CN-121980689-A - Method for optimizing UGKWP method based on mixed precision data to verify spacecraft design

CN121980689ACN 121980689 ACN121980689 ACN 121980689ACN-121980689-A

Abstract

The invention discloses a method for verifying spacecraft design based on a mixed precision data optimization UGKWP method, which relates to the field of spacecraft design, and comprises the steps of S1, taking shape data after spacecraft design as input data of a UGKWP method, performing simulation calculation based on the input data by a S2 and UGKWP method to obtain theoretical values of an external flow field of a spacecraft in a multi-scale environment, and verifying performance of the spacecraft based on the theoretical values obtained in the S2 to guide the spacecraft shape design optimization in the next step, wherein the UGKWP method reduces the occupation amount of unit macroscopic quantity and microscopic physical quantity to a memory in the simulation process through a mixed precision data optimization and precision compensation algorithm. Based on the characteristics of different physical quantities and different using modes in a calculation program, the invention stores microscopic physical quantities in different low-precision data forms and designs different precision compensation algorithms to ensure conservation of the physical quantities and stable promotion of calculation.

Inventors

YU HANG
WU JUNLIN
Peng Aoping
LI LANGQUAN

Assignees

中国空气动力研究与发展中心超高速空气动力研究所

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. A method for optimizing UGKWP a method based on hybrid accuracy data to verify a spacecraft design, comprising: s1, taking shape data after spacecraft design as input data of UGKWP method; S2, carrying out simulation calculation based on input data by a UGKWP method to obtain a theoretical value of an external flow field of the spacecraft in a multi-scale environment; s3, verifying performance of the spacecraft based on the theoretical value obtained in the S2 so as to guide the appearance design optimization of the spacecraft in the next step; The UGKWP method reduces the occupation amount of the unit macroscopic quantity and microscopic physical quantity to the memory in the simulation process through a mixed precision data optimization and precision compensation algorithm.
2. The method for optimizing UGKWP the method based on mixed precision data to verify a spacecraft design of claim 1, wherein the mixed precision data comprises bool data, int16 data, FP16 data, and FP64 data.
3. The method of optimizing UGKWP a method based on hybrid precision data to verify a spacecraft design of claim 1, wherein in the UGKWP method, the unit macro-scale update form optimized based on hybrid precision data is characterized by the following formula: in the above-mentioned method, the step of, And Macroscopic amounts of the i-cell at times n and n +1 respectively, For this purpose the volume of the unit, Representing the area of the grid interface, In order to balance the flux, Representing the resolved transport flux(s), Is the net particle flux in the time step, and Wherein m is, And e represents mass, velocity and energy respectively, Represents the physical quantity carried by the microscopic particles expressed in the form of low precision of FP16, an P is shorthand for particle, Indicating the moment at which the movement of the particles ends, N (i) indicating all the grid planes of the i cell, Indicating the moment before the particle movement, FP16 indicates the quantity in a low-precision form; For the particle flux term in time steps and corrected, and , wherein, Representing the microscopic physical information carried by the equivalent particles, and Eqv denotes an equivalent.
4. The method for hybrid precision data based optimization UGKWP method to verify a spacecraft design of claim 1, wherein the microscopic physical quantities consist essentially of a particle time of flight prtTf, a particle mass prtMass, a particle velocity prtVloc, and a particle position prtPos; In the UGKWP method, two data of the particle flight time prtTf and the particle quality prtMass are respectively optimized through a data fixed-point strategy based on high-precision correction; in the UGKWP method, the particle position prtPos data is optimized by a scale-independent relative coordinate transformation method; In the UGKWP method, the particle velocity prtVloc data is optimized by a low-precision dynamic quantization method.
5. The method for hybrid precision data based optimization UGKWP method to validate a spacecraft design of claim 4, wherein said high precision corrected data localization strategy is: Based on a data localization mode, storing two data of particle flight time prtTf and particle quality prtMass by utilizing low-precision data int16, and introducing a group of accompanying correction arrays of high-precision FP64 to correct the stored localized low-precision particle quality prtMass data; wherein the array length of the accompanying correction array is the same as the grid number of the calculation domain, and the accompanying correction array comprises three parts of mass correction, momentum correction and energy correction.
6. The method of hybrid precision data based optimization UGKWP method to validate spacecraft designs of claim 5, wherein the data localization strategy for particle time of flight prtTf is characterized by: Dynamically calculating quantization factor 2 m for quantization according to maximum value of original high-precision FP64 array so as to make fixed-point form after quantized particle flight time prtTf The following relationship should be satisfied: In the above-mentioned method, the step of, Is the maximum value that can be represented by the low-precision data int16, For the current step of time, 2 m is the quantization factor that dynamically calculates the maximum value of the original high-precision FP64 array, and 。
7. The method for hybrid precision data based optimization UGKWP method to validate a spacecraft design of claim 6, wherein said data localization strategy of particle mass prtMass is: After the original high-precision data is quantized and amplified by using a quantization factor 2 m , the integer part of the original high-precision data is stored in a semi-precision quality array in the form of int16, and the momentum and energy corresponding to the truncated decimal part are accumulated in a high-precision correction array of a corresponding grid to be used as quality correction, momentum correction and energy correction respectively; Wherein, since the accompanying correction array is independently involved in the calculation of the particle flux and the statistics of the macroscopic quantity, and updated in each iteration step, after the quantization factor 2 m is updated according to the particle mass extremum, the low-precision mass array of the collision-free P left particle is updated according to the new quantization factor, and in the updating process, the truncation error of the P left particle due to precision conversion and the truncation error of the new sampling particle P samp in the quantization process are accumulated in the accompanying correction array.
8. The method for validating spacecraft design based on hybrid precision data optimization UGKWP method as claimed in claim 7, wherein after updating of the companion correction array is completed, the mass, momentum and energy recorded in the companion correction array in each grid are equivalent to one companion simulation particle P eqv , and the transport time of P eqv is sampled as with other actual simulation particles; Wherein the mass and velocity of P eqv are derived from the macroscopic quantities accumulated in the accompanying correction array, and the correction flux generated across the grid interface during the motion of P eqv Is regarded as free transport flux Is involved in the calculation of the net flux of the unit and in the statistics of the partial macroscopic quantity of the particles In this case, the non-collided P eqv is included in the statistical category.
9. The method for verifying a spacecraft design based on the hybrid precision data optimization UGKWP method of claim 4, wherein the scale-independent relative coordinate transformation method is to store particle position information by using low precision FP16 data, and decompose the particle position information into a high precision unit center absolute coordinate and a low precision particle relative coordinate and record the two coordinates respectively; in the particle tracking process, the low-precision particle relative coordinates are obtained through the following calculation : In the above-mentioned method, the step of, Is the absolute coordinates of the center of the unit with high precision, Representing the projection vector of the control volume, Absolute coordinates of the particle position; When the position relation of the particles in the unit grids is changed, the relative coordinates of the corresponding particles need to be updated according to the central absolute coordinates and the projection quantity of the new grids, and if the particles collide, low-precision particle data corresponding to the collision particles are removed from the memory.
10. The method for validating spacecraft design based on hybrid precision data optimization UGKWP as defined in claim 4, wherein said low precision dynamic quantization method is to split the micro-particle velocity into unit macro-velocity and sampling velocity and store them separately; The macro speed is always kept in a double-precision form, the sampling speed is represented by a low-precision FP16 data set which can meet the preset normal distribution characteristic, the low-precision data is obtained by processing the particle speed data in a low-precision quantification mode, and the low-precision quantification refers to the reduction of storage requirements through the cutting off of low-order effective digits of the high-precision data and the preset precision loss; when an accurate velocity value is required, the sampling velocity and the unit macroscopic velocity can be synthesized and restored to a high-precision microscopic particle velocity by the following formula: In the above-mentioned method, the step of, Is the velocity of the micro particles, and the velocity of the micro particles meets the average value of Standard deviation is Is a normal distribution of (c).

Description

Method for optimizing UGKWP method based on mixed precision data to verify spacecraft design Technical Field The invention relates to the field of spacecraft design. More particularly, the present invention relates to a method for optimizing UGKWP methods based on hybrid accuracy data to verify spacecraft designs. Background With the development of aerospace technology, the space domain explored by human beings has been expanded from a traditional dense atmospheric environment to the near space, outer space and even deep space fields. Accordingly, the mission airspace of the aircraft also extends from a traditional single river basin to a multi-scale river basin containing continuous and thin flows. For example, a gliding hypersonic aircraft needs to perform maneuvering tasks in a sub-orbital space where transitional flow and even free molecular flow predominate after traversing continuous and slippery domains in the ascending section, and a reusable carrier undergoes a multi-scale flow evolution process from free molecular flow, to transitional flow, and ultimately to continuous flow as it returns from outer space to the earth's atmosphere. The UGKWP (Unified Gas-KINETIC WAVE-part) method is a Gas dynamic method that focuses on multi-scale flow problem simulation. The UGKWP method requires not only the macro conservation of the recording unit, but also the storage of microscopic information (including mass, velocity, spatial position, etc.) of a large number of analog particles during the calculation. Typically, tens to hundreds of analog particles are contained in each grid cell, and the number of particles is further increased to reduce statistical noise when dealing with low flow problems. This large-scale storage requirement for microscopic particle data places higher demands on the memory capacity of the computing device by the UGKW method. The memory of the CPU server can be flexibly expanded by adding a memory bank, but on a heterogeneous parallel processor represented by a GPU, the memory particles are usually directly welded on a substrate, so that users can hardly expand arbitrarily as required. Therefore, the algorithm has an excessive demand on the memory, which not only limits the application scale, but also occupies more bandwidth to cause performance bottleneck. In the field of scientific computing, numerical operations using double-precision data have become a "real" standard. This convention is developed primarily based on two key factors, namely, firstly, the double-precision floating point number has a very large numerical representation range according to the IEEE754 standard, and secondly, the double-precision floating point number can provide 15-17 decimal significant digits, and the high-precision characteristic can effectively avoid the problems of complex operation and truncation error accumulation generated in the frequent iteration process. This is especially true for UGKWP methods, where UGKWP implements flow field simulation by analyzing distribution functions and simulating particles, and the moment solving process and particle parameters include many small amounts, which can be ensured not to be truncated in the calculation process only by using double-precision data, compared with the conventional macroscopic-amount-based CFD method. However, each double-precision floating point number requires 8 bytes of memory space, which is significantly more costly than single-precision floating point numbers (4 bytes) and half-precision floating point numbers (2 bytes). Therefore, the use of the mixed precision data becomes the most direct mode of the compression algorithm memory. Disclosure of Invention It is an object of the present invention to address at least the above problems and/or disadvantages and to provide at least the advantages described below. To achieve these objects and other advantages and in accordance with the purpose of the invention, a method for optimizing UGKWP a method to verify a spacecraft design based on hybrid accuracy data is provided, comprising: s1, taking shape data after spacecraft design as input data of UGKWP method; S2, carrying out simulation calculation based on input data by a UGKWP method to obtain a theoretical value of an external flow field of the spacecraft in a multi-scale environment; s3, verifying performance of the spacecraft based on the theoretical value obtained in the S2 so as to guide the appearance design optimization of the spacecraft in the next step; The UGKWP method reduces the occupation amount of the unit macroscopic quantity and microscopic physical quantity to the memory in the simulation process through a mixed precision data optimization and precision compensation algorithm. Preferably, the mixed precision data includes a bool data, an int16 data, an FP16 data, and an FP64 data. Preferably, in the UGKWP method, the unit macro-scale update form optimized based on the hybrid precision data is characterized by the following