Research On Measurement Of Controller's Alertness Prior To Watch

Air traffic control work is important and complex,suggesting that there are vast mental and physical work need to be done by controllers to ensure the aviation safety,however,the 24/7job requirements of the Radar control work make it possible for the controller to be on duty at any time,which makes it difficult to avoid the situation of fatigue before taking up the post,and inadequate alertness of the controllers has become a hidden safety hazard in the operation of air traffic control.Therefore,in this paper,alertness identification and prediction models based on the PVT test and the analysis of voice data were proposed,to provide theoretical support for scheduling.After comparing the characteristics of various alertness measurement methods,this paper chose the PVT test,which is widely used in alertness measure,as the measurement method in the prediction model.Firstly,according to the change of PVT data,the logistic function containing the unknown parameters ? and ? is selected as fitting function,and the optimal values of parameters ? and ? are determined by Markov Chain Monte Carlo sampling method.The model can excavate the historical alertness data of the controller,realize the alertness prediction of the controller on duty,and provide the guidance suggestion to the scheduling.Then,in view of the large amount of radiotelephony data generated by control work,this paper analyzes acoustic features of the radiotelephony data,including short-term average energy,short-term average amplitude,short-term average zero-crossing rate,pitch frequency,average speech rate,Melt Frequency Cepstrum Coefficient(MFCC),and response time and used PVT test to correlate alertness levels with speech data.Then based on those acoustic vectors,Support Vector Machine(SVM)is introduced to verify different parameters' combination effectiveness.After the train and classify upon SVM,the model realized the alertness classification based on the voice data,that is,low alertness recognition,and compare the prediction results of the prediction models to verify the prediction effect.The results shows that the alertness probability prediction model proposed in this paper can effectively excavate the controller's alertness change rules and provide theoreticalguidance for the controller's optimal scheduling.The alertness identification model can effectively identify the inadequate alertness based on the voice data of the controllers,and combined with the predictive model,to prevent the controller from performing control work in a non-alert state.