Research On Shortwave Communication Effectiveness Evaluation Based On Cloud Neural Network

Shortwave communication is widely used in military and civilian fields.Ionosphere-reflection channel is the main channel in shortwave communication, it istime-varying and complex, the reliability of the shortwave communication is seriouslyaffected by it. Currently “Shortwave Communication Frequency Forecast” is the mainauxiliary method in the process of establishing reliable shortwave communication. Thismethod only considers of the frequency and the result of it is an average value over a longperiod. It is not flexible and comprehensive enough from the point of application. Thedissertation proposes a new auxiliary method: shortwave communication effectivenessevaluation. There are two problems which need to be resolved in the method: Firstly, thereare certain factors and uncertain factors in the effectiveness evaluation of shortwavecommunication and the certain factors also exist acquisition-uncertainty. How to portray theuncertainty of factors? How to convert the factors to the evaluating values? Secondly,many factors affects the effectiveness of the shortwave communication and the relationshipsbetween effectiveness and factors are complex and non-linear. How to get a reasonableevaluation from lots of complex nonlinear factors?In this dissertation, the improved cloud neural network is proposed to evaluate theeffectiveness of shortwave communication, the network is based on existing cloud neuralnetworks. In the process of determining the number and the digital characteristics of thecloud model, the existing cloud neural networks adopt the man-determine approach or thecloud-transform-based approach to determine them. But the input dimension of shortwavecommunication effectiveness evaluation is larger and the training sample data distributionof shortwave communication effectiveness evaluation is dispersed, in this case, adopting theman-determine approach will affect the accuracy of the results, adopting thecloud-transform-based approach will greatly increase the time complexity. This dissertationadopts the center-determine method of traditional RBF neural network to solve the problem.In the process of learning the uncertainty rules which in the samples, the existing cloudneural networks learning is not thorough enough.. This dissertation according to the idea of "multi-trigger and superimposed" designs the improved learning algorithm.Experimental results show that the improved cloud neural network not only has thenonlinear approximation ability of neural network, but also has the portraying uncertaintyability of cloud model, with the help of it, the uncertainty of shortwave communicationeffectiveness evaluation’s samples can also be transmitted into the network, the network canlearn both the certainty rules and uncertainty rules from the samples. Compared with theexisting cloud neural networks, the prediction accuracy is improved, there are obviousadvantages in dealing with the problem of " large input dimension and disperse sample datadistribution". The improved cloud neural network can effectively evaluate the effectivenessof the shortwave communication, be able to play a good auxiliary role in the establishmentof reliable shortwave communication.