CN-121997706-A - Method and system for supplementing soft X-ray photon counting rate through deep learning based on physical driving

CN121997706ACN 121997706 ACN121997706 ACN 121997706ACN-121997706-A

Abstract

The application provides a method and a system for supplementing soft X-ray photon counting rate by deep learning based on physical driving, wherein the method comprises the steps of inputting preprocessed measurement data into a trained photon counting rate interpolation model and outputting the supplemented photon counting rate; the photon counting rate interpolation model comprises a first auxiliary sub-network, a second auxiliary sub-network and a main network, wherein the first auxiliary sub-network is used for converting an input ion abundance ratio into SWCX scale factors, the second auxiliary sub-network is used for converting an input geomagnetic index into neutral hydrogen density, the first auxiliary sub-network and the second auxiliary sub-network comprise 1 TEBs and 1 fully-connected layer, and the main network comprises 3 one-dimensional convolution layers, two TEBs and two fully-connected layers which are sequentially connected. The method has the advantages that the photon counting rate is estimated more comprehensively and accurately, the complement result is more accurate in value, and the physical meaning of the complement result is constrained to accord with the radiation theory.

Inventors

Ou Feiyang
LI DALIN
SUN TIANRAN
ZHANG YINGJIE
WANG RONGCONG

Assignees

中国科学院国家空间科学中心

Dates

Publication Date: 20260508
Application Date: 20251222

Claims (9)

1. A method for deep learning complement soft X-ray photon count rate based on physical driving, comprising: Inputting the preprocessed measurement data into a trained photon counting rate interpolation model, and outputting the completed photon counting rate; the photon counting rate interpolation model comprises a first auxiliary sub-network, a second auxiliary sub-network and a main network; the first auxiliary sub-network is used for converting the input ion abundance ratio into SWCX scale factors; the second auxiliary sub-network is used for converting the input geomagnetic index into neutral hydrogen density; The first auxiliary sub-network and the second auxiliary sub-network comprise 1 transducer encoder block and 1 full connection layer; The main network comprises 3 one-dimensional convolution layers, two transducer encoder blocks and two full connection layers which are connected in sequence.
2. The method for depth learning and supplementing soft X-ray photon counting rate based on physical driving according to claim 1, wherein the 3 one-dimensional convolution layers are causal convolution layers, the expansion rates are respectively 1, 2 and 4, and the convolution kernel size of each layer is 3.
3. The physical drive based deep learning complement soft X-ray photon counting rate method of claim 1 wherein the data processing of the two fully connected layers comprises: the first fully connected layer uses the ReLU activation function to compress the embedding of each time step into a low-dimensional potential space; the second fully connected layer performs linear projection for each time step, generating an initial output sequence of the model.
4. The method of claim 1, wherein the measurement data includes a number density of protons in solar wind, a speed of solar wind, an ion abundance ratio, a geomagnetic index, and a photon count rate of 2.5-5.0 keV energy bands.
5. The method of physical drive based deep learning complement soft X-ray photon count rate of claim 4 wherein the calculation of the photon count rate interpolation model comprises: inputting an ion abundance ratio into the first auxiliary subnetwork, outputting SWCX scale factors; Inputting geomagnetic indexes into the second auxiliary sub-network, and outputting neutral hydrogen density; The method comprises the steps of splicing the number density of protons in solar wind, the speed of the solar wind, the ion abundance ratio, the geomagnetic index, the photon count rate of 2.5-5.0 keV energy bands, the SWCX scale factor and the neutral hydrogen density along characteristic dimensions, and broadcasting the spliced solar wind data set identification, the binary mask for distinguishing missing areas and the fixed sine position codes to a common embedding space and combining the spliced solar wind data set identification, the binary mask for distinguishing missing areas and the fixed sine position codes to form a final multivariable input tensor; Inputting the multivariable input tensor into the main network, and outputting an initial complement result, namely a complete photon counting rate time sequence; and intercepting the central part from the initial result sequence according to the set step number to obtain a final output result, namely a time sequence of the photon counting rate after the completion of focusing on the key area.
6. The method of claim 1, wherein the photon count rate interpolation model further comprises an auxiliary physical loss module for calculating a physical theoretical value of photon count rate; the processing procedure of the auxiliary physical loss module comprises the following steps: mapping each sub-data set identifier into an embedded vector, and obtaining sub-data set correlation coefficients through a full connection layer And ; Photon count rate from 2.5-5.0 keV energy bands using one-dimensional convolution layer Extracting constant component in background signal with 2.5-5.0keV energy section ; By means of Subtracting out Obtaining time-varying soft proton signals in background signals of 2.5-5.0keV energy sections ; Calculating the background signal under the target energy section of 0.5-0.7keV by using the following method : ; The soft X-ray photon counting rate of 0.5-0.7 keV energy section is calculated by using the following method : ; Wherein, the Is SWCX scale factors; Is neutral hydrogen density; is solar wind proton density; Is the solar wind speed; Is the thermal velocity; Is the actual line of sight of the satellite.
7. The physical drive based deep learning soft X-ray photon count rate method of claim 6, wherein the training process of the photon count rate interpolation model comprises: Preprocessing a data set of a target sequence comprising the number density of protons in solar wind, the speed of solar wind, the ion abundance ratio, the geomagnetic index, the photon count rate of 2.5-5.0 keV energy bands and the photon count rate of 0.5-0.7 keV energy segments; Dividing the preprocessed data set into a training set and a testing set according to a set proportion; inputting the ion abundance ratio in the training set into the first auxiliary sub-network, and outputting SWCX scale factors; Inputting geomagnetic indexes in the training set into the second auxiliary sub-network, and outputting neutral hydrogen density; The method comprises the steps of splicing the number density of protons in solar wind, the speed of the solar wind, the ion abundance ratio, the geomagnetic index, the photon count rate of 2.5-5.0 keV energy bands, the SWCX scale factor and the neutral hydrogen density along characteristic dimensions, and broadcasting the spliced solar wind data set identification, the binary mask for distinguishing missing areas and the fixed sine position codes to a common embedding space and combining the spliced solar wind data set identification, the binary mask for distinguishing missing areas and the fixed sine position codes to form a final multivariable input tensor; inputting the multivariable input tensor into the main network, and outputting a potential photon counting sequence with a first set step number; inputting the potential photon counting sequence of the first set step number into the auxiliary physical loss module, and outputting the photon counting rate and the auxiliary physical loss after the completion; Combining the numerical main loss of the main network and the auxiliary physical loss of the auxiliary physical loss module into a loss function of the photon counting rate interpolation model; And through a back propagation algorithm, according to the error of the complement value and the true value calculated by the loss function, the weight and the bias of each layer are reversely and iteratively adjusted along the network layer, and the loss function is minimized so as to improve the complement accuracy of the photon counting rate interpolation model.
8. The physical drive based deep learning complement soft X-ray photon count rate method of claim 1 wherein the preprocessing comprises: time alignment and synchronization are carried out on all data sources; removing invalid values of the data and carrying out normalization processing; and sliding the processed data according to a set step length to form a time sequence data set with a fixed window length.
9. A physical drive-based deep learning complement soft X-ray photon count rate system implemented based on the method of any one of claims 1-8, the system comprising: A preprocessing module for preprocessing the measurement data, and And the photon counting rate complementing module is used for inputting the data output by the preprocessing module into the trained photon counting rate interpolating model and outputting the complemented photon counting rate.

Description

Method and system for supplementing soft X-ray photon counting rate through deep learning based on physical driving Technical Field The application belongs to the fields of soft X-ray, solar wind charge exchange, data complement, time sequence and deep learning, and particularly relates to a method and a system for deep learning complement soft X-ray photon counting rate based on physical driving. Background Photon counting rates are well accepted for their high signal-to-noise ratio and efficient resource utilization, playing a vital role in space physics. In soft X-ray research, photon counting rate is used as a basic observation index, so that accurate quantification of soft X-ray radiation intensity and time variation is possible, and extensive researches such as solar flare dynamics, star canopy plasma evolution, soft X-ray spatial distribution high-resolution imaging and the like are supported. This makes accurate photon counting indispensable for driving the development of physical interpretation and observation accuracy. Even so, the photon count rate of soft X-rays is often contaminated during spatial detection. For example, EPICs mounted on XMM-Newton satellites are subject to magnetic layer reconnection events when viewed, and low energy solar protons with energies less than 300keV are accelerated, scattered with soft X-ray photons, and reach the focal plane. Thus, soft proton contamination is caused, so that the observed data is contaminated, and the true value of the soft X-ray photon counting rate cannot be accurately reflected. These soft proton flares are abrupt and have a great impact. In XMM-Newton observations, the signal from soft proton contamination can be above 1000% of the stationary background signal and can last from about 100 seconds to several hours, affecting up to 40% of the observation time. The distribution of the solar protons is similar to that of the true soft X-ray photons, and is difficult to directly distinguish from the pure soft X-rays due to spatial or morphological characteristics. Thus, a large amount of space-science observation data is contaminated. During periods of soft proton contamination, the absolute count rate and its time distribution may be altered, masking the physical characteristics. The use of contaminated data may distort the interpretation of soft X-rays, introduce systematic errors, and reduce the reliability of scientific conclusions. The photon count rate data of contaminated soft X-rays cannot be used directly. Contaminated fragment data will typically be labeled and excluded at the time of investigation, creating gaps in photon count rate time series. Thus, it is particularly important to complement this portion of the missing data. After completion, not only can the signal be recovered, but also the time continuity can be recovered, the data integrity is enhanced, and the reliability in downstream analysis is ensured by relying on the complete and physically consistent photon time sequence. Researchers have developed a number of kits to support photon count rate analysis. Of these, stingray is a widely used Python library that provides the basic functions for spectral timing analysis. Although it focuses on photon count rate time series, it is not suitable for processing incomplete data. It lacks built-in functionality for gap filling or missing data modeling, which limits its application in actual observation scenarios. This highlights the necessity of complementary methods to reconstruct the continuous sequence prior to analysis, especially when signal integrity is critical for downstream tasks. Most existing methods for processing soft X-ray photon count rate data are mainly focused on signal denoising and spectral analysis, but do not directly solve the interpolation problem of missing fragments. Thus, these techniques cannot recover a continuous sequence of photon count rates from incomplete and irregularly sampled observations. Disclosure of Invention The application aims to overcome the defect that the prior art cannot accurately complement incomplete photon counting rate observation data and accords with physical significance. In order to achieve the above object, the present application provides a method for supplementing soft X-ray photon count rate based on physical driving deep learning, comprising: Inputting the preprocessed measurement data into a trained photon counting rate interpolation model, and outputting the completed photon counting rate; the photon counting rate interpolation model comprises a first auxiliary sub-network, a second auxiliary sub-network and a main network; the first auxiliary sub-network is used for converting the input ion abundance ratio into SWCX scale factors; the second auxiliary sub-network is used for converting the input geomagnetic index into neutral hydrogen density; The first auxiliary sub-network and the second auxiliary sub-network comprise 1 transducer encoder block and 1 full connection la