| Distributed technology has now become the development direction of intelligent systems.Combining different application scenarios and requirements,building a specific distributed system can greatly improve the operating efficiency of the system.However,the traditional distributed system is mainly applied to the algorithm training inside the computer.At the hardware level,due to the variety of hardware devices,single structure and unreasonable data cache,it is difficult to form coordination,which poses a great challenge to the construction of distributed system.According to the above problem,this paper based on the research of distributed system,application scenarios for emergency rescue at sea,by 9 assembly hardware simulation of rescue equipment,rescue equipment and design of collaborative cluster participation,realize the extracted from the complex sea environment huge amounts of multi-source heterogeneous data information,automatic generation of rescue plan,support the rescue commander efficient decision-making and action.To realize the federated learning distributed training system for rescue plan,the main research contents are as follows:(1)A federated learning based distributed node system training framework is proposed.9 hardware devices were assembled and configured to collaborate in networking,and a system network architecture combining Ad-Hoc and WiFi was designed.It effectively solves the problem of insufficient ability of cooperative training and communication difficulty of nodes.(2)A cache backup strategy based on node importance is proposed to improve global information redundancy.When a new piece of information is found,the problem of data loss in the maritime rescue environment is effectively solved by training and calculating the cache backup information of nodes with high importance,so as to ensure that the global information is not lost in the case of node disconnection.(3)Put forward the rescue plan generation strategy based on matching algorithm.Generate a reasonable rescue plan according to the system objectives in specific scenarios.By using the global information converged by each node and combining with the historical cache rescue scheme in the database,we define reasonable rules for matching,train and generate the final rescue scheme,and get the correct system result.(4)the combination of the above several kinds of structures,realize the whole system hardware and software platform,the strategy and algorithm and specific tests,proved that the system in a particular simulation scenarios,can effectively solve the multi-node cooperative coordination ability is insufficient,data leakage problems easily,and can correct operation system as a result,meet the demand of the operation of the particular scenario. |
