A Study And Implementation On Aircraft Collision Detection Technology

Air conflict detection is the method, which aim at recording the trajectories ofplanes, judging their future movement, and then calculating the possibility of crashduring a period of time. The main idea of this thesis is to estimate the intent informationof a plane by using real-time flight data from radar. By comparing the results of all theadjacent planes, we can infer that whether there is a collision between the planes in thefuture. In order to implement this, we need a state predictor.The main method is the3rdgeneration artificial neural networks—Spiking neuralnetworks. It can transform the input data into a series of fire-times of neurons. By theadjustment and computing, the neural networks will simulate the trajectory of eachplane. Though the neuronal computation process, it could pick up a large number ofpredictions, and then transform it into the form of the probability distribution by using aappropriate method, then use the relative entropy to describe the errors between realoutput and expect output. In the end, we update the parameters of neural networks byerror back propagation method, it could make neural network more familiar with thereal aircraft flight path.After the trajectory prediction, we can judge the conflict. We should reduce thecomputational complexity by a filter at first, and then we can compute the spatialdistance among planes by using a method named proximity. At last, we can finallypredict the conflict by comparing it with a standard.This thesis come up with a complete, innovative aircraft collision detectionalgorithm through a large number of experiments, analyzes, summarize. Especially, thecreativity works on Spiking Neural Network Model, the probability distribution errorcalculation methods, and derivation a new method for neural network parameteradjustment mechanism; it may dedicate a little meager force for neural networks andlearning algorithm.