Research On The Intelligent Assessment Methods For Collaborative Simulation Training In Marine Engine Room

The Ministry of Transport of China issued the national seafarers' development plan from 2016 to 2020 according to the strategic goal of building China into a maritime power, and put forward the task of promoting the electronic and intelligent assessment for seafarers. The purpose is to improve the quality of the seafarers with standardized and normalized assessment for the competence, reduce the occurrence of the ship engine-room accident caused by the human factors, and promote the maritime traffic safety. The dissertation focuses on the intelligent assessment methods for collaborative simulation training in marine engine room. The improvement of the operation training mode and the assessment system for simulation training for seafarers,the application of the marine engine room accidents in the training and assessment, and the evaluation system and methods for engine room simulator are studied. The main research work is as follows.In order to meet the special requirements of the new mandatory competence standard of-the engine-room resource management, this dissertation studies the simulation training and assessment model of the marine engine room based on the method of "man - machine - environment system engineering". The cooperative training model of the "role - task - resource" system of the ship is established, and the task centered cooperative training mode with linkage between deck and engine departments is put forward. This model overcomes the shortcomings of the traditional engine room simulator in the integrated full mission training and the objective assessment. This dissertation also analyzes the marine engine room accidents,studies the relationship among the accidents, human error factors and engine room resource management factor,designs targeted situation and task, improves the assessment system, and focuses on the competence of reducing the engine room accidents in training and assessment. The training and assessment have more practical value especially for marine college students who lack seagoing service.Based on the automatic comprehensive evaluation of the engine room simulator,the intelligent algorithm is introduced into the assessment of the training of the marine engine room, and an improved intelligent comprehensive evaluation method based on genetic algorithm optimization is proposed. The method includes building the linkage knowledge base between bridge and engine rpom, constructing evaluation index membership function library and target functions for different optimization, using entropy weight method and the historical assessment data to adjust the weights dynamically and optimizing them by genetic algorithm, obtaining the fuzzy relationship matrix according to results of system parameters checked in real time and the membership functions, and producing the assessment results by multiple fuzzy comprehensive evaluations. The effect of genetic algorithm optimization is analyzed comparably in the example. The assessment results are objective and the error is the smallest.In order to further improve the intelligence and objectivity of the assessment of the actual operation training in the marine engine room, the regression problem of the intelligent assessment for marine engine room operation is studied with deep learning technique, and the intelligent assessment methods based on the deep learning are put forward. Features of mass sample data are transformed using Sparse Automatic Encoder(SAE). Sample features are learned deeply and used for classification evaluation, thus the better evaluation model is achieved after repeated training using effective data.According to the characteristics of assessment for practical engine-room operation, the method for detemining the hierarchical network structure of Deep Belief Networks(DBN) is provided. The Restricted Boltzmann Machines are trained by greedy training algorithm, and the assessment model comes into being after the network fine-tuning using BP algorithm finally. In the simulation experiments, the prediction results of the deep auto-encoder network, BP neural network and DBN model are verified comparably. It is found that the assessment error of DBN is the smallest, the maximum error is 2.45 point, and the minimum error is 0.09 point. It avoids the problem that the multilayer neural network is trapped in the local optimum. This method is available for carrying out intelligent assessment combined with engine-room simulator, automatic engine-room or intelligent engine-room of real ship.On the basis of drafting part of the performance standards for the engine room simulator, this dissertation also studies the performance evaluation system of the engine room simulator, and puts forward the efficiency evaluation method of the engine room simulator based on the cloud model. The change of the cloud center of gravity is measured by calculating the weighted deviations, and A VLCC marine engine room simulator is evaluated in the experiment. This dissertation provides a feasible evaluation method for the certification of marine engine room simulator in China, and makes some basic research work for the promotion of intelligent assessment based on engine room simulator.