Mirror Effect Based Learning Systems to Predict Human Errors - Application to the Air Traffic Control

Resilience of a human-machine system such as air traffic control relates to the successful control of its instability. This instability can be due to the temporal variability or the magnitude of functional task demands, and this may provoke human error occurrence. The paper proposes an original knowledge based learning approach for human error prediction by taking into account the temporal variability of task demands. The reinforcement of the knowledge base applies the concept of the reffect by considering different feedback strategies to handle the knowledge content. When a sequence of information from the knowledge base is similar to new inputs, this sequence is used for predicting human errors. The vector composed by new inputs and the real observed human error replaces then this sequence or it is considered as a new knowledge. This principle is the mirror effect that consists in mirroring a part or an entire content of a new input vector into the current knowledge base. Algorithms of the proposed mirror effect based learning systems are proposed in order to predict human errors by making correlations between the temporal variability of task demand and the occurrence of human errors. They are applied in air traffic control. The results show that the knowledge based on task demand evolution and its reinforcement by mirror effects are suitable for predicting human error occurrence. (C) 2016, IFAC (International Federation Control) Hosting By Elsevier Ltd. All rights reserverd.