Research On PSC Targeting Model Based On Intelligent Optimization Algorithms

PSC targeting model is a hot spot research direction developing rapidly in recent years. Based on the analysis of PSC targeting mechanisms and algorithms of primary MOU organizations in the maritime society, we propose nonlinear evaluation factors selecting algorithms and classification algorithms based on intelligent optimization algorithms, especially targeting models in emphasis, and preliminarily establish a set of scientific PSC targeting mathematical model. Detailed contents are listed as:On one hand, researches of attributes reduction on PSC targeting factors include:Investigation is made upon the attributes reduction on PSC targeting factors exploring Rough Set theory. We apply Rough Set theory toolbox RSES to reduce attribute. Due to the high error-compatibility capacity, it can save more time to calculate effectively and select more important factors to assess ship’s safety. The rules are concise, saving resources and effective.Investigation is made upon the combination of Rough Set theory and Analytic Hierarchy Process (AHP), proposing improved PSC targeting factors algorithm to solve the problem of the probability of multiple attribute reduction sets based on Rough Set theory solely. The algorithm introduce the concept of importance into non-core attributes with the help of discernibility matrix, on the premise of not degrade the effective classification information, and adopt attribute importance in order to find the minimum attribute set. Combining the advantage of Rough Set, discernibility matrix and AHP to reduce attributes, the algorithm can avoid being affected by subjectivity when existing multiple attribute sets.On the other hand, researches on PSC targeting mathematics algorithms and models include:A novel PSC targeting algorithm is proposed exploring (Backward Propagation, BP) neural network and (Radial Base Function, RBF) neural network to train samples and construct PSC evaluation model, based on that we have been get PSC targeting factors adopting Rough Set theory. Simulation results show that the proposed algorithm can effectively combine the advantage of BP neural network and Rough Set theory, with a concise rule and approaching global minimum. We find that practical and effective, but also having a broad application prospect. Compared with BP neural network, though the structure RBF method is simple, easy to programming and with less training times, but running efficiency is not very well and the error is larger. The reason of relative larger error is maybe due to the k-mean clustering algorithm only adopt sample input, and centre width of every hidden layer is same with a initial value according to experience.As to the disadvantage of neural network easy to drop in local minimum, an improved particle swarm-BP neural network PSC targeting algorithm is proposed. The algorithm can adaptively adjust Inertia weights and update speed and position according to premature convergence degree and individual fitness value, aimed to the problem of BP such as slower convergence speed and easy to fall into local minimum value. This paper explores improved PSO algorithm to train BP network, applying to PSC ship-selecting. Testing results show that this algorithm improves the performance on speed of convergence and precision of convergence.As to the probability of better identification effects with suitable parameters and enough samples, but it’s a difficult point of determine hidden layer number and learning rate when exploring neural network which is confinned according to practical situation in actual PSC targeting works, we propose a rapid risk evaluation model based on the combination of improved PSO and method, and apply it into demonstration analysis. The results show that the algorithm serves to do rapid ship classification, and the accuracy is97.619%, the time complexity is reduced efficiently. The algorithm is very suitable for limited inspect resource to do rapid ship classification, which is has certain practical value for PSC targeting.