Publications
103 results found
Wu X, Gan P, Li J, et al., 2023, A long short-term memory neural network-based error estimator for three-dimensional dynamically adaptive mesh generation, Physics of Fluids, Vol: 35, ISSN: 1070-6631
Adaptive meshes are pivotal in numerical modeling and simulation, offering a means to efficiently, precisely, and flexibly represent intricate physical phenomena, particularly when grappling with their intricacies and varying scales. However, the transition from two dimensions (2D) to three dimensions (3D) poses a substantial challenge, as the computational demands of dynamically adaptive mesh techniques increase exponentially. Addressing this challenge effectively, we turn to the cutting-edge realm of artificial intelligence and neural networks. In our study, we harness the innovative power of a long short-term memory (LSTM) neural network as an error estimator for adapting unstructured meshes in both 2D and 3D scenarios. This LSTM network predicts the evolution of the adaptive grid based on specified variables, presenting itself as an artificial intelligence-driven architecture to optimize the adaptive criterion for the target variable. This is achieved by establishing a direct correspondence between the Riemann metric and these variables. To demonstrate the practical applicability of our approach, we seamlessly integrate the LSTM error estimator into the 3D adaptive atmospheric model Fluidity-Atmosphere (Fluidity-Atmos), thereby enabling real-time mesh adaptation during numerical simulations. We assess the effectiveness of this method in terms of simulation precision and computational efficiency through a series of experiments in both 2D and 3D settings. Our results not only reveal that the mesh patterns generated by the LSTM error estimator within Fluidity-Atmos closely resemble those produced by traditional error estimators but also underscore its superior performance in enhancing simulation accuracy. Notably, as the number of nodes increases, the LSTM mesh generator substantially reduces CPU time requirements by up to 50% in 3D cases compared to the conventional mesh generator within Fluidity-Atmos, highlighting its remarkable computational efficiency.
Fu R, Xiao D, Navon IM, et al., 2023, A non‐linear non‐intrusive reduced order model of fluid flow by auto‐encoder and self‐attention deep learning methods, International Journal for Numerical Methods in Engineering, Vol: 124, Pages: 3087-3111, ISSN: 0029-5981
This paper presents a new nonlinear non-intrusive reduced-order model (NL-NIROM) that outperforms traditional proper orthogonal decomposition (POD)-based reduced order model (ROM). This improvement is achieved through the use of auto-encoder (AE) and self-attention based deep learning methods. The novelty of this work is that it uses stacked auto-encoder (SAE) network to project the original high-dimensional dynamical systems onto a low dimensional nonlinear subspace and predict fluid dynamics using an self-attention based deep learning method. This paper introduces a new model reduction neural network architecture for fluid flow problem, as well as, a linear non-intrusive reduced order model (L-NIROM) based on POD and self-attention mechanism. In the NL-NIROM, the SAE network compresses high-dimensional physical information into several much smaller sized representations in a reduced latent space. These representations are expressed by a number of codes in the middle layer of SAE neural network. Then, those codes at different time levels are trained to construct a set of hyper-surfaces using self-attention based deep learning methods. The inputs of the self-attention based network are previous time levels' codes and the outputs of the network are current time levels' codes. The codes at current time level are then projected back to the original full space by the decoder layers in the SAE network. The capability of the new model, NL-NIROM, is demonstrated through two test cases: flow past a cylinder, and a lock exchange. The results show that the NL-NIROM is more accurate than the popular model reduction method namely POD based L-NIROM.
Cheng M, Fang F, Navon IM, et al., 2023, Assessing uncertainty and heterogeneity in machine learning-based spatiotemporal ozone prediction in Beijing-Tianjin- Hebei region in China, Science of the Total Environment, Vol: 881, Pages: 1-13, ISSN: 0048-9697
Accurate prediction of spatiotemporal ozone concentration is of great significance to effectively establish advanced early warning systems and regulate air pollution control. However, the comprehensive assessment of uncertainty and heterogeneity in spatiotemporal ozone prediction remains unknown. Here, we systematically analyze the hourly and daily spatiotemporal predictive performances using convolutional long short term memory (ConvLSTM) and deep convolutional generative adversarial network (DCGAN) models over the Beijing-Tianjin-Hebei region in China from 2013 to 2018. In extensive scenarios, our results show that the machine learning-based (ML-based) models achieve better spatiotemporal ozone concentration prediction performance with multiple meteorological conditions. A further comparison to the air pollution model-Nested Air Quality Prediction Modelling System (NAQPMS) and monitoring observations, the ConvLSTM model demonstrates the practical feasibility of identifying high ozone concentration distribution and capturing spatiotemporal ozone variation patterns at a high spatial resolution (here 15 km × 15 km).
Xu T, Yao R, Du C, et al., 2023, A quantitative evaluation model of outdoor dynamic thermal comfort and adaptation: a year-long longitudinal field study, Building and Environment, Vol: 237, Pages: 1-14, ISSN: 0360-1323
The understanding of human outdoor thermal comfort demand and thermal adaptation contributes to sustainable urban design as well as city resilience in the context of human health and wellbeing. Humans' past thermal experience influence their outdoor thermal comfort. However, the quantitative relationship between the past thermal experience and outdoor thermal comfort is still not clear. This study aims to reveal quantitative relations of the impact of people's past thermal experience on adaptive thermal comfort and to develop a new outdoor adaptive thermal comfort model. A year-long longitudinal questionnaire survey along with a combination of outdoor thermal environment campaigns was carried out in Chongqing, China. It began on August 15, 2020, and finished on August 19, 2021. Through the analysis of 2240 valid responses to the questionnaire survey, the outdoor thermal adaptation characteristic and dynamic thermal comfort evaluation of the respondents were revealed. The results show that the quantified temperature of past outdoor thermal experience is the quadratic correlation with the thermal sensitivity coefficient and deviation constant, and the linear correlated with outdoor thermal demands. Based on the quantitative analysis, a new outdoor adaptive thermal comfort model has been developed as a function of the exponentially weighted sum of historical mean air temperature series (MeanTrm). The outdoor adaptive thermal comfort zones by 80% and 90% satisfactions thereby have been first drawn based on the Universal Thermal Climate Index (UTCI). The study developed a methodology for the evaluation of dynamic outdoor thermal comfort which can be used for different climate regions.
Abubakkar-Waziri H, Kalaiarasan G, Wawman R, et al., 2023, SARS-CoV2 in public spaces in West London UK during COVID-19 pandemic, BMJ Open Respiratory Research, Vol: 10, ISSN: 2052-4439
Background: Spread of SARS-CoV2 by aerosol is considered an important mode of transmission over distances >2 m, particularly indoors.Objectives: We determined whether SARS-CoV2 could be detected in the air of enclosed/semi-enclosed public spaces.Methods and analysis: Between March 2021 and December 2021 during the easing of COVID-19 pandemic restrictions after a period of lockdown, we used total suspended and size-segregated particulate matter (PM) samplers for the detection of SARS-CoV2 in hospitals wards and waiting areas, on public transport, in a university campus and in a primary school in West London.Results: We collected 207 samples, of which 20 (9.7%) were positive for SARS-CoV2 using quantitative PCR. Positive samples were collected from hospital patient waiting areas, from hospital wards treating patients with COVID-19 using stationary samplers and from train carriages in London underground using personal samplers. Mean virus concentrations varied between 429 500 copies/m3 in the hospital emergency waiting area and the more frequent 164 000 copies/m3 found in other areas. There were more frequent positive samples from PM samplers in the PM2.5 fractions compared with PM10 and PM1. Culture on Vero cells of all collected samples gave negative results.Conclusion: During a period of partial opening during the COVID-19 pandemic in London, we detected SARS-CoV2 RNA in the air of hospital waiting areas and wards and of London Underground train carriage. More research is needed to determine the transmission potential of SARS-CoV2 detected in the air.
Cheng M, Fang F, Navon IM, et al., 2023, Ensemble Kalman filter for GAN-ConvLSTM based long lead-time forecasting, Journal of Computational Science, Vol: 69, Pages: 1-16, ISSN: 1877-7503
Data-driven machine learning techniques have been increasingly utilized for accelerating nonlinear dynamic system prediction. However, machine learning-based models for long lead-time forecasts remain a significant challenge due to the accumulation of uncertainty along the time dimension in online deployment. To tackle this issue, the ensemble Kalman filter (EnKF) has been introduced to machine learning-based long-term forecast models to reduce the uncertainty of long lead-time forecasts of chaotic dynamic systems. Both the deep convolutional generative adversarial network (DCGAN) and convolutional long short term memory (ConvLSTM) are used for learning the complex nonlinear relationships between the past and future states of dynamic systems. Using an iterative Multi-Input Multi-Output (MIMO) algorithm, the two-hybrid forecast models (DCGAN-EnKF and ConvLSTM-EnKF) are able to yield long lead-time forecasts of dynamic states. The performance of the hybrid models has been demonstrated by one-level and two-level Lorenz 96 models. Our results show that the use of EnKF in ConvLSTM and DCGAN models successfully corrects online model errors and significantly improves the real-time forecasting of dynamic systems for a long lead-time.
Arcucci R, Xiao D, Fang F, et al., 2023, A reduced order with data assimilation model: Theory and practice, Computers and Fluids, Vol: 257, Pages: 1-12, ISSN: 0045-7930
Numerical simulations are extensively used as a predictive tool to better understand complex air flows and pollution transport on the scale of individual buildings, city blocks and entire cities. Fast-running Non-Intrusive Reduced Order Model (NIROM) for predicting the turbulent air flows has been proved to be an efficient method to provide numerical forecasting results. However, due to the reduced space on which the model operates, the solution includes uncertainties. Additionally, any computational methodology contributes to uncertainty due to finite precision and the consequent accumulation and amplification of round-off errors. Taking into account these uncertainties is essential for the acceptance of any numerical simulation. In this paper we combine the NIROM method with Data Assimilation (DA), the main question is how to incorporate data (e.g. from physical measurements) in models in a suitable way, in order to improve model predictions and quantify prediction uncertainty. Here, the focus is on the prediction of nonlinear dynamical systems (the classical application example being weather forecasting). DA is an uncertainty quantification technique used to incorporate observed data into a prediction model in order to improve numerical forecasted results. The Reduced Order Data Assimilation (RODA) model we propose in this paper achieves both efficiency and accuracy including Variational DA into NIROM. The model we present is applied to the pollutant dispersion within an urban environment.
Cai S, Fang F, Wang Y, 2023, Nonstationary seismic-well tying with time-varying wavelets, Geophysics, Vol: 88, Pages: M145-M155, ISSN: 0016-8033
Seismic-well tying is an important technique for correlating well-logging curves in depth with seismic traces in time. An appropriate seismic-well tying technique must account for two types of nonstationarity: the nonstationary time errors in the synthetic seismic trace caused by the inaccurate time-depth relationship established based on sonic-logging velocity and the nonstationary seismic signals due to the time-varying wavelets during wave propagation. The nonstationary problems related to the time-depth relationship and the time-varying wavelets are interrelated in seismic-well tying procedure. We implemented a nonstationary seismic-well tying method by iteratively updating the time-depth relationship and estimating the time-varying wavelets. From the estimated time-varying wavelets, we also estimated a Q-value by assuming that the subsurface medium has a constant Q at depth and used the constant Q to constrain the variation of the seismic wavelet during propagation. Then, we used the improved time-depth relationship and time-varying wavelets with the Q constraint for further iterations. In the iterative procedure, we quantified the accuracy of the seismic-well tying result using the correlation coefficient between the synthetic and the true seismic trace in each iteration and evaluated the reliability using the normalized mean-square errors among the wavelets estimated in different iterations.
Li J, Li Y, Steppeler J, et al., 2023, Challenges and prospects for numerical techniques in atmospheric modeling, Bulletin of the American Meteorological Society, Vol: 104, Pages: 449-455, ISSN: 0003-0007
Wu X, Abubakar-Waziri H, Fang F, et al., 2023, Modeling for understanding of coronavirus disease-2019 (COVID-19) spread and design of an isolation room in a hospital, Physics of Fluids, Vol: 35, Pages: 1-9, ISSN: 1070-6631
We have modeled the transmission of coronavirus 2019 in the isolation room of a patient suffering from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) at the Royal Brompton Hospital in London. An adaptive mesh computational fluid dynamics model was used for simulation of three-dimensional spatial distribution of SARS-CoV-2 in the room. The modeling set-up is based on data collected in the room during the patient stay. Many numerical experiments have been carried out to provide an optimal design layout of the overall isolation room. Our focus has been on (1) the location of the air extractor and filtration rates, (2) the bed location of the patient, and (3) consideration of the health and safety of the staff working in the area.
Kumar P, Kalaiarasan G, Bhagat RK, et al., 2022, Active air monitoring for understanding the ventilation and infection risks of SARS-CoV-2 transmission in public indoor spaces, Atmosphere, Vol: 13, Pages: 1-24, ISSN: 2073-4433
Indoor, airborne, transmission of SARS-CoV-2 is a key infection route. We monitored fourteen different indoor spaces in order to assess the risk of SARS-CoV-2 transmission. PM2.5 and CO2 concentrations were simultaneously monitored in order to understand aerosol exposure and ventilation conditions. Average PM2.5 concentrations were highest in the underground station (261 ± 62.8 μgm−3), followed by outpatient and emergency rooms in hospitals located near major arterial roads (38.6 ± 20.4 μgm−3), the respiratory wards, medical day units and intensive care units recorded concentrations in the range of 5.9 to 1.1 μgm−3. Mean CO2 levels across all sites did not exceed 1000 ppm, the respiratory ward (788 ± 61 ppm) and the pub (bar) (744 ± 136 ppm) due to high occupancy. The estimated air change rates implied that there is sufficient ventilation in these spaces to manage increased levels of occupancy. The infection probability in the medical day unit of hospital 3, was 1.6-times and 2.2-times higher than the emergency and outpatient waiting rooms in hospitals 4 and 5, respectively. The temperature and relative humidity recorded at most sites was below 27 °C, and 40% and, in sites with high footfall and limited air exchange, such as the hospital medical day unit, indicate a high risk of airborne SARS-CoV-2 transmission.
Wu P, Qiu F, Feng W, et al., 2022, A non-intrusive reduced order model with transformer neural network and its application, PHYSICS OF FLUIDS, Vol: 34, ISSN: 1070-6631
- Author Web Link
- Cite
- Citations: 3
Cheng M, Fang F, Navon IM, et al., 2022, Spatio-temporal hourly and daily ozone forecasting in china using a hybrid machine learning model: autoencoder and generative adversarial networks, Journal of Advances in Modeling Earth Systems, Vol: 14, Pages: 1-26, ISSN: 1942-2466
Efficient and accurate real-time forecasting of national spatial ozone distribution is critical to the provision of effective early warning. Traditional numerical air quality models require a high computational cost associated with running large-scale numerical simulations. In this work, we introduce a hybrid model (VAE-GAN) combining a generative adversarial network (GAN) with a variational autoencoder (VAE) to learn the dynamic ozone distributions in spatial and temporal spaces. The VAE-GAN model can not only decipher the complex nonlinear relationship between the inputs (the past states/ozone and meteorological factors) and outputs (ozone), but also provide ozone forecasts for a long lead-time beyond the training period. The performance of VAE-GAN is demonstrated in hourly and daily spatio-temporal ozone forecasts over China. The training datasets from 2013 to 2017 and validation datasets from 2018 to 2019 are the collection of data from the air quality reanalysis datasets. With the use of VAE, large dataset sizes are decreased by three orders of magnitude, enabling hourly and daily forecasts to be computed in seconds. Results show that the VAE-GAN achieves a reasonable accuracy in the prediction of both the spatial and temporal evolution patterns of hourly and daily ozone fields, as compared to the Nested Air Quality Prediction Modeling System (commonly used in China), the reanalysis data and observations during the validation period. Thus, the VAE-GAN is a cost-effective tool for large data-driven predictions, which can potentially reinforce air pollution prediction efforts in providing risk assessment and management in a timely manner.
Steppeler J, Li J, Fang F, et al., 2022, The o2o3 Local Galerkin Method Using a Differentiable Flux Representation, JOURNAL OF THE METEOROLOGICAL SOCIETY OF JAPAN, Vol: 100, Pages: 9-27, ISSN: 0026-1165
Steppeler J, Li J, Zhu J, et al., 2022, The o2o3 Local Galerkin Method Using a Differentiable Flux Representatio, Journal of the Meteorological Society of Japan, Vol: 100, Pages: 9-27, ISSN: 0026-1165
The spectral element (SE) and local Galerkin (LG) methods may be regarded as variants and generalizations of the classic Galerkin approach. In this study, the second-order spectral element (SE2) method is compared with the alternative LG scheme referred to as o2o3, which combines a second-order field representation (o2) with a third-order representation of the flux (o3). The full name of o2o3 is o2o3C0C1, where the continuous basis functions in C0-space are used for the field representation and the piecewise third-order differentiable basis functions in C1-space are used for the flux approximation. The flux in o2o3 is approximated by a piecewise polynomial function that is both continuous and differentiable, contrary to several Galerkin and LG schemes that use either continuous or discontinuous basis functions for flux approximations. We show that o2o3 not only has some advantages of SE schemes but also possesses third-order accuracy similar to o3o3 and SE3, while SE2 possesses second-order accuracy and does not show superconvergence. SE3 has an approximation order greater than or equal to three and uses the irregular Gauss–Lobatto collocation grid, whereas SE2 and o2o3 have a regular collocation grid; this constitutes an advantage for physical parameterizations and follow-up models, such as chemistry or solid-earth models. Furthermore, o2o3 has the technical simplicity of SE2. The common features (accuracy, convergence, and numerical dispersion relations) and differences between these schemes are described in detail for one-dimensional homogeneous advection tests. A two-dimensional test for cut cells indicates the suitability of o2o3 for realistic applications.
Cai S, Wang Y, Fang F, 2022, NONSTATIONARY SEISMIC-WELL TIE USING DYNAMIC TIME WARPING, Pages: 2949-2953
Seismic-well tie can be viewed as a labelling procedure for seismic data, in which well-logging data with relatively accurate stratigraphical and petrophysical information are correlated to seismic data. This labeled data can then be used as input data for deep learning methods in different tasks, such as seismic processing, interpretation, and inversion. It is difficult to conduct the seismic-well tie procedure. Nonstationary time-shifts commonly exist between the synthetic seismogram constructed from well-logging data and the seismic trace at the vicinity of the well location. These nonstationary time-shifts are caused by the inaccuracy of velocities used to build depth-to-time relations. Nonstationarity also exists in seismic data, which is caused by time-varying wavelets in viscoelastic media. This work proposes a nonstationary seismic-well tie method to automatically correlate well-logging data with seismic data. The nonstationary time-shifts are estimated using the Dynamic Time Warping algorithm. Numerical simulation identifies that the proposed method can successfully tie well-logging data to seismic data by estimating and correcting the nonstationary time-shifts iteratively. Accurate seismic velocities can be obtained by iteratively adjusting the initial logged sonic velocities using a nonstationary velocity adjustment operator in this process. From tied reflectivity series and seismic data, accurate nonstationary wavelets can be estimated.
Li J, Fang F, Steppeler J, et al., 2021, Demonstration of a three-dimensional dynamically adaptive atmospheric dynamic framework for the simulation of mountain waves, Meteorology and Atmospheric Physics, Vol: 133, Pages: 1627-1645, ISSN: 0177-7971
In this paper, Fluidity-Atmosphere, representative of a three-dimensional (3D) non-hydrostatic Galerkin compressible atmospheric dynamic framework, is generated to resolve large-scale and small-scale phenomena simultaneously. This achievement is facilitated by the use of non-hydrostatic equations and the adoption of a flexible 3D dynamically adaptive mesh where the mesh is denser in areas with higher gradients of variable solutions and relatively sparser in the rest of the domain while maintaining promising accuracy and reducing computational resource requirements. The dynamic core is formulated based on anisotropic tetrahedral meshes in both the horizontal and vertical directions. The performance of the adaptive mesh techniques in Fluidity-Atmosphere is evaluated by simulating the formation and propagation of a non-hydrostatic mountain wave. The 2D anisotropic adaptive mesh shows that the numerical solution is in good agreement with the analytic solution. The variation in the horizontal and vertical resolutions has a strong impact on the smoothness of the results and maintains convergence even at high resolutions. When the simulation is extended to 3D, Fluidity-Atmosphere shows stable and symmetric results in the benchmark test cases. The flows over a bell-shaped mountain are resolved quite smoothly. For steep mountains, Fluidity-Atmosphere performs very well, which shows the potential of using 3D adaptive meshes in atmospheric modeling. Finally, as an alternative cut-cell mesh in Fluidity-Atmosphere, the anisotropic adaptive mesh coupled with the Galerkin method provides an alternative accurate representation of terrain-induced flow.
Cheng M, Fang F, Navon IM, et al., 2021, A real-time flow forecasting with deep convolutional generative adversarial network: Application to flooding event in Denmark, Physics of Fluids, Vol: 33, Pages: 1-14, ISSN: 1070-6631
Real-time flood forecasting is crucial for supporting emergency responses to inundation-prone regions. Due to uncertainties in the future (e.g., meteorological conditions and model parameter inputs), it is challenging to make accurate forecasts of spatiotemporal floods. In this paper, a real-time predictive deep convolutional generative adversarial network (DCGAN) is developed for flooding forecasting. The proposed methodology consists of a two-stage process: (1) dynamic flow learning and (2) real-time forecasting. In dynamic flow learning, the deep convolutional neural networks are trained to capture the underlying flow patterns of spatiotemporal flow fields. In real-time forecasting, the DCGAN adopts a cascade predictive procedure. The last one-time step-ahead forecast from the DCGAN can act as a new input for the next time step-ahead forecast, which forms a long lead-time forecast in a recursive way. The model capability is assessed using a 100-year return period extreme flood event occurred in Greve, Denmark. The results indicate that the predictive fluid flows from the DCGAN and the high fidelity model are in a good agreement (the correlation coefficient
Zheng J, Wu X, Fang F, et al., 2021, Numerical study of COVID-19 spatial-temporal spreading in London, PHYSICS OF FLUIDS, Vol: 33, ISSN: 1070-6631
- Author Web Link
- Cite
- Citations: 14
Wang Y, Ding X, Hu K, et al., 2021, Feasibility of DEIM for retrieving the initial field via dimensionality reduction, JOURNAL OF COMPUTATIONAL PHYSICS, Vol: 429, ISSN: 0021-9991
- Author Web Link
- Cite
- Citations: 4
Fang F, 2021, Numerical and Data-Driven Modelling in Coastal, Hydrological and Hydraulic Engineering, WATER, Vol: 13
- Author Web Link
- Cite
- Citations: 1
Cheng M, Fang F, Pain CC, et al., 2020, An advanced hybrid deep adversarial autoencoder for parameterized nonlinear fluid flow modelling, Computer Methods in Applied Mechanics and Engineering, Vol: 372, Pages: 1-19, ISSN: 0045-7825
Considering the high computation cost required in conventional computation fluid dynamic simulations, machine learning methods have been introduced to flow dynamic simulations in years, aiming on reducing CPU time. In this work, we propose a hybrid deep adversarial autoencoder (VAE-GAN) to integrate generative adversarial network (GAN) and variational autoencoder (VAE) for predicting parameterized nonlinear fluid flows in spatial and temporal dimensions. High-dimensional inputs are compressed into the low-dimensional representations by nonlinear functions in a convolutional encoder. In this way, the predictive fluid flows reconstructed in a convolutional decoder contain the dynamic fluid flow physics of high nonlinearity and chaotic nature. In addition, the low-dimensional representations are applied to the adversarial network for model training and parameter optimization, which enables fast computation process. The capability of the hybrid VAE-GAN is illustrated by varying inputs on a flow past a cylinder test case as well as a second case of water column collapse. Numerical results show that this hybrid VAE-GAN has successfully captured the spatio-temporal flow features with CPU speed-up of three orders of magnitude. These promising results suggest that the hybrid VAE-GAN can play a critical role in efficiently and accurately predicting complex flows in future research efforts.
Cheng M, Fang F, Kinouchi T, et al., 2020, Long lead-time daily and monthly streamflow forecasting using machine learning methods, Journal of Hydrology, Vol: 590, Pages: 1-13, ISSN: 0022-1694
Long lead-time streamflow forecasting is of great significance for water resources planning and management in both the short and long terms. Despite of some studies using machine learning methods in streamflow forecasting, only few studies have been conducted to explore long lead-time forecasting capabilities of these methods, and gain an insight into systematic comparison of model forecasting performance in both the short and long terms. In this work, an artificial neural network (ANN) and a long short term memory (LSTM), a powerful tool for learning long-term temporal dependencies and capturing nonlinear relationship, have been adopted to forecast streamflow at daily and monthly scales for a long lead-time period. For long lead-time streamflow forecasting, a recursive forecasting procedure, which takes the last one-step-ahead forecast as a new input for the next-step-ahead forecast, is used in the ANN and LSTM forecasting systems. Two models are trained and validated for streamflow forecasting using the rainfall and runoff datasets collected from the Nan River Basin and Ping River Basin, Thailand, covering the period 1974 to 2014. To further explore the impact of parameter settings on model performance, two parameters, i.e. the length of time lag and the number of maximum epochs, are examined in the ANN and LSTM models. The main findings are highlighted here. First, with an optimal setting up of model parameters, both the ANN and LSTM model can provide accurate daily forecasting (up to 20 days ahead). Second, in comparison to the ANN model, the LSTM model exhibits better model performance in long lead-time daily forecasting, but less satisfactory in multi-monthly forecasting due to lack of large monthly training dataset. Third, the selection of the length of the time lag and number of maximum epochs used in both ANN and LSTM modelling are the key for long lead-time streamflow forecasting at daily and monthly scales. These findings suggest that the LSTM could be ad
Cheng M, Fang F, Pain CC, et al., 2020, Data -driven modelling of nonlinear spatio-temporal fluid flows using a deep convolutional generative adversarial network, Computer Methods in Applied Mechanics and Engineering, Vol: 365, Pages: 1-18, ISSN: 0045-7825
Deep learning techniques for fluid flow modelling have gained significant attention in recent years. Advanced deep learning techniques achieve great progress in rapidly predicting fluid flows without prior knowledge of the underlying physical relationships. However, most of existing researches focused mainly on either sequence learning or spatial learning, rarely on both spatial and temporal dynamics of fluid flows (Reichstein et al., 2019). In this work, an Artificial Intelligence (AI) fluid model based on a general deep convolutional generative adversarial network (DCGAN) has been developed for predicting spatio-temporal flow distributions. In deep convolutional networks, the high-dimensional flows can be converted into the low-dimensional “latent” representations. The complex features of flow dynamics can be captured by the adversarial networks. The above DCGAN fluid model enables us to provide reasonable predictive accuracy of flow fields while maintaining a high computational efficiency. The performance of the DCGAN is illustrated for two test cases of Hokkaido tsunami with different incoming waves along the coastal line. It is demonstrated that the results from the DCGAN are comparable with those from the original high fidelity model (Fluidity). The spatio-temporal flow features have been represented as the flow evolves, especially, the wave phases and flow peaks can be captured accurately. In addition, the results illustrate that the online CPU cost is reduced by five orders of magnitude compared to the original high fidelity model simulations. The promising results show that the DCGAN can provide rapid and reliable spatio-temporal prediction for nonlinear fluid flows.
Zheng J, Fang F, Wang Z, et al., 2020, A new anisotropic adaptive mesh photochemical model for ozone formation in power plant plumes, ATMOSPHERIC ENVIRONMENT, Vol: 229, ISSN: 1352-2310
- Author Web Link
- Cite
- Citations: 3
Steppeler J, Li J, Navon IM, et al., 2020, Medium range forecasts using cut-cells: a sensitivity study, METEOROLOGY AND ATMOSPHERIC PHYSICS, Vol: 132, Pages: 171-179, ISSN: 0177-7971
- Author Web Link
- Cite
- Citations: 2
Arcucci R, Casas CQ, Xiao D, et al., 2020, A Domain Decomposition Reduced Order Model with Data Assimilation (DD-RODA), Conference on Parallel Computing - Technology Trends (ParCo), Publisher: IOS PRESS, Pages: 189-198, ISSN: 0927-5452
Li J, Steppeler J, Fang F, et al., 2019, Potential numerical techniques and challenges for atmospheric modeling, Bulletin of the American Meteorological Society, Vol: 125, Pages: ES239-ES242, ISSN: 0003-0007
Xiao D, Fang F, Heaney CE, et al., 2019, A domain decomposition method for the non-intrusive reduced order modelling of fluid flow, Computer Methods in Applied Mechanics and Engineering, Vol: 354, Pages: 307-330, ISSN: 0045-7825
In this paper we present a new domain decomposition non-intrusive reduced order model (DDNIROM) for the Navier–Stokes equations. The computational domain is partitioned into subdomains and a set of local basis functions is constructed in each subdomain using Proper Orthogonal Decomposition (POD). A radial basis function (RBF) method is then used to generate a set of hypersurfaces for each subdomain. Each local hypersurface represents, not only the fluid dynamics over the subdomain to which it belongs, but also the interactions with the surrounding subdomains. This implicit coupling between the subdomains provides the global coupling necessary to enforce incompressibility and is a means of providing boundary conditions for each subdomain.The performance of this DDNIROM is illustrated numerically by three examples: flow past a cylinder, and air flow over 2D and 3D street canyons. The results show that the DDNIROM exhibits good agreement with the high-fidelity full model while the computational cost is reduced by several orders of magnitude. The domain decomposition (DD) method provides the flexibility to choose different numbers of local basis functions for each subdomain depending on the complexity of the flow therein. The fact that the RBF surface representation takes input only from its current subdomain and the surrounding subdomains, means that, crucially, there is a reduction in the dimensionality of the hypersurface when compared with a more traditional, global NIROM. This comes at the cost of having a larger number of hypersurfaces.
Hu R, Fang F, Pain CC, et al., 2019, Rapid spatio-temporal flood prediction and uncertainty quantification using a deep learning method, Journal of Hydrology, Vol: 575, Pages: 911-920, ISSN: 0022-1694
Recently accrued attention has been given to machine learning approaches for flooding prediction. However, most of these studies focused mainly on time-series flooding prediction at specified sensors, rarely on spatio-temporal prediction of inundations. In this work, an integrated long short-term memory (LSTM) and reduced order model (ROM) framework has been developed. This integrated LSTM-ROM has the capability of representing the spatio-temporal distribution of floods since it takes advantage of both ROM and LSTM. To reduce the dimensional size of large spatial datasets in LSTM, the proper orthogonal decomposition (POD) and singular value decomposition (SVD) approaches are introduced. The LSTM training and prediction processes are carried out over the reduced space. This leads to an improvement of computational efficiency while maintaining the accuracy. The performance of the LSTM-ROM developed here has been evaluated using Okushiri tsunami as test cases. The results obtained from the LSTM-ROM have been compared with those from the full model (Fluidity). In predictive analytics, it is shown that the results from both the full model and LSTM-ROM are in a good agreement whilst the CPU cost using the LSTM-ROM is decreased by three orders of magnitude compared to full model simulations. Additionally, prescriptive analytics has been undertaken to estimate the uncertainty in flood induced conditions. Given the time series of the free surface height at a specified detector, the corresponding induced wave conditions along the coastline have then been provided using the LSTM network. Promising results indicate that the use of LSTM-ROM can provide the flood prediction in seconds, enabling us to provide real-time predictions and inform the public in a timely manner, reducing injuries and fatalities.
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.