Dynamic System Modeling Method Of Non-contact Grab System Based On Deep Learning

With the intensification of the energy crisis,the development of solar energy is increasingly important.As the wafer becomes thinner and larger It challenges the automatic handling of photovoltaic wafers in the production process.In practical engineering applications,due to Bernoulli The uneven suction of the suction cup leads to impact during the handling process.From the perspective of control The scene further improves the stability of the wafer handling process.Therefore,for the control needs in the actual project Establishing a more accurate system model is an important problem to be solved urgently.From the perspective of engineering The short-term and short-term memory neural network model based on grey wolf algorithm optimization and the autoregressive moving average length The short-term memory neural network hybrid model is used to model the Bernoulli non-contact suction cup grabbing system To verify the validity of the model,an experimental data collection platform is built,and the accuracy of the model is verified through open-loop control.The research contents of this paper are summarized as follows:First,establish a neural network based on Long Short-Term Memory(LSTM)Network wafer grabbing system model,discuss the influence of manually configuring network parameters on model error,and establish a comparison Experiment to verify the effectiveness of the model.Aiming at the disadvantage that network parameters need to be manually configured and verified,a Grey Wolf Optimizer(GWO)is proposed The optimized LSTM neural network uses GWO algorithm to optimize the parameters of LSTM and realize automatic configuration and verification of network parameters.Secondly,for the non-contact handling system,the traditional LSTM model is improved and a A prediction framework combining the traditional linear model ARIMA with the nonlinear model LSTM is constructed,This model realizes the effective integration of linear prediction and nonlinear prediction,and forms a kind of comprehensive prediction model A new combined forecasting model: ARIMA-LSTM.Number of data collected through the experimental platform It is reported that the model training set and test set are constructed,and the performance index analysis is compared with the traditional model The experimental results show that this method can improve the accuracy of model identification.Third,in order to verify the availability of the model,start with the static and steady-state characteristics of the system,and design the model verification letter The difference between the identification model and the real system is analyzed from different angles.At the same time,the controller is designed for the model.The simulation results show that on the two identification models,the controller It can make the model output track the reference output very well.Control signal of Bernoulli suction cup handling system through simulation calculation The open-loop experiment shows that the model has certain usability in the real system.