Search

CN-121980939-A - Machine learning-based method for determining on-orbit background of hard X-ray detector array

CN121980939ACN 121980939 ACN121980939 ACN 121980939ACN-121980939-A

Abstract

The invention provides an on-orbit background determination method of a hard X-ray detector array based on machine learning, which is oriented to high-energy detection equipment such as a hard X-ray imager of' Quadrai-Yi-No. "ASO-S, and belongs to the field of spatial science data processing. The method comprises the steps of integrating observation data of a monitoring probe in a detector array, satellite orbit magnetic field parameters and external space environment parameters to construct multi-source feature vectors, training and optimizing a machine learning model by using a signal-free data set, and inputting a target period feature vector into the model to realize high-precision prediction of a scientific observation probe background. According to the invention, the monitoring probe data is taken as an anchor point, and the dynamic environment parameters are combined, so that individual background estimation with high time resolution is provided for each probe, and the extraction precision of the target signal is obviously improved. The method has outstanding advantages in processing the data affected by the radiation band particles, and effectively expands the scientific sample size applicable to energy spectrum and imaging analysis.

Inventors

  • LIU WEI
  • SU YANG
  • CHEN WEI
  • CHEN DENGYI
  • ZHANG ZHE
  • LI ZHENTONG
  • YU FU
  • JIANG XIANKAI

Assignees

  • 中国科学院紫金山天文台

Dates

Publication Date
20260505
Application Date
20260123

Claims (8)

  1. 1. An on-orbit background determination method of a hard X-ray detector array based on machine learning, wherein a hard X-ray imager comprises a first type detector for scientific observation and a second type detector sensitive to environmental background, and the method is characterized by comprising the following steps: Step 1, multi-source data are acquired and aligned according to uniform time stamps, wherein the multi-source data comprise HXI instrument data, satellite platform data and external space environment data, and HXI instrument data are energy spectrum data from a second type of detector; step 2, processing the multi-source data obtained in the step 1 to construct a feature vector for a machine learning model; Step 3, constructing and training a machine learning model, wherein the machine learning model is used for learning a nonlinear mapping relation from the feature vector in the step 2 to an output target value, and the target value is actually measured energy spectrum data of each probe in the first type of detector; And 4, constructing a feature vector according to scientific data containing a target signal event, inputting the feature vector into a trained machine learning model, and outputting a predicted energy spectrum of each probe in the first type of detector as an environment background.
  2. 2. The method for determining the on-orbit background of the hard X-ray detector array based on machine learning as claimed in claim 1, wherein in the step 1, the satellite platform data is obtained by obtaining engineering parameters of a satellite, calculating the orbit parameters and the magnetic coordinate parameters of the satellite through coordinate conversion, and generating an area mark of the satellite, and the external space environment data is obtained by obtaining solar wind parameters and geomagnetic indexes from a public data source.
  3. 3. The method for determining the on-orbit background of a hard X-ray detector array based on machine learning as claimed in claim 2, wherein the engineering parameters comprise ECI coordinates and geomagnetic field vectors, the orbit parameters comprise geographic longitude and latitude and altitude of each moment of a satellite, the magnetic coordinate parameters comprise geocentric L shell values, magnetic latitude and magnetic places of each moment of the satellite, and Boolean marks describing whether the satellite is in a radiation zone, a polar region or an SAA region are generated as regional marks according to the geocentric L shell values and the magnetic latitude.
  4. 4. The method for determining the on-orbit background of a hard X-ray detector array based on machine learning as set forth in claim 2, wherein in step 2, the processing of the data acquired in step 1 is specifically: and aligning and combining all HXI instrument data, satellite platform data and external space environment data according to the world time with second-level precision to generate a time sequence data table containing all data.
  5. 5. The method for determining the on-orbit background of a hard X-ray detector array based on machine learning according to claim 2, wherein in the step 2, the feature vector for the machine learning model comprises the count rate of all energy bins of the probe in the second type of detector, all orbit parameters, magnetic coordinate parameters and region markers, time-corrected external spatial environment data and derived features.
  6. 6. The method for determining the on-orbit background of a hard X-ray detector array based on machine learning as set forth in claim 5, wherein said derivative features comprise a ratio of high energy segment to low energy segment count rates of the probe in the second type of detector and a product of a centroid L shell value and a Kp index.
  7. 7. The method of claim 1, wherein in step 3, during training of the machine learning model, feature vectors corresponding to non-target signal events are screened out as inputs of the machine learning model, and measured energy spectrum data of each probe in the first type of detector at the same time is used as output target values of the machine learning model.
  8. 8. The method of claim 1, wherein in step 3, the machine learning model is a multi-layer perceptron neural network, and the training process uses a loss function that performs logarithmic transformation on the output target value and weights samples of a specific physical area.

Description

Machine learning-based method for determining on-orbit background of hard X-ray detector array Technical Field The invention belongs to the technical field of spatial science data processing, and particularly relates to a hard X-ray detector array on-orbit background determination method based on machine learning. Background In high-energy celestial physical observation, particularly in the observation of high-energy transient phenomena such as solar flare and gamma ray storm by using a detector (such as a hard X-ray imager) carried by an artificial satellite, a signal received by the detector usually consists of two parts, namely an effective physical signal from a target celestial body and a background signal from a satellite orbit space environment. The environmental background signal for the high-inclination near-earth orbit is mainly generated by the interaction of high-energy charged particles, cosmic rays and the like in an earth radiation band with a detector and surrounding substances thereof, and the intensity and energy spectrum form of the environmental background signal are dynamically changed along with the factors such as the orbit position of a satellite, the space weather condition and the like. In order to obtain a clean target signal, the environmental background must be accurately subtracted from the original observations. In the prior art, a "modeling method" or a "background interval method" is commonly used. The background interval method is to deduct the background by referring to the background interval (such as before the occurrence of an explosion event or near the orbit) which is adjacent to the time or space, when the method is applied to the solar synchronous orbit satellites such as ASO-S, the data of the corresponding time period of 48 hours before or after the occurrence of the target event (such as solar flare) is selected as the current environmental background to be deducted according to the assumption that the data can return to the approximately same position under the earth coordinate system every 48 hours. However, this prior art has the following significant drawbacks: 1. The accuracy is not enough, the satellite does not precisely return to the same position after 48 hours, the orbit drift can cause the difference of space environments, the background is not completely the same, the translation alignment of a few seconds to tens of seconds (aiming at ASO-S) is often needed to be manually carried out, and the operation is complex and the subjectivity is strong. 2. The dynamic adaptation is poor-the particle flux of the near-earth space environment, especially the radiation band and the polar region, changes rapidly with time. Even in the same geographic location, there may be a large difference in background at different times. The background around 48 hours can not truly reflect the background condition at the current moment. In addition, if an explosion event occurs around 48 hours, the data cannot be used as a background. 3. Radiation band internal failure-as satellites traverse the earth radiation band (e.g., the south atlantic anomaly SAA or polar region), the particle environment is extremely intense and dynamic, the background signal increases dramatically and changes rapidly. At this point, the 48-hour translation method is almost completely disabled and does not provide a reliable background estimate. 4. Individual differences in detectors neglect that for an array of sub-detectors (e.g., HXI of the 99 probes), there are differences in the environmental background each senses due to the different specific location of each probe on the satellite platform and the different surrounding material shielding environment. The existing methods generally fail to model the individual background of each probe. The other two methods also involve the use of a dedicated background detector, but this method also does not reflect the background distribution throughout the array. Therefore, there is a need for an environmental background determination method that overcomes the above-described drawbacks, achieves dynamics, high accuracy, and has detector array coverage. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a hard X-ray detector array on-orbit background determination method based on machine learning, which aims to solve the problems of inaccurate background subtraction, poor dynamic adaptability, incapability of processing data in a radiation band and the like in the background technology. In order to achieve the above purpose, the present invention adopts the following technical scheme: the invention provides a machine learning-based method for determining the on-orbit background of a hard X-ray detector array, wherein a hard X-ray imager comprises a first type detector for scientific observation and a second type detector sensitive to the environmental background, and the method comprises the following steps: Step 1, multi-so