DAS same-well monitoring real-time microseismic effective event identification method based on deep learning

A DAS same-well monitoring real-time microseismic effective event identification method based on deep learning, comprising:

• constructing a DAS-based horizontal well microseismic monitoring system;
• constructing a training data set comprising microseismic event data, pipe wave data and background noise data with different types of labels;
• constructing a signal identification module composed of a deep neural network, wherein the deep neural network comprises a convolution sub-network and an attention mechanism sub-network;
• input data of the signal identification module uses two convolution layers to extract basic features, with a data size of 256×256, and a convolution kernel of 32×3×3;
• the convolution sub-network comprises 2 convolution layers, with a data size of 256×256, and a convolution kernel of 32×3×3, where first dimension is a number of convolution, second dimension and third dimension are convolution size;
• the attention mechanism sub-network comprises 1 convolution layer and 2 pooling layers, the convolution layer has a data size of 256×256, and a convolution kernel of 32×3×3;
• the 2 pooling layers comprise 1 average pooling layer and 1 maximum pooling layer, the 2 pooling layers are connected to each other by a connected layer, and then use the convolution layer with a data size of 128×128, and a convolution kernel of 64×3×3;
• the convolution sub-network and the attention mechanism sub-network are connected to each other by a connected layer, and then two convolution layers with a data size of 128×128, and a convolution kernel of 64×3×3 are used;
• then, 2 fully connected layers with a size of 1024×1 are used;
• an input of the signal identification module comprises the microseismic event data, pipe wave data and background noise data with different types of labels, and an output of the signal identification module comprises a predicted label value, network parameters are updated by using an error of the predicted label value and an accurate label value;
• the deep neural network uses cross entropy as a loss function to calculate an error of the deep neural network;
• training the signal identification module by using the training data set;
• preprocessing actual monitoring data, inputting the preprocessed data into the signal identification module to obtain an output result;
• marking microseismic events identified in the output result, and updating the marked microseismic events into the training data set;
• and
• adjusting and updating the signal identification module.