Long Short-term Memory Network Model Research Articles

Application of LSTM-based Body Fatigue Detection Algorithm in Tai Chi Training

Tai Chi training sessions are often lengthy, and athletes are prone to experiencing fatigue during the process. Timely detection of body fatigue can help athletes prevent injuries caused by excessive fatigue. This study combines Long Short-Term Memory (LSTM) networks with facial muscle activity detection models to propose a novel fatigue detection algorithm. In this algorithm, the limitations of LSTM networks in capturing future information are addressed by introducing an improved LSTM network model and combining attention mechanisms to highlight the important features of physical fatigue. This study utilizes hyperspectral imaging technology to extract real-time muscle fatigue signals from the faces of subjects in the dataset. Performance validation of the proposed model shows that it effectively extracts facial fatigue features with a detection accuracy of 97.67% and a recall rate of 96.78%, outperforming other existing models in this field. The model constructed through research has excellent performance and has broad application prospects in current and future technological development due to its high flexibility and adaptability, providing support and innovation momentum for different industries.

A novel bidirectional LSTM network model for very high cycle random fatigue performance of CFRP composite thin plates

This study proposes a novel Double-layer Bidirectional Long Short-Term Memory (BiLSTM) neural network model, shorted as TDA-BiLSTM, which integrates Transfer learning and Attention mechanisms. The model aims to predict the fatigue life of carbon fiber reinforced polymer (CFRP) thin plate structures subjected to very high cycle random vibration fatigue loads. Distinct from conventional servo-hydraulic and ultrasonic fatigue testing methods, this research pioneers the use of a vibration table for very high cycle fatigue (VHCF) testing to fill the gap in the field. By training and validating data across various life ranges, the TDA-BiLSTM model demonstrates significant advantages in training speed and predicting accuracy. Its bidirectional structure and attention mechanism effectively capture complex patterns and long-term dependencies in sequence data. Experimental results indicate that the TDA-BiLSTM model achieves significantly lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) compared to Long Short-Term Memory (LSTM)and Long Short-Term Memory with Transfer learning (Tr-LSTM) models across different life ranges on the ε-N curve, indicating higher accuracy and stability in strain life prediction tasks. Additionally, an analysis of typical damage areas using Scanning Electron Microscopy (SEM) reveals the failure mechanisms of CFRP plates under very high cycle vibration fatigue loads.

Cloud computing load prediction method based on CNN-BiLSTM model under low-carbon background

With the establishment of the "double carbon” goal, various industries are actively exploring ways to reduce carbon emissions. Cloud data centers, represented by cloud computing, often have the problem of mismatch between load requests and resource supply, resulting in excessive carbon emissions. Based on this, this paper proposes a complete method for cloud computing carbon emission prediction. Firstly, the convolutional neural network and bidirectional long-term and short-term memory neural network (CNN-BiLSTM) combined model are used to predict the cloud computing load. The real-time prediction power is obtained by real-time prediction load of cloud computing, and then the carbon emission prediction is obtained by power calculation. Develop a dynamic server carbon emission prediction model, so that the server carbon emission can change with the change of CPU utilization, so as to achieve the purpose of low carbon emission reduction. In this paper, Google cluster data is used to predict the load. The experimental results show that the CNN-BiLSTM combined model has good prediction effect. Compared with the multi-layer feed forward neural network model (BP), long short-term memory network model (LSTM ), bidirectional long short-term memory network model (BiLSTM), modal decomposition and convolution long time series neural network model (CEEMDAN-ConvLSTM), the MSE index decreased by 52%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} , 50%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}, 34%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} and 45%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} respectively.

A Soft Sensor for Flow Estimation and Uncertainty Analysis Based on Artificial Intelligence: A Case Study of Water Supply Systems

The fourth industrial revolution has transformed the industry, with information technology playing a crucial role in this shift. The increasing digitization of industrial systems demands efficient sensing and control methods, giving rise to soft sensors that have the potential to replace traditional physical sensors in order to reduce costs and enhance efficiency. This study explores the implementation of an artificial neural network (ANN) based soft sensor model in a water supply system to predict flow rates within the system. The soft sensor is centered on a Long Short-Term Memory (LSTM) artificial neural network model using Monte Carlo dropout to reduce uncertainty and improve estimation performance. Based on the results of this work, it is concluded that the proposed soft sensor (with Monte Carlo dropout) can predict flow rates more precisely, contributing to the reduction in water losses, as well as cost savings. This approach offers a valuable solution for minimizing water losses and ensuring the efficient use of this vital resource. Regarding the use of soft sensors based on LSTM neural networks with a careful choice of Monte Carlo dropout parameters, when compared to the multilayer perceptron model, the LSTM model with Monte Carlo dropout showed better mean absolute error, root mean square error, and coefficient of determination: 0.2450, 0.3121, and 0.996437 versus 0.2556, 0.3522, and 0.9954. Furthermore, this choice of Monte Carlo dropout parameters allowed us to achieve an LSTM network model capable of reducing uncertainty to 1.8290, keeping the error metrics also at low levels.

Error Analysis and Correction of FENGYUN-4A GIIRS Temperature Profile Products in Summer over the Qinghai–Tibet Plateau

To understand the applicability of the temperature profile product of the FENGYUN-4A (FY-4A) geostationary interferometric infrared detector (GIIRS) in summer over the Qinghai–Tibet Plateau and to improve product quality, the error of GIIRS temperature products was analyzed based on radiosonde data. A long short-term memory network model (LSTM) was used to correct the GIIRS temperature profile product in summer over the plateau at a high altitude (above 500 hPa), and further evaluation of the corrected product was conducted. The results show that summertime GIIRS temperature retrievals over the Qinghai–Tibet Plateau had a positive bias above 150 hPa and a negative bias below 150 hPa, resulting in an overall negative bias. The root mean square error was between 2 and 2.9 K, and the root mean square error was relatively large at 100 hPa and above. The LSTM established in this study could effectively correct the GIIRS temperature over the plateau. The correlation and root mean square error of the corrected GIIRS temperature and the radiosonde observation temperature were significantly improved. Using a trained LSTM correction model to correct the hourly GIIRS temperature can improve the accuracy and usability of the product. After correction, the average bias of the GIIRS temperature compared to the ERA5 temperature product was reduced from −0.26 K to 0.06 K, and the root mean square error was reduced from 2.25 K to 1.25 K. The correction model can be applied to different seasons, and it can also correct the GIIRS temperature in larger areas based on other high-precision and high-resolution data, achieving good results, thus indicating that the correction model has universal applicability.

Short-Term Air Traffic Flow Prediction Based on CEEMD-LSTM of Bayesian Optimization and Differential Processing

With the rapid development of China’s civil aviation, the flow of air traffic in terminal areas is also increasing. Short-term air traffic flow prediction is of great significance for the accurate implementation of air traffic flow management. To enhance the accuracy of short-term air traffic flow prediction, this paper proposes a short-term air traffic flow prediction model based on complementary ensemble empirical mode decomposition (CEEMD) and long short-term memory (LSTM) of the Bayesian optimization algorithm and data differential processing. Initially, the model performs CEEMD on the short-term air traffic flow series. Subsequently, to improve prediction accuracy, the data differencing is employed to stabilize the time series. Finally, the smoothed sequences are, respectively, input into the LSTM network model optimized by the Bayesian optimization algorithm for prediction. After data reconstruction, the final short-term flow prediction result is obtained. The model proposed in this paper is verified by using the data from Shanghai Pudong International Airport. The results show that the evaluation indexes of the prediction accuracy and fitting degree of the model, RMSE (Root Mean Square Error), MAE (Mean Absolute Error), and R2 (Coefficient of Determination), are 0.336, 0.239, and 97.535%, respectively. Compared to other classical time-series prediction models, the prediction accuracy is greatly improved, which can provide a useful reference for short-term air traffic flow prediction.

A novel deep learning rainfall–runoff model based on Transformer combined with base flow separation

ABSTRACT Precise long-term runoff prediction holds crucial significance in water resource management. Although the long short-term memory (LSTM) model is widely adopted for long-term runoff prediction, it encounters challenges such as error accumulation and low computational efficiency. To address these challenges, we utilized a novel method to predict runoff based on a Transformer and the base flow separation approach (BS-Former) in the Ningxia section of the Yellow River Basin. To evaluate the effectiveness of the Transformer model and its responsiveness to the base flow separation technique, we constructed LSTM and artificial neural network (ANN) models as benchmarks for comparison. The results show that Transformer outperforms the other models in terms of predictive performance and that base flow separation significantly improves the performance of the Transformer model. Specifically, the performance of BS-Former in predicting runoff 7 days in advance is comparable to that of the BS-LSTM and BS-ANN models with lead times of 4 and 2 days, respectively. In general, the BS-Former model is a promising tool for long-term runoff prediction.

Forecasting Model of Sustainable Financial Market Trend Based on Neural Network Algorithm

Recent advances in computer technology have resulted in the ongoing gathering of massive amounts of data and information. Because the financial market creates so much real-time data, including transaction records, we have a great potential to get important insights from analysing that data, especially in the banking industry. As a result, the goal of this work is to use the financial data that is now accessible to create a unique stock market prediction model. We employ the deep learning approach because of its exceptional capacity to learn from large datasets. This study proposes a hybrid approach that integrates the Archimedes optimisation algorithm (AOA) with a long short-term memory (LSTM) network. So far, heuristic-based trial and error has been extensively used to estimate the temporal window size and architectural components of long short-term memory networks. This paper investigates the temporal properties of financial market data by providing a systematic way to selecting the topology and time window size for the LSTM network. The experimental results demonstrate that the hybrid LSTM network and AOA model outperform the benchmark model.

Deep learning for predictive window operation modeling in open-plan offices

This study explores how the past, both short and long-term, affects the predictive window operation modeling in open-plan offices. To achieve this, the study proposes a deep learning method that uses long short-term memory (LSTM) artificial neural networks. The proposed model is an ensemble of LSTM networks that utilize relevant indoor environmental data from multiple sensory systems and weather data from a nearby weather station to predict the window state with different predictive horizons. The drawbacks of such networks are extensive training time and the lack of interpretability. Hence, Bayesian optimization is proposed for efficient hyperparameter tuning of the network. Furthermore, explainability through a model agnostic approach is utilized for feature importance ranking.The performance of the LSTM networks was compared to a dense feedforward neural network (DFNN) on an imbalanced dataset to evaluate the effect of the long-term past. Handling imbalanced window operation data was improved by employing class weight and early stopping, leading to a higher true positive rate, marginally better F1-score, and fewer network training epochs. The results indicate that a stateful bi-directional LSTM (Bi-LSTM) network models the data sequence more efficiently in aggregated comparison than other LSTM types and DFNN. However, the DFNN slightly outperformed LSTM networks in interval-by-interval comparison, with an F1-score of 0.88 when predicting the window state 10 min ahead, shifting to 0.82 for a 60-minute predictive horizon. Additionally, the study showed that the ensemble model performed better in terms of F1-score than the individual models.

A novel solar irradiance forecasting method based on multi-physical process of atmosphere optics and LSTM-BP model

Prediction of solar irradiance is crucial for minimizing energy costs and ensuring high power quality in electrical power grids that incorporate distributed solar photovoltaic generation. Traditional methods have focused mainly on time series prediction of irradiance or studies of the relationship between irradiance and environmental factors at a given moment. This paper presents a novel global horizontal irradiance forecasting method based on multi-physical process of atmosphere optics and LSTM-BP model. In the proposed method, cloud fraction, cloud albedo and aerosol optical thickness are selected as critical characteristic factors in the sunlight-atmosphere interaction. Long short-term memory (LSTM) is utilized to extract temporal features of the critical characteristic factors in various physical processes of atmospheric optics and predict their values. Subsequently, back propagation (BP) neural network is constructed to investigate the relationship between these characteristic factors and global horizontal irradiance under a given point in time. The global horizontal irradiance is predicted by combining the predicted values of critical characteristic factors and BP neural network for modelling irradiance decay process. To demonstrate the superior performance of the proposed model, the results obtained by the proposed method are compared with those of the LSTM network model for time series forecasting, in which six error evaluation indicators are used in the comparison. The proposed method for solar irradiance forecasting is verified in different time scales and under various weather conditions and shows better accuracy comparing with time series forecasting methods.

A Comparative Study of LSTM and Temporal Convolutional Network Models for Semisubmersible Platform Wave Runup Prediction

Abstract Wave runup prediction is necessary for offshore structure designs and early warnings. Data-driven methods based on machine learning have inspired reduced-order solutions for wave–structure interaction problems. This study provides the quantification of deep learning algorithms’ potential for wave runup prediction. Two prominent deep learning models were utilized to predict the wave runups along the fore column of semisubmersible under head seas. The long short-term memory (LSTM) and the temporal convolutional networks (TCNs) were comprehensively compared based on the datasets from a model test carried out in the deep ocean basin. The LSTM and TCN model structures were optimized to compare prediction accuracy and computational complexity reasonably. The results reveal that (1) both developed TCN and LSTM models had a satisfied prediction accuracy of over 90%. Their predictions were extended to 10 s into the future with accuracies over 80% and 45%, respectively. (2) With the noise-extended datasets, the LSTM model was robust to noises, while the TCN model showed better prediction performance on the extreme wave runup events. (3) The incident wave and dominant rotation provided the major information for wave runup prediction. TCN and LSTM models’ prediction accuracies were 91.5% and 89.3% based on the simplified input tensors composed of incident wave and pitch. The comparison showed the great potential of the TCN model to predict the nonlinear wave runup with less time and memory costs. The input tensors’ design and optimization based on physical understanding also play a significant role in the prediction performance.

Reduced-order modelling of unsteady cavitating flow around a Clark-Y hydrofoil

This paper proposes a novel approach that combines Proper Orthogonal Decomposition (POD) reduced-order system with Long Short-Term Memory (LSTM) neural network to predict flow velocity. Large Eddy Simulation (LES) is used to simulate the cavitating flow around a NACA66 hydrofoil. POD is adopted to reduce the dimensionality of the high-dimensional data. It was found that 66.81% of the flow field energy and dominant coherent structures can be captured with first eight POD modes. The LSTM network model was further used to predict the temporal data of the POD mode coefficients, and the error of the predicted coefficients was within an acceptable range. The reconstructed flow field agrees well with the real flow field and the cavitation development has also been well illustrated. This method provides a promising and efficient alternative for flow prediction and has potential for applications in fluid dynamics, aerospace engineering, and hydrodynamics.

Residual LSTM-based short duration forecasting of polarization current for effective assessment of transformers insulation

The empirical application of polarization and depolarization current (PDC) measurement of transformers facilitates the extraction of critical insulation-sensitive parameters. This technique, rooted in time-domain dielectric response analysis, forms the bedrock for parameterization and insulation modeling. However, the inherently time-consuming nature of polarization current measurements renders them susceptible to data corruption. This article explores deep-learning-based short-duration techniques for forecasting polarization current to address this limitation. By incorporating spatial shortcuts, the residual long short-term memory (LSTM) network facilitates the seamless propagation of spatial and temporal gradients. Furthermore, the relative forecasting assessment of the proposed residual LSTM model’s performance is made against traditional LSTM, attention LSTM, gated recurrent units (GRU), and convolutional neural network (CNN) models. Thus, optimal model selection strategies are evaluated based on their capability to capture extended dependencies and short-term information present in the data. In addition, the Monte Carlo dropout prediction is employed to estimate uncertainty in polarization current forecasts. The findings demonstrate that the proposed residual LSTM network model for polarization current forecasting yields the lowest error metrics and maintains prediction consistency over the testing duration. Thus, the proposed approach significantly reduces PDC measurement time, providing an effective means to develop proactive maintenance strategies for evaluating the insulation condition of transformers.

Comparative Evaluation of Deep Learning Techniques in Streamflow Monthly Prediction of the Zarrine River Basin

Predicting monthly streamflow is essential for hydrological analysis and water resource management. Recent advancements in deep learning, particularly long short-term memory (LSTM) and recurrent neural networks (RNN), exhibit extraordinary efficacy in streamflow forecasting. This study employs RNN and LSTM to construct data-driven streamflow forecasting models. Sensitivity analysis, utilizing the analysis of variance (ANOVA) method, also is crucial for model refinement and identification of critical variables. This study covers monthly streamflow data from 1979 to 2014, employing five distinct model structures to ascertain the most optimal configuration. Application of the models to the Zarrine River basin in northwest Iran, a major sub-basin of Lake Urmia, demonstrates the superior accuracy of the RNN algorithm over LSTM. At the outlet of the basin, quantitative evaluations demonstrate that the RNN model outperforms the LSTM model across all model structures. The S3 model, characterized by its inclusion of all input variable values and a four-month delay, exhibits notably exceptional performance in this aspect. The accuracy measures applicable in this particular context were RMSE (22.8), R2 (0.84), and NSE (0.8). This study highlights the Zarrine River’s substantial impact on variations in Lake Urmia’s water level. Furthermore, the ANOVA method demonstrates exceptional performance in discerning the relevance of input factors. ANOVA underscores the key role of station streamflow, upstream station streamflow, and maximum temperature in influencing the model’s output. Notably, the RNN model, surpassing LSTM and traditional artificial neural network (ANN) models, excels in accurately mimicking rainfall–runoff processes. This emphasizes the potential of RNN networks to filter redundant information, distinguishing them as valuable tools in monthly streamflow forecasting.

Research on Malware Detection Technology for Mobile Terminals Based on API Call Sequence

With the development of the Internet, the types and quantities of malware have grown rapidly, and how to identify unknown malware is becoming a new challenge. The traditional malware detection method based on fixed features is becoming more and more difficult. In order to improve detection accuracy and efficiency for mobile terminals, this paper proposed a malware detection method for mobile terminals based on application programming interface (API) call sequence, which was characterized by the API call sequence and used a series of feature preprocessing techniques to remove redundant processing of the API call sequence. Finally, the recurrent neural network method (RNN) was used to build the model and perform detection and verification. Furthermore, this paper constructed a malware detection model based on a two-way recurrent neural network and used the two-way long short-term memory network model (LSTM) to train the data set containing 5986 malware samples and 5065 benign software samples to obtain the final detection model and its parameters. Finally, the feature vector of the APK file to be detected was passed into the model and obtained the detection results. The experimental results indicated that the detection accuracy of this method can reach 93.68%.

Stabilization temperature prediction in carbon fiber production using empirical mode decomposition and long short-term memory network

Carbon fiber holds significant promise as a sustainable material with diverse applications. The production of carbon fiber involves a critical process known as oxidative stabilization. Optimizing this process necessitates accurate prediction of the stabilization temperature. However, predicting the stabilization temperature based on easily available heater power data encounters formidable challenges due to factors like nonlinearity, non-uniformity, time delay and time-varying effect. To address these issues, this study proposed an innovative method for predicting the stabilization temperature using empirical mode decomposition (EMD) and long short-term memory (LSTM) network. The distinctive aspect of this proposed method lies in the utilization of two EMD-based reconstruction approaches to preprocess the heater power training and testing data. This helps mitigate the nonlinearity between stabilization temperature and heater power output during the machine learning network training phase and alleviate the time-varying effect of heater power testing data relative to the training data during the network testing phase. Considering the hysteresis effect of heater power output on stabilization temperature, an LSTM network was integrated for accurate stabilization temperature prediction. The proposed EMD-LSTM network model was implemented for stabilization temperature predictions in six oxidation ovens of a carbon fiber production line. The average metrics for predicted results using this method are notably superior, with values of 0.0153 °C for root-mean-square error (RMSE), 0.3711 for relative RMSE (rRMSE), 0.0117 °C for mean absolute error (MAE), and 0.2838 for relative MAE (rMAE), surpassing the results obtained from the LSTM network model trained and tested using unreconstructed data. The comparative analysis further demonstrates the advantageous performance of the proposed EMD-LSTM network model compared to models based on five widely used machine learning networks. This study offers valuable insights for optimizing the oxidative stabilization process in carbon fiber production, thereby reducing waste heat and exhaust emissions, and promoting cleaner carbon fiber production.

PTS-LSTM: Temperature prediction for fused filament fabrication using thermal image time series

In-situ thermography for Fuse Filament Fabrication (FFF) processes reveals the dynamic thermal behavior during printing. The data collected are thermal image time series. Their infrared (IR) intensity is visual evidence of heat-affected zone (HAZ) temperatures, which can be leveraged to train deep learning models, e.g., Long Short-Term Memory (LSTM), for real-time temperature prediction and process monitoring. Nonetheless, the data collection method and printing path may pose challenges for data modeling. Typically, the IR camera has a fixed position while the HAZ moves per the predetermined printing path. Consequently, the HAZ shifts in images and the features extracted from these thermal images show a “periodic” behavior over time. Such periodic patterns do not reflect any useful information about HAZ temperatures. Instead, they are noise hiding and interrupting the true temperature information, thus must be removed before using the data to train an LSTM model for temperature prediction. This study integrates a time series model, i.e., ARIMA, with Stacked LSTM to build a pTS-LSTM model that eliminates noisy patterns and predicts temperatures during FFF printing. The case study results show the outperformance of pTS-LSTM over conventional LSTM and classic Recurrent Neural Network models. pTS-LSTM is demonstrated to be promising for in-situ process monitoring with low-quality thermal images. In FFF practices, pTS-LSTM will be a preferred option over the commonly used deep learning models for thermal-image-based temperature prediction.

Short-term power load comparison based on time series and neural networks considering multiple features

In order to achieve a certain balance between power supply and production demand, as well as to ensure the operation of social machinery and multiple markets, accurate power load forecasting is indispensable. A load prediction method based on the Convolutional Neural Network (CNN) - the Long Short-Term Memory (LSTM) and the CNN-gated Recurrent Unit (GRU) is proposed. The multi-feature load forecasting is constructed by the date factor, the weather factor, the load factor and the electricity price factor. The advantage of CNN in feature extraction of the data set is used to establish a high-dimensional relationship with load, optimize the input LSTM and GRU network model, train each group of the neural network model, and output load prediction value. Combining the load data of a certain area with its meteorological factors, the CNN-LSTM and CNN-GRU methods are tested and compared with the single-network models of Back Propagation (BP), LSTM and GRU. By comparing it with other popular algorithms, it is proved that the presented model has superior computational efficiency and prediction accuracy.

Aircraft parameter estimation using a stacked long short-term memory network and Levenberg-Marquardt method

To effectively estimate the unknown aerodynamic parameters from the aircraft’s flight data, this paper proposes a novel aerodynamic parameter estimation method incorporating a stacked Long Short-Term Memory (LSTM) network model and the Levenberg-Marquardt (LM) method. The stacked LSTM network model was designed to realize the aircraft dynamics modeling by utilizing a frame of nonlinear functional mapping based entirely on the measured input–output data of the aircraft system without requiring explicit postulation of the dynamics. The LM method combines the already-trained LSTM network model to optimize the unknown aerodynamic parameters. The proposed method is applied by using the real flight data, generated by ATTAS aircraft and a bio-inspired morphing Unmanned Aerial Vehicle (UAV). The investigation reveals that for the two different flight data, the designed stacked LSTM network structure can maintain the efficacy of the network prediction capability only by appropriately adjusting the dropout rates of its hidden layers without changing other network parameters (i.e., the initial weights, initial biases, number of hidden cells, time-steps, learning rate, and number of training iterations). Besides, the proposed method’s effectiveness and potential are demonstrated by comparing the estimated results of the ATTAS aircraft or the bio-inspired morphing UAV with the corresponding reference values or wind-tunnel results.

Predicting and explaining karst spring dissolved oxygen using interpretable deep learning approach

AbstractDissolved oxygen (DO) is one of the most important indicators of water quality and an essential measure for the aquatic organisms and the local ecosystem. DO concentrations in karst spring flow involves complex hydrological processes because of the heterogeneous nature of the karst system. In this study, an interpretable and explainable model that integrates the shapley additive explanations (SHAP) algorithm with the long short‐term memory network model (LSTM) is proposed to evaluate the contributions of karst spring discharge, precipitation, water temperature, and specific conductance to DO concentrations in karst spring flow. The hybrid model can predict the temporal fluctuations of DO levels and provide a robust characterization of DO behaviours. To demonstrate the applicability of the proposed model, we adopted the hydrological and meteorological data from Barton Springs, situated within a highly karstified segment of the Edwards aquifer characterized by extensive fractures and openings. The optimal prediction performance is achieved with a 14‐day time step, which is considered as the effective response time between various hydrological processes and DO concentrations at Barton Springs. It reveals that the influence of karst spring discharge, precipitation, water temperature, and specific conductance in previous 14 days collectively contribute to the current DO concentration in karst spring flow. The SHAP values of input features provide both local and global explanations, demonstrating the magnitude and the direction of each feature's impact on DO levels in karst spring flow. In descending order, the contributions of various hydrological processes to DO are ranked as follows: precipitation, karst spring discharge, temperature, and specific conductance. Precipitation and discharge exhibit positive SHAP values, indicating that increases in these hydrological processes contribute to higher DO levels in the karst flow. Water temperature and specific conductance have negative SHAP values, suggesting that higher water temperature and specific conductance will lead to decreased DO levels in the karst flow.