Abstract: The feed water pump system is an indispensable equipment in thermal power plants, and the state trend prediction of the feed water pump system helps to maintain the normal operation of thermal power plants. In order to reduce the loss caused by the failure of feed water pump system, this paper establishes a hybrid model of particle swarm optimization (PSO) algorithm and long short-term memory neural network (LSTM) on the basis of principal component analysis (PCA) to predict the state trend and fault warning of feed water pump system. The method firstly uses PCA to downscale the multidimensional parameters and extract the main features; secondly, the extracted feature parameters are inputted into the LSTM model optimized by PSO, and the warning is given when the prediction result exceeds the threshold value. The experimental results show that the PCA-PSO-LSTM model possesses high prediction accuracy and good stability.

    Key words: feed water pump system; principal component analysis; particle swarm optimization algorithm; long short-term memory neural network; trend prediction; fault warning