Temporal Prediction Research Articles

TG-SPRED: Temporal Graph for Sensorial Data PREDiction

This study introduces an innovative method aimed at reducing energy consumption in sensor networks by predicting sensor data, thereby extending the network’s operational lifespan. Our model, Temporal Graph Sensor Prediction (TG-SPRED), predicts readings for a subset of sensors designated to enter sleep mode in each time slot, based on a non-scheduling-dependent approach. This flexibility allows for extended sensor inactivity periods without compromising data accuracy. TG-SPRED addresses the complexities of event-based sensing—a domain that has been somewhat overlooked in existing literature—by recognizing and leveraging the inherent temporal and spatial correlations among events. It combines the strengths of Gated Recurrent Units and Graph Convolutional Networks to analyze temporal data and spatial relationships within the sensor network graph, where connections are defined by sensor proximities. An adversarial training mechanism, featuring a critic network employing the Wasserstein distance for performance measurement, further refines the predictive accuracy. Comparative analysis against six leading solutions using four critical metrics—F-score, energy consumption, network lifetime, and computational efficiency—showcases our approach’s superior performance in both accuracy and energy efficiency.

Parallel framework of a multi-graph convolutional network and gated recurrent unit for spatial–temporal metro passenger flow prediction

Metro passenger flow prediction is a critical problem in metro transport systems. However, recent studies have either overlooked spatial information on the metro network or primarily focused on modeling spatial dependencies using only the physical topology. To achieve accurate metro passenger flow (inflow and outflow at each station of a network) prediction, this study proposes a joint prediction model that combines the multi-graph convolution network and the gated recurrent unit (GRU). In addition to exploring location topology relationships, the proposed model selects two non-Euclidean spatial dependencies in metro passenger flow prediction to design essential graph elements as part of the stacked spatial block. Three spatial relationships (adjacency, similarity, and correlation) are integrated in parallel with the GRU network. The metro passenger flow prediction framework ASC-GRU (adjacency, similarity, correlation, and gated recurrent unit) is designed to mitigate the distortion of results during the capturing of passenger flow spatial–temporal features. Finally, ASC-GRU is tested using two datasets from the Hangzhou and Shanghai metro networks in China, and the error metrics of different models are compared and analyzed to verify the effectiveness and feasibility of ASC-GRU. The test results demonstrate that the proposed model outperforms other baseline models in passenger flow prediction over long time intervals and large networks. In particular, compared with the best performance of the baselines, the average reduction is around 3%, 12% and 13% in metrics of MAPE, MAE and RMSE, respectively.

Predictive coding networks for temporal prediction.

One of the key problems the brain faces is inferring the state of the world from a sequence of dynamically changing stimuli, and it is not yet clear how the sensory system achieves this task. A well-established computational framework for describing perceptual processes in the brain is provided by the theory of predictive coding. Although the original proposals of predictive coding have discussed temporal prediction, later work developing this theory mostly focused on static stimuli, and key questions on neural implementation and computational properties of temporal predictive coding networks remain open. Here, we address these questions and present a formulation of the temporal predictive coding model that can be naturally implemented in recurrent networks, in which activity dynamics rely only on local inputs to the neurons, and learning only utilises local Hebbian plasticity. Additionally, we show that temporal predictive coding networks can approximate the performance of the Kalman filter in predicting behaviour of linear systems, and behave as a variant of a Kalman filter which does not track its own subjective posterior variance. Importantly, temporal predictive coding networks can achieve similar accuracy as the Kalman filter without performing complex mathematical operations, but just employing simple computations that can be implemented by biological networks. Moreover, when trained with natural dynamic inputs, we found that temporal predictive coding can produce Gabor-like, motion-sensitive receptive fields resembling those observed in real neurons in visual areas. In addition, we demonstrate how the model can be effectively generalized to nonlinear systems. Overall, models presented in this paper show how biologically plausible circuits can predict future stimuli and may guide research on understanding specific neural circuits in brain areas involved in temporal prediction.

Transition Propagation Graph Neural Networks for Temporal Networks.

Researchers of temporal networks (e.g., social networks and transaction networks) have been interested in mining dynamic patterns of nodes from their diverse interactions. Inspired by recently powerful graph mining methods like skip-gram models and graph neural networks (GNNs), existing approaches focus on generating temporal node embeddings sequentially with nodes' sequential interactions. However, the sequential modeling of previous approaches cannot handles the transition structure between nodes' neighbors with limited memorization capacity. In detail, an effective method for the transition structures is required to both model nodes' personalized patterns adaptively and capture node dynamics accordingly. In this article, we propose a method, namely t ransition p ropagation g raph n eural n etworks (TIP-GNN), to tackle the challenges of encoding nodes' transition structures. The proposed TIP-GNN focuses on the bilevel graph structure in temporal networks: besides the explicit interaction graph, a node's sequential interactions can also be constructed as a transition graph. Based on the bilevel graph, TIP-GNN further encodes transition structures by multistep transition propagation and distills information from neighborhoods by a bilevel graph convolution. Experimental results over various temporal networks reveal the efficiency of our TIP-GNN, with at most 7.2% improvements of accuracy on temporal link prediction. Extensive ablation studies further verify the effectiveness and limitations of the transition propagation module. Our code is available at https://github.com/doujiang-zheng/TIP-GNN.

Continuous-Time User Preference Modelling for Temporal Sets Prediction

Continuous-Time User Preference Modelling for Temporal Sets Prediction

Detection of Threshold-Level Stimuli Modulated by Temporal Predictions of the Cerebellum.

The cerebellum has the reputation of being a primitive part of the brain that mostly is involved in motor coordination and motor control. Older lesion studies and more recent electrophysiological studies have, however, indicated that it is involved in temporal perception and temporal expectation building. An outstanding question is whether this temporal expectation building cerebellar activity has functional relevance. In this study, we collected magnetoencephalographic data from 30 healthy participants performing a detection task on at-threshold stimulation that was presented at the end of a sequence of temporally regular or irregular above-threshold stimulation. We found that behavioral detection rates depended on the degree of irregularity in the sequence preceding it. We also found cerebellar responses evoked by above-threshold and at-threshold stimulation. The evoked responses to at-threshold stimulation differed significantly, depending on whether it was preceded by a regular or an irregular sequence. Finally, we found that detection performance across participants correlated significantly with the differences in cerebellar evoked responses to the at-threshold stimulation, demonstrating the functional relevance of cerebellar activity in sensory expectation building. We furthermore found evidence of thalamic involvement, as indicated by responses in the beta band (14-30 Hz) and by significant modulations of cerebello-thalamic connectivity by the regularity of the sequence and the kind of stimulation terminating the sequence. These results provide evidence that the temporal expectation building mechanism of the cerebellum, what we and others have called an internal clock, shows functional relevance by regulating behavior and performance in sensory action that requires acting and integrating evidence over precise timescales.

An accurate and efficient deep learning model for spatio-temporal prediction of interfacial flows

Aiming at fast and accurate spatiotemporal prediction of interfacial flow fields, a novel deep learning model combining Convolutional AutoEncoder (CAE) and long short-term memory with spatial and temporal attention (LSTM-STA) is proposed and named as CAE-LSTM-STA (hybrid model of CAE and LSTM-STA) in this article. To enable a fast calculation process, CAE is first utilized to compress the high-dimensional snapshots of flow fields into the low-dimensional latent space by its encoder. Then, the latent space serves as the input feature for LSTM-STA to temporally evolve the state of the low-dimensional latent space. Finally, the low-dimensional latent space at future time steps is fed back into the decoder of CAE to derive the full-order snapshots of the flow fields. The prediction performance of the proposed model is evaluated via two representative benchmark cases, including the dam break case and the rising bubble case. The capability of CAE in dimension reduction is found to be much better than that of the widely used Proper Orthogonal Decomposition, while LSTM-STA is observed that outperforms the original LSTM in multivariant temporal prediction. These promising results indicate that the proposed CAE-LSTM-STA model is able to effectively capture and advance the spatiotemporal characteristics of the interfacial flow fields, therefore making it an advanced surrogate model for fast and precise generation of the temporally continuous interfacial flow fields.

Sequence prediction of missing data of spatio-temporal series of marine environmental elements based on ST-ResNet model

Spatial data of marine environmental elements are of great significance to the marine industry and marine safety, and they are also affected by external factors. This results in a new method for spatial and temporal sequence prediction of marine environmental elements based on the ST-ResNet model. Specifically, the temporal distance, period, and trend of marine regions are modelled by residual networks, and different residual sub-networks are used for each characteristic, in each sub-network, different weights are assigned to different branches and regions by simulating the changes of marine data, and the output is finally fitted to obtain the output. The ST-ResNet model was validated on the ERA5 dataset as an example, and compared with other prediction models, the ST-ResNet model achieved better results, which verified the effectiveness of the method in predicting the temporal and spatial sequences of marine environmental elements.

STDiff: Spatio-Temporal Diffusion for Continuous Stochastic Video Prediction

Predicting future frames of a video is challenging because it is difficult to learn the uncertainty of the underlying factors influencing their contents. In this paper, we propose a novel video prediction model, which has infinite-dimensional latent variables over the spatio-temporal domain. Specifically, we first decompose the video motion and content information, then take a neural stochastic differential equation to predict the temporal motion information, and finally, an image diffusion model autoregressively generates the video frame by conditioning on the predicted motion feature and the previous frame. The better expressiveness and stronger stochasticity learning capability of our model lead to state-of-the-art video prediction performances. As well, our model is able to achieve temporal continuous prediction, i.e., predicting in an unsupervised way the future video frames with an arbitrarily high frame rate. Our code is available at https://github.com/XiYe20/STDiffProject.

CoSTA: End-to-End Comprehensive Space-Time Entanglement for Spatio-Temporal Video Grounding

This paper studies the spatio-temporal video grounding task, which aims to localize a spatio-temporal tube in an untrimmed video based on the given text description of an event. Existing one-stage approaches suffer from insufficient space-time interaction in two aspects: i) less precise prediction of event temporal boundaries, and ii) inconsistency in object prediction for the same event across adjacent frames. To address these issues, we propose a framework of Comprehensive Space-Time entAnglement (CoSTA) to densely entangle space-time multi-modal features for spatio-temporal localization. Specifically, we propose a space-time collaborative encoder to extract comprehensive video features and leverage Transformer to perform spatio-temporal multi-modal understanding. Our entangled decoder couples temporal boundary prediction and spatial localization via an entangled query, boasting an enhanced ability to capture object-event relationships. We conduct extensive experiments on the challenging benchmarks of HC-STVG and VidSTG, where CoSTA outperforms existing state-of-the-art methods, demonstrating its effectiveness for this task.

Abstract 6223: Personalized MRI-informed forecasting of prostate cancer progression during active surveillance

Abstract Low and intermediate-risk prostate cancer (PCa) patients are eligible for active surveillance (AS), whereby treatment is delayed until progression to higher-risk disease is detected via longitudinal multiparametric magnetic resonance imaging (mpMRI) scans, biopsies, and prostate specific antigen (PSA) tests. However, AS relies on a population-based observational paradigm in which standard testing frequencies may not match the patient’s tumor dynamics, resulting in a delayed diagnosis of progressive disease. To address these issues, we propose to advance AS towards a patient-specific predictive paradigm by leveraging computational tumor forecasts obtained with a biomechanistic model informed by mpMRI and clinical data collected during standard AS. Here, we present a retrospective study with n = 16 PCa cases. All patients had three mpMRI scans obtained over 2.6 to 5.6 years during AS. Our biomechanistic model describes PCa growth in terms of the dynamics of tumor cell density as a combination of tumor cell mobility and net proliferation. The model is defined on the patient’s prostate geometry, which is segmented on T2-weighted MRI data. Tumor cell density estimates are derived from apparent diffusion coefficient (ADC) maps obtained from diffusion-weighted MRI data. To facilitate modeling, the longitudinal imaging datasets are non-rigidly co-registered for each patient. We initialize the model with the tumor cell density map obtained from the ADC map of the first scan. The model is parameterized by minimizing the model-data mismatch in tumor cell density at the second scan date, and then we perform a tumor forecast up to the third scan date. Finally, we build a logistic classifier from a panel of model-based biomarkers calculated from the personalized model forecasts (e.g., tumor volume, total proliferation activity) at the times of histopathological assessment (i.e., biopsy, surgery) to classify tumors as low risk (Gleason score 3+3, n = 16) or intermediate-high risk (Gleason score ≥ 3+4, n = 15). We obtained a concordance correlation coefficient (CCC) for tumor volume of 0.89 at both model calibration and forecasting horizon. The spatial fit of tumor cell density yielded a median Dice score and CCC of 0.80 and 0.59 at the second mpMRI date, and of 0.76 and 0.60 at the third mpMRI date, respectively. The logistic classifier of PCa risk yielded an area under the ROC curve of 0.90 and operates at optimal sensitivity, specificity, and accuracy of 86.7%, 93.8%, and 90.3%, respectively. The time trajectories of PCa risk obtained from the temporal predictions of the model-based biomarkers enabled identification of PCa progression by more than 1 year. Thus, while further development and validation over larger cohorts are required, these results suggest that our forecasting technology is a promising tool to predict PCa progression in AS and, hence, optimally guide monitoring and treatment decisions during AS. Citation Format: Guillermo Lorenzo Gomez, Chengyue Wu, Joshua P. Yung, John F. Ward, Hector Gomez, Alessandro Reali, Thomas E. Yankeelov, Aradhana M. Venkatesan, Thomas J. Hughes. Personalized MRI-informed forecasting of prostate cancer progression during active surveillance [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6223.

A deep contrastive framework for unsupervised temporal link prediction in dynamic networks

In dynamic networks, temporal link prediction aims to predict the appearance and disappearance of links in future snapshots based on the network structure we have observed. It also plays a crucial role in network analysis and predicting the behavior of the dynamic system. However, most existing studies only focus on supervised temporal link prediction problems, i.e., taking part of the links in future snapshots as supervised information. The ones that can solve the unsupervised temporal link prediction problem are mainly based on matrix decomposition, which lack the capability to automatically extract nonlinear spatial and temporal features from dynamic networks. The most challenging part of this problem is to extract the inherent evolution of the patterns hidden in dynamic networks in unsupervised ways. Inspired by the application and achievement of contrastive learning in network representation learning, we propose a novel deep Contrastive framework for unsupervised Temporal Link Prediction (CTLP). Our framework is based on a deep encoder-decoder architecture, which can capture the nonlinear structure and temporal features automatically and can predict future links of subsequent snapshots of dynamic networks in an unsupervised manner. Besides, CTLP could handle the multi-step temporal link prediction problem of dynamic networks through attenuation modeling across the snapshots. Extensive experiments on temporal link prediction show that our CTLP framework significantly outperforms state-of-the-art unsupervised methods, and even outperforms the supervised methods in some cases.

Audiovisual speech asynchrony asymmetrically modulates neural binding

Previous psychophysical and neurophysiological studies in young healthy adults have provided evidence that audiovisual speech integration occurs with a large degree of temporal tolerance around true simultaneity. To further determine whether audiovisual speech asynchrony modulates auditory cortical processing and neural binding in young healthy adults, N1/P2 auditory evoked responses were compared using an additive model during a syllable categorization task, without or with an audiovisual asynchrony ranging from 240 ms visual lead to 240 ms auditory lead. Consistent with previous psychophysical findings, the observed results converge in favor of an asymmetric temporal integration window. Three main findings were observed: 1) predictive temporal and phonetic cues from pre-phonatory visual movements before the acoustic onset appeared essential for neural binding to occur, 2) audiovisual synchrony, with visual pre-phonatory movements predictive of the onset of the acoustic signal, was a prerequisite for N1 latency facilitation, and 3) P2 amplitude suppression and latency facilitation occurred even when visual pre-phonatory movements were not predictive of the acoustic onset but the syllable to come. Taken together, these findings help further clarify how audiovisual speech integration partly operates through two stages of visually-based temporal and phonetic predictions.

Meibomian glands segmentation in infrared images with limited annotation.

To investigate a pioneering framework for the segmentation of meibomian glands (MGs), using limited annotations to reduce the workload on ophthalmologists and enhance the efficiency of clinical diagnosis. Totally 203 infrared meibomian images from 138 patients with dry eye disease, accompanied by corresponding annotations, were gathered for the study. A rectified scribble-supervised gland segmentation (RSSGS) model, incorporating temporal ensemble prediction, uncertainty estimation, and a transformation equivariance constraint, was introduced to address constraints imposed by limited supervision information inherent in scribble annotations. The viability and efficacy of the proposed model were assessed based on accuracy, intersection over union (IoU), and dice coefficient. Using manual labels as the gold standard, RSSGS demonstrated outcomes with an accuracy of 93.54%, a dice coefficient of 78.02%, and an IoU of 64.18%. Notably, these performance metrics exceed the current weakly supervised state-of-the-art methods by 0.76%, 2.06%, and 2.69%, respectively. Furthermore, despite achieving a substantial 80% reduction in annotation costs, it only lags behind fully annotated methods by 0.72%, 1.51%, and 2.04%. An innovative automatic segmentation model is developed for MGs in infrared eyelid images, using scribble annotation for training. This model maintains an exceptionally high level of segmentation accuracy while substantially reducing training costs. It holds substantial utility for calculating clinical parameters, thereby greatly enhancing the diagnostic efficiency of ophthalmologists in evaluating meibomian gland dysfunction.

Spatial–Temporal Differentiation and Trend Prediction of Coupling Coordination Degree of Port Environmental Efficiency and Urban Economy: A Case Study of the Yangtze River Delta

Green development is a primary path for ports and cities to achieve a low-carbon transition under the Sustainable Development Goals and a powerful driving force to elevate regional port–city relations to a high level of coordination. In this paper, twenty port cities in the Yangtze River Delta (YRD) were selected and port environmental efficiency (PEE) was calculated through the window SBM model, while the EW-TOPSIS model was used to evaluate high-quality urban economic development (HED). The coupling coordination degree (CCD) model, the kernel density model, GIS spatial analysis, and the grey prediction model were used to further explore the spatial–temporal dynamic evolution and prediction of the CCD between PEE and HED. The results suggested that: (1) PEE fluctuation in the YRD is increasing, with a trend of seaports achieving higher PEE than river ports; (2) HED in the YRD shows upward trends, and the polarization of individual cities is obvious; (3) Temporally, the CCD in the YRD has risen from 0.438 to 0.518. Shanghai consistently maintains intermediate coordination, and Jiangsu has experienced the most significant increase in CCD. Spatially, CCD is led by Lianyungang, Suzhou, Shanghai, and Ningbo-Zhoushan, displaying a decreasing distribution pattern from east to west. The projection for 2026 suggests that all port cities within the YRD will have transitioned to a phase of orderly development. To enhance the coordination level in the YRD, policymakers should consider the YRD as a whole to position the ports functionally and manage them hierarchically, utilize the ports to break down resource boundaries to promote the synergistic division of labor among cities, and then tilt the resources towards Anhui.

PixRevive: Latent Feature Diffusion Model for Compressed Video Quality Enhancement.

In recent years, the rapid prevalence of high-definition video in Internet of Things (IoT) systems has been directly facilitated by advances in imaging sensor technology. To adapt to limited uplink bandwidth, most media platforms opt to compress videos to bitrate streams for transmission. However, this compression often leads to significant texture loss and artifacts, which severely degrade the Quality of Experience (QoE). We propose a latent feature diffusion model (LFDM) for compressed video quality enhancement, which comprises a compact edge latent feature prior network (ELPN) and a conditional noise prediction network (CNPN). Specifically, we first pre-train ELPNet to construct a latent feature space that captures rich detail information for representing sharpness latent variables. Second, we incorporate these latent variables into the prediction network to iteratively guide the generation direction, thus resolving the problem that the direct application of diffusion models to temporal prediction disrupts inter-frame dependencies, thereby completing the modeling of temporal correlations. Lastly, we innovatively develop a Grouped Domain Fusion module that effectively addresses the challenges of diffusion distortion caused by naive cross-domain information fusion. Comparative experiments on the MFQEv2 benchmark validate our algorithm's superior performance in terms of both objective and subjective metrics. By integrating with codecs and image sensors, our method can provide higher video quality.

Optimising the surveillance of Aedes aegypti in Brazil by selecting smaller representative areas within an endemic city.

Arboviruses, such as dengue (DENV), zika (ZIKV), and chikungunya (CHIKV), constitute a growing urban public health threat. Focusing on Aedes aegypti mosquitoes, their primary vectors, is crucial for mitigation. While traditional immature-stage mosquito surveillance has limitations, capturing adult mosquitoes through traps yields more accurate data on disease transmission. However, deploying traps presents logistical and financial challenges, demonstrating effective temporal predictions but lacking spatial accuracy. Our goal is to identify smaller representative areas within cities to enhance the early warning system for DENV outbreaks. We created Sentinel Geographic Units (SGUs), smaller areas of 1 km2 within each stratum, larger areas, with the aim of aligning the Trap Positivity Index (TPI) and Adult Density Index (ADI) with their respective strata. We conducted a two-step evaluation of SGUs. First, we examined the equivalence of TPI and ADI between SGUs and strata from January 2017 to July 2022. Second, we assessed the ability of SGU's TPI and ADI to predict DENV outbreaks in comparison to Foz do Iguaçu's Early-Warning System, which forecasts outbreaks up to 4 weeks ahead. Spatial and temporal analyses were carried out, including data interpolation and model selection based on Akaike information criteria (AIC). Entomological indicators produced in small SGUs can effectively replace larger sentinel areas to access dengue outbreaks. Based on historical data, the best predictive capability is achieved 2 weeks after infestation verification. Implementing the SGU strategy with more frequent sampling can provide more precise space-time estimates and enhance dengue control. The implementation of SGUs offers an efficient way to monitor mosquito populations, reducing the need for extensive resources. This approach has the potential to improve dengue transmission management and enhance the public health response in endemic cities.

Temporal pattern-aware QoS prediction by Biased Non-negative Tucker Factorization of tensors

Dynamic quality of service (QoS) data contain rich temporal patterns of user-service interactions, which are vital for better understanding user behaviors and service conditions. Canonical polyadic (CP)-based latent factorization model has proven to be capable of capturing such patterns. However, it models the relations among latent features of user, service and time in a rigid and unnatural way, causing its failures in capturing the complex patterns when the target QoS data become massive. To address that issue, this paper proposes a Biased Non-negative Tucker Factorization of 3D tensors (BNTucF) model with the four-fold ideas: (1) utilizing a core tensor for modeling the complex interactions among latent features; (2) incorporating linear biases into the model for accurate descriptions on QoS fluctuation; (3) constraining the model to be non-negative for describing QoS non-negativity; (4) deducing a single latent factor-dependent, multiplicative updating scheme for training the model in an efficient density-oriented way. Empirical studies demonstrate that the proposed BNTucF can learn complex dynamic user-service interaction patterns more accurately, hence achieving accurate predictions on missing QoS data.

Deep learning-based dynamic forecasting method and application for ultra-deep fractured reservoir production

Addressing the complex challenges in dynamic production forecasting for the deep-ultra-deep fractured carbonate reservoirs in the Tarim Basin’s Tahe Oilfield, characterized by numerous influencing factors, strong temporal variations, high non-linearity, and prediction difficulties, We proposes a prediction method based on Gated Recurrent Unit networks (GRU). Initially, the production data and influencing factors are subjected to dimensionality reduction using Pearson correlation coefficient and principal component analysis methods to obtain multi-attribute time series data. Subsequently, deep learning modeling of time series data is conducted using Gated Recurrent Unit networks. The model is then optimized using the Optuna algorithm and applied to the dynamic production forecasting of the deep-ultra-deep fractured carbonate reservoirs in the Tahe Oilfield. The results demonstrate that the Gated Recurrent Unit network model optimized by Optuna excels in the dynamic production forecasting of the Tahe fractured carbonate reservoirs. Compared with the traditional method, the mean absolute error (MAE), the root mean square error (MSE) and the mean absolute percentage error (MAPE) are reduced by 0.04, 0.1 and 1.1, respectively. This method proves to be more adaptable to the production forecasting challenges of deep fractured reservoirs, providing an effective means to enhance model performance. It holds significant practical value and importance in guiding the development of fractured reservoirs.

Link prediction by continuous spatiotemporal representation via neural differential equations

With the continuous advancement of data science and machine learning, temporal link prediction has emerged as a crucial aspect of dynamic network analysis, providing significant research and application potential across various domains. While deep learning techniques have achieved remarkable results in temporal link prediction, most existing studies have focused on discrete model frameworks. These frameworks face limitations in capturing deep structural features and effectively aggregating temporal information. To address these limitations, we draw inspiration from neural differential equations to propose a Continuous Temporal Graph Neural Differential Equation (CTGNDE) network model for temporal link prediction. Specifically, we design a spatial graph Ordinary Differential Equation (ODE) to capture the spatial correlations inherent in complex spatiotemporal information. Then we employ Neural Controlled Differential Equation (Neural CDE) to learn complex evolution patterns and effectively aggregate temporal information. Finally, we characterize completely continuous and more accurate hidden state trajectories by coupling spatial and temporal messages. Experiments conducted on 10 real-world network datasets validated the superior performance of the CTGNDE model over the state-of-the-art baselines.