Research On Prediction Methods For Deck-Motion Of Air-Carrier

Under the complex sea condition, six degree freedom-motion of large ship will be produced with the excitation of waves, winds and other interference, which brings negative effects on the landing safety of the carrier-based aircraft, the control of weapons as well as the crews and equipment. When these motions cannot be effectively controlled, research on extremely short-term prediction technologies on deck-motion is significant. Extremely short-term prediction on ship motion means predicting the deck-motion of the following few (usually more than ten) seconds with historical deck-motion data that including current motion data. Statistical predicting, convolution, periodic graph, power spectrum autocorrelation, Kalman filter are common methods for the ship extremely short-term prediction. There are some shortcomings when above methods applied for real time prediction because accurate mechanical models and statistical parameters are always needed but not easy to obtain. Aiming to predict deck-motion of air-carrier accurately and in real time, prediction methods based on Grey theory, BP neural network and online sequential extreme learning machine with particle swarm optimization are successively introduced and studied in this dissertation. The main work is as follows:Firstly, researches on deck-motion are introduced and a detailed deck-motion model based on the superposition of multiple sine waves is studied. The negative effect of deck-motion on landing and take-off of aircraft is analyzed, and the coupling process between ship swinging and lever-arm is introduced. Also the needs of preparing and on-line replacing methods for samples used for prediction model training are analyzed.Secondly, grey prediction for deck-motion is introduced. The mechanism of classical grey prediction model GM(1,1) is analyzed. For the inherent defects of GM(1,1) model, discrete grey prediction model DGM and the improved algorithm NDGM are introduced successively. Simulation results indicated that the method with NDGM can effectively predict the motion of sine waves, but cannot accurately predict that of the superposition of sine waves.Thirdly, the neural network prediction method for the deck-motion is introduced, and the features, developments about it are also studied. After analyzed the mechanism of neural network and BP model (a most representative feed-forward neural network), gene algorithm is brought to get optimized parameters for BP. Simulation results show that the method based on GA-BP can effectively predict the deck-motion but at costs of long training time. That means GA-BP cannot be used for real-time deck-motion prediction.Finally, aiming to shorten training time in traditional feed-forward neural network, single-hidden layer feed-forward neural network-ELM is introduced and the background, value and advantages about ELM are analyzed. The mechanism of standard ELM is studied, and for using ELM with batch training of real-time data, online sequential extreme learning machine -OS-ELM is proposed. Aiming to get optimized parameter for OS-ELM, PSO algorithm is further introduced. Simulation results show that the method based on PSO-OS-ELM can effectively predict deck-motion.