Research On Multi-step Thrust Allocation Optimization Algorithm Of Vessels

Entering the era of ocean economy, various countries have advanced their exploitation for various marine resources to deeper sea and harsher environment, new marine operations such as dynamic positioning, vessels coordination and offshore rescue are developed and widely needed, all these needs have made new challenges to the performance of propulsion system on board. Propulsion system with redundant thrusters has been the dominant choice for various ships such as drilling ships and surveillance ship. Hence, to improve the control precision, stability and maneuverability, it is of great importance to design intelligent thrust allocation algorithm that utilize the redundant thrusters wisely and efficiently.Subject to mechanical constraints, thruster forces and angles are slow to follow input commands. Traditional thruster allocation optimization algorithms are carried out independently at each time step at fixed frequency, they only consider the limited feasible regions for thruster states within one step imposed by mechanical rate constraints. This scheme overlooked the long term optimization in broader feasible region for thruster states and may lead to degraded thrust configuration hence suffering from lower power efficiency and maneuverability. Viewing thrust allocation as a long term multi-step process, a novel algorithm is developed in this thesis to improve long term performance of thrusters: genetic algorithm is first exploited to obtain global optimum in boarder feasible region across multiple steps, then adaptive dynamic programming is used to obtain the change process from current thrust states to the optimum one. To realize the algorithm, following research achievements is discussed:(1) Broader feasible region introduces local minimums that would fail traditional deterministic optimization methods. Optimum thruster states’ sensitivity to thrust forces and angles is analyzed and pseudo-inverse method is incorporated into the mutation operator to maintain population diversity and reduce computation cost, moreover adaptive mutation mesh grid is used to improve the local search. Simulation result shows that modified genetic algorithm obtains better global optimum that traditional SQP method, especially in broad feasible regions across several seconds.(2) In order to solve the multi-step process optimization, a nonlinear optimal control model with input saturation and terminal constraints is developed. A HJB equation formation for the constrained problem is researched and proposed. The training and solving schemes of adaptive dynamic programming is proposed is realize the terminal constraints. Simulation result shows this algorithm improves overall power efficiency and maneuverability compared with traditional SQP method.Lastly, a hardware-in-the-loop simulation platform is discussed and developed in this thesis. It includes embedded systems on model ship and control and monitor terminal on laptop. With simple operations, the model ship can demonstrate waypoint tracking and dynamic positioning in outdoor lakes and stay resistant to wind and waves.