Intelligent injection-production control optimization method and system for gas storage based on reinforcement learning

Abstract

The application discloses a gas storage intelligent injection-production control optimization method and system based on reinforcement learning, relates to the technical field of gas storage intelligent injection-production, and comprises the following steps: determining an initial well group maximum spatial span based on inter-well pressure disturbance decay characteristics; performing local curvature decomposition and discretization on the span to generate multiple groups of candidate well group maximum spatial spans; mapping the well group maximum spatial span to injection-production collaborative control targets such as well group connectivity structure and well number upper limit, and respectively applying the injection-production collaborative control targets to gas storage injection-production; extracting features of injection-production processes under different span conditions and obtaining span scores through reinforcement learning; and analyzing and screening the best well group maximum spatial span based on the span score curve and using the best well group maximum spatial span for gas storage injection-production control.

Description

Technical Field

[0001] This invention relates to the field of intelligent gas injection and production technology for gas storage facilities, and more specifically, to a method and system for intelligent injection and production control optimization of gas storage facilities based on reinforcement learning. Background Technology

[0002] With the increasing demands for large-scale operation and refined peak-shaving scheduling of natural gas storage facilities, the injection and production control of existing storage facilities is gradually evolving from manual, experience-based scheduling to intelligent and model-based approaches. Current technologies typically adjust the injection and production rhythm based on operating parameters such as bottom-hole pressure, injection / production volume, and storage capacity, combined with numerical simulations, rule bases, or optimization algorithms, to achieve safe and stable operation. However, existing intelligent injection and production technologies mostly focus on single wells or static well groups as control objects, failing to adequately characterize the dynamic coupling relationships between wells, making it difficult to achieve coordinated optimization control at the well group level under complex operating conditions.

[0003] In the coordinated operation of multiple wells in a gas storage facility, the maximum spatial span of the well group is a crucial parameter describing the spatial relationships between wells within the group. This parameter determines which injection and production wells are included in the same coordinated control unit. The maximum spatial span of the well group directly affects the propagation range of pressure disturbances between wells, the synchronization of injection and production responses, and the evolution characteristics of the pressure field within the well group. It serves as a key bridge connecting the physical coupling between wells and the injection and production scheduling strategy. In existing technologies, the maximum spatial span of the well group often exists as an implicit parameter, and its setting method directly determines the well group division results and the effectiveness of subsequent injection and production coordinated control.

[0004] There is no single optimal value for the maximum spatial span of a well cluster; both excessively large and small spans can negatively impact intelligent injection and production in gas storage facilities. When this span varies, the inter-well pressure coupling strength, injection and production load distribution, and command execution consistency within the well cluster all change significantly, affecting the stability, efficiency, and safety of the injection and production process. Therefore, the maximum spatial span of a well cluster essentially constitutes a control factor that needs to be weighed, systematically influencing the effectiveness of intelligent injection and production in gas storage facilities.

[0005] When the maximum spatial span of a well group is too low, the well group tends to be fragmented, limiting the number of wells within a single group and thus reducing the scope of injection-production coordination. Under these conditions, the propagation of pressure disturbances between wells within the group is restricted, making it difficult to effectively coordinate injection and production commands, and easily leading to inconsistent injection and production rhythms between well groups. Simultaneously, an excessively small well group size increases the number of scheduling units, raises control complexity, and weakens the overall coordination capability of injection and production scheduling, thereby reducing the overall operational efficiency of the gas storage facility.

[0006] When the maximum spatial span of a well group is too high, the number of injection and production wells within the group increases significantly, the spatial distance between wells widens, and the pressure coupling relationship between wells tends to weaken. Under these conditions, the response differences of different wells to the same injection and production command increase, easily leading to pressure imbalances within the well group and uneven distribution of injection and production loads. Simultaneously, large-span well groups are more prone to local anomalies being masked by the overall situation during injection and production, increasing the risk of bottom hole pressure anomalies and command execution mismatch, thus affecting the safety and controllability of the injection and production process.

[0007] In existing intelligent injection and production technologies for gas storage facilities, the setting of the maximum spatial span of well groups largely relies on empirical rules, fixed distance thresholds, or static partitioning strategies. These technologies lack dynamic evaluation and optimization mechanisms based on actual injection and production response effects, making it difficult to simultaneously consider inter-well pressure coordination and injection and production execution stability. This control method, which lacks systematic consideration and adaptive adjustment of the maximum spatial span of well groups, easily leads to insufficient injection and production efficiency, increased pressure fluctuations, and unstable intelligent scheduling effects, thus hindering further improvements in intelligent injection and production technologies for gas storage facilities.

[0008] To address the above problems, this invention proposes a solution. Summary of the Invention

[0009] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for optimizing intelligent injection and production control of gas storage facilities based on reinforcement learning. By quantitatively constructing, controllably mapping, and using reinforcement learning scoring to screen the maximum spatial span of the well group based on the attenuation characteristics of inter-well pressure disturbances, the present invention addresses the problems in existing intelligent injection and production systems for gas storage facilities, such as the reliance on empirically set spatial scales for well groups, difficulty in balancing inter-well interference and insufficient coordination, and resulting in insufficient stability and coordination in injection and production execution.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] The intelligent injection-production control optimization method for gas storage based on reinforcement learning includes the following steps: obtaining the initial maximum spatial span of the well group based on inter-well pressure disturbance attenuation analysis; discretizing the initial maximum spatial span of the well group based on the local curvature decomposition of the inter-well pressure disturbance attenuation curve to generate several second maximum spatial spans of the well group to be screened; constructing a connected subgraph of the well group based on the spatial coverage relationship between the wells, mapping the maximum spatial span of the well group to the injection-production coordinated execution control target, wherein the injection-production coordinated execution control target includes the upper limit combination of the number of wells corresponding to the maximum spatial span of the well group; mapping the maximum spatial span of the second well group and applying it to the gas storage injection and production; extracting features from the injection and production processes of different gas storages and training a score based on reinforcement learning to generate a span score; performing curve analysis based on the span score and the corresponding maximum spatial span of the second well group, screening out the optimal maximum spatial span of the second well group, and performing gas storage injection and production.

[0012] The intelligent injection and production control optimization system for gas storage based on reinforcement learning includes a well group maximum spatial span initialization module, a well group maximum spatial span discretization construction module, a well number limit combination acquisition module, a gas storage injection and production module, a span scoring module, and a gas storage injection and production optimization module. The well group maximum spatial span initialization module is used to obtain the initial maximum spatial span of the well group based on inter-well pressure disturbance attenuation analysis. The well group maximum spatial span discretization construction module is used to discretize the initial maximum spatial span of the well group based on the local curvature decomposition of the inter-well pressure disturbance attenuation curve, generating several second maximum spatial spans of the well group to be screened. The well number limit combination acquisition module is used to obtain the maximum spatial span of the well group based on the well number attenuation curve. The system constructs a well cluster connected subgraph representing the spatial coverage relationship between wells, mapping the maximum spatial span of the well cluster to the injection-production coordinated execution control objective. This objective includes an upper limit combination of the number of wells corresponding to the maximum spatial span of the well cluster. A gas storage injection-production module maps the maximum spatial span of the second well cluster and applies it to gas storage injection-production. A span scoring module extracts features from different gas storage injection-production processes and trains a score based on reinforcement learning to generate a span score. A gas storage injection-production optimization module performs curve analysis based on the span score and the corresponding maximum spatial span of the second well cluster, selecting the optimal maximum spatial span of the second well cluster for gas storage injection-production.

[0013] The technical effects and advantages of the intelligent injection and production control optimization method and system for gas storage based on reinforcement learning in this invention are as follows:

[0014] 1. This invention constructs a customized acquisition and discretization mechanism for the maximum spatial span of a well group based on inter-well pressure disturbance attenuation analysis and local curvature decomposition of attenuation curves. This allows the spatial scale of the well group to no longer rely on empirical settings or static rules, but instead directly originates from the physical correlation characteristics of inter-well pressure response during the actual injection and production process of the gas storage facility. By further mapping the maximum spatial span of the well group into directly executable injection and production collaborative control objectives such as the well group connectivity subgraph and the upper limit of the number of wells, a quantitative linkage of "spatial scale—well group structure—injection and production execution" is achieved. This transforms the maximum spatial span of the well group from an abstract parameter into a practically controllable and applicable injection and production control constraint, significantly improving the adaptability and relevance of well group division and collaborative injection and production strategies to the formation response characteristics of the gas storage facility.

[0015] 2. This invention applies the maximum spatial span of different well groups to the actual gas storage injection and production process. It constructs feature vectors for the injection and production process based on characteristics highly correlated with the spatial scale of the well group, such as pressure coordination, pressure imbalance, abnormal fluctuations, injection and production load dispersion, and execution mismatch. Reinforcement learning is introduced to comprehensively score and train the injection and production execution performance under various span conditions, thereby avoiding the one-sided judgment of the merits of the well group's spatial span based on a single indicator or static threshold. Through customized analysis of the span scoring curve, the optimal maximum spatial span of the well group is automatically selected, balancing sufficient pressure coupling with well group independence. This enables the gas storage injection and production control to achieve a dynamic balance between suppressing excessive inter-well interference and avoiding insufficient coordination, improving the stability, consistency, and overall execution efficiency of injection and production scheduling. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the intelligent injection and production control optimization method for gas storage based on reinforcement learning, as described in this invention.

[0017] Figure 2 This is a schematic diagram of the intelligent injection and production control optimization system for gas storage based on reinforcement learning, as described in this invention. Detailed Implementation

[0018] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1, Figure 1 The present invention provides an intelligent injection and production control optimization method for gas storage facilities based on reinforcement learning, comprising the following steps:

[0020] S1, the maximum spatial span of the initial well group is obtained based on the inter-well pressure disturbance attenuation analysis;

[0021] In this embodiment, obtaining the initial maximum spatial span of the well group based on inter-well pressure disturbance attenuation analysis specifically involves:

[0022] The historical injection and production records of multiple injection and production wells in the gas storage facility are obtained, the target wells where injection and production volume changes are identified, and the bottom hole pressure change sequence of the target wells before and after the injection and production change time is extracted.

[0023] Centered on the target well, adjacent wells are selected from near to far according to the spatial distance between them. The bottom pressure change sequence of each adjacent well within the same time window as the target well is extracted, and the inter-well pressure response sequence corresponding to the target well and each adjacent well is constructed.

[0024] Calculate the inter-well spatial distance corresponding to each inter-well pressure response sequence, and sort the inter-well pressure response sequences in ascending order of inter-well spatial distance to generate an inter-well spatial distance sequence;

[0025] In the inter-well spatial distance sequence, the maximum pressure response amplitude of each adjacent well pressure change sequence within the target well injection and production change time window is calculated, and the maximum pressure response amplitude is associated with the corresponding inter-well spatial distance to generate an inter-well spatial distance-pressure response amplitude sequence.

[0026] Using the inter-well spatial distance as the independent variable and the pressure response amplitude as the dependent variable, a decay sequence of pressure response amplitude as a function of inter-well spatial distance is constructed, mapped to a curve, and the rate of change of pressure response amplitude between adjacent distance points is calculated.

[0027] The pressure response attenuation inflection point is determined at the position where the rate of change of the pressure response amplitude changes from a gradual change to a continuous decrease, and the inter-well space distance corresponding to the inflection point is extracted.

[0028] The inter-well spatial distance corresponding to the inflection point is determined as the maximum spatial scale in which there is still a significant pressure response correlation between wells within the well group, and is used as the initial maximum spatial span of the well group for well group division.

[0029] In this embodiment, the historical injection and production record refers to the collection of injection and production status data of each injection and production well, recorded chronologically during the long-term operation of the gas storage facility. This record includes at least information on the changes in injection and production conditions of each well at different points in time, as well as bottom hole pressure data collected synchronously. The historical injection and production record provides the basic data source for subsequent identification of injection and production volume change events and extraction of pressure response sequences; its temporal continuity ensures that the pressure disturbance propagation process can be completely captured.

[0030] In this embodiment, the target wells experiencing changes in injection and production volume refer to those wells that exhibit significant adjustments to their injection and production conditions in historical injection and production records. These adjustments include increases or decreases in injection and production volume or switching of injection and production states triggered by the injection and production condition control system. By identifying such target wells, it can be ensured that the pressure changes upon which subsequent analyses are based are not random fluctuations, but rather caused by specific injection and production disturbances, thereby making the pressure responses between wells comparable and traceable.

[0031] In this embodiment, the bottom hole pressure change sequence refers to a data sequence of bottom hole pressure changes over time, continuously collected within a preset time window around the moment of change in the injection and production volume of the target well. This sequence is used to characterize the direct pressure response of injection and production disturbances at the single-well scale. Its time structure reflects the amplitude characteristics and trend of pressure changes, and serves as the basis for subsequently constructing the inter-well pressure response relationship.

[0032] In this embodiment, adjacent wells refer to other injection and production wells that are spatially distant from the target well and whose bottomhole pressure change sequences can be obtained within the same time window. Adjacent wells are not designated based on manual experience, but are selected sequentially according to a rule of sorting wells from closest to furthest in terms of spatial distance, thereby ensuring that the pressure disturbance propagation analysis has clear spatial directionality and hierarchy.

[0033] In this embodiment, the inter-well pressure response sequence refers to the set of well-to-well pressure response data formed by extracting the bottom-hole pressure change sequence of adjacent wells within the same time window, using the injection-production changes of the target well as the disturbance source. This sequence is used to reflect the propagation effect of injection-production disturbances under different inter-well distance conditions, and its variation amplitude and synchronicity directly reflect the degree of inter-well connectivity.

[0034] In this embodiment, the inter-well spatial distance refers to the physical distance between the target well and each adjacent well, calculated based on the spatial location of the wellhead or well trajectory. This distance value is used to establish the correspondence between pressure response and spatial location, and is the spatial scale basis for constructing the pressure disturbance attenuation sequence.

[0035] In this embodiment, the inter-well spatial distance sequence refers to an ordered set of distances formed by arranging the inter-well spatial distances between the target well and each adjacent well in ascending order. This sequence provides a sequential constraint for the analysis of pressure response amplitude changes with spatial distance, giving subsequent attenuation characteristics a basis for continuous and monotonic analysis.

[0036] In this embodiment, the maximum pressure response amplitude refers to the maximum pressure change relative to the starting point of the change in the bottom hole pressure change sequence of adjacent wells within the time window corresponding to the injection-production change of the target well. This amplitude value is used to quantify the strongest response of a single adjacent well to injection-production disturbances, thereby avoiding interference from instantaneous noise in the judgment of pressure response.

[0037] In this embodiment, the well-to-well spatial distance-pressure response amplitude sequence refers to a data sequence formed by correlating the spatial distance between each adjacent well with its corresponding maximum pressure response amplitude. This sequence is used to uniformly describe the mapping relationship between spatial distance changes and pressure response intensity, and serves as the direct data basis for constructing pressure disturbance attenuation curves.

[0038] In this embodiment, the pressure response amplitude decay sequence refers to the overall trend of gradually weakening pressure response amplitude, with the inter-well spatial distance as the direction of change. This decay sequence, mapped as a continuous curve, allows for a comprehensive characterization of the strength variations of pressure disturbances during spatial propagation.

[0039] In this embodiment, the rate of change of pressure response amplitude refers to the characterization of the degree of change in pressure response amplitude between adjacent distance points as the spatial distance between wells increases. This rate of change reflects the rate at which the pressure response weakens with increasing distance, and its trend can reveal the location where a significant turning point occurs in the propagation capability of pressure disturbances.

[0040] In this embodiment, the pressure response attenuation inflection point refers to the position where the rate of change of the pressure response amplitude changes from a relatively flat state to a continuously weakening state. The inter-well spatial distance corresponding to this position reflects the spatial boundary of the structural change in the pressure disturbance propagation capability. This inflection point is not set by a fixed threshold, but is obtained by identifying the changing trend, thus exhibiting adaptability.

[0041] In this embodiment, the initial maximum spatial span of the well group refers to the spatial scale determined by the inter-well spatial distance corresponding to the inflection point of pressure response decay. This scale indicates that significant pressure response correlations still exist between wells within this range. This span serves as the initial scale for well group division, providing a physical constraint basis for subsequent well group structure construction and injection-production synergy analysis.

[0042] It should be noted that the above process of obtaining the initial maximum spatial span of the well group based on the inter-well pressure disturbance attenuation analysis assumes that the propagation of injection and production disturbances within the gas storage facility has temporal continuity and spatial attenuation characteristics. Therefore, by analyzing actual injection and production change events that have occurred in historical injection and production records, it is possible to avoid relying on manually set empirical distances or fixed thresholds, thereby making the obtained initial maximum spatial span of the well group adaptable to the actual geological conditions and operating status of the gas storage facility.

[0043] It should be noted that in the process of constructing the inter-well pressure response sequence and pressure response amplitude decay curve, the same target well may correspond to multiple injection and production change events. By comprehensively processing the pressure decay characteristics formed by multiple injection and production changes, the impact of a single abnormal condition on the initial judgment of the maximum spatial span of the well group can be reduced, thereby improving the stability and reliability of the well group spatial scale determination results.

[0044] S2, based on the local curvature decomposition of the inter-well pressure disturbance attenuation curve, the maximum spatial span of the initial well group is discretized to generate several maximum spatial spans of the second well group to be screened;

[0045] In this embodiment, the local curvature decomposition based on the inter-well pressure disturbance attenuation curve discretizes the initial maximum spatial span of the well group, generating several second maximum spatial spans of the well group to be screened, specifically as follows:

[0046] The location of the inter-well spatial distance corresponding to the maximum spatial span of the initial well group is obtained based on the inter-well spatial distance-pressure response amplitude attenuation curve.

[0047] Using the inter-well spatial distance as the independent variable and the pressure response amplitude as the dependent variable in the inter-well spatial distance-pressure response amplitude decay curve, the curve is subjected to second-order difference calculation to generate a curvature sequence along the direction of inter-well spatial distance.

[0048] In the curvature sequence, with the position corresponding to the maximum spatial span of the initial well group as the center, an interval in which the absolute value of curvature continuously increases is selected in the direction smaller than the maximum spatial span of the initial well group, and an interval in which the absolute value of curvature rapidly decreases is selected in the direction larger than the maximum spatial span of the initial well group, thus constructing an effective curvature change interval;

[0049] Within the effective range of curvature change, the inter-well spatial distance-pressure response amplitude attenuation curve is sampled at equal intervals according to the inter-well spatial distance to obtain multiple discrete inter-well spatial distance nodes;

[0050] For each discrete well-space distance node, calculate its corresponding curvature value, and sort the discrete well-space distance nodes from largest to smallest according to the curvature value;

[0051] The sorted discrete well-to-well spatial distance nodes are used as candidate values ​​for the spatial scale of the well group under different pressure disturbance attenuation characteristics to construct several second-level maximum spatial spans of the well group.

[0052] In this embodiment, the well-to-well spatial distance-pressure response amplitude attenuation curve refers to the continuous curve formed by establishing a mapping relationship between the well-to-well spatial distance and the corresponding maximum pressure response amplitude in S1, with the well-to-well spatial distance as the direction of change and the pressure response amplitude as the response intensity. This curve is used to comprehensively depict the change in the spatial propagation capability of injection-production disturbances from strong to weak, and serves as the basic data object for subsequent curvature analysis and scale discretization.

[0053] In this embodiment, the location of the inter-well spatial distance corresponding to the initial maximum spatial span of the well group refers to the curve position point on the pressure response amplitude decay curve that corresponds to the initial maximum spatial span value of the well group determined in S1. This position point is used to anchor the spatial reference point in the curve where the pressure disturbance transitions from significant correlation to weak correlation, providing a central reference for subsequent local analysis.

[0054] In this embodiment, the second-order difference calculation refers to the re-analysis of the adjacent variation trends of the pressure response amplitude curve along the inter-well spatial distance direction, used to characterize the change in the curvature of the curve with distance. This calculation method, by describing the changes in the variation trend, allows for the quantitative identification of locations in the curve where the pressure response decay rate changes significantly, without the need to introduce manual thresholds.

[0055] In this embodiment, the curvature sequence refers to an ordered data sequence formed by arranging the second-order difference calculation results in order of spatial distance between wells. This sequence is used to describe the bending intensity of the pressure response amplitude decay curve at different spatial distances, and its numerical changes reflect the structural changes in the spatial characteristics of pressure disturbance propagation.

[0056] In this embodiment, the absolute value of curvature refers to the result of retaining only the magnitude of the bending intensity corresponding to each position in the curvature sequence without distinguishing its direction. By using the absolute value of curvature, the degree of significant change in the pressure response decay trend can be uniformly measured, avoiding interference from differences in the direction of local fluctuations in the curve on scale identification.

[0057] In this embodiment, the effective curvature variation range refers to the spatial distance range formed by selecting representative curvature variation segments on both sides of the position corresponding to the maximum spatial span of the initial well group, centered on that position, based on the characteristics of the absolute value change of curvature. The interval smaller than the direction corresponding to the maximum spatial span of the initial well group is used to reflect the variation characteristics of the pressure disturbance still in the significant propagation stage, while the interval larger than the direction corresponding to that span is used to reflect the variation characteristics of the pressure disturbance in the rapid decay stage.

[0058] In this embodiment, the equidistant sampling refers to sampling points along the inter-well spatial distance direction at fixed intervals within the effective range of curvature change. This sampling method transforms the continuous curve into a set of discrete and spatially uniform distance nodes, facilitating subsequent comparative analysis at different spatial scales.

[0059] In this embodiment, the discrete well-to-well spatial distance nodes refer to a set of specific well-to-well spatial distance values ​​obtained through equally spaced sampling. Each node corresponds to a specific location on the attenuation curve. This set of nodes is used to represent candidate scale points for pressure disturbance attenuation characteristics at different spatial scales and is the direct source of candidate values ​​for the subsequent construction of well group spatial span.

[0060] In this embodiment, the curvature value corresponding to a discrete node refers to the curvature intensity value at the node location after mapping each discrete well-to-well spatial distance node back to a curvature sequence. This curvature value is used to quantify the significance of the pressure response attenuation trend change at that spatial scale, thereby distinguishing the structural feature differences between different candidate scales.

[0061] In this embodiment, curvature value sorting refers to arranging all discrete well-to-well spatial distance nodes from largest to smallest based on the curvature value corresponding to the discrete nodes. This sorting allows the spatial scales with the most obvious pressure disturbance attenuation characteristics to be identified first, providing a structured basis for subsequent screening.

[0062] In this embodiment, the second maximum spatial span of the well group refers to a set of spatial span values ​​formed by using the sorted discrete well-to-well spatial distance nodes as candidate values ​​for the well group spatial scale. Each second maximum spatial span of the well group corresponds to a different pressure disturbance attenuation structural feature, which is used to construct the well group structure in subsequent steps and participate in the injection-production effect evaluation.

[0063] It should be noted that in S2, the pressure response amplitude decay curve is decomposed locally by local curvature decomposition. Instead of discretizing the entire curve uniformly, it is expanded locally around the physically meaningful spatial scale of the initial well group maximum spatial span determined in S1. This ensures that the maximum spatial span of the generated second well group is within the effective range of actual pressure disturbance propagation, thus avoiding the generation of candidate scales that do not match the actual operating state of the gas storage.

[0064] It should be noted that by combining the effective range of curvature change with equal-interval sampling, the continuous spatial scale can be transformed into a finite number of discrete candidate values ​​without relying on manual experience in setting the step size. This makes the subsequent injection and sampling process comparison, reinforcement learning training, and span scoring process have controllable computational complexity, while maintaining a full ability to express the attenuation characteristics of pressure disturbance.

[0065] S3, constructing a well group connected subgraph based on the spatial coverage relationship between wells, mapping the maximum spatial span of the well group to the injection-production coordinated execution control target, wherein the injection-production coordinated execution control target includes the upper limit combination of the number of wells corresponding to the maximum spatial span of the well group;

[0066] In this embodiment, the well group connectivity subgraph construction based on the spatial coverage relationship between wells maps the maximum spatial span of the well group to the injection-production coordinated execution control target. The injection-production coordinated execution control target includes the upper limit combination of the number of wells corresponding to the maximum spatial span of the well group, specifically:

[0067] Obtain the wellhead spatial coordinates of all injection and production wells in the gas storage facility, calculate the inter-well spatial distance between any two wells, and construct the inter-well spatial distance matrix;

[0068] For each maximum spatial span of the second well group, well pairs in the inter-well spatial distance matrix that are less than or equal to the maximum spatial span of the second well group are marked as connectable well pairs;

[0069] Using injection and production wells as nodes and the connectable well pairs as undirected edges, a well group connectivity graph corresponding to the maximum spatial span of the second well group is constructed.

[0070] The well group connectivity graph is decomposed into connectivity components to generate several independent well group connectivity subgraphs.

[0071] In each well group connected subgraph, the number of nodes in the subgraph is counted to obtain the upper limit of the number of wells in a single well group under the maximum spatial span of the second well group.

[0072] The maximum spatial span of the second well group is associated with the number of connected subgraphs of the corresponding well group and the upper limit of the number of wells in each connected subgraph of the well group to generate a combination relationship of maximum spatial span of well group and upper limit of number of wells.

[0073] The relationship between the maximum spatial span of the well group and the upper limit of the number of wells is used as the mapping result of the injection-production coordinated execution control objective.

[0074] In this embodiment, the inter-well spatial coverage relationship refers to the mutual coverage or interaction relationship formed between different injection and production wells due to the spatial distance meeting the maximum spatial span constraint of a given well group. This coverage relationship is not directly equivalent to pressure connectivity, but rather is used to characterize the inter-well relationship that allows them to be included in the same well group's collaborative control range on a spatial scale, and is the basic spatial constraint condition for subsequent well group structure construction.

[0075] In this embodiment, the inter-well spatial distance matrix refers to the structured data calculated based on the spatial coordinates of the wellheads of all injection and production wells within the gas storage facility, and stored uniformly in the form of a two-dimensional matrix. This matrix is ​​used to quickly determine whether well pairs meet the spatial connectivity conditions under different maximum spatial spans of well groups, and serves as the original data source for constructing well group connectivity maps.

[0076] In this embodiment, the second maximum spatial span of the well group refers to the spatial scale parameters of multiple candidate well groups generated by discretization based on the pressure disturbance attenuation curve in the previous step. Each second maximum spatial span of the well group serves as an independent spatial threshold for constructing the corresponding well group connectivity structure, and does not represent the final well group scale adopted.

[0077] In this embodiment, a connectable well pair refers to a set of well pairs whose inter-well spatial distance is less than or equal to the maximum spatial span of a given second well group. This label only reflects accessibility on a spatial scale and does not involve injection-production conditions or pressure response strength; it is a basic determination result used for graph structure modeling.

[0078] In this embodiment, the well cluster connectivity graph refers to an undirected graph structure constructed using injection and production wells as graph nodes and connectable well pairs as undirected edges. This connectivity graph reflects the spatial collaborative network structure that may be formed between injection and production wells within a gas storage facility under the constraint of the maximum spatial span of a specific well cluster.

[0079] In this embodiment, connected component decomposition refers to the process of performing structural analysis on the constructed well cluster connectivity graph, dividing the set of nodes in the graph that are reachable from each other through paths into several disconnected subgraphs. Each connected component corresponds to a well cluster unit that can independently implement injection-production coordinated control at the current spatial scale.

[0080] In this embodiment, the well cluster connectivity subgraph refers to a well cluster structure obtained through connectivity component decomposition, where every pair of internal nodes has a spatial connection path and is structurally completely isolated from other subgraphs. This subgraph serves as the basic unit for subsequent injection-production coordinated execution and scale control.

[0081] In this embodiment, the upper limit of the number of wells in a single well group refers to the number of injection and production wells included in the connected subgraph of a single well group under the condition of the maximum spatial span of a certain second well group. This number reflects the maximum number of wells that a single well group can be included in coordinated injection and production control at this spatial scale.

[0082] In this embodiment, the maximum spatial span of a well group and the upper limit of the number of wells are combined to form a mapping result by associating the maximum spatial span of each second well group with the number of connected subgraphs of its corresponding well group and the upper limit of the number of wells in each subgraph. This combination relationship is used to characterize the changing features of the well group organization structure at different spatial scales.

[0083] In this embodiment, the injection-production coordinated execution control objective refers to the set of target parameters formed by using the maximum spatial span of the well group as the control input constraint and the corresponding upper limit combination of the number of wells as the execution boundary condition, which is used to guide the coordinated organization and scale control of the gas storage injection-production process.

[0084] It should be noted that the process of constructing a well group connected subgraph based on the spatial coverage relationship between wells in this embodiment is a priori organization of the injection and production wells of the gas storage facility at the spatial structure level. Its purpose is to transform the continuous spatial scale problem into a discrete well group structure problem, thereby providing operable structured input conditions for the subsequent multi-scheme comparison, reinforcement learning scoring and optimal scale selection in the injection and production process.

[0085] It should be noted that the above-mentioned combination relationship between the maximum spatial span of the well group and the upper limit of the number of wells is not directly used as the final injection and production control scheme, but as a constraint and evaluation reference condition in the process of coordinated injection and production. It is combined with the span score results obtained by subsequent feature extraction and reinforcement learning scoring of the injection and production process to ultimately achieve the optimal selection of the spatial scale of the well group and the adaptive optimization of the gas storage injection and production strategy.

[0086] S4, respectively, maps the maximum spatial span of the second well group and applies it to gas storage injection and production;

[0087] In this embodiment, the mapping of the maximum spatial span of the second well group and its application to gas storage injection and production specifically involves:

[0088] Based on the maximum spatial span of each second well group, read its corresponding well group connected subgraph and the upper limit of the number of wells in the well group connected subgraph;

[0089] Within each well group connected subgraph, the injection and production wells within the well group are sorted according to the spatial location of the wellhead and the historical injection and production role identifier, and the injection and production wells not exceeding the upper limit of the number of wells are selected as the set of coordinated injection and production wells of the well group under the current span conditions.

[0090] For each injection and production well in the coordinated injection and production well set, its historical injection and production operating condition type is inherited, and it is determined as a gas injection well or a gas production well, respectively, and the roles of injection and production wells are kept from conflicting within the same well group;

[0091] Using the coordinated injection and production well set as the execution unit, the overall injection and production plan of the gas storage is divided into well groups to generate a well group-level injection and production execution plan corresponding to the maximum spatial span of the second well group;

[0092] During the injection and production process, injection and production instructions are issued synchronously according to the well group-level injection and production execution plan, and the coordinated start and stop status of injection and production wells within the well group is maintained during the injection and production process;

[0093] Record the bottom hole pressure changes, injection and production volume changes, and well group-level execution status of each well group throughout the entire injection and production cycle under the maximum spatial span condition of the second well group, forming gas storage injection and production process data under the corresponding span condition.

[0094] In this embodiment, the maximum spatial span of the second well group refers to the spatial scale parameters of multiple candidate well groups obtained by discretization based on the inter-well pressure disturbance attenuation curve in the previous step. Each maximum spatial span of the second well group corresponds to a set of determined well group connectivity subgraph structure and upper limit constraints on the number of wells, which are applied as independent injection and production execution mapping conditions in this step.

[0095] In this embodiment, the well group connectivity subgraph refers to a well group structural unit constructed from the spatial coverage relationship between wells and decomposed into connectivity components under the condition of a specific second well group's maximum spatial span. The injection and production wells within this subgraph satisfy the connectivity conditions for coordinated injection and production on a spatial scale, and are the basic units for splitting and organizing the injection and production execution plan.

[0096] In this embodiment, the upper limit on the number of wells refers to the maximum number of injection and production wells allowed to be included in the collaborative injection and production execution within a single well group connected subgraph, under the condition of the maximum spatial span of the corresponding second well group. This upper limit on the number of wells is derived from the node statistics of the well group connected subgraph and is used to constrain the actual scale of the number of wells participating in collaborative injection and production, so as to avoid excessive concentration of well groups.

[0097] In this embodiment, the wellhead spatial location refers to the wellhead coordinates of the injection and production wells in the gas storage plane or three-dimensional space, used to characterize the spatial relative positional relationship of injection and production wells within the well group. In this step, this spatial location information is used to sort the injection and production wells within the well group to ensure spatial consistency and controllability in the selection of wells within the well group.

[0098] In this embodiment, the historical injection-production role identifier refers to the role attribute formed by marking the long-term gas injection or gas production functions undertaken by each injection-production well based on historical operational data. This identifier reflects the main functional positioning of the well in past injection-production cycles and is used to maintain the continuity of operational logic when determining the allocation of injection-production roles within the well group.

[0099] In this embodiment, the collaborative injection-production well set refers to a set of injection-production wells selected from a single well group's connected subgraph, ordered by the spatial location of the wellheads and combined with historical injection-production role identifiers, and whose number does not exceed the upper limit of the number of wells. This set constitutes the well group members that actually participate in the collaborative injection-production execution under the current maximum spatial span of the second well group.

[0100] In this embodiment, the historical injection-production operating condition type refers to the stable operating condition attributes formed by injection and production wells during their historical operation, including their operating type as either injection or production wells. In this step, this operating condition type is directly inherited to avoid operational instability caused by frequently switching the injection and production functions of wells within the same well group.

[0101] In this embodiment, the roles of injection and production wells do not conflict. This means that within the connected subgraph of the same well group, the injection and production wells in the coordinated injection and production well set maintain the functional distinction between injection wells and production wells in terms of role allocation. There is no situation where the same well is given contradictory injection and production instructions at the same time, thereby ensuring the feasibility of coordinated execution of the well group.

[0102] The well-group-level injection and production execution scheme in this embodiment refers to an injection and production control scheme formed by breaking down the overall injection and production plan of the gas storage facility according to the well group structure, with the coordinated injection and production well set as the execution unit. This scheme clarifies the allocation and execution sequence of injection and production tasks for each well group under the condition of the maximum spatial span of the current second well group.

[0103] In this embodiment, the coordinated start-stop state refers to the operational state in which each injection and production well within the well group is uniformly controlled for start-stop according to the well group-level injection and production execution plan during the injection and production process. This state is used to ensure the coordination of injection and production wells within the well group on a time scale and to avoid unexpected disturbances to the inter-well pressure field caused by independent start-stop of a single well.

[0104] In this embodiment, the gas storage injection and production process data refers to the bottom hole pressure changes, injection and production volume changes, and well group-level execution status information collected and recorded throughout the entire injection and production cycle under the condition of the maximum spatial span of a specific second well group. This process data serves as the source of raw input data for subsequent feature extraction and reinforcement learning scoring training.

[0105] It should be noted that in this embodiment, the maximum spatial span of the second well group is mapped and applied to the gas storage injection and production process, not to directly determine the final operation plan, but to construct a sample set of injection and production processes that can be used for unified evaluation by executing well group-level injection and production processes with consistent structure and comparable parameters under multiple spatial scale conditions.

[0106] It should be noted that by maintaining the inheritance of injection and production roles and the consistency of well group organization under different maximum spatial span conditions of the second well group, and only changing the spatial scale of the well group and the upper limit of the number of wells, the influence of spatial scale factors on the stability and coordination of the injection and production process can be effectively isolated, providing a clear and comparable decision-making environment for subsequent span scoring based on reinforcement learning.

[0107] S5 extracts features from the injection and production processes of different gas storage facilities and trains scores based on reinforcement learning to generate a span score.

[0108] In this embodiment, the step of extracting features from the injection and production processes of different gas storage facilities and generating a span score based on reinforcement learning is specifically as follows:

[0109] For each second well group with the maximum spatial span, read the corresponding gas storage injection and production process data, and group the injection and production processes according to the well group connectivity subgraph;

[0110] Within each well group connected subgraph, the synchronous correlation coefficient of the bottom hole pressure change sequence of any two wells during the injection and production cycle is statistically analyzed, and the mean value of the correlation coefficient within the well group is calculated as a pressure coordination feature.

[0111] Within each well group connected subgraph, the maximum bottom hole pressure difference of each injection and production well during the injection and production cycle is calculated, and the maximum bottom hole pressure difference is used as the pressure imbalance characteristic.

[0112] Within each well group connected subgraph, the number of wells experiencing abnormal bottom hole pressure fluctuations during the injection and production cycle is counted, and the proportion of the number of abnormal wells to the upper limit of the number of wells in the well group is used as the pressure anomaly proportion feature.

[0113] Within each well group connected subgraph, the deviation between the actual injection and production volume of each injection and production well and the average injection and production volume of the well group during the injection and production cycle is calculated, and the mean square value of the deviation is calculated as the discrete feature of the injection and production load.

[0114] Within each well group connected subgraph, the number of wells that failed to execute injection and production commands synchronously due to insufficient pressure coupling is counted, and this number is used as the injection and production execution mismatch characteristic.

[0115] The pressure coordination characteristics, pressure imbalance characteristics, pressure anomaly ratio characteristics, injection-production load dispersion characteristics, and injection-production execution mismatch characteristics are combined in a fixed order to form the injection-production process feature vector corresponding to the maximum spatial span of the second well group.

[0116] The feature vectors of the injection and production process corresponding to the maximum spatial span of all second well groups are constructed into a state set, and the maximum spatial span of the second well group is used as the action index;

[0117] Using the completion rate of well cluster injection and production commands after the injection and production process is completed as the reward signal, reinforcement learning training samples are constructed.

[0118] A reinforcement learning model is trained based on the state set, action index, and reward signal to obtain a scoring model for outputting the maximum spatial span score of the well group.

[0119] Input the feature vector of the injection and production process corresponding to the maximum spatial span of each second well group into the scoring model, and output the corresponding span score;

[0120] The scoring model is a weighted summation model, the output of which is a span score. The weights are each feature, the training objective is to train the specific values ​​of the weights, and the trained weight values ​​are assigned to the weighted summation model to obtain the trained scoring model.

[0121] In this embodiment, the gas storage injection and production process data refers to the complete set of operational data recorded under the maximum spatial span of a specific second well group. This data includes the changes in bottom hole pressure of each injection and production well within the well group over time, the execution status of injection and production volumes, and the execution status of well group-level injection and production commands, which are used to reflect the actual operational effect under this span condition.

[0122] In this embodiment, grouping the injection and production process by well group connectivity subgraph means dividing the injection and production process data under the maximum spatial span condition of the same second well group into several independent data subsets according to the well group connectivity subgraph structure constructed in the previous steps, so that each well group connectivity subgraph corresponds to a sample of injection and production process that can be evaluated independently.

[0123] In this embodiment, the bottom hole pressure change sequence refers to the bottom hole pressure time series data continuously collected from a single injection-production well during the injection-production cycle. It is used to reflect the response behavior of the well to pressure disturbances during the injection-production process and serves as the basic data source for subsequently constructing features such as pressure coordination and pressure imbalance.

[0124] In this embodiment, the synchronization correlation coefficient refers to a statistical measure used to characterize the degree of synchronization between the bottom-hole pressure change sequences of two wells on a time scale. This synchronization correlation coefficient reflects the consistency of the pressure change trends of the two wells within the same injection-production cycle and is used to characterize whether the pressure response between the wells has synergistic characteristics.

[0125] In this embodiment, the pressure coordination feature refers to the average index obtained by statistically analyzing the correlation coefficients of bottom hole pressure synchronization for all well pairs within a single well group connected subgraph. This feature is used to comprehensively characterize the overall pressure coordination level within the well group under the condition of the maximum spatial span of the current second well group.

[0126] In this embodiment, the maximum bottomhole pressure difference refers to the extreme difference in the range of bottomhole pressure variation among the injection and production wells in the well cluster connectivity sub-diagram during the injection-production cycle. This difference reflects the degree of uneven pressure distribution within the well cluster and is an important basis for assessing whether a significant pressure gradient exists within the well cluster.

[0127] In this embodiment, the pressure imbalance feature refers to directly using the maximum bottom hole pressure difference as the quantitative result to characterize the pressure imbalance state within the well group caused by insufficient spatial span or pressure coupling. The larger the value, the more obvious the pressure distribution difference within the well group.

[0128] In this embodiment, abnormal fluctuations in bottom hole pressure refer to the behavior of a certain injection-production well where the bottom hole pressure changes significantly deviate from its historical stable change pattern during the injection-production cycle. This abnormality is usually manifested as an unexpected and drastic increase or decrease in a short period of time.

[0129] In this embodiment, the pressure anomaly proportion feature refers to the ratio of the number of wells experiencing abnormal bottomhole pressure fluctuations within a single well group connected subgraph to the upper limit of the number of wells in that well group. This feature is used to characterize the extent to which abnormal pressure behavior spreads within the well group.

[0130] In this embodiment, the actual injection and production volume refers to the actual gas injection or production volume data executed by the injection and production well during the injection and production cycle, which is used to reflect the actual load of the well during the execution of the injection and production command.

[0131] In this embodiment, the average injection-production volume of the well group refers to the average level obtained by statistically analyzing the actual injection-production volume of all injection-production wells within the injection-production cycle in a single well group connected subgraph, and is used as a reference benchmark to measure the degree of deviation of the load of a single well.

[0132] In this embodiment, the injection-production load dispersion characteristic refers to the dispersion index formed by the degree of deviation between the actual injection-production volume of each injection-production well and the average injection-production volume of the well group within the well group connectivity subgraph. This characteristic reflects whether the injection-production load distribution within the well group is balanced.

[0133] In this embodiment, the injection-production execution mismatch characteristic refers to the statistical results showing that, during the injection-production cycle, due to insufficient inter-well pressure coupling, some injection and production wells failed to synchronously complete the injection and production commands according to the well group-level injection-production execution plan. This characteristic is used to reflect the inadequacy of the well group's collaborative execution capability.

[0134] In this embodiment, the injection and production process feature vector refers to a multi-dimensional feature representation formed by combining pressure coordination features, pressure imbalance features, pressure anomaly ratio features, injection and production load discrete features, and injection and production execution mismatch features in a preset fixed order. It is used to completely describe the injection and production behavior features under the maximum spatial span condition of a certain second well group.

[0135] In this embodiment, the state set refers to the set composed of the feature vectors of the injection and production process corresponding to the maximum spatial span of all second well groups. This set is used to characterize the state space of the gas storage injection and production process under different spatial span conditions.

[0136] In this embodiment, the action index refers to mapping the maximum spatial span of different second well groups to optional action identifiers in reinforcement learning, enabling the model to select and evaluate among different spatial span conditions.

[0137] In this embodiment, the feedback signal refers to the feedback signal formed based on the completion rate of the well group injection and production command after the injection and production process is completed. It is used to reflect the overall effect of the current maximum spatial span of the second well group in actual execution.

[0138] The reinforcement learning model in this embodiment refers to a decision model trained based on the state set, action index, and reward signal. Its goal is to learn the mapping relationship between different injection feature combinations and spatial span conditions.

[0139] In this embodiment, the span score refers to the comprehensive evaluation result output by the trained scoring model for the maximum spatial span of each second well group, which is used to quantify the injection-production synergy and execution effect under the spatial span condition.

[0140] The weighted summation model in this embodiment refers to a model that takes each feature in the feature vector of the injection and collection process as input, assigns weights obtained through training to each feature, performs linear summation, and finally outputs a span score. This model determines the relative importance of each feature in the score through training.

[0141] It should be noted that the weighted summation scoring model trained by reinforcement learning in this embodiment is not introduced by introducing a complex nonlinear network structure, but rather by adaptively training the feature weights through a reinforcement learning mechanism, so that the scoring results can truly reflect the impact of different well group spatial span conditions on the overall coordination of the injection and production process.

[0142] It should be noted that by explicitly using the maximum spatial span of the second well group as an action index in the reinforcement learning framework, the model can learn by comparison under different spatial scale conditions during training. This avoids static evaluation based solely on the results of a single injection and production process, and improves the stability and generalization ability of the span score under different operating conditions.

[0143] S6. Based on the span score and the corresponding maximum spatial span of the second well group, curve analysis is performed to select the optimal maximum spatial span of the second well group for gas storage injection and production.

[0144] In this embodiment, the step of performing curve analysis based on the span score and the corresponding maximum spatial span of the second well group to select the optimal maximum spatial span of the second well group for gas storage injection and production is as follows:

[0145] Obtain the maximum spatial span of all second well groups and their corresponding span scores, and arrange the span scores in ascending order of the maximum spatial span of the second well groups to construct the maximum spatial span-span score sequence of the well groups;

[0146] Using the maximum spatial span of the second well group as the independent variable and the span score as the dependent variable, the maximum spatial span-span score sequence of the well group is smoothed to generate a continuous score curve;

[0147] Calculate the rate of change of scores between adjacent span points on the continuous scoring curve to generate a first-order score change sequence;

[0148] In the first-order change sequence of the rating, the position where the rate of change of the rating changes from positive to negative is identified, and the candidate maximum position of the rating curve is determined.

[0149] For the second well group maximum spatial span corresponding to the candidate maximum value position, a preset span neighborhood is extended to the left and right adjacent span points, and the mean score and variance of the score within the neighborhood are calculated.

[0150] Remove the maximum spatial span of the second well group corresponding to the candidate maxima whose score variance is greater than the preset stability threshold, and retain the candidate spans with stable scores;

[0151] Among the retained candidate spans, the maximum spatial span of the second well group with the largest span score is selected and determined as the optimal maximum spatial span of the second well group.

[0152] Input the maximum spatial span of the optimal second well group into the well group connectivity subgraph construction module to generate the corresponding well group connectivity structure and the upper limit combination of the number of wells.

[0153] Based on the combination of the well cluster connectivity structure and the upper limit of the number of wells, the injection and production scheduling control of the gas storage facility is executed.

[0154] In this embodiment, the second maximum spatial span of the well group refers to the spatial scale parameters of multiple candidate well groups obtained by discrete construction through the preceding steps and verification and scoring through the injection and production process. Each span corresponds to a maximum allowable spatial distance between wells, which is used to constrain the well group connectivity structure and the range of coordinated injection and production.

[0155] In this embodiment, the span score refers to the quantitative evaluation result output by the scoring model on the comprehensive execution effect of the maximum spatial span of each second well group in the actual injection and production process. The score reflects the comprehensive level of well group pressure coordination, load balancing and command execution stability under the spatial span condition.

[0156] In this embodiment, the maximum spatial span of the well group—span score sequence refers to an ordered data sequence formed by associating the maximum spatial span of all second well groups with their corresponding span scores and arranging them in ascending order of spatial span. This sequence is used to characterize the overall trend of the impact of span changes on the scoring results.

[0157] In this embodiment, the smoothing process refers to the continuous processing of the discrete maximum spatial span of the well group—the span score sequence, so that the score changes with the spatial span to form a continuous trend curve without obvious abrupt changes, thereby reducing the interference of fluctuations in a single injection and production process on the overall judgment.

[0158] In this embodiment, the continuous scoring curve refers to the continuous curve of the span score obtained after smoothing as it changes with the maximum spatial span of the well group, which is used to reflect the overall evolution trend of the score at different spatial scales.

[0159] In this embodiment, the rate of change of the score refers to the direction and magnitude of the score change between two adjacent spatial span points on a continuous score curve, and is used to characterize whether the score is rising, stable or falling as the spatial span increases.

[0160] In this embodiment, the first-order change sequence of the rating refers to the change trend sequence formed by arranging the rate of change of ratings between adjacent span points in a continuous rating curve in order of spatial span, which is used to help identify the turning point of the rating trend.

[0161] In this embodiment, the position where the rate of change of the score changes from positive to negative refers to the position in the first-order change sequence of the score that changes from a continuously rising state to a falling state as the score spans the spatial range. This position corresponds to the local maximum region of the score curve.

[0162] In this embodiment, the candidate maximum position refers to one or more possible positions of the second well group with the maximum spatial span corresponding to the score peak value, identified by the score change rate turning from positive to negative, and used as candidate objects for subsequent stability screening.

[0163] In this embodiment, the span neighborhood refers to the local span interval formed within the maximum spatial span range of several second well groups adjacent to a candidate maximum value position, and is used to evaluate the scoring stability of the candidate span under adjacent conditions.

[0164] In this embodiment, the average score refers to the average level obtained by statistically analyzing the span scores corresponding to all adjacent spans within the span neighborhood, which is used to characterize the overall score performance of the candidate maximum value in the local range.

[0165] In this embodiment, the rating variance refers to the degree of dispersion of the ratings corresponding to each adjacent span relative to the mean of the neighborhood ratings within the span neighborhood, and is used to reflect whether there are significant fluctuations in the ratings around that span.

[0166] In this embodiment, the preset stability threshold is a stability judgment standard used to determine whether the score fluctuation is acceptable. This threshold is used to distinguish whether the score changes with the span and is in a stable and controllable state, thereby excluding spatial span candidate values ​​that are sensitive to changes in working conditions.

[0167] In this embodiment, the candidate span with stable scores refers to the second well group's maximum spatial span whose score variance does not exceed a preset stability threshold within its span neighborhood. The scores corresponding to this type of span remain relatively stable under adjacent scale changes.

[0168] In this embodiment, the optimal second well group maximum spatial span refers to the spatial span parameter that achieves the maximum span score among all candidate spans with stable scores. This span achieves a comprehensive optimal balance between performance superiority and stability.

[0169] In this embodiment, the well group connectivity subgraph construction module refers to a functional module used to reconstruct the inter-well connectivity relationship, generate the well group connectivity structure, and the corresponding upper limit combination of the number of wells based on the selected maximum spatial span of the well group.

[0170] In this embodiment, the well group connectivity structure refers to the well group topology structure formed by injection and production well nodes and their connectable relationships under the constraint of the maximum spatial span of the optimal second well group, which is used to guide the well group-level injection and production organization method.

[0171] In this embodiment, the maximum number of wells is configured to allow the maximum number of wells to participate in coordinated injection and production within each well group's connected subgraph in the well group connectivity structure, thereby constraining the size of a single well group.

[0172] In this embodiment, the injection and production scheduling control refers to the process of splitting the gas storage injection and production plan at the well group level, issuing instructions and controlling execution based on the well group connectivity structure and the upper limit combination of the number of wells, so as to realize the injection and production scheduling that operates under the optimal spatial span conditions.

[0173] Example 2, Figure 2 This invention presents a reinforcement learning-based intelligent injection and production control optimization system for gas storage facilities, comprising a well group maximum spatial span initialization module, a well group maximum spatial span discretization construction module, a well number upper limit combination acquisition module, a gas storage injection and production module, a span scoring module, and a gas storage injection and production optimization module. The well group maximum spatial span initialization module is used to obtain the initial maximum spatial span of the well group based on inter-well pressure disturbance attenuation analysis. The well group maximum spatial span discretization construction module is used to discretize the initial maximum spatial span of the well group based on the local curvature decomposition of the inter-well pressure disturbance attenuation curve, generating several second maximum spatial spans of the well group to be screened. The well number upper limit combination acquisition module is used for... Based on the construction of a well cluster connected subgraph based on the spatial coverage relationship between wells, the maximum spatial span of the well cluster is mapped to the injection-production coordinated execution control target. The injection-production coordinated execution control target includes the upper limit combination of the number of wells corresponding to the maximum spatial span of the well cluster. The gas storage injection-production module is used to map the maximum spatial span of the second well cluster and apply it to the gas storage injection-production. The span scoring module is used to extract features from different gas storage injection-production processes and train the scoring based on reinforcement learning to generate a span score. The gas storage injection-production optimization module is used to perform curve analysis based on the span score and the corresponding maximum spatial span of the second well cluster, select the optimal maximum spatial span of the second well cluster, and perform gas storage injection-production.

[0174] In the embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0175] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0176] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0177] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0178] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0179] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0180] In the embodiments provided in this disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0181] It should be noted that, in this disclosure, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element limited by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0182] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for optimizing intelligent injection and production control in gas storage facilities based on reinforcement learning, characterized in that, Includes the following steps: The maximum spatial span of the initial well group was obtained based on the inter-well pressure disturbance attenuation analysis. Based on the local curvature decomposition of the inter-well pressure disturbance attenuation curve, the maximum spatial span of the initial well group is discretized and constructed to generate several maximum spatial spans of the second well group to be screened. Based on the construction of a well group connected subgraph based on the spatial coverage relationship between wells, the maximum spatial span of the well group is mapped to the injection-production coordinated execution control target, which includes the upper limit combination of the number of wells corresponding to the maximum spatial span of the well group; The maximum spatial span of the second well group was mapped and applied to gas storage injection and production. Feature extraction was performed on the injection and production processes of different gas storage facilities, and a score was generated based on reinforcement learning. Based on the span score and the corresponding maximum spatial span of the second well group, curve analysis is performed to select the optimal maximum spatial span of the second well group for gas storage injection and production. The specific steps for training the scoring based on reinforcement learning are as follows: The feature vectors of the injection and production process corresponding to the maximum spatial span of all second well groups are constructed into a state set, and the maximum spatial span of the second well group is used as the action index; Using the completion rate of well cluster injection and production commands after the injection and production process is completed as the reward signal, reinforcement learning training samples are constructed. Based on the state set, action index, and reward signal, a reinforcement learning model is trained to obtain a scoring model for outputting the maximum spatial span score of the well group.

2. The intelligent injection and production control optimization method for gas storage based on reinforcement learning according to claim 1, characterized in that, The method for obtaining the initial maximum spatial span of the well group based on inter-well pressure disturbance attenuation analysis is as follows: The historical injection and production records of multiple injection and production wells in the gas storage facility are obtained, the target wells where injection and production volume changes are identified, and the bottom hole pressure change sequence of the target wells before and after the injection and production change time is extracted. Centered on the target well, adjacent wells are selected from near to far according to the spatial distance between them. The bottom pressure change sequence of each adjacent well within the same time window as the target well is extracted, and the inter-well pressure response sequence corresponding to the target well and each adjacent well is constructed. Calculate the inter-well spatial distance corresponding to each inter-well pressure response sequence, and sort the inter-well pressure response sequences in ascending order of inter-well spatial distance to generate an inter-well spatial distance sequence; In the inter-well spatial distance sequence, the maximum pressure response amplitude of each adjacent well pressure change sequence within the target well injection and production change time window is calculated, and the maximum pressure response amplitude is associated with the corresponding inter-well spatial distance to generate an inter-well spatial distance-pressure response amplitude sequence. Using the inter-well spatial distance as the independent variable and the pressure response amplitude as the dependent variable, a decay sequence of pressure response amplitude as a function of inter-well spatial distance is constructed, mapped to a curve, and the rate of change of pressure response amplitude between adjacent distance points is calculated. The pressure response attenuation inflection point is determined at the position where the rate of change of the pressure response amplitude changes from a gradual change to a continuous decrease, and the inter-well spatial distance corresponding to the inflection point is extracted. The inter-well spatial distance corresponding to the inflection point is determined as the maximum spatial scale in which there is still a significant pressure response correlation between wells within the well group, and is used as the initial maximum spatial span of the well group for well group division.

3. The intelligent injection and production control optimization method for gas storage based on reinforcement learning according to claim 2, characterized in that, The local curvature decomposition based on the inter-well pressure disturbance attenuation curve discretizes the initial well group's maximum spatial span, generating several second well group maximum spatial spans to be screened. Specifically: The location of the inter-well spatial distance corresponding to the maximum spatial span of the initial well group is obtained based on the inter-well spatial distance-pressure response amplitude attenuation curve. Using the inter-well spatial distance as the independent variable and the pressure response amplitude as the dependent variable in the inter-well spatial distance-pressure response amplitude decay curve, the curve is subjected to second-order difference calculation to generate a curvature sequence along the direction of inter-well spatial distance. In the curvature sequence, with the position corresponding to the maximum spatial span of the initial well group as the center, an interval in which the absolute value of curvature continuously increases is selected in the direction smaller than the maximum spatial span of the initial well group, and an interval in which the absolute value of curvature rapidly decreases is selected in the direction larger than the maximum spatial span of the initial well group, thus constructing an effective curvature change interval; Within the effective range of curvature change, the inter-well spatial distance-pressure response amplitude attenuation curve is sampled at equal intervals according to the inter-well spatial distance to obtain multiple discrete inter-well spatial distance nodes; For each discrete well-space distance node, calculate its corresponding curvature value, and sort the discrete well-space distance nodes from largest to smallest according to the curvature value; The sorted discrete well-to-well spatial distance nodes are used as candidate values ​​for the spatial scale of the well group under different pressure disturbance attenuation characteristics to construct several second-level maximum spatial spans of the well group.

4. The intelligent injection-production control optimization method for gas storage based on reinforcement learning according to claim 3, characterized in that, The well cluster connectivity subgraph construction based on the spatial coverage relationship between wells maps the maximum spatial span of the well cluster to the injection-production coordinated execution control objective. The injection-production coordinated execution control objective includes the upper limit combination of the number of wells corresponding to the maximum spatial span of the well cluster, specifically: Obtain the wellhead spatial coordinates of all injection and production wells in the gas storage facility, calculate the inter-well spatial distance between any two wells, and construct the inter-well spatial distance matrix; For each second well group's maximum spatial span, well pairs in the inter-well spatial distance matrix that are less than or equal to the second well group's maximum spatial span are marked as connectable well pairs; Using injection and production wells as nodes and the connectable well pairs as undirected edges, a well group connectivity graph corresponding to the maximum spatial span of the second well group is constructed. The well group connectivity graph is decomposed into connectivity components to generate several independent well group connectivity subgraphs. In each well group connected subgraph, the number of nodes in the subgraph is counted to obtain the upper limit of the number of wells in a single well group under the maximum spatial span of the second well group. The maximum spatial span of the second well group is associated with the number of connected subgraphs of the corresponding well group and the upper limit of the number of wells in each connected subgraph of the well group to generate a combination relationship of maximum spatial span of well group and upper limit of number of wells. The relationship between the maximum spatial span of the well group and the upper limit of the number of wells is used as the mapping result of the injection-production coordinated execution control objective.

5. The intelligent injection-production control optimization method for gas storage based on reinforcement learning according to claim 4, characterized in that, The mapping of the maximum spatial span of the second well group and its application to gas storage injection and production are specifically as follows: Based on the maximum spatial span of each second well group, read its corresponding well group connected subgraph and the upper limit of the number of wells in the well group connected subgraph; Within each well group connected subgraph, the injection and production wells within the well group are sorted according to the spatial location of the wellhead and the historical injection and production role identifier, and the injection and production wells not exceeding the upper limit of the number of wells are selected as the set of coordinated injection and production wells of the well group under the current span conditions. For each injection and production well in the coordinated injection and production well set, its historical injection and production operating condition type is inherited, and it is determined as a gas injection well or a gas production well, respectively, and the roles of injection and production wells are kept from conflicting within the same well group; Using the coordinated injection and production well set as the execution unit, the overall injection and production plan of the gas storage is divided into well groups to generate a well group-level injection and production execution plan with the maximum spatial span of the second well group. During the injection and production process, injection and production instructions are issued synchronously according to the well group-level injection and production execution plan, and the coordinated start and stop status of injection and production wells within the well group is maintained during the injection and production process; Record the bottom hole pressure changes, injection and production volume changes, and well group-level execution status of each well group throughout the entire injection and production cycle under the maximum spatial span condition of the second well group, forming gas storage injection and production process data under the corresponding span condition.

6. The intelligent injection-production control optimization method for gas storage based on reinforcement learning according to claim 5, characterized in that, The process involves extracting features from the injection and production processes of different gas storage facilities, and then using reinforcement learning to train a scoring system to generate a range score. Specifically: For each second well group with the maximum spatial span, read the corresponding gas storage injection and production process data, and group the injection and production processes according to the well group connectivity subgraph; Within each well group connected subgraph, the synchronous correlation coefficient of the bottom hole pressure change sequence of any two wells within the injection and production cycle is statistically analyzed, and the mean value of the correlation coefficient within the well group is calculated as a pressure coordination feature. Within each well group connected subgraph, the maximum bottom hole pressure difference of each injection and production well during the injection and production cycle is calculated, and the maximum bottom hole pressure difference is used as the pressure imbalance characteristic. Within each well group connected subgraph, the number of wells experiencing abnormal bottom hole pressure fluctuations during the injection and production cycle is counted, and the proportion of the number of abnormal wells to the upper limit of the number of wells in the well group is used as the pressure anomaly proportion feature. Within each well group connected subgraph, the deviation between the actual injection and production volume of each injection and production well and the average injection and production volume of the well group during the injection and production cycle is calculated, and the mean square value of the deviation is calculated as the discrete feature of the injection and production load. Within each well group connected subgraph, the number of wells that failed to execute injection and production commands synchronously due to insufficient pressure coupling is counted, and this number is used as the injection and production execution mismatch characteristic. The pressure coordination characteristics, pressure imbalance characteristics, pressure anomaly ratio characteristics, injection-production load dispersion characteristics, and injection-production execution mismatch characteristics are combined in a fixed order to form the injection-production process feature vector corresponding to the maximum spatial span of the second well group. Input the feature vector of the injection and production process corresponding to the maximum spatial span of each second well group into the scoring model, and output the corresponding span score. The scoring model is a weighted summation model, the output of which is a span score. The weights are each feature, the training objective is to train the specific values ​​of the weights, and the trained weight values ​​are assigned to the weighted summation model to obtain the trained scoring model.

7. The intelligent injection-production control optimization method for gas storage based on reinforcement learning according to claim 6, characterized in that, The process involves performing curve analysis based on the span score and the corresponding maximum spatial span of the second well group to select the optimal maximum spatial span for gas storage injection and production. Specifically: Obtain the maximum spatial span of all second well groups and their corresponding span scores, and arrange the span scores in ascending order of the maximum spatial span of the second well groups to construct the maximum spatial span-span score sequence of the well groups; Using the maximum spatial span of the second well group as the independent variable and the span score as the dependent variable, the maximum spatial span-span score sequence of the well group is smoothed to generate a continuous score curve; Calculate the rate of change of scores between adjacent span points on the continuous scoring curve to generate a first-order score change sequence; In the first-order change sequence of the rating, the position where the rate of change of the rating changes from positive to negative is identified, and the candidate maximum position of the rating curve is determined. For the second well group maximum spatial span corresponding to the candidate maximum value position, a preset span neighborhood is extended to the left and right adjacent span points, and the mean score and variance score within the neighborhood are calculated. Remove the maximum spatial span of the second well group corresponding to the candidate maxima whose score variance is greater than the preset stability threshold, and retain the candidate spans with stable scores; Among the retained candidate spans, the maximum spatial span of the second well group with the largest span score is selected and determined as the optimal maximum spatial span of the second well group. Input the maximum spatial span of the optimal second well group into the well group connectivity subgraph construction module to generate the corresponding well group connectivity structure and the upper limit combination of the number of wells. Based on the combination of the well cluster connectivity structure and the upper limit of the number of wells, the injection and production scheduling control of the gas storage facility is executed.

8. A reinforcement learning-based intelligent injection and production control optimization system for gas storage facilities, used to implement the reinforcement learning-based intelligent injection and production control optimization method for gas storage facilities as described in any one of claims 1-7, characterized in that, It includes a module for initializing the maximum spatial span of a well group, a module for discretizing the maximum spatial span of a well group, a module for obtaining the upper limit of the number of wells, a module for gas storage injection and production, a span scoring module, and a module for optimizing gas storage injection and production. The module for initializing the maximum spatial span of the well group is used to obtain the initial maximum spatial span of the well group based on the analysis of pressure disturbance attenuation between wells. The maximum spatial span discretization construction module for well groups is used to discretize the initial maximum spatial span of well groups based on the local curvature decomposition of the inter-well pressure disturbance attenuation curve, and generate several second maximum spatial spans of well groups to be screened. The module for obtaining the upper limit combination of well numbers is used to construct a well group connected subgraph based on the spatial coverage relationship between wells, and to map the maximum spatial span of the well group to the injection and production coordinated execution control target. The injection and production coordinated execution control target includes the upper limit combination of well numbers corresponding to the maximum spatial span of the well group. The gas storage injection and production module is used to map the maximum spatial span of the second well group and apply it to the gas storage injection and production. The span scoring module is used to extract features from the injection and production processes of different gas storage facilities, and to train the scoring based on reinforcement learning to generate span scores. The gas storage injection and production optimization module is used to perform curve analysis based on the span score and the corresponding maximum spatial span of the second well group, and to select the optimal maximum spatial span of the second well group for gas storage injection and production.

Images