A method and system for determining frictional resistance of a fracturing fluid on site

Abstract

The application provides a fracturing fluid friction resistance field determination method and system, relates to the technical field of fracturing fluid performance parameter optimization, and comprises the following steps: deploying a sensor network at a fracturing construction site, collecting fluid parameters and geological environment parameters in real time; performing feature engineering processing based on the collected data, calculating derived features such as Reynolds number, dynamic viscosity adjustment factor and formation pressure gradient; training a prediction model by using a machine learning algorithm; inputting real-time data into the prediction model to predict the friction resistance under the current fracturing fluid conveying state; and automatically adjusting the fracturing fluid formula or pumping parameters according to the prediction result. The application realizes real-time monitoring and dynamic prediction of the fracturing fluid friction resistance, continuously optimizes the prediction model through a feedback learning mechanism, solves the problems of poor timeliness and low accuracy of the traditional method, significantly improves the adaptability to complex geological environments, and greatly improves the intelligent level and economic benefits of fracturing operations.

Description

Technical Field

[0001] This invention relates to the field of fracturing fluid performance parameter optimization technology, specifically to a method and system for determining the frictional resistance of fracturing fluid in the field. Background Technology

[0002] Fracturing fluid plays a crucial role in oil and gas extraction, used to improve formation permeability and increase oil and gas well production. Fracturing fluid is delivered to the formation under high pressure generated by surface equipment, causing the formation to fracture and extend, creating fractures. It also carries proppant into the fractures to keep them open, forming flow channels with conductivity.

[0003] The frictional resistance of fracturing fluid in pipelines directly affects pumping efficiency and construction costs. Excessive frictional resistance leads to increased pump pressure consumption, limited discharge capacity, and even equipment damage. Accurately measuring the frictional resistance of fracturing fluid is crucial for optimizing fracturing operations, guiding adjustments to fracturing fluid formulations and optimizing pumping parameters to ensure efficient and economical delivery of fracturing fluid to the target formation. In the stimulation of unconventional reservoirs such as shale oil and gas and tight oil and gas, fracturing operations involve high discharge rates, large quantities of fluids, long durations, and diverse fluid types, placing increasingly stringent requirements on the monitoring of fracturing fluid performance.

[0004] In existing technologies, the determination of fracturing fluid frictional resistance mainly involves offline testing using laboratory-prepared samples. On-site, samples are manually taken from the mixing truck, and basic performance parameters such as pH and viscosity of the fracturing fluid are measured using pH test paper and a six-speed rotational viscometer. This method has significant drawbacks: first, it lacks timeliness, as the time from sampling to obtaining test results is relatively long, failing to reflect changes in fracturing fluid performance in real time; second, it has low accuracy, with large human operational errors, and the results obtained using a six-speed rotational viscometer for low-viscosity slickwater fracturing fluid exhibit significant errors; third, it cannot provide online tracking and evaluation, preventing technicians from promptly monitoring fracturing fluid performance indicators and making timely adjustments to address performance changes.

[0005] Traditional friction resistance measurement devices, consisting of a storage tank unit, a power pump, and test pipelines, have identical orifice diameters in each test pipeline and a fixed shear rate range, making them unsuitable for testing under varying flow rates and viscosities. Indoor pipeline friction meters have long pipelines, occupy a large space, and suffer significant wear from the proppant in the sand-carrying fluid, resulting in high maintenance costs. Existing devices only provide passive measurements of friction resistance; after testing, data analysis and parameter adjustments are required manually, leading to low automation and an inability to meet the requirements of continuous on-site operation.

[0006] The performance of fracturing fluids is influenced by a variety of factors, including additive concentration, temperature, pressure, and shear rate, all of which are dynamically changing during actual operations. Measurements taken at a single point in time cannot fully reflect the performance evolution of the fracturing fluid throughout the entire delivery process. Especially under complex geological conditions, parameters such as formation pressure, porosity, and permeability have a significant impact on the flow characteristics of fracturing fluids, and traditional testing methods cannot incorporate these geological parameters, thus limiting prediction accuracy.

[0007] Chinese patent document CN105203295A discloses a measuring device and method for measuring the frictional resistance of fracturing fluid. It discloses a technical solution that expands the shear rate testing range by setting test pipelines with different orifice diameters, which has the technical effect of improving the applicability of the measuring device. However, it still has the problems of offline static measurement, inability to reflect the dynamic changes of fracturing fluid performance in real time during construction, and lack of intelligent control capabilities.

[0008] Chinese patent document CN120195343A discloses an online detection and control method and system for fracturing fluid performance. It discloses a technical solution that uses sensors to detect parameters such as pH, viscosity, salinity, and drag reduction and controls them through a remote control center. This method improves the timeliness and accuracy of detection. However, it still has problems such as data processing remaining at the level of simple threshold judgment, lacking predictive capabilities based on machine learning, and being unable to actively adapt to changes in complex geological environments. Summary of the Invention

[0009] The purpose of this invention is to provide a method and system for determining the friction resistance of fracturing fluid in real time, dynamically predicting the trend of performance changes, and automatically optimizing construction parameters.

[0010] To achieve the above objectives, the present invention provides the following technical solution: A method for determining the frictional resistance of fracturing fluid in the field includes the following steps; S1: Deploy a sensor network at the fracturing site; S2: The sensor network is used to collect fluid parameters during the fracturing fluid delivery process in real time and to obtain geological environment parameters; S3: Based on the collected fluid parameters and geological environment parameters, feature engineering is performed to obtain input features; S4: Based on the input features extracted in S3 and the measured friction resistance data collected in historical fracturing operations, a machine learning algorithm is used to train a prediction model and establish a nonlinear mapping relationship between the input features and the friction resistance; wherein, the input features include derived features reflecting the non-Newtonian fluid properties of the fracturing fluid and derived features reflecting the influence of geological parameters; S5: Input the real-time collected data into the trained prediction model to predict the fracturing fluid friction resistance under the current fracturing fluid delivery state in real time; S6: Adjust the fracturing fluid formulation or pumping parameters according to the predicted results of the fracturing fluid friction resistance so that the friction resistance reaches the preset target range; and feed back the adjusted actual friction resistance measured value to the prediction model to perform incremental training or retraining of the prediction model to continuously optimize the prediction model.

[0011] Further: In S1, the sensor network includes a pressure sensor, a temperature sensor, a flow sensor, a differential pressure friction resistance measuring device, and a geological parameter acquisition device.

[0012] The differential pressure friction resistance measuring device includes an upstream pressure sensor and a downstream pressure sensor spaced at a predetermined distance on the fracturing fluid delivery pipeline. By measuring the pressure difference between the two ends of the same pipe segment, and combining this with the pipe segment length and inner diameter, the frictional pressure drop per unit pipe length is calculated, thus characterizing the frictional resistance of the fracturing fluid in the pipeline in real time. Specifically, the upstream and downstream pressure sensors are installed at two cross-sectional positions at a known distance on the fracturing fluid delivery pipeline. The predetermined distance is determined based on the nominal diameter of the pipeline, preferably 50-200 times the inner diameter of the pipe, to ensure sufficient flow field development and accurate measurement results. The upstream and downstream pressure sensors synchronously collect the pressure values ​​P at both ends of the pipe segment. up and P down The pressure difference ΔP = P up - P down This is the frictional pressure drop of that pipe section.

[0013] The geological parameter acquisition device includes a downhole pressure gauge and a logging data interface. The downhole pressure gauge is installed at a preset depth inside the wellbore to monitor bottomhole flowing pressure and formation pressure in real time. The logging data interface receives static geological parameters such as formation porosity and permeability obtained from previous well logging interpretations, using them as prior input data for the prediction model. It should be noted that formation porosity and permeability are static geological parameters obtained through previous well logging interpretations and are input into the prediction model as known prior parameters during fracturing operations; they do not need to be collected in real time during the operation. Formation pressure, however, is monitored in real time using the downhole pressure gauge.

[0014] Further: In S2, the fluid parameters include fluid viscosity, flow rate, temperature, and density, and the geological environment parameters include formation pressure and porosity. Wherein: Fluid viscosity is collected in real time by an online viscosity sensor installed on the fracturing fluid delivery pipeline. The online viscosity sensor adopts the vibration or rotation viscosity measurement principle and can continuously measure the apparent viscosity of the fracturing fluid under high pressure flow conditions in the pipeline. The flow velocity is collected in real time by an electromagnetic flow meter or an ultrasonic flow meter, and the average flow velocity is calculated based on the cross-sectional area of ​​the pipe. Temperature is collected in real time by thermocouples or platinum resistance temperature sensors installed on the outer wall of the pipe or inserted inside the pipe. Density is collected in real time by a Coriolis mass flow meter or an online density meter, or calculated based on the mass concentration of each component in the fracturing fluid formulation; Formation pressure is obtained through real-time monitoring using downhole pressure gauges; Porosity, a static geological parameter obtained through prior well logging interpretation, is provided as prior input data to the prediction model. Before fracturing operations, the porosity of the target formation is interpreted and evaluated based on well logging data, and the interpretation results are stored in the data acquisition module and retrieved during prediction model calculations.

[0015] Furthermore: the feature engineering process in S3 includes: S31: Clean and normalize the original data to obtain the processed data; S32: Calculate derived features based on fluid dynamics principles for the processed data; S33: Based on the derived features, after removing features with a correlation greater than a preset threshold through correlation analysis, the feature set is obtained after encoding, and the features in the feature set are used as input features.

[0016] Furthermore: the derived feature includes the Reynolds number, and the formula for calculating the Reynolds number is:

[0017] in, It is fluid density. It's the flow rate. It is the pipe diameter. It is fluid viscosity. It is the Reynolds number.

[0018] Furthermore, the derived feature also includes a dynamic viscosity adjustment factor, which is calculated based on the Carreau-Yasuda model, and the calculation formula is as follows:

[0019] in, It is the apparent viscosity. and These are the viscosities at zero shear rate and infinite shear rate, respectively. It's a relaxation time. For shear rate, and These are the parameters of the Carreau-Yasuda model; the parameters η0 and η of the Carreau-Yasuda model are... ∞ λ, α, and n were determined by rheological experiments on fracturing fluid samples.

[0020] Furthermore, the derived feature also includes the formation pressure gradient, which is calculated using the following formula:

[0021] in, For formation pressure gradient, and These represent the formation pressures at two different depths. It is the perpendicular distance between these two points.

[0022] Furthermore, the feature engineering process in S3 also includes extracting time series features and using sine transform sin(2πt / T) and cosine transform cos(2πt / T) to capture periodic influencing factors, where t is the current time and T is the period.

[0023] Furthermore, a feature importance evaluation step is included between steps S3 and S4, employing a feature importance scoring method based on random forests. The calculation formula is as follows:

[0024] Where I is the impurity of a node, t is the branch number of the tree, r is the total number of branches in the tree, F is the feature, and E[I(t)|F] is the average impurity of nodes after feature F is split; features with importance scores below the preset importance threshold are removed, and features with high importance scores are retained for model training.

[0025] Further: The machine learning algorithm in S4 includes at least one of random forest, gradient boosting tree, neural network or long short-term memory network; the training process of the prediction model includes: dividing the historical dataset into training set and validation set according to a preset ratio, using input features as model input and measured friction resistance value as model output label, calculating prediction error through loss function and updating model parameters using optimization algorithm until the model's performance index on the validation set reaches the preset requirements; the performance index includes coefficient of determination R² and root mean square error RMSE.

[0026] Further: The adjustment in S6 includes: adjusting at least one of the fracturing fluid additive concentration, pumping rate or pressure according to the deviation between the predicted frictional resistance value and the preset target range through a feedback learning mechanism; and feeding back the adjusted actual frictional resistance measured value to the prediction model, and performing incremental training or retraining on the prediction model to continuously optimize the prediction model.

[0027] A fracturing fluid friction resistance field determination system for implementing the above method includes: Sensor networks are used to acquire fluid parameters and geological environment parameters in real time; The data acquisition module, connected to the sensor network, is used to collect sensor data to obtain fluid parameters and geological environment parameters during the fracturing fluid delivery process, and to perform preprocessing. The prediction module, connected to the data acquisition module, has a built-in prediction model trained based on machine learning algorithms, which is used to predict the frictional resistance of fracturing fluid in real time based on the extracted input features. The prediction model is obtained by training on measured frictional resistance data collected in historical fracturing operations and corresponding input features. The input features include derived features reflecting the non-Newtonian fluid properties of fracturing fluid and derived features reflecting the influence of geological parameters. The control module is connected to the prediction module and generates adjustment instructions based on the deviation between the prediction result and the preset target range. The execution module, connected to the control module, is used to perform operations to adjust the fracturing fluid formulation or pumping parameters according to the adjustment instructions.

[0028] Furthermore: the sensor network includes a pressure sensor, a temperature sensor, a flow sensor, a differential pressure friction resistance measuring device, and a geological parameter acquisition device; the differential pressure friction resistance measuring device includes an upstream pressure sensor and a downstream pressure sensor installed at a predetermined distance on the fracturing fluid delivery pipeline; the geological parameter acquisition device includes a downhole pressure gauge and a logging data interface.

[0029] Furthermore, the control module also includes a feedback learning unit, which feeds back the adjusted actual operation results to the prediction module for continuous optimization of the prediction model.

[0030] Compared with the prior art, the present invention has the following advantages: I. This invention deploys a sensor network at the fracturing site to collect fluid and geological environmental parameters in real time. By combining feature engineering and machine learning algorithms to construct a predictive model, it achieves dynamic prediction and automatic control of fracturing fluid friction resistance. Compared to traditional offline testing methods, this invention solves the problems of poor timeliness and low accuracy, enabling continuous monitoring of fracturing fluid performance changes during construction, timely detection of anomalies, and automatic parameter adjustment.

[0031] Second, this invention introduces multi-dimensional features such as Reynolds number, dynamic viscosity adjustment factor, and formation pressure gradient, and accurately describes the viscosity variation law of non-Newtonian fluids through the Carreau-Yasuda model. By incorporating geological parameters into the prediction model, it significantly improves the prediction accuracy of friction resistance in complex geological environments. Compared with existing technologies that only measure a single parameter, this invention can comprehensively reflect the evolution of physicochemical properties during fracturing fluid transportation, providing a scientific basis for optimizing formulations.

[0032] Third, this invention adopts a feedback learning mechanism to feed the adjusted actual operation effect back to the prediction model, thereby achieving continuous optimization of the model. Compared with the traditional method that relies on human experience to make judgments, this invention automatically identifies the pattern of parameter changes through machine learning algorithms, and can adapt to different working conditions without frequent manual intervention, thus greatly improving the level of intelligence in fracturing operations.

[0033] Fourth, this invention reduces additive waste, lowers equipment maintenance costs, and shortens operation time by precisely controlling the fracturing fluid formulation and pumping parameters, resulting in a significant reduction in overall operating costs and demonstrating good economic benefits and practical value. Attached Figure Description

[0034] Figure 1 A flowchart of a method for determining the frictional resistance of fracturing fluid in the field, provided by the present invention; Figure 2 This is a schematic diagram of a fracturing fluid friction resistance determination system provided by the present invention. Detailed Implementation

[0035] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Example 1 like Figure 1 As shown, the present invention provides a method for determining the frictional resistance of fracturing fluid in the field, comprising the following steps: S1: Deploy a sensor network at the fracturing site; S2: The sensor network is used to collect fluid parameters during the fracturing fluid delivery process in real time and to obtain geological environment parameters; S3: Based on the collected fluid parameters and geological environment parameters, feature engineering is performed to obtain input features; After feature engineering, the extracted input features are used to train the prediction model. For historical fracturing operation data, derived features such as dynamic viscosity adjustment factor, Reynolds number, formation pressure gradient, and time series features are calculated using the same method as in step S3. These features, along with the corresponding measured friction resistance values, constitute the training sample set.

[0037] S4: Based on the input features extracted in S3 and the measured friction resistance data collected in historical fracturing operations, a machine learning algorithm is used to train a prediction model and establish a nonlinear mapping relationship between the input features and the friction resistance; wherein, the input features include derived features reflecting the non-Newtonian fluid properties of the fracturing fluid and derived features reflecting the influence of geological parameters; S5: Input the real-time collected data into the trained prediction model to predict the fracturing fluid friction resistance under the current fracturing fluid delivery state in real time; S6: Adjust the fracturing fluid formulation or pumping parameters according to the predicted results of the fracturing fluid friction resistance so that the friction resistance reaches the preset target range; and feed back the adjusted actual friction resistance measured value to the prediction model to perform incremental training or retraining of the prediction model to continuously optimize the prediction model.

[0038] The feedback learning mechanism works as follows: After adjusting at least one of the fracturing fluid additive concentration, pumping rate, or pressure based on the output of the prediction model, the system continuously collects the adjusted actual operating data, including the adjusted measured value of frictional resistance and the corresponding operating parameters. The adjusted measured value of actual frictional resistance is compared with the predicted value of the prediction model to calculate the prediction deviation. When the prediction deviation exceeds a preset threshold, the system adds the newly collected data as incremental training samples to the training dataset to perform incremental training or periodic retraining of the prediction model, updating the model parameters so that the prediction model can adapt to new operating conditions and changes in fracturing fluid formulation, thereby achieving continuous optimization of the model.

[0039] The preset threshold is set according to the engineering accuracy requirements, preferably with a relative error of no more than 10%. When the prediction deviation of multiple consecutive data points exceeds the preset threshold, the model retraining process is triggered.

[0040] It should be noted that the training process of the prediction model specifically includes the following steps: (1) Data preparation: Collect a complete dataset of historical fracturing operations. Each data record includes input features and output labels. Input features include the original fluid parameters (viscosity, velocity, temperature, density) after feature engineering, geological environment parameters (formation pressure, porosity), and calculated derived features (Reynolds number, dynamic viscosity adjustment factor, formation pressure gradient, time series features, etc.). The output label is the frictional resistance value actually measured under the corresponding working conditions in this operation.

[0041] (2) Dataset partitioning: The historical dataset is randomly divided into a training set and a validation set according to a preset ratio. The preset ratio is preferably 80% for the training set and 20% for the validation set. The training set is used for learning and optimizing model parameters, and the validation set is used to evaluate the generalization performance of the model during training to prevent overfitting.

[0042] (3) Model selection: Select a machine learning algorithm capable of handling complex nonlinear relationships as the basic architecture of the prediction model. Optional algorithms include, but are not limited to, random forests, gradient boosting trees, neural networks, or long short-term memory networks (LSTM). Neural networks achieve nonlinear transformation of input features through multiple layers of nonlinear activation functions; random forests capture nonlinear interactions between features through the combination of multiple decision trees; and long short-term memory networks handle nonlinear dependencies in time series through gating mechanisms and recurrent structures.

[0043] (4) Model Training: Input the preprocessed training set feature data into the model, and randomly initialize the model parameters. The model calculates the predicted friction resistance value based on the current parameters and compares it with the actual friction resistance label. The prediction error is calculated using a loss function, preferably mean squared error (MSE). The error is backpropagated through an optimization algorithm to update the model parameters and reduce the loss function value. The optimization algorithm includes stochastic gradient descent (SGD) or the Adam optimizer. Repeat the above forward calculation and parameter update process until the model loss on the training set no longer decreases significantly.

[0044] (5) Model Validation and Hyperparameter Tuning: During training, the model performance is evaluated periodically using a validation set. Evaluation metrics include the coefficient of determination (R²) and root mean square error (RMSE). The model hyperparameters are adjusted based on the validation set performance, including but not limited to decision tree depth, learning rate, and number of neural network layers. The model that performs best on the validation set is selected as the final prediction model.

[0045] The physical basis of the nonlinear mapping relationship lies in the following: fracturing fluid, as a non-Newtonian fluid, exhibits a dynamic viscosity variation with shear rate, which can be reflected by the dynamic viscosity adjustment factor calculated using the Carreau-Yasuda model. Formation pressure gradients alter the flow path and resistance distribution of the fracturing fluid, while temperature changes simultaneously affect both fluid viscosity and formation porosity. In actual fracturing operations, these multiple factors are coupled—formation pressure affects the flow velocity of the fracturing fluid, which in turn affects the shear rate, ultimately altering the dynamic viscosity. Under different shear rate conditions, the fracturing fluid exhibits different rheological properties, and the frictional resistance corresponding to the same viscosity value varies under different formation pressure environments. This coupling effect of multiple factors results in highly nonlinear characteristics in frictional resistance prediction, necessitating the use of machine learning algorithms capable of capturing these interactions. By simultaneously inputting features such as the dynamic viscosity adjustment factor, Reynolds number, and formation pressure gradient into the prediction model, the machine learning algorithm can automatically learn the interactions between these features, establishing a frictional resistance prediction model that comprehensively considers the dynamic characteristics of the fluid and the influence of the geological environment.

[0046] Example 2 like Figure 2 As shown, the present invention also provides a fracturing fluid friction resistance field determination system for implementing the above method, comprising: Sensor networks are used to acquire fluid parameters and geological environment parameters in real time; The data acquisition module, connected to the sensor network, is used to collect sensor data to obtain fluid parameters and geological environment parameters during the fracturing fluid delivery process, and to perform preprocessing. The prediction module, connected to the data acquisition module, has a built-in prediction model trained based on machine learning algorithms, which is used to predict the frictional resistance of fracturing fluid in real time based on the extracted input features. The prediction model is obtained by training on measured frictional resistance data collected in historical fracturing operations and corresponding input features. The input features include derived features reflecting the non-Newtonian fluid properties of fracturing fluid and derived features reflecting the influence of geological parameters. The control module is connected to the prediction module and generates adjustment instructions based on the deviation between the prediction result and the preset target range. The execution module, connected to the control module, is used to perform operations to adjust the fracturing fluid formulation or pumping parameters according to the adjustment instructions.

[0047] According to one specific embodiment of this example, the sensor network includes a pressure sensor, a temperature sensor, a flow sensor, a differential pressure friction resistance measuring device, and a geological parameter acquisition device.

[0048] The differential pressure friction resistance measuring device includes an upstream pressure sensor and a downstream pressure sensor spaced at a predetermined distance on the fracturing fluid delivery pipeline. By measuring the pressure difference between the two ends of the same pipe segment, and combining this with the pipe segment length and inner diameter, the frictional pressure drop per unit pipe length is calculated, thus characterizing the frictional resistance of the fracturing fluid in the pipeline in real time. Specifically, the upstream and downstream pressure sensors are installed at two cross-sectional positions at a known distance on the fracturing fluid delivery pipeline. The predetermined distance is determined based on the nominal diameter of the pipeline, preferably 50-200 times the inner diameter of the pipe, to ensure sufficient flow field development and accurate measurement results. The upstream and downstream pressure sensors synchronously collect the pressure values ​​P at both ends of the pipe segment. up and P down The pressure difference ΔP = P up - P down This is the frictional pressure drop of that pipe section.

[0049] The geological parameter acquisition device includes a downhole pressure gauge and a logging data interface. The downhole pressure gauge is installed at a preset depth inside the wellbore to monitor bottomhole flowing pressure and formation pressure in real time. The logging data interface receives static geological parameters such as formation porosity and permeability obtained from previous well logging interpretations, using them as prior input data for the prediction model. It should be noted that formation porosity and permeability are static geological parameters obtained through previous well logging interpretations and are input into the prediction model as known prior parameters during fracturing operations; they do not need to be collected in real time during the operation. Formation pressure, however, is monitored in real time using the downhole pressure gauge.

[0050] According to a specific implementation of this embodiment, the control module further includes a feedback learning unit, which feeds back the adjusted actual operation effect to the prediction module; when the prediction deviation exceeds a preset threshold, the newly collected data is used as incremental training samples to perform incremental training or retraining on the prediction model in order to continuously optimize the prediction model.

[0051] Example 3 The following is a specific embodiment for further explanation, and the specific implementation steps are as follows: 1. Deployment of smart sensor networks; Deploy advanced sensors: Deploy high-precision pressure sensors, temperature sensors, flow meters, and new friction sensors in key locations. These sensors should have wireless transmission capabilities to send data to the central processing unit in real time.

[0052] Geological feature monitoring: Geological sensors are deployed simultaneously to monitor parameters such as formation pressure and porosity, providing geological environment information for the model.

[0053] 2. Real-time data acquisition and transmission; Build a data platform: Establish a cloud data platform to collect real-time data from various sensors, including but not limited to fracturing fluid characteristics, pumping pressure, flow rate changes, temperature, and geological parameters.

[0054] Data preprocessing: Cleaning and normalizing the raw data to prepare for subsequent analysis.

[0055] 3. Machine learning model development; Feature engineering: Based on historical data and physical principles, key parameters are selected as input features, such as fluid viscosity, flow velocity, temperature, and formation characteristics.

[0056] Model training: Deep learning or reinforcement learning algorithms, such as neural networks, random forests or long short-term memory networks (LSTM), are used to train the model to predict frictional resistance under different working conditions.

[0057] Real-time prediction: The model is connected to a real-time data stream to enable real-time prediction of frictional resistance under the current fracturing fluid delivery conditions.

[0058] 4. Adaptive optimization strategy; Dynamic adjustment strategy: Based on model prediction results, automatically adjust fracturing fluid formulation (such as additive concentration), pumping rate, pressure, etc., to minimize frictional resistance.

[0059] Feedback learning mechanism: The adjusted actual operation effect is fed back to the model, and the prediction model and adjustment strategy are continuously iterated and optimized to form a closed-loop control.

[0060] 5. Safety and efficiency monitoring; Risk warning system: An outlier detection module is set up to issue an early warning signal as soon as a possible equipment failure or operational risk is predicted.

[0061] Efficiency assessment: Regularly assess fracturing operation efficiency, including cost savings and production increases, and continuously verify the effectiveness of optimization strategies.

[0062] Furthermore, regarding feature engineering: for the step of selecting key parameters as input features based on historical data and physical principles, such as fluid viscosity, flow velocity, temperature, and formation characteristics, we also provide specific improvements: Feature engineering extracts the most valuable information from raw data and transforms it into input features that machine learning algorithms can understand. For the scenario of predicting fracturing fluid friction resistance, the following is a detailed implementation plan for feature engineering: 1. Data collection and organization; First, collect detailed data records of historical fracturing operations, including but not limited to: Fluid properties: type, viscosity, and density of fracturing fluid.

[0063] Pumping parameters: pumping rate (flow rate), pressure, and pumping time.

[0064] Environmental conditions: ground temperature, downhole temperature, and ambient temperature during operation.

[0065] Formation characteristics: rock porosity, permeability, formation pressure.

[0066] Operational results: Actual measured frictional resistance values, production increase effect, etc.

[0067] 2. Data preprocessing; Missing value handling: Check if there are missing values ​​in the dataset, and handle them by deleting or imputing (mean, median or specific model prediction) as appropriate.

[0068] Outlier detection and handling: Use statistical methods (such as box plots and Z-scores) to identify and handle outliers to avoid them negatively impacting model training.

[0069] Data standardization / normalization: Since the dimensions and scales of different features may vary greatly, numerical features are standardized (e.g., z-score standardization) or normalized (e.g., min-max scaling) so that all features are on the same scale, which facilitates model learning.

[0070] 3. Feature selection; 3.1 Guided by physical principles; Based on fluid mechanics principles, it is clear which parameters directly affect frictional resistance: Fluid viscosity: directly affects the internal friction force during fluid flow.

[0071] Flow velocity: The higher the velocity, the greater the frictional resistance usually is.

[0072] Temperature: affects fluid viscosity, which in turn affects frictional resistance.

[0073] Formation characteristics, such as porosity and permeability, indirectly affect fluid flow paths and resistance.

[0074] 3.2 Correlation analysis; Use correlation coefficient matrices (such as Pearson coefficients) to analyze the correlation between features, eliminate highly correlated features, and reduce multicollinearity problems.

[0075] For non-linear relationships, consider using methods such as scatter plots and interactive information to further explore the relationships between features.

[0076] 3.3 Feature Construction; Derivative features: Based on physical knowledge, new features can be created. For example, the "Reynolds number" can be constructed as an indicator of the fluid flow state.

[0077] Periodicity characteristics: If the working time affects the frictional resistance (such as temperature changes caused by the diurnal temperature difference), periodicity characteristics can be extracted from the time series.

[0078] 4. Feature importance assessment; Feature selection techniques (such as recursive feature elimination and tree-based feature importance) are used to evaluate the contribution of each feature to the model's prediction performance, and the most effective feature set is further selected.

[0079] 5. Feature combination and encoding; For categorical features (such as fracturing fluid type), use unique thermal coding or tag coding.

[0080] Considering the interactions between features, non-linear relationships can be captured by creating feature interaction terms.

[0081] After completing the above steps, the resulting feature set can be used as input to the machine learning model for subsequent model training and prediction. Feature engineering is an iterative process that may require repeated adjustments and optimizations based on the model's performance.

[0082] The specific formula is as follows: 1. Formulas for calculating physical parameters; In feature engineering, some features need to be calculated using physical formulas. These calculations help to better reflect the essential characteristics of fluid flow.

[0083] Reynolds number: A dimensionless number that measures the state of fluid flow, reflecting the ratio of inertial forces to viscous forces, and is crucial for determining the transition between laminar and turbulent flow. The formula for calculating the Reynolds number is:

[0084] in, It is fluid density. It's the flow rate. It is the pipe diameter. It refers to fluid viscosity. The Reynolds number is an important parameter in predicting frictional drag because it is directly related to the choice of flow drag model (such as laminar or turbulent flow model).

[0085] Nusselt number: A dimensionless number describing convective heat transfer efficiency. Although directly related to the heat transfer process, it can also reflect the influence of fluid flow characteristics on heat transfer in some cases, indirectly affecting fracturing fluid temperature control. The formula for calculating the Nusselt number is:

[0086] here, It is the convective heat transfer coefficient. It refers to thermal conductivity. In certain situations, when considering the indirect effect of temperature on frictional resistance, the Nusselt number may become an auxiliary characteristic.

[0087] 2. Reasons for feature selection design; Fluid viscosity ( ): Directly affects frictional resistance. The higher the viscosity, the greater the internal friction during flow. Therefore, it is the core input of the prediction model.

[0088] Flow rate ( High-speed flow leads to greater kinetic energy, which in turn increases friction with the pipe wall and affects frictional resistance.

[0089] Temperature: Temperature affects fluid viscosity. Generally speaking, as temperature increases, viscosity decreases, which affects the magnitude of frictional resistance.

[0090] Formation characteristics (such as porosity and permeability): Although these parameters are not directly measured fluid properties, they have a significant impact on the flow path and resistance of fracturing fluid in the formation, indirectly determining the fracturing effect and the required pressure level.

[0091] 3. Example of feature construction: Time series features; Assuming that operating time and weather conditions (such as temperature) have a periodic impact on frictional resistance, time feature extraction is used to convert the time data into periodic features, for example, using a sine transform. Sum and cosine transform This is used to represent hourly variations within a day, capturing the impact of temperature fluctuations caused by diurnal temperature differences on frictional resistance.

[0092] in, It is the current time (hour). The interval is 24 hours, which maps time information to a periodic space, helping the model capture periodic patterns in the time series.

[0093] 4. Feature importance assessment methods; Feature importance based on random forests: The random forest algorithm can naturally provide feature importance scores, which are measured by calculating the average reduction in impurity of a feature in the decision tree node split.

[0094]

[0095] in, Here, t represents the impurity of a node, t is the branch number of the tree, and r is the total number of branches in the tree. It is a feature The average impurity of nodes after splitting. Importance(F) represents the importance score of feature F, that is, the degree to which the feature contributes to the prediction results in the random forest model. A higher score means that the feature is more important in the prediction.

[0096] By using the above method, not only were directly relevant features selected based on physical principles, but the input feature set was also optimized by constructing new features and evaluating feature importance, enabling the model to more accurately understand and predict the frictional resistance of fracturing fluid in actual operations.

[0097] Furthermore, based on the above, we can further refine and supplement some features and corresponding calculation formulas to more comprehensively cover various factors affecting the frictional resistance of fracturing fluid, while enhancing the predictive ability of the model.

[0098] New features and formulas: 1. Dynamic viscosity adjustment factor; Considering the dynamic effects of additives in fracturing fluids, the apparent viscosity of the fluid changes with the shear rate during pumping. A dynamic viscosity adjustment factor based on the Carreau-Yasuda model is introduced to describe this non-Newtonian fluid characteristic.

[0099]

[0100] in, It is the apparent viscosity. and These are the viscosities at zero shear rate and infinite shear rate, respectively. It's a relaxation time. For shear rate, and These are parameters of the Carreau-Yasuda model. α is the transition parameter of the Carreau-Yasuda model, used to control the smoothness of the viscosity change curve when transitioning from a low shear rate region to a high shear rate region. n is the power-law exponent, describing the flow behavior of the fluid at high shear rates. When n < 1, it indicates shear thinning characteristics; when n > 1, it indicates shear thickening characteristics; and when n = 1, it indicates a Newtonian fluid. This dynamic adjustment factor can more accurately reflect the true viscosity of the fracturing fluid under different flow conditions.

[0101] 2. Formation pressure gradient; Variations in formation pressure gradients can also affect the resistance to fracturing fluid flow, especially in complex geological structures. Pressure gradients can be calculated by measuring or estimating the pressure at different formation depths.

[0102]

[0103] here, and These represent the formation pressures at two different depths. This is the vertical distance between these two points. This feature helps the model understand the indirect effect of formation pressure on frictional resistance.

[0104] It should be noted that the dynamic viscosity adjustment factor is designed because many fracturing fluids contain non-Newtonian additives, and their viscosity changes with the shear rate. Accurately reflecting this dynamic characteristic can effectively improve the model's simulation accuracy of actual flow conditions, especially during the pumping phase at high shear rates. The formation pressure gradient is introduced because it directly affects the flow path and required pressure of fracturing fluid in the formation, especially during the fracturing propagation stage. Different pressure gradients alter the fluid distribution within the fracture network, thus influencing frictional resistance. This feature helps the model better adapt to complex geological environments and improves the geographic relevance of predictions.

[0105] Through these improvements, we have further refined the feature set to better adapt to the complex physical phenomena in the fracturing fluid flow process, thereby improving the accuracy and applicability of the model predictions.

[0106] The following example, using fracturing fluid prepared on-site in a well in the Jinlong 2 well area, illustrates the specific testing process: The main steps of the testing process include: equipment preparation, testing procedures, and data recording and analysis.

[0107] 1. Equipment preparation; Test setup: Build a circulation system that can simulate the actual fracturing fluid flow conditions, including a high-pressure pump, flow meter, pressure sensor, temperature controller, and a representative delivery pipe (such as the same material and diameter as that used in actual operations).

[0108] Fracturing fluid samples: Four types of fluids were prepared on-site: stock solution, gel, low viscosity, and high viscosity. These four fluids play different roles in the fracturing of this well, in order to evaluate the impact of additives in different fluid formulations on frictional resistance.

[0109] Data recording equipment: Equipped with a high-precision data acquisition system for real-time recording of key parameters such as fluid flow rate, pressure difference, and temperature.

[0110] 2. Testing process; 2.1 System preheating and calibration; Start the circulation system and perform a pre-run with clean water to ensure all equipment is working properly, while also calibrating the system, including verifying the accuracy of the flow meter and pressure sensor.

[0111] 2.2 Basic data recording; Without adding fracturing fluid, the pressure drop and flow rate of clean water as it passes through the delivery pipe are recorded as baseline data.

[0112] 2.3 Fracturing fluid testing; Single-sample test: Select a fracturing fluid sample, inject it into the circulation system, maintain the set flow rate, and record the pressure drop, flow rate, and temperature under these conditions.

[0113] Multivariate testing: Change the flow rate or temperature and repeat the above test to evaluate the effect of these factors on frictional resistance.

[0114] Comparison of different samples: Change the fracturing fluid sample and repeat step 2.3 to collect data under different formulations.

[0115] 3. Data recording and analysis; Calculate the friction resistance coefficient: Calculate the friction resistance coefficient of the fracturing fluid under various conditions according to the Darcy-Weisbach equation.

[0116] Comparative analysis: Compare the differences in frictional resistance between different fracturing fluid formulations, flow rates, and temperatures to identify the optimal configuration.

[0117] Optimization recommendations: Based on the data analysis results, we propose suggestions for adjusting the fracturing fluid formulation and optimizing the pumping parameters to reduce frictional resistance and improve fracturing efficiency.

[0118] The following uses historical operation data from a well in the Jinlong 2 well area as an example to illustrate the feature calculation and prediction verification process of the prediction model.

[0119] 1. Initial parameter settings; Fluid density ρ = 1000 kg / m³, flow velocity v = 3 m / s, pipe diameter D = 0.1 m, fluid viscosity μ = 0.05 Pa·s, shear rate = 500 s - ¹, Current time t = 10 hours (period T = 24 hours), formation pressure P(z1) = 30 MPa at depth z1 = 1000 m, formation pressure P(z2) = 40 MPa at depth z2 = 1500 m, z2 - z1 = 500 m.

[0120] Carreau-Yasuda model parameters (determined through rheological experiments for the fracturing fluid formulation used in this well): zero-shear viscosity η0 = 1.0 Pa·s, infinite-shear viscosity η ∞ = 0.001 Pa·s, relaxation time λ = 0.1 s, transition parameter α = 2, power law exponent n = 0.5.

[0121] The measured frictional resistance f under this working condition actual = 1.2 MPa.

[0122] 2. Calculation of derived features; Reynolds number calculation: Re = ρ·v·D / μ = 1000 × 3 × 0.1 / 0.05 = 6000 Calculation of dynamic viscosity adjustment factor (based on Carreau-Yasuda model): Calculate the intermediate term: (λ· ) α = (0.1 × 500)² = 50² = 2500 Calculate the exponent: (n - 1) / α = (0.5 - 1) / 2 = -0.25 Calculate the bracketed terms: [1 + (λ· ) α ] ((n-1) / α) = [1 + 2500] (-0.25) = 2501 (-0.25) ≈ 0.1414 Therefore: η app = η ∞ + (η0- η ∞ ) × 0.1414 = 0.001 + (1.0 - 0.001) × 0.1414 ≈0.001 + 0.999 × 0.1414 ≈ 0.1423 Pa·s Take the calculated result η of the dynamic viscosity adjustment factor. app = 0.1423 Pa·s is used as the input feature value. It should be noted that this value is the result of the dynamic viscosity adjustment factor. During model training and prediction, it is normalized and preprocessed in the same way as other features. The normalization benchmark is determined by the historical job dataset.

[0123] Formation pressure gradient calculation: ΔP / Δz = (P(z2) - P(z1)) / (z2- z1) = (40 - 30) MPa / 500 m = 0.02MPa / m = 20 kPa / m; Time series feature calculation: sin(2π × 10 / 24) = sin(5π / 6) = sin(π - π / 6) = sin(π / 6) = 0.5 (exact value); cos(2π × 10 / 24) = cos(5π / 6) = -cos(π / 6) = -√3 / 2 ≈ -0.866; 3. Model prediction and evaluation; The calculated eigenvectors [Re = 6000, η] app = 0.1423 Pa·s, ΔP / Δz = 20 kPa / m, sin = 0.5, cos = -0.866, and other original features] Input the pre-trained prediction model.

[0124] The model performs forward computation (for tree models, this involves traversing multiple decision trees; for neural networks, this involves matrix operations and nonlinear activation) and outputs the predicted frictional resistance value f corresponding to that set of features. pred = 1.15 MPa.

[0125] Verification and evaluation: Absolute error: |f actual -f pred | = |1.2 - 1.15| = 0.05 MPa; Relative error: |1.2 - 1.15| / 1.2 × 100% ≈ 4.17%; This error is within an acceptable range for engineering purposes (less than the preset threshold of 10%), validating the model's predictive effectiveness for this data point. Statistical results of such errors (e.g., mean absolute error (MAE), root mean square error (RMSE)) across the entire validation set will be used to comprehensively evaluate the model's overall predictive accuracy and generalization ability.

[0126] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for determining the frictional resistance of fracturing fluid in the field, characterized in that: Includes the following steps; S1: Deploy a sensor network at the fracturing site; S2: The sensor network is used to collect fluid parameters during the fracturing fluid delivery process in real time and to obtain geological environment parameters; S3: Based on the collected fluid parameters and geological environment parameters, feature engineering is performed to obtain input features; S4: Based on the input features extracted in S3 and the measured friction resistance data collected in historical fracturing operations, a machine learning algorithm is used to train a prediction model and establish a nonlinear mapping relationship between the input features and the friction resistance; wherein, the input features include derived features reflecting the non-Newtonian fluid properties of the fracturing fluid and derived features reflecting the influence of geological parameters; S5: Input the real-time collected data into the trained prediction model to predict the fracturing fluid friction resistance under the current fracturing fluid delivery state in real time; S6: Adjust the fracturing fluid formulation or pumping parameters according to the predicted results of the fracturing fluid friction resistance so that the friction resistance reaches the preset target range. The feature engineering process in S3 includes: S31: Clean and normalize the original data to obtain the processed data; S32: Calculate derived features based on fluid dynamics principles for the processed data; S33: Based on the derived features, after removing features with a correlation greater than a preset threshold through correlation analysis, the feature set is obtained after encoding, and the features in the feature set are used as input features. The adjustment in S6 includes: adjusting at least one of the fracturing fluid additive concentration, pumping rate, or pressure according to the deviation between the predicted frictional resistance value and the preset target range through a feedback learning mechanism; and feeding back the adjusted actual measured value of frictional resistance to the prediction model, and performing incremental training or retraining on the prediction model to continuously optimize the prediction model.

2. The method for determining the frictional resistance of fracturing fluid in the field according to claim 1, characterized in that: In step S1, the sensor network includes a pressure sensor, a temperature sensor, a flow sensor, a differential pressure friction resistance measuring device, and a geological parameter acquisition device. The differential pressure friction resistance measuring device includes an upstream pressure sensor and a downstream pressure sensor spaced at a preset distance on the fracturing fluid delivery pipeline. It obtains the frictional pressure drop of the pipeline segment by measuring the pressure difference between the two ends of the same pipeline segment. The geological parameter acquisition device includes a downhole pressure gauge for real-time monitoring of formation pressure and a logging data interface for receiving static geological parameters obtained from previous logging interpretation.

3. The method for determining the frictional resistance of fracturing fluid in the field according to claim 1, characterized in that: In S2, the fluid parameters include fluid viscosity, flow rate, temperature, and density, and the geological environment parameters include formation pressure and porosity. The fluid viscosity, flow rate, temperature, and density are collected in real time by online sensors installed on the fracturing fluid delivery pipeline, the formation pressure is obtained in real time by downhole pressure gauges, and the porosity is a static geological parameter obtained through previous well logging interpretation and is provided as prior input data to the prediction model.

4. The method for determining the frictional resistance of fracturing fluid in the field according to claim 1, characterized in that: The derived feature includes the Reynolds number, which is calculated using the following formula: in, It is the fluid density. It's the flow rate. It is the pipe diameter. It is fluid viscosity. It is the Reynolds number.

5. The method for determining the frictional resistance of fracturing fluid in the field according to claim 1, characterized in that: The derived feature also includes a dynamic viscosity adjustment factor, which is calculated based on the Carreau-Yasuda model, and the calculation formula is as follows: in, It is the apparent viscosity. and These are the viscosities at zero shear and infinite shear rates, respectively. It's a relaxation time. Shear rate, and These are the parameters of the Carreau-Yasuda model. α is the transition parameter of the Carreau-Yasuda model, used to control the smoothness of the viscosity change curve when transitioning from a low shear rate region to a high shear rate region; n is the power-law exponent, describing the flow behavior of the fluid at high shear rates; the parameters η0 and ηn of the Carreau-Yasuda model are... ∞ λ, α, and n were determined by rheological experiments on fracturing fluid samples.

6. The method for determining the frictional resistance of fracturing fluid in the field according to claim 1, characterized in that: The derived feature also includes the formation pressure gradient, which is calculated using the following formula: in, For formation pressure gradient, and These represent the formation pressures at two different depths. It is the perpendicular distance between these two points.

7. The method for determining the frictional resistance of fracturing fluid in the field according to claim 1, characterized in that: The feature engineering process in S3 also includes extracting time series features and using sine transform sin(2πt / T) and cosine transform cos(2πt / T) to capture periodic influencing factors, where t is the current time and T is the period.

8. The method for determining the frictional resistance of fracturing fluid in the field according to claim 1, characterized in that: Between steps S3 and S4, a feature importance evaluation step is included, employing a feature importance scoring method based on random forests. The calculation formula is as follows: Where I is the impurity of a node, t is the branch number of the tree, r is the total number of branches in the tree, F is the feature, E[I(t)|F] is the average impurity of nodes after feature F is split, and Importance(F) represents the importance score of feature F; features with importance scores below the preset importance threshold are removed, and features with high importance scores are retained for model training.

9. The method for determining the frictional resistance of fracturing fluid in the field according to claim 1, characterized in that: The machine learning algorithm in S4 includes at least one of random forest, gradient boosting tree, neural network, or long short-term memory network; the training process of the prediction model includes: dividing the historical dataset into training set and validation set according to a preset ratio, using input features as model input and measured friction resistance value as model output label, calculating prediction error through loss function and updating model parameters using optimization algorithm until the model's performance indicators on the validation set reach the preset requirements; the performance indicators include coefficient of determination R² and root mean square error RMSE.

10. A fracturing fluid friction resistance field determination system for implementing the method of any one of claims 1 to 9, characterized in that, include: Sensor networks are used to acquire fluid parameters and geological environment parameters in real time; The data acquisition module, connected to the sensor network, is used to collect sensor data to obtain fluid parameters and geological environment parameters during the fracturing fluid delivery process, and to perform preprocessing. The prediction module, connected to the data acquisition module, has a built-in prediction model trained based on machine learning algorithms, which is used to predict the frictional resistance of fracturing fluid in real time based on the extracted input features. The prediction model is obtained by training on measured frictional resistance data collected in historical fracturing operations and corresponding input features. The input features include derived features reflecting the non-Newtonian fluid properties of fracturing fluid and derived features reflecting the influence of geological parameters. The control module is connected to the prediction module and generates adjustment instructions based on the deviation between the prediction result and the preset target range. The execution module, connected to the control module, is used to perform operations to adjust the fracturing fluid formulation or pumping parameters according to the adjustment instructions.

11. The fracturing fluid friction resistance field determination system according to claim 10, characterized in that: The sensor network includes pressure sensors, temperature sensors, flow sensors, differential pressure friction resistance measuring devices, and geological parameter acquisition devices; the differential pressure friction resistance measuring devices include upstream pressure sensors and downstream pressure sensors installed at preset distances on the fracturing fluid delivery pipeline; the geological parameter acquisition devices include downhole pressure gauges and logging data interfaces.

12. The fracturing fluid friction resistance field determination system according to claim 10, characterized in that: The control module also includes a feedback learning unit, which feeds back the adjusted actual operation effect to the prediction module. When the prediction deviation exceeds a preset threshold, the newly collected data is used as an incremental training sample to perform incremental training or retraining on the prediction model to continuously optimize the prediction model.

Images