The Bidirectional Information Fusion Using an Improved LSTM Model

Algorithm 1

The soft computing of the coal quality analyses by using DFAS-LSTM model.

Input: Data of the historical conventional measurement points of the coal-fired power plants Xh; Real-time conventional measurement data of the coal-fired power plants Xr; Coal quality test data of the coal-fired power plants yr;
Output: Real-time coal quality data in the furnace of the coal-fired power plants y;
(1)	Remove noise from historical data Xh and filter it;
(2)	Standardize the data, process the standard data by PCA and ICA algorithm to obtain data Xi;
(3)	Initialize weight parameters, batch the data to get Xib, and input Xib into DFAS-LSTM;
(4)	Use alertness mechanism to process data Xib and obtain data Xa;
(5)	Data Xa input to LSTM, which is based on improved activation function and fusion structure;
(6)	Use the coal quality test data yr to compare with the output of the neural network to obtain the cost function C;
(7)	Use the optimizer to optimize the cost function C by updating the weight parameters of the neural network;
(8)	After the model is stable, the optimization ends and the model parameters are solidified;
(9)	Take Xr as input, output coal quality information y in real time.