An intelligent prediction method and system for dynamic evolution of an inter-well communication path in an oil reservoir

Abstract

The application provides an intelligent prediction method and system for dynamic evolution of interwell connectivity path based on permeability change, and relates to the technical field of intelligent oil and gas field development. The application preprocesses static seismic multi-attributes, initial permeability and production dynamic data, and extracts multi-scale features; a dynamic permeability and static prior fusion mechanism is established to generate a dynamic weight connectivity graph; on this basis, reinforcement learning path search is adopted, and dynamic re-planning is realized through time sequence rolling to obtain interwell connectivity path and migration law; multi-objective optimization of path length and cumulative flow capacity is combined to form a non-dominated solution set and a Pareto frontier to provide candidate schemes for different engineering preferences; and finally, the spatio-temporal evolution prediction results of the path and the matching adjustment suggestions are output. The application realizes the fusion of static and dynamic data, breaks through the limitation of relying only on static seismic to depict the reservoir, and significantly improves the development effect and decision-making scientificity of marine sandstone reservoirs.

Description

Technical Field

[0001] This invention relates to the field of intelligent oil and gas field development technology, and in particular to an intelligent prediction method and system for the dynamic evolution of inter-well connectivity paths in oil reservoirs. Background Technology

[0002] Existing methods for studying inter-well connectivity are mostly based on static seismic data or single-point-of-time analysis, lacking the ability to predict dynamic evolution and struggling to address permeability variations in complex reservoirs. This method significantly improves upon these methods in data fusion, model design, and path optimization, enhancing the accuracy and operability of predictions.

[0003] Marine sandstone reservoirs are one of the world's major oil and gas resource types, characterized by strong reservoir heterogeneity and complex sedimentary environments. As oilfield development progresses and reservoirs enter the medium-to-high water-cut stage, the study of well connectivity becomes crucial for improving recovery rates and optimizing development strategies. However, the development of marine sandstone reservoirs faces severe challenges. Due to diverse sedimentary environments and significant reservoir heterogeneity, accurate evaluation of well connectivity is difficult. During development, reservoir properties (such as permeability) change, affecting or altering well connectivity and remaining oil distribution, further increasing the complexity of reservoir development and its adjustments.

[0004] To accurately and efficiently evaluate inter-well connectivity and thus guide reservoir development strategies, the use of artificial intelligence (AI) methods for evaluating inter-well connectivity is currently a hot technological trend. However, AI methods have some drawbacks:

[0005] 1. Insufficient feature extraction: The dynamic permeability change data and static seismic data were not effectively integrated, resulting in insufficient ability to extract dynamic permeability change features;

[0006] 2. Insufficient quantitative analysis: Current artificial intelligence methods mostly predict the inter-well connectivity paths at a single point in time, lacking the ability to predict the spatiotemporal evolution of inter-well connectivity paths. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide an intelligent prediction method and system for the dynamic evolution of inter-well connectivity paths in oil reservoirs. This method improves upon traditional methods for characterizing three-dimensional reservoir space based on static seismic data by establishing a fusion mechanism between dynamic permeability and static seismic data. An optimal path search algorithm based on reinforcement learning is employed to predict inter-well connectivity paths and their evolution. The technical achievements of this invention can improve the development performance of marine sandstone reservoirs.

[0008] To achieve the above objectives, the present invention provides the following solution:

[0009] A method for intelligent prediction of dynamic evolution of inter-well connectivity paths in oil reservoirs, comprising:

[0010] It gathers static seismic multi-attribute data, initial permeability data, and production dynamic data, completes noise processing, missing value processing, outlier processing, and duplicate value processing, and extracts multi-scale spatial features through an encoder that combines convolution and downsampling with residual structure.

[0011] Based on the static seismic multi-attribute model, a static connectivity prior for simulating river channels and corridors is constructed. The initial permeability and the time series reflecting changes in formation pressure, water content, particle release and migration intensity are used as model inputs and associated with the static connectivity prior.

[0012] A permeability evolution relationship was established, starting from the initial permeability and driven by the combined effects of formation pressure, water aquifer and particle migration. Regression and recursive fitting were used and online updates were introduced to obtain the permeability field over time.

[0013] A cross-modal attention mechanism is introduced into the deep residual network. Guided by the multi-scale spatial features, the phase-by-phase permeability field is fused with the static seismic multi-attribute to generate a dynamic weighted connectivity graph that characterizes the weights of geometric shape and flow capacity.

[0014] Based on a heuristic path search algorithm, the optimal connectivity path is obtained from the injection well to the production well on the dynamic weighted connectivity graph in each period, and replanning is performed on a rolling basis over time to obtain the path migration pattern and the location of key channels.

[0015] With the optimization objectives of connected path length and cumulative flow capacity, a permeability lower limit constraint and a flow bottleneck constraint are introduced. A multi-objective evolutionary algorithm with non-dominated sorting is used to solve the problem and obtain a non-dominated solution set. The non-dominated solution set is the set of Pareto optimal solutions, and the non-dominated solution set is represented as the Pareto front in the objective space.

[0016] Based on engineering preferences, a connectivity path scheme is selected from the non-dominated solution set, and the spatiotemporal evolution results of the inter-well connectivity path and corresponding adjustment suggestions are output to guide reservoir development.

[0017] Preferably, the encoder is formed by a multi-scale structure consisting of convolution, downsampling and residual connection. The encoder is used to extract the multi-scale spatial features and serve as the feature basis for cross-modal fusion and path evaluation.

[0018] Preferably, the time dependence of the time series is modeled by a sequence modeling unit, which includes a bidirectional long short-term memory network or a temporal convolutional network to characterize the changing patterns of production dynamics over time.

[0019] Preferably, the parameter update of the permeability evolution relationship adopts recursive least squares and introduces an online update mechanism, which is jointly corrected by the previous period's estimation result and the current period's production dynamics, and the period-by-period permeability field is obtained according to time.

[0020] Preferably, the dynamic weighted connectivity graph simultaneously represents the geometric shape and flow capacity weight of the connected path, and serves as the weighted graph for path search and the evaluation benchmark for multi-objective optimization.

[0021] Preferably, the evaluation function of the heuristic path search algorithm consists of the traveled cost and the remaining distance estimate. The node expansion order is determined according to the evaluation value from smallest to largest, and rolling replanning is performed in periods. In each period, the optimal connectivity path is obtained from the injection well to the production well on the dynamic weighted connectivity graph.

[0022] Preferably, in the path search stage based on the heuristic path search algorithm, multiple candidate paths are generated by setting weight preferences for static connectivity priors and dynamic flow capabilities in the evaluation function, so as to reflect different emphases on static geometry and dynamic capabilities.

[0023] Preferably, the multi-objective optimization takes the length of the connected path and the cumulative flow capacity as the optimization objectives, sets the lower limit constraint of permeability and the flow bottleneck constraint, obtains the non-dominated solution set through non-dominated screening, and selects the path scheme from the non-dominated solution set for reservoir development adjustment according to engineering preferences.

[0024] Preferably, the heuristic path search algorithm is the A* optimal path search algorithm.

[0025] A smart prediction system for the dynamic evolution of inter-well connectivity paths in oil reservoirs, comprising:

[0026] The data preprocessing unit is used to collect static seismic multi-attribute data, initial permeability data, and production dynamic data. It completes noise processing, missing value processing, outlier processing, and duplicate value processing. It also extracts multi-scale spatial features through an encoder that combines convolution and downsampling with a residual structure.

[0027] The static prior construction unit is used to construct a static connectivity prior for simulating river channels and corridors based on the multiple attributes of static earthquakes. It takes the initial permeability and the time series reflecting changes in formation pressure, water content, particle release and migration intensity as model inputs and associates them with the static connectivity prior.

[0028] The permeability evolution unit is used to establish the permeability evolution relationship starting from the initial permeability and driven by the combined effects of formation pressure, water aquifer and particle migration. It adopts regression and recursive fitting and introduces online updates to obtain the permeability field of each period over time.

[0029] A cross-modal fusion unit is used to introduce a cross-modal attention mechanism into the deep residual network. Guided by the multi-scale spatial features, the phase-by-phase permeability field is fused with the static seismic multi-attribute to generate a dynamic weighted connectivity graph that characterizes the weights of geometric shape and flow capacity.

[0030] The path search unit is used to find the optimal connectivity path from the injection well to the production well on the dynamic weighted connectivity graph in each period based on the heuristic path search algorithm, and to perform replanning on a rolling basis over time to obtain the path migration pattern and the location of key channels.

[0031] The multi-objective optimization unit is used to optimize the length of the connected path and the cumulative flow capacity. It introduces the lower limit constraint of permeability and the flow bottleneck constraint, and uses a non-dominated sorting multi-objective evolutionary algorithm to solve the problem and obtain a non-dominated solution set. The non-dominated solution set is the set of Pareto optimal solutions, and the non-dominated solution set is represented as the Pareto front in the objective space.

[0032] The results output unit is used to select a connectivity path scheme in the non-dominated solution set according to engineering preferences, and output the spatiotemporal evolution results of the inter-well connectivity path and corresponding adjustment suggestions to guide reservoir development.

[0033] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0034] Background methods often rely solely on static seismic attributes to infer inter-well geometric connectivity, or on empirical interpretations based solely on production curves. This disconnect between static geometry and dynamic flow easily leads to channel identification errors. This invention extracts multi-scale spatial features through data preprocessing and constructs a static connectivity prior using multiple static seismic attributes. Simultaneously, it incorporates initial permeability and time series reflecting formation pressure changes, water content changes, and particle release and migration intensity as model inputs, correlating them with the static connectivity prior. This achieves integrated modeling of static geometry and dynamic flow, reducing path identification instability and inaccurate time-series response caused by information fragmentation from the outset.

[0035] Traditional connectivity inference methods often approximate permeability as a constant or slow-moving variable, failing to reflect the blockage and unblocking processes caused by pressure, water content, and particle migration, leading to time-varying distortions in channel effectiveness. This invention establishes a permeability evolution relationship starting from an initial permeability level and driven by the combined effects of formation pressure, water content, and particle migration. It employs regression and recursive fitting with online updates to obtain a time-series permeability field, allowing inter-well connectivity to be corrected in real-time according to changes in operating conditions and physical properties. This improves the sensitivity and timeliness of identifying channel opening, decay, and recovery.

[0036] Existing fusion methods often use fixed weights or simple superposition, making it difficult to simultaneously highlight geometric connectivity and flow capacity, leading to biased path evaluation. This invention introduces a cross-modal attention mechanism into a deep residual network. Guided by multi-scale spatial features, it fuses phase-by-phase permeability fields with static seismic multi-attributes to generate a dynamic weighted connectivity map representing the weights of geometric morphology and flow capacity. This map unifies morphology and capacity within the same weight space, providing a geologically and engineering-sensitive evaluation benchmark for subsequent path search, thereby reducing misjudgments of "connected morphology but weak flow capacity" or "strong flow capacity but discontinuous morphology."

[0037] The optimal path at a fixed time point is prone to failure after injection-production adjustments, and common methods lack continuous tracking of channel migration. This invention, based on a heuristic path search algorithm, seeks the optimal connectivity path from the injection well to the production well on a dynamic weighted connectivity graph in each period, and performs replanning over time, directly outputting the path migration pattern and key channel locations. This allows for continuous time-series correction of the optimal path, improving the adaptability and robustness of the solution to changes in operating conditions such as well network adjustments, water cut increases, and pressure recovery.

[0038] Single-objective optimization methods tend to favor certain indicators while neglecting overall development effectiveness. This invention uses connected path length and cumulative flow capacity as optimization objectives, introduces lower limit constraints on permeability and flow bottleneck constraints, and employs a non-dominated sorting multi-objective evolutionary algorithm to obtain a non-dominated solution set (represented as the Pareto front in the objective space). The algorithm then selects a solution based on engineering preferences. This mechanism presents the feasible optimal boundary in set form, facilitating the selection of solutions that balance path accessibility and flow capacity under different development preferences, thus enhancing the interpretability and engineering usability of the results. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the static background and initial state provided for an embodiment of the present invention;

[0042] Figure 3 This is a time-series diagram of the dynamic variable of pressure P provided in an embodiment of the present invention;

[0043] Figure 4This is a time-series diagram of the dynamic variable of water content Sw provided in an embodiment of the present invention;

[0044] Figure 5 This is a time-series diagram illustrating the dynamic variable of particle concentration C provided in an embodiment of the present invention;

[0045] Figure 6 A schematic diagram of the regression coefficients of the penetration rate time series model provided in this embodiment of the invention;

[0046] Figure 7 This is a schematic diagram of the final penetration rate provided in an embodiment of the present invention;

[0047] Figure 8 The fusion component and weight map at t=0 provided in this embodiment of the invention;

[0048] Figure 9 The fusion component and weight map provided for t=10 in this embodiment of the invention;

[0049] Figure 10 The fusion component and weight map for t=19 provided in this embodiment of the invention;

[0050] Figure 11 This is a schematic diagram of the optimal path on W(t=0) provided in an embodiment of the present invention;

[0051] Figure 12 This is a schematic diagram of the optimal path on W(t=10) provided in an embodiment of the present invention;

[0052] Figure 13 This is a schematic diagram of the optimal path on W(t=19) provided in an embodiment of the present invention;

[0053] Figure 14 A schematic diagram of the Pareto front provided for an embodiment of the present invention;

[0054] Figure 15 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The purpose of this invention is to provide an intelligent prediction method and system for the dynamic evolution of inter-well connectivity paths in oil reservoirs. By integrating static seismic priors and dynamic permeability evolution, and combining cross-modal attention, rolling path search and multi-objective optimization, it can achieve dynamic and accurate prediction of inter-well connectivity paths and multi-scheme decision support, significantly improving the timeliness, accuracy and engineering applicability of oil reservoir development.

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides an intelligent prediction method for the dynamic evolution of inter-well connectivity paths in oil reservoirs, comprising:

[0059] Step 100: Collect static seismic multi-attribute data, initial permeability data, and production dynamic data; complete noise processing, missing value processing, outlier processing, and duplicate value processing; and extract multi-scale spatial features through an encoder combining convolution and downsampling with residual structure.

[0060] Step 200: Construct a static connectivity prior for simulating river channels and corridors based on the multi-attribute static seismic data. Use the initial permeability and time series reflecting changes in formation pressure, water content, particle release and migration intensity as model inputs and associate them with the static connectivity prior.

[0061] Step 300: Establish a permeability evolution relationship starting from the initial permeability and driven by the combined effects of formation pressure, water aquifer and particle migration. Use regression and recursive fitting and introduce online updates to obtain the permeability field for each period over time.

[0062] Step 400: Introduce a cross-modal attention mechanism in the deep residual network, guided by multi-scale spatial features, and fuse the phase-by-phase permeability field with static seismic multi-attributes to generate a dynamic weighted connectivity graph that characterizes the weights of geometric shape and flow capacity.

[0063] Step 500: Based on the heuristic path search algorithm, the optimal connectivity path is obtained from the injection well to the production well on the dynamic weighted connectivity graph of each period, and replanning is performed in a rolling manner over time to obtain the path migration pattern and the location of key channels.

[0064] Step 600: Taking the length of the connected path and the cumulative flow capacity as optimization objectives, we introduce the lower limit constraint of permeability and the flow bottleneck constraint, and use the non-dominated sorting multi-objective evolutionary algorithm to solve the problem and obtain the non-dominated solution set; the non-dominated solution set is the set of Pareto optimal solutions, and the non-dominated solution set is represented as the Pareto front in the objective space.

[0065] Step 700: Select a connectivity path scheme from the non-dominated solution set based on engineering preferences, and output the spatiotemporal evolution results of the inter-well connectivity path and corresponding adjustment suggestions to guide reservoir development.

[0066] Specifically, in step 100 of this embodiment, data preprocessing is performed, including data collection, data cleaning, and feature extraction. Data collection needs to acquire as comprehensive a range of geological and production dynamic data of the reservoir as possible. Production dynamic data includes production indicators for individual wells and well groups, such as well network type, fluid production, oil production, water production, water cut, and oil-water interface; injection data for injection wells, such as injection volume, injection pressure, and hysteresis response time of the injection-production process; and key time node data, such as water breakthrough time, rate of water cut increase, and time points of significant fluctuations in production indicators. Geological data includes reservoir depth, effective thickness, porosity, initial permeability, and perforation parameters. Data cleaning includes noise removal, missing value removal, outlier removal, and duplicate value removal. Feature extraction refers to using convolutional layers and downsampling to acquire spatial features at different scales during the encoding stage, and enhancing nonlinear modeling capabilities through residual structures to ensure the trainability of deep networks.

[0067] Furthermore, this embodiment proposes a single-parameter causal denoising technique based on well network topology-temporal joint regularization for noise processing. The core idea is to treat the production dynamics of each well within the same reservoir as a multivariate temporal field on a "well network map," using a topological Laplace constrained by the physical proximity between wells and the causal intensity of injection and production for spatial consistency constraints, while simultaneously suppressing short-term noise with temporal difference constraints. Both are controlled by the same regularization intensity, thus achieving "cross-well consistency and temporal smoothing" with a single parameter, and avoiding geological boundary ambiguity caused by excessive smoothing. This method does not introduce subjective weights; the edge weights between wells are derived from objective well spacing and injection-production coherence.

[0068] Specifically, the optimization objective and the formula for the closed-form solution in this embodiment are as follows:

[0069]

[0070] Construction of "causal consistency" for well network diagram weights (data-driven, zero hyperparameter):

[0071] .

[0072] After data collection and cleaning, the production dynamics of multiple wells are aligned by time and stitched together into an observation vector. A well network diagram based on well spacing and injection-production coherence is constructed. The graphical Laplace and time difference operators are calculated, and the above linear equations are solved to obtain the denoised multi-well time series, which is used as the encoder input. This suppresses spikes and isolated anomalies without disrupting sand body boundaries and maintains the causal consistency of injection-production responses. In the formula of the above denoising process in this embodiment, This represents the observation vector stitched together by well and time. The obtained denoised vector; It serves as the sole regularity strength parameter, simultaneously controlling both spatial and temporal constraints. The graph Laplace matrix is ​​constructed based on the well spacing and injection-production coherence. Let be the weighted adjacency matrix of the graph; for The degree matrix; For well With well The right to the side; The distance between wells; The average well spacing across the entire field is used for scale normalization; For injection well The injection volume or injection pressure sequence; For production wells The liquid or oil production sequence; Pearson correlation coefficient; Physical lag in closed intervals The search variables within the scope are given by the time lag range of field experience and are not used as model parameters. For time-by-time first-order difference, it is a block diagonal time difference operator; Let be the identity matrix with the same dimension as the unknown vector.

[0073] In this embodiment of the invention, static seismic multi-attribute and initial permeability data are first obtained as static background information. For example... Figure 2 As shown, static attribute A1 exhibits a locally high Gaussian clumping structure, while static attribute A2 displays an overall smooth distribution. These two attributes are correlated and weighted to form a static connectivity prior, used to simulate the potential channel and corridor orientations within the reservoir. Simultaneously, a heterogeneous initial permeability field k0 is introduced. This permeability field reflects the spatial differences in reservoir pore structure and physical properties, providing a baseline state for subsequent dynamic evolution simulations.

[0074] Building upon the static background, this invention further constructs temporal inputs for multiple types of dynamic variables. For example... Figure 3 As shown, the pressure P gradually decreases around the injection end and fluctuates slightly over time, truly reflecting the pressure transmission law during fluid displacement; as Figure 4 As shown, the water saturation Sw increases over time at the production end, reflecting the dynamic characteristics of water channeling and water cut rise during the injection-production displacement process; for example... Figure 5As shown, the particle concentration C is positively correlated with the pressure gradient and gradually decreases over time, which is used to describe the particle release and migration effects during seepage. The introduction of the above dynamic time series enables this invention to not only rely on static connectivity judgment, but also to realistically characterize the evolution of inter-well connectivity over time.

[0075] By comprehensively inputting the aforementioned static multi-attribute data, initial permeability, and dynamic variables P, Sw, and C, this invention achieves an organic combination of static priors and dynamic temporal sequences. Static connectivity priors ensure the representation of the geometry of potential channels and corridors, while dynamic variables provide multidimensional constraints on pressure transfer, water uplift, and particle migration during displacement. The fusion of these two in the deep model forms a dynamic weighted connectivity graph, providing comprehensive input conditions for subsequent path search and optimization, thereby significantly improving the accuracy and timeliness of well-to-well connectivity path prediction.

[0076] Furthermore, this embodiment considers that the marine sandstone reservoir rock has relatively loose cementation, and during water injection development, the reservoir rock and clay easily release microparticles. These released microparticles migrate with the fluid, altering the original pore structure of the reservoir, forming high-permeability channels or blocking the reservoir. Therefore, it is necessary to use actual reservoir cores and fluid samples from the field to conduct water drive physics experiments, obtaining the dynamic evolution equation of permeability during water drive development. This can quantify the spatiotemporal distribution of permeability changes with microparticle concentration at different development stages. To improve the model's feasibility, this embodiment extends the dynamic evolution equation of permeability into a time-series form:

[0077] .

[0078] in The initial static permeability, For formation pressure, This represents the water saturation level. (Concentration of particles in the fluid). As a multivariate nonlinear function, it can be fitted using machine learning regression models (such as XGBoost and LSTM) from both experimental and field data. A stress-sensitive effect term is introduced into the equation to account for the impact of formation pressure changes on permeability. Simultaneously, an online update mechanism is incorporated, utilizing Kalman filtering to update the permeability field in real time after acquiring the latest production data.

[0079] Furthermore, such as Figure 6 and Figure 7 As shown, the time-series fitting of the penetration rate in this embodiment is as follows:

[0080] (1) Fit using recursive least squares (RLS approximation), the formula is: ;

[0081] (2) Update parameters online and periodically to obtain .

[0082] (3) Use the fitting results to recursively predict k t+1 get .

[0083] in, For a moment Penetration rate; For a moment Penetration rate; This represents the logarithmic change in permeability between adjacent time points; The constant term regression coefficient; The coefficient representing the influence of particulate concentration; For a moment The concentration of particles; The influence coefficient of water saturation; For a moment Water saturation; This is the pressure influence coefficient; For a moment Formation pressure; As the reference pressure; For the regression parameter vector; This represents the predicted permeability value obtained based on the fitting results.

[0084] Furthermore, this embodiment employs a deep residual network algorithm to fuse dynamic permeability data with static seismic multi-attribute data, requiring the dynamic permeability model to be embedded as a key parameter in the model. The specific steps are as follows:

[0085] (1) Expand the input window: On the basis of static seismic data, add the initial permeability distribution field and dynamic image factor field, and input static seismic data and dynamic permeability data at the same time.

[0086] (2) Feature interaction module design: The distribution of connected paths characterized by seismic attributes (static) constrains the spatial range of permeability changes, and the weight of channel permeability is adjusted by real-time particle concentration (dynamic), so that the fused attributes not only reflect the geometric shape of the connected paths, but also reflect the dynamic changes of their seepage capacity.

[0087] (3) Optimize the output window: The output results include the three-dimensional spatial structure of the initial connected path and the dynamic permeability field characterized by static seismic data. The fused attributes are used for three-dimensional spatial rasterization, providing the "dynamic evolution of connected path" parameter basis for subsequent path search algorithms.

[0088] (4) Based on the deep residual network, this embodiment adds a cross-modal attention mechanism to realize the deep feature interaction between dynamic permeability data and static seismic multi-attribute data. Spatial features are extracted by a 3D convolutional network, and temporal dependent features are modeled by Bi-LSTM or TCN, outputting a dynamic weighted connectivity map, which not only includes the geometry of the connected paths, but also reflects the flow capacity weights.

[0089] In this embodiment of the invention, static seismic attributes A1 and A2 are first linearly combined, which is equivalent to using an approximate 1×1 convolution in a deep network to superimpose different attributes to generate a static geometric component S_geom. For example... Figures 8 to 10 As shown, S_geom can highlight the potential river and corridor orientations, providing a geometric basis for connectivity. Through this static geometric component, the model can identify the areas where potential channels exist within the overall spatial pattern, providing geometric constraints for further discrimination under dynamic displacement conditions.

[0090] Building upon the geometric components, this invention further constructs a dynamic capability component S_dyn. This component is a weighted linear combination of initial permeability, water saturation, and particle concentration, reflecting the reservoir's flow capability at the current moment. Figures 8 to 10 As shown, at different time sections of t=0, t=10, and t=19, S_dyn exhibits dynamic changes with rising water cut and particle migration. High-value areas represent channels with strong flow capacity, while low-value areas characterize areas of capacity decay. This dynamic capacity component allows connectivity assessment to no longer rely on a single static condition, but rather reflects the real-time evolution of inter-well seepage.

[0091] Based on this, the present invention couples static geometric components with dynamic capability components through a gating fusion mechanism, and forms a dynamic weight graph W through gentle weight allocation. For example... Figures 8 to 10 As shown on the right, the fused weighted graph W can balance the stability of static geometry with the temporal changes in dynamic capabilities at different time sections, demonstrating the migration and strength evolution of connectivity paths over time. This fusion result preserves both the constraints of static geological structure on connectivity and reflects the real-time state of dynamic fluid displacement, thus generating a time-varying connectivity weighted graph that can be directly used for path search and optimization.

[0092] Optionally, the design idea of ​​the reinforcement learning-based well connectivity path search algorithm in this embodiment is as follows:

[0093] The A* algorithm is a heuristic search algorithm capable of quickly and accurately finding the shortest path from multiple paths in a given 3D space. In a 3D reservoir space search environment fused with dynamic permeability and static multi-attribute seismic data, the A* algorithm uses a heuristic function to estimate the distance from any node to the target node, thereby improving search efficiency. Its underlying logic is to comprehensively evaluate the total cost from the starting point to the endpoint through an evaluation function, achieving the optimal search path from the injection well (starting point) to the endpoint (production well). In marine sandstone reservoirs, reservoir permeability changes in real time during development, and the seepage space is complex and variable. This algorithm can effectively adapt to these characteristics, enabling the prediction of the evolution of inter-well connectivity paths in marine sandstone reservoirs.

[0094] Building upon deep residual networks, this embodiment adds a cross-modal attention mechanism to enable deep feature interaction between dynamic permeability data and static seismic multi-attribute data. Spatial features are extracted by a 3D convolutional network, while temporal dependent features are modeled using Bi-LSTM or TCN, outputting a dynamic weighted connectivity map that includes not only the geometry of connected paths but also reflects flow capacity weights.

[0095] The key to the A* optimal path search algorithm lies in establishing an evaluation function to define the path cost. This cost consists of two parts: the actual cost and the estimated cost, as shown in formula (1):

[0096] (1)

[0097] in, Let $C$ be the total cost of the path. For reality, the actual cost from the starting point to the current position is the cumulative distance of the traversed path, reflecting the actual consumption of path search; It stands for forecast, which represents the estimated cost from the current location to the destination, usually calculated using Euclidean distance.

[0098] Estimated cost The calculation method is formula (2):

[0099] (2)

[0100] in , and These are the differences between the coordinates of the starting point and the ending point, respectively.

[0101] In this embodiment of the invention, a path search cost function is first defined on the fused dynamic weight graph Wt. The cost consists of the cumulative cost of the already traveled paths and the estimated cost of the production wells yet to be reached, ensuring the optimality of the path in terms of geometry and flow capacity. Then, the A* algorithm is used to search on W(t=0), as follows: Figure 11 As shown, an optimal path is formed from the injection well to the production well, traversing the high-weight region. This path exhibits convergence along the potential channel, indicating that the fusion of static geometry and dynamic capabilities can correctly guide path generation. The illustrative expression for the cost function is: .

[0102] During the time-series progression, this invention recalculates the optimal path at each time point through a rolling replanning mechanism. For example... Figure 12 As shown, when time progresses to t=10, although the overall weight distribution is similar to that at t=0, due to the evolution of the dynamic capability components, the local path selection undergoes a slight shift, and the path automatically adjusts within the high-weight segment. This process avoids the problem of path failure caused by the solidification of solutions at a single moment, and realizes channel tracking and path updating that evolves over time.

[0103] As time progresses further to t=19, as Figure 13 As shown, while the optimal path maintains a consistent overall trend, new migrations and detours occur in local segments, reflecting the migration patterns of seepage channels under dynamic reservoir changes. By performing A* search and dynamic replanning on each Wt period, this invention can continuously output the optimal interconnection path sequence between wells and explicitly provide the migration characteristics and key locations of the path over time, providing a reliable basis for optimizing subsequent development measures.

[0104] Furthermore, the multi-objective optimization method for inter-well connectivity paths in this embodiment is as follows:

[0105] Marine sandstone reservoirs exhibit dramatic particle release and migration, significant variations in reservoir permeability, and strong heterogeneity. Therefore, connectivity paths obtained solely through path search algorithms may not be the truly optimal paths and require further optimization. Inter-well connectivity path optimization is essentially a multi-objective, multi-constraint optimization problem. This embodiment considers two indicators—shortest connectivity path and maximum flow volume—for connectivity path optimization.

[0106] (1) Optimization objective based on shortest connected path :

[0107]

[0108] in, Indicates the number of path nodes; , and These are the coordinates of the status nodes along the path.

[0109] (2) Optimization objective based on maximum flow volume :

[0110]

[0111] in, The number of cave nodes; This represents the state of a cave within the path.

[0112] To achieve multi-objective optimization, this embodiment employs the NSGA-II algorithm to solve the dual-objective problem of "shortest path + maximum flow volume," and introduces a lower limit constraint on penetration rate and a flow bottleneck constraint. The output is a set of Pareto optimal solutions, allowing engineers to select the optimal path under different operating conditions.

[0113] Furthermore, the steps of the multi-objective trade-off (Pareto front) in this embodiment are as follows:

[0114] (1) Scan α / β (prefer static vs. dynamic) to obtain multiple candidate paths;

[0115] (2) Calculate the length L and cumulative flow F of each path (integrate over W);

[0116] (3) Perform non-dominated screening to obtain the Pareto solution set.

[0117] In this embodiment of the invention, to simultaneously consider the geometric optimality of the path and the engineering value of its flow capacity, static and dynamic preference factors are set up, and multiple candidate paths are generated by scanning different weight combinations. For each candidate path, its path length and cumulative flow on the fused weight graph are calculated, and both are used as dual-objective indicators for evaluation. The path length reflects the simplicity of geometric connectivity, while the cumulative flow reflects the path's advantage in seepage capacity.

[0118] After obtaining the indices of all candidate paths, a non-dominated ranking method is used for screening, removing inferior solutions that are simultaneously inferior to other paths, and retaining the effective paths that make up the Pareto solution set. The resulting Pareto front is as follows: Figure 14 As shown, this illustrates the trade-off boundary between path length and cumulative flow. This frontier allows engineers to choose between shorter paths or higher flow rates based on actual needs, achieving multi-objective optimization and decision support for connectivity paths.

[0119] As an example, the steps for implementing multi-objective optimization and decision support for connected paths in this embodiment are as follows:

[0120] In this embodiment, within a selected time segment, the ratios of "static preference factors" and "dynamic preference factors" are scanned, with values ​​ranging from 0 to 1 and a step size of 0.1, forming a set of ratio sequences, such as 1:0, 0.8:0.2, 0.6:0.4, 0.4:0.6, 0.2:0.8, and 0:1. The static preference factor refers to the numerical weight emphasizing the contribution of static geological and structural information in the fused weight map, while the dynamic preference factor refers to the numerical weight emphasizing the contribution of dynamic information such as pressure, water content, and particle migration. For each ratio, a corresponding weight field is generated using the fused weight map, and a heuristic path search algorithm is used to find the optimal connectivity path between the injection well and the production well, resulting in a set of candidate paths. Two indicators are calculated for each candidate path: path length and cumulative flow. Path length refers to the cumulative step size of the candidate path in the grid space, reflecting the geometric simplicity; cumulative flow refers to the total value obtained by summing the values ​​point-by-point along the path on the fused weight map, reflecting the overall flow capacity of the path. For example, when the ratio is 0.6:0.4, the resulting path length is approximately 78 to 79, and the cumulative traffic is approximately 47 to 48; when the ratio is 0.2:0.8, the path length is approximately 78, and the cumulative traffic is approximately 49.

[0121] This embodiment performs non-dominated screening on all candidate paths. Non-dominated screening means that under multi-index evaluation, if a path is not inferior to another path in both path length and cumulative flow, and is strictly superior in at least one of them, then the path is retained as a valid solution; if a path is simultaneously superior to another path, then it is eliminated. After screening, a set of several valid solutions is obtained, forming a trade-off boundary, i.e., the frontier, in the index space. The frontier is the boundary line formed by all valid solutions in the index space, characterizing the optimal trade-off between different objectives. Taking one experiment as an example, a total of 20 candidate paths were generated. After comparison and elimination, 13 valid solutions were finally retained. These valid solutions are arranged sequentially on the plane, with path lengths concentrated around 78, and cumulative flow gradually increasing from 45 to 50, clearly showing the trade-off between "shorter" and "more efficient".

[0122] This embodiment sets selection rules based on different engineering preferences and selects specific solutions from the leading edge. For example, when the engineering objective is to prioritize ensuring production capacity, a threshold of "cumulative flow rate not less than 48" is set, along with a constraint of "path length not exceeding 78.5". The path with the largest cumulative flow rate among the leading edge points that meet these conditions is selected as the recommended solution. When the engineering objective is to prioritize reducing pressure drop or construction difficulty, "path length not exceeding 78" is set as the primary rule, and the path with the highest cumulative flow rate among the leading edge points that meet this condition is selected as the recommended solution. The finally selected path outputs its coordinate sequence on the grid, the key high-weight sections it passes through, the percentage gain compared to adjacent candidate paths, and provides adjustment suggestions for the injection-production well network and completion measures, realizing the direct conversion from multi-objective optimization results to engineering decision instructions.

[0123] like Figure 15 As shown, corresponding to the above method, this embodiment also provides an intelligent prediction system for the dynamic evolution of inter-well connectivity paths in reservoirs, including:

[0124] The data preprocessing unit is used to collect static seismic multi-attribute data, initial permeability data, and production dynamic data. It completes noise processing, missing value processing, outlier processing, and duplicate value processing. It also extracts multi-scale spatial features through an encoder that combines convolution and downsampling with a residual structure.

[0125] The static prior construction unit is used to construct a static connectivity prior for simulating river channels and corridors based on the multiple attributes of static earthquakes. It takes the initial permeability and the time series reflecting changes in formation pressure, water content, particle release and migration intensity as model inputs and associates them with the static connectivity prior.

[0126] The permeability evolution unit is used to establish the permeability evolution relationship starting from the initial permeability and driven by the combined effects of formation pressure, water aquifer and particle migration. It adopts regression and recursive fitting and introduces online updates to obtain the permeability field of each period over time.

[0127] A cross-modal fusion unit is used to introduce a cross-modal attention mechanism into the deep residual network. Guided by the multi-scale spatial features, the phase-by-phase permeability field is fused with the static seismic multi-attribute to generate a dynamic weighted connectivity graph that characterizes the weights of geometric shape and flow capacity.

[0128] The path search unit is used to find the optimal connectivity path from the injection well to the production well on the dynamic weighted connectivity graph in each period based on the heuristic path search algorithm, and to perform replanning on a rolling basis over time to obtain the path migration pattern and the location of key channels.

[0129] The multi-objective optimization unit is used to optimize the length of the connected path and the cumulative flow capacity. It introduces the lower limit constraint of permeability and the flow bottleneck constraint, and uses a non-dominated sorting multi-objective evolutionary algorithm to solve the problem and obtain a non-dominated solution set. The non-dominated solution set is the set of Pareto optimal solutions, and the non-dominated solution set is represented as the Pareto front in the objective space.

[0130] The results output unit is used to select a connectivity path scheme in the non-dominated solution set according to engineering preferences, and output the spatiotemporal evolution results of the inter-well connectivity path and corresponding adjustment suggestions to guide reservoir development.

[0131] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0132] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for intelligent prediction of dynamic evolution of inter-well connectivity paths in oil reservoirs, characterized in that, include: It gathers static seismic multi-attribute data, initial permeability data, and production dynamic data, completes noise processing, missing value processing, outlier processing, and duplicate value processing, and extracts multi-scale spatial features through an encoder that combines convolution and downsampling with residual structure. Based on the static seismic multi-attribute model, a static connectivity prior for simulating river channels and corridors is constructed. The initial permeability and the time series reflecting changes in formation pressure, water content, particle release and migration intensity are used as model inputs and associated with the static connectivity prior. A permeability evolution relationship was established, starting from the initial permeability and driven by the combined effects of formation pressure, water aquifer and particle migration. Regression and recursive fitting were used and online updates were introduced to obtain the permeability field over time. A cross-modal attention mechanism is introduced into the deep residual network. Guided by the multi-scale spatial features, the phase-by-phase permeability field is fused with the static seismic multi-attribute to generate a dynamic weighted connectivity graph that characterizes the weights of geometric shape and flow capacity. Based on a heuristic path search algorithm, the optimal connectivity path is obtained from the injection well to the production well on the dynamic weighted connectivity graph in each period, and replanning is performed on a rolling basis over time to obtain the path migration pattern and the location of key channels. With the optimization objectives of connected path length and cumulative flow capacity, a permeability lower limit constraint and a flow bottleneck constraint are introduced. A multi-objective evolutionary algorithm with non-dominated sorting is used to solve the problem and obtain a non-dominated solution set. The non-dominated solution set is the set of Pareto optimal solutions, and the non-dominated solution set is represented as the Pareto front in the objective space. Based on engineering preferences, a connectivity path scheme is selected from the non-dominated solution set, and the spatiotemporal evolution results of the inter-well connectivity path and corresponding adjustment suggestions are output to guide reservoir development.

2. The intelligent prediction method for dynamic evolution of inter-well connectivity paths in reservoirs according to claim 1, characterized in that, The encoder is formed by a multi-scale structure consisting of convolution, downsampling, and residual connections. The encoder is used to extract the multi-scale spatial features and serve as the feature basis for cross-modal fusion and path evaluation.

3. The intelligent prediction method for dynamic evolution of inter-well connectivity paths in reservoirs according to claim 1, characterized in that, The time dependence of the time series is modeled by a sequence modeling unit, which includes a bidirectional long short-term memory network or a temporal convolutional network, to characterize the changes in production dynamics over time.

4. The intelligent prediction method for dynamic evolution of inter-well connectivity paths in reservoirs according to claim 1, characterized in that, The parameter update of the permeability evolution relationship adopts recursive least squares and introduces an online update mechanism. The estimation results of the model parameters in the permeability evolution relationship in the previous period are jointly corrected with the current period's production dynamics to obtain the period-by-period permeability field over time.

5. The intelligent prediction method for dynamic evolution of inter-well connectivity paths in reservoirs according to claim 1, characterized in that, The dynamic weighted connectivity graph simultaneously represents the geometric shape and flow capacity weight of the connected paths, and serves as the weighted graph for path search and the evaluation benchmark for multi-objective optimization.

6. The intelligent prediction method for dynamic evolution of inter-well connectivity paths in reservoirs according to claim 1, characterized in that, The evaluation function of the heuristic path search algorithm consists of the traveled cost and the remaining distance estimate. The node expansion order is determined according to the evaluation value from smallest to largest, and rolling replanning is performed in periodic units. In each period, the optimal connectivity path is obtained from the injection well to the production well on the dynamic weighted connectivity graph.

7. The intelligent prediction method for dynamic evolution of inter-well connectivity paths in reservoirs according to claim 6, characterized in that, In the path search phase based on the heuristic path search algorithm, multiple candidate paths are generated by setting weight preferences for static connectivity priors and dynamic flow capabilities in the evaluation function to reflect different emphases on static geometry and dynamic capabilities.

8. The intelligent prediction method for dynamic evolution of inter-well connectivity paths in reservoirs according to claim 5, characterized in that, The multi-objective optimization takes the length of the connected path and the cumulative flow capacity as the optimization objectives, sets the lower limit constraint of permeability and the flow bottleneck constraint, obtains the non-dominated solution set through non-dominated screening, and selects the path scheme from the non-dominated solution set for reservoir development adjustment according to engineering preferences.

9. The intelligent prediction method for dynamic evolution of inter-well connectivity paths in reservoirs according to claim 1, characterized in that, The heuristic path search algorithm is the A* optimal path search algorithm.

10. A smart prediction system for the dynamic evolution of inter-well connectivity paths in an oil reservoir, characterized in that, include: The data preprocessing unit is used to collect static seismic multi-attribute data, initial permeability data, and production dynamic data. It completes noise processing, missing value processing, outlier processing, and duplicate value processing. It also extracts multi-scale spatial features through an encoder that combines convolution and downsampling with a residual structure. The static prior construction unit is used to construct a static connectivity prior for simulating river channels and corridors based on the multiple attributes of static earthquakes. It takes the initial permeability and the time series reflecting changes in formation pressure, water content, particle release and migration intensity as model inputs and associates them with the static connectivity prior. The permeability evolution unit is used to establish the permeability evolution relationship starting from the initial permeability and driven by the combined effects of formation pressure, water aquifer and particle migration. It adopts regression and recursive fitting and introduces online updates to obtain the permeability field of each period over time. A cross-modal fusion unit is used to introduce a cross-modal attention mechanism into the deep residual network. Guided by the multi-scale spatial features, the phase-by-phase permeability field is fused with the static seismic multi-attribute to generate a dynamic weighted connectivity graph that characterizes the weights of geometric shape and flow capacity. The path search unit is used to find the optimal connectivity path from the injection well to the production well on the dynamic weighted connectivity graph in each period based on the heuristic path search algorithm, and to perform replanning on a rolling basis over time to obtain the path migration pattern and the location of key channels. The multi-objective optimization unit is used to optimize the length of the connected path and the cumulative flow capacity. It introduces the lower limit constraint of permeability and the flow bottleneck constraint, and uses a non-dominated sorting multi-objective evolutionary algorithm to solve the problem and obtain a non-dominated solution set. The non-dominated solution set is the set of Pareto optimal solutions, and the non-dominated solution set is represented as the Pareto front in the objective space. The results output unit is used to select a connectivity path scheme in the non-dominated solution set according to engineering preferences, and output the spatiotemporal evolution results of the inter-well connectivity path and corresponding adjustment suggestions to guide reservoir development.

Images