Randomly quantized seismic data reconstruction using LSTM network

A system for performing a seismic survey of a geological formation, comprising:

• a seismic transducer located at a surface of the geological formation, wherein the seismic transducer is configured to generate a seismic source wavelet within the geological formation;
• an array of seismic sensors located at a surface of the geological formation, wherein the array of seismic sensors is configured to generate seismic traces, wherein the seismic traces include the seismic source wavelet convolved with seismic waves reflected from the geological formation including noise;
• a transmitter operatively connected to each of the seismic sensors of the array of seismic sensors, wherein the transmitter is configured to wirelessly transmit the seismic traces;
• a receiver configured to receive the seismic traces and generate a set of seismic time-series traces;
• a computing device including electrical circuitry, a memory storing program instructions, and a processor configured to execute the program instructions;
• a quantizer operatively connected to the processor, wherein the quantizer is configured to compress the set of seismic time series traces by randomly selecting a subset of samples of the set of seismic time-series traces;
• a trained long short-term memory (LSTM) neural network based autoencoder operatively connected to the processor, wherein the trained LSTM neural network based autoencoder is configured to deconvolve the randomly selected subset of samples and generate a deconvolved randomly selected subset of samples;
• and
• a computer screen operatively connected to the computing device, wherein the computing device is configured to reconstruct the seismic waves reflected from the geological formation from the deconvolved randomly selected subset of samples and render the reconstructed seismic waves reflected from the geological formation on the computer screen.

A method for performing a seismic survey of a volume of a below ground substrate, comprising:

• selecting an array of seismic sensors for performing the seismic survey;
• installing the array of seismic sensors on a ground surface above the volume of a below ground substrate in a pattern configured to detect reflected seismic waves from areas of interest within the volume;
• generating, with a seismic transducer, a seismic source wavelet within the volume;
• receiving, by an array of seismic sensors located at a surface of the geological formation, the seismic source wavelet convolved with seismic waves reflected from the geological formation including noise;
• generating, by the array of seismic sensors, seismic traces by transducing the seismic source wavelet convolved with the seismic waves reflected from the geological formation including the noise into electrical signals;
• transmitting, by each one of the array of seismic sensors, the seismic traces;
• receiving, by a receiver, the seismic traces from each one of the array of seismic sensors;
• generating, by the receiver, a set of seismic time-series traces;
• recording, in a memory of a computing device connected to the receiver, the set of seismic time-series traces;
• compressing, by the computing device, the set of seismic time series traces by quantizing a randomly selected subset of samples of the set of seismic time-series traces;
• deconvolving, by a trained long short-term memory (LSTM) neural network based autoencoder connected to the computing device, the randomly selected subset of samples, wherein deconvolving comprises removing the seismic source wavelet and the noise from each seismic time-series trace of the randomly selected subset of samples;
• generating, by the trained LSTM neural network based autoencoder, a deconvolved randomly selected subset of samples;
• reconstructing, by the computing device, the reflected seismic traces from the deconvolved randomly selected subset of samples;
• comparing, by the computing device, an amplitude of each reflected seismic trace to a threshold value and identifying spikiness when the magnitude exceeds the threshold value;
• determining, by the computing device, a pattern of the spikiness;
• comparing, by the computing device, the pattern of the spikiness to a database storing records which relate patterns of spikiness to structures in the volume;
• and
• displaying, on a computer screen of the computing device, the reflected seismic traces with the identification of the structures in the volume.