Research And Implementation Of Main Steam Temperature Prediction System Based On Deep Learning

During the operation of large-scale thermal power plants,main steam temperature is one of the important parameters.Establishing an accurate prediction model for main steam temperature is the key to predictive control of main steam temperature,and it is of great significance to ensure the safe,stable and normal operation of unit equipment.However,the main steam temperature has the characteristics of delayed extension,non-linearity,time-varying,and many influencing factors.The accuracy of prediction models established by traditional mechanism-based methods is low.With the rapid development of intelligent control technology and deep learning technology,applying deep learning technology to predictive control of main steam temperature is a new research direction.It is in line with the "Made in China 2025" action plan for industrial intelligence.Based on the above background and research status,this paper has carried out the research of the main steam temperature prediction system based on deep learning:(1)Aiming at the problem that the main steam temperature is difficult to predict due to non-linearity,time-varying,and many influencing factors,an improved first-order autoregressive gate recurrent unit main steam temperature prediction model is proposed.First,analyze the main steam temperature system data and use the grey relational analysis method to select the five main influence indexes that affect the main steam temperature.Then three time series data prediction methods are summarized,The improvement is made on the basis of the sequence-to-sequence prediction method most suitable for the main steam temperature prediction.Finally,the idea of autoregressive model is combined,and the main steam temperature prediction model is constructed with the gate recurrent unit neural network as the core.(2)Aiming at the problem of low accuracy of main steam temperature prediction,an improved autoregressive gate recurrent unit main steam temperature prediction model with self-attention mechanism is further proposed.First,conduct an in-depth analysis of the main steam temperature,find its unique characteristics and existing problems,and add linear mapping to the network in a targeted manner.Then,a self-attention mechanism is introduced to highlight the degree of influence of the input at different times on the main steam temperature,so as to solve the problem of low prediction accuracy caused by the delay of the main steam temperature.Finally,a large number of experiments were carried out using historical operating data of 800MW thermal power units in a power plant.Choosing the best model,through the display diagram and related evaluation indicators,the four time series forecasting models are compared with the improved model proposed in this article.When the error threshold is set to 1 celsius,the improved model proposed in this paper has the highest accuracy.(3)Based on the above research content,this paper designs and implements a main steam temperature prediction system based on deep learning.The system includes 5 modules:data storage,data processing,model training,model testing,and main steam temperature prediction.Simplifies the user’s operation and enhances the interactive experience.