Research On Radar Stealth Statistical Characteristics Analysis And Prediction Of Fighter

Radar cross section(RCS)is a physical quantity reflecting radar stealth performance.The radar stealth performance of fighter will directly affect its combat effectiveness and survivability in air combat.The study of RCS statistical characteristics of fighter can accurately and effectively obtain the statistical law of fighter and provide reference for the identification of stealth fighter.In actual combat,radar is often unable to detect the target from all angles.Research on RCS sequence prediction can estimate the unknown Angle of the target,which can better meet the needs of radar for target detection in the real battle field.The main work of this paper are as follows:(1)The cumulative distribution function(CDF)is commonly used to describe the statistical rule of fighter’s RCS,but the kernel density estimation algorithm based on fixed bandwidth is difficult to meet the accuracy requirement considering CDF estimation.In this paper,an adaptive kernel density estimation(AKDE)algorithm based on least square cross validation and integral squared error criterion is proposed to optimize the bandwidth.The adaptive optimal bandwidth of the algorithm can take into account the global data,thus improving the accuracy of CDF estimation.Simulation results show that AKDE algorithm improves the estimation accuracy by more than 50%compared with traditional algorithm,and can obtain the statistical characteristics of fighter more accurately.At the same time,AKDE algorithm has a significant advantage in CDF estimation when solving RCS data less than1m~2.(2)In HH and VV polarization,the variation of RCS sequence has the characteristics of disorder and nonlinear.In order to solve this problem,an algorithm based on machine learning and neural network is proposed to predict the mapping relationship between single station RCS and incident Angle.Based on the auto-regressive integrated moving average model(ARIMA),bayesian information criterion(BIC)was introduced to optimize the order of model,and the residual value was combined with the optimal nonlinear auto-regressive with exogenous inputs(NARX)network,and the ARIMA-NARX prediction algorithm based on BIC was proposed.Simulation results show that the proposed prediction algorithm is 60%more accurate than the traditional prediction algorithm,and it is not restricted by frequency and polarization.(3)The variation law of RCS sequence is smaller and more complex in the cross-polarization mode,and the existing neural network algorithm is not effective in the prediction of RCS sequence.To solve this problem,this paper introduces the output value of LSTM hidden layer into TPA mechanism which pays more attention to historical data information,and proposes TPA-LSTM algorithm which can obtain high prediction accuracy.The parameter optimization of TPA-LSTM combined model is carried out for 6 cases with 3bands and 2 polarization modes.Finally,simulation results show that the prediction accuracy of the proposed algorithm is more than 20%higher than the existing algorithm.