Design of a Computational Heuristic to Solve the Nonlinear Liénard Differential Model

In this study,the design of a computational heuristic based on the nonlinear Liénard model is presented using the efficiency of artificial neural networks(ANNs)along with the hybridization procedures of global and local search approaches.The global search genetic algorithm(GA)and local search sequential quadratic programming scheme(SQPS)are implemented to solve the nonlinear Liénard model.An objective function using the differential model and boundary conditions is designed and optimized by the hybrid computing strength of the GA-SQPS.The motivation of the ANN procedures along with GA-SQPS comes to present reliable,feasible and precise frameworks to tackle stiff and highly nonlinear differentialmodels.The designed procedures of ANNs along with GA-SQPS are applied for three highly nonlinear differential models.The achieved numerical outcomes on multiple trials using the designed procedures are compared to authenticate the correctness,viability and efficacy.Moreover,statistical performances based on different measures are also provided to check the reliability of the ANN along with GASQPS.

A Back Propagation-Type Neural Network Architecture for Solving the Complete n ×n Nonlinear Algebraic System of Equations

The objective of this research is the presentation of a neural network capable of solving complete nonlinear algebraic systems of n equations with n unknowns. The proposed neural solver uses the classical back propagation algorithm with the identity function as the output function, and supports the feature of the adaptive learning rate for the neurons of the second hidden layer. The paper presents the fundamental theory associated with this approach as well as a set of experimental results that evaluate the performance and accuracy of the proposed method against other methods found in the literature.

Swarming Computational Approach for the Heartbeat Van Der Pol Nonlinear System

The present study is related to design a stochastic framework for the numerical treatment of the Van der Pol heartbeat model(VP-HBM)using the feedforward artificial neural networks(ANNs)under the optimization of particle swarm optimization(PSO)hybridized with the active-set algorithm(ASA),i.e.,ANNs-PSO-ASA.The global search PSO scheme and local refinement of ASA are used as an optimization procedure in this study.An error-based merit function is defined using the differential VP-HBM form as well as the initial conditions.The optimization of the merit function is accomplished using the hybrid computing performances of PSO-ASA.The designed performance of ANNs-PSO-ASA is implemented for the numerical treatment of the VP-HBM dynamics by fluctuating the pulse shape adjustment terms,external forcing factor and damping coefficient with fixed ventricular contraction period.To perform the correctness of the present scheme,the obtained numerical results through the designed ANN-PSO-ASA will be compared with the Adams numerical method.The statistical investigations with larger dataset are provided using the“mean absolute deviation”,“Theil’s inequality coefficient”and“variance account for”operators to perform the applicability,reliability,and effectiveness of the designed ANNs-PSO-ASA scheme for solving the VP-HBM.

基于物理信息神经网络的电磁场计算方法

物理信息神经网络(PINNs)将偏微分方程(PDEs)及其定解条件编码进网络中,使PDEs残差最小化的同时逼近定解条件,实现PDEs的求解。由于电磁场计算时存在局部高梯度问题、含源方程引发的训练困难问题和高对比系数界面识别问题等,PINNs在用于电磁场方程求解时训练效率低、计算精度不高,因而目前应用较少。该文对于PINNs在电磁场中的训练困难问题进行理论分析,提出了针对电磁场PDEs形式和神经网络架构的修改方法,实现了基于PINNs的静电场和稳恒磁场求解,计算结果准确性较好。将该方法推广到方程更加复杂的频域涡流场求解中,求解结果表明PINNs可以在复杂的频域方程上保证良好的精度。该研究工作为实现电磁场快速计算提供了新思路。

宁波陆埠镇平原河网水系结构对水动力影响研究

平原河网的水动力条件对维持区域生态平衡和生物多样性具有关键作用,而水系结构直接关系到河网的连通性和流动性。以宁波陆埠镇河网为例,在概括水系结构现状的基础上,基于浅水方程组理论,构建二维河网水动力模型,通过实测数据对模型进行验证。通过水系连通方式构造3种河网结构,模拟分析2种流动条件下,河网中水体流动的规律。结果表明:洋溪河在流入和流出条件下,流速最小值分别增至0.102 m·s^(-1)和0.154 m·s^(-1),白鹤桥横河、孙家横河、沿西河和周家湾浦的流速也有不同程度的增加,仓前河部分河段出现水体几乎不流动现象,说明水系连通对改善河网水动力有积极的作用,如果水系过度连通,在水流交汇的局部区域就可能出现水体滞留现象。因此,在实施水系连通工程前应该做好评估工作。

基于神经网络的弹性力学位移计算方法研究

常规的弹性力学问题是以有限元方法为主进行数值求解的,网格划分对计算结果影响较大,从而在求解力学问题时存在一定的局限性。神经网络作为一种通用数值逼近器,可以用来实现对弹性力学偏微分方程的求解。本文建立了基于物理约束的神经网络偏微分方程求解模型,以常见的二维平面压缩计算为例,实现了该方法的求解和验证,为智能化弹性力学求解提供了新思路与方法。

Particle dispersion modeling in ventilated room using artificial neural network

Due to insufficiency of a platform based on experimental results for numerical simulation validation using computational fluid dynamic method(CFD) for different geometries and conditions,in this paper we propose a modeling approach based on the artificial neural network(ANN) to describe spatial distribution of the particles concentration in an indoor environment.This study was performed for a stationary flow regime.The database used to build the ANN model was deducted from bibliography literature and composed by 261 points of experimental measurement.Multilayer perceptron-type neural network(MLP-ANN) model was developed to map the relation between the input variables and the outputs.Several training algorithms were tested to give a choice of the Fletcher conjugate gradient algorithm(TrainCgf).The predictive ability of the results determined by simulation of the ANN model was compared with the results simulated by the CFD approach.The developed neural network was beneficial and easy to predict the particle dispersion curves compared to CFD model.The average absolute error given by the ANN model does not reach 5%against 18%by the Lagrangian model and 28% by the Euler drift-flux model of the CFD approach.

Sediment transport following water transfer from Yangtze River to Taihu Basin

To meet the increasing ：need of fresh water and to improve the water quality of Taihu Lake, water transfer from the Yangtze River was initiated in 2002. This study was performed to investigate the sediment distribution along the river course following water transfer. A rainfall-runoff model was first built to calculate the runoff of the Taihu Basin in 2003. Then, the flow patterns of river networks were simulated using a one-dimensional river network hydrodynamic model. Based on the boundary conditions of the flow in tributaries of the Wangyu River and the water level in Taihu Lake, a one-dimensional hydrodynamic and sediment transport numerical model of the Wangyu River was built to analyze the influences of the inflow rate of the water transfer and the suspended sediment concentration （SSC） of inflow on the sediment transport. The results show that the water transfer inflow rate and SSC of inflow have significant effects on the sediment distribution. The higher the inflow rate or SSC of inflow is, the higher the SSC value is at certain cross-sections along the ：river course of water transfer. Higher inflow rate and SSC of inflow contribute to higher sediment deposition per kilometer and sediment thickness. It is also concluded that a sharp decrease of the inflow velocity at the entrance of the Wangyu River on the river course of water transfer induces intense sedimentation at the cross-section near the Changshu hydro-junction. With an increasing distance from the Changshu hydro-junction, the sediment deposition and sedimentation thickness decrease gradually along the river course.

ESTIMATION OF ATTRACTION DOMAIN AND EXPONENTIAL CONVERGENCE RATE OF CONTINUOUS FEEDBACK ASSOCIATIVE MEMORY

Analytical techniques and Liapunov method were used for the estimation of the attraction domain of memory patterns and local exponential stability of neural networks. The results were used to design efficient continuous feedback associative memory neural networks. The neural network synthesis procedure ensured the gain of large exponential convergence rate without reduction of the attraction domain.

基于深度学习的水体生境图像分类与质量评价——以长三角一体化先行启动区为例

【目的】面对长三角一体化地区生态高质量修复与智能化监测的更高诉求,开展基于深度学习的水体生境质量评价,旨在探索风景园林数字技术的前沿领域,为长三角一体化地区的水体生态修复与生态绿色发展提供信息化与智能化的技术支撑。【方法】采用基于深度学习的图像分类方法,通过卷积神经网络(convolutional neural networks,CNN)训练,实现大范围、大批量的水体生境卫星图像智能识别、分类与评价。【结果】构建了长三角一体化先行启动区水体生境卫星图像数据集,训练了水体生境质量评价深度学习模型,对研究范围内全域的水体生境进行了高精度、自动化的生境质量评价。【结论】深度学习模型能够长时序、大范围地对水体生境进行质量评价,为水体生境的修复实践提供技术支撑,未来可实现对长三角一体化示范区水体生境质量的跟踪监测。

感潮河网口门处舾装码头工程对引排水的影响

与内陆地区相比,城市感潮河网在防洪排涝和水环境保护等方面面临较大压力,沿江水闸利用潮差进行内河引排水是提升感潮河网防洪排涝能力、改善水环境的有效途径,但临江工业的发展与感潮河网引排水容易形成矛盾。以南通农场闸闸口附近的惠生重工3#舾装码头新建工程为背景,采用数学模型研究舾装码头工程对感潮河网引排水的影响,提出一种将船舶概化成活动闸门的船舶阻水作用分析的等效处理方法,并验证其合理性。研究认为舾装码头工程口门拓宽疏浚措施对感潮河网排水排涝有利、对感潮河网引水不利,口门泊船对感潮河网引排水影响甚微。

基于LeNet-5的试卷手写分数无分割识别方法

研究了LeNet-5网络识别试卷计分栏手写分数的无分割识别算法。先将高拍仪获取的试卷计分栏图像运用ROI区域选择和连通域处理等图像处理算法将其中手写的各分数截取出来,形成单独的图像,然后将图像归一化,并将格式转化为csv文件,后把这些csv数据输入训练好的LeNet-5网络实现识别。在训练LeNet-5网络时,不断修改该网络的batch、learning rate、epoch和权重数量等参数,提高识别效果。对于无分割字符识别,卷积神经网络需进行100分类(即0-99,不考虑100分的试卷),将全连接层神经元数量设置为100。为了提高试卷识别分数效果,将试卷计分栏各题号的理论总分也输入卷积神经网络,并将该数据作为误分类的一个筛选,提高了分类准确性。采用以上系列方法,对500份学生试卷的手写分数识别,正确率达93.20%。将以上方法应用于试卷分数统计系统,可以大幅度提高试卷分数的统计速度并降低教师的劳动强度。

基于“分段-组合”残差神经网络的超声速氢气零维点火计算方法

受限于发动机燃烧数值模拟需要长时间超级计算机运行的问题,发展了一种基于“分段-组合”残差神经网络的氢气零维点火计算方法。以氢气零维点火算例为基础,基于自主研发的高超声速内外流耦合数值模拟软件AHL3D构建数据集。数据集中输入变量为超声速工况下的温度、压强及8种组分质量分数的初始状态值,输出变量为3000个时刻点的温度、压强及8种组分质量分数状态值。构建了一种“分段”训练、“组合”预测的残差神经网络框架。算法首先将高维输入数据进行降维训练,再将“分段”模型预测后的参数冻结形成“组合”模型。与氢燃料直接计算相比,实验结果表明“分段-组合”残差神经网络可显著提升计算效率,对于11组分29反应的反应动力学模型可获得9.13倍的计算加速比,均方根误差降到了7.85×10^(-5),氢燃料参数的预测精度都高于98%,计算效率及精度优于现有的神经网络燃烧计算方法。

平原河网城市极端暴雨洪涝风险数值模拟方法研究

受全球气候变化和人类活动影响,极端暴雨事件频繁发生,2021年郑州“7·20”特大暴雨更是敲响了城市洪涝风险的警钟。针对平原河网城市极端暴雨洪涝风险数值模拟方法问题,提出流域性模型与研究区域精细化模型耦合嵌套的建模方法,结合郑州市降雨量与当地雨水情特征给定边界条件。基于该模拟方法,构建了水文—水动力一维、二维耦合洪涝分析模型,并以无锡市为例,模拟不同倍比郑州市暴雨量级下洪涝过程。从淹没风险分布特性和河网水位变化规律两方面,揭示了平原河网城市遭遇极端暴雨的洪涝灾害特征,研究成果可为平原河网城市极端暴雨数值模拟提供方法支撑,并为城市防洪和发展规划提供理论依据,最大限度减轻灾害损失。

简化护卫舰的尾流场优化研究

小型舰船在航行中,机库后方产生的不稳定回流区会影响舰载机在甲板上的作业。为了改善舰面流场,保障舰载机在甲板上的安全起降,基于改进的简化护卫舰外形(MSFS),首先结合计算流体力学和人工神经网络方法,预测不同的机库长度L和宽度W对应的回流区大小,建立两种神经网络下的回流区响应面,然后在该响应面上使用粒子群算法得到优化的舰船结构。结果表明:回流区长度随着L和W的变化呈现出明显的非线性变化;与数值模拟结果相比,误差逆传播神经网络得到的回流区长度最大相对误差约为0.9%,径向基函数神经网络得到的最大相对误差约为2.2%;两种神经网络下得到的预测回流区响应面在整体上较为相似,且在两个预测响应面上优化得到的最小回流区长度都为1.93H,和数值模拟结果之间的相对误差约为0.5%。

STUDY AND APPLICATION OF STEADY FLOW AND UNSTEADY FLOW MATHEMATICAL MODEL FOR CHANNEL NETWORKS

Based on the Preissmann implicit scheme for the one-dimensional Saint-Venant equation, the mathematical model for one-dimensional fiver networks and canal networks was developed and the key issues on the model were expatiated particularly in this article. This model applies the method of three-steps solution for chaunel-junction-channel to simulate the river networks, and the Gauss elimination method was used to calculate the sparse matrix. This model was applied to simulate the tree-type irrigation canal networks, complex looped channel networks and the Lower Columbia Slough networks. The results of water level and discharge agree with the data from the Adlul and field data. The model is proved to be robust for simulating unsteady flows in river networks with various degrees of complex structure. The calculated results show that this model is useful for engineering applications in complicated river networks. Future research was recommended to focus on setting up ecological numerical model of water quality in river networks and canal networks.

A 2-D HYDRODYNAMIC MODEL FOR THE RIVER,LAKE AND NETWORK SYSTEM IN THE JINGJIANG REACH ON THE UNSTRUCTURED QUADRANGLES

The main river, the Dongting Lake and river networks in the Jingjiang reach of the Yangtze River constitute a complex water system, for which a full 2-D hydrodynamic model is established instead of the traditional 1-D or compound models for simulation of such complex systems, based on the latest developments of computer technologies and numerical methods. To better handle irregular boundaries and keep the computation cost well in a reasonable limit, unstructured grids of moderate scale are used. In addition, a dynamic boundary tracking method is proposed to simulate variable flow domains at different floods, especially, when the moderate scale gird can not describe flows in narrow river-network channels at low water levels. The t9 semi-implicit method and the Eulerian-Lagrangian Method （ELM） are adopted, which make the model unconditionally stable with respect to the gravity wave speed and Courant number restrictions. Properties and efficiency of the model are discussed, and it is concluded that the new model is robust and efficient enough for the simulation of a big, complex water system. Validation tests show that the simulation results agree well with field data. It takes about 0.96 h to complete the computation of a 76 d flood, which indicates that the model is efficient enough for engineering applications.

基于卷积注意力网络的网格质量评价方法

将深度卷积神经网络引入网格质量评价问题有望代替网格工程师完成繁杂的网格质量评价工作,节省计算流体力学数值模拟的人力成本,但现有方法的准确率和效率仍需要提高.因此,本文提出一种基于卷积注意力网络的网格质量评价方法.首先,本文提出在轻量级卷积神经网络模型中嵌入通道注意力的方式以同时提升准确率和效率;其次,设计了一个神经网络模型CANet用于网格质量评价任务;最后,通过Z-Score标准化对数据进行预处理,解决输入数据分布不一致的问题,以进一步提升准确率.实验结果表明,与现有方法相比,CANet可以达到更优的准确率97.06%,并且在效率上也有至少34.9%的提升.

Reliability Analysis of Electrical System of Computer Numerical Control Machine Tool Based on Bayesian Networks

The core of computer numerical control(CNC) machine tool is the electrical system which controls and coordinates every part of CNC machine tool to complete processing tasks, so it is of great significance to strengthen the reliability of the electrical system. However, the electrical system is very complex due to many uncertain factors and dynamic stochastic characteristics when failure occurs. Therefore, the traditional fault tree analysis(FTA) method is not applicable. Bayesian network(BN) not only has a unique advantage to analyze nodes with multiply states in reliability analysis for complex systems, but also can solve the state explosion problem properly caused by Markov model when dealing with dynamic fault tree(DFT). In addition, the forward causal reasoning of BN can get the conditional probability distribution of the system under considering the uncertainty;the backward diagnosis reasoning of BN can recognize the weak links in system, so it is valuable for improving the system reliability.

感潮河网水环境容量数值计算

对感潮河网水环境容量和污染物允许排放量的概念进行了分析 ,考虑了感潮河网水环境容量的时空动态变化及其影响因素 ,提出了基于感潮河网水质模型的水环境容量数值计算方法 .该方法不仅反映了在感潮河网地区水流运动复杂、流态不稳定、流向多变的特点 ,而且反映了感潮河网水环境容量受边界水质变化、河段水质目标不一致、水利工程运行等多种因素影响下的时空变化规律 .该方法经过实例计算验证 ,效果良好 .