Research On Temperature Control Method For Biolysozyme Fermentation Based On LSTM Neural Network

Lysozyme is an anti-infective substance with antiviral,antibacterial,hemostatic,pain-relieving properties and accelerates tissue recovery.In controlling the fermentation temperature of lysozyme,the traditional Proportional,Integral,Derivative(PID)control algorithm is difficult to adjust the parameters when applied to a complex system with large hysteresis and time-varying nonlinearity,which affects the control effect.Theoretical analysis shows that heuristic algorithms and neural networks can improve the speed of controller parameter tuning by their excellent merit-seeking ability,thus improving the difficulty of parameter tuning and the accuracy and stability of the control system.The paper focuses on the parameter tuning of PID controllers based on heuristic algorithms and neural networks,and the main contents and innovations are as follows:(1)In order to accurately simulate the temperature system of lysozyme fermenter,the basic dynamic equations of fermenter are established in this paper based on the intrinsic mechanism of fermenter,so as to give the mechanical temperature model of fermenter described by dynamic equations.The step response curve method is used to empirically model the fermenter temperature system through the measured temperature data of the fermenter.After comparing and analyzing the above two models with the lysozyme fermentation temperature test data collected in the field,it is concluded that the empirical modeling can more accurately describe the temperature system of the fermenter.(2)To address the problem of large overshoot and slow stabilization rate generated in the process of controlling temperature,the paper proposes a PID parameter adjustment method based on Long Short-Term Memory(LSTM)network.The method utilizes the ability of LSTM neural network to learn the information of time series and obtain the trend of change between error series under continuous sampling time,so as to adjust the network weights more reasonably and realize the fast rectification of PID parameters Kp、Ki and Kd.By incorporating the PID algorithm into the back propagation of the LSTM neural network,the PID controller parameters can change with the environment,thus improving the adaptability of the fermenter temperature system to the external environment as well as its stability.(3)The paper proposes a whale optimization algorithm(LRWOA)based on the Levy flight strategy and the random swim strategy.The algorithm incorporates the Levy flight strategy to improve the global optimization capability of the algorithm and the random swim strategy to improve the local optimization capability of the algorithm,which overcomes the problem that the whale optimization algorithm tends to fall into local extremes too early when dealing with complex problems.The experimental results show that the algorithm has a stronger optimization-seeking ability compared with the traditional optimization algorithm.(4)In order to improve the instability problem arising from the initialization parameters of LSTM neural networks,the paper proposes a PID parameter rectification method based on LRWOALSTM.The method makes use of the good optimization-seeking ability of the LRWOA algorithm to overcome the problem that may fall into the local optimal solution in advance brought by the randomized parameters,reduce the problem that the neural network training divergence does not converge to jump back and forth horizontally,and improve the stability of the initialization of the LSTM neural network parameters,thus speeding up the tuning of the PID controller parameters.The experimental results show that the LRWOA-LSTM-PID control method proposed in this paper improves the accuracy and stability of lysozyme fermentation temperature control.