Automatic ship berthing using artificial neural network trained by consistent teaching data using nonlinear programming method

Ship handling during berthing is considered as one of the most sophisticated tasks that a ship master has to face. The presence of current and wind make it even more complicated to execute, especially when ship approaches to a pier in low speed. To deal with such phenomenon, only experienced human brain decides the necessary action taken depending on situation demand. So automation in berthing is still far beyond imagination. But, if the human brain can be replicated by any suitable artificial intelligence technique to perform the same action that human brain does during berthing, then automatic ship berthing is possible. In this research artificial neural network is used for that purpose. To enhance its learnability, consistent teaching data based on the virtual window concept are created to ensure optimal steering with the help of nonlinear programming language (NPL) method. Then instead of centralized controller, two separate feed forward neural networks are trained using Lavenberg–Marquardt algorithm in backpropagation technique for command rudder angle and propeller revolution output respectively. The trained ANNs are then verified for their workability in no wind condition. On the other hand, separate ANNs are trained with reconstructed teaching data considering gust wind disturbances. To deal with any abrupt condition, ANN followed by PD controller is also introduced in case of command rudder angle output whose effectiveness is well verified not only for teaching data but also in case of non-teaching data and different gust wind distributions. Ship berthing has always demanded a sophisticated controller due to the requirement of multiple input and output parameters including data of environmental disturbances. To achieve that purpose, different kinds of controllers like fuzzy theory and knowledge based system are tried by many researchers but each of such controllers has some limitations when using for berthing. Like in case of fuzzy theory, it is needed to define the fuzzy rule but to define proper rule is very tough for berthing since any unpredictable situation may arise including environmental disturbances which cannot be pre-implemented as a rule. On the other hand in case knowledge based system or expert system, every possible situation needs to be included as written statement with corresponding solution which makes the controller more rigid instead of robust. Thus human knowledge judgement should be regarded as a best alteration in such cases. That is why, among different kinds of controllers investigated for berthing, ANN has some privileges like it …