High Computational Efficiency Research Articles

Snowfall deposition in mountainous terrain: a statistical downscaling scheme from high-resolution model data on simulated topographies

One of the primary causes of non-uniform snowfall deposition on the ground in mountainous regions is the preferential deposition of snow, which results from the interaction of near-surface winds with topography and snow particles. However, producing high-resolution snowfall deposition patterns can be computationally expensive due to the need to run full atmospheric models. To address this, we developed two statistical downscaling schemes that can efficiently downscale near-surface, low-resolution snowfall data to fine-scale snow deposition accounting for the effect of preferential deposition in mountainous regions. Our approach relies on a comprehensive, model database generated using 3D wind fields from an atmospheric model and a preferential deposition model on several thousand simulated topographies covering a broad range in terrain characteristics. Both snowfall downscaling schemes rely on fine-scale topographic scaling parameters and low-resolution wind speed as input. While one scheme, referred to as the “wind scheme”, further necessitates fine-scale vertical wind components, a second scheme, referred to as the “aspect scheme”, does not require fine-scale temporal input. We achieve this by additionally downscaling near-surface vertical wind speed solely using topographic scaling parameters and low-resolution wind direction. We assess the performance of our downscaling schemes using an independent subset of the model database on simulated topographies, model data on actual terrain, and spatially measured new snow depth obtained through a photogrammetric drone survey following a snowfall event on previously snow-free ground. While the assessments show that our downscaling schemes perform well (relative errors ≤ ±3% with modeled and ≤ ±6% with measured snowfall deposition), they also demonstrate comparable results to benchmark downscaling models. However, our schemes notably outperform the benchmark models in representing fine-scale patterns. Our downscaling schemes possess several key features, including high computational efficiency, versatility enabled by the comprehensive model database, and independence from fine-scale temporal input data (aspect scheme), indicating their potential for widespread applicability. Therefore, our downscaling schemes for near-surface snowfall and vertical wind speed can be beneficial for various applications at fine grid resolutions such as in atmospheric and climate sciences, snow hydrology, glaciology, remote sensing, and avalanche sciences.

An algorithm/hardware co‐optimized method to accelerate CNNs with compressed convolutional weights on FPGA

SummaryConvolutional neural networks (CNNs) have shown remarkable advantages in a wide range of domains at the expense of huge parameters and computations. Modern CNNs still tend to be more complex and larger to achieve better inference accuracy. However, the complex and large structures of CNNs could slow down the inference speed. Recently, Compressing the convolutional weights to be sparse by pruning the unimportant parameters has been demonstrated as an efficient way to reduce the computations of CNNs. On the other hand, field‐programmable gate arrays (FPGAs) have been a popular hardware platform to accelerate CNN inference. In this paper, we propose an algorithm/hardware co‐optimized method for accelerating CNN inference on FPGAs. For the algorithm, we take advantage of unstructured and structured parameter sparsifying methods to achieve high sparsity and keep the regularity of convolutional weights. Correspondingly, hardware‐friendly index representations of sparse convolutional weights are proposed. For the hardware architecture, we propose row‐wise input‐stationary dataflow, which is tightly coupled with the algorithm. A row‐wise computing engine (RConv Engine) is proposed, which is based on the dataflow. Inside the RConv Engine, the scalar‐vector structure is applied to implement the basic processing elements (PEs). To flexibly calculate the feature map with various sizes, the PEs are organized in a 2D structure with two work modes. The experimental results demonstrate that our co‐optimized method implements high sparsity of convolutional weights, and the computing engine achieves high computation efficiency. Compared with other accelerators, our co‐optimized method implements a 10.9 speedup on FPS at most with the highest sparsity of convolutional weights and negligible accuracy loss.

Pyramid hybrid pooling quantization for efficient fine-grained image retrieval

Deep hashing approaches, including deep quantization and deep binary hashing, have become a common solution to large-scale image retrieval due to their high computation and storage efficiency. Most existing hashing methods cannot produce satisfactory results for fine-grained retrieval, because they usually adopt the outputs of the last CNN layer to generate binary codes. Since deeper layers tend to summarize visual clues, e.g., texture, into abstract semantics, e.g., dogs and cats, the feature produced by the last CNN layer is less effective in capturing subtle but discriminative visual details that mostly exist in shallow layers. To improve fine-grained image hashing, we propose Pyramid Hybrid Pooling Quantization (PHPQ). Specifically, we propose a Pyramid Hybrid Pooling (PHP) module to capture and preserve fine-grained semantic information from multi-level features, which emphasizes the subtle discrimination of different sub-categories. Besides, we propose a learnable quantization module with a partial codebook attention mechanism, which helps to optimize the most relevant codewords and improves the quantization. Comprehensive experiments on two widely-used public benchmarks, i.e., CUB-200-2011 and Stanford Dogs, demonstrate that PHPQ outperforms state-of-the-art methods.

Properly shortening design time scale of medium-deep borehole heat exchanger for high building heating performances with high computational efficiency

The medium-deep borehole heat exchanger (MDBHE) is a clean solution for building heating with renewable geothermal energy. The long-term performance evaluation with the design time scale same as the life span of the MDBHE is generally adopted by the MDBHE design for high heating performances. However, the long-term performance evaluation largely increases the computational load of the MDBHE design, resulting in practical inconvenience. This study proposes the idea that properly shortening the design time scale could maintain the information of the long-term performance evaluation for the MDBHE design with high heating performances while reducing the computational load. The idea is verified by the MDBHE design for building heating in the severe cold zone of China. The multiple design parameters of the MDBHE are simultaneously optimized considering climate change. The results show that under different geological conditions, economic conditions, and system safety requirements, compared with the life span of the MDBHE of 30–50 years, the design time scale properly shortened to 10 years achieves similar high heating performances regarding the energy consumption, cost, and system safety, and reduces the computational load by 66.1 %–79.3 %. This study contributes to high computational-efficient and high-performance design of the MDBHE for building heating.

Salinity Inversion of Flat Sea Surface Based on Deep Neural Network

The distribution and change of sea surface salinity (SSS) have an important influence on the sea dynamic environment, marine ecological environment, global water cycle, and global climate change. Satellite remote sensing is the only practical way to continuously observe SSS over a wide area and for a long period of time. The salinity retrieval model of flat sea surface, which primarily includes empirical model and iterative model, is the key to retrieving satellite SSS products. The empirical models have high computational efficiency but low inversion accuracy, while the iterative models have high inversion accuracy but low computational efficiency. In order to reconcile the contradiction between the computational efficiency and inversion accuracy of existing models, this paper proposes a universal deep neural network (DNN) model architecture and corresponding training scheme, and provides 3 DNN models with extremely high computational efficiency and high inversion accuracy. The inversion error range, the root mean square error (RMSE), and the mean absolute error (MAE) of the DNN models on 311,121 sets of data have decreased by more than 40 times, 150 times, and 150 times, respectively, compared to the empirical model. The computational efficiency of the DNN models on 420,903 sets of data has improved by more than 100,000 times compared to the iterative model. Therefore, the algorithm developed in this paper can effectively solve the contradiction between the computational efficiency and inversion accuracy of existing models, and provide a theoretical support for high-precision and high-efficiency salinity inversion research.

An empirical formula of nuclear <i>β</i>-decay half-lives

Nuclear <i>β</i>-decay half-lives play an important role not only in nuclear physics, but also in astrophysics. The <i>β</i>-decay half-lives of many nuclei involved in the astrophysical rapid neutron-capture (r -process) still cannot be measured experimentally, so the theoretical predictions of nuclear <i>β</i>-decay half-lives are inevitable for r-process studies. Theoretical models for studying the nuclear <i>β</i>-decay half-lives include the empirical formula, the gross theory, the quasiparticle random phase approximation (QRPA), and the shell model. Compared with other theoretical models of <i>β</i>-decay half-lives, the empirical formula has high computational efficiency, and its prediction accuracy can be improved by introducing more and more physical information. In this work, an empirical formula without free parameters is proposed to calculate the nuclear <i>β</i>-decay half-lives based on the Fermi theory of <i>β</i> decay. By including the pairing effect, the shell effect, and the isospin dependence, the newly proposed empirical formula significantly improves the accuracy of predicting the nuclear <i>β</i>-decay half-life. For the nuclei with half-lives less than 1 second, the root-mean-square deviation of the common logarithms of the nuclear <i>β</i>-decay half-life predicted by the new empirical formula from the experimental data decreases to 0.220, which is improved by about 54% compared with that by the empirical formula without free parameters, even better than those by other existing empirical formulas and microscopic QRPA approaches. In the unknown region, the nuclear <i>β</i>-decay half-lives predicted by the new empirical formula are generally shorter than those predicted by the microscopic models in the light nuclear region, while those predicted by the new empirical formula in the heavy nuclear region are generally in agreement with those predicted by the microscopic models. The half-lives of neutron-rich nuclei on the nuclear chart are then predicted by the new empirical formula, providing nuclear <i>β</i>-decay half-life inputs for the r-process simulations.

Mechanism and rule of microwave pulse response of two-stage PIN limiter

<sec>This paper aims to analyze the failure mechanism of the two-stage PIN limiter after having been injected by a microwave pulse. A two-stage PIN limiter model with high computational efficiency and accuracy is built by using the method of field-circuit collaborative simulation. Using this model, the temperature change of the PIN diodes during the injection of microwave pulse is simulated. The melting temperature of the PIN diode is selected as the failure criterion of the PIN limiter. The time and energy required for the failure of the PIN limiter under injection of microwave pulses with different frequencies and amplitudes are simulated. Furthermore, the mechanisms that trigger off these effects are analyzed. The relationship between the microwave pulse parameters and the PIN limiter failure time is summarized by using an empirical formula.</sec><sec>According to the simulation results, the temperature change of the second-stage PIN diode is relatively small compared with that of the first-stage. During the injection of the microwave pulse, the failure time and energy consumption of limiter show a certain regularity with the variation of microwave pulse amplitude and frequency, and this work discusses this regularity from the following three aspects. Firstly, the failure time and energy consumption decrease in a similar trend with frequency increasing. And with the increase in signal amplitude, the failure time and energy consumption tend to stabilize. Secondly, the increase in the signal amplitude leads failure time to decrease, which is similar to the relationship between failure time and the signal frequency mentioned before. But as the signal’s amplitude increases, the energy consumption first increases and then decreases slightly when the amplitude reaches about 900 V. Based on the theoretical analysis and the physical image of the two-stage PIN limiter, the reasons for these effects can be explained as the changes in the I-region's impedance and heat distribution change caused by electric field changes. Thirdly, the failure time and energy consumption show different sensitivities to different parameters of the microwave pulse. The signal frequency change has greater influence on the energy consumption than the signal amplitude change, while the amplitude change can exert a greater influence on the failure time than the frequency change.</sec>

Energy-Efficient Sleep Apnea Detection Using a Hyperdimensional Computing Framework Based on Wearable Bracelet Photoplethysmography.

Sleep apnea syndrome (SAS) is a common sleep disorder, which has been shown to be an important contributor to major neurocognitive and cardiovascular sequelae. Considering current diagnostic strategies are limited with bulky medical devices and high examination expenses, a large number of cases go undiagnosed. To enable large-scale screening for SAS, wearable photoplethysmography (PPG) technologies have been used as an early detection tool. However, existing algorithms are energy-intensive and require large amounts of memory resources, which are believed to be the major drawbacks for further promotion of wearable devices for SAS detection. In this paper, an energy-efficient method of SAS detection based on hyperdimensional computing (HDC) is proposed. Inspired by the phenomenon of chunking in cognitive psychology as a memory mechanism for improving working memory efficiency, we proposed a one-dimensional block local binary pattern (1D-BlockLBP) encoding scheme combined with HDC to preserve dominant dynamical and temporal characteristics of pulse rate signals from wearable PPG devices. Our method achieved 70.17% accuracy in sleep apnea segment detection, which is comparable with traditional machine learning methods. Additionally, our method achieves up to 67× lower memory footprint, 68× latency reduction, and 93× energy saving on the ARM Cortex-M4 processor. The simplicity of hypervector operations in HDC and the novel 1D-BlockLBP encoding effectively preserve pulse rate signal characteristics with high computational efficiency. This work provides a scalable solution for long-term home-based monitoring of sleep apnea, enhancing the feasibility of consistent patient care.

A Physics-Guided Graph Convolution Neural Network for Optimal Power Flow

The data-driven method with strong approximation capabilities and high computational efficiency provides a promising tool for optimal power flow (OPF) calculation with stochastic renewable energy. However, the topology change dramatically increases the learning difficulties and the demand for learning samples. In this work, we propose a physics-guided graph convolution neural network (GCNN) for OPF calculation with consideration of varying topologies, including the physics-guided graph convolution kernel, feature construction, and loss function formulation. Specifically, a physics-embedded graph convolution kernel is derived by aggregating the features from local neighborhoods utilizing the nodal OPF model formulation. An iterative feature construction method is also developed that encodes both the physical feature and practical constraints into the node vector. Finally, a correlative learning loss function to optimize the unbalanced power injection is developed. Extensive numerical results carried out on various IEEE test systems show that the prediction accuracy of OPF using the proposed method under varying topology changes can be improved by an average of 13.30% and up to 32.63% compared with state-of-the-art methods.

H2CT: Asynchronous Distributed Key Generation With High Computational Efficiency and Threshold Security in Blockchain Network

H²CT: Asynchronous Distributed Key Generation With High Computational Efficiency and Threshold Security in Blockchain Network

Spatial Reliability Enhanced Correlation Filter: An Efficient Approach for Real-Time UAV Tracking

Traditional discriminative correlation filter (DCF) has received great popularity due to its high computational efficiency. However, the lightweight framework of DCF cannot promise robust performance when the tracker faces appearance variations within the background. These unpredictable appearance variations always distract the filter. Most existing DCF-based trackers either utilize deep convolutional features or incorporate additional constraints to elevate tracking robustness. Despite some improvements, both of them hamper the tracking speed and can only roughly alleviate the distractions of appearance variations. In this paper, a novel spatial reliability enhanced learning strategy is proposed to handle the problems aforementioned. By monitoring the variation of response produced in detection phase, a dynamic reliability map is generated to indicate the reliability of each background subregion. Then, label adjustment is conducted to repress the distractions of these unreliable areas. Compared with the conventional way of constraint where a new term is always added to realize the desired goal, label adjustment is simultaneously more efficient and effective. Moreover, to promise the accuracy and dependability of the reliability map, an adaptively updated response pool recording reliable historical response values is proposed. Extensive and exhaustive experiments on three challenging unmanned aerial vehicle (UAV) benchmarks, i.e., UAV123@10fps, DTB70 and UAVDT, which totally include 243 video sequences, validate the superiority of the proposed method against other state-of-the-art trackers and exhibit a remarkable generality in a variety of scenarios. Meanwhile, the tracking speed of 65.2FPS on a cheap CPU makes it suitable for real-time UAV applications.

Fast Quasi-Optimal Power Flow of Flexible DC Traction Power Systems

This paper proposes a quasi-optimal power flow (OPF) algorithm for flexible DC traction power systems (TPSs). Near-optimal OPF solutions can be solved with high computational efficiency by quasi-OPF. Unlike conventional OPF algorithms, quasi-OPF does not utilize mathematical optimization algorithms but adopts a new methodology. First, we adopt a new modeling method and successfully reveal the physical meaning of OPF solutions in flexible DC TPSs. Then, by converting the physical meaning of OPF solutions into mathematical expressions, a simple mapping from the power flow solution to the near-optimal OPF solution is obtained, and the quasi-OPF algorithm is designed based on power flow and this mapping. Since calculating power flow is computationally cheap and the mapping is based on simple arithmetic, the quasi-OPF algorithm can solve OPF with much less execution time, achieving subsecond level calculation and a speed-up of 57 times compared to the primal-dual interior point method. The effectiveness is verified by mathematical proofs and a case study with Beijing Metro Line 13. This study provides insight into the physical meaning of OPF solutions and is a powerful tool for flexible DC TPSs to analyze the effects of coordinated control, design real-time coordinated control strategies, and solve operational problems in planning.

Blockchain-Based Secure and Efficient Secret Image Sharing With Outsourcing Computation in Wireless Networks

Secret Image Sharing (SIS) is the technology that shares any given secret image by generating and distributing <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</i> shadow images in the way that any subset of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k</i> shadow images can restore the secret image. However, in the existing SIS schemes, the shadow images will be easily tampered and corrupted during the communication, which will pose serious security issues. Recently, blockchain has emerged as a promising paradigm in the field of data communication and information security. To securely communicate and effectively protect the secret image data in wireless networks, we propose a Blockchain-based Secure and Efficient Secret Image Sharing (BC-SESIS) scheme with outsourcing computation in wireless networks. In the proposed BC-SESIS scheme, the shadow images are encrypted and stored in the blockchain to prevent them from being tampered and corrupted. The identity authentication-enabled smart contract is deployed to achieve the ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k, n</i> ) threshold for secret image restoring. Furthermore, to reduce the computational burden of smart contract and users, an efficient outsourcing computation method is designed to outsource the restoring task, which is securely implemented by agent miners in the encryption domain. Theoretical analysis and extensive experiments demonstrate that the BC-SESIS scheme can achieve desirable communication security and high computational efficiency in the wireless networks.

Quantitative analysis method of laser-induced breakdown spectroscopy based on temperature iterative correction of self-absorption effect

Laser-induced breakdown spectroscopy (LIBS) is an ideal real-time on-line method of detecting minor elements in alloys. However, in the case of laser-produced high-density plasma, the self-absorption is usually an undesired effect because it not only reduces the true line intensity, leading the line intensity to become nonlinear with the increase of emitting species content, but also affects the characterization parameters of the plasma, and finally affects the accuracy of quantitative analysis. Since the plasma electron temperature (&lt;i&gt;T&lt;/i&gt;), radiation particle number density and absorption path length (&lt;i&gt;Nl&lt;/i&gt;) determine the degree of self-absorption and affect the corrected spectral line intensity, a new self-absorption correction method is proposed based on temperature iteration. The initial &lt;i&gt;T&lt;/i&gt; is obtained by using this method through spectral line intensity, and the self-absorption coefficient SA is calculated based on the initial &lt;i&gt;Nl&lt;/i&gt; parameter to correct the spectral line intensity. Then a new &lt;i&gt;T&lt;/i&gt; is obtained from the new spectral line intensity and the new SA is calculated to further correct the spectral line intensity. Through continuous calculation and correction of these two parameters, self-absorption correction is finally achieved. The experimental results of alloy steel samples show that the linearity of Boltzmann plot is increased from 0.867 without self-absorption correction to 0.974 with self-absorption correction, and the linear correlation coefficient &lt;i&gt;R&lt;/i&gt;&lt;sup&gt;2&lt;/sup&gt; of the single variable calibration curve for Mn element increases from 0.971 to 0.997. The relative error of elemental content measurement is improved from 4.32% without self-absorption correction to 1.23% with self-absorption correction. Compared with the commonly applied self-absorption correction methods, this method has obvious advantages of simpler programming, higher computation efficiency, and its independence of the availability or accuracy of Stark broadening coefficients. Moreover, this method can directly obtain the radiation particle number density and absorption path length, which is beneficial to the diagnosis and quantitative analysis of plasma.

Explicit and unconditionally stable finite-difference time-domain subgridding algorithm based on hanging variables

Due to the limitation of the stability condition, the finite-difference time-domain (FDTD) method cannot efficiently deal with electromagnetic problems containing fine structures. The explicit and unconditionally stable (EUS) FDTD method can eliminate the constraint of the stability condition and improve the simulation efficiency of fine structures by filtering out the unstable modes for the system matrix. However, the EUS-FDTD method needs to solve the eigenvalues of the numerical system matrix, and the symmetry of the numerical system matrix needs to be ensured when the subgridding scheme is used to discretize targets containing fine structures. The existing EUS-FDTD subgridding method encounters some problems such as complex implementation and insufficient accuracy. In order to solve the above problems, in this work, the hanging variables subgridding (HVS) algorithm is applied to the EUS-FDTD algorithm. Starting from the symmetry of the system matrix, the stability of the hanging variables subgridding algorithm is proven, and a high-precision, stable, and easy-to-implement HVS-EUS-FDTD scheme is proposed. Numerical examples of the radiation of linear magnetic currents in free space, electromagnetic scattering of multiple dielectric objects, and a three-dimensional cavity containing a medium demonstrate the stability, high accuracy and efficiency of the proposed method. Numerical experiments show that the computational efficiency of the HVS-EUS-FDTD algorithm can be improved hundreds of times compared with that of the uniform fine grid FDTD algorithm, and the highest computational efficiency can be improved up to ratio (the size ratio of coarse grid to fine grid) times compared with that of the HVS-FDTD algorithm.

Smoothing interpolation of five-axis tool path with less feedrate fluctuation and higher computation efficiency

Five-axis machine tools are widely used in aerospace, die and mold industries. Most of the time, the tool path is described by connecting a series of short linear segments. The tangency discontinuity of the linear segments leads to the discontinuity of the kinematic profiles, hence generating discontinuous interpolation points, and resulting in fluctuation of machine motion, which reduces the machining efficiency and causes excessive tracking and contouring errors. This paper aims to generate smooth interpolation points at servo sample intervals with the proposed G2 continuous five-axis corner rounding algorithm suitable for reducing feedrate fluctuation along the smoothed toolpath and improving machining efficiency, and the modified jerk-limited feedrate planning algorithm suitable for eliminating feedrate fluctuation within the feedrate planning units and improving interpolation computation efficiency. In this paper, the five-axis linear segments are first smoothed by curvature-optimized cubic Bezier splines, which reduces the number of the feedrate planning units along the smoothed toolpath and improves machining efficiency. Then, the jerk-limited feedrate planning algorithm is modified to ensure that the time interval of each different scheduled feedrate phase is an integer multiple of the axis position control loop closure time (i.e. servo sample interval), which eliminates the feedrate fluctuation within each feedrate planning unit and reduces the interpolation computation load of position command in each interpolation period (i.e. servo sample interval). At last, the effectiveness of the proposed algorithm is demonstrated with simulations and experiments following a five-axis, curved tool path on a CNC (Computer Numerical Control) machine tool.

Computational genetic epidemiology: Leveraging HPC for large-scale AI models based on Cyber Security

To better understand disease susceptibility and prevention, computational genetic epidemiology is leading research. This paper introduces GenomeMinds, a breakthrough method for scaling large-scale AI models for disease risk prediction. HPC was used to develop the method. GenomeMinds is compared to six standard methods to demonstrate its benefits. GenomeMinds' incredible potential is shown by real-world performance assessments. These measures evaluate data processing speed, forecast accuracy, scalability, computer efficiency, privacy, and ethics. GenomeMinds benefits are shown via scatter plots, which visually compare data. According to the data, GenomeMinds may revolutionize computational genetic epidemiology by doing well across all criteria. GenomeMinds has faster data processing, better prediction accuracy, stronger scalability, higher computational efficiency, enhanced privacy and security, and a comprehensive ethical awareness.

Multirate Modeling and Predictive Control for WBG-Device-Based High-Switching-Frequency Power Converters

With the growth of wide bandgap devices, it is necessary to exploit the high switching frequency benefits to improve the power converter's performance, thus requiring a higher sampling/interrupt frequency in digital signal processors. However, such a short interrupt time duration imposes big computational difficulty in the execution of programming code, especially using the model predictive control (MPC). Thus, most of the existing MPCs are applied with switching frequencies below 20 kilohertz, which cannot exploit the full potential of the wide bandgap device-based power converters. To solve this challenge, this article proposes a multi-rate MPC scheme, where the trigger of interrupt and switching device transition can be performed at different rates. Compared to conventional MPCs, the main uniqueness of the proposed multi-rate MPC is that the high-dimension control sequence is solved and applied within each interrupt interval. Therefore, the increased switching frequency objective can be easily achieved with a low sampling/interrupt frequency configuration, which also significantly relieves the digital execution of the heavy interrupt tasks. The proposed method shows a more optimized control input and higher computational efficiency over the multi-rate finite-control-set MPC counterpart. A silicon carbide inverter-fed AC motor drive system is used to verify the proposed multi-rate MPC. The results show the improved system performance with the combined advantages of both the high switching frequency and MPC strategy.

Scenario Reduction Network Based on Wasserstein Distance With Regularization

Power systems with high penetration of renewable energy contain various uncertainties. Scenario-based optimization problems need a large number of discrete scenarios to obtain a reliable approximation for the probabilistic model. It is important to choose typical scenarios and ease the computational burden. This paper presents a scenario reduction network model based on Wasserstein distance. Entropy regularization is used to transform the scenario reduction problem into an unconstrained problem. Through an explicit neural network structure design, the output of the scenario reduction network corresponds to the Sinkhorn distance function. The scenario reduction network can generate the typical scenario set through unsupervised learning. An efficient algorithm is proposed for continuous/discrete scenario reduction. The superiority of the scenario reduction network model is verified through case studies. The numerical results highlight the high accuracy and computational efficiency of the proposed model over state-of-the-art models, and it makes an ideal candidate for large-scale scenario reduction problems.

Modeling hydrogen–diesel dual direct injection combustion with FGM and transported PDF

This work extends the transported probability density function (PDF) method to model hydrogen–diesel dual direct injection (H2DDI) combustion, where a H2 jet is ignited by a small pilot diesel jet flame. The work is motivated by previous H2DDI engine tests where stable compression ignition engine operation was reported with up to 90 % H2 supply by energy share. To help explain this high performance and provide a tool to help further engine optimization, the Eulerian Monte Carlo Fields (EMCF) solution method is employed with which a transported PDF equation is solved in combination with Flamelet Generated Manifold (FGM) for high accuracy and computational efficiency. In all cases, the pilot fuel ignites before interacting with the H2 jet and thus reaction rate and chemical composition are computed as the weighted average of the corresponding values taken for two separated FGM tables generated for the two fuels. Pressure measurements and high-speed schlieren imaging performed in a preburn-type optical constant volume combustion chamber (CVCC) with n-heptane (nC7H16) pilot fuel were used to validate the EMCF+FGM model. The application focus is how H2 ignition and combustion is affected when the nC7H16 is injected prior to or after the main H2 jet. EMCF+FGM computed heat release rate and flame structure were compared with experimental data and results from conventional FGM model based on presumed PDF. When applied to H2DDI combustion, the EMCF+FGM model successfully reproduces flame structures and heat release rate under different injection strategies, including the transition towards partially-premixed combustion when the nC7H16 pilot follows the main H2 injection. Moreover, analysis of heat release rate from different stochastic fields can provide a useful indication about cyclic variability and combustion stability.