Research On The Body Attitude Prediction Of A Self-propelled Anti-aircraft Gun Based On Neural Network

When the self-propelled anti-aircraft gun is fired continuously,the complex attitude change of the vehicle body caused by the load impact is one of the important factors affecting its shooting accuracy.In order to acquire the attitude of anti-aircraft gun vehicle in real time,this paper presents a method of vehicle body attitude prediction combining dynamic simulation and neural network technology.In any shooting Angle condition,the vehicle body vibration attitude information can be quickly and accurately acquired,which provides accurate basis for real-time adjustment of fire control and has important reference value for improving the continuous shooting accuracy of self-developed anti-aircraft gun.The main research contents and conclusions of this paper are as follows:(1)Based on the research of the existing vehicle attitude acquisition methods,combined with the vibration characteristics of the self-developed anti-aircraft gun vehicle body and the requirement of the attitude solution time,the overall technical scheme for building the vehicle attitude prediction model was proposed.(2)Complete the construction of full gun dynamics model based on multi-body system theory.Considering the flexible deformation of the frame is an important factor that causing the attitude change of the body,the part of the frame is treated with the finite element technique.The dynamic model is verified by experimental data,and the accuracy of the model is improved by modifying the dynamic parameters to ensure the reliability of sample data sources.(3)Taking the range of local radiative Angle as the research object,the influence of sample data distribution on the prediction accuracy of the network model is analyzed to determine the best way to construct the sample base.Design macro commands to achieve simulation automation.Through data cleaning,data extraction and normalization,the data obtained by simulation are preprocessed to complete the construction of sample base.(4)According to the characteristics of the sample data,the LM-BP algorithm was designed as the core of the initial structure parameters of the network,and the convergence state of the network and the attenuation ratio of the mean square error were taken as the targets to optimize the parameters during the training process.The generalization of the final model is verified by non-sample data and the algorithm is implemented by hardware and software based on DSP.