Prediction Results Research Articles

Self-Supervised Pretraining Based on Noise-Free Motion Reconstruction and Semantic-Aware Contrastive Learning for Human Motion Prediction

Human motion prediction is to forecast future human motions based on the observed ones. Currently, a prediction model always includes an encoder to extract the features of the observed motions and a decoder for motion prediction with the extracted features. However, existing encoders have limited abilities of anti-interference and semantic attribute learning; thus, the prediction results are not fully satisfactory. To solve this problem, we propose two novel pretext tasks, i.e., noise-free motion reconstruction and semantic-aware contrastive learning, to implement self-supervised pretraining on the encoder. The former extracts the features of a noise-added motion to reconstruct a noise-free motion. This pretext task can improve the ability of anti-interference of the encoder. The latter shortens the distance between two motions with the same motion category while widening the distance between two motions from different motion categories. This pretext task enables the encoder to capture the semantic commonality of the motions with the same motion category, thereby enhancing the encoder's ability of semantic attribute learning. After implementing the self-supervised pretraining, the pretrained encoder is connected with a decoder to build a prediction model called ReLe-GCN, which is fine-tuned for human motion prediction. The results on public datasets show the high accuracies of ReLe-GCN in both short-term and long-term motion predictions.

Research on prediction method of sensorless measurement of centrifugal pump operational state based on hybrid model

Centrifugal pumps are widely used in various fields such as industrial production and urban water supply. In order to achieve the monitoring of operational state of pumping systems, as well as the problems of inability to install sensors or high cost of sensors in practical engineering, the sensorless estimation method of centrifugal pump operational state provides a new opportunity for the monitoring and control of pumping systems. In this paper, a hybrid model based on QP model and neural network model is proposed to estimate the flow rate of rational pump by dividing the speed region. Taking a centrifugal pump as the research object, the operational points of the pump at different rotational speeds are acquired by experiments. Using the proposed method, a prediction model is established to predict the operational state of the pump. The prediction results are verified through experiments, and the error analysis of the prediction results is carried out. The results show that the proposed hybrid model can improve the prediction accuracy of centrifugal pump operational state and has certain practical value of engineering.

StructuralDPPIV: a novel deep learning model based on atom structure for predicting dipeptidyl peptidase-IV inhibitory peptides.

Diabetes is a chronic metabolic disorder that has been a major cause of blindness, kidney failure, heart attacks, stroke, and lower limb amputation across the world. To alleviate the impact of diabetes, researchers have developed the next generation of anti-diabetic drugs, known as dipeptidyl peptidase IV inhibitory peptides (DPP-IV-IPs). However, the discovery of these promising drugs has been restricted due to the lack of effective peptide-mining tools. Here, we presented StructuralDPPIV, a deep learning model designed for DPP-IV-IP identification, which takes advantage of both molecular graph features in amino acid and sequence information. Experimental results on the independent test dataset and two wet experiment datasets show that our model outperforms the other state-of-art methods. Moreover, to better study what StructuralDPPIV learns, we used CAM technology and perturbation experiment to analyze our model, which yielded interpretable insights into the reasoning behind prediction results. The project code is available at https://github.com/WeiLab-BioChem/Structural-DPP-IV.

Systematic analysis and expression of Gossypium ATG8 family reveals the roles of GhATG8f responding to salt stress in cotton.

Comprehensive analysis of Gossypium ATG8 family indicates that GhATG8f could improve salt tolerance of cotton by increasing SOD, POD and CAT activity and proline accumulation. In plants, autophagy is regulated by several genes that play important roles in initiating and controlling the process. ATG8, functioning as a protein similar to ubiquitin, is involved in crucial tasks throughout the autophagosome formation process. In this research, we conducted an extensive and all-encompassing investigation of 64 ATG8 genes across four varieties of cotton. According to the subcellular localization prediction results, 49 genes were found in the cytoplasm, 6 genes in the chloroplast, 1 gene in the peroxisome, 5 genes in the nucleus, and 3 genes in the extracellular region. Phylogenetic analysis categorized a total of 5 subfamilies containing sixty-four ATG8 genes. The expression of the majority of GhATG8 genes was induced by salt, drought, cold, and heat stresses, as revealed by RNA-seq and real-time PCR. Analysis of cis-elements in the promoters of GhATG8 genes revealed the predominant presence of responsive elements for plant hormones and abiotic stress, suggesting that GhATG8 genes might have significant functions in abiotic stress response. Furthermore, we additionally performed a gene interaction network analysis for the GhATG8 proteins. The salt stress resistance of cotton was reduced due to the downregulation of GhATG8f expression, resulting in decreased activity of CAT, SOD, and POD enzymes, as well as decreased fresh weight and proline accumulation. In summary, our research is the initial exploration of ATG8 gene components in cotton, providing a basis for future investigations into the regulatory mechanisms of ATG8 genes in autophagy and their response to abiotic stress.

Estimation of SOH of Lithium-Ion Batteries Based on PSO-Bi GRU-Attention Network

The State of Health (SOH) of lithium-ion batteries is a critical parameter that characterizes their actual lifespan, and its accurate assessment ensures the safe and reliable operation of batteries. However, in practical applications, SOH cannot be directly measured. To further improve the accuracy of SOH estimation for lithium-ion batteries, this study employs the Particle Swarm Optimization (PSO) algorithm to search for the optimal hyperparameters of the Bidirectional Gated Recurrent Unit (Bi GRU) neural network, enabling the prediction of time series information. Additionally, Attention Mechanism (AM) is integrated to allocate weights to the prediction results, resulting in the SOH prediction for lithium-ion batteries. The propose model is validated using the B0005 battery from the NASA lithium battery dataset. Experimental results demonstrate that, compared to the Bi GRU-Attention and Bi GRU models, the propose model reduces the Root Mean Square Error (RMSE) by 52.34% and 66.88%, respectively.

Modelling and predicting online vaccination views using bow-tie decomposition.

Social media has become increasingly important in shaping public vaccination views, especially since the COVID-19 outbreak. This paper uses bow-tie structure to analyse a temporal dataset of directed online social networks that represent the information exchange among anti-vaccination, pro-vaccination and neutral Facebook pages. Bow-tie structure decomposes a network into seven components, with two components, strongly connected component (SCC) and out-periphery component (OUT), emphasized in this paper: SCC is the largest strongly connected component, acting as an 'information magnifier', and OUT contains all nodes with a directed path from a node in SCC, acting as an 'information creator'. We consistently observe statistically significant bow-tie structures with different dominant components for each vaccination group over time. In particular, the anti-vaccination group has a large OUT, and the pro-vaccination group has a large SCC. We further investigate changes in opinions over time, as measured by fan count variations, using agent-based simulations and machine learning models. Across both methods, accounting for bow-tie decomposition better reflects information flow differences among vaccination groups and improves our opinion dynamics prediction results. The modelling frameworks we consider can be applied to any multi-stance temporal network and could form a basis for exploring opinion dynamics using bow-tie structure in a wide range of applications.

Development of Honey Badger-Cat Swarm Optimisation-Based Parallel Cascaded Deep Network for Software Bug Prediction Framework

Software bug prediction is mainly used for testing and code inspection. So, software bug prediction is carried out by network measures over the decades. But, the classical fault prediction method failed to obtain the semantic difference among various programs. Thus, it degrades the working of the prediction model, which is designed using these aspects. It is necessary to obtain the semantic difference for designing the prediction model accurately and effectively. In a software defect prediction system, it faces many difficulties in identifying the defect modules like correlation, irrelevance aspects, data redundancy, and missing samples or values. Consequently, many researchers are designed to detect software bug prediction that categorises faulty as well as non-faulty modules with software matrices. But, there are only a few works focussed to mitigate the class imbalance problem in bug prediction. In order to overcome the problem, it is required to develop an efficient software bug prediction method with the enhanced classifier. For this experimentation, the input data are taken from the standard online data sources. Initially, the input data undergo pre-processing phase and then, the pre-processed data are provided as input to the feature extraction by utilising the Auto-Encoder. These obtained features are utilised in getting the optimal fused features with the help of a new Hybrid Honey Badger Cat Swarm Algorithm (HHBCSA). Finally, these features are fed as input to the Optimised Parallel Cascaded Deep Network (OPCDP), where the “Extreme Learning Machine (ELM) and Deep Belief Network (DBN)” are used for the prediction of software bugs, in which the parameters from both classifiers are optimised by proposed HHBCSA algorithm. From the investigations, the recommended software bug prediction method offers a quicker bug prediction result, which helps to detect and remove the software bug easily and accurately.

Error prediction of a capacitor voltage transformer using dilated causal convolution and LSTM

To timely and accurately complete the error prediction of capacitor voltage transformers, this paper proposes an error prediction algorithm for capacitor voltage transformers that combines dilated causal convolution and LSTM. The algorithm is divided into two parts: one part updates the state of the input sequence using LSTM, while the other part extracts features through dilated causal convolution, maintaining data causality about the original sequence. Finally, the outputs of these two parts are combined to make the obtained sequence feature information more accurate and enriched. Experiments show that the MSE between the error prediction results obtained by this method and the actual error samples of capacitor voltage transformers is only 0.0003, which can accurately and efficiently complete the error prediction of capacitor voltage transformers.

Effect of zero-valent iron (ZVI) and biogas slurry reflux on methane production by anaerobic digestion of waste activated sludge.

This study aimed to improve anaerobic digestion (AD) efficiency through the addition of zero-valent iron (ZVI) and biogas slurry. This paper demonstrated that methane production was most effectively promoted at a biogas slurry reflux ratio of 60%. The introduction of ZVI into anaerobic systems does not enhance its bioavailability. However, both biogas slurry reflux and the combination of ZVI with biogas slurry reflux increase the relative abundance of microorganisms involved in the direct interspecific electron transfer (DIET) process. Among them, the dominant microorganisms Methanosaeta, Methanobacterium, Methanobrevibacter, and Methanolinea accounted for over 60% of the total methanogenic archaea. The Tax4Fun function prediction results indicate that biogas slurry reflux and the combination of ZVI with biogas slurry reflux can increase the content of key enzymes in the acetotrophic and hydrotrophic methanogenesis pathways, thereby strengthening these pathways. The corrosion of ZVI promotes hydrogen production, and the biogas slurry reflux provided additional alkaline and anaerobic microorganisms for the anaerobic system. Their synergistic effect promoted the growth of hydrotrophic methanogens and improved the activities of various enzymes in the hydrolysis and acidification phases, enhanced the system's buffer capacity, and prevented secondary environmental pollution. PRACTITIONER POINTS: Optimal methane production was achieved at a biogas slurry reflux ratio of 60%. Biogas slurry reflux in anaerobic digestion substantially reduced discharge. ZVI addition in combination with biogas slurry reflux facilitates the DIET process.

The effect of bta-miR-1296 on the proliferation and extracellular matrix synthesis of bovine mammary fibroblasts.

Mammary fibrosis in dairy cows is a chronic condition caused by mastitis, and can lead to serious culling of dairy cows resulting in huge economic losses in the dairy industry. MicroRNAs (miRNAs) exert an important role in regulating mammary gland health in dairy cows. This study investigated whether exosomal miRNAs in mammary epithelial cells can regulate the proliferation of bovine mammary fibroblasts (BMFBs) in mastitis. Liposome transfection technology was used to construct a cellular model of the overexpression and inhibition of miRNAs. The STarMir software, dual luciferase reporter gene test, real-time quantitative PCR (qRT-PCR), a Cell Counting Kit-8 (CCK-8), and a Western Blot and plate clone formation test were used to investigate the mechanism by which bta-miR-1296 regulates the proliferation of BMFBs. Target gene prediction results revealed that glutamate-ammonia ligase was a direct target gene by which bta-miR-1296 regulates cell proliferation. It was found that bta-miR-1296 significantly inhibited the proliferation of BMFBs. After BMFBs were transfected with a bta-miR-1296 mimic, mRNA expression in the extracellular matrix (ECM), α-smooth muscle actin (α-SMA), collagen type I alpha 1 chain (COL1α1) and collagen type III alpha 1 chain (COL3α1), and various cell growth factors (basic fibroblast growth factor (bFGF), vascular endothelial growth factor (VEGF), platelet-derived growth factor-BB (PDGF-BB), and transforming growth factor-β1 (TGF-β1)) were down-regulated, and the expressions of α-SMA, COL1α1, COL3α1, phospho-extracellular regulated protein kinases, phospho-protein kinaseB, TGF-β1, and phospho-Smad family member3 proteins were inhibited. In conclusion, bta-miR-1296 can inhibit the proliferation of BMFBs and the synthesis of ECM in BMFBs, thus affecting the occurrence and development of mammary fibrosis in dairy cows and laying the foundation for further studies to clarify the regulatory mechanism of mammary fibrosis.

Correlation between Surface Integrity and Low Cycle Fatigue Life of Machined Inconel 718

This work aims to improve the surface integrity and LCF life of machined Inconel 718. The correlation between the LCF life of Inconel 718 and various states of machined surface integrity is explored. In this paper, the surface integrity of Inconel 718 specimens is enhanced by low-plasticity burnishing (LPB). The LCF life of specimens with different surface integrity is predicted using a micro–macro finite element method (FEM). Firstly, the LCF specimens with different surface integrity are machined by turning (turned specimen), polishing (matrix specimen), and LPB process (LPBed specimen). Secondly, the LCF experiment is carried out to reveal the effect of surface integrity on LCF life. Finally, the LCF micro–macro FEM model is proposed to predict the LCF of machined Inconel 718 specimens. The representative volume element (RVE) model is established based on the measured surface integrity and microstructure of Inconel 718 specimens. The effect of surface integrity on LCF life is transformed to equivalent load. The micro–macro FEM model combined with Tanaka–Mura dislocation crack initiation theory and extended finite element method (XFEM) is applied to predict the LCF life of the machined specimens. The study results show that the LCF life of LPBed specimens can be improved by 90.5% and 36.1% compared with that of turned specimens and polished matrix specimens, respectively. The errors between FEM prediction results and experimental results are 13.1%, 9.2%, and 12.2%, respectively. The proposed micro–macro FEM model could be utilized to predict the LCF life of Inconel 718 with different surface integrities, and to apply the LCF life prediction further in industry.

A condition fitting method of temperature field for tunnel structure based on test measurements

Accurate temperature distribution of the tunnel lining under a fire scenario is crucial for conducting reliable thermal–mechanical investigations, as it directly impacts thermal stress and property degradation. In this regard, achieving consistency between simulation and actual response is imperative. However, the precision of most simulated temperature fields is hindered by a phenomenon known as the 'temperature platform', which occurs around the vaporization point of water. To this end, a Condition Fitting Method (CFM) is proposed in this paper. In CFM, optimal parameters are adjusted based on local historical measurements, enabling the acquisition of a detailed temperature field for the entire tunnel structure in real-time or even forward time step. Based on a preliminary segment fire test, the implementation of CFM successfully captures the temporal characteristics of the temperature platform on certain survey points. The results of both the current and forward time steps obtained from CFM demonstrate close agreement with the reference measurements obtained from the fire test, indicating consistency throughout the entire process. The maximum average error of the calculation result on survey points is 50.4 °C while the value for prediction result is 65.3 °C. For further investigation, a parametric study is conducted to identify the most crucial parameters. Additionally, numerical verification is performed to assess the feasibility of the proposed method and to compare the efficiency of different optimal algorithms. The results indicate that the proposed method has overcome the drawback of conventional simulation method that the temperature deviate from actual values at early stage. This achievement ensures that the proposed method provides a reliable temperature field for further investigations.

Whole-genome sequencing and evolutionary analysis of the wild edible mushroom, Morchella eohespera.

Morels (Morchella, Ascomycota) are an extremely desired group of edible mushrooms with worldwide distribution. Morchella eohespera is a typical black morel species, belonging to the Elata clade of Morchella species. The biological and genetic studies of this mushroom are rare, largely hindering the studies of molecular breeding and evolutionary aspects. In this study, we performed de novo sequencing and assembly of the M. eohespera strain m200 genome using the third-generation nanopore sequencing platform. The whole-genome size of M. eohespera was 53.81 Mb with a contig N50 of 1.93 Mb, and the GC content was 47.70%. A total of 9,189 protein-coding genes were annotated. Molecular dating showed that M. eohespera differentiated from its relative M. conica at ~19.03 Mya (million years ago) in Burdigalian. Evolutionary analysis showed that 657 gene families were contracted and 244 gene families expanded in M. eohespera versus the related morel species. The non-coding RNA prediction results showed that there were 336 tRNAs, 76 rRNAs, and 45 snRNAs in the M. eohespera genome. Interestingly, there was a high degree of repetition (20.93%) in the M. eohespera genome, and the sizes of long interspersed nuclear elements, short interspersed nuclear elements, and long terminal repeats were 0.83 Mb, 0.009 Mb, and 4.56 Mb, respectively. Additionally, selection pressure analysis identified that a total of 492 genes in the M. eohespera genome have undergone signatures of positive selection. The results of this study provide new insights into the genome evolution of M. eohespera and lay the foundation for in-depth research into the molecular biology of the genus Morchella in the future.

Comparison of prediction results of Dendrolimus punctatus larvae peak occurrence based on eight prediction models

AbstractIn order to clarify the prediction accuracy of eight models for predicting the peak occurrence of the first generation larvae of Dendrolimus punctatus and provide basis for the pest control, a catastrophe prediction model was established based on the peak occurrence of the first generation larvae of Dendrolimus punctatus in Qianshan City, Anhui Province from 1983 to 2016, and compared with other seven prediction models. Comparing the forecasting results in 2015 and 2016 with actual value and taking 1 head/plant as the error standard, the errors of multiple regression models with six factors as independent variables, namely, peak occurrence of pupae in overwintering generation, peak occurrence of eggs in the first generation, cumulative population in overwintering generation, peak occurrence of adults in overwintering generation, rainfall in early April and parasitic rate of Trichogrammatid in the first generation eggs of Dendrolimus punctatus, were 0.21 heads/plant and 0.23 heads/plant with accuracy rate of 100%. The errors of stepwise regression model with the same six factors were 0.23 head/plant and 0.29 head/plant. The prediction accuracy of artificial BP neural network model, Markov chain model, contingency table model, stationary time series model, and fuzzy comprehensive evaluation model was 100%, but variance period extrapolation model had an accuracy rate of 88%. The accuracy of catastrophe prediction model was related to the selection of catastrophe threshold. Comprehensive comparison of the above eight models, multiple regression, stepwise regression, artificial BP neural network, Markov chain model, stationary time series model, and catastrophe prediction model were more accurate.

Explanatory Object Part Aggregation for Zero-Shot Learning.

Zero-shot learning (ZSL) aims to recognize objects from unseen classes only based on labeled images from seen classes. Most existing ZSL methods focus on optimizing feature spaces or generating visual features of unseen classes, both in conventional ZSL and generalized zero-shot learning (GZSL). However, since the learned feature spaces are suboptimal, there exists many virtual connections where visual features and semantic attributes are not corresponding to each other. To reduce virtual connections, in this paper, we propose to discover comprehensive and fine-grained object parts by building explanatory graphs based on convolutional feature maps, then aggregate object parts to train a part-net to obtain prediction results. Since the aggregated object parts contain comprehensive visual features for activating semantic attributes, the virtual connections can be reduced by a large extent. Since part-net aims to extract local fine-grained visual features, some attributes related to global structures are ignored. To take advantage of both local and global visual features, we design a feature distiller to distill local features into a master-net which aims to extract global features. The experimental results on AWA2, CUB, FLO, and SUN dataset demonstrate that our proposed method obviously outperforms the state-of-the-arts in both conventional ZSL and GZSL tasks.

Convolutional neural network based data interpretable framework for Alzheimer’s treatment planning

Alzheimer’s disease (AD) is a neurological disorder that predominantly affects the brain. In the coming years, it is expected to spread rapidly, with limited progress in diagnostic techniques. Various machine learning (ML) and artificial intelligence (AI) algorithms have been employed to detect AD using single-modality data. However, recent developments in ML have enabled the application of these methods to multiple data sources and input modalities for AD prediction. In this study, we developed a framework that utilizes multimodal data (tabular data, magnetic resonance imaging (MRI) images, and genetic information) to classify AD. As part of the pre-processing phase, we generated a knowledge graph from the tabular data and MRI images. We employed graph neural networks for knowledge graph creation, and region-based convolutional neural network approach for image-to-knowledge graph generation. Additionally, we integrated various explainable AI (XAI) techniques to interpret and elucidate the prediction outcomes derived from multimodal data. Layer-wise relevance propagation was used to explain the layer-wise outcomes in the MRI images. We also incorporated submodular pick local interpretable model-agnostic explanations to interpret the decision-making process based on the tabular data provided. Genetic expression values play a crucial role in AD analysis. We used a graphical gene tree to identify genes associated with the disease. Moreover, a dashboard was designed to display XAI outcomes, enabling experts and medical professionals to easily comprehend the prediction results.

The influence of training datasets on LSTM-based irradiance prediction

In solar photovoltaic power generation, the prediction of solar irradiance is essential for minimizing energy costs and ensuring the provision of high-quality electricity. Deep learning models have recently gained popularity in the field, as numerous scholars have successfully employed them to predict solar irradiance. In line with this, this paper proposes three distinct methods for dividing the training dataset. Subsequently, these methods are employed in predicting solar irradiance using the LSTM-based model. Furthermore, an error analysis of the prediction results is conducted for each of the three models. The optimal training dataset division method is determined and proposed based on a comparison of the sizes of the three models using six error evaluation indexes.

Information Propagation Prediction Based on Spatial–Temporal Attention and Heterogeneous Graph Convolutional Networks

With the development of deep learning and other technologies, the research of information propagation prediction has also achieved important research achievements. However, the existing information diffusion studies either focus on the attention relationships of users or they predict the information according to the diffusion relationships of users, which makes the prediction results have certain limitations. Therefore, a prediction model has been proposed spatial–temporal attention heterogeneous graph convolutional networks (STAHGCNs). First, we use GCN to learn user influence relationships and user behavior relationships, and we propose a user representation fusion mechanism to learn the user characteristics. Second, to account for the dynamics of user behavior, a temporal attention mechanism strategy is used to encode time into the heterogeneous graph to obtain a more expressive user representation. Finally, the obtained user representation is input into the multihead attention mechanism for information propagation prediction. Experimental results performed on the Twitter, Douban, Digg, and Memetracker datasets have shown that the proposed STAHGCN model increased by 8.80% and 6.74% at hits@N and map@N, respectively, which are significantly better than the original latest DyHGCN model. The proposed STAHGCN model effectively integrates spatial factors, such as time factor, user influence, and behavior, which greatly improves the accuracy of information propagation prediction and has great significance for rumor monitoring and malicious account detection.

A meta-analysis of diabetes risk prediction models applied to prediabetes screening.

To provide a systematic overview of diabetes risk prediction models used for prediabetes screening to promote primary prevention of diabetes. The Cochrane, PubMed, Embase, Web of Science and China National Knowledge Infrastructure (CNKI) databases were searched for a comprehensive search period of 30 August 30, 2023, and studies involving diabetes prediction models for screening prediabetes risk were included in the search. The Quality Assessment Checklist for Diagnostic Studies (QUADAS-2) tool was used for risk of bias assessment and Stata and R software were used to pool model effect sizes. A total of 29 375 articles were screened, and finally 20 models from 24 studies were included in the systematic review. The most common predictors were age, body mass index, family history of diabetes, history of hypertension, and physical activity. Regarding the indicators of model prediction performance, discrimination and calibration were only reported in 79.2% and 4.2% of studies, respectively, resulting in significant heterogeneity in model prediction results, which may be related to differences between model predictor combinations and lack of important methodological information. Numerous models are used to predict diabetes, and as there is an association between prediabetes and diabetes, researchers have also used such models for screening the prediabetic population. Although it is a new clinical practice to explore, differences in glycaemic metabolic profiles, potential complications, and methods of intervention between the two populations cannot be ignored, and such differences have led to poor validity and accuracy of the models. Therefore, there is no recommended optimal model, and it is not recommended to use existing models for risk identification in alternative populations; future studies should focus on improving the clinical relevance and predictive performance of existing models.

Spatiotemporal evolution of deformation and LSTM prediction model over the slope of the deep excavation section at the head of the South-North Water Transfer Middle Route Canal

Slope deformation is one of the focal issues of concern during the normal operation and maintenance of the South-North Water Transfer Middle Route Project. To study the slope deformation evolution in the deep excavation section at the head of the canal, we applied 88 views of Sentinel-1A ascending image data from 2017 to 2019 and MT-InSAR(Multi-temporal InSAR) deformation monitoring technology to obtain long-time series deformation rates and cumulative deformation fields over the slope in the study area. Based on the analysis of the time-series monitoring data of the deformation field sample points, a LSTM (Long Short Term Memory Network) slope deformation predictive model was constructed to predict the slope deformation for the next 12 months at 12 sample points of the deep excavation slope. The impact of rainfall on slope deformation was investigated, and the reliability of the LSTM model was verified by using the measured data. The results show that the average annual deformation rate of the slope ranges from 10mm/a to 25mm/a, the maximum cumulative deformation is about 60 mm, and the slope of the excavated section is generally in an uplifted state. The rainfall-induced repeated uplift or subsidence of the canal slopes together with the peak deformation was closely related to the amount of rainfall during the wet season, and the longer the duration of the wet season, the more obvious the crest. Among the12 sample sites, the minimum and maximum deformation predicted using the LSTM model were 51.7 mm and 73.9 mm respectively, with the lowest correlation coefficient of 0.994 and the highest of 0.999. The maximum and minimum values of RMSE (Root Mean Square Error) were 4.4 mm and 3.6 mm respectively, indicating reliable prediction results. The results of the study can provide reference for the prevention and control of geological hazards in the South-North Water Transfer Canal.