Method and system for managing carbon dioxide supplies and supercritical turbines using machine learning

A system, comprising:

• a plurality of wells in a geological region of interest, wherein the plurality of wells comprises a production well, a first injection well, and a second injection well;
• a plurality of supercritical carbon dioxide (sCO2) turbines;
• a first control system coupled to the first injection well;
• a second control system coupled to the production well;
• a third control system coupled to an sCO2 turbine among the plurality of sCO2 turbines;
• a fourth control system coupled to the second injection well;
• a carbon dioxide management network;
• and
• a carbon dioxide manager comprising a computer processor, wherein the carbon dioxide manager is connected to the first control system, the second control system, the third control system, the fourth control system, and the plurality of wells over the carbon dioxide management network,
• wherein the first control system is configured to perform, using the first injection well, a stimulation operation based on injecting a predetermined amount of carbon dioxide into a subsurface of the geological region of interest,
• wherein the predetermined amount of carbon dioxide comprises a gas tracer that is used to detect one or more carbon dioxide leakages from the geological region of interest,
• wherein the carbon dioxide manager is configured to:
• obtain gas tracer data for the geological region of interest from the plurality of wells,
• wherein the gas tracer data is acquired using a plurality of chromatography-based sensors disposed in the plurality of wells;
• obtain turbine data regarding the plurality of sCO2 turbines, wherein the turbine data comprises historical electric-power data for the plurality of sCO2 turbines;
• obtain acquired production data from at least one production well among the plurality of wells;
• obtain carbon emission data for the plurality of wells;
• determine predicted production data and predicted carbon emission data for the production well using a first machine-learning model, the gas tracer data, the turbine data, and the carbon emission data;
• and
• determine error data based on a mismatch between first predicted production data and the acquired production data;
• update the first machine-learning model to produce an updated machine-learning model,
• wherein the first machine-learning model is updated using the error data and a machine-learning algorithm comprising a backward propagation algorithm;
• determine, using the first machine-learning model, predicted carbon dioxide demand data for the sCO2 turbine for a period of time based on the historical electric-power data,
• wherein the first machine-learning model comprises a non-linear autoregressive neural network with an exogenous input (NARX) model, and
• wherein the NARX model comprises at least one hidden layer and outputs the predicted production data and the predicted carbon dioxide demand data;
• and
• transmit one or more commands to the third control system based on the predicted production data, the predicted carbon dioxide demand data, and the predicted carbon emission data;
• and
• wherein the third control system is configured to:
• distribute, in response to the third control system receiving the one or more commands, a flow of carbon dioxide between the second injection well and the sCO2 turbine based on the predicted production data, the predicted carbon dioxide demand data, and the predicted carbon emission data, and
• generate electric power using the sCO2 turbine and a portion of the flow of carbon dioxide that is distributed to the sCO2 turbine, wherein generating the electric power is performed using a Brayton cycle process,
• wherein the one or more commands achieve a predetermined production rate at the production well and a predetermined carbon footprint, and
• wherein the predetermined carbon footprint corresponds to carbon emissions that are produced by the production well, the second injection well, and the sCO2 turbine.