# US 11,441,404 B2 · Recurrent neural network model for bottomhole pressure and temperature in stepdown analysis > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,441,404 B2 - Document ID: US11441404B2 - Status: Granted - Kind: B2 - Application number: 17/043,129 - Filed: 2018-04-12 - Published: 2021-01-21 - Granted: 2022-09-13 - Earliest claimed: 2018-04-12 - Assignees: Landmark Graphics Corporation - Applicants: Landmark Graphics Corporation - Inventors: Srinath Madasu; Yogendra Narayan Pandey; Keshava Rangarajan - Web page: https://patentgrep.com/patent/US11441404B2 ## CPC - E21B 43/26: by forming crevices or fractures - E21B 47/06: Measuring temperature or pressure - G06N 3/044: Recurrent networks, e.g. Hopfield networks - G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] - G06N 3/063: using electronic means - G06N 3/084: Backpropagation, e.g. using gradient descent - G06N 3/09: Supervised learning - E21B: EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A method for fracturing a formation is provided. Real-time fracturing data is acquired from a well bore during fracturing operation. The real-time fracturing data is processed using a recurrent neural network trained using historical data from analogous wells. A real-time response variable prediction is determined using the processed real-time fracturing data. Fracturing parameters for the fracturing operation are adjusted in real-time based on the real-time response variable prediction. The fracturing operation is performed using the fracturing parameters that were adjusted based on the real-time response variable prediction. ## Claims ### Claim 1 A method for fracturing a formation, the method comprising: acquiring real-time fracturing data from a well bore during fracturing operation; processing the real-time fracturing data using a recurrent neural network trained using historical data from analogous wells; determining a real-time response variable prediction using the processed real-time fracturing data; estimating a total number of open perforations in the well bore and tortuosity friction in the well bore based on the determined real-time response variable; adjusting fracturing parameters for the fracturing operation in real-time based on the real-time response variable prediction, the estimated total number of open perforations in the well bore, and the estimated tortuosity friction in the well bore; and performing the fracturing operation using the fracturing parameters that were adjusted based on the real-time response variable prediction. ### Claim 2 (depends on 1) The method of claim 1, wherein the response variable comprises at least one of a bottomhole pressure of a well bore and temperature of the well bore. ### Claim 3 (depends on 1) The method of claim 1, wherein the real-time fracturing data comprises multidimensional spatial-temporal data. ### Claim 4 (depends on 1) The method of claim 1, wherein the step of determining the real-time response variable prediction further comprises determining temporal correlations within the processed real-time fracturing data. ### Claim 5 (depends on 1) The method of claim 1, wherein the recurrent neural network comprises one or more long short-term memory (LSTM) layers, wherein the one or more LSTM layers are arranged in a sequence, and wherein each of the one or more LSTM layers is configured to perform operations comprising: receiving a layer input, wherein the layer input is the real-time fracturing data or a layer output generated by a preceding LSTM layer in the sequence; generating an LSTM output based on the layer input and a previous projected output; and generating, by processing through the respective LSTM layer, a projected output by applying a matrix of current values of weights to project the LSTM output. ### Claim 6 (depends on 5) The method of claim 5, wherein the recurrent neural network comprises one or more Gated Recurrent Units (GRU) layers arranged in sequence. ### Claim 7 (depends on 1) The method of claim 1, wherein the recurrent neural network comprises LSTM layers and wherein each LSTM layer has weights and biases associated therewith. ### Claim 8 A system for fracturing a formation, the system comprising: a downhole tool having a plurality of sensors thereon configured to acquire real-time fracturing data from a well bore; a processor in data communication with the downhole tool; and a memory device coupled to the processor, the memory device containing a set of instructions that, when executed by the processor, cause the processor to: acquire the real-time fracturing data from the well bore during fracturing operation; process the real-time fracturing data using a recurrent neural network trained using historical data from analogous wells; determine a real-time response variable prediction using the processed real-time fracturing data; estimate a total number of open perforations in the well bore and tortuosity friction in the well bore based on the determined real-time response variable; and adjust fracturing parameters for the fracturing operation in real-time based on the real-time response variable prediction, the estimated total number of open perforations in the well bore, and the estimated tortuosity friction in the well bore. ### Claim 9 (depends on 8) The system of claim 8, wherein the response variable comprises at least one of a bottomhole pressure of a well bore and temperature of the well bore. ### Claim 10 (depends on 8) The system of claim 8, wherein the real-time fracturing data comprises multidimensional spatial-temporal data. ### Claim 11 (depends on 8) The system of claim 8, wherein the set of instructions that causes the processor to determine the real-time response variable prediction further causes the processor to determine temporal correlations within the processed real-time fracturing data. ### Claim 12 (depends on 8) The system of claim 8, wherein the recurrent neural network comprises one or more long short-term memory (LSTM) layers, wherein the one or more LSTM layers are arranged in a sequence, and wherein each of the one or more LSTM layers is configured to perform operations comprising: receiving a layer input, wherein the layer input is the real-time fracturing data or a layer output generated by a preceding LSTM layer in the sequence; generating an LSTM output based on the layer input and a previous recurrent projected output; and generating, by processing through the respective LSTM layer, a projected output by applying a matrix of current values of weights to project the LSTM output. ### Claim 13 (depends on 8) The system of claim 8, wherein the recurrent neural network comprises one or more Gated Recurrent Units (GRU) layers arranged in sequence.