Product quality prediction method based on dual-channel information complementary fusion stacked auto-encoder

A product quality prediction method based on a dual-channel information complementary fusion stacked auto-encoder, comprising:

• step 1:
• collecting sensor data and product quality data of an industrial process to obtain a training data set Sl={xi, yi}i=1M, where, x represents an input sample, y represents a sample label, and M represents a number of labeled samples;
• step 2:
• constructing a dual-channel information complementary fusion stacked auto-encoder model for product quality prediction, and training the dual-channel information complementary fusion stacked auto-encoder model by using the training data set;
• wherein the dual-channel information complementary fusion stacked auto-encoder model comprises a stacked auto-encoder, a top-down fusion channel, a down-top fusion channel, and a gating layer;
• the top-down fusion channel comprises a plurality of information complementary fusion layers;
• the information complementary fusion layer uses hidden layer information of the stacked auto-encoder to perform information fusion, and uses the fused information to obtain an output value of the corresponding information complementary fusion layer;
• the fused information is transmitted to a lower layer;
• the down-top fusion channel comprises a plurality of information complementary fusion layers;
• the information complementary fusion layer uses the hidden layer information of the stacked auto-encoder to perform information fusion, and uses the fused information to obtain an output value of the corresponding information complementary fusion layer;
• the fused information is transmitted from the lower layer to an upper layer;
• the gating layer is used for collecting the fused information and the output values of a plurality of information complementary fusion layers in two fusion channels, then calculating weights of the output values by using the fused information, and finally performing weighted sum using the weights and the output values to obtain an overall prediction result of the dual-channel information complementary fusion stacked auto-encoder model;
• and
• step 3:
• collecting industrial field work data and inputting the industrial field work data into the trained dual-channel information complementary fusion stacked auto-encoder model, so that the trained dual-channel information complementary fusion stacked auto-encoder model outputs a corresponding product quality prediction result.