Identification Of Main Steam Temperature Object Based On LSTM

The main steam temperature of thermal power units has the characteristics of large delay,large inertia,non-linearity and time-varying,and is an important monitoring object in the production process of thermal power plants.Main steam temperature control is an important link to ensure the safety and smooth operation of the u nit and to improve economic efficiency.The mechanism identification method is to conduct a lot of systematic analysis on the controlled object,rely on the empirical formula to set the parameters,and the model data identification result is poor in generalization ability.The analysis of a large amount of data on the spot and the mining of valuable information for the prediction of the main steam temperature model have a broader meaning.In this paper,Long Short-Term Memory Network(LSTM)is used to identify the main steam temperature system model,and a multi-input and double-output main steam temperature model is established for a 1000 MW ultra-supercritical unit.First filter and normalize the collected data.PCA processing is performed on the field data,and the data dimension after PCA processing is obtained through the decision solution.Take MAE as the model’s error evaluation index.Then,the principle of LSTM is summarized,and the design steps of recursive multi-step prediction model and non-recursive multi-step prediction model are introduced.Since the non-recursive model needs to establish a neural network model according to the length of the prediction time,this paper determines Model design steps.The neural network optimization algorithm is selected as Adam,and the main steam temperature system is identified based on LSTM.According to the adjustment of the parameters and structure of LSTM,the main steam temperature system model of data input dimension change,LSTM parameter change,LSTM structure change,and fully connected layer activation function change were established respectively.At the same time,a model that can better reflect the system characteristics was given.Main steam temperature model.Finally,the steam temperature system model based on LSTM is compared with the steam temperature system model based on BP network.The results show that the BP algorithm has a fast operation speed,the identification model has a large loss function value and low accuracy;the loss function of the main steam temperature control model based on LSTM is small,the error index MAE of the training data is small,and the model fits High;the main steam temperature control strategy that combines the LSTM neural network algorithm and PCA data preprocessing has great field application value.