The physical information neural networks(PINNs)are an important tool for solving partial differen-tial equations.In the numerical solution of partial differential equations,equations with discontinuous solutions are currently a research challenge.The PINNs and their existing improved algorithms often cannot capture the characteristics of discontinuities.However,many discontinuous problems need to be considered in the field of fluid mechanics.In response to the shortcomings of the PINNs in handling equations with discontinuous solu-tions,an adaptive gradient-annihilated PINNs(AGA-PINNs)method was proposed to solve the Burgers equa-tions and the Allen-Cahn equations with discontinuous solutions.The gradient related weight functions were uti-lized to optimize the loss function,and the training points were dynamically adjusted based on the current re-sidual distribution during the training process to gradually enhance the model's learning ability in discontinuous regions.The experimental results show that,the AGA-PINNs method significantly improves the accuracy of pre-dicted physical quantities compared to the traditional PINNs,the gradient-enhanced physics-informed neural networks(gPINNs),the gradient-annihilated physics-informed neural networks(GA-PINNs)method,and the high-order deep Galerkin(DG)method.In solving the Burgers equation and the Allen-Cahn equation,the mean square errors decrease by approximately 1 order of magnitude,accurately reproducing the characteristics of shock waves in the Burgers equation and the phase separation phenomena in the Allen-Cahn equation.
Core analysis is the cornerstone of reservoir evaluation.The digital rock physics(DRP)enables the quantitative characterization of reservoir microstructures and flow properties through 3D imaging and numerical simulation.The evolution of DRP paradigms was reviewed from the 1st-generation"What you see is what you get"(geometric characterization)and the 2nd-generation"What you calculate is what you get"(numerical simulation),to the emerging 3rd-generation"What you learn is what you get"(intelligent prediction).The 1st 2 paradigms face 2 core bottlenecks:the difficulty in determining the representative elementary volume(REV)and the prohibitive computational costs of numerical simulations.The progress of the 3rd-generation intelligent prediction paradigm was systematically summarized,with the focus on structural properties,single-phase per-meability,and multiphase flow properties(relative permeability,capillary pressure,and wettability).Centered on deep learning,this paradigm offers revolutionary pathways to overcome traditional limitations:on one hand,the generative adversarial networks(GANs)and the super-resolution(SR)techniques were employed for data-driven reconstruction to mitigate REV challenges;on the other hand,the efficient surrogate models,such as the convolutional neural networks(CNNs),were constructed to accelerate property prediction by orders of magnitude.Finally,the current challenges regarding data dependence,model generalization,and physical inter-pretability were discussed to outlines future directions,including integrating physical mechanisms into AI mod-els(physics-informed AI),promoting multi-scale data fusion,and constructing reservoir"digital twins"to support the development of smart oilfields.
The non-Fourier heat transfer characteristics of couple stress fluid in a microchannel were investiga-ted,and effects of the couple stress parameters,the Hartmann number,the Joule heating effect and the 2-phase lag time parameters on the temperature distributions and heat transfer characteristics of the fluid were analyzed through establishment of the relevant governing equations combined with the non-Fourier heat transfer model.The results show that,the above parameters significantly influence the fluidity and heat diffusion char-acteristics of the couple stress fluid.The increase of the couple stress parameter aggravates the temperature gra-dient;the Hartmann number enhances the magnetic field confinement and inhibits the heat conduction;the Joule heating effect promotes the temperature gradient and highlights the non-Fourier effect;and the change of the biphasic hysteresis time parameter has a significant modulation effect on the temperature peak and the re-sponse speed.
The modified Allen-Cahn equation with the mean curvature source term can effectively model curva-ture-driven physical processes.However,the nonlinear term and gradient magnitude make it difficult to design efficient numerical schemes.An efficient numerical scheme was proposed for solving the equation.Based on the Strang operator splitting method,the original equation was discretized in time into three subproblems:the non-linear equation was solved analytically;the mean curvature equation was discretized using the second-order Runge-Kutta method combined with central differences to establish a fully discrete explicit scheme,while the heat equation was discretized with the Crank-Nicolson method and solved via the alternating direction implicit(ADI)fast scheme.Theoretical analysis shows that,the proposed scheme achieves second order convergence accuracy.Finally,numerical experiments were presented to validate the convergence rate and effectiveness of the scheme.
With the widespread implementation of zero-output transformation of low-pressure cylinders in heat-ing units,the low-pressure cylinders of steam turbines often need to operate at extremely low flow rates.To get a deeper understanding of the flow behavior under this working condition,the low-pressure cylinder of a steam turbine in a certain power plant was studied,a numerical calculation model for the final-stage flow channel was constructed,and the structure of the final-stage flow field and the temperature changes on the surface of the moving blades were inspected through simulation analysis of the flow state and aerodynamic performance in the cylinder under variable working conditions,especially at low flow rates.The research results indicate that,at low flow rates,the gas separation and backflow will occur in the final stage of the low-pressure cylinder.The separation initially occurs at the moving blade root and gradually spreads to the blade top as the flow rate de-creases.Local vortices emerge in the channel,and a negative attack angle appears at the moving blade inlet,which significantly hinders the normal steam flow.When the load drops to 15％THA,a local high-temperature zone will appear at the steam outlet edge on the top of the last stage stator blade,showing the blower heating effect.As the flow rate further decreases,the high-temperature area will keep expanding and the maximum temperature will continue to rise.When the load decreases to 10％THA,the maximum surface temperature of the moving blade will rise by 40.29％compared with the rated working condition.This research provides a refer-ence basis for the safe operation of the low-pressure cylinder of the heating unit after the zero-output transfor-mation.
