CN-122021337-A - Unsteady state liquid drop evaporation rate prediction method based on physical information neural network

CN122021337ACN 122021337 ACN122021337 ACN 122021337ACN-122021337-A

Abstract

The invention discloses a non-steady-state liquid drop evaporation rate prediction method based on a physical information neural network, which relates to the technical field of intelligent prediction of fluid mechanics and comprises the following steps of: and carrying out periodic rule analysis on the unsteady state liquid drop evaporation rate prediction, carrying out sectional recording on the pulsation process of the airflow speed, the ambient temperature and the combustion pressure, generating an evaporation change curve, and extracting the phase position and the energy amplitude sequence of each combustion period to form an evaporation period diagram. According to the method, dynamic time synchronization of the evaporation rate of the liquid drops is realized through periodic rule analysis and phase drift identification, time drift caused by error accumulation is eliminated, prediction stability and physical consistency are improved, energy and time response are coordinated through a time resonance list and a rhythm control scheme, numerical resonance is weakened, and the prediction result is ensured to be kept continuous and accurate under the unsteady combustion condition.

Inventors

ZHOU TAOTAO
WEI JIANGJUN
ZHANG YU
QIU LIANG
HUA YANG
WANG TAO
QIAN YEJIAN
ZHUANG YUAN

Assignees

合肥工业大学

Dates

Publication Date: 20260512
Application Date: 20260212

Claims (10)

1. The unsteady state liquid drop evaporation rate prediction method based on the physical information neural network is characterized by comprising the following steps of: performing periodic rule analysis on the unsteady state liquid drop evaporation rate prediction, performing sectional recording on the pulsation process of the airflow speed, the ambient temperature and the combustion pressure to generate an evaporation change curve, and extracting the phase position and the energy amplitude sequence of each combustion period to form an evaporation period diagram; comparing the evaporation change curves of all periods section by section based on the evaporation period diagram, identifying phase offset points and energy mutation points in the evaporation rate curve, extracting key time nodes and generating a phase drift record table; Backtracking a time error source based on a phase drift record table, calculating the energy accumulation state of the offset node, determining a resonance trigger zone, a time lag zone and a time lead zone according to energy response characteristics, and forming a time resonance list according to the resonance trigger zone, the time lag zone and the time lead zone; Constructing a periodic stable control strategy according to the time resonance list, setting a phase buffering sequence, a time yielding threshold value and an energy stable interval, and generating a rhythm control scheme; And executing a rhythm control scheme, and adjusting the data sampling period and the execution rhythm in the evaporation prediction process in real time through a coordinated adjustment mechanism of a reverse time window, an energy absorption area and an airflow cooling curtain, so as to weaken error energy accumulation and inhibit a numerical resonance effect.
2. The method for predicting the evaporation rate of an unstable droplet based on a physical information neural network according to claim 1, wherein the evaporation cycle pattern forming step is as follows: Dynamically recording time sequence characteristics of air flow speed, ambient temperature and combustion pressure in a combustion environment, and collecting transient responses of an air flow speed field, a thermal field and a pressure field in a time sequence to form a continuous air flow speed time sequence, a continuous temperature time sequence and a continuous pressure time sequence; The time series data of the air flow speed, the temperature and the pressure are synchronously arranged, an evaporation response associated data set is constructed, and an evaporation change curve is formed by taking the phase relation of the combustion pressure as a reference; dividing the combustion period into equal time intervals, and calculating a phase position and energy amplitude sequence of an evaporation rate curve by taking a pressure pulsation peak value as a reference to form a liquid drop evaporation response characteristic matrix; the phase position and the energy amplitude sequence are integrated in time sequence, and an evaporation periodic chart is drawn to show the periodic variation rule of the evaporation rate and the energy distribution trend.
3. The method for predicting the evaporation rate of an unstable droplet based on a physical information neural network according to claim 2, wherein the phase drift record table generating step comprises the following steps: taking each combustion period evaporation change curve in the evaporation period diagram as an object, and comparing the sections by sections according to a time sequence, so that adjacent period curves keep a corresponding relationship in the same phase interval and realize time axis alignment; carrying out cycle-by-cycle tracking on the positions of the peaks and the troughs of the evaporation change curve after alignment, identifying phase offset points of the evaporation rate along with time, and determining the offset direction and the offset degree; After the phase offset points are identified, continuously scanning the energy distribution state of the evaporation change curve, calibrating energy mutation points and judging the energy accumulation direction; And carrying out time sequencing and pairing arrangement on the phase shift points and the energy mutation points to generate a phase shift record table so as to express the time sequence shift characteristic of unsteady evaporation.
4. The method for predicting the evaporation rate of the unsteady liquid drops based on the physical information neural network according to claim 3, wherein when a phase drift record table is generated, phase offset points and energy mutation points are paired and arranged according to a time sequence and cycle numbers, an evaporation response structured record is established by taking time as a main axis, and the phase offset direction, the energy change amplitude and the time position of each cycle are recorded.
5. A method for predicting the evaporation rate of an unstable droplet based on a physical information neural network according to claim 3, wherein the time resonance list forming process is as follows: performing backtracking arrangement on time nodes in the phase drift record table, and remapping phase offset points and energy mutation points onto an evaporation change curve time axis according to a periodic sequence to form a continuous offset sequence; After the time path backtracking is completed, calculating the energy accumulation state of each offset node, and distinguishing an energy accumulation node from an energy release node according to the energy change trend; Identifying a resonance triggering area, a time delay area and a time advance area according to the energy response characteristics, and continuously calibrating the positions and the ranges of the resonance triggering area, the time delay area and the time advance area on a time axis; And sorting the identified starting time, ending time, energy accumulation direction and phase response relation of each time interval to form a time resonance list so as to represent the time energy coupling mode of the evaporation rate.
6. The method for predicting the evaporation rate of an unstable droplet based on a physical information neural network according to claim 5, wherein in the process of forming a time resonance list, the formation rule of a periodic error is determined by comparing the arrangement sequence of a resonance trigger area with an energy accumulation trend, and the time span of a time lag area and a time lead area is combined to analyze the time stability interval of an evaporation response as the basis of setting a phase buffering sequence and a time yield threshold in the subsequent rhythm control.
7. The method for predicting the evaporation rate of an unstable droplet based on a physical information neural network according to claim 5, wherein the step of generating the cadence control scheme is as follows: Based on a resonance triggering area, a time delay area and a time advance area in the time resonance list, establishing a phase buffering sequence according to an energy accumulation direction and a phase shift trend; Comparing the response amplitude and the energy change rate of each time interval according to the phase buffering sequence, and setting a time yielding threshold to limit the priority relation of energy transfer; Carrying out partition analysis on the energy distribution state according to the time yielding threshold and the phase buffering sequence, and determining an energy stability interval to maintain the balance state of time and energy; And integrating the phase buffering sequence, the time yielding threshold and the energy stabilizing interval to generate a rhythm control scheme to adjust the time rhythm and energy response balance of the evaporation rate prediction process.
8. The method for predicting the evaporation rate of an unstable droplet based on a physical information neural network according to claim 7, wherein in the process of establishing the phase buffering sequence, a time interval in which the energy accumulation direction is identical and the phase shift direction is opposite is set as a phase delay buffer, a time interval in which the energy release direction is identical to the phase shift direction is set as a phase advance buffer, and the evaporation rate prediction maintains dynamic balance between the time response and the energy transfer in successive periods by alternate arrangement of time delay and advance.
9. The method for predicting the evaporation rate of an unsteady droplet based on a physical information neural network according to claim 7, wherein the step of executing a cadence control scheme to adjust the data sampling period and the execution cadence in the evaporation prediction process in real time by a coordinated adjustment mechanism of a reverse time window, an energy absorption region and an air flow cooling curtain is as follows: Establishing a reverse time window mechanism according to a phase buffering sequence and a time yield threshold set in the rhythm control scheme, and adjusting an evaporation prediction time axis through time backtracking to balance the phase progress of the time sequence; an energy absorption region is constructed on the basis of realizing time backtracking adjustment in a reverse time window, and energy accumulation in the evaporation rate prediction process is locally buffered and dissipated to maintain energy transfer balance; constructing an airflow cooling curtain according to the energy distribution state of the energy absorption area, and controlling heat energy transfer in the evaporation rate prediction process through the synergistic effect of the airflow and the heat exchange process; under the condition that the reverse time window, the energy absorption area and the airflow cooling curtain are operated cooperatively, the data sampling period and the execution rhythm are adjusted in real time so as to weaken error energy accumulation and inhibit numerical resonance effect.
10. The method for predicting the non-steady-state liquid drop evaporation rate based on a physical information neural network according to claim 9, wherein the coordinated adjustment of a reverse time window, an energy absorption region and an air flow cooling curtain is performed through the linkage of time feedback and energy balance, the reverse time window dynamically adjusts the time backtracking amplitude according to the energy accumulation trend, the energy absorption region forms a buffer layer in the energy transmission process, and the air flow cooling curtain controls the liquid drop surface heat exchange intensity through air flow disturbance.

Description

Unsteady state liquid drop evaporation rate prediction method based on physical information neural network Technical Field The invention relates to the technical field of intelligent prediction of fluid mechanics, in particular to an unsteady state liquid drop evaporation rate prediction method based on a physical information neural network. Background The unsteady state liquid drop evaporation rate prediction based on the physical information neural network refers to that under the strong unsteady state spray combustion environment, physical rules such as energy conservation, mass conservation, interface transmission and the like in the liquid drop evaporation process are directly embedded into a training structure of the neural network, and multi-source heterogeneous data from experimental observation, numerical simulation and sensing monitoring are subjected to high-dimensional cleaning, noise reduction and feature extraction by combining a big data processing technology, so that the network not only depends on data driving learning, but also can automatically meet the constraint of a physical equation in the training process, and therefore, the high-precision prediction of the evaporation rate is realized under complex transient conditions such as airflow pulsation, temperature fluctuation, multi-liquid drop interaction and the like. According to the method, the residual terms of the evaporation kinetic equation and the boundary consistency constraint are introduced into the loss function, so that the network output result is balanced among data fitting, physical consistency and statistical robustness, the limitation that the traditional empirical model can not accurately reflect unsteady heat and mass transfer behaviors is broken through, and an intelligent prediction approach integrating big data driving and physical driving characteristics is provided for combustion simulation and spray optimization. Simulation and spray optimization provide an intelligent prediction approach with both data driving and physical driving characteristics. The prior art has the defects that in the prior art, the liquid drop evaporation rate prediction based on a physical information neural network is stable in a multi-hypothesis time stepping process, but under the periodic combustion pulsation condition, the air flow, the temperature and the pressure can show periodic fluctuation, so that the evaporation rate continuously oscillates with time. When the network predicts in successive time steps, small time errors and phase shifts accumulate continuously in the periodic cycle, which is prone to numerical resonance phenomena, resulting in non-physical amplification or attenuation of the evaporation rate of the predicted output. Along with the continuous progress of the combustion period, error energy is gradually overlapped, and finally a predicted sequence is completely deviated from a real change trend after a plurality of periods, abnormal conditions such as waveform distortion, rhythm drift and even energy inversion appear, and the real evaporation rule of liquid drops in an unsteady combustion environment is difficult to accurately reflect, so that the physical consistency and engineering usability of a model are reduced. The above information disclosed in the background section is only for enhancement of understanding of the background of the disclosure and therefore it may include information that does not form the prior art that is already known to a person of ordinary skill in the art. Disclosure of Invention The invention aims to provide an unsteady state liquid drop evaporation rate prediction method based on a physical information neural network, so as to solve the problems in the background art. In order to achieve the purpose, the invention provides the following technical scheme that the unsteady state liquid drop evaporation rate prediction method based on the physical information neural network comprises the following steps: performing periodic rule analysis on the unsteady state liquid drop evaporation rate prediction, performing sectional recording on the pulsation process of the airflow speed, the ambient temperature and the combustion pressure to generate an evaporation change curve, and extracting the phase position and the energy amplitude sequence of each combustion period to form an evaporation period diagram; comparing the evaporation change curves of all periods section by section based on the evaporation period diagram, identifying phase offset points and energy mutation points in the evaporation rate curve, extracting key time nodes and generating a phase drift record table, and providing basic data for subsequent time error analysis; Backtracking a time error source based on a phase drift record table, calculating the energy accumulation state of the offset node, determining a resonance trigger zone, a time delay zone and a time advance zone according to energy response characteristics, formin