High Prediction Accuracy Research Articles

Hyperspectral Imaging Combined with Machine Learning Can Be Used for Rapid and Non-Destructive Monitoring of Residual Nitrite in Emulsified Pork Sausages

The non-destructive and rapid monitoring system for residual nitrite content in processed meat products is critical for ensuring food safety and regulatory compliance. This study was performed to investigate the application of hyperspectral imaging in combination with machine learning algorithms to predict and monitor residual nitrite concentrations in emulsified pork sausages. The emulsified pork sausage was formulated with 1.5% (w/w) sodium chloride, 0.3% (w/w) sodium tripolyphosphate, 0.5% (w/w) ascorbic acid, and sodium nitrite at concentrations of 0, 30, 60, 90, 120, and 150 mg/kg, based on total sample weight. Hyperspectral imaging measurements were conducted by capturing images of the cross-sections and lateral sides of sausage samples in a linescan mode, covering the spectral range of 1000–2500 nm. The analysis revealed that higher nitrite concentrations could influence the protein matrix and hydrogen-bonding capacities, which might cause increased reflectance at approximately 1080 nm and 1280 nm. Machine learning models, including XGBoost, CATboost, and LightGBM, were employed to analyze the hyperspectral data. XGBoost demonstrated the best performance, achieving an R2 of 0.999 and a root mean squared error of 0.095, highlighting its high predictive accuracy. This integration of hyperspectral imaging with advanced machine learning algorithms offers a non-destructive and real-time method for monitoring residual nitrite content in processed meat products, noticeably improving quality control processes in the meat industry. Additionally, real-time implementation in industrial settings could further streamline quality control and enhance operational efficiency. Further research should focus on validating these findings with larger sample sizes and more diverse datasets to ensure robustness.

Machine Learning‐Aided Discovery of Low‐Pt High Entropy Intermetallic Compounds for Electrochemical Oxygen Reduction Reaction

Advancing the design of cathode catalysts to significantly maximize platinum utilization and augment the longevity has emerged as a formidable challenge in the field of fuel cells. Herein, we rationally design a high entropy intermetallic compound (HEIC, Pt(FeCoNiCu)3) for catalyzing oxygen reduction reaction (ORR) by an efficient machine learning stategy, where crystal graph convolutional neural networks are employed to expedite the multicomponent design. Based on a dataset generated from first‐principles calculations, the model can achieve a high prediction accuracy with mean absolute errors of 0.003 for surface strain and 0.011 eV atom‐1 for formation energy. In addition, we identify two chemical features (atomic size difference and mixing enthalpy) as new descriptors to explore advanced ORR catalysts. The carbon supported Pt(FeCoNiCu)3 catalyst with small particle size is successfully synthesized by a freeze‐drying‐annealing technology, and exhibits ultrahigh mass activity (4.09 A mgPt‐1) and specific activity (7.92 mA cm‐2). Meanwhile, The catalyst also shows significantly enhanced electrochemical stability which can be ascribed to the sluggish difussion effect in the HEIC structure. Beyond offering a promising low‐Pt electrocatalysts for fuel cell cathode, this work offers a new paradigm to rationally design advanced catalysts for energy storage and conversion devices.

Machine Learning-Aided Discovery of Low-Pt High Entropy Intermetallic Compounds for Electrochemical Oxygen Reduction Reaction.

Advancing the design of cathode catalysts to significantly maximize platinum utilization and augment the longevity has emerged as a formidable challenge in the field of fuel cells. Herein, we rationally design a high entropy intermetallic compound (HEIC, Pt(FeCoNiCu)3) for catalyzing oxygen reduction reaction (ORR) by an efficient machine learning stategy, where crystal graph convolutional neural networks are employed to expedite the multicomponent design. Based on a dataset generated from first-principles calculations, the model can achieve a high prediction accuracy with mean absolute errors of 0.003 for surface strain and 0.011 eV atom-1 for formation energy. In addition, we identify two chemical features (atomic size difference and mixing enthalpy) as new descriptors to explore advanced ORR catalysts. The carbon supported Pt(FeCoNiCu)3 catalyst with small particle size is successfully synthesized by a freeze-drying-annealing technology, and exhibits ultrahigh mass activity (4.09 A mgPt-1) and specific activity (7.92 mA cm-2). Meanwhile, The catalyst also shows significantly enhanced electrochemical stability which can be ascribed to the sluggish difussion effect in the HEIC structure. Beyond offering a promising low-Pt electrocatalysts for fuel cell cathode, this work offers a new paradigm to rationally design advanced catalysts for energy storage and conversion devices.

Data-driven insights into the properties of liquisolid systems based on machine learning algorithms

Liquisolid systems (LS) represent a formulation approach where liquid drug or its dispersion is transformed into a powder with good flowability and compactibility, leading to enhanced drug dissolution and bioavailability. Many research groups have focused on the preparation and investigation of LS, leading to a higher need for comprehensive evaluation of factors impacting LS characteristics. The aim of this work was to investigate the applicability of machine learning algorithms in the LS evaluation, using data mined from published literature, and provide an insight into critical factors governing the liquisolid system performance.The dataset was prepared using publication search engines and relevant keywords, with a total of 425 formulations included in the database. The database focused on preparation methods, formulation parameters, and liquisolid system characteristics. Subsequently, critical properties of the liquisolid system, i.e. flowability, compact hardness, and drug dissolution, were analyzed using machine learning algorithms, including Gradient Boosting, Adaptive Boosting and Random Forest.In addition to conventional preparation methods and excipients, novel technologies (fluid bed preparation, extrusion/spheronization) and materials (Neusilin®, Fujicalin®, and Syloid®) enhanced the properties of liquisolid systems. The analysis revealed that formulation factors, such as carrier and coating agent type and content, liquid phase load, model drug type and content, as well as preparation method, significantly influenced liquisolid system characteristics. The models developed exhibited high prediction accuracy when applied on test data (higher than 80 %). This indicates that the machine learning models may provide an insight into the critical attributes affecting the LS performance and may be used as a valuable tool in the development and optimization of these samples.

Degradation mechanism and strength prediction of aeolian sand concrete under sulphate dry-wet cycles

In this paper, aeolian sand (AS) with different contents (0 %, 25 %, 50 %, 75 % and 100 %) was used as fine aggregate to prepare aeolian sand concrete (ASC). The effects of AS content and dry-wet cycle (DWC) times on the durability of ASC under sulfate (5 % Na2SO4) erosion conditions were investigated by DWC tests. The mass loss rate, compressive strength loss rate and relative dynamic elastic modulus (RDEM) were used as the macroscopic performance evaluation indexes, while the damage degradation mechanism of ASC was revealed by combining the microscopic technical means such as SEM, XRD, TG-DSC and NMR. The results show that the macroscopic properties of ASC under the action of DWC of sulphate all show the first increase and then decrease, as when the dosage of AS is below 50 %, the "bead effect" and "padding effect" of AS can further improve the internal densification of ASC, thus effectively resisting external sulphate erosion. During sulphate erosion, SO42- firstly reacts with cement hydration product Ca(OH)2 to generate erosion products such as Ettringite (AFt) and fills them within the original defects, with short-term rnhancement of macroscopic properties. As the DWC continue, the excessive erosion products accumulated, resulting in the destruction of pore structure and extension through the phenomenon, the matrix compactness is reduced, and the macroscopic properties degraded. The grey entropy correlation analysis yielded that the grey entropy correlation grade of bound fluid saturation, harmless pore percentage and less harmful pore percentage with compressive strength were all above 0.8, and the GM(1,4) strength prediction model established on the basis of the above pore structure parameters has a high prediction accuracy, and the relative errors between the predicted strength values and the actual strength values are within 10 %, and the research results can provide certain theoretical support for the study of the durability of ASC in sulfate erosion environment.

Developing a Hybrid ARIMA-XGBOOST Model for Analysing Mobile Money Transaction Data in Kenya

This study formulated a hybrid ARIMA-XGBOOST model using the dataset from the Central Bank of Kenya and the objective was to formulate a Hybrid ARIMA-XGBOOST models to capture the patterns and dynamics in Mobile Money Transactions in Kenya. The study was motivated by the gaps that existed from the previous studies which lacked the ability to model both linear and non-linear trends in the data across time period. Since 2007 when Mpesa was invented, the rate of Mobile Money transactions has rose steadily and this required a complex model that will provide a clear trend. To accurately model the dynamics of Mobile Money Transactions in Kenya, this study presented a complex Hybrid model. The model was formulated by combining Autoregressive Integrated Moving Averages (ARIMA) and Extreme Gradient Boosting (XGBOOST). The ARIMA captures the linear trends of the data while the XGBOOST models the non-linear part and trained to model the ARIMA residuals. The model parameters were evaluated using Mean Absolute Error, Root Mean Square Error among others and confirmed to be accurate and reliable. Based on the findings from the ADF coefficient, the stationary condition was met (P-value=0.01), and therefore we proceeded to develop the ARIMA models. Initial diagnoses included model identification and examination of autocorrelation to determine the ARIMA configurations, whereas the Box-Jenkins test confirmed the models' adequacy (P-value=2.220e-16). Based on Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Percentage Error (MPE), which are all below 1 percent, indicates that the performance of the models had high prediction accuracy.

Prediction of pre-eclampsia with machine learning approaches: Leveraging important information from routinely collected data

BackgroundGlobally, pre-eclampsia (PE) is a leading cause of maternal and perinatal morbidity and mortality. PE prediction using routinely collected data has the advantage of being widely applicable, particularly in low-resource settings. Early intervention for high-risk women might reduce PE incidence and related complications. We aimed to replicate our machine learning (ML) published work predicting another maternal condition (gestational diabetes) to (1) predict PE using routine health data, (2) identify the optimal ML model, and (3) compare it with logistic regression approach. MethodsData were from a large health service network with 48,250 singleton pregnancies between January 2016 and June 2021. Supervised ML models were employed. Maternal clinical and medical characteristics were the feature variables (predictors), and a 70/30 data split was used for training and testing the model. Predictive performance was assessed using area under the curve (AUC) and calibration plots. Shapley value analysis assessed the contribution of feature variables. ResultsThe random forest approach provided excellent discrimination with an AUC of 0.84 (95% CI: 0.82–0.86) and highest prediction accuracy (0.79); however, the calibration curve (slope of 1.21, 95% CI 1.13–1.30) was acceptable only for a threshold of 0.3 or less. The next best approach was extreme gradient boosting, which provided an AUC of 0.77 (95% CI: 0.76–0.79) and well-calibrated (slope of 0.93, 95% CI 0.85–1.01). Logistic regression provided good discrimination performance with an AUC of 0.75 (95% CI: 0.74–0.76) and perfect calibration. Nulliparous, pre-pregnancy body mass index, previous pregnancy with prior PE, maternal age, family history of hypertension, and pre-existing hypertension and diabetes were the top-ranked features in Shapley value analysis. ConclusionTwo ML models created the highest-performing prediction using routinely collected data to identify women at high risk of PE, with acceptable discrimination. However, to confirm this result and also examine model generalisability, external validation studies are needed in other settings, utilising standardised prognostic factors.

Development of an improved mass transfer model for condensation with dynamic feedback regulation

In the field of condensation research, accurate and efficient numerical simulation models are highly desired. One of the widely used phase change mass transfer models is known as the Lee model. However, this model has limitations due to uncertainties in the mass transfer intensity factor, which can lead to simulation failures and impede large-scale engineering applications. This study presents an improved mass transfer model which employs a tuning function to dynamically and accurately modify the mass transfer intensity factor in each two-phase cell via feedback adjustment. This approach enables immediate and customized adjustment of the factor for each grid within the limit of the tuning function, significantly reducing the errors caused by repeated artificial adjustments and related uncertainties. The adjustment capabilities of the hyperbolic tangent and softsign functions are evaluated. The softsign function is superior in both regulation efficiency and accuracy. The model's accuracy is verified by comparing with previous experiments and simulations, as well as by the one-dimensional Stefan problem. The proposed model achieves high prediction accuracy, reduces the risk of simulation dispersion, and significantly enhances computational efficiency with the iteration number reduced by nearly half under certain conditions compared to the Lee model. Furthermore, the new model shows robustness as the prediction accuracy is insensitive to variations in critical parameters. The improved model is expected to provide valuable assistance for future research and applications in flow condensation with outstanding computational accuracy, efficiency, and stability.

A cGAN-based fatigue life prediction of 316 austenitic stainless steel in high-temperature and high-pressure water environments

The thermo-mechanical-chemical coupling effect presents significant challenges in accurately predicting the fatigue life of 316 austenitic stainless steel in high-temperature and high-pressure water environments (referred to hereafter as environmental fatigue). The complexity of environmental fatigue experiments results in limited and dispersed data, further making the life prediction difficult. Traditional fatigue life prediction models are often constrained by specific loading conditions and do not adequately account for the complex environmental influences. To address these issues, this paper proposes a novel environmental fatigue life prediction model of 316 stainless steel utilizing conditional Generative Adversarial Networks. The proposed model incorporates critical environmental factors, loading conditions and stacking fault energy, allowing direct prediction of environmental fatigue life. A comparative analysis on the predicted and experimental results reveals that the cGAN-based model significantly improves the prediction accuracy, reducing the fatigue life prediction error from a factor of 5 to within 3. To quantify the uncertainty in fatigue life prediction, the Monte Carlo Dropout method is employed to enable a probabilistic assessment of fatigue life. Furthermore, four environmental and loading conditions are established to evaluate the model’s extrapolation capability. The results demonstrate that the probabilistic fatigue assessment effectively captures data distribution and achieves high prediction accuracy.

A neural network model for predicting the temperature of insulated overhead lines

In this work, we develop a neural network model based on experimental heating and cooling curves of insulated wires of overhead lines under changes in wind speed and its direction relative to the wire axis. Insulated SIP-3 wire used in overhead lines was investigated. A multilayer perceptron neural network was employed to model the process of heating and cooling of insulated wire under changes in wind speed and its direction. The average absolute error was taken as a criterion for evaluating the prediction accuracy of wire core and insulation temperatures. The main parameters of the developed neural network model include the number of hidden layers and neurons in each hidden layer, the degree of regulation, and the regulatory rigidity of the model. The modelled heating and cooling curves of insulated wire were compared with those obtained experimentally. The average absolute error was equal to 1.74 and –4.08°C for the predicted core and insulation temperatures, respectively. The difference between the heating curves at low wind speeds was found to range within 9°C. It was shown that an increase in wind speed leads to a decrease in the difference between the curves. Our analysis showed that neural network models used for predicting variations in the temperature of insulated overhead lines should be trained using a larger number of input parameters. This is the main prerequisite for high prediction accuracy of such models, when the difference between the simulated and experimental data does not exceed 5%.

An efficient predicting approach for blade frequency line-spectrum noise of marine centrifugal pumps

The paper introduces an innovative and efficient calculation method for the spectral flow noise of the pump system based on the spatial distribution dipole, which predicts the blade frequency radiation noise at the pipe flange of the pump system under different operating conditions. First, the unsteady flow simulation of the centrifugal pump is calculated using computational fluid dynamics (CFD). Then, the pressure on each segmented blade surface is decomposed into lift and drag forces. After that, each segment is decomposed into several dipole point sources based on the blade noise theory. Finally, the simplified dipole point sources are considered sound sources, and the blade frequency line spectrum noise at the pipe flange of the pump system is predicted using acoustic-structure coupling. The results indicate that the simplified method provides high prediction accuracy while significantly improving the computational efficiency of blade frequency line spectrum noise calculation for centrifugal pumps.

Prediction and analysis of interior sound quality of high-speed trains based on XGBoost

This paper primarily investigates the objective quantification of subjective perceptions of sound quality in high-speed trains. The study measures and analyzes the sound quality in high-speed trains under high-speed operating conditions and calculates psychoacoustic parameters. A subjective evaluation of the interior sound quality is conducted using the grading scale method. An Extreme Gradient Boosting (XGBoost) algorithm is utilized to construct a predictive model for sound quality in high-speed trains and get high predictive accuracy for sound quality evaluation. Employing AI-based evaluation models to evaluate the sound quality in high-speed trains offers a more effective method for studying the interior sound quality of high-speed trains.

Predicting the properties of metamaterials consisting of curved-wall triangles using ensemble neural networks with interpretability

Machine learning has emerged as a promising tool for predicting the properties of metamaterials, owing to its substantially faster prediction speed compared to conventional methods. However, the lack of interpretability of black-box machine learning models has impeded their adoption in this field. In this work, we develop an ensemble of neural networks (NNE) for predicting the mechanical properties of metamaterials consisting of curved-wall triangles. Our method achieves high prediction accuracy for the effective Young's modulus and Poisson's ratio for most samples. The average relative errors of the NNE are less than 5% for predicting the dimensionless Young's modulus and less than 10% for predicting Poisson's ratios. Additionally, our method is orders of magnitude faster than finite element simulations. More importantly, by explaining the predictions of NNE using SHapley Additive exPlanations (SHAP), we are able to quantify the relationships between geometric features and metamaterial properties. This reveals insights into parameter couplings and nonlinear effects. New physical insights can thus be obtained and new design rules discovered through this machine learning interpretability approach. Overall, our interpretable machine learning model can accelerate the design and discovery of metamaterials while providing physical insights into the inner workings of the models.

An on-demand collaborative edge caching strategy for edge–fog–cloud environment

In this paper, we tackle the critical challenges of content edge caching, such as limited storage capacity, content popularity prediction, dynamic user demand, and user privacy, issues that most existing studies only address partially. We present an innovative Genetic Algorithm-based On-demand Collaborative Edge Caching mechanism (GAOCEC), which introduces a multi-tiered caching architecture integrating cloud, fog, and edge computing. To enhance caching efficiency and minimize system cost, a novel on-demand caching quota mechanism is proposed that dynamically allocates cache resources to edge servers. To strengthen user privacy protection during content popularity prediction, a CNN-BiLSTM-based Federated Learning algorithm (CBFL) is presented that ensures high prediction accuracy without the need to upload local data to the cloud. We also refine the genetic algorithm for content placement by fine-tuning various parameter sets to identify the optimal balance between latency reduction and caching cost. Our experimental results validate the effectiveness of our approach, demonstrating increased cache hit rates, decreased content response times, and an overall improvement in system efficiency. This work provides a comprehensive, adaptive, and privacy-preserving solution for the edge–fog–cloud environment.

Prediction of the Strength of the Concrete-Filled Tubular Steel Columns Using the Artificial Intelligence

Introduction. The machine learning algorithms are highly promising for predicting the load-bearing capacity of the building structures. The paper aims at building the predictive models for calculating the strength of the concrete-filled steel tubular (CFST) columns to enable a highly accurate prediction of the ultimate loads for the entire possible range of parameters affecting the load-bearing capacity of the eccentrically compressed columns.Materials and Methods. The article studies the eccentrically compressed short concrete-filled steel tubular (CFST) columns of circular cross-section. Model input parameters: column outer diameter, pipe wall thickness, yield strength of steel, compressive strength of concrete, relative eccentricity. Output parameters: the ultimate loads without taking into account and taking into account the random eccentricities. The models were trained on synthetic data generated based on the theoretical principles of the limit equilibrium method. Two machine learning models were built. When training the first model, the ultimate loads were determined at a given eccentricity of the longitudinal force without taking into account the additional random eccentricity. When training the second model, the additional random eccentricity was taken into account. The effect of the features on the model predictions was assessed using the Feature Importance function. The Optuna method was used to select the hyperparameters. The machine learning models were implemented in the Jupiter Notebook environment using the Gradient Boosting learning method. The total volume of the training sample was 179 025 samples.Results. The importance of the features most affecting the predictive values of the model have been determined. For both models, the outer diameter of the column and the relative eccentricity have proved to be the most important features, which is consistent with the existing experience of designing and calculating such structures. Optimisation of the hyperparameters using the Grid Search method enabled getting the improved results. The high accuracy of prediction has been ascertained by the low values of the regression metrics: MSE = 9.024; MAE = 9.250; MAPE = 0.004 — for the model built without taking into account the additional random eccentricity; MSE = 8.673; MAE = 8.673; MAPE = 0.004 — for the model built taking into account the additional random eccentricity.Discussion and Conclusion. The developed Gradient Boosting models for predicting the ultimate loads of the eccentrically compressed short concrete-filled steel tubular (CFST) columns of circular cross-section, both without taking into account and taking into account the additional random eccentricities, have demonstrated high accuracy and stability of prediction, they can be applied for assessing the strength of the columns during design and construction, which will reduce the time and resources involved in physical testing. In the future, it is planned to expand the data range by including other materials, different cross-section geometries of the columns and a slenderness parameter, which may improve the generalization ability of the model.

A defect classification algorithm for gas tungsten arc welding process based on unsupervised learning and few-shot learning strategy

Welding defect prediction is the foundation for ensuring welding quality in gas tungsten arc welding (GTAW). In the prediction process, method based on molten pool vision is the most effective. Since the classification of molten pool defects relies on a substantial volume of labeled data, it is challenging for the models to be applied industrially. This paper presents an algorithm, FS-Classifier, that can achieve high prediction accuracy based on a limited amount of labeled data. The FS-Classifier comprises two stages: Firstly, an unsupervised training approach named RaP is designed to pre-train the feature extractor using extensive unlabeled daily datasets. The RaP consists of a rotation angle prediction task and a position prediction task, which ensure that the network focuses on salient features and precise elements, respectively. Secondly, the support vectors constructed from limited labeled data are used for the feature classifier. The input data is classified to certain class by computing its distances to support vector. The model achieves an accuracy of 94.5 % on the private dataset and 92.8 % on the public dataset for the six classes of defects using 5 % of labeled data volume. In addition, comparative experiments show that our method only requires 5 % of labeled data to achieve accuracy comparable to traditional supervised learning methods. The proposed algorithm addresses the issue of relying on a substantial amount of labeled data in welding process defect classification.

Research on wind turbine icing prediction data processing and accuracy of machine learning algorithm

Studying the icing problem of wind turbine blades is crucial for optimizing wind farm operation and maintenance. Traditional machine learning algorithms like Random Forest and Support Vector Machines have limitations in handling complex time series data and capturing long-term dependencies, leading to insufficient accuracy and generalization. To address these gaps, this paper proposes a comprehensive approach involving advanced machine learning frameworks and feature engineering techniques. This paper based on the original SACDA dataset, four distinct types of processing are performed on the prediction data via PCA dimensionality reduction and the introduction of new features; meanwhile, the four types mentioned above of datasets are trained and predicted using the Gated Recycling Unit model (GRU), Random Forest (RF), GA-BP Neural Network (BP), and Extreme Learning Machine (ELM) models. The results indicate that the prediction accuracies of all four types of models are more satisfactory, with the RF model having the highest prediction accuracy and the overall accuracy remaining above 99 percent, the dataset + RF model after dimensionality reduction of the original sensitive features having the highest prediction accuracy and speed.

Development and Validation of an Early Recurrence Predictive Model for Intrahepatic Cholangiocarcinoma Based on a Systematic Review and Meta-Analysis of 17 Cohorts

BackgroundIntrahepatic cholangiocarcinoma (ICC) is a rare malignant tumor with a high incidence of recurrence after surgery. This study aims to identify the early recurrence (ER) risk factors in ICC and develop an ER predictive model for ICC. MethodA computerized search was performed in Embase, Web of Science, Cochrane Library, and PubMed to identify studies focusing on the risk factors associated with ER in ICC. Based on the results of the meta-analysis, an ER predictive model was developed and validated using an external cohort of patients diagnosed with ICC through postoperative pathological examinations who received curative-intent surgery at West China Hospital of Sichuan University. ResultsA total of 4,534 ICC patients from 17 studies were included in the meta-analysis. The results indicated that gender, preoperative serum level of CA199, CEA, tumor size, tumor number, differentiation, macrovascular invasion, microvascular invasion, perineural invasion, lymph node metastasis, and AJCC T stage were ER risk factors for ICC. All risk factors were scored according to their weightings, and a risk predictive model was established. The area under the curve of the validation cohort was 0.902, indicating high predictive accuracy. Additionally, the calibration curve demonstrated that the predicted ER rate aligned well with the actual ER rate in the validation cohort. ConclusionThe predictive model accurately assesses the risk of ER in ICC patients undergoing surgical resection, providing valuable prognostic insights and guiding the tailoring of individualized treatment strategies to improve patient outcomes.

The physical information LSTM surrogate model for establishing a digital twin model of reciprocating air compressors

Reciprocating air compressors play a crucial role in industrial production processes. However, due to the complex structure and long operating time of reciprocating air compressors, real-time monitoring to grasp the operating status of reciprocating air compressors has become particularly important. Digital twin is a technology that can reflect the behavior of physical entities in real time, accurately predicting and evaluating the operation status of reciprocating air compressors. However, the establishment of a digital twin model for reciprocating air compressors requires a significant amount of computational resources, which can result in the inability to meet the requirements of real-time performance evaluation. To overcome this limitation, this paper proposes a method for constructing a digital twin model of reciprocating air compressors based on a surrogate model. The surrogate model is constructed based on a long short-term memory neural network with physical information(PILSTM). This model can accurately describe the changes in cylinder pressure by combining physical information. According to the characteristics of cylinder pressure changes, regularization formulas are added to ensure the smoothness of the predicted pressure. The experimental results show that the digital twin model based on the surrogate model has high prediction accuracy and real-time performance. Therefore, this model provides a new method for monitoring the operating status of reciprocating air compressors.

Dual-stream transformer-attention fusion network for short-term carbon price prediction

Accurate prediction of carbon price provides important references for relevant companies, investors, and policymakers. However, the nonlinear, non-stationary, and random of carbon price time series pose the current prediction models a challenging task for carbon price prediction. Considering deep learning, especially for Transformer, has got a promising space in time series prediction. Therefore, this study develops a dual-stream Transformer-attention fusion network (DTF-Net), which contains three modules: multi-scale extraction, dual-stream Transformer, and attention fusion module. Firstly, the external variables of atmospheric pollution and the target variable of carbon price are taken as inputs, and the multi-scale extraction module is constructed via multiple one-dimensional convolutions with kernels to mine features on different time scales, enhancing feature engineering. Then, inspired by the idea of “divide and conquer”, the dual-stream Transformer module is applied to independently capture multivariate internal relationships and univariate temporal dependencies, improving feature learning. Finally, the attention fusion module designs the attention mechanism to generate real-time weights and dynamically integrate the above features, highlighting core features. In summary, the proposed DTF-Net network has not only relatively high prediction accuracy but also refined designs and clear multi-layer functions. Five experiments under two carbon price datasets from Hubei and Beijing carbon markets in China show the average improvements of mean absolute percentage error (MAPE) are 63.70 % and 64.55 %, 54.51 % and 53.03 %, and 57.04 % and 52.24 % for recursive, parallel and hybrid methods, respectively. The proposed DTF-Net outperforms eighteen benchmark models and is an addition to predicting carbon prices.