An artificial intelligence-based oil well production prediction method

Abstract

The present invention relates to an oil well production prediction method based on artificial intelligence. The prediction method described in the present invention includes: real-time collecting data of downhole fluid ratio, pressure, pipe diameter and temperature through downhole and surface multi-modal sensors; calculating wellbore flow characteristic parameters based on multiphase fluid mechanics formulas; using a multiphase flow simulation algorithm to couple multi-source data to generate a full-chain fluid distribution result; optimizing the full-chain fluid distribution result to generate a production distribution prediction value; fusing multi-source data through a physics-informed neural network to generate a full-chain fluid parameter simulation result; and finally using the ensemble Kalman filter algorithm to achieve real-time data assimilation and output a dynamic simulation result. The oil production prediction method provided by the present invention solves problems such as the separation of downhole and surface data, the lag of the static boundary of the model, and the over-simplification of physical equations in traditional methods, improves the prediction accuracy and real-time performance in complex oil well scenarios, and provides intelligent decision-making support for oilfield development.

Description

Technical Field

[0001] The present invention relates to the technical field of oil resource exploration, and particularly relates to an oil well production prediction method based on artificial intelligence. Background Art

[0002] The prediction of oil well production is a core link in oilfield development and production management. The key lies in the accurate modeling and dynamic analysis of the entire chain flow process from downhole to the ground. Traditional oil well production prediction methods mainly rely on physical models and empirical formulas. By collecting limited parameters such as downhole pressure and temperature, and combining with multiphase flow theory to simulate wellbore flow, the production distribution is then deduced. However, the oilfield production system has high complexity, involving downhole fluid dynamics, surface pipeline network transmission, and multi-parameter coupling effects. Traditional methods often have difficulty in taking into account the integration of multi-source heterogeneous data and real-time requirements.

[0003] In the prior art, oil well production prediction models usually face the following problems: First, the fragmentation of downhole and surface data leads to difficult parameter coupling. For example, it is difficult to synchronously calculate the wellbore flow characteristics and pipeline network pressure distribution; Second, traditional numerical simulation methods rely on static boundary conditions and cannot respond to changes in production data in real time, resulting in prediction results lagging behind actual working conditions; Third, the accuracy of multiphase flow models is limited by the simplified assumptions of physical equations and it is difficult to accurately capture the dynamic interaction effects of oil, gas, and water three-phase fluids. These problems make the prediction error of traditional methods relatively large in complex oil well scenarios and it is difficult to support real-time optimization decisions. Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide an oil well production prediction method based on artificial intelligence that can achieve dynamic coupling and real-time synchronization of full-chain parameters.

[0005] The purpose of the present invention is achieved by the following solutions:

[0006] In the first aspect, the present invention provides an oil well production prediction method based on artificial intelligence, including the following steps:

[0007] S1: Based on downhole multimodal sensors and surface pipeline network multimodal sensors, collect downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data in real time. The downhole fluid ratio data is used to indicate the three-phase fluid ratio of the oil phase, gas phase, and water phase;

[0008] S2: Calculate wellbore flow characteristic parameters based on multiphase fluid mechanics formulas for the downhole fluid ratio data and downhole pressure data, and generate wellbore flow characteristic parameters. The wellbore flow characteristic parameters are used to indicate the velocity distribution and density distribution of the fluid in the wellbore;

[0009] S3: Based on the multiphase flow simulation algorithm, perform pressure and temperature coupling calculations on the pipeline network temperature data, pipe diameter parameters, and wellbore flow characteristic parameters to generate the full-chain fluid distribution results;

[0010] S4: Based on the finite difference method and the backpropagation algorithm, perform dynamic boundary optimization processing on the full-chain fluid distribution results to generate the predicted value of the optimized production distribution;

[0011] S5: Based on the physics-informed neural network, perform multi-parameter fusion processing on the downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data to generate the full-chain fluid parameter simulation results;

[0012] S6: Based on the ensemble Kalman filter algorithm, perform data assimilation processing on the predicted value of the production distribution and the full-chain fluid parameter simulation results to generate the dynamic simulation results synchronized with the actual production, and the dynamic simulation results are used to predict the changing trend of the oil well production.

[0013] In one embodiment, S2 of the oil well production prediction method based on artificial intelligence provided by the present invention specifically includes the following steps:

[0014] S21: Based on the holdup equation and the velocity gradient equation, calculate the downhole pressure data to obtain the initial velocity distribution data of the fluid in the wellbore;

[0015] S22: Based on the sliding window technique, perform sliding window averaging processing on the initial velocity distribution data to generate the smoothed fluid velocity distribution data;

[0016] S23: Based on the physics-informed neural network, perform fluid mechanics constraint processing on the fluid velocity distribution data and the downhole fluid ratio data to generate the fluid density distribution data in the wellbore;

[0017] S24: Integrate and process the fluid velocity distribution data and the fluid density distribution data to generate the wellbore flow characteristic parameters.

[0018] In one embodiment, S3 of the oil well production prediction method based on artificial intelligence provided by the present invention specifically includes the following steps:

[0019] S31: Based on the machine learning algorithm, train the multiphase flow model with the historical production data of the oil well and the historical pipeline network parameters to generate the multiphase flow simulation kernel;

[0020] S32: Based on the multiphase flow simulation kernel, perform pipe section pressure drop calculations on the pipeline network temperature data, pipe diameter parameters, and wellbore flow characteristic parameters to generate the pipeline network fluid pressure distribution;

[0021] S33: Identify whether the wellbore velocity distribution of the wellbore flow characteristic parameters exceeds a preset flow velocity safety threshold. If it exceeds, optimize the wellhead flow rate constraint for the fluid pressure distribution in the pipeline network based on the adaptive particle swarm algorithm to generate updated boundary conditions;

[0022] S34: Based on the numerical iteration method, perform pressure-flow synchronous coupling on the updated boundary conditions and the wellbore flow characteristic parameters to generate the full-chain fluid distribution result.

[0023] In one embodiment, step S4 of the oil well production prediction method based on artificial intelligence provided by the present invention specifically includes the following steps:

[0024] S41: Divide the full-chain fluid distribution result into multiple spatio-temporal grid units, and discretize the mass flow rate distribution of each spatio-temporal grid unit through the finite difference method to generate discretized flow rate distribution data;

[0025] S42: Based on the long short-term memory neural network, dynamically adjust the allocation ratio weights of the flow rate distribution data and the downhole fluid ratio data to generate an initial predicted value of the production distribution;

[0026] S43: Identify whether the deviation between the initial predicted value and the downhole fluid ratio data exceeds a preset deviation threshold. If it exceeds, perform iterative update of the allocation weights on the initial predicted value based on the backpropagation algorithm to generate an optimized predicted value of the production distribution.

[0027] In one embodiment, step S42 of the oil well production prediction method based on artificial intelligence provided by the present invention specifically includes the following steps:

[0028] S421: Normalize the downhole fluid ratio data to generate standardized fluid ratio data;

[0029] S422: Input the standardized fluid ratio data into the long short-term memory neural network, and extract the time series features of the wellbore and pipeline network allocation through the gating mechanism to generate initial allocation ratio weights;

[0030] S423: Optimize the wellbore-pipeline network allocation ratio of the initial allocation ratio weights and the flow rate distribution data through the dynamic programming algorithm to generate an initial predicted value of the production distribution.

[0031] In one embodiment, step S6 of the oil well production prediction method based on artificial intelligence provided by the present invention specifically includes the following steps:

[0032] S61: Based on the Fourier transform algorithm, extract the main frequency components in the pressure and temperature change trends of the production distribution predicted value to generate frequency domain feature data;

[0033] S62: Supplement the missing data points of the frequency-domain feature data through cubic spline interpolation to generate complete digital twin model data;

[0034] S63: Use the Ensemble Kalman Filter algorithm to perform data fusion on the full-chain fluid parameter simulation results and digital twin model data to generate dynamic simulation results synchronized with actual production.

[0035] In a second aspect, the present invention provides an oil well production prediction system based on artificial intelligence, including

[0036] A data acquisition module for real-time collecting downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data based on downhole multi-modal sensors and above-ground pipeline network multi-modal sensors. The downhole fluid ratio data is used to indicate the three-phase fluid ratio of the oil phase, gas phase, and water phase;

[0037] A characteristic parameter calculation module for calculating wellbore flow characteristic parameters based on downhole fluid ratio data and downhole pressure data using multiphase fluid mechanics formulas to generate wellbore flow characteristic parameters, which are used to indicate the velocity distribution and density distribution of the fluid in the wellbore;

[0038] A pressure-temperature coupling module for performing pressure-temperature coupling calculations on pipeline network temperature data, pipe diameter parameters, and wellbore flow characteristic parameters based on multiphase flow simulation algorithms to generate full-chain fluid distribution results;

[0039] A dynamic optimization module for performing dynamic boundary optimization processing on the full-chain fluid distribution results based on the finite difference method and the backpropagation algorithm to generate an optimized production distribution prediction value;

[0040] A multi-parameter fusion module for performing multi-parameter fusion processing on downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data based on a physics-informed neural network to generate full-chain fluid parameter simulation results;

[0041] A synchronous simulation generation module for performing data assimilation processing on the production distribution prediction value and the full-chain fluid parameter simulation results based on the Ensemble Kalman Filter algorithm to generate dynamic simulation results synchronized with actual production, which are used to predict the changing trend of oil well production.

[0042] In a third aspect, the present application provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements any one of the above-mentioned artificial intelligence-based oil well production prediction methods.

[0043] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, any of the above-described artificial intelligence-based oil well production prediction methods is implemented.

[0044] In summary, an artificial intelligence-based oil well production prediction method provided by the present invention can comprehensively, accurately and real-timely simulate and predict an oil well production system through technical methods such as multi-modal sensor data acquisition, multiphase fluid mechanics calculation, multiphase flow simulation, dynamic boundary optimization, multi-parameter fusion and data assimilation, thereby improving the accuracy of oil well production prediction, providing a reliable basis for production decision-making. At the same time, this method can realize dynamic monitoring and optimization adjustment of the production system, contribute to improving production efficiency and reducing production costs, and integrates physical models and data-driven methods, thereby ensuring the accuracy and reliability of simulation results and providing strong support for the intelligent production management of oil wells.

[0045] For better understanding and implementation, the present invention will be described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a schematic flowchart of an artificial intelligence-based oil well production prediction method provided by an embodiment of the present application;

[0047] Figure 2 is a schematic flowchart of generating an optimized production distribution prediction value provided by an embodiment of the present application;

[0048] Figure 3 is a schematic structural diagram of an artificial intelligence-based oil well production prediction system provided by another embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] To facilitate understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Preferred embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, these embodiments are provided to make the understanding of the disclosure of the invention more thorough and comprehensive.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs. The terms used in the description of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the related listed items.

[0051] In one embodiment, as Figure 1As shown, a method for predicting oil well production based on artificial intelligence is provided. In this embodiment, an example is given where this method is applied to a terminal. It can be understood that this method can also be applied to a server, or to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0052] S1: Based on downhole multi-modal sensors and above-ground pipeline network multi-modal sensors, collect downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data in real time. The downhole fluid ratio data is used to indicate the three-phase fluid ratio of the oil phase, gas phase, and water phase.

[0053] Specifically, various types of sensors can be installed at different depths in the oil well, including but not limited to pressure sensors, temperature sensors, fluid composition sensors, etc., for real-time monitoring of data such as the pressure and temperature of the downhole fluid and the ratio of the three-phase fluid of oil, gas, and water. These sensors are connected to the ground data acquisition system through wired or wireless communication methods, and transmit the collected data to the ground in real time.

[0054] At the same time, in the above-ground oil and gas pipeline network, various sensors are also installed, such as pressure sensors, temperature sensors, flow sensors, etc. The above-ground pipeline network multi-modal sensors mainly collect pipeline network temperature data and pipe diameter parameters. The pipeline network temperature data can reflect the temperature change of the fluid in the pipeline network, which is of great significance for understanding the thermodynamic properties and flow characteristics of the fluid. The pipe diameter parameter is the structural parameter of the pipeline network, which directly affects the flow resistance and flow distribution of the fluid in the pipeline network. Through high-precision measurement techniques, these sensors can obtain the above data in real time and accurately, and transmit them to the data processing system, providing reliable data support for subsequent calculations and analyses.

[0055] S2: Calculate the wellbore flow characteristic parameters based on the downhole fluid ratio data and the downhole pressure data using multi-phase fluid mechanics formulas, and generate wellbore flow characteristic parameters, which are used to indicate the velocity distribution and density distribution of the fluid in the wellbore.

[0056] Specifically, the multiphase hydrodynamics formula is a mathematical expression that describes the flow law of multiphase fluids in pipelines or wellbores, comprehensively considering factors such as the interaction between phases, flow velocity, density, and pressure of the fluids. Specifically, by substituting the collected proportion data of the oil phase, gas phase, and water phase, as well as the downhole pressure data, into these formulas, the system can calculate key parameters such as the velocity distribution and density distribution of the fluids in the wellbore. Among them, the velocity distribution parameter can reflect the change in the flow velocity of different-phase fluids in the wellbore, which is directly significant for determining the flow state and flow rate of the fluids. The density distribution parameter reflects the density change of the fluids at different positions and different phase states, which is crucial for accurately calculating the volume flow rate and mass flow rate of the fluids. These wellbore flow characteristic parameters provide the necessary basic data for the subsequent coupled calculation of pipeline network pressure and temperature, ensuring the accurate modeling and analysis of the entire flow system.

[0057] S3: Based on the multiphase flow simulation algorithm, perform the coupled calculation of pressure and temperature on the pipeline network temperature data, pipe diameter parameters, and wellbore flow characteristic parameters to generate the full-chain fluid distribution results.

[0058] Specifically, a multiphase flow simulation algorithm based on the finite element method or the multiphase flow network method can be selected. Taking the finite element method as an example, the pipeline network system is divided into multiple tiny finite element units, and the fluid flow in each unit is discretized to establish the corresponding control equations. Specifically, the wellbore flow characteristic parameters (fluid velocity, density distribution) obtained by the system are used as the boundary conditions at the pipeline network inlet. Combining the pipeline network temperature data and pipe diameter parameters, the fluid flow in the pipeline network is simulated and calculated. During the calculation process, the influence of the variation of fluid physical property parameters (such as viscosity, specific heat capacity, etc.) with temperature is considered and corrected through looking up tables or empirical formulas.

[0059] For each pipeline network unit, calculate the pressure drop and temperature change of the fluid simultaneously. The pressure drop calculation considers factors such as the fluid flow velocity, pipe diameter, and fluid viscosity, and is calculated using formulas such as the Darcy - Weisbach formula; the temperature change calculation considers factors such as the heat exchange between the fluid and the pipeline, the adiabatic compression or expansion of the fluid, and is solved using the energy conservation equation. Integrate the calculation results of all units to obtain the pressure and temperature distribution of the entire pipeline network system, thereby generating the full-chain fluid distribution results, including information such as the flow state, pressure, and temperature of the fluid in the wellbore and pipeline network.

[0060] S4: Based on the finite difference method and the backpropagation algorithm, perform dynamic boundary optimization processing on the full-chain fluid distribution results to generate the predicted value of the optimized production distribution.

[0061] Specifically, the finite difference method is used to discretize the full-chain fluid distribution results, converting continuous fluid distribution parameters (such as pressure, temperature, fluid velocity, etc.) into discrete grid point data and establishing corresponding difference equations; the backpropagation algorithm is used to optimize the boundary condition parameters in the difference equations. The error between the predicted production distribution value and the actually measured production data is used as the objective function, and the boundary condition parameters are adjusted through the backpropagation algorithm to minimize the objective function.

[0062] Specifically, the system uses the finite difference method to preliminarily process the full-chain fluid distribution results to obtain an initial predicted production distribution value. Then, the predicted value is compared with the actual production data to calculate the error. The error is backpropagated to the boundary condition parameters, and the boundary condition parameters are adjusted according to a certain learning rate and optimization algorithm (such as the gradient descent method). This process is repeated until the error meets the preset threshold, and finally an optimized predicted production distribution value is obtained.

[0063] S5: Based on the physics-informed neural network, multi-parameter fusion processing is performed on the downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data to generate full-chain fluid parameter simulation results.

[0064] Specifically, the physics-informed neural network is a deep learning method that integrates physical laws into the neural network architecture. By adding a constraint term of the physical equation to the loss function of the neural network, the network can not only learn the data but also meet the requirements of physical laws. In this step, the physics-informed neural network is used to perform multi-parameter fusion processing on the downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data.

[0065] Specifically, the system takes the above multi-source data such as downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data as the input of the neural network, and through the forward propagation and backpropagation processes of the network, learns the complex relationships and internal laws between the data. During the training process, the network continuously adjusts the weight and bias parameters so that the output full-chain fluid parameter simulation results not only conform to the statistical characteristics of the data but also satisfy the constraints of physical laws, thereby improving the accuracy and reliability of the simulation results.

[0066] S6: Based on the ensemble Kalman filter algorithm, data assimilation processing is performed on the predicted production distribution value and the full-chain fluid parameter simulation results to generate a dynamic simulation result synchronized with actual production, and the dynamic simulation result is used to predict the changing trend of oil well production.

[0067] Specifically, the Ensemble Kalman Filter is a filtering algorithm based on Bayesian statistics, which can estimate and update the state of a system by fusing model prediction data and actual observation data. In this step, the Ensemble Kalman Filter algorithm is used to perform data assimilation on the predicted production distribution values and the simulation results of the full-chain fluid parameters. First, based on the predicted production distribution values and the simulation results of the full-chain fluid parameters, the state vector and covariance matrix of the system are constructed. Then, using the actual production data as the observation data, the Ensemble Kalman Filter algorithm is used to update the state vector to obtain the dynamic simulation results synchronized with the actual production. This result can reflect the production status of the oil well in real time and is used to predict the change trend of the oil well production, providing a scientific basis for the development and production management of the oil field.

[0068] In summary, the oil well production prediction method based on artificial intelligence provided by the present invention can comprehensively, accurately and real-time simulate and predict the oil well production system through technical methods such as multi-modal sensor data acquisition, multiphase fluid mechanics calculation, multiphase flow simulation, dynamic boundary optimization, multi-parameter fusion and data assimilation, thereby improving the accuracy of oil well production prediction and providing a reliable basis for production decision-making. At the same time, this method can realize the dynamic monitoring and optimization adjustment of the production system, contribute to improving production efficiency and reducing production costs, and integrates physical models and data-driven methods to ensure the accuracy and reliability of simulation results, providing strong support for the intelligent production management of oil wells.

[0069] In one of the embodiments, S2 of the oil well production prediction method based on artificial intelligence provided by the present invention specifically includes the following steps:

[0070] S21: Calculate the downhole pressure data based on the holdup equation and the flow velocity gradient equation to obtain the initial velocity distribution data of the fluid in the wellbore.

[0071] Specifically, the holdup equation is used to describe the volume fraction distribution of the fluid in the wellbore, and the flow velocity gradient equation is used to describe the variation law of the fluid velocity with position. By combining these two equations, a mathematical relationship between the fluid velocity and the downhole pressure can be established. Among them, the holdup equation and the flow velocity gradient equation can be expressed as:

[0072] ;

[0073] Among them, is the holdup, is the liquid volume flow rate, is the cross-sectional area of the wellbore, is the fluid velocity, is the height direction, is the pressure gradient, is the fluid viscosity; the system substitutes the collected downhole pressure data into the above equation, solves the velocity distribution of the fluid through numerical calculation methods, determines the volume fraction of the fluid according to the holdup equation, and then combines the velocity gradient equation and uses numerical methods such as the finite difference method to calculate the fluid velocity at different height positions in the wellbore to obtain the initial velocity distribution data.

[0074] S22: Perform sliding window averaging on the initial velocity distribution data based on the sliding window technique to generate smoothed fluid velocity distribution data.

[0075] Specifically, the sliding window technique is a data smoothing method that moves a fixed-size window over a data sequence and performs averaging or other statistical processing on the data within the window to reduce data fluctuations and noise. In this step, using sliding window averaging can effectively smooth short-term fluctuations and outliers in the initial velocity distribution data, making the data smoother and more stable. The system sets a fixed-size window, for example, the window size is 5 data points, and slides the window sequentially over the initial velocity distribution data sequence. For each window position, calculate the average value of the data points within the window and replace the value of the data point at the center position of the window with this average value. In this way, smooth the entire initial velocity distribution data to generate smoothed fluid velocity distribution data.

[0076] S23: Perform fluid mechanics constraint processing on the fluid velocity distribution data and downhole fluid ratio data based on the physics-informed neural network to generate the fluid density distribution data in the wellbore.

[0077] Specifically, the physics-informed neural network is a deep learning method that incorporates physical laws into the neural network training process. By adding a constraint term of the physical equation to the loss function, it ensures that the output result of the neural network not only conforms to the statistical characteristics of the data but also meets the requirements of physical laws. In this step, using the physics-informed neural network to process the fluid velocity distribution data and downhole fluid ratio data can ensure that the generated fluid density distribution data conforms to the basic principles of fluid mechanics. Specifically, the system takes the fluid velocity distribution data and downhole fluid ratio data as the input of the neural network and designs a network structure containing multiple hidden layers and neurons. During the training process of the network, in addition to using the traditional loss function (such as mean square error) to measure the difference between the predicted value and the true value, the basic equations of fluid mechanics (such as the mass conservation equation, momentum conservation equation, etc.) are also added as constraint terms. By using an optimization algorithm (such as the gradient descent method) to adjust the weight and bias parameters of the network, the output result of the network not only conforms to the characteristics of the input data but also meets the constraint conditions of fluid mechanics, thereby generating the fluid density distribution data in the wellbore.

[0078] S24: Integrate and process the fluid velocity distribution data and the fluid density distribution data to generate wellbore flow characteristic parameters.

[0079] Specifically, the integration and processing process includes aligning and fusing the velocity distribution data and the density distribution data in space and time to ensure that the two are combined under the same coordinate system and time series. Through mathematical operations and data fusion techniques, the velocity and density data are integrated into a comprehensive flow characteristic parameter. This parameter can comprehensively describe the flow state of the fluid in the wellbore, including the velocity distribution, density distribution of the fluid, and the mutual relationship between the two. The wellbore flow characteristic parameters are important basic data for subsequent oil production prediction. They can provide detailed information about the fluid flow characteristics for the model and help improve the accuracy and reliability of production prediction.

[0080] In one embodiment, step S3 of the oil well production prediction method based on artificial intelligence provided by the present invention specifically includes the following steps:

[0081] S31: Train a multiphase flow model based on machine learning algorithms using the historical production data of the oil well and the historical pipeline network parameters to generate a multiphase flow simulation kernel.

[0082] Specifically, the machine learning algorithms in the present invention mainly refer to algorithms such as neural networks and support vector machines that can learn patterns and rules from historical data. The historical production data of the oil well includes information such as past oil well production, fluid ratio, pressure, temperature, etc., and the historical pipeline network parameters cover data such as the pipe diameter, length, material of the pipeline network, and the corresponding temperature, pressure, etc. After preprocessing these historical data through steps such as cleaning and normalization, they are input into the machine learning model for training. The model continuously adjusts its own parameters during the training process to minimize the error between the predicted value and the actual value, and finally generates a multiphase flow simulation kernel that can accurately describe the flow characteristics of multiphase fluids in the pipeline network. This kernel is essentially a trained model that can predict key information such as the flow state and pressure distribution of the fluid in the pipeline network based on the input current data.

[0083] S32: Calculate the pressure drop of the pipe section based on the multiphase flow simulation kernel, the pipeline network temperature data, the pipe diameter parameters, and the wellbore flow characteristic parameters to generate the pipeline network fluid pressure distribution.

[0084] Specifically, the system estimates the fluid pressure drop in each pipe segment by inputting the above data into the multiphase flow simulation kernel and utilizing the multiphase flow laws learned in the kernel. The pipe network temperature data affects physical properties such as the viscosity and density of the fluid, the pipe diameter parameter determines the flow resistance of the fluid, and the wellbore flow characteristic parameters give the initial state of the fluid when it enters the pipe network. Considering these factors comprehensively, the multiphase flow simulation kernel can calculate the pressure drop of each pipe segment, and then generate a fluid pressure distribution map of the entire pipe network. This pressure distribution map visually shows the pressure changes of the fluid in various parts of the pipe network, providing a basis for subsequent flow rate constraint optimization and the generation of the full-chain fluid distribution results.

[0085] S33: Identify whether the wellbore velocity distribution of the wellbore flow characteristic parameters exceeds a preset flow velocity safety threshold. If it exceeds, optimize the wellhead flow rate constraint of the pipe network fluid pressure distribution based on the adaptive particle swarm algorithm to generate updated boundary conditions.

[0086] Specifically, the system extracts the wellbore velocity distribution data from the wellbore flow characteristic parameters and compares it with the preset flow velocity safety threshold. The flow velocity safety threshold is set according to the production requirements and safety standards of the oil well. Exceeding this threshold may lead to unstable flow in the wellbore or other production problems.

[0087] If it is identified that the wellbore velocity distribution exceeds the safety threshold, it is necessary to optimize the pipe network fluid pressure distribution. Preferably, the adaptive particle swarm algorithm can be used. The adaptive particle swarm algorithm is an intelligent optimization algorithm that dynamically adjusts the velocity and position of particles by simulating the flight and optimization process of particles in the search space to find the optimal solution. In the present invention, the system uses this algorithm to optimize the wellhead flow rate constraint. By adjusting the wellhead flow rate, the flow velocity in the wellbore is brought back to the safe range while minimizing the impact on the overall production efficiency. The optimized result is the updated boundary condition, which will be used as an important basis for subsequent calculations.

[0088] S34: Based on the numerical iteration method, perform pressure-flow synchronous coupling on the updated boundary conditions and the wellbore flow characteristic parameters to generate the full-chain fluid distribution results.

[0089] Specifically, the numerical iteration method is a numerical method that approximates the true solution through continuous iterative calculations. In this step, the updated boundary conditions and wellbore flow characteristic parameters are used as the initial inputs, and the pressure and flow rate distributions of the fluid in the wellbore and pipeline network are gradually calculated through the numerical iteration method. During the iteration process, the estimated values of pressure and flow rate are continuously adjusted so that the pressure and flow rate distributions after iteration satisfy the basic equations and boundary conditions of fluid mechanics until the set convergence accuracy is achieved. The finally generated full-chain fluid distribution results comprehensively consider various factors such as the pressure, flow rate, velocity, and density of the fluid in the wellbore and pipeline network, and can accurately reflect the distribution of the fluid in the entire production system. This result provides comprehensive and accurate fluid flow information for subsequent oil production prediction, helps improve the accuracy and reliability of the prediction, and thus provides a scientific basis for the production management of the oilfield, guiding the formulation and optimization of production decisions.

[0090] In one embodiment, as Figure 2 shown, step S4 of a method for predicting oil well production based on artificial intelligence provided by the present invention specifically includes the following steps:

[0091] S41: Divide the full-chain fluid distribution results into multiple spatio-temporal grid cells, and discretize the mass flow rate distribution of each spatio-temporal grid cell through the finite difference method to generate discretized flow rate distribution data.

[0092] Specifically, the Long Short-Term Memory Neural Network (LSTM) is a neural network structure that can effectively process time series data. It controls the flow of information through a gating mechanism and can capture long-term dependencies in the data. In this step, the system uses the discretized flow rate distribution data and downhole fluid ratio data as the inputs of the LSTM network. The network dynamically adjusts the production allocation ratio weights of each grid cell at different time steps by learning the patterns and rules in the historical data. This means that the network can automatically determine the contribution ratio of each grid cell to the total production at different times according to the dynamic changes of fluid flow and the ratio of different phase fluids. In this way, the LSTM network can generate an initial prediction value of the production distribution considering time and space factors, providing a preliminary result for subsequent prediction optimization.

[0093] S42: Dynamically adjust the allocation ratio weights of the flow rate distribution data and downhole fluid ratio data based on the Long Short-Term Memory Neural Network to generate an initial prediction value of the production distribution.

[0094] Specifically, the initial prediction value of the production distribution is generated through the following steps:

[0095] S421: Normalize the downhole fluid proportion data to generate standardized fluid proportion data.

[0096] Specifically, the normalization process is to convert data with different ranges and dimensions into a unified standard range, usually [0, 1] or [-1, 1]. In the present invention, a suitable normalization method, such as min-max normalization, can be used to process the downhole fluid proportion data. Min-max normalization linearly maps the data to a specified range, and the formula is:

[0097] ;

[0098] where, is the normalized data, is the original data, and are the minimum and maximum values of the data respectively. The normalized fluid proportion data can eliminate the dimensional differences between different phase fluid proportion data, and improve the training efficiency and prediction accuracy of the neural network.

[0099] S422: Input the standardized fluid proportion data into a long short-term memory neural network, and extract the temporal features of the wellbore and pipeline network distribution through the gating mechanism to generate the initial allocation proportion weights.

[0100] Specifically, the gating mechanism of the long short-term memory neural network includes an input gate, a forget gate, and an output gate. The long short-term memory neural network controls the information input at the current moment, the retention of previous information, and the output of information at the current moment through the gating mechanism respectively. During the training process, the network automatically learns how to extract the key features of fluid distribution in the wellbore and pipeline network according to the temporal changes of the standardized fluid proportion data. These features reflect the flow trends and interactions of different phase fluids at different times. Based on the extracted features, the network further generates the initial production allocation proportion weights, which represent the contribution proportions of each phase fluid to the total production at different time and space positions.

[0101] S423: Optimize the wellbore-pipeline network allocation proportion for the initial allocation proportion weights and flow distribution data through the dynamic programming algorithm to generate the initial predicted value of the production distribution.

[0102] Specifically, the dynamic programming algorithm is a method for solving the optimal solution by decomposing complex problems into multiple sub-problems. In this step, the dynamic programming algorithm is used to optimize the initial allocation ratio weights and flow distribution data. This algorithm can consider the flow distribution conditions at different locations and times, as well as the proportional relationship of each phase fluid, and find the optimal allocation ratio weights that make the overall production prediction value closest to the actual situation. Specifically, the system takes the initial allocation ratio weights and flow distribution data as the input of the dynamic programming algorithm, and iteratively solves the optimal allocation ratio weights by establishing a state transition equation and an optimization objective function. Then, based on the optimized weights and flow distribution data, the production distribution conditions at each location and time are calculated to generate an initial prediction value of the production distribution. This prediction value comprehensively considers the flow characteristics and proportional relationship of the fluid, providing a basis for the subsequent optimization of the prediction value.

[0103] S43: Identify whether the deviation between the initial prediction value and the downhole fluid ratio data exceeds a preset deviation threshold. If it exceeds, iteratively update the allocation weights of the initial prediction value based on the backpropagation algorithm to generate an optimized production distribution prediction value.

[0104] Specifically, the system compares the generated initial prediction value with the actual downhole fluid ratio data and calculates the deviation between the two. And verify the deviation between the initial prediction value and the downhole fluid ratio data through a preset deviation threshold determined according to production experience and accuracy requirements. If the deviation between the two exceeds the preset deviation threshold, it indicates that there is a large difference between the initial prediction value and the actual situation, and optimization is required.

[0105] The backpropagation algorithm is an optimization algorithm based on gradient descent and is commonly used in the training of neural networks. In this step, the backpropagation algorithm is applied to the optimization of the initial prediction value. By calculating the error gradient between the initial prediction value and the actual value, the backpropagation adjusts the weight parameters in the LSTM network. Continuously iterate and update the weights to reduce the deviation between the optimized production distribution prediction value and the actual downhole fluid ratio data until the convergence condition is met or the preset number of iterations is reached, and finally obtain a more accurate optimized production distribution prediction value.

[0106] The above-provided oil well production prediction method based on artificial intelligence can achieve dynamic boundary optimization processing of the full-chain fluid distribution results through the comprehensive application of various technical methods such as the finite difference method, long short-term memory neural network, dynamic programming algorithm, and backpropagation algorithm. It can also dynamically adjust the production distribution prediction according to real-time data, improving the accuracy and timeliness of the prediction results. At the same time, this method combines neural networks and optimization algorithms to fully explore the temporal features and spatial distribution laws in the data, providing more reliable data support for oil well production prediction. Combined with the iterative update mechanism, it can ensure that the prediction results can continuously approach the actual production situation, providing a strong basis for production decision-making.

[0107] In one of the embodiments, step S6 of the oil well production prediction method based on artificial intelligence provided by the present invention specifically includes the following steps:

[0108] S61: Extract the main frequency components in the pressure and temperature change trends from the production distribution prediction values based on the Fourier transform algorithm to generate frequency domain feature data.

[0109] Specifically, the Fourier transform is a mathematical transformation method that converts a time-domain signal into a frequency-domain signal. In this step, the system takes the production distribution prediction values as the time-domain signal and applies the Fourier transform algorithm to convert it into a frequency-domain signal. By analyzing the frequency-domain signal, the main frequency components in the pressure and temperature change trends can be identified. These main frequency components represent the most significant frequency components in the change trends. In the frequency domain, the system extracts the main frequency components as the frequency domain feature data according to the energy distribution and amplitude size of the signal. These frequency domain feature data can reflect the main periodicity and regularity of the pressure and temperature changes in the production distribution prediction values, providing a basis for the subsequent generation of digital twin model data.

[0110] S62: Supplement the missing data points of the frequency domain feature data through the cubic spline interpolation method to generate complete digital twin model data.

[0111] Specifically, cubic spline interpolation is an interpolation method based on cubic polynomials. This algorithm can accurately estimate the values of missing data points while maintaining the smoothness of the interpolation function. In the present invention, the system analyzes the frequency-domain characteristic data, determines the positions and quantities of the missing data points, and uses the known data points to establish a cubic spline interpolation function. This function is composed of a cubic polynomial between each pair of adjacent known data points and satisfies the conditions that the function values, first-order derivatives, and second-order derivatives are continuous at the known data points. Subsequently, the system calculates the frequency-domain characteristic values at the missing data points by solving the interpolation function, thereby generating complete digital twin model data. The digital twin model data is an accurate mapping of the actual production system in the virtual space, which contains comprehensive information on the predicted production distribution values in both the frequency domain and the time domain, providing complete data support for subsequent data fusion and dynamic simulation.

[0112] S63: Use the ensemble Kalman filter algorithm to perform data fusion on the simulation results of the full-chain fluid parameters and the digital twin model data, generating dynamic simulation results synchronized with the actual production.

[0113] Specifically, the ensemble Kalman filter is a filtering algorithm based on Bayesian statistics, which can estimate and update the state of the system by fusing model prediction data and actual observation data. In this step, the system takes the simulation results of the full-chain fluid parameters as the model prediction data and the digital twin model data as the actual observation data, and applies the ensemble Kalman filter algorithm for data fusion. The system performs fusion processing on the simulation results of the full-chain fluid parameters and the digital twin model data through the ensemble Kalman filter algorithm. The algorithm calculates the optimal state estimate value according to the covariance relationship between the two and the characteristics of the observation noise, generating dynamic simulation results synchronized with the actual production. This result synthesizes the information of model prediction and actual observation, can more accurately reflect the production state and change trend of the oil well, and provides strong support for real-time optimization decisions.

[0114] In summary, a method for predicting oil well production based on artificial intelligence provided by the present invention can achieve frequency-domain feature extraction, data supplementation, and data fusion of the predicted production distribution values through the comprehensive application of various technical methods such as Fourier transform, cubic spline interpolation, and ensemble Kalman filter, and finally generate dynamic simulation results synchronized with the actual production. Through this process, the present invention can extract key frequency-domain features from the production prediction data to provide important frequency information for subsequent analysis; and by interpolating to supplement missing data, the integrity and usability of the data can be improved; at the same time, different sources of data are effectively combined through the data fusion method, thereby enhancing the accuracy and reliability of the simulation results. In addition, this aspect also reacts in real time to the state changes of the oil well production system through the dynamic simulation results, providing strong support for production decisions and production prediction, and helping to improve production efficiency and economic benefits.

[0115] In a second aspect, as Figure 3 shown, the present invention provides an oil well production prediction system 700 based on artificial intelligence, and the system is configured with the following modules:

[0116] A data acquisition module 710, configured to collect downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data in real time based on downhole multimodal sensors and above-ground pipeline network multimodal sensors, and the downhole fluid ratio data is used to indicate the three-phase fluid ratio of the oil phase, gas phase, and water phase;

[0117] A characteristic parameter calculation module 720, configured to calculate wellbore flow characteristic parameters based on downhole fluid ratio data and downhole pressure data according to multiphase fluid mechanics formulas, and generate wellbore flow characteristic parameters, where the wellbore flow characteristic parameters are used to indicate the velocity distribution and density distribution of the fluid in the wellbore;

[0118] A pressure-temperature coupling module 730, configured to perform pressure-temperature coupling calculation on pipeline network temperature data, pipe diameter parameters, and wellbore flow characteristic parameters based on a multiphase flow simulation algorithm, and generate a full-chain fluid distribution result;

[0119] A dynamic optimization module 740, configured to perform dynamic boundary optimization processing on the full-chain fluid distribution result based on the finite difference method and the backpropagation algorithm, and generate an optimized production distribution prediction value;

[0120] A multi-parameter fusion module 750, configured to perform multi-parameter fusion processing on downhole fluid ratio data, downhole pressure data, pipe diameter parameters, and pipeline network temperature data based on a physics-informed neural network, and generate a full-chain fluid parameter simulation result;

[0121] A synchronous simulation generation module 760, configured to perform data assimilation processing on the production distribution prediction value and the full-chain fluid parameter simulation result based on the ensemble Kalman filter algorithm, and generate a dynamic simulation result synchronized with actual production, where the dynamic simulation result is used to predict the changing trend of the oil well production.

[0122] In summary, an oil well production prediction system provided by the present invention can achieve comprehensive, accurate, and real-time simulation and prediction of an oil well production system through technical methods such as multi-modal sensor data acquisition, multiphase fluid mechanics calculation, multiphase flow simulation, dynamic boundary optimization, multi-parameter fusion, and data assimilation, thereby improving the accuracy of oil well production prediction, providing a reliable basis for production decision-making. At the same time, the system can achieve dynamic monitoring and optimization adjustment of the production system, contribute to improving production efficiency and reducing production costs, and integrates physical models and data-driven methods, thereby ensuring the accuracy and reliability of simulation results and providing strong support for the intelligent production management of oil wells.

[0123] Preferably, the characteristic parameter calculation module 720 provided by the present invention is configured with the following units:

[0124] A velocity calculation unit 721, configured to calculate downhole pressure data based on the holdup equation and the flow velocity gradient equation to obtain initial velocity distribution data of the fluid in the wellbore;

[0125] A velocity smoothing unit 722, configured to perform a moving window averaging process on the initial velocity distribution data based on the moving window technique to generate smoothed fluid velocity distribution data;

[0126] A density generation unit 723, configured to perform a fluid mechanics constraint process on the fluid velocity distribution data and the downhole fluid ratio data based on a physics-informed neural network to generate fluid density distribution data in the wellbore;

[0127] A parameter integration unit 724, configured to integrate the fluid velocity distribution data and the fluid density distribution data to generate wellbore flow characteristic parameters.

[0128] Preferably, the pressure-temperature coupling module 730 provided by the present invention is configured with the following units:

[0129] A multiphase flow model training unit 731, configured to train a multiphase flow model based on machine learning algorithms using historical production data of the oil well and historical pipeline network parameters to generate a multiphase flow simulation kernel;

[0130] A pipe section pressure drop calculation unit 732, configured to calculate the pipe section pressure drop based on the multiphase flow simulation kernel for pipeline network temperature data, pipe diameter parameters, and wellbore flow characteristic parameters to generate a pipeline network fluid pressure distribution;

[0131] A boundary condition optimization unit 733, configured to identify whether the wellbore velocity distribution of the wellbore flow characteristic parameters exceeds a preset flow velocity safety threshold. If it exceeds, the wellhead flow rate constraint optimization is performed on the pipeline network fluid pressure distribution based on the adaptive particle swarm algorithm to generate updated boundary conditions;

[0132] A coupling result generation unit 734, configured to perform pressure-flow synchronous coupling on the updated boundary conditions and the wellbore flow characteristic parameters based on the numerical iteration method to generate a full-chain fluid distribution result.

[0133] Preferably, the dynamic optimization module 740 provided by the present invention is configured with the following units:

[0134] A flow rate discretization unit 741, configured to divide the full-chain fluid distribution result into multiple spatio-temporal grid units, and perform a discretization process on the mass flow rate distribution of each spatio-temporal grid unit through the finite difference method to generate discretized flow rate distribution data;

[0135] A production prediction generation unit 742, configured to dynamically adjust the allocation ratio weights of flow distribution data and downhole fluid ratio data based on a long short-term memory neural network, and generate an initial predicted value of production distribution;

[0136] Preferably, the production prediction generation unit 742 is configured with the following sub-units:

[0137] A data normalization unit 7421, configured to normalize the downhole fluid ratio data to generate normalized fluid ratio data;

[0138] A weight generation unit 7422, configured to input the normalized fluid ratio data into a long short-term memory neural network, extract the time series features of wellbore and pipeline network allocation through a gating mechanism, and generate an initial allocation ratio weight;

[0139] A predicted value generation unit 7423, configured to optimize the wellbore-pipeline network allocation ratio of the initial allocation ratio weight and flow distribution data through a dynamic programming algorithm, and generate an initial predicted value of production distribution.

[0140] A production prediction optimization unit 743, configured to identify whether the deviation between the initial predicted value and the downhole fluid ratio data exceeds a preset deviation threshold. If it exceeds, iterative update of the allocation weight of the initial predicted value is performed based on the backpropagation algorithm to generate an optimized production distribution predicted value.

[0141] Preferably, the synchronous simulation generation module 760 provided by the present invention is configured with the following units:

[0142] A frequency domain feature extraction unit 761, configured to extract the main frequency components in the pressure and temperature change trends of the production distribution predicted value based on the Fourier transform algorithm to generate frequency domain feature data;

[0143] A data supplement unit 762, configured to supplement the missing data points of the frequency domain feature data through cubic spline interpolation to generate complete digital twin model data;

[0144] A data fusion unit 763, configured to perform data fusion on the full-chain fluid parameter simulation results and digital twin model data by using the ensemble Kalman filter algorithm to generate a dynamic simulation result synchronized with actual production.

[0145] In one embodiment, the present application further provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the above-mentioned artificial intelligence-based oil well production prediction method.

[0146] In one embodiment, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the above-mentioned artificial intelligence-based oil well production prediction method.

[0147] In the description of this specification, the description referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0148] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0149] As described above, the above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various changes or substitutions, and these should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An oil well production prediction method based on artificial intelligence, characterized in that: The following steps are involved: S1: Based on the downhole multimodal sensor and the ground pipe network multimodal sensor, downhole fluid ratio data, downhole pressure data, pipe diameter parameters and pipe network temperature data are collected in real time, and the downhole fluid ratio data is used to indicate the ratio of the three-phase fluid of the oil phase, the gas phase and the water phase; S2: calculating wellbore flow characteristic parameters on the downhole fluid ratio data and the downhole pressure data based on a multiphase fluid mechanics formula to generate wellbore flow characteristic parameters, wherein the wellbore flow characteristic parameters are used to indicate velocity distribution and density distribution of the fluid in the wellbore; S3: performing pressure and temperature coupling calculation on the pipe network temperature data, the pipe diameter parameters and the wellbore flow characteristic parameters based on a multiphase flow simulation algorithm to generate a full-chain fluid distribution result; S4: Based on the finite difference method and the back propagation algorithm, dynamic boundary optimization processing is performed on the fluid distribution result of the whole chain to generate an optimized output distribution prediction value; S5: performing multi-parameter fusion processing on the downhole fluid ratio data, the downhole pressure data, the pipe diameter parameter and the pipe network temperature data based on a physical information neural network to generate a full-chain fluid parameter simulation result; S6: Based on the ensemble Kalman filter algorithm, data assimilation processing is performed on the production distribution prediction value and the full-chain fluid parameter simulation results to generate dynamic simulation results synchronized with actual production, and the dynamic simulation results are used to predict the change trend of oil well production.

2. The oil well production prediction method according to claim 1, characterized in that: The S2 includes: S21: Calculating the downhole pressure data based on the liquid holdup equation and the velocity gradient equation to obtain initial velocity distribution data of the fluid in the wellbore; S22: performing sliding window averaging processing on the initial velocity distribution data based on a sliding window technique to generate smoothed fluid velocity distribution data; S23: performing fluid mechanics constraint processing on the fluid velocity distribution data and the downhole fluid ratio data based on a physical information neural network to generate fluid density distribution data in the wellbore; S24: Integrate the fluid velocity distribution data and the fluid density distribution data to generate wellbore flow characteristic parameters.

3. The oil well production prediction method according to claim 1, characterized in that: The S3 includes: S31: Train the multiphase flow model based on the historical production data of oil wells and historical pipe network parameters based on machine learning algorithms to generate a multiphase flow simulation kernel; S32: Calculating the pipe section pressure drop based on the pipe network temperature data, the pipe diameter parameters and the wellbore flow characteristic parameters based on the multiphase flow simulation kernel to generate a pipe network fluid pressure distribution; S33: Identify whether the wellbore velocity distribution of the wellbore flow characteristic parameter exceeds a preset velocity safety threshold, and if so, perform wellhead flow constraint optimization on the pipeline network fluid pressure distribution based on an adaptive particle swarm algorithm to generate updated boundary conditions; S34: Based on the numerical iteration method, the updated boundary conditions and the wellbore flow characteristic parameters are subjected to pressure-flow synchronous coupling to generate the full-chain fluid distribution result.

4. The oil well production prediction method according to claim 1, characterized in that: The S4 includes: S41: Divide the whole chain fluid distribution result into a plurality of space-time grid units, discretize the mass flow distribution of each space-time grid unit by finite difference method, and generate discretized flow distribution data; S42: dynamically adjusting the distribution weights of the flow distribution data and the downhole fluid ratio data based on a long short-term memory neural network to generate an initial prediction value of the production distribution; S43: Identify whether the deviation between the initial prediction value and the downhole fluid ratio data exceeds a preset deviation threshold. If so, iteratively update the weights assigned to the initial prediction value based on a back propagation algorithm to generate an optimized production distribution prediction value.

5. The oil well production prediction method according to claim 4, characterized in that: The S42 includes: S421: performing normalization processing on the downhole fluid ratio data to generate standardized fluid ratio data; S422: inputting the standardized fluid ratio data into a long short-term memory neural network, extracting the time series characteristics of the wellbore and pipe network allocation through a gating mechanism, and generating an initial allocation ratio weight; S423: Optimizing the wellbore-pipeline network allocation ratio of the initial allocation ratio weight and the flow distribution data through a dynamic programming algorithm to generate an initial predicted value of the production distribution.

6. The oil well production prediction method according to any one of claims 1 to 5, characterized in that: The S6 includes: S61: extracting the main frequency components in the pressure and temperature change trends of the production distribution prediction value based on the Fourier transform algorithm to generate frequency domain feature data; S62: Supplement the missing data points of the frequency domain feature data by using a cubic spline interpolation method to generate complete digital twin model data; S63: Using the ensemble Kalman filter algorithm to fuse the full-chain fluid parameter simulation results and the digital twin model data, to generate dynamic simulation results synchronized with actual production.

7. An oil well production prediction system based on artificial intelligence, characterized in that: The system comprises: A data acquisition module, used to collect downhole fluid ratio data, downhole pressure data, pipe diameter parameters and pipe network temperature data in real time based on downhole multimodal sensors and ground pipe network multimodal sensors, wherein the downhole fluid ratio data is used to indicate the ratio of three-phase fluids of oil phase, gas phase and water phase; a characteristic parameter calculation module, used to calculate wellbore flow characteristic parameters on the downhole fluid ratio data and the downhole pressure data based on a multiphase fluid mechanics formula, and generate wellbore flow characteristic parameters, wherein the wellbore flow characteristic parameters are used to indicate velocity distribution and density distribution of the fluid in the wellbore; A pressure-temperature coupling module, used to perform pressure-temperature coupling calculation on the pipe network temperature data, the pipe diameter parameters and the wellbore flow characteristic parameters based on a multiphase flow simulation algorithm to generate a full-chain fluid distribution result; A dynamic optimization module, used for performing dynamic boundary optimization processing on the fluid distribution results of the whole chain based on the finite difference method and the back propagation algorithm to generate an optimized output distribution prediction value; A multi-parameter fusion module, used for performing multi-parameter fusion processing on the downhole fluid ratio data, the downhole pressure data, the pipe diameter parameter and the pipe network temperature data based on a physical information neural network to generate a full-chain fluid parameter simulation result; The synchronous simulation generation module is used to perform data assimilation processing on the production distribution prediction value and the full-chain fluid parameter simulation results based on the ensemble Kalman filter algorithm to generate dynamic simulation results synchronized with actual production. The dynamic simulation results are used to predict the change trend of oil well production.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images