 2000 and 2020, with greater seasonal fluctuations in the Northern Hemisphere than in the Southern Hemisphere. This product is one of some remote sensing-based global high-precision long-term XCO2 datasets and an important tool to help advance the understanding of climate change and carbon balance, but also to detect CO2 concentration anomalies. The dataset can be obtained publicly at doi:https://doi.org/10.5281/zenodo.10083102 (Guan and Sun, 2023).

Intraflow temporal correlation-based network traffic prediction

Intraflow temporal correlation-based network traffic prediction

Drilling Rate of Penetration Prediction Based on CBT-LSTM Neural Network.

Due to the uncertainty of the subsurface environment and the complexity of parameters, particularly in feature extraction from input data and when seeking to understand bidirectional temporal information, the evaluation and prediction of the rate of penetration (ROP) in real-time drilling operations has remained a long-standing challenge. To address these issues, this study proposes an improved LSTM neural network model for ROP prediction (CBT-LSTM). This model integrates the capability of a two-dimensional convolutional neural network (2D-CNN) for multi-feature extraction, the advantages of bidirectional long short-term memory networks (BiLSTM) for processing bidirectional temporal information, and the dynamic weight adjustment of the time pattern attention mechanism (TPA) for extracting crucial information in BiLSTM, effectively capturing key features in temporal data. Initially, data are denoised using the Savitzky-Golay filter, and five correlation coefficient methods are employed to select input features, with principal component analysis (PCA) used to reduce model complexity. Subsequently, a sliding window approach transforms the time series into a two-dimensional structure to capture dynamic changes, constructing the model input. Finally, the ROP prediction model is established, and search methods are utilized to identify the optimal hyperparameter combinations. Compared with other neural networks, CBT-LSTM demonstrates superior performance metrics, with MAE, MAPE, RMSE, and R2 values of 0.0295, 0.0357, 9.3101%, and 0.9769, respectively, indicating the highest predictive capability. To validate the model's robustness, noise was introduced into the training data, and results show stable performance. Furthermore, the model's predictive results for other wells achieved R2 values of 0.95, confirming its strong generalization ability. This method provides a new solution for ROP prediction in real-time drilling operations, assisting drilling engineers in better planning their operations and reducing drilling cycles.

Neural encoding of melodic expectations in music across EEG frequency bands.

The human brain tracks regularities in the environment and extrapolates these to predict future events. Prior work on music cognition suggests that low-frequency (1-8Hz) brain activity encodes melodic predictions beyond the stimulus acoustics. Building on this work, we aimed to disentangle the frequency-specific neural dynamics linked to melodic prediction uncertainty (modelled as entropy) and prediction error (modelled as surprisal) for temporal (note onset) and content (note pitch) information. By using multivariate temporal response function (TRF) models, we re-analysed the electroencephalogram (EEG) from 20 subjects (10 musicians) who listened to Western tonal music. Our results show that melodic expectation metrics improve the EEG reconstruction accuracy in all frequency bands below the gamma range (< 30 Hz). Crucially, we found that entropy contributed more strongly to the reconstruction accuracy enhancement compared to surprisal in all frequency bands. Additionally, we found that the encoding of temporal, but not content, information metrics was not limited to low frequencies, rather it extended to higher frequencies (> 8Hz). An analysis of the TRF weights revealed that the temporal predictability of a note (entropy of note onset) may be encoded in the delta- (1-4Hz) and beta-band (12-30 Hz) brain activity prior to the stimulus, suggesting that these frequency bands associate with temporal predictions. Strikingly, we also revealed that melodic expectations selectively enhanced EEG reconstruction accuracy in the beta band for musicians, and in the alpha band (8-12 Hz) for non-musicians, suggesting that musical expertise influences the neural dynamics underlying predictive processing in music cognition.

Temporal variability and predictability predict alpine plant community composition and distribution patterns.

One of the most reliable features of natural systems is that they change through time. Theory predicts that temporally fluctuating conditions shape community composition, species distribution patterns, and life history variation, yet features of temporal variability are rarely incorporated into studies of species-environment associations. In this study, we evaluated how two components of temporal environmental variation-variability and predictability-impact plant community composition and species distribution patterns in the alpine tundra of the Southern Rocky Mountains in Colorado (USA). Using the Sensor Network Array at the Niwot Ridge Long-Term Ecological Research site, we used in situ, high-resolution temporal measurements of soil moisture and temperature from 13 locations ("nodes") distributed throughout an alpine catchment to characterize the annual mean, variability, and predictability in these variables in each of four consecutive years. We combined these data with annual vegetation surveys at each node to evaluate whether variability over short (within-day) and seasonal (2- to 4-month) timescales could predict patterns in plant community composition, species distributions, and species abundances better than models that considered average annual conditions alone. We found that metrics for variability and predictability in soil moisture and soil temperature, at both daily and seasonal timescales, improved our ability to explain spatial variation in alpine plant community composition. Daily variability in soil moisture and temperature, along with seasonal predictability in soil moisture, was particularly important in predicting community composition and species occurrences. These results indicate that the magnitude and patterns of fluctuations in soil moisture and temperature are important predictors of community composition and plant distribution patterns in alpine plant communities. More broadly, these results highlight that components of temporal change provide important niche axes that can partition species with different growth and life history strategies along environmental gradients in heterogeneous landscapes.

STIA-DJANet: Spatial–Temporal Intention-Aware vessel trajectory prediction based on Dual-Joint Attention Network for e-navigation

Harnessing high-precision vessel trajectory prediction holds great promise for enhancing autonomous and safe navigation towards e-navigation. However, many technical researchers and maritime managers realize that only accurate prediction of vessel trajectory is insufficient, extraction and recognition of navigation intentions from vessel around is still crucial issue and gap in current prediction models. In response to solve this shortcoming, we propose an intention-aware model for vessel trajectory prediction, which leverages both of spatial–temporal intentions and dual joint attentions network, namely STIA-DJANet. First, the double intentions of spatial–temporal aware is designed to extract the social relationships between the own vessel trajectory and neighboring trajectories at each timestamp, effectively capturing dynamic changes in their interactions. Second, feature fusion of dual joint attentions for double intentions is proposed to extract navigation features from the intention-aware module, which can be integrated into trajectory prediction. In experiments, numerical results and comparisons on three real-world datasets demonstrate that our framework outperforms state-of-the-art models in terms of the average displacement error (ADE) and final displacement error (FDE) metrics. The average improvements were 42.852% and 44.726%, respectively. Reliable prediction results can significantly advance the development of e-navigation systems development.

A novel recovery controllability method on temporal networks via temporal lost link prediction

Abstract Temporal networks are essential in representing systems where interactions between elements evolve over time. A crucial aspect of these networks is their controllability the ability to guide the network to a desired state through a set of control inputs. However, as these networks evolve, links between nodes can be lost due to various reasons, such as network failures, disruptions, or attacks. The loss of these links can severely impair the network’s controllability, making it challenging to recover desired network functions. In this paper, while investigating the destructive effects of various attacks on controllability processes in temporal networks, a new controllability recovery method is proposed, in which it prevents disruptions in this type of network processes by predicting lost links. In the proposed method, using network embedding and feature extraction, the dissimilarity of the nodes is calculated and then the missing links are predicted by designing a neural network. The results of the implementation of the proposed method on the datasets have demonstrates that the proposed method performed better than other conventional methods.

Adversarial nonnegative matrix factorization for temporal link prediction

Temporal link prediction has been extensively studied and widely applied in various applications, aiming to predict future network links based on the historical networks. However, most existing methods ignore the behavior of previous network updating information in temporal networks. To address these issues, we propose a novel link prediction model based on adversarial nonnegative matrix factorization, which fuses graph representation and adversarial learning to perform temporal link prediction. Specifically, we add a bounded adversary matrix to the input matrix to provide the robustness against real perturbations. Then, our model fully exploits the impact of snapshots by using communicability. Simultaneously, we utilize the cosine similarity to extract the node similarity and map it to low-dimensional latent representation to preserve the local structure. Additionally, we provide effective updating rules to learn the parameters of this model. Extensive experiments results on six real-world networks demonstrate that the proposed method outperforms several classical and the state-of-art matrix-based methods.

Attention-Linear Trajectory Prediction

Recently, a large number of Transformer-based solutions have emerged for the trajectory prediction task, but there are shortcomings in the effectiveness of Transformers in trajectory prediction. Specifically, while position encoding preserves some of the ordering information, the self-attention mechanism at the core of the Transformer has its alignment invariance that leads to the loss of temporal information, which is crucial for trajectory prediction. For this reason, we design a simple and efficient strategy for temporal information extraction and prediction of trajectory sequences using the self-attention mechanism and linear layers. The experimental results show that the strategy can improve the average accuracy by 15.31%, effectively combining the advantages of the linear layer and the self-attention mechanism, while compensating for the shortcomings of the Transformer. Additionally, we conducted an empirical study to explore the effectiveness of the linear layer and sparse self-attention mechanisms in trajectory prediction.

The structure and statistics of language jointly shape cross-frequency neural dynamics during spoken language comprehension

Humans excel at extracting structurally-determined meaning from speech despite inherent physical variability. This study explores the brain’s ability to predict and understand spoken language robustly. It investigates the relationship between structural and statistical language knowledge in brain dynamics, focusing on phase and amplitude modulation. Using syntactic features from constituent hierarchies and surface statistics from a transformer model as predictors of forward encoding models, we reconstructed cross-frequency neural dynamics from MEG data during audiobook listening. Our findings challenge a strict separation of linguistic structure and statistics in the brain, with both aiding neural signal reconstruction. Syntactic features have a more temporally spread impact, and both word entropy and the number of closing syntactic constituents are linked to the phase-amplitude coupling of neural dynamics, implying a role in temporal prediction and cortical oscillation alignment during speech processing. Our results indicate that structured and statistical information jointly shape neural dynamics during spoken language comprehension and suggest an integration process via a cross-frequency coupling mechanism.

Spatial–Temporal-Correlation-Constrained Dynamic Graph Convolutional Network for Traffic Flow Forecasting

Accurate traffic flow prediction in road networks is essential for intelligent transportation systems (ITS). Since traffic data are collected from the road network with spatial topological and time series sequences, the traffic flow prediction is regarded as a spatial–temporal prediction task. With the powerful ability to model the non-Euclidean data, the graph convolutional network (GCN)-based models have become the mainstream framework for traffic forecasting. However, existing GCN-based models either use the manually predefined graph structure to capture the spatial features, ignoring the heterogeneity of road networks, or simply perform 1-D convolution with fixed kernel to capture the temporal dependencies of traffic data, resulting in insufficient long-term temporal feature extraction. To solve those issues, a spatial–temporal correlation constrained dynamic graph convolutional network (STC-DGCN) is proposed for traffic flow forecasting. In STC-DGCN, a spatial–temporal embedding encoder module (STEM) is first constructed to encode the dynamic spatial relationships for road networks at different time steps. Then, a temporal feature encoder module with heterogeneous time series correlation modeling (TFE-HCM) and a spatial feature encoder module with dynamic multi-graph modeling (SFE-DCM) are designed to generate dynamic graph structures for effectively capturing the dynamic spatial and temporal correlations. Finally, a spatial–temporal feature fusion module based on a gating fusion mechanism (STM-GM) is proposed to effectively learn and leverage the inherent spatial–temporal relationships for traffic flow forecasting. Experimental results from three real-world traffic flow datasets demonstrate the superior performance of the proposed STC-DGCN compared with state-of-the-art traffic flow forecasting models.

Small-scale geographic differences in multiple-driver environmental variability can modulate contrasting phenotypic plasticity despite high levels of gene flow

Climate change is altering not only the mean conditions of marine environments, but also their temporal variability and predictability. As these alterations are not uniform across seascapes, their biological effects are expected to accentuate intra-specific differences in the adaptive capacity (e.g., plasticity and evolutionary potential) of natural populations. To test this theoretical framework, we assessed the phenotypic and genetic profiles of mussel from three study sites across a multi-driver heterogeneous environmental mosaic in Chilean Patagonia. Our study reveals that temporal variability, predictability, and exposure to extreme events (low pH/low salinity), collectively, can modulate the plasticity and optimal conditions of mussels. Despite these phenotypic differences, we observed low genetic differentiation, likely resulting from significant gene flow induced by aquaculture, ultimately diminishing variation among individuals from different geographic areas. Our findings underscore how variability and predictability are essential factors shaping phenotypic diversity, even at small spatial scales. Balancing these factors could enhance species resilience and ecological success, crucial for biodiversity conservation amidst climate change.

Commuter exposure to smoke and particulate matter air pollution in Auckland during the 2019 New Zealand International Convention Centre fire

ABSTRACT Human mobility typically exhibits temporal and spatial predictability. However, during hazardous events, roadways, footpaths and public transport networks can be disrupted by detours, closures and congestion. Urban fires, exemplified by the October 2019 New Zealand International Convention Centre (NZICC) fire in Auckland, New Zealand, are on the rise, posing threats to life, property and mobility. In this study, we employ geospatial analyses to investigate the impact of the NZICC fire on human mobility, encompassing both driving and walking. We generate predicted surfaces of particulate matter (PM2.5 and PM10) from seven local fixed air monitoring stations to estimate network-based exposure and inhalation dosage during travel. High levels of air pollution during the fire exceeded baseline concentrations, surpassing National Environmental Standards for Air Quality (NESAQ) and World Health Organization (WHO) limits for both PM2.5 and PM10. Private car commuters faced delays and congestion due to road closures, prolonging smoke exposure. Pedestrians, including those accessing essential public transportation infrastructure, experienced unavoidable exposure to smoke along the fastest routes from NZICC to places like Britomart train station and the Downtown Ferry. We recommend increasing the availability and dissemination of air pollutant monitoring data to enhance public awareness of the health risks associated with smoke exposure.

Serial dependencies for externally and self-generated stimuli.

Our senses are constantly exposed to external stimulation. Part of the sensory stimulation is produced by our own movement, like visual motion on the retina or tactile sensations from touch. Sensations caused by our movements appear attenuated. The interpretation of current stimuli is influenced by previous experiences, known as serial dependencies. Here we investigated how sensory attenuation and serial dependencies interact. In Experiment 1, we showed that temporal predictability causes sensory attenuation. In Experiment 2, we isolated temporal predictability in a visuospatial localization task. Attenuated stimuli are influenced by serial dependencies. However, the magnitude of the serial dependence effects varies, with greater effects when the certainty of the previous trial is equal to or greater than the current one. Experiment 3 examined sensory attenuation's influence on serial dependencies. Participants localized a briefly flashed stimulus after pressing a button (self-generated) or without pressing a button (externally generated). Stronger serial dependencies occurred in self-generated trials compared to externally generated ones when presented alternately but not when presented in blocks. We conclude that the relative uncertainty in stimulation between trials determines serial dependency strengths.

Fifty Percent of the Time, Tones Come Every Time: Stronger Prediction Error Effects on Neurophysiological Sensory Attenuation for Self-generated Tones.

The N1/P2 amplitude reduction for self-generated tones in comparison to external tones in EEG, which has recently also been described for action observation, is an example of the so-called sensory attenuation. Whether this effect is dependent on motor-based or general predictive mechanisms is unclear. Using a paradigm, in which actions (button presses) elicited tones in only half the trials, this study examined how the processing of the tones is modulated by the prediction error in each trial in a self-performed action compared with action observation. In addition, we considered the effect of temporal predictability by adding a third condition, in which visual cues were followed by external tones in half the trials. The attenuation result patterns differed for N1 and P2 amplitudes, but neither showed an attenuation effect beyond temporal predictability. Interestingly, we found that both N1 and P2 amplitudes reflected prediction errors derived from a reinforcement learning model, in that larger errors coincided with larger amplitudes. This effect was stronger for tones following button presses compared with cued external tones, but only for self-performed and not for observed actions. Taken together, our results suggest that attenuation effects are partially driven by general predictive mechanisms irrespective of self-performed actions. However, the stronger prediction-error effects for self-generated tones suggest that distinct motor-related factors beyond temporal predictability, potentially linked to reinforcement learning, play a role in the underlying mechanisms. Further research is needed to validate these initial findings as the calculation of the prediction errors was limited by the design of the experiment.