The Time-domain Airborne Electromagnetic System Flying Parameter Fitting Based On RBF Neural Network

Time domain airborne electromagnetic system (Airborne Time-domainElectromagnetic Systems, abbreviated as ATEM) has a large magnetic moment launch,space exploration, high resolution, probing depth, higher detection efficiencyadvantages, it is widely used in geological survey, mineral exploration, waterresources quick survey and other fields. However, during the actual detection ofairborne electromagnetic system in flight, launch-receiving coil, flight speed, air,wind and other factors, result in real-time changes in their posture, causing airborneelectromagnetic system to produce geometric errors, affecting the aviationelectromagnetic measurement accuracy and data interpretation accuracy.According to the problems above and combined with the research project"Multi-modal aviation transient electromagnetic system calibration algorithm toachieve" of the Institute of Electronics,based on a time-domain airborneelectromagnetic theory, this paper proposed using RBF (Radial Basis Function)neuralnetwork fitting method for time domain airborne flight system parameter fitting. Thefollowing research has been conducted：1.We calculate the theoretical derivation and calculate the electromagneticresponse of the non-magnetic conductive sphere, cylinder, and conductivenonmagnetic conductive sheet of the limited time-domain electromagneticresponse conductor. Describes the calculation is based on the time-domain airborneelectromagnetic system testing theory and electromagnetic coil in response toabnormal formula. Ultimately determine the correspondence between the abnormalrelationship between the coil and the wire conductor, conclusion： as long as thelimited time of abnormal coil conductor with constants is same, then you can simulatea limited abnormal coil conductor.2.Based on the basic theory of electromagnetic field, and mutual Neumannformula derived in detail anywhere coil inductance calculation formula. Changesaffect the position of the coil inductance value directly, thereby affecting the entireelectromagnetic system response data. The time domain airborne electromagneticsystem anywhere in the coil inductance coil is calculated into the abnormal responsebased on the formula, the electromagnetic response of the final derived formula.Combined with the actual system operating parameters are calculated based onseveral sets of typical abnormal coil domain airborne electromagnetic response.3.According to the principle of the principal component analysis, thetime-domain airborne electromagnetic system flight parameters were calculated andsorted contribution rate. Response time domain airborne electromagnetic system largeamount of data, input parameters more, proposed the use of RBF neural networkmethod for flight parameters to fit a combination of these characteristics. RBF neuralnetwork has a high degree of fit fast fitting without overfitting the advantage,compared to least-squares fitting results finalized after fitting program. According tothe actual situation given the parameters and structure of the flow chart diagram ofRBF neural network fit.4.Combined with theoretical simulation laboratory data, field data and dynamicflight simulation experiment field actual flight test data, the use of the proposedprogram were carried out fitting. Laboratory of theoretical simulation single theoretical data anomalies less than the mean absolute error20nV/m2, goodness offit is0.91. Theoretical data anomalies of the double mean absolute error is160nV/m2,the goodness of fit is about0.87. Dynamic simulation flight test of goodness of fit isabout0.82, Tongbai field experiment measured the actual flight data fitting coil centerin abnormal positions relative error is less than1%, the goodness of fit is0.97, theentire cross-section survey line fitting the relative error is less than±5%, with anaverage of2.5percent, the goodness of fit is0.91. The results show that RBF neuralnetwork fitting is effective.Therefore, the RBF neural network fitting can achieve better fit airborneelectromagnetic system flight parameters for fast processing airborne electromagneticsystem mass measured data, it provides a new method.