A comprehensive analysis method for start-up pressure and remaining oil distribution in heterogeneous reservoirs

Abstract

The invention discloses a method for comprehensively analyzing the starting pressure and remaining oil distribution of a heterogeneous reservoir. The invention relates to the technical field of reservoir analysis, including the use of multi-scale data fusion, intelligent algorithm optimization, and dynamic monitoring and feedback mechanisms to comprehensively analyze the starting pressure and the remaining oil distribution law. The invention has the advantages that: high-precision data such as permeability and porosity obtained by micro-scale core analysis are used as the basic input for numerical simulation of oil reservoirs, so that the simulation results are closer to the actual oil reservoir seepage conditions, and combined with macro-scale logging seismic data, the influence of reservoir heterogeneity and geological structure on fluid flow is better considered, the accuracy and reliability of numerical simulation of oil reservoirs are improved, and more powerful support is provided for the formulation and optimization of oil reservoir development plans. The porosity, oil saturation and other data provided by the micro-core analysis are combined with the reservoir volume and structural information reflected by the macro-logging seismic data.

Description

Technical Field

[0001] The invention relates to the technical field of reservoir analysis, in particular to a comprehensive analysis method for the starting pressure and residual oil distribution of a heterogeneous reservoir. Background Art

[0002] Heterogeneous reservoirs refer to reservoirs with spatial variations in rock physical properties (such as porosity and permeability), fluid properties (such as density and viscosity) and geological structure (such as formation thickness and fault distribution). Such variations can occur at different scales, from the microscopic pore scale to the macroscopic reservoir scale. Heterogeneous reservoirs are widely present in oil and gas reservoirs. Their complex geological structure and fluid seepage characteristics bring many challenges to oil and gas production. Accurately grasping the starting pressure and remaining oil distribution of heterogeneous reservoirs is extremely critical for optimizing development plans and improving recovery rates.

[0003] Research on residual oil distribution at the microscale mainly focuses on fluid distribution and displacement mechanism at the pore scale, while research at the macroscale focuses on the residual oil distribution characteristics of the entire reservoir. There is a lack of effective scale conversion methods between the two, making it difficult to accurately predict the macroscopic residual oil distribution based on microscopic mechanisms, and it is also difficult to deeply understand the formation and distribution mechanism of microscopic residual oil from macroscopic phenomena. To this end, we propose a comprehensive analysis method for the start-up pressure and residual oil distribution of heterogeneous reservoirs. Summary of the invention

[0004] The purpose of the present invention is to provide a comprehensive analysis method for the starting pressure and remaining oil distribution of heterogeneous reservoirs.

[0005] In order to solve the problems raised in the above background technology, the present invention provides the following technical solutions: a method for comprehensive analysis of startup pressure and remaining oil distribution in heterogeneous reservoirs, comprising using multi-scale data fusion, intelligent algorithm optimization, and dynamic monitoring and feedback mechanism to comprehensively analyze the startup pressure and remaining oil distribution law. The specific operation steps of the method for comprehensive analysis of startup pressure and remaining oil distribution in heterogeneous reservoirs are as follows:

[0006] Step 1: Collect micro-scale core analysis data, meso-scale well logging data, and macro-scale seismic data;

[0007] Step 2: Nest the unit models of micro, meso and macro scales to form a reservoir model, and integrate the collected production dynamic data with the geological static data;

[0008] Step 3: Map the microscopic, mesoscopic and macroscopic scale data using the data conversion model, train the model using the mapped data, and train the model to predict the starting pressure;

[0009] Step 4: Use multi-scale simulation equations to simulate the flow and distribution of heterogeneous residual oil at the micro and macro levels;

[0010] Step 5: All optical fiber sensor data are fused and processed, characteristic parameters are extracted from the processed data, and the pressure and temperature changes of the heterogeneous reservoir are monitored in real time;

[0011] Step 6: Assimilate the real-time collected production data and integrate it into the model, and provide real-time feedback and adaptive updates to the analysis model.

[0012] As a further solution of the present invention: in the step 1, core samples are obtained according to drilling, and micro-pore structure analysis is performed on the core samples using micro-CT technology to obtain the lithology, pore structure, permeability and porosity physical properties of the core. According to the pore size, the flow units are divided into three types: micropores with a radius of less than 0.1um, small holes with a radius between 0.1um and 1um, and large holes with a radius of more than 1um. The scattering intensity of gamma rays after the interaction with formation materials is measured by density logging to determine the mesoscopic formation distribution. Flow units based on sedimentary microphases are divided within the formation. The division of mesoscopic flow units reflects the heterogeneity between different sedimentary microphases in the same formation. Seismic detectors are arranged on the ground to excite and receive artificial seismic waves, and the refraction and reflection signals of seismic waves in the formation are recorded and analyzed. The seismic data show the macroscopic structure of the underground formation, including the reservoir structure, the change in formation thickness, and the connectivity between different strata. The macroscopic flow units of the entire reservoir are divided. The division of macroscopic flow units grasps the fluid flow trend of the reservoir and the macroscopic pattern of remaining oil distribution as a whole.

[0013] As a further solution of the present invention: in the step 2, the flow unit models of microscopic, mesoscopic and macroscopic scales are integrated into a multi-scale flow unit nested reservoir model. In the model, the microscopic flow unit is nested in the mesoscopic flow unit, and the mesoscopic flow unit is nested in the macroscopic flow unit. The scales influence and restrict each other through the established association relationship. The production dynamic data of the oil well, including the change of the oil well production, bottom hole pressure and water content over time, are collected. By randomly perturbing the model parameters, a group of set members is generated. Each set member is time-advanced according to the reservoir model. During the time advancement process, the reservoir model is used to predict the production dynamic data corresponding to each set member, and the difference between the predicted production dynamic data of each set member and the actual observed data is calculated. The weight of each set member is determined by combining the prediction error covariance matrix and the observation error covariance matrix. The calculation formula is:

[0014]

[0015] Among them, [P f ] m,n represents the elements of the forecast error covariance matrix, represents the mth variable in the predicted state vector of the i-th set member, represents the mth variable in the ensemble average prediction state vector, represents the nth variable in the predicted state vector of the i-th set member, Represents the nth variable in the ensemble average prediction state vector, N represents the number of all members in the ensemble, the diagonal elements of the prediction error covariance matrix represent the variance of the prediction errors of each state parameter, and the off-diagonal elements show the correlation between the prediction errors of different parameters. The prediction error covariance matrix is ​​used to update the weights when estimating the reservoir state, and the ensemble members are weighted averaged to make the reservoir model more consistent with the actual production situation and enable the continuous integration of production dynamic data and geological static data.

[0016] As a further solution of the present invention: In the step 3, a microscopic, mesoscopic and macroscopic data conversion model based on a neural network is established. Since the dimensions and numerical ranges of the microscopic, mesoscopic and macroscopic data are different, the formula is used:

[0017]

[0018] Among them, x new represents the normalized value, x represents the original data, and x min Represents the minimum value of the data, x max Represents the maximum value of the data. The above formula can be used to map different microscopic, mesoscopic and macroscopic data to the (0,1) interval. The processed microscopic core data, mesoscopic well data and macroscopic seismic data are divided into training set, validation set and test set. The training set is used to train the neural network model, accounting for 60%-80% of the total data set. The validation set is used to adjust the hyperparameters of the model, including the number of hidden layers, the number of neurons and the learning rate, accounting for 10%-20% of the total data set. The test set is used to evaluate the final performance of the model, accounting for 10%-20% of the total set. During the training process, the training set data is input into the model, and iterative training is performed according to the steps of forward propagation, loss calculation, back propagation and weight update.

[0019] As a further solution of the present invention: In step 3, each iteration becomes a training round epoch, and the error between the output and the real micro data is calculated, and the formula is:

[0020]

[0021] Among them, n represents the number of samples, y i Represents real micro data, represents the micro data predicted by the neural network, B represents the average deviation between the predicted value of the model and the true value. After each epoch, the loss value on the validation set is calculated, and the changing trend of the validation loss is observed to monitor the fitting effect of the model. The correlation coefficient R is used to evaluate the performance of the model on the test set. After evaluation, the model is applied to actual micro, meso and macro data conversion. According to geological exploration data and production dynamic data of nearby oil wells, the model is used to predict the start-up pressure of different well locations and different development stages.

[0022] As a further solution of the present invention: In the step 4, the collected reservoir geological data, including core analysis data, well logging data and seismic data, are used to describe the spatial distribution pattern of geological variables using a multi-point geostatistical method, and the remaining oil distribution in the reservoir is predicted by a multi-scale simulation equation, the equation being:

[0023]

[0024] Among them, T p represents the temperature of the particle, T f represents the temperature of the fluid, r represents the radius of the particle, k pf represents the heat transfer coefficient between particles and fluid, q represents the heat release rate of chemical reaction on the interface per unit area, S represents the surface of the fluid, represents the unit external normal vector of the surface at the calculation point, k represents the thermal conductivity, represents the temperature gradient, represents the velocity of the fluid, ρ represents the density of the fluid, c p Represents the specific heat capacity of the fluid. The above equation shows that the mass flux of particle dissolution and adsorption on the interface between the particle surface and the surrounding fluid is equal at the micro and macro scales. The finite element method is used to divide the solution area into a finite number of units. The approximate solution is constructed on each unit to approximate the true solution of the entire area, accurately simulating the flow and distribution of residual oil in heterogeneous reservoirs at the micro and macro scales.

[0025] As a further solution of the present invention: in the step 4, in the numerical simulation of the remaining oil distribution, the relevant parameters in the remaining oil distribution simulation are encoded, and a multi-level encoding structure is adopted to incorporate the geological parameters, fluid parameters and development operation parameters of the remaining oil distribution into the encoding system, and a certain number of individuals are randomly generated to form an initial population, each individual represents a remaining oil distribution situation, and the remaining oil saturation distribution in the initial population is initialized with a bias by referring to the statistical law of the remaining oil distribution in the produced oil reservoirs at different development stages. The initialization function is:

[0026] F=(w1R+w2C -1 +w3T -1 +w4E -1)×f region ×f policy ×f tech

[0027] Among them, R represents the remaining oil recovery rate, C represents the development cost, T represents the development cycle, E represents the environmental impact index, w1, w2, w3 and w4 represent the weight coefficients, w1+w2+w3+w4=1, f region represents the regional difference factor, f policy represents the policy factor, f tech represents the technical feasibility factor, f region 、f policy 、f tech ∈[0, 1], F represents the fitness index of an individual under multiple objectives. The selection operator is used to select individuals with higher fitness F from the current population as parents. The selected parent individuals are randomly paired, and each gene position of the parent individuals is exchanged using uniform crossover. The offspring individuals obtained through the crossover operation are randomly selected from the offspring individuals. Gene positions on the chromosomes are mutated to change their gene values. The mutation operation increases the diversity of the population and prevents the algorithm from converging to the local optimal solution too early. After each iteration, the optimal individual in the population that meets the termination condition is output as the result of the remaining oil distribution simulation.

[0028] As a further solution of the present invention: in step five, the optical fiber sensor is buried near the reservoir, the sampling frequency of pressure and temperature is set to 1 Hz-5 Hz, the optical fiber sensor converts the perceived changes in temperature and pressure into changes in optical signals, the optical fiber sensor is connected to the data acquisition equipment and the monitoring system through an optical fiber jumper, the monitoring system processes and analyzes the received data, fuses the data from all the optical fiber sensors, extracts characteristic parameters related to the reservoir pressure and temperature changes from the processed data, and monitors the reservoir pressure and temperature changes in real time.

[0029] As a further solution of the present invention: in the step six, according to the parameters monitored in real time during the production process, different production data have different dimensions and numerical ranges, and a formula is used to map different parameters to the (0,1) interval, and the preprocessed production data is used to estimate and train the parameters of the neural network model, and the real-time production data is assimilated and integrated into the model to achieve adaptive updating of the model, and update conditions are set according to the degree of change of the production data, the quality of the data and the performance indicators of the model. When the deviation degree B between the newly collected production data and the model prediction data is greater than 0.1, the model update is triggered, and the updated model is verified using data that is not used for model training and updating, and the prediction results of the model are compared with the actual production data, including the deviation degree B and the correlation coefficient R, to evaluate the performance improvement after the model update.

[0030] By adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are:

[0031] 1. The high-precision data such as permeability and porosity obtained by the micro-scale core analysis of the present invention are used as the basic input for numerical simulation of oil reservoirs, so that the simulation results are closer to the actual oil reservoir seepage situation. Combined with the macro-scale logging seismic data, the influence of reservoir heterogeneity and geological structure on fluid flow is better considered, the accuracy and reliability of numerical simulation of oil reservoirs are improved, and more powerful support is provided for the formulation and optimization of oil reservoir development plans. The porosity, oil saturation and other data provided by the micro-core analysis are combined with the reservoir volume and structural information reflected by the macro-logging seismic data, so that the reserves of the oil reservoir can be calculated more accurately, avoiding the reserve estimation error that may be caused by relying solely on single-scale data, and providing a more accurate basis for the rational development and utilization of oil and gas resources;

[0032] 2. The present invention optimizes the key parameters of the numerical simulation model by adopting a genetic algorithm to determine the encoding method and fitness function of the parameters. The fitness function is based on the matching degree between the simulation results and the actual production data (such as oil well production, water content, etc.). Through the selection, crossover and mutation operations of the genetic algorithm, the optimal parameter combination is continuously iterated and searched. This optimization method can effectively improve the prediction accuracy of the numerical simulation model for the distribution of residual oil in heterogeneous reservoirs, avoid simulation errors caused by improper parameter selection, and provide a reliable basis for formulating reasonable residual oil production strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Schematic diagram of the steps for comprehensive analysis of the starting pressure and remaining oil distribution of a novel heterogeneous reservoir in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0035] The present invention provides a method for comprehensively analyzing the startup pressure and the remaining oil distribution of a heterogeneous reservoir, comprising using multi-scale data fusion, intelligent algorithm optimization, and a dynamic monitoring and feedback mechanism to comprehensively analyze the startup pressure and the remaining oil distribution law. The specific operation steps of the method for comprehensively analyzing the startup pressure and the remaining oil distribution of a heterogeneous reservoir are as follows:

[0036] Step 1: Collect micro-scale core analysis data, meso-scale well logging data, and macro-scale seismic data;

[0037] Step 2: Nest the unit models of micro, meso and macro scales to form a reservoir model, and integrate the collected production dynamic data with the geological static data;

[0038] Step 3: Map the microscopic, mesoscopic and macroscopic scale data using the data conversion model, train the model using the mapped data, and train the model to predict the starting pressure;

[0039] Step 4: Use multi-scale simulation equations to simulate the flow and distribution of heterogeneous residual oil at the micro and macro levels;

[0040] Step 5: All optical fiber sensor data are fused and processed, characteristic parameters are extracted from the processed data, and the pressure and temperature changes of the heterogeneous reservoir are monitored in real time;

[0041] Step 6: Assimilate the real-time collected production data and integrate it into the model, and provide real-time feedback and adaptive updates to the analysis model.

[0042] In one embodiment of the present invention: in step 1, core samples are obtained according to drilling, and micro-pore structure analysis is performed on the core samples using micro-CT technology to obtain the lithology, pore structure, permeability and porosity physical properties of the core. According to the pore size, the flow units are divided into three types: micropores with a radius of less than 0.1um, small holes with a radius between 0.1um and 1um, and large holes with a radius of more than 1um. The scattering intensity of gamma rays after the interaction with formation materials is measured by density logging to determine the mesoscopic formation distribution. Flow units based on sedimentary microphases are divided within the formation. The division of mesoscopic flow units reflects the heterogeneity between different sedimentary microphases in the same formation. Seismic detectors are arranged on the ground to excite and receive artificial seismic waves, and the refraction and reflection signals of seismic waves in the formation are recorded and analyzed. The seismic data show the macroscopic structure of the underground formation, including the reservoir structure, the change in formation thickness, and the connectivity between different strata. The macroscopic flow units of the entire reservoir are divided. The division of macroscopic flow units grasps the fluid flow trend of the reservoir and the macroscopic pattern of remaining oil distribution as a whole.

[0043] In one embodiment of the present invention: in step 2, the flow unit models of micro, meso and macro scales are integrated into a multi-scale flow unit nested reservoir model. In the model, the micro flow unit is nested in the meso flow unit, and the meso flow unit is nested in the macro flow unit. The scales influence and restrict each other through the established association relationship. The production dynamic data of the oil well, including the change of oil well production, bottom hole pressure and water content over time, are collected. By randomly perturbing the model parameters, a group of set members are generated. Each set member is time-advanced according to the reservoir model. During the time advancement process, the reservoir model is used to predict the production dynamic data corresponding to each set member, and the difference between the predicted production dynamic data of each set member and the actual observed data is calculated. The weight of each set member is determined by combining the prediction error covariance matrix and the observation error covariance matrix. The calculation formula is:

[0044]

[0045] Among them, [P f ] m,n represents the elements of the forecast error covariance matrix, represents the mth variable in the predicted state vector of the i-th set member, represents the mth variable in the ensemble average prediction state vector, represents the nth variable in the predicted state vector of the i-th set member, Represents the nth variable in the ensemble average prediction state vector, N represents the number of all members in the ensemble, the diagonal elements of the prediction error covariance matrix represent the variance of the prediction errors of each state parameter, and the off-diagonal elements show the correlation between the prediction errors of different parameters. The prediction error covariance matrix is ​​used to update the weights when estimating the reservoir state, and the ensemble members are weighted averaged to make the reservoir model more consistent with the actual production situation and enable the continuous integration of production dynamic data and geological static data.

[0046] In one embodiment of the present invention: In step 3, a microscopic, mesoscopic and macroscopic data conversion model based on a neural network is established. Since the dimensions and numerical ranges of microscopic, mesoscopic and macroscopic data are different, the formula is used:

[0047]

[0048] Among them, x new represents the normalized value, x represents the original data, and x min Represents the minimum value of the data, x maxRepresents the maximum value of the data. The above formula can be used to map different microscopic, mesoscopic and macroscopic data to the (0,1) interval. The processed microscopic core data, mesoscopic well data and macroscopic seismic data are divided into training set, validation set and test set. The training set is used to train the neural network model, accounting for 60%-80% of the total data set. The validation set is used to adjust the hyperparameters of the model, including the number of hidden layers, the number of neurons and the learning rate, accounting for 10%-20% of the total data set. The test set is used to evaluate the final performance of the model, accounting for 10%-20% of the total set. During the training process, the training set data is input into the model, and iterative training is performed according to the steps of forward propagation, loss calculation, back propagation and weight update.

[0049] In one embodiment of the present invention: In step 3, each iteration becomes a training round epoch, and the error between the output and the real micro data is calculated, and the formula is:

[0050]

[0051] Among them, n represents the number of samples, y i Represents real micro data, represents the micro data predicted by the neural network, B represents the average deviation between the predicted value of the model and the true value. After each epoch, the loss value on the validation set is calculated, and the changing trend of the validation loss is observed to monitor the fitting effect of the model. The correlation coefficient R is used to evaluate the performance of the model on the test set. After evaluation, the model is applied to actual micro, meso and macro data conversion. According to geological exploration data and production dynamic data of nearby oil wells, the model is used to predict the start-up pressure of different well locations and different development stages.

[0052] In one embodiment of the present invention: In step 4, the collected reservoir geological data, including core analysis data, well logging data and seismic data, are used to describe the spatial distribution pattern of geological variables using a multi-point geostatistical method, and the remaining oil distribution in the reservoir is predicted by a multi-scale simulation equation, the equation being:

[0053]

[0054] Among them, T p represents the temperature of the particle, T f represents the temperature of the fluid, r represents the radius of the particle, k pf represents the heat transfer coefficient between particles and fluid, q represents the heat release rate of chemical reaction on the interface per unit area, S represents the surface of the fluid, represents the unit external normal vector of the surface at the calculation point, k represents the thermal conductivity, represents the temperature gradient, represents the velocity of the fluid, ρ represents the density of the fluid, cp Represents the specific heat capacity of the fluid. The above equation shows that the mass flux of particle dissolution and adsorption on the interface between the particle surface and the surrounding fluid is equal at the micro and macro scales. The finite element method is used to divide the solution area into a finite number of units. The approximate solution is constructed on each unit to approximate the true solution of the entire area, accurately simulating the flow and distribution of residual oil in heterogeneous reservoirs at the micro and macro scales.

[0055] In one embodiment of the present invention: in step 4, in the numerical simulation of remaining oil distribution, the relevant parameters in the remaining oil distribution simulation are encoded, and a multi-level encoding structure is adopted to incorporate the geological parameters, fluid parameters and development operation parameters of the remaining oil distribution into the encoding system, and a certain number of individuals are randomly generated to form an initial population, each individual represents a remaining oil distribution situation, and the remaining oil saturation distribution in the initial population is initialized with a bias by referring to the statistical law of the remaining oil distribution of the produced oil reservoirs at different development stages. The initialization function is:

[0056] F=(w1R+w2C -1 +w3T -1 +w4E -1 )×f region ×f policy ×f tech

[0057] Among them, R represents the remaining oil recovery rate, C represents the development cost, T represents the development cycle, E represents the environmental impact index, w1, w2, w3 and w4 represent the weight coefficients, w1+w2+w3+w4=1, f region represents the regional difference factor, f policy represents the policy factor, f tech represents the technical feasibility factor, f region 、f policy 、f tech ∈[0, 1], F represents the fitness index of an individual under multiple objectives. The selection operator is used to select individuals with higher fitness F from the current population as parents. The selected parent individuals are randomly paired, and each gene position of the parent individuals is exchanged using uniform crossover. The offspring individuals obtained through the crossover operation are randomly selected from the offspring individuals. Gene positions on the chromosomes are mutated to change their gene values. The mutation operation increases the diversity of the population and prevents the algorithm from converging to the local optimal solution too early. After each iteration, the optimal individual in the population that meets the termination condition is output as the result of the remaining oil distribution simulation.

[0058] In one embodiment of the present invention: in step five, a fiber optic sensor is buried near the reservoir, and the sampling frequency of pressure and temperature is set to 1 Hz-5 Hz. The fiber optic sensor converts the perceived changes in temperature and pressure into changes in optical signals. The fiber optic sensor is connected to a data acquisition device and a monitoring system via a fiber optic jumper. The monitoring system processes and analyzes the received data, fuses the data from all fiber optic sensors, extracts characteristic parameters related to reservoir pressure and temperature changes from the processed data, and monitors the pressure and temperature changes of the reservoir in real time.

[0059] In one embodiment of the present invention: In step six, according to the parameters monitored in real time during the production process, different production data have different dimensions and numerical ranges, and a formula is used to map different parameters to the (0,1) interval. The preprocessed production data is used to estimate and train the parameters of the neural network model, and the real-time production data is assimilated and integrated into the model to achieve adaptive updating of the model. The update conditions are set according to the degree of change of the production data, the quality of the data and the performance indicators of the model. When the deviation degree B between the newly collected production data and the model prediction data is greater than 0.1, the model update is triggered, and the updated model is verified using data that is not used for model training and updating. The prediction results of the model are compared with the actual production data, including the degree of deviation B and the correlation coefficient R, and the performance improvement of the model after the update is evaluated.

[0060] Embodiment 1: During the data assimilation process, the prediction error covariance matrix is ​​used to determine the weight of the model prediction when updating the reservoir state estimate. When the prediction error covariance matrix is ​​large, it means that the uncertainty of the model prediction is high. Relative to the observed data, the weight of the model prediction part in the update process will be reduced. On the contrary, if the prediction error covariance matrix is ​​small, the credibility of the model prediction is high, and its weight will increase accordingly. When updating the estimated value of the reservoir pressure, if the pressure prediction error covariance matrix is ​​small, it means that the model predicts the pressure more accurately. Then, when combined with the observed pressure data for updating, the pressure value predicted by the model will be given a larger weight. When calculating the error between the output and the actual micro data, the closer the B value is to 0, the higher the prediction accuracy of the model. By adjusting the weights and other parameters of the neural network, the value of B is continuously reduced.

[0061] Embodiment 2: When evaluating the correlation between predicted data and real data, the closer the absolute value of R is to 1, the stronger the correlation between predicted data and real data, and the better the model performance. After evaluation, the model performance meets expectations and is applied to actual micro- and macro-data conversion. In new undeveloped areas or areas with less data, existing logging and seismic data are used to predict key parameters such as micro-porosity and permeability through trained models, providing a basis for reservoir evaluation and oil and gas development program design. At the same time, the model can also be interpreted to analyze which features of macro data play a key role in the prediction of micro data. By calculating feature importance or conducting feature selection experiments, it is found that resistivity logging data contributes more to the prediction of micro-porosity, which can help geologists and engineers better understand the intrinsic connection between data and further optimize data collection and analysis strategies.

[0062] Embodiment 3, the genetic algorithm optimizes the parameters of the remaining oil distribution simulation by continuously iterating the selection, crossover and mutation operations. In each generation of the population, the reservoir parameter combination corresponding to each individual is numerically simulated, and its fitness value is calculated. Genetic operations are performed according to the fitness value to generate a new generation of population. The maximum number of iterations is set to 100 generations. As the number of iterations increases, the individuals in the population gradually evolve toward a better reservoir parameter combination, that is, the simulated remaining oil distribution matches the actual production data better and better. The optimized remaining oil distribution simulation results are compared with the actual remaining oil distribution. On a typical section of the reservoir, the simulated remaining oil saturation contour is compared with the actual remaining oil saturation contour obtained by the reservoir monitoring technology. If the two are consistent in shape, position and value, it means that the optimized simulation can accurately reflect the distribution of the remaining oil.

[0063] As attached Figure 1 As shown, the model is used to convert the collected micro-scale and macro-scale data, integrate the collected dynamic data with the geological static data, and simulate and predict the start-up pressure and remaining oil distribution.

[0064] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

[0065] In the description of the specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0066] The above contents are merely examples and explanations of the present invention. Various modifications or additions to the specific embodiments described or replacements in similar ways by technicians in the technical field shall fall within the protection scope of the present invention as long as they do not deviate from the invention or exceed the scope defined by the claims.

[0067] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for comprehensive analysis of startup pressure and remaining oil distribution in heterogeneous reservoirs, comprising the use of multi-scale data fusion, intelligent algorithm optimization, and dynamic monitoring and feedback mechanism to comprehensively analyze the startup pressure and remaining oil distribution law, characterized in that: The specific steps are as follows: Step 1: Collect micro-scale core analysis data, meso-scale well logging data, and macro-scale seismic data; Step 2: Nest the unit models of micro, meso and macro scales to form a reservoir model, and integrate the collected production dynamic data with the geological static data; Step 3: Map the microscopic, mesoscopic and macroscopic scale data using the data conversion model, train the model using the mapped data, and train the model to predict the starting pressure; Step 4: Use multi-scale simulation equations to simulate the flow and distribution of heterogeneous remaining oil at the micro and macro levels. Use the collected reservoir geological data, including core analysis data, well logging data, and seismic data, and use multi-point geostatistical methods to describe the spatial distribution pattern of geological variables. Use multi-scale simulation equations to predict the distribution of remaining oil in the reservoir. The equation is: Among them, T p represents the temperature of the particle, T f represents the temperature of the fluid, r represents the radius of the particle, k pf represents the heat transfer coefficient between particles and fluid, q represents the heat release rate of chemical reaction on the interface per unit area, S represents the surface of the fluid, represents the unit external normal vector of the surface at the calculation point, k represents the thermal conductivity, represents the temperature gradient, represents the velocity of the fluid, ρ represents the density of the fluid, c p Represents the specific heat capacity of the fluid. Through the above equation, we can get that the mass flux of particle dissolution and adsorption at the interface between the particle surface and the surrounding fluid is equal at the micro and macro scales. The finite element method is used to divide the solution area into a finite number of units. By constructing an approximate solution on each unit to approximate the true solution of the entire area, the flow and distribution of the remaining oil in the heterogeneous reservoir at the micro and macro scales can be accurately simulated. Step 5: All optical fiber sensor data are fused and processed, characteristic parameters are extracted from the processed data, and the pressure and temperature changes of the heterogeneous reservoir are monitored in real time; Step 6: Assimilate the real-time collected production data and integrate it into the model, and provide real-time feedback and adaptive updates to the analysis model.

2. A method for comprehensive analysis of heterogeneous reservoir starting pressure and remaining oil distribution according to claim 1, characterized in that: In the step 1, core samples are obtained by drilling, and micro-pore structure analysis is performed on the core samples using micro-CT technology to obtain the lithology, pore structure, permeability and porosity physical properties of the core. According to the pore size, the flow units are divided into three types: micropores with a radius of less than 0.1um, small pores with a radius between 0.1um and 1um, and large pores with a radius of more than 1um. Density logging is used to measure the scattering intensity after the interaction between gamma rays and formation materials to determine the mesoscopic formation distribution. Flow units based on sedimentary microphases are divided within the formation. The division of mesoscopic flow units reflects the heterogeneity between different sedimentary microphases in the same formation. Seismic detectors are arranged on the ground to excite and receive artificial seismic waves, and the refraction and reflection signals of seismic waves in the formation are recorded and analyzed. The seismic data show the macroscopic structure of the underground formation, including the reservoir structure, the change in formation thickness, and the connectivity between different strata. The macroscopic flow units of the entire reservoir are divided. The division of macroscopic flow units grasps the fluid flow trend of the reservoir and the macroscopic pattern of the remaining oil distribution as a whole.

3. A method for comprehensive analysis of heterogeneous reservoir starting pressure and remaining oil distribution according to claim 2, characterized in that: In the step 2, the flow unit models of micro, meso and macro scales are integrated into a multi-scale flow unit nested reservoir model. In the model, the micro flow unit is nested in the meso flow unit, and the meso flow unit is nested in the macro flow unit. The scales influence and restrict each other through the established association relationship. The production dynamic data of the oil well, including the change of oil well production, bottom hole pressure and water content over time, are collected. By randomly perturbing the model parameters, a group of set members are generated. Each set member is time-advanced according to the reservoir model. During the time advancement process, the reservoir model is used to predict the production dynamic data corresponding to each set member, and the difference between the predicted production dynamic data of each set member and the actual observed data is calculated. The prediction error covariance matrix and the observation error covariance matrix are combined to determine the weight of each set member. The calculation formula is: Among them, [P f ] m,n represents the elements of the forecast error covariance matrix, represents the mth variable in the predicted state vector of the i-th set member, represents the mth variable in the ensemble average prediction state vector, represents the nth variable in the predicted state vector of the i-th set member, Represents the nth variable in the ensemble average prediction state vector, N represents the number of all members in the ensemble, the diagonal elements of the prediction error covariance matrix represent the variance of the prediction errors of each state parameter, and the off-diagonal elements show the correlation between the prediction errors of different parameters. The prediction error covariance matrix is ​​used to update the weights when estimating the reservoir state, and the ensemble members are weighted averaged to make the reservoir model more consistent with the actual production situation and enable the continuous integration of production dynamic data and geological static data.

4. A method for comprehensive analysis of heterogeneous reservoir starting pressure and remaining oil distribution according to claim 2, characterized in that: In the step 3, a microscopic, mesoscopic and macroscopic data conversion model based on a neural network is established. Since the dimensions and numerical ranges of microscopic, mesoscopic and macroscopic data are different, the formula is used: Among them, x new represents the normalized value, x represents the original data, and x min Represents the minimum value of the data, x max Represents the maximum value of the data. The above formula can be used to map different microscopic, mesoscopic and macroscopic data to the (0,1) interval. The processed microscopic core data, mesoscopic well data and macroscopic seismic data are divided into training set, validation set and test set. The training set is used to train the neural network model, accounting for 60%-80% of the total data set. The validation set is used to adjust the hyperparameters of the model, including the number of hidden layers, the number of neurons and the learning rate, accounting for 10%-20% of the total data set. The test set is used to evaluate the final performance of the model, accounting for 10%-20% of the total set. During the training process, the training set data is input into the model, and iterative training is performed according to the steps of forward propagation, loss calculation, back propagation and weight update.

5. A method for comprehensive analysis of heterogeneous reservoir starting pressure and remaining oil distribution according to claim 4, characterized in that: In step 3, each iteration becomes a training round epoch, and the error between the output and the real micro data is calculated using the formula: Among them, n represents the number of samples, y i Represents real micro data, represents the micro data predicted by the neural network, B represents the average deviation between the predicted value of the model and the true value. After each epoch, the loss value on the validation set is calculated, and the changing trend of the validation loss is observed to monitor the fitting effect of the model. The correlation coefficient R is used to evaluate the performance of the model on the test set. After evaluation, the model is applied to actual micro, meso and macro data conversion. According to geological exploration data and production dynamic data of nearby oil wells, the model is used to predict the start-up pressure of different well locations and different development stages.

6. The method for comprehensive analysis of heterogeneous reservoir starting pressure and remaining oil distribution according to claim 1, characterized in that: In the step 4, in the numerical simulation of the remaining oil distribution, the relevant parameters in the remaining oil distribution simulation are encoded, and a multi-level encoding structure is adopted to incorporate the geological parameters, fluid parameters and development operation parameters of the remaining oil distribution into the encoding system, and a certain number of individuals are randomly generated to form an initial population, each individual represents a remaining oil distribution situation, and the remaining oil saturation distribution in the initial population is initialized with a bias by referring to the statistical law of the remaining oil distribution of the produced oil reservoirs at different development stages. The initialization function is: F=(w1R+w2C -1 +w3T -1 +w4E -1 )×f region ×f policy ×f tech Among them, R represents the remaining oil recovery rate, C represents the development cost, t represents the development cycle, E represents the environmental impact index, w1, w2, w3 and w4 represent the weight coefficients, w1+w2+w3+w4=1, f region represents the regional difference factor, f policy represents the policy factor, f tech represents the technical feasibility factor, f region 、f policy 、f tech ∈[0, 1], F represents the fitness index of an individual under multiple objectives. The selection operator is used to select individuals with higher fitness F from the current population as parents. The selected parent individuals are randomly paired, and each gene position of the parent individuals is exchanged using uniform crossover. The gene positions on the chromosomes of the offspring individuals obtained through the crossover operation are randomly selected for mutation operations to change their gene values. The mutation operation increases the diversity of the population and prevents the algorithm from converging to the local optimal solution too early. After each iteration, the optimal individual in the population that meets the termination condition is output as the result of the remaining oil distribution simulation.

7. The method for comprehensive analysis of heterogeneous reservoir starting pressure and remaining oil distribution according to claim 1, characterized in that: In the step five, the optical fiber sensor is buried near the reservoir, the sampling frequency of pressure and temperature is set to 1 Hz-5 Hz, the optical fiber sensor converts the perceived changes in temperature and pressure into changes in optical signals, the optical fiber sensor is connected to the data acquisition equipment and the monitoring system through an optical fiber jumper, the monitoring system processes and analyzes the received data, fuses the data from all the optical fiber sensors, extracts characteristic parameters related to the reservoir pressure and temperature changes from the processed data, and monitors the pressure and temperature changes of the reservoir in real time.

8. A method for comprehensive analysis of heterogeneous reservoir starting pressure and remaining oil distribution according to claim 4, characterized in that: In step six, according to the parameters monitored in real time during the production process, different production data have different dimensions and numerical ranges. The different parameters are mapped to the interval (0, 1) using a formula, and the preprocessed production data is used to estimate and train the parameters of the neural network model. The real-time production data is assimilated and integrated into the model to achieve adaptive updating of the model. The updating conditions are set according to the degree of change of the production data, the quality of the data and the performance indicators of the model. When the deviation degree B between the newly collected production data and the model prediction data is greater than 0.1, the model update is triggered. The updated model is verified using data that is not used for model training and updating. The prediction results of the model are compared with the actual production data, including the deviation degree B and the correlation coefficient R, to evaluate the performance improvement of the updated model.

Images