Research On Weight Estimation Model And Algorithm For Launching Ship Based On Deep Learning

The application of ship airbag launching technology is becoming wider and wider,but the security problems that follow are increasing.In the process of ship launching,how to control accurately the center of gravity is a major problem to ensure the safety of the launching project.In addition,accidents caused by the deviation of the ship's actual gravity center from the position of the design gravity center occur from time to time after launching.Therefore,how to estimate accurately the ship's weight and center of gravity during shipbuilding is another problem in the shipbuilding industry.The reason for this is how to accurately estimate the weight of the ship and its components during the shipbuilding process and the ship's launching process is the key to the above two problems.At present,the actual ship weight of the launching process is often replaced by the designed ship weight,and the ship weight under construction is often calculated by the cumulative method.In view of the above methods,such as large error,cumbersome operation and rough calculation,this paper proposes a model and algorithm for estimating the weight of launching ship based on deep learning,which can accurately estimate the actual ship weight and the weight of each component.The main work of this paper is divided into four parts:(1)Acquisition and preprocessing of airbag state data: Using the experimental equipment of the research group to build the launching airbag state monitoring platform to obtain real-time state data of the airbag,and through data preprocessing and data visualization,data purification and correlation research are realized.The estimation algorithm model provides a clean high-dimensional feature data set.(2)Feature extraction of high-dimensional data: The underlying deep learning network model of the estimation model is used to achieve dimensionality reduction of the data set.This paper designs a deep stack sparse autoencoder(DSSAE)network,trains the network through a combination of supervised and unsupervised learning,the parameter structure experiment and grid search method are used to determine the network structure and find the optimal parameters to achieve high-level feature extraction.(3)Regression estimation of the weight of the launching ship.The weight estimation of the launching ship is estimated by the model's top-level support vector regression machine(SVR)model.This paper studies the optimization algorithm of SVR model parameters through a large number of experiments.Two optimization algorithms,genetic algorithm(GA)and Particle swarm optimization algorithm(PSO)are applied to find the optimal parameters of SVR.Hundreds of iterative experiments prove that PSO is beneficial to improve the accuracy of the SVR model.(4)Construction and implementation of the experimental platform for estimating the weight of the launching ship: The platform consists of two parts: the airbag state monitoring platform and the DSSAE-PSO-SVR estimation algorithm.After setting up the experimental environment,the latter uses the fusion model DSSAE-PSO-SVR algorithm to carry out the estimation experiment,and compares it with DBN-PSO-SVR,BPNN two deep learning algorithms and PSO-SVR regression algorithm.The experiment proves that the DSSAE-PSO-SVR 5-layer network has good robustness and the average relative error of the algorithm is 4.21%.The average relative errors of the other three algorithms are 6.36%,10.36%,and 14.72%,respectively.Finally,the algorithm model is embedded in the software production interface to realize the data of the input airbag state in the software system,and the weight of the launching ship is output.During the test,the system runs smoothly and the estimation result reaches the expected accuracy.Through the above four parts of work,this paper builds and implements the weight estimation model of the launching ship.The model can be used not only to estimate the weight of the hull and its components in ship construction and launching,but also to estimate the weight of the ship as the shipyard weight.In addition,the model is equally applicable to the weight estimation of large objects other than ships.