Research On Post Mission Compensation Method Of Airborne Gravity Gradiometry Based On Deep Learning

Compared with gravity signals,gravity gradient signals are more sensitive to distance changes,have higher resolution,and have more components in space,so they can provide richer and more comprehensive gravity field information.it can reflect the fine changes of local geological characteristics,and plays an important role in many scientific fields,such as space science,marine science,inertial navigation,aerospace and national defense technology.Airborne gravity gradiometry can collect gravity gradient signals with low cost and high efficiency,and is not affected by many harsh terrain,so the development of airborne gravity gradiometer is of great significance.The airborne gravity gradiometer has been developed early in foreign countries and has been commercialized.The related research in China started relatively late.At present,the "13th five-year Plan" project aims at the engineering and practicality of gravity gradiometer,and error compensation is one of the key technologies.Although there is a mature error compensation process abroad,through the post mission compensation technology,the response model for various parameters in the measurement process of the gradiometer is used to compensate the error.However,because of the technical blockade,our country can only develop its own error compensation technology.Previous studies in China have mainly focused on the establishment of error models for error compensation for all kinds of non-ideal factors of gradient meters,but on the one hand,it is difficult to model because they do not understand how some parameters affect the measurement of gradient meters.On the other hand,many error-related parameters are difficult to calibrate,and the existing modeling errors may not be compensated.In actual measurement,the types of error compensation that can be carried out by the gradient meter prototype are relatively limited.It limits the further improvement of the effect of error compensation.Deep learning is a hot research field of computer science in recent years.In deep learning,the neural network is used to learn the rules directly from the data,so there is no need to understand the specific influence mechanism of various parameters,and there is no need to manually calibrate the error-related parameters.The use of deep learning for error compensation is expected to further improve the compensation effect.In this paper,an post mission compensation method based on depth learning is proposed,that is,a neural network used to predict the output noise of the gradiometer is trained by using the parameters in the airborne gravity gradiometry and the output noise of the gradiometer.The predicted output noise is used to compensate the output signal of the gradiometer.In order to study the effect of post mission compensation based on deep learning,the measurement equation of gradiometer considering all kinds of non-ideal factors is derived,and according to the measurement equation of gradiometer,the output noise of the gradiometer is simulated by using the simulated gradiometer parameter signal,and then the simulation data and the measured data are used to compensate the post error based on depth learning.The results show that the complex laws contained in the data can be learned by depth learning,and the output noise of the gradiometer predicted by neural network is very close to that of the simulated and actual gradiometer.The noise is reduced by more than an order of magnitude after compensation by using the predicted gradiometer output noise.For the measured data,the error level of compensation is close to the target accuracy of the gradient meter prototype.To sum up,the post mission compensation method based on deep learning can effectively compensate the error in airborne gravity gradiometry.Under the current condition of the gradient meter prototype,the post mission compensation method based on deep learning is effective,and it is expected to be used alone or in combination with the modeling method to further improve the accuracy of the gradient meter prototype.