Research On Motion Control Of Bionic Robotic Fish Based On LSTM Neural Network

Fish,as a kind of creature with an early origin on earth,has evolved unparalleled swimming ability through hundreds of millions of years of evolution and natural selection,which is of great significance to the research of underwater vehicles.Deep learning has played a great role in improving the autonomy and perception of robotic fish underwater.In this paper,the robotic fish is taken as the research object,the deep learning algorithm is applied to the research of robotic fish motion control,and the robotic fish platform is built.The robotic fish motion control model was established and analyzed,and the model was optimized and improved The feasibility of deep learning for the robotic fish motion control method was verified through simulation and experiment.The specific research contents are as followsFirst,it summarizes the technical background,research value and significance of robotic fish,expounds the development of learning algorithms on robotic fish,summarizes the development of technical methods,and finally briefly introduces the chapter structure and content of this articleSecondly,the mechanical structure of the robotic fish used in this article and the research platform of the bionic robotic fish system are introduced in detail.How to control the robotic fish movement,how to communicate with the robotic fish,how to monitor the robotic fish swimming modalities,and return real-time data are given a detailed description.The software and hardware parts of the robotic fish system platform are also described.Experimental data analysis of the robotic fish swimming modals is doneAdditionally,in order to study the application of neural networks to robotic fish,a robotic fish motion control model based on long-short memory neural network was established.The modeling steps,possible problems and parameter settings were described in detail.The experimental design of the motion state of the robotic fish was carried out,and the simulation and experimental results were analyzed.Furthermore,the long-short-term memory network is compared with the backward propagation neural network and the recursive pulse neural network,and the advantages and disadvantages of the neural network are compared through simulation and experimental analysis.Analysis of the experimental results shows that training the same experimental data,the long-short-term memory neural network can converge faster,and the error is smaller,which verifies the effectiveness of the modelFinally,the robotic fish motion control model is optimized and improved for the two evaluation indexes of the training time of the loss function value.On the basis of the previous long-term and short-term memory neural network model,the robotic fish is trained for movement.By changing the model parameters,the straight swim and turning movement of the robotic fish are observed,and the simulation and experimental design for the influence of different parameters on the movement state of the robotic fish are carried out.The running results are compared to verify the validity of the model.