Fracturing fluid powder metering deviation adaptive compensation feeding regulation method

Abstract

The application discloses a fracturing fluid powder metering deviation self-adaptive compensation feeding regulation method, and belongs to the field of fracturing fluid supply regulation, which comprises the following steps: S1, constructing a standardized multi-dimensional disturbance characteristic matrix and a multi-category fracturing powder cross-powder reference property calibration library; S2, constructing a sub-item metering deviation model, outputting an instantaneous mechanism metering deviation estimated value and a disturbance cause classification label; S3, building a TFT time sequence fusion deep learning network, outputting a long-time sequence advanced flow deviation prediction sequence; S4, constructing a three-dimensional coupled self-adaptive control model, and rolling optimization is used to solve a three-dimensional collaborative compensation control vector; and S5, issuing a synchronous regulation instruction after hardware safety limiting. The fracturing fluid powder metering deviation self-adaptive compensation feeding regulation method can accurately quantify and compensate the metering deviation caused by multi-coupling disturbances under all working conditions, and can also consider high-precision metering and continuous unattended construction.

Description

Technical Field

[0001] This invention relates to the field of fracturing fluid supply control technology, and in particular to a method for adaptive compensation of fracturing fluid powder metering deviation in supply control. Background Technology

[0002] During hydraulic fracturing operations in oil and gas wells, the precise and continuous supply of various dry powder additives, such as guar gum, breaker, and anti-swelling agent, directly determines the concentration of fracturing fluid, its proppant-carrying capacity, and the reservoir stimulation effect. On-site powder supply systems are susceptible to multiple time-varying disturbances, including ambient temperature and humidity, vehicle transport vibration, powder agglomeration and bridging, powder type switching, and changes in silo material load, resulting in significant instantaneous and steady-state metering deviations. Traditional supply control relies on manual offline calibration and closed-loop adjustment by a single actuator, which cannot adapt to scenarios involving multiple coupled disturbances and alternating powders. This leads to poor metering accuracy, frequent shutdowns for manual calibration, and severely restricts the efficiency and quality of continuous fracturing operations. Summary of the Invention

[0003] The purpose of this invention is to provide an adaptive compensation method for fracturing fluid powder metering deviation, thereby solving the above-mentioned technical problems.

[0004] To achieve the above objectives, the present invention provides an adaptive compensation method for fracturing fluid powder metering deviation control, comprising the following steps: S1. Simultaneously collect multi-source field sensor raw signals including total powder mass, powder density, temperature and humidity, triaxial vibration, screw speed, and discharge gate displacement. After Kalman filtering for noise reduction, multi-disturbance feature derivation and normalization processing, a standardized multi-dimensional disturbance feature matrix is ​​obtained. Combined with laboratory offline physical property test data, a cross-powder benchmark physical property calibration library for multiple types of fracturing powder is established. S2. Based on the standardized multidimensional disturbance feature matrix and cross-powder reference property calibration library output by S1, a sub-item measurement deviation model is constructed using Biot fluid-structure interaction theory and BK mixing mode powder damage equation. A third-order extended state observer is used to uniformly identify all time-varying composite disturbances in real time, and output instantaneous mechanism measurement deviation estimates and disturbance cause classification labels. S3. Based on the standardized multidimensional perturbation feature matrix output by S1 and the cross-powder reference property calibration library, as well as the instantaneous mechanism measurement deviation estimate and perturbation cause classification label output by S2, a TFT time-series fusion deep learning network with built-in variable selection network and causal mask multi-head attention is built to separate the static inherent property of powder and time-varying dynamic perturbation features. The instantaneous mechanism measurement deviation estimate of S2 is embedded to construct a physical constraint hybrid loss function, and the long-time series leading flow deviation prediction sequence is output. S4. Based on the instantaneous timing measurement deviation estimate of S2 output, the long-term time-series advanced flow deviation prediction sequence of S3 output, and the cross-powder reference physical property calibration library of S1 output, a three-dimensional coupled adaptive control model of screw speed, silo vibration frequency, and discharge gate opening is constructed. According to the disturbance cause classification label of S2, the adjustment weights of multiple actuators are adaptively allocated, and the three-dimensional collaborative compensation control vector is solved by rolling optimization. S5. Based on the three-dimensional collaborative compensation control vector output by S4, the actual drive setting values ​​of screw speed, hopper vibration frequency and discharge gate opening are obtained by superimposing the reference steady-state control quantities of each powder. After hardware safety limiting, the screw, vibrator and discharge gate servo mechanism are synchronously controlled. After the control is completed, a new round of sensor data is collected and input into S1 to realize closed-loop acquisition iteration.

[0005] Therefore, the present invention employs the above-mentioned adaptive compensation feeding control method for fracturing fluid powder metering deviation, which has the following beneficial effects: 1. Collect data from multi-source sensors across the entire domain and construct a standardized disturbance feature matrix. Build a cross-powder property calibration library to fully cover the static properties of powders and all time-varying dynamic disturbances on site, achieving quantitative characterization of disturbances under all working conditions. 2. By combining Biot fluid-structure interaction and BK powder damage equation, a component measurement deviation model is established. Combined with a third-order extended state observer, composite time-varying disturbances are uniformly identified, and instantaneous mechanism deviations and disturbance classification labels are accurately output, so as to realize the traceability of disturbance root causes and the component quantification of deviations. 3. Construct a TFT time-series fusion network with built-in variable selection network and causal mask multi-head attention, separate static physical properties from time-varying perturbations, embed the mechanism model to construct a physical constraint hybrid loss function, effectively avoid the poor generalization problem of pure data-driven models, and output a long-series flow deviation prediction sequence to provide feedforward compensation; 4. Establish a three-dimensional coupled model-free adaptive controller for screw speed, vibration frequency, and discharge gate opening. Adaptively allocate adjustment weights for each actuator based on disturbance labels, perform rolling optimization and collaborative compensation of control quantities, and achieve multi-mechanism linkage for differentiated disturbance elimination, resulting in smooth adjustment and fast convergence speed. 5. Based on the powder density, the powder switching conditions are identified. The online adaptive update of the physical property calibration library is adopted by transfer learning. After the update, the parameters are synchronously distributed to each module. There is no need to stop the machine for manual recalibration, which ensures uninterrupted continuous fracturing operations. In summary, this invention solves the problems of time-varying steady-state and instantaneous composite deviations in metering caused by single-screw control with only one dimension of correction, powder moisture absorption and agglomeration, silo bridging, vehicle vibration, and material switching. It achieves dynamic adaptive compensation under all working conditions without manual shutdown for calibration, and the relative metering error is stably controlled within ±0.3%.

[0006] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0007] Figure 1 This is a flowchart of an adaptive compensation feeding control method for fracturing fluid powder metering deviation according to the present invention. Detailed Implementation

[0008] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0009] It should be noted that the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0010] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0011] like Figure 1 As shown, a method for adaptive compensation of fracturing fluid powder metering deviation and feed control includes the following steps: S1. Simultaneously collect multi-source field sensor raw signals including total powder mass, powder density, temperature and humidity, triaxial vibration, screw speed, and discharge gate displacement. After Kalman filtering for noise reduction, multi-disturbance feature derivation and normalization processing, a standardized multi-dimensional disturbance feature matrix is ​​obtained. Combined with laboratory offline physical property test data, a cross-powder benchmark physical property calibration library for multiple types of fracturing powders is established.

[0012] S2. Based on the standardized multidimensional disturbance feature matrix and cross-powder reference property calibration library output by S1, a sub-item measurement deviation model is constructed using Biot fluid-structure interaction theory and BK mixing mode powder damage equation. A third-order extended state observer is used to uniformly identify all time-varying composite disturbances in real time, and output instantaneous mechanism measurement deviation estimates and disturbance cause classification labels.

[0013] S3. Based on the standardized multidimensional perturbation feature matrix output by S1 and the cross-powder reference property calibration library, as well as the instantaneous mechanism measurement deviation estimate and perturbation cause classification label output by S2, a TFT time-series fusion deep learning network with built-in variable selection network and causal mask multi-head attention is built to separate the static inherent properties of powder and time-varying dynamic perturbation features. The instantaneous mechanism measurement deviation estimate of S2 is embedded to construct a physical constraint hybrid loss function, and the long-time-series leading flow deviation prediction sequence is output.

[0014] S4. Based on the instantaneous timing measurement deviation estimate of S2 output, the long-term leading flow deviation prediction sequence of S3 output, and the cross-powder reference physical property calibration library of S1 output, a three-dimensional coupled adaptive control model of screw speed, silo vibration frequency, and discharge gate opening is constructed. According to the disturbance cause classification label of S2, the adjustment weights of multiple actuators are adaptively allocated, and the three-dimensional collaborative compensation control vector is solved by rolling optimization.

[0015] S5. Based on the three-dimensional collaborative compensation control vector output by S4, the actual drive setting values ​​of screw speed, hopper vibration frequency and discharge gate opening are obtained by superimposing the reference steady-state control quantities of each powder. After hardware safety limiting, the screw, vibrator and discharge gate servo mechanism are synchronously controlled. After the control is completed, a new round of sensor data is collected and input into S1 to realize closed-loop acquisition iteration.

[0016] Step S1 specifically includes the following steps: S11. Read the current system-set target powder supply flow rate. The globally unified sampling period is dynamically determined based on the adaptive sampling interval formula of the flow rate. : ; In the formula: The maximum sampling interval of the system is 2.0s, which is the baseline value for low displacement under no-load conditions. This is the conversion factor between flow rate and sampling interval.

[0017] When calculated When the sampling period is less than 0.5s, the minimum sampling period is forcibly locked at 0.5s to ensure the accuracy of disturbance capture under high displacement.

[0018] S12, according to the calculation Periodically and synchronously acquire raw signals from multiple sources of field sensors. These raw signals include the total mass of fracturing fluid powder stored in the silo in real time. Real-time bulk density of fracturing fluid powder at the silo outlet under flow conditions Real-time temperature of the on-site operating environment of the fracturing fluid powder feeding equipment and humidity Triaxial real-time vibration acceleration of fracturing fluid powder feeding equipment Real-time operating speed of the powder conveying screw Real-time opening stroke of the discharge gate on the hopper discharge port .

[0019] S13. Perform Kalman filtering on the original signals from the multi-source field sensors to obtain the filtered set of basic physical quantities. .

[0020] in, These are the total mass of the filtered stored powder, the powder bulk density, the operating environment temperature, the operating environment humidity, the vibration acceleration of the x, y, and z axes, the screw speed, and the opening of the discharge gate.

[0021] S14. Derivation of multidimensional perturbation characteristics based on the set of fundamental physical quantities: Vibration amplitude disturbance characteristics derived .

[0022] Humidity-related powder density perturbation deviation characteristics derived .

[0023] Temperature and humidity reference offset perturbation characteristics derived .

[0024] Characteristics of residual mass load disturbance deviation in silos .

[0025] Screw speed operating condition disturbance deviation characteristics .

[0026] Discharge gate opening throttling disturbance deviation characteristics .

[0027] In the formula: The overall vibration disturbance amplitude of the equipment; This represents the powder density perturbation offset. For the first Dry basis bulk density of fracturing powder; This represents the offset due to ambient temperature disturbance. This represents the offset due to environmental relative humidity disturbance. To calibrate the temperature and humidity reference constants for laboratory standards; This refers to the deviation of the powder load in the silo. The rated steady-state filling reference mass of the silo; This refers to the screw speed disturbance deviation; The screw speed is calibrated to the current steady-state reference of the powder. This refers to the disturbance deviation of the discharge gate opening. The discharge gate opening is the steady-state reference value for the system.

[0028] S15. Concatenate the complete original perturbation vector at a single moment, and integrate all basic physical quantities and derived perturbation features to obtain the perturbation features. : ; In the formula, This is a transpose operation.

[0029] S16, Disturbance characteristics Normalization mapping to The interval is used to obtain the normalized perturbation eigenvector at a single time step. .

[0030] S17. Temporal sliding window splicing generates a standardized multidimensional perturbation feature matrix.

[0031] S171, Set the length of the timing sliding window There are [number] sampling points, and the total window duration is [time]. It covers the powder flow time variation range of about 30 seconds.

[0032] S172. According to the sampling time from farthest to most recent, the continuous The output of S16 at each moment Vertical sequential concatenation generates a standardized multidimensional perturbation feature matrix. : .

[0033] S18. Offline physical property testing combined with online iterative updates to construct a cross-powder benchmark physical property calibration library for multiple types of fracturing powders.

[0034] Step S18 specifically includes the following steps: S181, in Under standard offline laboratory conditions, steady-state flowability tests were conducted on three types of fracturing powders: guar gum thickener, breaker, and anti-swelling agent. A fixed set of reference parameters was determined for each type of powder, resulting in a reference parameter vector for each powder type. : ; In the formula: For the first Powder-based base bulk density, , Representative guar gum powder, Represents debonding agent, Represents anti-swelling powder; It exhibits Type I fracture toughness; It exhibits type II fracture toughness; For the first Equivalent elastic modulus of powder-like agents; Poisson's ratio; For the first The internal friction coefficient of powder-like materials; This is the linear calibration coefficient for screw flow rate-speed. This refers to the critical vibration amplitude for bridging the silo.

[0035] S182. Integrate the reference parameter vectors of the three types of powders to generate an initial global multi-category fracturing powder cross-powder reference property calibration library. : .

[0036] Step S2 specifically includes the following steps: S21. Identify the current online powder type. ;extract Real-time normalized density feature inversion ,and middle Perform a difference comparison; if the difference is less than the judgment threshold, lock the current powder identifier. .

[0037] S22, Based on real-time powder quality stored in the silo Construct the differential equation for the instantaneous powder discharge flow rate: ; In the formula: Supply flow rate for raw instantaneous powder without disturbance compensation.

[0038] The negative sign indicates that the mass of powder in the silo continuously decreases as the material is discharged, and the dimensions on both sides are unified as the mass of powder per unit time.

[0039] S23. Combining the BK mixed fracture damage model, a partial deviation model for the attenuation of powder fracture toughness induced by temperature and humidity is constructed.

[0040] S231, Introduce humidity offset Corrected real-time fracture toughness of Type I and II powders: ; In the formula: , The fracture toughness of the powder after real-time moisture absorption is classified as Type I and Type II fracture toughness. , These are the humidity-toughness attenuation correction factors for fracture toughness of type I and type II, respectively.

[0041] S232. Real-time calculation of Type I and II critical fracture energies based on plane strain equivalent modulus: ; In the formula: This is the equivalent elastic modulus of the current powder plane strain; , These are the critical energies for real-time Type I fracture and Type II fracture, respectively.

[0042] S233, Modify the BK mixing mode damage equation to adapt the dry powder agglomeration flowability damage characterization: ; In the formula: The critical energy for comprehensive flow damage of powder in real time; , These are the current powder opening energy release rate and shear energy release rate, respectively; The breaking index is the dry powder mixture.

[0043] S234. Constructing flow component deviations: ; In the formula: The flow measurement deviation is caused solely by humidity disturbances; This is the humidity deviation weighting coefficient; The baseline comprehensive damage energy of powder under standard temperature and humidity conditions.

[0044] S24. Construct calculation models for the partial deviation of vehicle vibration and the partial deviation of silo bridging respectively; ; ; In the formula: Measurement deviations are induced solely by equipment vibration; This is the vibration disturbance weighting coefficient; Powder bridging can cause flow measurement errors; This is the bridge erection deviation correction factor; This represents the maximum mechanical opening of the discharge gate; This is an indicator function; it outputs 1 if the condition within the parentheses is true, and 0 if it is false.

[0045] The following factors are considered together: flow rate deviation, vehicle vibration deviation, and silo bridging deviation. ; In the formula: This represents the cumulative deviation of the individual items.

[0046] S25. Establish a second-order state-space model for powder feeding and construct a third-order extended state observer (ESO) to identify the total unmodeled perturbation.

[0047] S251. Define system state variables; Represents instantaneous powder flow rate. Represents the rate of change of flow rate. The representative model does not model all composite time-varying total disturbances; a set of state equations is constructed as follows: ; In the formula: The inherent nonlinear dynamics of powder delivery; The gain is controlled by the screw reference. This is the current standard screw control quantity for powder formulations; The system outputs instantaneous powder flow rate.

[0048] S252, Real-time estimation of the total composite disturbance using the iterative equations of the third-order extended state observer. : ; In the formula: For flow observation residuals; The observer outputs a flow rate estimate; The observed value is the rate of change of flow rate; To obtain an estimate of the total composite disturbance from observation; All are observer gain coefficients; This is a nonlinear filter function used to suppress high-frequency noise in weighing. These are the filter shape and width constants, respectively; , , They are respectively , , The first derivative.

[0049] S253. By combining the superimposed deviations of the individual components and estimating the total composite disturbance, the instantaneous mechanism measurement deviation estimate is obtained: ; In the formula: This is the instantaneous measurement deviation estimate.

[0050] S26. Determine the cause of the disturbance based on the sub-item deviation threshold, and output the classification label. : like and ,but This represents a single humidity-induced moisture absorption disturbance.

[0051] like and ,but Single vehicle-mounted vibration disturbance.

[0052] like and ,but Single silo bridging disturbance.

[0053] If any two or more of the other items exceed the threshold, Multiple disturbance coupling conditions.

[0054] in, These are the critical thresholds for judging the component deviation, vehicle vibration component deviation, and silo bridging component deviation, respectively.

[0055] Step S3 specifically includes the following steps; S31. Unified loading of multi-source pre-input and hierarchical splitting of static and dynamic features for preprocessing.

[0056] S311, Synchronous loading step S1 outputs a standardized multidimensional perturbation feature matrix. and cross-powder reference property calibration library And step S2 outputs the instantaneous mechanism measurement deviation estimate. And disturbance cause classification labels .

[0057] S312. Extraction of current powder static physical property characteristics: Call the current powder identifier obtained in step S2. ,from Retrieving static property vectors ; It is a static characteristic, changing only when the powder is replaced, and remaining constant over a continuous time series for the same powder.

[0058] S313, Temporal dynamic perturbation feature extraction; output matrix from step S1 Extracting the normalized perturbation vector time-by-time This constitutes the basic dynamic time sequence set within the window. ; This is the index for discrete time-series sampling times.

[0059] S314. Temporal Physical Prior Feature Concatenation: For each sampling time within the window... Output step S2 at that moment. , spliced ​​to At the tail, a single-time composite dynamic feature vector is generated. : .

[0060] S315, Global Temporal Input Tensor Assembly: Assemble all windows Stacking the data in chronological order from oldest to newest, generating the original dynamic input tensor of the TFT network. : ; S316. Feature-based hierarchical caching and differentiated storage: Static features are stored separately. TFT network original dynamic input tensor (dynamic timing tensor) The buffers are divided into two paths, which are then fed into the static and dynamic branches of the network to select variables.

[0061] S32. Adaptive feature weight selection and nonlinear mapping are performed using the TFT built-in variable selection network.

[0062] S321. For dynamically input tensors Each internal dimension feature Independently construct a dedicated GRN mapping unit: ; In the formula: For the first The 3D features are dynamically output after nonlinear transformation by the GRN mapping unit; for Inner Dimensional original composite dynamic characteristics; For gated jump residual networks, This is the set of trainable weight parameters for the network.

[0063] S322, Static properties With the transformed dynamic output feature set After fusion, the input is synchronously fed into a global feature evaluation gated residual network, which outputs the original feature weight vector. ; In the formula: This is the original vector of unnormalized feature importance; Gated residual networks are used for global feature evaluation; The total dimension of the composite features at a single time step.

[0064] S323. Perform Softmax normalization to obtain adaptive feature weights; ; In the formula: For the first 3D feature normalization adaptive weights, interval The sum of all dimension weights equals 1; It is an exponential activation function; for At this moment, the first The unnormalized original importance weights of the GRN network output by the static-dynamic joint feature evaluation are used to evaluate the dimensional features. At this moment, the first The unnormalized original importance weights of the GRN network output by the static-dynamic joint feature evaluation are obtained from the dimensional features.

[0065] S324, Weighted fusion simplified dynamic time series features; ; In the formula: The simplified time-varying dynamic feature vector is obtained by filtering and weighting through a variable selection network, and perturbation dimensions with very low contribution to measurement bias are automatically removed.

[0066] S325, Static features With simplified dynamic features Time-by-time splicing generates hybrid temporal features that are fed into the multi-head attention module. .

[0067] S33. Use causal mask multi-head attention module to mine long- and short-term temporal dependencies.

[0068] S331, Feature QKV linear projection: Mixed temporal features of the window Perform a linear transformation to generate the query matrix. Key matrix Value matrix .

[0069] S332. Constructing a temporal causal mask matrix Temporal causal mask matrix Assign values ​​to all elements of the upper triangle The lower triangle and diagonal are set to 0 to strictly shield features of future moments and avoid leakage of time sequence information.

[0070] S333, Constructing a formula for calculating single-head causal attention: ; In the formula: Output the temporal feature matrix for single-head attention; For single-head feature dimension hyperparameters; To query the key inner product temporal correlation matrix; This is a scaling factor to prevent gradient vanishing caused by excessively large inner product values.

[0071] S334, Multi-head stitching and dimensionality reduction fusion: Eight independent attention heads are set up in parallel. The outputs of each group are stitched along the feature dimension and then compressed through a single linear layer to output a global temporal hidden layer feature matrix. This vector simultaneously contains long- and short-term time-series correlation features of static physical properties, historical dynamic disturbances, and step S2 mechanism deviations.

[0072] S34. Construct a physical constraint hybrid loss function using the instantaneous mechanical measurement deviation estimate embedded in S2.

[0073] S341. Predictive Output Layer Definition: Defines the future output of the network. Step ahead measurement deviation 60 steps correspond to a 30-second long time series prediction window.

[0074] S342. Construct a data-driven standard MSE loss term (fitting the actual flow deviation label). ; In the formula: The loss is for fitting pure data; This refers to the total number of steps to predict ahead; For TFT timing fusion deep learning network Predicting measurement deviations step by step; This represents the true value of the actual flow deviation measured on-site.

[0075] S343, the mechanism deviation obtained in S2 As a priori physical constraint, the predicted values ​​are limited to not deviate from the fluid dynamics mechanism range: ; In the formula: The loss is due to the consistency constraint of the mechanism.

[0076] S344. Constructing a hybrid loss function based on physical constraints: ; In the formula: To calculate the global total loss for network training, backpropagation is used to update all network weights. For physical constraint balance weights, and .

[0077] S345. Model Optimization Solution: Minimize using the Adam gradient descent optimizer. An early stopping strategy was set to prevent overfitting, and offline pre-training and online incremental fine-tuning of the TFT temporal fusion deep learning network were completed.

[0078] S35. The inference stage outputs a long-term leading flow deviation prediction sequence.

[0079] S351, S31 is now assembled. and static features The data is fed into the forward propagation of the TFT temporal fusion deep learning network after training convergence.

[0080] S352. The output layer outputs 60 consecutive time-series lead-measure prediction values, which are then arranged in ascending order of prediction time to generate a complete prediction sequence. ; In the formula: This is a long-term time series predictive sequence of leading flow deviation; The next short-term leading deviation; Based on The metering deviation of powder supply flow rate in the second sampling period is predicted by the TFT deep learning network based on real-time historical window data. Based on Based on historical window data, the TFT deep learning network predicts the powder supply flow rate measurement deviation for the 60th sampling period in the future.

[0081] Step S4 specifically includes the following steps; S41, Synchronous reading step S2 output instantaneous mechanism measurement deviation And disturbance cause classification labels Step S3 outputs a long-term time-series lead flow deviation prediction sequence. Step S1 outputs the cross-powder reference property calibration library. and real-time process setting target supply flow .

[0082] S412, Based on screw foundation calibration coefficient Initialize the system reference steady-state control variables: ; In the formula: The reference steady-state speed of the screw; The reference vibration frequency for the silo; This is the reference opening degree for the discharge gate; This is the calibration coefficient for the current powder flow rate and rotation speed; , All of these are general steady-state reference constants of the system.

[0083] S413. Initialize the initial value of the three-dimensional control increment vector. Initially, all components are set to 0; among them, for The three-dimensional collaborative compensation control increment vector obtained by solving at each time step; for Incremental adjustment of screw speed compensation at all times; for Incremental adjustment of vibration frequency compensation in the silo at all times; for The opening of the discharge gate is adjusted and increased in real time.

[0084] At the same time, initialize the initial value of the 3D pseudo gradient identification vector. ; for The updated 3D time-varying pseudo-gradient row vector is adaptively identified at every moment. These are the pseudo-gradient components of screw speed, hopper vibration frequency, and discharge gate opening, respectively. , , , , , These are the property matching weighting coefficients for screw gradient, vibration gradient, and discharge gate gradient, respectively.

[0085] S42. Establish a three-dimensional coupled discrete dynamic linearization model for powder feeding, which combines screw speed, vibration frequency, and discharge gate opening.

[0086] S421, Definition For system output variables; where, for Real-time measurement of powder flow rate.

[0087] S422, Define the three-dimensional control increment vector , for The optimal three-dimensional collaborative compensation control increment vector obtained by the rolling optimization solution at any time; They are respectively The screw speed compensation adjustment increment, hopper vibration frequency compensation adjustment increment, and discharge gate opening compensation adjustment increment are all measured at any given time.

[0088] S423. Constructing the discrete dynamic equations of time-varying pseudo-gradients coupled with multiple actuators: ; In the formula: Predict powder flow rate for the next moment; The current three-dimensional time-varying pseudo-gradient row vector represents the coupling influence weights of single-unit adjustment of screw speed, vibration frequency, and discharge gate opening on flow rate.

[0089] S424. Calculate the change in flow rate at the current moment: ; In the formula: This represents the flow rate difference between adjacent sampling periods.

[0090] S43, Online adaptive identification and update of three-dimensional coupled pseudo-gradient vectors.

[0091] S431. Construct a pseudo-gradient iterative identification and update formula; ; In the formula: for The 3D pseudo-gradient vector is updated at each time step; To identify the pseudo gradient vector in the previous time step; To identify the iteration step size, a constant of 0.7 is fixed. Regularization coefficient, a constant To prevent the denominator from being 0; The squared norm of the control increment vector from the previous step.

[0092] S432, Pseudo-gradient component amplitude constraint: After identification, ... Set upper and lower boundaries separately to avoid gradient drift caused by abrupt changes in physical properties.

[0093] S44. Adaptively allocate actuator adjustment constraint weights based on S2 disturbance labels; S441. Initialize the basic fluctuation constraint weights: , , , These are the basic adjustment penalty coefficients for screw speed, vibration frequency, and discharge gate opening, respectively.

[0094] S442, Read the disturbance cause classification label output in step S2. Adjust the constraint weights of each institution according to the labels: like : ; like : ; like : ; like : ; in, , , These are the real-time penalty coefficients for screw speed, vibration frequency, and discharge gate opening, respectively. S443, Output the adaptive constraint weight triplet at the current time. .

[0095] S45, Fusion and Construct a rolling optimization cost function with actuator fluctuation penalty: ; in, ; In the formula: The total uncompensated flow deviation of the system; This represents the current measured powder flow rate; Optimize the objective function for the current moment.

[0096] S46. Regarding the cost function about Find the partial derivatives and set them to zero to derive the analytical solution for the optimal control increment: ; In the formula: This is the optimal three-dimensional compensation increment vector; This is the controller iteration step size, a constant of 0.85; It is a three-dimensional pseudo-gradient transpose vector; This represents the total adjustment constraint coefficient.

[0097] To each Set upper and lower limits for positive and negative adjustment, and cut off directly to the boundary value when the threshold is exceeded to avoid damage to the actuator due to over-range.

[0098] Step S5 specifically includes the following steps; S51. Read the three-dimensional optimal compensation increment vector output from step S4. and the benchmark steady-state control quantities of the three types of actuators ; Synchronously read system hardware safety range constraint thresholds: ; In the formula: These are the lower and upper limits of the screw speed, respectively; These are the lower and upper limits of the vibration frequency, respectively; These are the lower and upper limits of the discharge gate opening, respectively.

[0099] S52. By superimposing the reference value and the compensation increment, the original unlimited drive setting value is obtained: ; ; ; In the formula: The target rotational speed of the screw without a safety limit; The target vibration frequency for which no safety limit has been applied; The target opening for material discharge without safety limits.

[0100] S53. Execute safety limiting constraints and generate final drive settings: ; ; ; In the formula: This is the legal screw drive speed after amplitude limiting; ; Legal vibration driving frequency after amplitude limiting; This refers to the legal opening degree of the discharge gate after the amplitude is limited.

[0101] Package and execute the final 3D driving instruction vector. .

[0102] S54. Synchronously send drive signals and control the three types of actuators in parallel.

[0103] S541, industrial computer will Convert to hardware-compatible standard analog levels or 485 communication commands.

[0104] S542 synchronously sends control commands to the screw frequency converter driver, the hopper vibration controller, and the discharge gate servo driver to achieve synchronous and coordinated adjustment of the three types of mechanisms.

[0105] S543, Set and adjust the stable waiting time Wait for the powder delivery flow rate to reach a new steady state to avoid instantaneous fluctuations interfering with the next round of sampling data.

[0106] S55. Closed-loop data retrieval: Send a new round of raw sensor data into step S1 to complete the full process iteration.

[0107] S551, Stable Waiting Time After completion, step S1 dynamic sampling logic is triggered to synchronously acquire all original analog signals.

[0108] S552, Collect the original sensor dataset with timestamps. The complete input is fed back to step S1, and S1 is executed again to achieve closed-loop acquisition iteration.

[0109] Step S5 is followed by S6, which involves monitoring changes in online powder density to identify powder switching conditions, using transfer learning to update the cross-powder reference property calibration library stored in S1, and simultaneously supplying the updated property parameters to S2, S3, S4, and S5 to complete adaptive calibration without manual shutdown, thereby continuously optimizing the accuracy of metering deviation compensation.

[0110] Step S6 specifically includes the following steps; S61. Monitor online powder density changes to identify powder switching conditions.

[0111] S611. Continuously read the filtered fracturing fluid powder bulk density output from step S1. And the standard reference density of guar gum, debunker and anti-swelling agent; , , ; S612. Calculate the absolute difference between the real-time density and the three reference densities: ; In the formula: These represent the differences between the real-time density and the baseline density of guar gum, degumming agent, and anti-swelling agent, respectively.

[0112] S613, Set powder switching judgment threshold And the judgment logic: If the current minimum density difference A powder switching event is detected, and the switching trigger flag is set. ;on the contrary If the powder is not replaced, skip this step in the subsequent update process; in this embodiment, it is set that... ; S614, if Record the old powder label before switching. Identify new powder labels (The powder category corresponds to the smallest difference), cache switch start time. .

[0113] S62. Collect the mean vector of steady-state temporal disturbance characteristics after powder switching.

[0114] S621, After powder switching is triggered, continue waiting. The steady-state runtime is during which the normalized perturbation feature vector output from each round of S1 is continuously received. .

[0115] S622. Constructing a steady-state feature dataset after the switch. The arithmetic mean of all steady-state eigenvectors is used to obtain the steady-state average perturbation eigenvector.

[0116] ; In the formula: This represents the steady-state average multidimensional perturbation deviation vector after powder switching; This represents the total number of sampled vectors within a 30-minute steady-state window, and .

[0117] S63. Transfer learning updates the corresponding category of powder reference physical property parameter vector.

[0118] S631, retrieve the original old powder reference parameter vector of S18. The new physical property baseline vector is calculated using the transfer learning incremental update formula: ; In the formula: To update the new powder reference property vector; ; Transfer learning updates step size.

[0119] S632, Parameter Boundary Constraint Verification: [This section appears to be incomplete and requires further context.] All internal physical property parameters are set to reasonable upper and lower limits for the oilfield, and those exceeding the boundary values ​​are forcibly truncated to the limit range.

[0120] S633, use the updated version Original covering S18 Generate and update the global cross-powder reference property calibration library. .

[0121] S64. Based on the updated global cross-powder reference property calibration library Repeat steps S2, S3, S4, and S5.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for adaptive compensation and control of fracturing fluid powder metering deviation, characterized in that: Includes the following steps: S1. Simultaneously collect multi-source field sensor raw signals including total powder mass, powder density, temperature and humidity, triaxial vibration, screw speed, and discharge gate displacement. After Kalman filtering for noise reduction, multi-disturbance feature derivation and normalization processing, a standardized multi-dimensional disturbance feature matrix is ​​obtained. Combined with laboratory offline physical property test data, a cross-powder benchmark physical property calibration library for multiple types of fracturing powder is established. S2. Based on the standardized multidimensional disturbance feature matrix and cross-powder reference property calibration library output by S1, a sub-item measurement deviation model is constructed using Biot fluid-structure interaction theory and BK mixing mode powder damage equation. A third-order extended state observer is used to uniformly identify all time-varying composite disturbances in real time, and output instantaneous mechanism measurement deviation estimates and disturbance cause classification labels. S3. Based on the standardized multidimensional perturbation feature matrix output by S1 and the cross-powder reference property calibration library, as well as the instantaneous mechanism measurement deviation estimate and perturbation cause classification label output by S2, a TFT time-series fusion deep learning network with built-in variable selection network and causal mask multi-head attention is built to separate the static inherent property of powder and time-varying dynamic perturbation features. The instantaneous mechanism measurement deviation estimate of S2 is embedded to construct a physical constraint hybrid loss function, and the long-time series leading flow deviation prediction sequence is output. S4. Based on the instantaneous timing measurement deviation estimate of S2 output, the long-term time-series advanced flow deviation prediction sequence of S3 output, and the cross-powder reference physical property calibration library of S1 output, a three-dimensional coupled adaptive control model of screw speed, silo vibration frequency, and discharge gate opening is constructed. According to the disturbance cause classification label of S2, the adjustment weights of multiple actuators are adaptively allocated, and the three-dimensional collaborative compensation control vector is solved by rolling optimization. S5. Based on the three-dimensional collaborative compensation control vector output by S4, the actual drive setting values ​​of screw speed, hopper vibration frequency and discharge gate opening are obtained by superimposing the reference steady-state control quantities of each powder. After hardware safety limiting, the screw, vibrator and discharge gate servo mechanism are synchronously controlled. After the control is completed, a new round of sensor data is collected and input into S1 to realize closed-loop acquisition iteration.

2. The method for adaptive compensation and control of fracturing fluid powder metering deviation according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11. Read the current system-set target powder supply flow rate. The globally unified sampling period is dynamically determined based on the adaptive sampling interval formula of the flow rate. : ； In the formula: This is the maximum sampling interval of the system, and the baseline value for low displacement under no-load conditions. This is the flow rate-sampling interval conversion factor; When calculated If the sampling period is less than 0.5s, the minimum sampling period is forcibly locked at 0.5s. S12, according to the calculation Periodically and synchronously acquire raw signals from multiple sources of field sensors. These raw signals include the total mass of fracturing fluid powder stored in the silo in real time. Real-time bulk density of fracturing fluid powder at the silo outlet under flow conditions Real-time temperature of the on-site operating environment of the fracturing fluid powder feeding equipment and humidity Triaxial real-time vibration acceleration of fracturing fluid powder feeding equipment Real-time operating speed of the powder conveying screw Real-time opening stroke of the discharge gate on the hopper discharge port ; S13. Perform Kalman filtering on the original signals from the multi-source field sensors to obtain the filtered set of basic physical quantities. ; in, These are the total mass of the filtered stored powder, the powder bulk density, the ambient temperature and humidity, the vibration acceleration along the x, y, and z axes, the screw speed, and the discharge gate opening. S14. Derivation of multidimensional perturbation characteristics based on the set of fundamental physical quantities: Vibration amplitude disturbance characteristics derived ; Humidity-related powder density perturbation deviation characteristics derived ; Temperature and humidity reference offset perturbation characteristics derived ; Characteristics of residual mass load disturbance deviation in silos ; Screw speed operating condition disturbance deviation characteristics ; Discharge gate opening throttling disturbance deviation characteristics ; In the formula: The overall vibration disturbance amplitude of the equipment; This represents the powder density perturbation offset. For the first Dry basis bulk density of fracturing powder; This represents the offset due to ambient temperature disturbance. This represents the offset due to environmental relative humidity disturbance. To calibrate the temperature and humidity reference constants for laboratory standards; This refers to the deviation of the powder load in the silo. The rated steady-state filling reference mass of the silo; This refers to the screw speed disturbance deviation; The screw speed is calibrated to the current steady-state reference of the powder. This refers to the disturbance deviation of the discharge gate opening. The discharge gate opening is the steady-state reference value for the system. S15. Concatenate the complete original perturbation vector at a single moment, and integrate all basic physical quantities and derived perturbation features to obtain the perturbation features. : ； In the formula, This is a transpose operation; S16, Disturbance characteristics Normalization mapping to The interval is used to obtain the normalized perturbation eigenvector at a single time step. ; S17. Temporal sliding window splicing to generate a standardized multidimensional perturbation feature matrix; S171, Set the length of the timing sliding window There are [number] sampling points, and the total window duration is [time]. ; S172. According to the sampling time from farthest to most recent, the continuous The output of S16 at each moment Vertical sequential concatenation generates a standardized multidimensional perturbation feature matrix. : ； S18. Offline physical property testing combined with online iterative updates to construct a cross-powder benchmark physical property calibration library for multiple types of fracturing powders.

3. The method for adaptive compensation and control of fracturing fluid powder metering deviation according to claim 2, characterized in that: Step S18 specifically includes the following steps: S181, in Under standard offline laboratory conditions, steady-state flowability tests were conducted on three types of fracturing powders: guar gum thickener, breaker, and anti-swelling agent. A fixed set of reference parameters was determined for each type of powder, resulting in a reference parameter vector for each powder type. : ； In the formula: For the first Powder-based base bulk density, , Representative guar gum powder, Represents de-colloidal agents, Represents anti-swelling powder; It exhibits Type I fracture toughness; It exhibits type II fracture toughness; For the first Equivalent elastic modulus of powder-like agents; Poisson's ratio; For the first The internal friction coefficient of powder-like materials; This is the linear calibration coefficient for screw flow rate-speed. The critical vibration amplitude for bridging in the silo; S182. Integrate the reference parameter vectors of the three types of powders to generate an initial global multi-category fracturing powder cross-powder reference property calibration library. : 。 4. The method for adaptive compensation and control of fracturing fluid powder metering deviation according to claim 3, characterized in that: Step S2 specifically includes the following steps: S21. Identify the current online powder type. ;extract Real-time normalized density feature inversion ,and middle Perform a difference comparison; if the difference is less than the judgment threshold, lock the current powder identifier. ; S22, Based on real-time powder quality stored in the silo Construct the differential equation for the instantaneous powder discharge flow rate: ； In the formula: Supply flow rate for raw instantaneous powder without disturbance compensation; S23. Combine the BK mixed fracture damage model to construct a partial deviation model for the attenuation of powder fracture toughness induced by temperature and humidity. S231, Introduce humidity offset Corrected real-time fracture toughness of Type I and II powders: ； In the formula: , The fracture toughness of the powder after real-time moisture absorption is classified as Type I and Type II fracture toughness. , These are the humidity-toughness attenuation correction factors for fracture toughness of type I and type II, respectively. S232. Real-time calculation of Type I and II critical fracture energies based on plane strain equivalent modulus: ； In the formula: This is the equivalent elastic modulus of the current powder plane strain; , These are the critical energy for real-time Type I fracture and the critical energy for Type II fracture, respectively. S233, Modify the BK mixing mode damage equation to adapt the dry powder agglomeration flowability damage characterization: ； In the formula: The critical energy for comprehensive flow damage of powder in real time; , These are the current powder opening energy release rate and shear energy release rate, respectively; The breaking index of dry powder mixture; S234. Constructing flow component deviations: ； In the formula: The flow measurement deviation is caused solely by humidity disturbances; This is the humidity deviation weighting coefficient; The baseline comprehensive damage energy of powder under standard temperature and humidity; S24. Construct calculation models for the partial deviation of vehicle vibration and the partial deviation of silo bridging respectively; ； ； In the formula: Measurement deviations are induced solely by equipment vibration; This is the vibration disturbance weighting coefficient; Powder bridging can cause flow measurement errors; This is the bridge erection deviation correction factor; This represents the maximum mechanical opening of the discharge gate; This is an indicator function; it outputs 1 if the condition inside the parentheses is true, and 0 if it is false. The following factors are considered together: flow rate deviation, vehicle vibration deviation, and silo bridging deviation. ； In the formula: This is the sum of the deviations from the individual items; S25. Establish a second-order state-space model for powder feeding and construct a third-order extended state observer (ESO) to identify the total unmodeled perturbation. S251. Define system state variables; Represents instantaneous powder flow rate. Represents the rate of change of flow rate. The representative model does not model all composite time-varying total disturbances; a set of state equations is constructed as follows: ； In the formula: The inherent nonlinear dynamics of powder delivery; The gain is controlled by the screw reference. This is the current standard screw control quantity for powder formulations; The system outputs instantaneous powder flow rate; S252, Real-time estimation of the total composite disturbance using the iterative equations of the third-order extended state observer. : ； In the formula: For flow observation residuals; The observer outputs a flow rate estimate; The observed value is the rate of change of flow rate; To obtain an estimate of the total composite disturbance from observation; All are observer gain coefficients; It is a nonlinear filtering function; These are the filter shape and width constants, respectively; , , They are respectively , , The first derivative; S253. By combining the superimposed deviations of the individual components and estimating the total composite disturbance, the instantaneous mechanism measurement deviation estimate is obtained: ； In the formula: This is the instantaneous measurement deviation estimate; S26. Determine the cause of the disturbance based on the sub-item deviation threshold, and output the classification label. : like and ,but This represents a single humidity-absorbing disturbance. like and ,but Single vehicle-mounted vibration disturbance; like and ,but Bridge disturbance in a single silo; If any two or more of the other items exceed the threshold, Multiple disturbance coupling conditions; in, These are the critical thresholds for judging the component deviation, vehicle vibration component deviation, and silo bridging component deviation, respectively.

5. The method for adaptive compensation and control of fracturing fluid powder metering deviation according to claim 4, characterized in that: Step S3 details Includes the following steps; S31. Unified loading of multi-source pre-input and hierarchical splitting of static and dynamic features for preprocessing; S311, Synchronous loading step S1 outputs a standardized multidimensional perturbation feature matrix. and cross-powder reference property calibration library And step S2 outputs the instantaneous mechanism measurement deviation estimate. And disturbance cause classification labels ; S312. Extraction of current powder static physical property characteristics: Call the current powder identifier obtained in step S2. ,from Retrieving static property vectors ; S313, Extraction of temporal dynamic perturbation features; Output matrix from step S1 Extracting the normalized perturbation vector time-by-time This constitutes the basic dynamic time sequence set within the window. ; This is an index for discrete time-series sampling times; S314. Temporal Physical Prior Feature Concatenation: For each sampling time within the window... Output step S2 at that moment. , spliced ​​to At the tail, a single-time composite dynamic feature vector is generated. : ； S315, Global Temporal Input Tensor Assembly: Assemble all windows Stacking the data in chronological order from oldest to newest, generating the original dynamic input tensor of the TFT network. : ； S316. Feature-based hierarchical caching and differentiated storage: Static features are stored separately. TFT network original dynamic input tensor The buffer is divided into two paths, which are respectively fed into the static and dynamic branches of the network to select variables; S32. Utilize the TFT built-in variable selection network to perform adaptive feature weight selection and nonlinear mapping; S321. For dynamically input tensors Each internal dimension feature Independently construct a dedicated GRN mapping unit: ； In the formula: For the first The 3D features are dynamically output after nonlinear transformation by the GRN mapping unit; for Inner Dimensional original composite dynamic characteristics; For gated jump residual networks, This is the set of trainable weight parameters for the network; S322, Static properties With the transformed dynamic output feature set After fusion, the input is synchronously fed into a global feature evaluation gated residual network, which outputs the original feature weight vector. ； In the formula: This is the original vector of unnormalized feature importance; Gated residual networks are used for global feature evaluation; The total dimension of the composite features at a single time step; S323. Perform Softmax normalization to obtain adaptive feature weights; ； In the formula: For the first 3D feature normalization adaptive weights; It is an exponential activation function; for At this moment, the first The unnormalized original importance weights of the GRN network output by the static-dynamic joint feature evaluation are used to evaluate the dimensional features. At this moment, the first The unnormalized original importance weights of the GRN network output by the static-dynamic joint feature evaluation are used to evaluate the dimensional features. S324, Weighted fusion simplified dynamic time series features; ； In the formula: The simplified time-series dynamic feature vector is obtained by filtering and weighting through a variable selection network. S325, Static features With simplified dynamic features Time-by-time splicing generates hybrid temporal features that are fed into the multi-head attention module. ; S33. Utilize causal masking multi-head attention module to mine long- and short-term temporal dependencies; S331, Feature QKV linear projection: Mixed temporal features of the window Perform a linear transformation to generate the query matrix. Key matrix Value matrix ; S332. Constructing a temporal causal mask matrix Temporal causal mask matrix Assign values ​​to all elements of the upper triangle Set the lower triangle and diagonal to 0; S333, Constructing a formula for calculating single-head causal attention: ； In the formula: Output the temporal feature matrix for single-head attention; For single-head feature dimension hyperparameters; To query the key inner product temporal correlation matrix; This is a scaling factor to prevent gradient vanishing caused by excessively large inner product values; S334, Multi-head stitching and dimensionality reduction fusion: Eight independent attention heads are set up in parallel. The outputs of each group are stitched along the feature dimension and then compressed through a single linear layer to output a global temporal hidden layer feature matrix. ; S34. Construct a physical constraint hybrid loss function using the instantaneous mechanical measurement deviation estimate embedded in S2; S341. Predictive Output Layer Definition: Defines the future output of the network. Step ahead measurement deviation ; S342. Constructing a data-driven standard MSE loss term ； In the formula: The loss is for fitting pure data; This refers to the total number of steps to predict ahead; For TFT timing fusion deep learning network Predicting measurement deviations step by step; This is the true value of the actual flow deviation measured on-site; S343, the mechanism deviation obtained in S2 As a priori physical constraint: ； In the formula: Loss due to mechanism consistency constraints; S344. Constructing a hybrid loss function based on physical constraints: ； In the formula: This represents the total global loss during network training. For physical constraint balance weights; S345. Model Optimization Solution: Minimize using the Adam gradient descent optimizer. An early stopping strategy was set up to complete the offline pre-training and online incremental fine-tuning of the TFT temporal fusion deep learning network; S35. The inference stage outputs a long-time time-series leading flow deviation prediction sequence. S351, S31 is now assembled. and static features The data is fed into the forward propagation of the converged TFT temporal fusion deep learning network. S352. The output layer outputs 60 consecutive time-series lead-measure prediction values, which are then arranged in ascending order of prediction time to generate a complete prediction sequence. ； In the formula: This is a long-term time series predictive sequence of leading flow deviation; ; The next short-term leading deviation; Based on The metering deviation of powder supply flow rate in the second sampling period is predicted by the TFT deep learning network based on real-time historical window data. Based on Based on historical window data, the TFT deep learning network predicts the powder supply flow rate measurement deviation for the 60th sampling period in the future.

6. The method for adaptive compensation and control of fracturing fluid powder metering deviation according to claim 5, characterized in that: Step S4 specific Includes the following steps; S41, Synchronous reading step S2 output instantaneous mechanism measurement deviation And disturbance cause classification labels Step S3 outputs a long-term time-series lead flow deviation prediction sequence. Step S1 outputs the cross-powder reference property calibration library. and real-time process setting target supply flow ; S412, Based on screw foundation calibration coefficient Initialize the system reference steady-state control variables: ； In the formula: The reference steady-state speed of the screw; The reference vibration frequency for the silo; This is the reference opening degree for the discharge gate; This is the calibration coefficient for the current powder flow rate and rotation speed; , All of these are system-wide steady-state reference constants; S413. Initialize the initial value of the three-dimensional control increment vector. Initially, all components are set to 0; among them, for The three-dimensional collaborative compensation control increment vector obtained by solving at each time step; for Incremental adjustment of screw speed compensation at all times; for Incremental adjustment of vibration frequency compensation in the silo at all times; for The discharge gate opening is constantly compensated and adjusted in increments; At the same time, initialize the initial value of the 3D pseudo gradient identification vector. ; for The updated 3D time-varying pseudo-gradient row vector is adaptively identified at every moment. These are the pseudo-gradient components of screw speed, hopper vibration frequency, and discharge gate opening, respectively. , , , , , These are the weighting coefficients for property matching of screw gradient, vibration gradient, and discharge gate gradient, respectively. S42. Establish a three-dimensional coupled discrete dynamic linearization model of powder feeding based on screw speed, vibration frequency, and discharge gate opening. S421, Definition For system output variables; where, for Real-time measurement of powder flow rate; S422, Define the three-dimensional control increment vector , for The optimal three-dimensional collaborative compensation control increment vector obtained by the rolling optimization solution at any time; They are respectively The screw speed compensation adjustment increment, the hopper vibration frequency compensation adjustment increment, and the discharge gate opening compensation adjustment increment at any given moment; S423. Constructing the discrete dynamic equations of time-varying pseudo-gradients coupled with multiple actuators: ； In the formula: Predict powder flow rate for the next moment; This is the current three-dimensional time-varying pseudo-gradient row vector; S424. Calculate the change in flow rate at the current moment: ； In the formula: The difference in flow rate between adjacent sampling periods; S43. Online adaptive identification and updating of three-dimensional coupled pseudo-gradient vectors; S431. Construct a pseudo-gradient iterative identification and update formula; ； In the formula: for The 3D pseudo-gradient vector is updated at each time step; To identify the pseudo gradient vector in the previous time step; To identify the iteration step size; The regularization coefficient is used. The squared norm of the control increment vector from the previous step; S432, Pseudo-gradient component amplitude constraint: After identification, ... Set the upper and lower boundaries respectively; S44. Adaptively allocate actuator adjustment constraint weights based on S2 disturbance labels; S441. Initialize the basic fluctuation constraint weights: , , , These are the basic adjustment penalty coefficients for screw speed, vibration frequency, and discharge gate opening, respectively. S442, Read the disturbance cause classification label output in step S2. Adjust the constraint weights of each institution according to the labels: like : ; like : ; like : ; like : ; in, , , These are the real-time penalty coefficients for screw speed, vibration frequency, and discharge gate opening, respectively. S443, Output the adaptive constraint weight triplet at the current time. ; S45, Fusion and Construct a rolling optimization cost function with actuator fluctuation penalty: ； in, ； In the formula: The total uncompensated flow deviation of the system; This represents the current measured powder flow rate; Optimize the objective function for the current moment; S46. Regarding the cost function about Find the partial derivatives and set them to zero to derive the analytical solution for the optimal control increment: ； In the formula: This is the optimal three-dimensional compensation increment vector; This represents the controller iteration step size; It is a three-dimensional pseudo-gradient transpose vector; This is the total adjustment constraint coefficient; To each Set upper and lower limits for positive and negative adjustment, and truncate directly to the boundary value when the threshold is exceeded.

7. The method for adaptive compensation and control of fracturing fluid powder metering deviation according to claim 6, characterized in that: Step S5 specifically includes the following steps; S51. Read the three-dimensional optimal compensation increment vector output from step S4. and the benchmark steady-state control quantities of the three types of actuators ; Synchronously read system hardware safety range constraint thresholds: ； In the formula: These are the lower and upper limits of the screw speed, respectively; These are the lower and upper limits of the vibration frequency, respectively; These are the lower and upper limits of the discharge gate opening, respectively; S52. By superimposing the reference value and the compensation increment, the original unlimited drive setting value is obtained: ； ； ； In the formula: The target rotational speed of the screw without a safety limit; The target vibration frequency for which no safety limit has been applied; The target opening for material discharge without a safety limit; S53. Execute safety limiting constraints and generate final drive settings: ； ； ； In the formula: This is the legal screw drive speed after amplitude limiting; Legal vibration drive frequency after amplitude limiting; This is the legal discharge gate opening after the amplitude is limited; Package and execute the final 3D driving instruction vector. ; S54. Synchronously send drive signals and control three types of actuators in parallel; S541, industrial computer will Convert to hardware-compatible standard analog levels or 485 communication commands; S542 synchronously sends control commands to the screw frequency converter driver, the hopper vibration controller, and the discharge gate servo driver to achieve synchronous and coordinated adjustment of the three types of mechanisms; S543, Set and adjust the stable waiting time ; S55. Closed-loop data retrieval: Send a new round of raw sensor data into step S1 to complete the full process iteration. S551, Stable Waiting Time After completion, step S1 dynamic sampling logic is triggered to synchronously acquire all original analog signals; S552, Collect the original sensor dataset with timestamps. The complete input is fed back to step S1, and S1 is executed again to achieve closed-loop acquisition iteration.

8. The method for adaptive compensation and control of fracturing fluid powder metering deviation according to claim 7, characterized in that: Step S5 is followed by S6, which involves monitoring changes in online powder density to identify powder switching conditions, using transfer learning to update the cross-powder reference property calibration library stored in S1, and simultaneously supplying the updated property parameters to S2, S3, S4, and S5 to complete adaptive calibration without manual shutdown, thereby continuously optimizing the accuracy of metering deviation compensation.

9. The method for adaptive compensation and control of fracturing fluid powder metering deviation according to claim 8, characterized in that: Step S6 specifically includes the following steps; S61. Monitor online powder density changes to identify powder switching conditions; S611. Continuously read the filtered fracturing fluid powder bulk density output from step S1. And the standard reference density of guar gum, debunker and anti-swelling agent; , , ; S612. Calculate the absolute difference between the real-time density and the three reference densities: ； In the formula: These are the differences between the real-time density and the baseline density of guar gum, debacterizing agent, and anti-swelling agent, respectively. S613, Set powder switching judgment threshold And the judgment logic: If the current minimum density difference A powder switching event is detected, and the switching trigger flag is set. ;on the contrary If the powder has not been replaced, skip this step and follow the updated process later. S614, if Record the old powder label before switching. Identify new powder labels Cache switching start time ; S62. Collect the mean vector of steady-state temporal disturbance characteristics after powder switching; S621, After powder switching is triggered, continue waiting. The steady-state runtime is during which the normalized perturbation feature vector output from each round of S1 is continuously received. ; S622. Constructing a steady-state feature dataset after the switch. The steady-state average perturbation eigenvector is obtained by taking the arithmetic mean of all steady-state eigenvectors: ； In the formula: This represents the steady-state average multidimensional perturbation deviation vector after powder switching; This represents the total number of sampled vectors within a 30-minute steady-state window, and ; S63. Update the corresponding category of powder reference physical property parameter vector through transfer learning; S631, retrieve the original old powder reference parameter vector of S18. The new physical property baseline vector is calculated using the transfer learning incremental update formula: ； In the formula: To update the new powder reference property vector; ; Transfer learning updates step size; S632, Parameter Boundary Constraint Verification: [This section appears to be incomplete and requires further context.] All internal physical property parameters are set to reasonable upper and lower limits for the oilfield site, and those exceeding the boundary values ​​are forcibly truncated to the limit range. S633, use the updated version Original covering S18 Generate and update the global cross-powder reference property calibration library. ; S64. Based on the updated global cross-powder reference property calibration library Repeat steps S2, S3, S4, and S5.

Images