Search

A multitask learning (MTL) model based on artificial neural networks (ANNs) is proposed in this study to improve the prediction accuracy and physical reliability of marine propeller hydrodynamic performance. The propeller's comprehensive geometric features are used as inputs, and the coefficients of quadratic polynomials for the thrust coefficient (KT) and torque coefficient (10KQ) curves are predicted as outputs. The loss function is customized through a positive gradient penalty of the curves to accelerate training. When the single-task and multitask models were compared, the prediction errors were reduced; KT decreased from 2.61% to 2.07%, 10 KQ decreased from 3.58% to 2.31%, and the efficiency (η) decreased from 3.04% to 2.00%. Non-physical fluctuations in the performance curves were effectively mitigated by the multitask model, yielding predicted curvatures which closely matched the experimental data. Strong generalization was demonstrated when the model was tested on unseen propellers, with deviations of 2.2% for KT, 4.6% for 10 KQ, and 3.8% for η. Finally, the model was applied to optimize the propeller design for a 325,000 ton very large ore carrier ship, where a Pareto front with 58 non-dominant solutions for the maximum speed and fluctuating pressure was successfully generated and effectively verified by the model's test results. The model enhanced the prediction of the propeller performance and contributed to optimization in the propeller's design. [ABSTRACT FROM AUTHOR]

A multitask learning (MTL) model based on artificial neural networks (ANNs) is proposed in this study to improve the prediction accuracy and physical reliability of marine propeller hydrodynamic performance. The propeller’s comprehensive geometric features are used as inputs, and the coefficients of quadratic polynomials for the thrust coefficient (KT) and torque coefficient (10KQ) curves are predicted as outputs. The loss function is customized through a positive gradient penalty of the curves to accelerate training. When the single-task and multitask models were compared, the prediction errors were reduced; KT decreased from 2.61% to 2.07%, 10 KQ decreased from 3.58% to 2.31%, and the efficiency (η) decreased from 3.04% to 2.00%. Non-physical fluctuations in the performance curves were effectively mitigated by the multitask model, yielding predicted curvatures which closely matched the experimental data. Strong generalization was demonstrated when the model was tested on unseen propellers, with deviations of 2.2% for KT, 4.6% for 10 KQ, and 3.8% for η. Finally, the model was applied to optimize the propeller design for a 325,000 ton very large ore carrier ship, where a Pareto front with 58 non-dominant solutions for the maximum speed and fluctuating pressure was successfully generated and effectively verified by the model’s test results. The model enhanced the prediction of the propeller performance and contributed to optimization in the propeller’s design.

In the present paper, the prediction of the performance of a centrifugal pump is investigated based on proper orthogonal decomposition (POD) reduced-order model (ROM). Firstly, the POD method is used to analyze the characteristics of the complex flow field inside the centrifugal pump, especially the eigenvalues of the velocity field and pressure field, and then the ROM of the flow field in the centrifugal pump is established then. After that, the established reduced-order model is used to predict the internal flow field. The results show that the POD mode field can reflect the spatial scale distribution of the flow field. The low-order flow field reflects the large-scale flow field structure, while the high-order flow field reflects the small-scale flow field structure. With the established ROM, the overall pump head (0D, 0-dimensional parameter) can be predicted with the relative error of less than 5%, and the standard root mean square error (normalized root mean square error, NRMSE) of pressure field and velocity field (3D, 3-dimensional parameter) is less than 4% and 8%, respectively. The efficiency and accuracy of both 0D and 3D parameters’ prediction motivate the fast prediction of multi-scale simulation of the large-scale industrial system in the future.

In modern smart cities, road networks are becoming more and more complicated, resulting in more complex format of graphs. This brings many challenges to the forecasting of traffic flow in road graphs. Most of traditional traffic flow forecasting methods ignored many implicit relationships inside road graphs. And this cannot be well suitable for modern road networks in smart cities. Besides, the operation of smart cities is accompanied with real-time big data stream. The running efficiency of forecasting methods is another important concern. To handle this issue, this paper proposes a graph deep learning-based fast traffic flow forecasting method in urban road networks. Firstly, the theory about graph convolution operations is deduced and can be used as the basis of a graph convolution network (GCN). Then, the whole road network is viewed as a complex road graph, and the GCN is introduced to establish a novel forecasting method for graph-level traffic flow. With roads being regarded as nodes and their relations being regarded as edges, graph-level forecasting can be realized with the use of the proposed method. Experiments are carried out on a standard real dataset to evaluate the proposal. The experimental results show a proper performance of the proposal.

Rapid prediction of urban flooding is an important measure to reduce the risk of flooding and to protect people's property. In order to meet the needs of emergency flood control, this paper constructs a rapid urban flood prediction model based on a machine learning approach. Using the simulation results of the hydrodynamic model as the data driver, a neural network structure combining convolutional neural network (CNN) and long and short-term memory network (LSTM) is constructed, taking into account rainfall factors, geographical data, and the distribution of the drainage network. The study was carried out with the central city of Zhoukou as an example. The results show that after the training of the hydrodynamic model and CNN−LSTM neural network model, it can quickly predict the depth of urban flooding in less than 10 s, and the average error between the predicted depth of flooding and the measured depth of flooding does not exceed 6.50%, which shows that the prediction performance of the neural network is good and can meet the seeking of urban emergency flood control and effectively reduce the loss of life and property. [ABSTRACT FROM AUTHOR]

Efficient prediction of indoor temperature distribution (ITD) is crucial for various building-related applications, particularly in real-time thermal management. Despite efforts from both physics-based and data-driven perspectives, ITD prediction still faces challenges, particularly in balancing prediction accuracy with computational speed and addressing the need for extensive high-quality data in practical applications. To address these challenges, this study develops a hybrid temperature distribution prediction model (HTDPM) that integrates deep learning with physical mechanisms, enabling the effective capture of spatial-temporal dependencies and uncertainties with limited input parameters, achieving a balance between prediction accuracy and computational efficiency. Meanwhile, a novel transfer learning approach is applied to HTDPM (HTDPM-TL), significantly reducing data requirements and enhancing model generalization across various practical scenarios. Besides, orthogonal and randomized experiments are designed to simulate multiple real-world scenarios, thus constructing an extensive source domain dataset. The HTDPM-TL was tested in various real-world scenarios with several baseline methods, demonstrating an average RMSE of 0.794 ∘C, a computational time of 0.289 s, and limited data volume. The computational speed was improved by three orders of magnitude compared to CFD simulations, and the prediction resolution was enhanced threefold compared to traditional data-driven models. These results highlight the potential of HTDPM-TL to achieve an optimal trade-off between prediction accuracy, computational efficiency, and data requirement. • A hybrid mechanism-based and data-driven model for efficient indoor temperature distribution prediction was developed. • A novel transfer learning approach for reducing data requirements and enhancing generalization was introduced. • The computational speed was improved by three orders of magnitude compared to baseline CFD simulations. • The prediction resolution was enhanced threefold compared to traditional data-driven models. [ABSTRACT FROM AUTHOR]

• Design of the ROR model framework based on partial differential operators. • Extraction of low-dimensional flow field features using an autoencoder. • Combination of flow field and spatial features in low-dimensional space using a cross-fitting algorithm. • Elimination of high-dimensional redundancies and noise, reducing dataset quality requirements. • Fast high-fidelity flow field data acquisition without prior physical knowledge. The real-time prediction of flow fields has scientific and engineering significance, although it is currently challenging. To address this issue, we propose a nonintrusive supervised reduced-order machine learning framework for flow-field reconstruction, referred to as ROR, to achieve real-time flow-field prediction. The model predicts a signed distance function of the domain and uses a typical flow field as feature extraction objects. Utilizing a cross-fit method, it efficiently combines these features, enabling rapid prediction of the full-order flow field. During the model validation phase, we assess the performance of our model by reconstructing steady-state two-dimensional indoor flows in different room layouts. The results indicate that our model accurately predicts the flow field in the target indoor layout within a short timeframe (approximately 5 s) and demonstrates robustness. To delve deeper into the model performance, we discuss the specific parameters of the model framework and test the effectiveness of the flow-field reconstruction under different air supply modes, with the results showing a mean squared error (MSE) of less than 1.5 %. Additionally, we compare our model with the fourier neural operator (FNO) model and find that it exhibited superior performance with the same number of training steps. The outcomes of this study not only bear significant theoretical implications for the field of flow-field prediction but also provide robust support for practical engineering applications. [ABSTRACT FROM AUTHOR]

In recent years, artificial intelligence (AI) technologies have been widely applied in many different fields including in the design, maintenance, and control of aero-engines. The air-cooled turbine vane is one of the most complex components in aero-engine design. Therefore, it is interesting to adopt the existing AI technologies in the design of the cooling passages. Given that the application of AI relies on a large amount of data, the primary task of this paper is to realize massive simulation automation in order to generate data for machine learning. It includes the parameterized three-dimensional (3-D) geometrical modeling, automatic meshing and computational fluid dynamics (CFD) batch automatic simulation of different film cooling structures through customized developments of UG, ICEM and Fluent. It is demonstrated that the trained artificial neural network (ANN) can predict the cooling effectiveness on the external surface of the turbine vane. The results also show that the design of the ANN architecture and the hyper-parameters have an impact on the prediction precision of the trained model. Using this established method, a multi-output model is constructed based on which a simple tool can be developed. It is able to make an instantaneous prediction of the temperature distribution along the vane surface once the arrangements of the film holes are adjusted.

In this work, a model is proposed for the computation of unbalanced magnetic pull (UMP) in PM machines. This model is suitable for use in dynamic simulations. However, the aim of this research is not dynamic modeling of the PM machines. The main idea is to provide a fast model to be integrated into the dynamic models of PM machines. To reduce the required computational burden, the UMP of PM machines is expressed as a second-order algebraic vector function of the stator currents. The function parameters are estimated by post-processing of the results of some magneto-static simulations. These parameters are functions of the machine geometrical data and the rotor position. The role of these parameters in the computation of the machine UMP is the same as the role of the machine inductances in computation of the electromagnetic torque. To investigate the capability of the developed model, the UMP of a concentric fractional-slot surface PM machine with diametrically asymmetric stator windings and the UMP of an eccentric PM-inset machine are predicted. The obtained results are compared in terms of the consumed time and the accuracy by means of finite element analysis. [ABSTRACT FROM AUTHOR]

• A Lagrangian-Markov chain model with coarse grids was proposed for fast prediction. • Detailed procedures and parameter determination approaches were provided. • The proposed model was tens to hundreds of times faster than the existing models. Fast calculation of person-to-person particle transport is essential for accelerating the evaluation and design of air distribution for reducing the risk of infection. This study developed a Lagrangian-Markov chain model with coarse grids for fast prediction of person-to-person particle transport in complex indoor environments. Detailed procedures and parameter determination approaches were developed. The proposed Lagrangian-Markov chain model was first validated with experimental data in two real-life cases of person-to-person particle transport, one in an aircraft cabin and the other in a COVID-19 isolation ward. The computing speed of the proposed model was then compared with the flux-based Markov chain, Eulerian, and Lagrangian models. The results show that the proposed Lagrangian-Markov chain model can predict person-to-person particle transport reasonably well in real-life cases with complex geometry and airflow fields. In terms of computing speed, the proposed Lagrangian-Markov chain model with coarse grids can be tens to hundreds of times faster than the three existing models for the two evaluation cases. With its fast computing speed, the Lagrangian-Markov chain model can be applied in the fast design of air distribution for real-life complex indoor environments. [ABSTRACT FROM AUTHOR]

• Markov chain model combined with CFD for multi-zone space proposed. • Exposure levels and deposition rates of particulate pollutants analyzed. • Proposed model may be of practical use for obtaining real-time predictions. Understanding the distribution and spread of indoor airborne contaminants between rooms in a multi-zone space is crucial for protecting human health and indoor environmental control. Predicting the transmission of suddenly released particulate pollutants from one room to another in a rapid and precise manner is important for reducing or even eliminating the risk of cross contamination between occupants. In this study, we developed a modified Markov chain model that considers the effect of gravity to predict the dispersion and deposition of aerosol particles in a two-zone chamber. To perform particle phase simulations, the proposed model couples the data obtained from a steady-state flow field using the computational fluid dynamics (CFD) software Fluent with some codes that we developed in the MATLAB environment. Two examples based on previously reported experimental data were used to validate the proposed model. The results indicate that the proposed model is suitable for modeling the dynamic processes where airborne particles are released from constant and pulsed contaminant sources in multi-zone spaces with reasonable accuracy and computational efficiency. After analyzing particle exposure in the two-zone chamber, we found that the accumulated exposure level in the zone with the constant contaminant source always had a higher rate of increase than the other zones. However, the accumulated exposure level for the pulsed (instantaneous) source was inversely proportional to the air exchange rate for all zones. Compared with traditional CFD models and multi-zone models, our proposed model can balance the accuracy and efficiency when predicting the spread of particulate contaminants in multi-zone spaces by choosing appropriate grid resolution (s). [ABSTRACT FROM AUTHOR]

This paper proposes a novel fast open circuit voltage prediction approach for Lithium-ion battery, which is potential to facilitate a convenient battery modeling and states estimation in the energy storage system. Open circuit voltage measurement suffers from a long relaxation time (several hours, even days) to reach the thermodynamic equilibrium of the battery. On the basis of the feedback control theory, the proposed multiple correction approach utilizes the constrained nonlinear optimization of the power function in each curve fitting step. The voltage measurement in a short period is divided into several segments to correct the voltage prediction multiple times with the feedback errors after each curve fitting. The similarity between the shape of the power function and the variation of the terminal voltage during the relaxation time is utilized. The proposed method can speed up the time-consuming open circuit voltage measurement and predict the open circuit voltage with high accuracy. Experimental tests on a LiFePO4 battery prove the validation and effectiveness of the proposed method in accurately predicting the open circuit voltage within a very short relaxation time (less than 15 min). [ABSTRACT FROM AUTHOR]

Abstract Thermophoresis is considered as one of the major causes leading to deposition of suspended submicron aerosols with the presence of temperature gradient. Prompt obtaining of information about particle deposition and spatial distribution is crucial to environment control. This study proposed a modified Markov chain model to include the effect of thermophoresis. A steady-state flow field and a constant temperature field are firstly established by using Computational Fluid Dynamics (CFD) tools. Then the data of flow and temperature fields are exported out as data files, which will be used by the proposed model to realize the particle phase simulation. A horizontal steady-state laminar duct flow case is used to validate the model. Results show that the proposed model is able to efficiently predict thermophoresis induced particle deposition and dispersion with reasonable accuracy. Detailed information also indicate that the particle deposition efficiency is dependent on both temperature gradients and particle diameters. However, the beginning of the particle escape (if any) is irrelevant to those two parameters. The highest deposition efficiency on the deposition surface is found to gradually increase over time and the location of highest deposition efficiency (peak) moves along with the main flow. The average speed of the moving peak is inversely proportional to temperature gradient, but the ultimate deposition efficiency is still proportional to temperature gradient. The proposed model is deemed a practical method in predicting aerosol particle deposition/transport as well as contaminant control in non-isothermal environment. Highlights • A modified Markov chain model is proposed to include thermophoresis. • Thermophoresis induced particle deposition and dispersion are efficiently predicted. • Deposition efficiency depends on both temperature gradient and particle diameter. • Location of the highest deposition efficiency on surface is identified. • The proposed method can be used to characterize other thermal process like fouling. [ABSTRACT FROM AUTHOR]

As potential carriers of hazardous pollutants, airborne particles may deposit onto surfaces due to gravitational settling. A modified Markov chain model to predict gravity induced particle dispersion and deposition is proposed in the paper. The gravity force is considered as a dominant weighting factor to adjust the State Transfer Matrix, which represents the probabilities of the change of particle spatial distributions between consecutive time steps within an enclosure. The model performance has been further validated by particle deposition in a ventilation chamber and a horizontal turbulent duct flow in pre-existing literatures. Both the proportion of deposited particles and the dimensionless deposition velocity are adopted to characterize the validation results. Comparisons between our simulated results and the experimental data from literatures show reasonable accuracy. Moreover, it is also found that the dimensionless deposition velocity can be remarkably influenced by particle size and stream-wise velocity in a typical horizontal flow. This study indicates that the proposed model can predict the gravity-dominated airborne particle deposition with reasonable accuracy and acceptable computing time. [ABSTRACT FROM AUTHOR]

Fast prediction of indoor temperature distribution is valuable to recognize the status and assist in rapid decision making and heat source control. Prediction using a superposition theorem based on the assumption of fixed flow field is an alternative method of performing fast prediction. However, little research revealed the methods’ reliability based on fixed flow field, and previous prediction methods primarily focused on the temperature at a steady state. In this paper, an algebraic expression was established to predict the temperature distribution based on the definition of transient accessibility indices for temperature. The prediction accuracy was mainly verified using a numerical method with 14 cases. It was concluded: (1) the proposed expression can perform fast prediction once the transient accessibility indices are prepared in advance; (2) the fixed flow field adopted in the proposed method should be built by a thermal scenario considering the heat source at a certain intensity, rather than a scenario with no heat source. The accuracy is acceptable for positions outside the heat source area; (3) there is no significant effect of the choice of supply air temperature utilized in building the fixed flow field on prediction accuracy. The research on prediction accuracy is helpful for a reasonable application of the proposed method in real projects. [ABSTRACT FROM AUTHOR]

In this study, a novel approach is proposed for fast prediction of dynamic voltage instability occurrence (as a short term phenomenon and/or a long term one) and voltage stability stiffness of the system, against load disturbances. The main contribution of this paper is in introducing a procedure for generating novel features to be applied to a pattern classifier, by which dynamic voltage stability status of a power system can be predicted. The proposed feature generation procedure only needs measured pre‐disturbance variables and disturbance severity provided by phasor measurement units as inputs whereas a set of output variables are derived from an unconstrained power flow program. Since the proposed method does not need any measured post disturbance data, the prediction task can be performed just after the disturbance. Thus, corrective actions can be executed in a short time after the disturbance to inhibit voltage instability. Moreover as no measured post‐disturbance data are needed, the proposed method can also be employed in preventive procedures for voltage stability enhancement and/or decreasing possibility of voltage instability occurrence. Training a decision tree based classifier with the proposed features and testing the method on a modified version of Nordic32 test system, the simulation results have demonstrated that the proposed method effectively predicts the status of dynamic voltage stability in the test system. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, we proposed a fast prediction algorithm for mode decision and motion estimation (ME) based on the similar macro block (FPSM). The so called similar macro block (MB) has the similar motion information on MB level in AVS-P2 codec. FPSM could efficiently eliminate some redundancy operations of mode decision, ME and rate distortion optimization (RDO) for MBs. What''s more, it could be conjunction with any other excellent algorithms on mode decision, ME and RDO to speed up the codec. Experiments show that, for those video sequences with little motion, the encoding time could be saved up to 60% with little loss of PSNR. Although we limit our discussion FPSM in AVS-p2, our method can also be used in other video coding standards. [Copyright &y& Elsevier]

Abstract: In this paper, we describe the principle, technique, and steps of fast predicting the outburst risk before coal un-covering in shaft and cross-cut, including the selection of indexes for the prediction, fast test of gas pressure, coring technique for obtaining an whole coal core, and the determination of outburst parameters. Using this technique, we can finish testing the original gas pressure of coal within 20 h, making it possible to accurately predict the outburst risk before coal un-covering in shaft and cross-cut in 3~5 days. In addition, we can, according to the prediction result, give out the residual pressure and the smallest pre-drainage rate of gas. The comparison of prediction data among more than 10 shafts and cross-cuts shows that the index of initial gas relief expansion energy can more accurately reflect the outburst risk of coal seams. [Copyright &y& Elsevier]

The utilization of buoyancy-driven natural ventilation in atrium buildings during transitional seasons helps create a healthy and comfortable indoor environment by bringing fresh air indoors. Among other factors, the air flow rate is a key parameter determining the ventilation performance of an atrium. In this study, a grey box modeling method is proposed and a prediction model is built for calculating the buoyancy-driven ventilation rate using three openings. This model developed from Bruce's neutral height-based formulation and conservation laws is supported with a theoretical structure and determined with 7 independent variables and 4 integrated parameters. The integrated parameters could be estimated from a set of simulated data and in the results, the error of the semi-empirical predictive equation derived from CFD (computational fluid dynamics) simulated data is controlled within 10%, which indicates that a reliable predictive equation could be established with a rather small dataset. This modeling method has been validated with CFD simulated data, and it can be applied extensively to similar buildings for designing an expected ventilation rate. The simplicity of this grey box modeling should save the evaluation time for new cases and help designers to estimate the ventilation performance and choose building optimal opening designs. [ABSTRACT FROM AUTHOR]