Method for rapidly identifying guidance law of intercepted aircraft based on GRU
An identification method and aircraft technology, applied in the direction of attitude control, etc., can solve the problems of long identification transition process, discount of engineering application value, limitation of aircraft maneuverability, etc., and achieve the effect of high accuracy and fast identification speed
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
experiment example
[0124] To build an identification model, the basic data used to construct the sample library are as follows:
[0125] Set the simulation step size of the relative kinematics model to 0.001s, the update frequency of the kinematic information measurement of the interceptor aircraft by our aircraft is 50Hz, and the input step size includes five cases of 10, 15, 20, 25, and 30, and the corresponding time spans are 0.2s, 0.3s, 0.4s, 0.5s, 0.6s. Finally, 160,000 training samples were extracted, each of which accounted for 20% of each time span, and each of the 10 different guidance laws accounted for 10%. Among the extracted samples, 10 were randomly selected for testing, and the rest were used for training. Among the neural network training parameters, the dropout failure rate is 5%, the number of batches is 3000, the number of iterations in each experiment is 2000, the initial learning rate of the network is 0.00025, and the learning rate decay rate is 0.85 every 100 iterations. ...
experiment example 1
[0127] When the input step size is 15, the number of hidden layers is 3, and the number of neurons in each hidden layer is 81, the corresponding identification models are obtained through GUR network, LSTM network, RNN network and BP network respectively.
[0128] The identification models obtained through different types of networks, the training effect and identification accuracy are as follows Figure 4 and Figure 5 shown in;
[0129] according to Figure 4 and Figure 5 It can be seen that the recognition accuracy rates of LSTM network and GRU network are 92.78% and 95.88% respectively, which are 3.45% and 6.55% higher than 89.33% of RNN network, and 1.44% and 4.5472% higher than that of BP network. It is proved that compared with LSTM and BP, GRU has greater advantages in dealing with timing-related problems.
experiment example 2
[0131] Robustness of the identification model obtained by different networks:
[0132] There will be errors when our aircraft measures the kinematic information of the enemy interceptor aircraft, and there will also be deviations between the relative motion model construction and the real model, both of which affect the input of the model as Gaussian white noise. Set the noise standard deviation of each input parameter, as shown in the table below, where noise condition 1 means that the sensor on our aircraft is working normally and the deviation between the constructed model and the actual physical model is small, and noise condition 2 means that the sensor cannot work normally or the constructed model is different from the actual physical model. The physical model has a large deviation, and the identification accuracy changes as follows Image 6 shown in.
[0133]
[0134] according to Image 6 It can be seen that the accuracy of each type of network after being affecte...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More