Improved control of propeller ventilation using an evidence reasoning rule based Adaboost.M1 approach

Abstract Marine propeller ventilation, which can occur in extreme seas, reduces energy efficiency of the ship, and can result in significant safety hazards during navigation. A major maritime challenge is to predict water levels during extreme seas and manage an appropriate control strategy for the propulsion system with fast response times. Here we report application of an evidence reasoning (ER) rule-based Adaboost.M1 approach to rapidly estimate propeller ventilation levels thereby improving propulsion management. The ER rule was used as the weak learner of Adaboost.M1 to extract torque signal from the propulsion motor and current signal from the propulsion inverter. First, K-means clustering was used to select the reference value sets of each weak learner. Second, the extracted signals were transformed into multiple pieces of identification evidence, which were combined through ER to estimate the ventilation level by present weak learner. Parameters of the weak learner were then updated according to the estimation results. Finally, the weak learners that had completed training were combined into a strong learner to estimate final ventilation levels. Experimental results showed that the estimation accuracy of the weak learner provided by the ER rule for Adaboost.M1 approach was more than 50%, which resulted in the accurate estimation by the strong learner. This approach provides an important technical advance to allow more effective switching control strategies for marine electric propulsion systems under different sea conditions.