Study On Extreme Learning Machine And Its Application In Wireless Spectrum Prediction

Extreme Learning Machine (ELM) is a new type of single layer feedforwardneural network proposed by Huang Guangbin. Compared with the traditional feedforward neural network, it has a simple structure, fast learning speed, global search ability and good generalization performance. But in the process of practical application, the number of hidden layer nodes could influence the result of practical problems. Too much of the hidden layer nodes will produce over fitting, and will unavoidably have little or no use of nodes to the practical problems. In order to solve this problem, a kind of Optimally Pruned Extreme Learning Machine (OP-ELM), by cutting through the hidden layer nodes in the original structure is proposed, and it improves the robustness and generalization of the classical Extreme Learning Machine.Currently, Cognitive Radio (CR) system is a key technology which used to solve the problem of spectrum resources shortage. The important reason of the shortage spectrum is the irrationality of spectrum allocation. Cognitive radio by sensing the primary user (PU) spectrum occupancy, so that second user (SU) makes full use of the spectrum holes which from primary user. It can be intelligent dynamic allocation of spectrum resources to achieve reliable communication services and improve spectrum utilization. Spectrum prediction is the key technology of Cognitive Radio. There are many traditional prediction methods, such as Markov chain method, regression analysis method and neural network method, etc. The traditional forecasting methods required a long time, therefore, it cannot meet the requirements of real-time spectrum predicted. The introduction of extreme learning machine, not only meet the requirements of real-time spectrum prediction, moreover, the robustness and generalization of the Optimally Pruned Extreme Learning Machine is better than the traditional machine learning, so it is more suitable for radio spectrum prediction problem.The main work is as follows.(1) System and deeply studies the principle and characteristics of Extreme Learning Machine. This article describes the mathematical model and training algorithm of extreme learning machine and several improved Extreme Learning Machine, and expound the fast characteristics of the Extreme Learning Machine from the theorem. Through the simulation experiment compares the Extreme Learning Machine with those of the traditional feedforward neural network to analysis the prediction performance.(2) In this paper, aiming at the problem that too much hidden layer nodes will affect the network performance, we adjusted the hidden layer node of the classic extreme learning machine, and construct the Optimally Pruned Extreme Learning Machine, then we analyze the performance of the Optimally Pruned Extreme Learning Machine through the benchmark experiments.(3) The Extreme Learning Machine is applied in cognitive radio spectrum prediction problems. In view of the problem of prediction accuracy and prediction speed in existing prediction methods, take the advantage of ELM’s simple, fast and the global optimal, use it predict the spectrum duration of primary users of cognitive radio system. This paper compares the performance between Extreme Learning Machine and several kinds of its improved model, the traditional feedforward neural networks and feedback neural network in spectrum prediction problem. The experiment results show that, compared with the traditional feedforward neural network and feedback neural network, Extreme Learning Machine especially Optimally Pruned Extreme Learning Machine obtained better performance both in prediction accuracy and prediction speed, and more suitable for wireless spectrum prediction problem.