Automatic control method based on artificial intelligence

Abstract

The invention discloses an automatic control method based on artificial intelligence, relates to the technical field of petroleum drilling, and is applied to a slurry pump. The method comprises the following steps: acquiring a slurry pump, determining fluid pulsation characteristics, mechanical wear characteristics and pump efficiency loss characteristics of the slurry pump according to a preset multi-physics field coupling model so as to construct a multi-modal characteristic vector, then generating an initial control instruction based on a preset fuzzy RBF neural network model, and finally verifying and correcting the initial control instruction based on a digital twinborn model so as to obtain a digital twinborn model. And generating a final control instruction to adjust the working state of the slurry pump according to the final control instruction, thereby realizing real-time monitoring and intelligent control of the running state of the slurry pump, being independent of artificial experience, improving timeliness and accuracy of slurry pump control, and further improving efficiency and safety of drilling operation.

Description

Technical Field

[0001] This application relates to the technical field of oil drilling, and particularly to an automatic control method based on artificial intelligence. Background Art

[0002] In the field of oil drilling, mud pumps are the core equipment for maintaining the circulation of drilling fluid and ensuring the stability of the wellbore. In the prior art, generally when abnormal fluctuations occur in the operating parameters of the mud pump (such as displacement and pressure), the parameters of the PID controller are adjusted through manual experience or the stroke frequency is manually corrected. The above methods rely on the subjective judgment of the operator and there is a response lag, which often leads to the working state of the mud pump not being adjusted to the optimal in time, affecting the efficiency and safety of drilling operations.

[0003] The above content is only used to assist in understanding the technical solution of this application, and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide an automatic control method based on artificial intelligence, aiming to improve the timeliness and accuracy of mud pump control.

[0005] To achieve the above object, this application proposes an automatic control method based on artificial intelligence, which is applied to a mud pump. The method includes: Collect the operating data of the mud pump; Perform dynamic normalization processing on the operating data to obtain corresponding preprocessed data; Obtain the fluid pulsation characteristics and mechanical wear characteristics of the mud pump according to the preprocessed data; Obtain the pump efficiency loss characteristics of the mud pump according to the preprocessed data and a preset multi-physical field coupling model; Construct a multi-modal feature vector based on the fluid pulsation characteristics, mechanical wear characteristics, and pump efficiency loss characteristics of the mud pump, and input it into a pre-trained variational autoencoder for feature dimension reduction to obtain a target feature vector; Input the target feature vector into a preset fuzzy RBF neural network model to obtain corresponding initial control instructions; the initial control instructions include a displacement compensation value and a stroke frequency adjustment amount; Pre-verify the initial control instructions based on a digital twin model to generate a safe operation domain including three-dimensional constraints of pressure-displacement-frequency; Dynamically correct the displacement compensation value and the stroke frequency adjustment amount according to the safe operation domain to generate a final control instruction, so as to adjust the working state of the mud pump according to the final control instruction.

[0006] In an embodiment, the step of performing dynamic normalization processing on the operating data includes: Determine a dynamic window according to a preset rule; Calculate the mean and standard deviation of the running data within the dynamic window; Substitute the mean and standard deviation into the following formula to standardize the running data: ; where, represents a single sampled value of the running data; represents the mean of the running data within the dynamic window; represents the standard deviation of the running data within the dynamic window; represents a very small constant.

[0007] In one embodiment, the running data includes pressure data and power spectral density data; The fluid pulsation characteristics include pressure pulsation intensity and main frequency energy ratio; The pressure pulsation intensity is calculated by the following formula: ; where, represents the pressure pulsation intensity, represents the number of sampling points within the dynamic window, where, , represents the dynamic window length, represents the sampling rate; represents the i-th sampling point of the preprocessed pressure data; represents the mean of the preprocessed pressure data within the dynamic window; The main frequency energy ratio is calculated by the following formula: ; where, represents the main frequency in the power spectral density, represents the power spectral density function; represents the highest effective frequency; represents the energy integral within the frequency interval ;

[0008] In one embodiment, the running data includes vibration data and acoustic emission data; The mechanical wear characteristics include acoustic emission energy accumulation and vibration intensity; The acoustic emission energy accumulation is calculated by the following formula: ; where, represents the accumulated energy of the preprocessed acoustic emission data within the dynamic window; represents the amplitude of the preprocessed acoustic emission signal at the -th sampling point; represents the sampling time interval; represents the number of sampling points within the dynamic window length, where , represents the dynamic window length, represents the sampling rate; The vibration intensity is calculated by the following formula: ; where represents the vibration intensity; represents the vibration data after envelope change at the i-th sampling point; represents the number of sampling points within the dynamic window length, where , represents the dynamic window length, represents the sampling rate.

[0009] In one embodiment, the pump efficiency loss characteristics include volumetric efficiency deviation and leakage equivalent area; The step of obtaining the pump efficiency loss characteristics of the mud pump according to the preprocessed data and a preset multi-physical field coupling model includes: Obtaining the theoretical displacement of the mud pump based on the preset multi-physical field coupling model; Obtaining the actual displacement of the mud pump based on the preprocessed data; Calculating the volumetric efficiency deviation and the leakage equivalent area based on the theoretical displacement, actual displacement, and preprocessed data.

[0010] In one embodiment, the operating data includes stroke frequency data, flow rate data, fluid data, and inlet and outlet pressure difference data; The theoretical displacement is calculated by the following formula: ; where represents the theoretical displacement, represents the stroke frequency; represents the piston cross-sectional area corresponding to the preset multi-physical field coupling model; represents the effective stroke length corresponding to the preset multi-physical field coupling model; represents the volumetric efficiency corresponding to the preset multi-physical field coupling model; represents the mechanical efficiency corresponding to the preset multi-physical field coupling model; represents the compressibility correction factor corresponding to the preset multi-physical field coupling model; The actual displacement is calculated based on the following formula: ; where, represents the actual displacement; represents the dynamic window length; represents the instantaneous flow rate; represents the integration variable.

[0011] In one embodiment, the volumetric efficiency deviation is calculated by the following formula: ; where, represents the actual displacement; represents the theoretical displacement; The leakage equivalent area is calculated by the following formula: ; where, represents the pressure difference between the inlet and outlet of the pump; represents the fluid density; represents the flow coefficient.

[0012] In one embodiment, the method further includes: Adopting a dual-time scale update strategy to perform online learning and update on the fuzzy RBF neural network, specifically including: Adjusting the network weights and membership function parameters based on the real-time error gradient to quickly respond to transient working conditions; and, Dynamically optimizing the network structure through sliding window statistical features to ensure long-term stability and suppress overfitting.

[0013] In one embodiment, the step of pre-verifying the initial control instruction based on the digital twin model to generate a safe operation domain including three-dimensional constraints of pressure-displacement-frequency includes: Receiving the initial control instruction and predicting the dynamic responses of pressure, displacement, and frequency based on the digital twin model; Extracting the pressure peak value, displacement fluctuation threshold, and resonance frequency forbidden zone based on the historical safety data of the digital twin model; Fitting the constraint relationships between the pressure peak value, displacement fluctuation threshold, resonance frequency forbidden zone and the predicted dynamic responses to generate a safe operation domain including three-dimensional constraints of pressure-displacement-frequency.

[0014] The automatic control method based on artificial intelligence proposed in this application collects the operation data of the mud pump, performs dynamic standardization processing on the operation data to obtain corresponding preprocessed data; then obtains the fluid pulsation characteristics and mechanical wear characteristics of the mud pump according to the preprocessed data; and obtains the pump efficiency loss characteristics of the mud pump according to the preprocessed data and a preset multi-physical field coupling model; then constructs a multi-modal feature vector based on the fluid pulsation characteristics, mechanical wear characteristics and pump efficiency loss characteristics of the mud pump, and inputs it into a pre-trained variational autoencoder for feature dimension reduction to obtain a target feature vector; and inputs the target feature vector into a preset fuzzy RBF neural network model to obtain corresponding initial control instructions; the initial control instructions include a displacement compensation value and a stroke frequency adjustment amount; then pre-verify the initial control instructions based on the digital twin model to generate a safe operation domain including three-dimensional constraints of pressure-displacement-frequency; finally, dynamically correct the displacement compensation value and the stroke frequency adjustment amount according to the safe operation domain to generate a final control instruction, so as to adjust the working state of the mud pump according to the final control instruction. That is, this solution collects the operation data of the mud pump, determines the fluid pulsation characteristics, mechanical wear characteristics and pump efficiency loss characteristics of the mud pump according to a preset multi-physical field coupling model, thereby constructing a multi-modal feature vector, then generates initial control instructions based on a preset fuzzy RBF neural network model, and finally verifies and corrects the initial control instructions based on the digital twin model to generate a final control instruction, so as to adjust the working state of the mud pump according to the final control instruction, realizing real-time monitoring and intelligent control of the operation state of the mud pump, not relying on manual experience, improving the timeliness and accuracy of mud pump control, and further improving the efficiency and safety of drilling operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0016] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic flowchart provided for an embodiment of the automatic control method based on artificial intelligence in this application; Figure 2 For Figure 1 the detailed flowchart of step S400 in Figure 3 For Figure 1 the detailed flowchart of step S700 in

[0018] The realization of the purpose, functional features, and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. Detailed implementation manners

[0019] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not used to limit this application.

[0020] To better understand the technical solutions of this application, the following will be described in detail in conjunction with the accompanying drawings of the specification and specific implementation manners.

[0021] The main solution of the embodiment of this application is as follows: collect the operation data of the mud pump, perform dynamic standardization processing on the operation data to obtain corresponding preprocessed data; then obtain the fluid pulsation characteristics and mechanical wear characteristics of the mud pump according to the preprocessed data; and obtain the pump efficiency loss characteristics of the mud pump according to the preprocessed data and a preset multi-physical field coupling model; then construct a multi-modal feature vector based on the fluid pulsation characteristics, mechanical wear characteristics, and pump efficiency loss characteristics of the mud pump, and input it into a pre-trained variational autoencoder for feature dimensionality reduction to obtain a target feature vector; and input the target feature vector into a preset fuzzy RBF neural network model to obtain corresponding initial control instructions; the initial control instructions include a displacement compensation value and a stroke frequency adjustment amount; then pre-verify the initial control instructions based on the digital twin model to generate a safe operation domain including three-dimensional constraints of pressure-displacement-frequency; finally, dynamically correct the displacement compensation value and the stroke frequency adjustment amount according to the safe operation domain to generate a final control instruction, so as to adjust the working state of the mud pump according to the final control instruction.

[0022] In this embodiment, for the convenience of description, the following will be described with an automatic control device based on artificial intelligence as the execution subject.

[0023] In the field of oil drilling, the mud pump is a core device for maintaining the circulation of drilling fluid and ensuring the stability of the wellbore. In the prior art, generally, when abnormal fluctuations occur in the operation parameters of the mud pump (such as displacement, pressure), the parameters of the PID controller are adjusted through manual experience or the stroke frequency is manually corrected. The above methods rely on the subjective judgment of the operator and have a response lag, often resulting in the working state of the mud pump not being adjusted to the optimal in time, affecting the efficiency and safety of the drilling operation.

[0024] The solution provided by this application collects the operation data of the mud pump, determines the fluid pulsation characteristics, mechanical wear characteristics, and pump efficiency loss characteristics of the mud pump according to a preset multi-physical field coupling model, thereby constructing a multi-modal feature vector. Then, an initial control instruction is generated based on a preset fuzzy RBF neural network model. Finally, the initial control instruction is verified and corrected based on the digital twin model to generate a final control instruction, so as to adjust the working state of the mud pump according to the final control instruction, realizing real-time monitoring and intelligent control of the operation state of the mud pump, not relying on manual experience, improving the timeliness and accuracy of mud pump control, and further improving the efficiency and safety of drilling operations.

[0025] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of implementing the above functions, an automatic control device based on artificial intelligence, etc. Hereinafter, an automatic control device based on artificial intelligence will be taken as an example to illustrate this embodiment and the following embodiments.

[0026] Based on this, an embodiment of this application provides an automatic control method based on artificial intelligence, which is applied to a mud pump. Referring to Figure 1 , the automatic control method based on artificial intelligence includes steps S100 to S800, where: Step S100: Collect the operation data of the mud pump.

[0027] In this embodiment, the operation data includes pressure data, power spectral density data, vibration data, acoustic emission data, stroke frequency data, flow rate data, fluid data, and inlet and outlet pressure difference data. Among them, the pressure data can be collected in real time through a pressure sensor, and the pressure sensor can be installed at the inlet and outlet of the mud pump to monitor the pressure change during the operation of the mud pump; the power spectral density data can be obtained by performing spectral analysis on the vibration signal, and the vibration signal can be collected through an acceleration sensor, and the acceleration sensor can be installed on the housing or key components of the mud pump to monitor the vibration condition during the operation of the mud pump; the acoustic emission data can be collected through an acoustic emission sensor, and the acoustic emission sensor can be installed on the key components of the mud pump to monitor the sound signal during the operation of the mud pump; the stroke frequency data can be collected through a stroke frequency sensor, and the stroke frequency sensor can be installed on the drive device of the mud pump to monitor the stroke frequency of the mud pump; the flow rate data can be collected through a flow rate sensor, and the flow rate sensor can be installed on the inlet and outlet pipelines of the mud pump to monitor the flow rate of the mud pump; the fluid data can be collected through a fluid parameter sensor, and the fluid parameter sensor can be installed at the inlet and outlet of the mud pump to monitor fluid parameters such as the density and viscosity of the mud; the inlet and outlet pressure difference data can be collected through a pressure difference sensor, and the pressure difference sensor can be installed on the inlet and outlet pipelines of the mud pump to monitor the pressure difference between the inlet and outlet of the mud pump. By collecting the operation data of the mud pump, it is convenient for subsequent data processing and analysis of the operation data, thereby providing a data basis for realizing the automatic control of the mud pump.

[0028] Step S200: Perform dynamic normalization processing on the operation data to obtain corresponding preprocessed data.

[0029] In this embodiment, since the operation data of the mud pump may be affected by various factors such as environmental temperature and mud properties, the data has fluctuations and uncertainties. Therefore, after obtaining the operation data, dynamic normalization processing is required to eliminate the influence of these factors and improve the accuracy and reliability of the data.

[0030] In a feasible implementation manner, the step S200 includes: Determine a dynamic window according to a preset rule.

[0031] In this embodiment, the preset rule can be set according to the actual operation situation and data characteristics of the mud pump. For example, the size and update frequency of the dynamic window can be determined according to factors such as the fluctuation range and change trend of the data. The specific rule is not limited here. In this embodiment, by setting a reasonable dynamic window, the timeliness and accuracy of the data can be ensured.

[0032] Calculate the mean and standard deviation of the operation data within the dynamic window.

[0033] In this embodiment, after determining the dynamic window, calculate the mean and standard deviation of the operation data within the dynamic window, and statistical analysis can be performed on the data within the window to obtain the overall distribution of the data. The mean reflects the average level of the data, while the standard deviation reflects the degree of dispersion of the data.

[0034] Substitute the mean and standard deviation into the following formula to standardize the operation data: ; where, represents a single sampling value of the operation data; represents the mean of the operation data within the dynamic window; represents the standard deviation of the operation data within the dynamic window; represents a minimum constant.

[0035] In this embodiment, the minimum constant can be a positive number close to zero, such as , etc., which is used to avoid the situation where the denominator is zero and ensure the stability of the standardization formula. The specific value of this minimum constant can be set according to the actual situation and is not limited here. By substituting the mean and standard deviation into the above formula, the data can be converted into a dimensionless form, thereby eliminating the dimensional differences between different data, enabling operation data from different sources and with different dimensions to be compared and analyzed on the same scale, and providing convenience for subsequent data processing and analysis.

[0036] In a feasible implementation manner, step S200 further includes denoising the operation data.

[0037] In this embodiment, the denoising process can be implemented by various methods such as filtering methods and wavelet transform methods. Through the denoising process, the noise and interference in the operation data can be effectively removed, making the data smoother and more stable, thereby further improving the accuracy and reliability of the data and providing a more accurate data basis for subsequent feature extraction.

[0038] Step S300, obtain the fluid pulsation characteristics and mechanical wear characteristics of the mud pump according to the preprocessed data.

[0039] In this embodiment, the fluid pulsation characteristics reflect the fluctuation of the fluid in the mud pump during operation, and are related to factors such as the flow state of the mud, the internal structure of the pump, and the operating parameters. By analyzing the fluid pulsation characteristics, the stability and efficiency of the mud pump during mud transportation can be understood, so as to judge whether there are potential faults or performance degradation. The mechanical wear characteristics reflect the wear degree of the mechanical components (such as bearings, gears, etc.) of the mud pump during long-term operation. Mechanical wear not only affects the performance of the mud pump, but may also cause serious faults and even lead to equipment scrapping.

[0040] In a feasible implementation manner, the fluid pulsation characteristics include the pressure pulsation intensity and the main frequency energy ratio. Among them, the pressure pulsation intensity is calculated by the following formula: ; Where, represents the pressure pulsation intensity, represents the number of sampling points in the dynamic window, where, , represents the dynamic window length, represents the sampling rate; represents the i-th sampling point of the preprocessed pressure data; represents the mean value of the preprocessed pressure data in the dynamic window.

[0041] The main frequency energy ratio is calculated by the following formula: ; Where, represents the main frequency in the power spectral density, represents the power spectral density function; represents the highest effective frequency; represents the energy integral in the frequency interval ;

[0042] In this embodiment, by calculating the pressure pulsation intensity and the main frequency energy ratio through the above formulas, the fluid pulsation situation of the mud pump can be quantitatively analyzed to reflect the fluid pulsation characteristics of the mud pump, and further provide parameters for the construction of subsequent multi-modal feature vectors.

[0043] In a feasible implementation manner, the mechanical wear characteristics include acoustic emission energy accumulation and vibration severity; The acoustic emission energy accumulation is calculated by the following formula: ; Where, represents the accumulated energy of the preprocessed acoustic emission data in the dynamic window; represents the preprocessed acoustic emission signal at the The amplitude of each sampling point; Indicates the sampling time interval; Indicates the number of sampling points within the dynamic window length, where , Indicates the dynamic window length, Indicates the sampling rate; The vibration intensity is calculated by the following formula: ; Where Indicates the vibration intensity; Indicates the vibration data after envelope change at the i-th sampling point; Indicates the number of sampling points within the dynamic window length, where , Indicates the dynamic window length, Indicates the sampling rate.

[0044] In this embodiment, by calculating the acoustic emission energy accumulation and vibration intensity, the mechanical wear condition of the mud pump can be quantitatively analyzed to reflect the mechanical wear characteristics of the mud pump, and further provide parameters for the construction of subsequent multi-modal feature vectors.

[0045] Step S400, according to the preprocessed data and a preset multi-physical field coupling model, obtain the pump efficiency loss characteristics of the mud pump.

[0046] In this embodiment, the pump efficiency loss characteristics refer to the situation where the pump efficiency of the mud pump decreases due to various factors during the actual working process. The pump efficiency loss not only affects the conveying capacity of the mud pump, but also may increase energy consumption and mechanical wear, reducing the overall performance and service life of the equipment. In order to accurately obtain the pump efficiency loss characteristics of the mud pump, a preset multi-physical field coupling model is introduced in this embodiment. This model comprehensively considers the interaction and influence among multiple physical fields such as fluid dynamics, solid mechanics, and thermodynamics involved in the working process of the mud pump, and can more comprehensively and accurately reflect the actual working state of the mud pump. In the specific implementation process, the preprocessed operation data can be input into the multi-physical field coupling model, and through model calculation and simulation analysis, the pump efficiency loss characteristic parameters of the mud pump can be obtained, so as to quantitatively evaluate the pump efficiency loss situation of the mud pump.

[0047] In a feasible implementation manner, the pump efficiency loss characteristics include volumetric efficiency deviation and leakage equivalent area. Refer to Figure 2 , step S400 includes steps S410 to S430, where: Step S410, based on the preset multi-physical field coupling model, obtain the theoretical displacement of the mud pump.

[0048] In this embodiment, the theoretical displacement refers to the displacement that the mud pump should have under ideal conditions, and its magnitude is related to factors such as the design parameters of the mud pump, working conditions, and medium characteristics. By obtaining the theoretical displacement, a benchmark value can be provided for subsequent evaluation of pump efficiency loss, so as to more accurately reflect the actual working state and performance of the mud pump.

[0049] In a feasible implementation manner, the theoretical displacement is calculated by the following formula: ; Where represents the theoretical displacement, represents the stroke frequency; represents the piston cross-sectional area corresponding to the preset multi-physical field coupling model; represents the effective stroke length corresponding to the preset multi-physical field coupling model; represents the volumetric efficiency corresponding to the preset multi-physical field coupling model; represents the mechanical efficiency corresponding to the preset multi-physical field coupling model; represents the compressibility correction factor corresponding to the preset multi-physical field coupling model.

[0050] In this embodiment, due to the double influence of the thermal expansion (thermal field) and elastic deformation (structural field) of the piston cross-sectional area, the piston cross-sectional area corresponding to the preset multi-physical field coupling model can specifically be the expansion deformation caused by the temperature rise calculated through thermo-structural coupling analysis, so as to obtain the actual piston cross-sectional area. Since the effective stroke of the piston is affected by the coupling action of the hydraulic load (flow field) and material creep (structural field), the effective stroke length corresponding to the preset multi-physical field coupling model can specifically be the actual effective stroke length calculated through fluid-structure coupling analysis. At the same time, the volumetric efficiency, mechanical efficiency, and compressibility correction factor corresponding to the preset multi-physical field coupling model can also be calculated and corrected by comprehensively considering the interactions and influences between multiple physical fields such as fluid dynamics, solid mechanics, and thermodynamics to ensure the accuracy and reliability of the theoretical displacement. By substituting parameters such as the stroke frequency, piston cross-sectional area, effective stroke length, volumetric efficiency, mechanical efficiency, and compressibility correction factor into the above formula, the theoretical displacement of the mud pump under ideal conditions can be calculated.

[0051] Step S420, obtaining the actual displacement of the mud pump based on the preprocessed data.

[0052] In this embodiment, the actual displacement is calculated based on the following formula: ; Where, represents the actual displacement; represents the dynamic window length; represents the instantaneous flow rate; Represents the integration variable.

[0053] In this embodiment, the instantaneous flow rate can be monitored in real time by a flow sensor, and its magnitude reflects the conveying capacity of the mud pump at a certain moment. By integrating the instantaneous flow rate within the dynamic window length, the actual displacement of the mud pump within a period of time can be obtained, thereby reflecting the actual working state and performance of the mud pump. By comparing and analyzing the actual displacement with the theoretical displacement, the pump efficiency loss of the mud pump can be evaluated, providing a basis for subsequent equipment maintenance and fault prediction.

[0054] Step S430, calculate the volumetric efficiency deviation and the leakage equivalent area based on the theoretical displacement, actual displacement, and preprocessed data.

[0055] In this embodiment, the volumetric efficiency deviation is calculated by the following formula: ; Where Represents the actual displacement; Represents the theoretical displacement; The leakage equivalent area is calculated by the following formula: ; Where Represents the pressure difference between the inlet and outlet of the pump; Represents the fluid density; Represents the flow coefficient.

[0056] In this embodiment, the fluid density is the density of the mud conveyed by the mud pump, and its magnitude is related to factors such as the composition, temperature, and pressure of the mud. The flow coefficient is a constant related to the internal structure of the mud pump and the fluid flow state, and it can be determined through experimental measurement or empirical formula. By substituting parameters such as the actual displacement, theoretical displacement, pressure difference between the inlet and outlet of the pump, fluid density, and flow coefficient into the above formulas, the volumetric efficiency deviation and the leakage equivalent area can be calculated respectively. The larger the volumetric efficiency deviation, the more serious the volumetric loss of the mud pump and the lower the pump efficiency. The leakage equivalent area quantifies the leakage situation of the mud pump during operation due to reasons such as poor sealing. The larger the leakage equivalent area, the more serious the leakage of the mud pump, which will also lead to a decrease in pump efficiency. By calculating the volumetric efficiency deviation and the leakage equivalent area, the pump efficiency loss of the mud pump can be reflected, providing a data basis for the construction of subsequent multi-modal feature vectors.

[0057] Step S500, construct a multi-modal feature vector based on the fluid pulsation characteristics, mechanical wear characteristics, and pump efficiency loss characteristics of the mud pump, and input it into a pre-trained variational autoencoder for feature dimensionality reduction to obtain a target feature vector.

[0058] In this embodiment, the fluid pulsation characteristics include the pressure pulsation intensity and the proportion of the main frequency energy, which reflect the stability of the fluid flow and the energy distribution of the mud pump; the mechanical wear characteristics include the acoustic emission energy accumulation and the vibration severity, which reveal the wear degree and state of the mechanical components of the mud pump; the pump efficiency loss characteristics are quantified and evaluated by the volumetric efficiency deviation and the leakage equivalent area to describe the decrease in the pump efficiency of the mud pump. By integrating the above-obtained fluid pulsation characteristics, mechanical wear characteristics, and pump efficiency loss characteristics of the mud pump, a multi-modal feature vector is constructed to comprehensively describe the working state of the mud pump. When constructing this multi-modal feature vector, the method of feature splicing can be adopted, that is, each eigenvalue is arranged and combined in a certain order to form a high-dimensional feature vector. However, high-dimensional feature vectors often contain redundant information and have a high computational complexity, which is not conducive to subsequent fault prediction and diagnosis. Therefore, a pre-trained variational autoencoder is introduced in this embodiment for feature dimensionality reduction.

[0059] In this embodiment, the variational autoencoder is a deep learning model that realizes feature dimensionality reduction by learning the latent representation of the data. In this embodiment, the pre-trained variational autoencoder has learned the latent distribution law of the working state data of the mud pump. Therefore, by inputting the integrated multi-modal feature vector into the pre-trained variational autoencoder, a low-dimensional target feature vector can be obtained. This target feature vector not only retains the key information in the original feature vector but also removes the redundant information, reduces the computational complexity, and provides a more concise and effective data basis for subsequent fault prediction and diagnosis.

[0060] Step S600: Input the target feature vector into a preset fuzzy RBF neural network model to obtain a corresponding initial control instruction; the initial control instruction includes a displacement compensation value and a stroke frequency adjustment amount.

[0061] In this embodiment, the fuzzy RBF neural network model is an intelligent model that combines fuzzy logic and a radial basis function (RBF) neural network. This model uses fuzzy logic to handle uncertainty and ambiguity, and at the same time uses the powerful non-linear mapping ability of the RBF neural network to achieve accurate modeling and prediction of the working state data of the mud pump. Before inputting the target feature vector into the fuzzy RBF neural network model, the model can be trained and optimized first to obtain a preset fuzzy RBF neural network model to ensure its accuracy and reliability. During the training process, a large amount of working state data of the mud pump can be used as training samples. By continuously adjusting the parameters and structure of the model, the model can accurately map the mapping relationship between the target feature vector and the initial control instruction. After the training is completed, the target feature vector is input into the preset fuzzy RBF neural network model, and the model can calculate the corresponding initial control instruction according to the input feature vector. The initial control instruction includes a displacement compensation value and a stroke frequency adjustment amount. Through this initial control instruction, the working state of the mud pump can be adjusted and optimized to improve its working efficiency and performance. The displacement compensation value is used to adjust the displacement of the mud pump to compensate for the decrease in displacement caused by pump efficiency loss; the stroke frequency adjustment amount is used to adjust the stroke frequency of the mud pump to optimize its working state and performance. By adjusting the displacement compensation value and the stroke frequency adjustment amount, the energy consumption and mechanical wear of the mud pump can be effectively reduced, and its overall performance and service life can be improved.

[0062] In a feasible implementation manner, the method further includes online learning and updating of the fuzzy RBF neural network by adopting a dual-time scale update strategy, specifically including: Adjusting the network weights and membership function parameters based on the real-time error gradient to quickly respond to transient working conditions; and, Dynamically optimizing the network structure by sliding window statistics features to ensure long-term stability and suppress overfitting.

[0063] It can be understood that during the actual operation of the mud pump, the working conditions may change frequently, which requires the control model to be able to respond quickly and adjust the control instructions. By adjusting the network weights and membership function parameters based on the real-time error gradient, it is possible to achieve immediate tracking and rapid adjustment of the changes in the working state of the mud pump, thereby improving the real-time performance and robustness of the model. At the same time, by dynamically optimizing the network structure through sliding window statistical features, the occurrence of overfitting can be effectively suppressed on the basis of ensuring the long-term stability of the model. The sliding window statistical features can capture the temporal characteristics and change trends of the mud pump working state data, thus providing strong support for the dynamic optimization of the network structure. In the specific implementation process, the size and update frequency of the sliding window can be reasonably set according to the actual operation data and working condition change characteristics of the mud pump to ensure the effect of online learning and update. In this embodiment, the fuzzy RBF neural network is updated online using a dual-time scale update strategy, which can further improve the adaptability and accuracy of the model.

[0064] Step S700, pre-verify the initial control instruction based on the digital twin model to generate a safe operation domain including three-dimensional constraints of pressure - displacement - frequency.

[0065] It can be understood that in actual applications, because under different mud pumps and working conditions, the same control instruction may produce different effects, there may be certain risks in directly implementing the initial control instruction on the mud pump. To ensure the safety and effectiveness of the control instruction, a digital twin model is introduced in this embodiment to pre-verify the initial control instruction. The digital twin model is a virtual simulation technology based on a physical model, which can reflect the actual working state and performance of the mud pump in real time. By inputting the initial control instruction into the digital twin model, the working state and performance changes of the mud pump under the corresponding control instruction can be simulated, so as to evaluate the rationality and feasibility of the control instruction.

[0066] In the digital twin model, the interaction and influence between multiple physical fields such as fluid dynamics, solid mechanics, and thermodynamics of the mud pump can be comprehensively considered to establish a three-dimensional simulation model of the mud pump. At the same time, combined with real-time sensor data, the simulation model is dynamically updated and optimized to ensure the accuracy and reliability of the simulation results. In this embodiment, a safe operation domain including three-dimensional constraints of pressure - displacement - frequency can be constructed by analyzing the historical safety data of the mud pump. The safe operation domain refers to the reasonable value range and mutual relationship between various parameters of the mud pump in the normal working state, which provides an important guarantee for the safe operation of the mud pump.

[0067] In this embodiment, through the pre-verification of the digital twin model, it is possible to evaluate whether the initial control instruction is within the safe operation domain. If not, the control instruction needs to be adjusted and optimized to ensure it is within the safe operation domain. At the same time, the digital twin model can also provide the optimal control strategy of the mud pump under different working conditions, providing strong support for the intelligent control of the mud pump. After the pre-verification is completed, the optimized control instruction can be sent to the mud pump to achieve precise control of the mud pump.

[0068] In a feasible implementation manner, referring to Figure 3 , step S700 includes steps S710 to S730, where: Step S710, receive the initial control instruction and predict the dynamic responses of pressure, displacement, and frequency based on the digital twin model.

[0069] In this embodiment, After receiving the initial control instruction, the digital twin model starts to simulate the working state of the mud pump under the corresponding control instruction. Based on the three-dimensional simulation model of the mud pump and real-time sensor data, the digital twin model can predict the dynamic responses of pressure, displacement, and frequency of the mud pump after adjusting the displacement compensation value and the stroke frequency adjustment amount. This step is a key link in the pre-verification. Through simulation and prediction, the impact of the control instruction on the working state of the mud pump can be intuitively seen, providing data support for the subsequent optimization of the control strategy.

[0070] Step S720, extract the pressure peak value, displacement fluctuation threshold, and resonance frequency forbidden zone based on the historical safety data of the digital twin model.

[0071] In this embodiment, the historical safety data records the change conditions and safety boundaries of each parameter of the mud pump during past operations. By analyzing the historical safety data, key safety indicators such as the pressure peak value, displacement fluctuation threshold, and resonance frequency forbidden zone that may occur during the operation of the mud pump can be extracted. The pressure peak value reflects the pressure-bearing capacity of the mud pump when it bears extreme working conditions, the displacement fluctuation threshold quantifies the reasonable range of displacement changes of the mud pump, and the resonance frequency forbidden zone reveals the resonance phenomenon that the mud pump may occur at specific frequencies.

[0072] Step S730, fit the constraint relationship between the pressure peak value, displacement fluctuation threshold, and resonance frequency forbidden zone and the predicted dynamic response to generate a safe operation domain including three-dimensional constraints of pressure-displacement-frequency.

[0073] In this embodiment, through data fitting, the peak pressure, displacement fluctuation threshold, and resonance frequency forbidden zone are correlated with the dynamic responses of pressure, displacement, and frequency predicted by the digital twin model, so as to reveal the constraint relationships among various parameters of the mud pump under different control instructions, and then a safe operation domain including three-dimensional constraints of pressure-displacement-frequency is generated. In the safe operation domain, each point represents the safe operation state of the mud pump under specific working conditions, and different points constitute the boundaries and ranges of the safe operation of the mud pump. By comparing the working state of the mud pump corresponding to the initial control instruction with the safe operation domain, it can be determined whether the initial control instruction meets the safety requirements. If the working state of the mud pump caused by the initial control instruction exceeds the boundary of the safe operation domain, the control instruction needs to be adjusted to ensure that the mud pump operates under safe and stable working conditions, thereby ensuring the working efficiency and performance of the mud pump.

[0074] Step S800, dynamically correct the displacement compensation value and the stroke frequency adjustment amount according to the safe operation domain to generate a final control instruction, so as to adjust the working state of the mud pump according to the final control instruction.

[0075] In this embodiment, after determining the safe operation domain including three-dimensional constraints of pressure-displacement-frequency, it is necessary to further dynamically correct the displacement compensation value and the stroke frequency adjustment amount in the initial control instruction to ensure that the working state of the mud pump after adjustment can be strictly maintained within the safe and stable boundaries.

[0076] In the specific implementation process, the working state of the mud pump corresponding to the initial control instruction can be compared with the safe operation domain. If the working pressure, displacement, or frequency of the mud pump caused by the initial control instruction exceeds the boundary of the safe operation domain, the displacement compensation value and the stroke frequency adjustment amount need to be adjusted accordingly according to the constraint relationships of the safe operation domain. The adjustment principle is to ensure that the working pressure, displacement, and frequency of the mud pump can return to the safe operation domain, and at the same time, the working efficiency and performance of the mud pump are maintained as much as possible. In this way, a final control instruction that more conforms to the actual operation requirements and safety requirements of the mud pump can be generated. The final control instruction not only considers the working efficiency and performance of the mud pump, but also fully considers the safety and stability of the mud pump, providing a strong guarantee for the intelligent control of the mud pump. After generating the final control instruction, it can be sent to the mud pump to achieve precise, safe, and stable control of the mud pump.

[0077] In this embodiment, by collecting the operation data of the mud pump and performing dynamic normalization processing on the operation data to obtain corresponding preprocessed data; then obtaining the fluid pulsation characteristics and mechanical wear characteristics of the mud pump according to the preprocessed data; and obtaining the pump efficiency loss characteristics of the mud pump according to the preprocessed data and a preset multi-physical field coupling model; then constructing a multi-modal feature vector based on the fluid pulsation characteristics, mechanical wear characteristics and pump efficiency loss characteristics of the mud pump, and inputting the multi-modal feature vector into a pre-trained variational autoencoder for feature dimensionality reduction to obtain a target feature vector; and inputting the target feature vector into a preset fuzzy RBF neural network model to obtain corresponding initial control instructions; the initial control instructions include a displacement compensation value and a stroke frequency adjustment amount; then pre-verifying the initial control instructions based on the digital twin model to generate a safe operation domain including three-dimensional constraints of pressure-displacement-frequency; finally, dynamically correcting the displacement compensation value and the stroke frequency adjustment amount according to the safe operation domain to generate a final control instruction, and adjusting the working state of the mud pump according to the final control instruction. That is, this solution realizes real-time monitoring and intelligent control of the operation state of the mud pump by collecting the operation data of the mud pump, determining the fluid pulsation characteristics, mechanical wear characteristics and pump efficiency loss characteristics of the mud pump according to a preset multi-physical field coupling model, thereby constructing a multi-modal feature vector, then generating initial control instructions based on a preset fuzzy RBF neural network model, and finally verifying and correcting the initial control instructions based on the digital twin model to generate a final control instruction, and adjusting the working state of the mud pump according to the final control instruction, without relying on manual experience, improving the timeliness and accuracy of mud pump control, and further improving the efficiency and safety of drilling operations.

[0078] The above are only some embodiments of the present application, and thus do not limit the patent scope of the present application. Any equivalent structural transformation made under the technical concept of the present application by using the content of the specification and drawings of the present application, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. An automatic control method based on artificial intelligence, applied to a mud pump, characterized in that The described method includes: Collecting the operation data of the mud pump; Performing dynamic normalization processing on the operation data to obtain corresponding preprocessed data; Obtaining the fluid pulsation characteristics and mechanical wear characteristics of the mud pump according to the preprocessed data; Obtaining the pump efficiency loss characteristics of the mud pump according to the preprocessed data and a preset multi-physical field coupling model; Constructing a multi-modal feature vector based on the fluid pulsation characteristics, mechanical wear characteristics, and pump efficiency loss characteristics of the mud pump, and inputting it into a pre-trained variational autoencoder for feature dimensionality reduction to obtain a target feature vector; Inputting the target feature vector into a preset fuzzy RBF neural network model to obtain corresponding initial control instructions; the initial control instructions include a displacement compensation value and a stroke frequency adjustment amount; Performing pre-verification on the initial control instructions based on a digital twin model to generate a safe operation domain including three-dimensional constraints of pressure-displacement-frequency; Dynamically correcting the displacement compensation value and the stroke frequency adjustment amount according to the safe operation domain to generate a final control instruction, so as to adjust the working state of the mud pump according to the final control instruction.

2. The automatic control method according to claim 1, characterized in that, The step of performing dynamic normalization processing on the operation data includes: Determining a dynamic window according to a preset rule; Calculating the mean and standard deviation of the operation data within the dynamic window; Substituting the mean and standard deviation into the following formula to normalize the operation data: ； in, Represents a single sample value of running data; Represents the mean of the running data in the dynamic window; Represents the standard deviation of the running data within the dynamic window; Represents a very small constant.

3. The automatic control method according to claim 2, characterized in that, The operation data includes pressure data and power spectral density data; The fluid pulsation characteristics include pressure pulsation intensity and main frequency energy ratio; The pressure pulsation intensity is calculated by the following formula: ； Among them, represents the pressure pulsation intensity, represents the number of sampling points within the dynamic window, where , represents the dynamic window length, represents the sampling rate; represents the i-th sampling point of the preprocessed pressure data; represents the mean value of the preprocessed pressure data within the dynamic window. The main frequency energy ratio is calculated by the following formula: ； Among them, represents the main frequency in the power spectral density, represents the power spectral density function; represents the highest effective frequency; represents the energy integral in the frequency interval within.

4. The automatic control method according to claim 2, wherein The operation data includes vibration data and acoustic emission data; The mechanical wear characteristics include acoustic emission energy accumulation and vibration intensity; The acoustic emission energy accumulation is calculated by the following formula: ； Among them, represents the accumulated energy of the preprocessed acoustic emission data within the dynamic window; represents the amplitude of the preprocessed acoustic emission signal at the th sampling point; represents the sampling time interval; represents the number of sampling points within the dynamic window length, where , represents the dynamic window length, represents the sampling rate; The vibration intensity is calculated by the following formula: ； Among them, represents the vibration intensity; represents the vibration data after envelope change at the i-th sampling point; represents the number of sampling points within the dynamic window length, where , represents the dynamic window length, represents the sampling rate.

5. The automatic control method according to claim 2, wherein The pump efficiency loss characteristics include volumetric efficiency deviation and leakage equivalent area; The step of obtaining the pump efficiency loss characteristics of the mud pump according to the preprocessed data and a preset multi-physical field coupling model includes: Obtaining the theoretical displacement of the mud pump based on a preset multi-physical field coupling model; Obtaining the actual displacement of the mud pump based on the preprocessed data; Calculating the volumetric efficiency deviation and the leakage equivalent area based on the theoretical displacement, actual displacement, and preprocessed data.

6. The automatic control method according to claim 5, characterized in that The operation data includes stroke frequency data, flow rate data, fluid data, and inlet and outlet pressure difference data; The theoretical displacement is calculated by the following formula: ； wherein represents the theoretical displacement; represents the stroke frequency; represents the piston cross-sectional area corresponding to the preset multi-physical field coupling model; represents the effective stroke length corresponding to the preset multi-physical field coupling model; represents the volumetric efficiency corresponding to the preset multi-physical field coupling model; represents the mechanical efficiency corresponding to the preset multi-physical field coupling model; represents the compressibility correction factor corresponding to the preset multi-physical field coupling model; The actual displacement is calculated based on the following formula: ； Among them, represents the actual displacement; represents the dynamic window length; represents the instantaneous flow rate; represents the integral variable.

7. The automatic control method according to claim 6, wherein The volumetric efficiency deviation is calculated by the following formula: ； Among them, represents the actual displacement; represents the theoretical displacement; The leakage equivalent area is calculated by the following formula: ； Among them, represents the pressure difference between the inlet and outlet of the pump; represents the fluid density; represents the flow coefficient.

8. The automatic control method according to claim 1, wherein The method further includes: Adopting a dual-time-scale update strategy to perform online learning and update on the fuzzy RBF neural network, specifically including: Adjusting the network weights and membership function parameters based on the real-time error gradient to quickly respond to transient working conditions; and, Dynamically optimizing the network structure by statistical features of a sliding window to ensure long-term stability and suppress overfitting.

9. The automatic control method according to claim 1, characterized in that, The steps of pre-verifying the initial control instruction based on the digital twin model to generate a safe operation domain including three-dimensional constraints of pressure - displacement - frequency are as follows: Receiving the initial control instruction and predicting the dynamic responses of pressure, displacement and frequency based on the digital twin model; Extracting the pressure peak value, displacement fluctuation threshold and resonance frequency forbidden zone based on the historical safety data of the digital twin model; Fitting the constraint relationships between the pressure peak value, displacement fluctuation threshold and resonance frequency forbidden zone and the predicted dynamic responses to generate a safe operation domain including three-dimensional constraints of pressure - displacement - frequency.

Images