Short-term Memory Neural Network Model Research Articles

Estimation of turnout irregularities using vehicle responses with improved BiLSTM and Gaussian process regression

Railway turnout irregularities pose a safety risk, but existing detection methods using expensive trains with inertial navigation systems and laser measurement devices are cost-prohibitive and time-consuming for high-speed railways. This paper proposes a low-cost and real-time solution for estimating railway turnout irregularities using vehicle body acceleration, allowing for continuous monitoring of dynamic changes. A Bayesian-optimized improved bidirectional long short-term memory neural network model (BO-BiLSTM) was proposed in this paper, utilizing the vehicle's vertical and lateral acceleration as input to achieve point estimation of irregularities, with an estimation error approximately 50% lower than that of traditional recurrent neural networks. In addition, an interval estimation model combining BO-BiLSTM and Gaussian process regression (BO-BiLSTM-GPR) was proposed. Compared to the traditional interval estimation method Bootstrap, under the condition of narrow interval width, GPR ensures that over 90% of the measured values fall within the estimated interval and provides a better solution to the trade-off between interval estimation reliability and uncertainty.

Runoff Prediction in the Xijiang River Basin Based on Long Short-Term Memory with Variant Models and Its Interpretable Analysis

The Long Short-Term Memory (LSTM) neural network model is an effective deep learning approach for predicting streamflow, and the investigation of the interpretability of deep learning models in streamflow prediction is of great significance for model transfer and improvement. In this study, four key hydrological stations in the Xijiang River Basin (XJB) in South China are taken as examples, and the performance of the LSTM model and its variant models in runoff prediction were evaluated under the same foresight period, and the impacts of different foresight periods on the prediction results were investigated based on the SHapley Additive exPlanations (SHAP) method to explore the interpretability of the LSTM model in runoff prediction. The results showed that (1) LSTM was the optimal model among the four models in the XJB; (2) the predicted results of the LSTM model decreased with the increase in foresight period, with the Nash–Sutcliffe efficiency coefficient (NSE) decreasing by 4.7% when the foresight period increased from one month to two months, and decreasing by 3.9% when the foresight period increased from two months to three months; (3) historical runoff had the greatest impact on streamflow prediction, followed by precipitation, evaporation, and the North Pacific Index (NPI); except evaporation, all the others were positively correlated. The results can provide a reference for monthly runoff prediction in the XJB.

Research on multi-lane energy-saving driving strategy of connected electric vehicle based on vehicle speed prediction

In order to enhance the energy-saving potential of electric vehicles, a lane change decision method based on vehicle-to-everything (V2X) is designed to further improve the economics of intelligent connected electric vehicles. Firstly, the traversal test of electric vehicles is conducted at different speeds and accelerations to construct an energy consumption cloud model that reflects the mapping relationship between electric vehicle speed, acceleration and power. Next, the traffic flow information from V2X is used to train the long short-term memory neural network model optimized by particle swarm optimization (PSO-LSTM) for the prediction of the future speed of the vehicle in front of each lane. Then, according to the established energy consumption cloud model, the power performance corresponding to the predicted vehicle speed is obtained. Finally, a lane change decision method based on analytic hierarchy process (AHP) is established, and it is applied in four typical parallel scenarios to verify the robustness and effectiveness of the decision method. Simulation tests were conducted in a simulated urban traffic environment, involving both single-lane change scenarios and continuous lane change scenarios. The results show that this method can accurately and effectively select the lane with the best economic performance. Compared with the driving strategy of selecting a fixed lane, the energy consumption can be improved by up to 27.2% in the continuous lane change scenario.

Modeling and optimization of the self-embrittle corrosive bifunctional detergent for corrosive deep decontamination of stainless steel surface by RAFT one-pot method based on machine learning and response surface methodology

Nuclear facilities generate lots of contaminated stainless steel metallic material during maintenance and decommissioning. As a new radioactive decontamination method, the self-embrittle decontamination method has the advantage of less secondary contaminants and being able to operate machinery remotely. Adding a certain amount of corrosive components to the self-embrittle compound detergent can achieve the dual functions of self-embrittlement and corrosive decontamination. It was used to optimized the response surface methodology (RSM) and train and evaluate four different machine learning models by the recorded data set. The purpose of the analysis was to quantify the accuracy of the corrosion decontamination effect of RSM model and four types of machine learning model. The results exhibits that the long short-term memory neural network (LSTM) model performs well. The prepared detergent can achieve the average corrosion depth of 5.9454 μm on stainless steel, which can satisfy the corrosion decontamination of radioactively contaminated stainless steel surfaces.

Online Reviews Sentiment Analysis and Product Feature Improvement with Deep Learning

The text mining of online reviews is currently a popular research direction of e-commerce and is considered the next blue ocean. Online reviews can dig out consumer preferences and provide theoretical guidance for the improvement of product features. However, current research mostly focuses on sentiment analysis methods and rarely involves feature extraction and large-scale data recognition. This article uses word segmentation technology to create a new feature extraction method. With the long short-term memory neural network and latent Dirichlet allocation topic model, we propose a product feature improvement model—CESC (Consumer online reviews–Extract short text–Sentiment analysis–Cluster feature). The model can derive the product features and attitudes that consumers prefer based on consumer online reviews and use it to improve product features. According to the experimental results of three electronic products sold on the e-commerce platform, the model can effectively dig out consumer preferences for online reviews. Enterprises can improve the quality of products and services, better meet the needs of consumers, promote consumers’ consumption, and achieve the enterprises’ goals and values.

Dynamic evaluation method for Time-variant reliability of structural safety of Concrete-faced rockfill dam

The evaluation of the structural safety risk for concrete-faced rockfill dams (CFRDs) is susceptible to the time-varying parameters and the environments of topography, geology, and operations. Traditional evaluation methods, due to their involvement in numerical simulation, cannot meet the current requirements for intelligent monitoring in terms of timeliness and practicality. Therefore, a dynamic evaluation method of reliability in CFRDs is proposed in this paper. Due to the nonlinear mapping relationship between the time-variant reliability and the monitoring characterizations of CFRDs, the long and short-term memory neural network (LSTM) and support vector machine (SVM) models are introduced to construct the potential functional relationships between the reliability sequences and the monitoring characterizations, and the matching between the accuracy of the two models and the frequency of monitoring characterizations has been fully studied, forming an adaptive evaluation model of safety risk for the CFRDs according to the different monitoring frequencies of different monitoring items. The application of Sanbanxi CFRD shows an average relative error of less than 1% and 5% for dam slope stability and slab cracking simulations, respectively. These results demonstrate that the LSTM and SVM models with adaptive monitoring frequency exhibit high fitting and prediction accuracy and applicability, and possess theoretical and engineering application value.

Future degradation trajectory prediction of lithium-ion battery based on a three-step similarity evaluation criterion for battery selection and transfer learning

Accurate remaining useful life (RUL) prediction is important for battery replacement and health maintenance. Due to the complex nonlinear nature of battery degradation, it is difficult to predict the long-term future degradation trend of the battery using its early historical data. Thus, an RUL prediction framework with transfer learning is proposed in this paper. Firstly, considering that the source domain batteries may have different distributions from target batteries due to the inherent inconsistencies between cells, a three-step similarity evaluation criterion is put forward to select similar batteries from the source domain to enhance the ability of positive migration. Secondly, to improve the prediction accuracy, a long short-term memory (LSTM) neural network model with an attention mechanism is developed based on the selected source domain batteries. Finally, a transfer strategy is introduced, the parameters of the LSTM layer and attention layer are shared and those of the rest layers are retrained with limited early historical data of the target battery. The experimental results show that the framework proposed in the paper can effectively predict the future degradation path of the target battery due to full use of the commonality of the batteries degradation and the characteristic of the target battery itself.

Simulation and Driving Factor Analysis of Satellite-Observed Terrestrial Water Storage Anomaly in the Pearl River Basin Using Deep Learning

Accurate estimation of terrestrial water storage (TWS) and understanding its driving factors are crucial for effective hydrological assessment and water resource management. The launches of the Gravity Recovery and Climate Experiment (GRACE) satellites and their successor, GRACE Follow-On (GRACE-FO), combined with deep learning algorithms, have opened new avenues for such investigations. In this study, we employed a long short-term memory (LSTM) neural network model to simulate TWS anomaly (TWSA) in the Pearl River Basin (PRB) from 2003 to 2020, using precipitation, temperature, runoff, evapotranspiration, and leaf area index (LAI) data. The performance of the LSTM model was rigorously evaluated, achieving a high average correlation coefficient (r) of 0.967 and an average Nash–Sutcliffe efficiency (NSE) coefficient of 0.912 on the testing set. To unravel the relative importance of each driving factor and assess the impact of different lead times, we employed the SHapley Additive exPlanations (SHAP) method. Our results revealed that precipitation exerted the most significant influence on TWSA in the PRB, with a one-month lead time exhibiting the greatest impact. Evapotranspiration, runoff, temperature, and LAI also played important roles, with interactive effects among these factors. Moreover, we observed an accumulation effect of precipitation and evapotranspiration on TWSA, particularly with shorter lead times. Overall, the SHAP method provides an alternative approach for the quantitative analysis of natural driving factors at the basin scale, shedding light on the natural dominant influences on TWSA in the PRB. The combination of satellite observations and deep learning techniques holds promise for advancing our understanding of TWS dynamics and enhancing water resource management strategies.

Prediction of Chinese-lantern submarine hose tension variation in extreme sea states based on data-driven methods

To predict the variation of Chinese-lantern type submarine hose tension in extreme sea states is of great significance for the monitoring, operation and lifetime assessment of catenary anchor leg mooring (CALM) systems. In this research, the machine learning algorithm in radial basis function (RBF) neural network and the deep learning algorithm in long short-term memory (LSTM) neural network model were used to predict the variation of submarine hose tension respectively, and the prediction performance of the two algorithms on the time-series variation of submarine hose tension was evaluated. During the neural network construction, the six-degree-of-freedom (6-DOF) motion of CALM buoy is taken as the training parameter, while the historical tension variation data is also integrated for the correction of the prediction results. The training data are collected by hydrodynamic coupling simulation of the established CALM system with submarine hoses. Different parameter values selection of the neural network model is investigated to determine the optimal structure of the RBF and LSTM neural network models, including spread, initial learning rate, number of nodes in the hidden layer and batch size. Since the direction of environmental loads has a large effect on the response of the submarine hose, four combinations of wind, wave and currents under 100-year self-existing sea state are considered in the research. Through the case study and sensitivity analysis, it is demonstrated that the non-linear relationship between 6-DOF motion and hose tension is varied for different environmental load directions and LSTM neural network is suitable for predicting the variation of Chinese-lantern type submarine hose tension in CALM systems under different combinations of extreme sea states.

Offshore wind speed assessment with statistical and attention-based neural network methods based on STL decomposition

This work proposes a novel offshore wind speed prediction approach by combining statistical method and attention-based neural network with seasonal-trend decomposition procedure with loess (STL). STL is utilized to decompose the processed data into season, trend and residual terms. Then, an attention-based long short-term memory neural network model (AT-LSTM), possessing the advantages of generalization and high-dimensional function approximation, is modeled to train season and residual terms with obvious volatility characteristics. And a hybrid model of auto-regressive integrated moving average (ARIMA) and LSTM is applied to predict the linear and nonlinear sequences in trend term with relatively gentle feature, yielding the proposed STL-AR-LSTM-ATLSTM model. Wherein, the proposed method is verified through sufficient pre-judgment experiments on season, trend and residual terms, as well as detailed multi-model comparison experiments. Finally, microcosmic prediction results and predicted statistical frequency distributions indicate that the new model has better prediction effect on offshore wind, compared to ARIMA, AT-LSTM, ARIMA-AT-LSTM models. Meanwhile, the presented model can reduce the lag problem of predicted values and perform well in the prediction of extreme values.

Prediction and Impact Analysis of Passenger Flow in Urban Rail Transit in the Postpandemic Era

In the postpandemic era, exploring the relationship between the daily new COVID-19 cases and passenger flow in urban rail transit can help effectively predict the impact of future pandemic situations on rail transit. In this study, based on a gated recurrent unit (GRU) neural network model, the daily passenger flow in urban rail transit in the postpandemic era was predicted, and the results were compared with those obtained using the long short-term memory (LSTM) neural network and other conventional time series analysis models such as SARIMA (seasonal autoregressive integrated moving average). Based on the trained GRU model, a partial dependence plot (PDP) was adopted to explore the quantitative relationship between the daily passenger flow and the daily new cases or weather attribute. The results showed that (1) the prediction accuracy of the GRU neural network model was 95.25%, which was the highest among the prediction models studied, indicating that the GRU could achieve the best performance. (2) The GRU model did not fluctuate significantly in the initial training stage, and its convergence rate was higher than that of the LSTM. (3) The number of daily new cases was negatively correlated with the daily passenger flow. For every new case on the previous day, the daily passenger flow fell by an average of 54,600 person-times. (4) Compared with no rain condition, the daily passenger flow decreased by 207,600 person-times on an average on rainy days. In summary, the neural network could achieve accurate prediction, while the PDP could compensate for the “black box” disadvantage of nonparametric models, owing to which the quantitative relationship between the number of new cases and daily passenger flow could be successfully explored. Our study can serve as a basis for demand prediction, operational organization, and policy implementation related to urban rail transit.

Early warning of exchange rate risk based on structural shocks in international oil prices using the LSTM neural network model

The Russia–Ukraine conflict has caused an external shock to the oil market as the exchange rate risk increased with significant depreciation of major currencies relative to the U.S. dollar (USD). In this paper, structural shocks to oil prices calculated by the Ready (2018) method are introduced in addition to traditional exchange rate predictor variables such as oil prices. Utilising a long short-term memory neural network model, we verify the early warning effect of structural shocks to international oil prices on exchange rate risk in the context of the Russia–Ukraine geo-conflict. We have several empirical findings. First, the variance decomposition method indicates that oil price shocks affect the euro-dollar exchange rate risk. After the conflict between Russia and Ukraine, the euro exchange rate risk continued to rise significantly, mainly affected by supply shocks. Second, compared with traditional prediction models, the LSTM neural network model has a higher prediction accuracy, which is conducive to improving the early warning performance of exchange rate risks and maintaining the stability of the foreign exchange market. Third, the introduction of oil price shock can fully extract and use the information from oil price changes, which significantly improves the learning performance of the oil price early warning model of exchange rate risk. Fourth, the introduction of oil price risk shocks in the whole sample, the introduction of oil price demand shocks before the Russia-Ukraine conflict, and the introduction of oil price supply shocks, especially negative supply shocks, after the conflict between Russia and Ukraine, can warn exchange rate risks more effectively. This study proposes a new method for early warning of exchange rate risk that can help cross-border enterprises and government policymakers to respond to unexpected events and effectively mitigate associated risks.

A deep-learning based model for fracture network characterization constrained by induced micro-seismicity and tracer test data in enhanced geothermal system

Fracture network developed in hot dry rock (HDR) primarily controls the flow and heat transport in enhanced geothermal system (EGS). Characterization of fracture network in deep geothermal reservoir is among the most challenging tasks. An interpretation framework for explicit delineating complex fracture network is proposed based on induced micro-seismic events, hydraulic stimulation and tracer test data, which are often available in EGS sites. It is processed by (1) the spatial distribution of fracture intensity, size and hydraulic aperture constrained by the hydraulic diffusivity derived from the spatio-temporal distribution of micro-seismic events, (2) numerical models and long short-term memory (LSTM) neural network models for the hydraulic stimulation and tracer test, and (3) parameter inversion model based on the multi-objective Harris hawks optimization (MOHHO) algorithm searching for optimal fracture network parameters minimizing the deviations between model predictions and field observations. This method has been applied to the Habanero EGS site, Australia. The prediction accuracy of the injection wellhead pressure and produced tracer concentration of the inversed micro-seismicity mapping DFN model is improved by 61.16% and 67.35%, respectively, compared with the stochastic DFN model. These results emphasize the importance to integrate induced micro-seismicity monitoring data for the characterization of hydraulically stimulated fracture network.

UNISON framework for user requirement elicitation and classification of smart product-service system

Smart product-service system (smart PSS) has become the manufacturers’ new revenue and sustainable competitiveness sources with the rise of digital servitization. To configure a customer-tailored smart PSS effectively, requirement analysis plays a significant role. Till now, the heterogeneity feature of smart PSS requirements, the usage of user-generated data, and data-driven paradigm have not been considered sufficiently. Hence, this study develops a UNISON framework based on the integration of deep learning approaches for eliciting and evaluating the user requirements for smart PSS with a systematic process, data-driven manner, and quantitative results. Adopting the user-centric perspective, this study employed massive user online reviews as the niche for requirement analysis. The bidirectional long short-term memory (Bi-LSTM) neural network and bi-term topic model (BTM) were employed to extract and elicit the smart PSS requirements in both product and service aspects. Further, a multi-dimensional and automatic requirement evaluation method integrating the sentiment analysis, IPA-Kano model, and opportunity algorithm was proposed for assessing, classifying, and prioritizing the identified requirement items. A smart PSS requirement landscape map was also developed to present the analysis results visually and guide decision-making. An empirical study has been conducted for the smart cleaning robot service system to prove the feasibility of the suggested approach. Enabled by the hybrid application of big data analytics and deep learning techniques, this study can empower the smart PSS provider to understand the user requirements comprehensively and upgrade the smart PSS design effectively.

A Novel Gait Phase Recognition Method Based on DPF-LSTM-CNN Using Wearable Inertial Sensors.

Gait phase recognition is of great importance in the development of rehabilitation devices. The advantages of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) are combined (LSTM-CNN) in this paper, then a gait phase recognition method based on LSTM-CNN neural network model is proposed. In the LSTM-CNN model, the LSTM layer is used to process temporal sequences and the CNN layer is used to extract features A wireless sensor system including six inertial measurement units (IMU) fixed on the six positions of the lower limbs was developed. The difference in the gait recognition performance of the LSTM-CNN model was estimated using different groups of input data collected by seven different IMU grouping methods. Four phases in a complete gait were considered in this paper including the supporting phase with the right hill strike (SU-RHS), left leg swimming phase (SW-L), the supporting phase with the left hill strike (SU-LHS), and right leg swimming phase (SW-R). The results show that the best performance of the model in gait recognition appeared based on the group of data from all the six IMUs, with the recognition precision and macro-F1 unto 95.03% and 95.29%, respectively. At the same time, the best phase recognition accuracy for SU-RHS and SW-R appeared and up to 96.49% and 95.64%, respectively. The results also showed the best phase recognition accuracy (97.22%) for SW-L was acquired based on the group of data from four IMUs located at the left and right thighs and shanks. Comparably, the best phase recognition accuracy (97.86%) for SU-LHS was acquired based on the group of data from four IMUs located at left and right shanks and feet. Ulteriorly, a novel gait recognition method based on Data Pre-Filtering Long Short-Term Memory and Convolutional Neural Network (DPF-LSTM-CNN) model was proposed and its performance for gait phase recognition was evaluated. The experiment results showed that the recognition accuracy reached 97.21%, which was the highest compared to Deep convolutional neural networks (DCNN) and CNN-LSTM.

Development of predictive algorithms for the wear resistance of denture teeth materials

ObjectivesTo investigate the wear resistance of conventional, CAD-milled and 3D-printed denture teeth in vitro with simulated aging. To use the collected data to train single time series sample model LSTM and provide proof of concept. MethodsSix denture teeth materials (Three Conventional; Double-cross linked PMMA (G1), Nanohybrid composite (G2), PMMA with microfillers (G3), CAD-milled (G4), two 3D-printed teeth (G5, G6) (Total n = 60) underwent simulation for 24 and 48 months of linear reciprocating wear using a universal testing machine (UFW200, NeoPlus) under 49 N load, 1 Hz and linear stroke of 2 mm in an artificial saliva medium. Single samples were parsed through Long Short-Term Memory (LSTM) neural network model using Python. To determine minimal simulation times, multiple data splits for training were trialled (10/20/30/40%). Scanning electron microscopy (SEM) was performed for material surface evaluation. Results3D printed tooth material (G5) had the lowest wear resistance (59 ± 35.71 μm) whereas conventional PMMA with microfillers (G3) shown the highest wear rate (303 ± 0.06 μm) after 48 months of simulation. The LSTM model successfully predicted up to 48 months wear using 30% of the collected data. Compared with the actual data, the model had a root-mean-square error range between 6.23 and 88.56 μm, mean-absolute-percentage-error 12.43-23.02% and mean-absolute-error 7.47-70.71 μm. SEM images revealed additional plastic deformations and chipping of materials, that may have introduced data artifacts. Conclusions3D printed denture teeth materials showed the lowest wear out of all studied for 48 months simulation. LSTM model was successfully developed to predict wear of various denture teeth. The developed LSTM model has the potential to reduce simulation duration and specimen number for wear testing of various dental materials, while potentially improving the accuracy and reliability of wear testing predictions. This work paves the way for generalized multi-sample models enhanced with empirical information.

Concrete Dam Deformation Prediction Model Research Based on SSA–LSTM

In the context of dam deformation monitoring, the prediction task is essentially a time series prediction problem that involves non-stationarity and complex influencing factors. To enhance the accuracy of predictions and address the challenges posed by high randomness and parameter selection in LSTM models, a novel approach called sparrow search algorithm–long short-term memory (SSA–LSTM) has been proposed for predicting the deformation of concrete dams. SSA–LSTM combines the SSA optimization algorithm with LSTM to automatically optimize the model’s parameters, thereby enhancing the prediction performance. Firstly, a concrete dam was used as an example to preprocess the historical monitoring data by cleaning, normalizing, and denoising, and due to the specificity of the data structure, multi-level denoising of abnormal data was performed. Second, some of the data were used to train the model, and the hyperparameters of the long and short-term memory neural network model (LSTM) were optimized by the SSA algorithm to better match the input data with the network structure. Finally, high-precision prediction of concrete dam deformation was carried out. The proposed model in this study significantly improves the prediction accuracy in dam deformation forecasting and demonstrates effectiveness in long-term time series deformation prediction. The model provides a reliable and efficient approach for evaluating the long-term stability of dam structures, offering valuable insights for engineering practices and decision-making.

Real-Time Intersection Turning Movement Flow Forecasting Using a Parallel Bidirectional Long Short-Term Memory Neural Network Model

Real-time forecasting of intersection turning movements is a critical requirement for predictive traffic signal control at signalized intersections. Several traffic flow forecasting models have been developed in the past; however, most of them have focused only on the road segment-level traffic instead of the turning movement flow (TMF) at intersections. Therefore, in this paper, we propose a new TMF forecasting model, which consists of a combination of two neural network models, namely the stacked bidirectional long short-term memory and the traditional multi-layer perceptron model; this combination will enable effective learning for both short- and long-term time-varying patterns. Moreover, extensive computational experiments, using two years of turning movement counts at 22 intersections in the city of Milton, Ontario, Canada, explore the performance advantage of the proposed model in comparison with several state-of-the-art base models for forecasting accuracy, robustness, and transferability.

Deep learning-based prediction for time-dependent chloride penetration in concrete exposed to coastal environment

The application of deep learning methods in civil engineering has gained significant attention, but its usage in studying chloride penetration in concrete is still in its early stages. This research paper focuses on predicting and analyzing chloride profiles using deep learning methods based on measured data from concrete exposed for 600 days in a coastal environment. The study reveals that Bidirectional Long Short-Term Memory (Bi-LSTM) and Convolutional Neural Network (CNN) models exhibit rapid convergence during the training stage, but fail to achieve satisfactory accuracy when predicting chloride profiles. Additionally, the Gate Recurrent Unit (GRU) model proves to be more efficient than the Long Short-Term Memory (LSTM) model, but its prediction accuracy falls short compared to LSTM for further predictions. However, by optimizing the LSTM model through parameters such as the dropout layer, hidden units, iteration times, and initial learning rate, significant improvements are achieved. The mean absolute error (MAE), determinable coefficient (R2), root mean square error (RMSE), and mean absolute percentage error (MAPE) values are reported as 0.0271, 0.9752, 0.0357, and 5.41%, respectively. Furthermore, the study successfully predicts desirable chloride profiles of concrete specimens at 720 days using the optimized LSTM model.

Prediction of molten pool temperature in laser solid forming based on PSO-LSTM

The temperature of the molten pool in Laser Solid Forming has a direct effect on the dimensional accuracy and mechanical properties of the parts. Accurate prediction of the melt pool temperature is important to ensure the stability of the melt pool temperature and to improve the forming accuracy and quality of the part. In order to accurately predict the melt pool temperature, this study proposes a melt pool temperature prediction method based on particle swarm optimization (PSO) optimised long short-term memory neural network (LSTM). Using IR camera to obtain melt pool temperature data and establish long short-term memory neural network melt pool temperature prediction model based on experimental data. Optimization of the initial learning rate and the number of hidden layer units of the long short-term memory neural network model using the particle swarm optimization algorithm to build a PSO-LSTM model for prediction of melt pool temperature. The results show that the PSO-LSTM prediction model outperforms the long short-term memory neural network and Ridge Regression models in all evaluation indicators and can achieve accurate prediction of melt pool temperature.