Sensor data-based linkage well water power self-adaptive regulation system

Abstract

The present application relates to pumping and injection well linkage regulation technical field, specifically relates to a kind of pumping and injection well linkage hydrodynamic self-adaptive regulation system based on sensor data, the system includes processor and memory, processor executes the computer program stored in memory to realize the following steps: obtaining optimal cluster, based on the lag time and lag correlation intensity of data pair in optimal cluster, obtain the predicted lag time and predicted lag correlation intensity between the pumping well and injection well existing hydraulic connection under each dimension;Based on the predicted lag time and predicted lag correlation intensity between the pumping well and injection well existing hydraulic connection under the same dimension, lag time network and lag correlation network under each dimension are respectively constructed, and pumping well and injection well in target area are adaptively linked hydrodynamic regulation based on lag time network and lag correlation network.The present application can significantly improve the stability and precision of pumping well and injection well linkage hydrodynamic regulation.

Description

Technical Field

[0001] This invention relates to the field of pumping and injection well linkage control technology, specifically to a pumping and injection well linkage hydrodynamic adaptive control system based on sensor data. Background Technology

[0002] In oilfield development, groundwater resource allocation, and geothermal energy utilization, the extraction and replenishment of underground fluids are usually balanced by the coordinated operation of pumping wells (extracting underground fluids) and injection wells (reinjecting treated fluids), or by the linkage of pumping wells and injection wells with hydrodynamic regulation, so as to maintain the stability of formation pressure and improve the efficiency of resource development. Traditionally, multiple sensors (such as pressure and flow sensors) are deployed at the well site for real-time monitoring and data acquisition. Operational status analysis and control are then based on this data. However, the seepage process of underground media is complex, and fluid propagation within the formation takes time. This leads to significant temporal differences or time lag effects in the hydrodynamic responses of pumping and injection wells. When directly relying on collected monitoring data for coordinated control of pumping and injection wells, the current approach is prone to misjudging the actual operational status of the wells due to the neglect of time lag, resulting in control lag, improper adjustment, and ultimately affecting the stability and accuracy of control. To ensure stability and accuracy, lag parameters are introduced during control, but these are fixed or preset. Therefore, existing technologies typically employ time-delay system modeling and other methods to perform lag correlation analysis on the time-series operational data of hydraulically linked pumping and injection wells. These methods introduce fixed or preset time... Lag parameters are used to quantify the time-series operational data of hydraulically connected pumping and injection wells, thereby characterizing the temporal differences in the hydrodynamic responses of pumping and injection wells. However, in actual underground fluid seepage processes, factors such as formation heterogeneity, differences in permeability conditions, and variations in the spatial distance between wells cause the time lag characteristics between pumping and injection wells to be non-fixed, dynamically changing with the operating status of the pumping and injection wells and formation conditions. Consequently, modeling methods based on fixed or preset lag parameters are difficult to accurately depict the real hydrodynamic propagation relationship or accurately quantify the temporal differences or time lag effects between hydraulically connected pumping and injection wells. They are also difficult to effectively characterize the dynamic coupling influence between multiple wells in complex underground environments, leading to problems such as lagging control response or inaccurate regulation, which affects the stability and accuracy of regulation. Therefore, how to adaptively link pumping and injection wells for hydrodynamic regulation to improve the stability and accuracy of regulation has become an urgent problem to be solved. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides an adaptive hydrodynamic control system for injection and extraction wells based on sensor data. The specific technical solution adopted is as follows: One embodiment of the present invention provides an adaptive hydrodynamic control system for injection and extraction wells based on sensor data, including a processor and a memory. The processor executes a computer program stored in the memory to perform the following steps: The system acquires multi-dimensional processed monitoring data sequences of pumping wells and injection wells in the target area, as well as characteristic data and local change intensity of characteristic data in each processed monitoring data sequence. The dimensions include at least the pressure, flow velocity, water level, and water quality parameters in the well. For pumping well a and injection well b that are hydraulically connected, the feature data in the processed monitoring data sequences of pumping well a and injection well b under the same dimension are combined in pairs to obtain data pairs under each dimension. Based on the length of the local sub-segment containing the data in each data pair, the alignment result of the local sub-segment, and the intensity of local change in the data in the data pair, the lag correlation strength of each data pair is obtained. Based on the difference in lag correlation strength, lag time difference, and time range relationship between any two data pairs under the same dimension, the target difference metric distance between any two data pairs under each dimension is obtained. Based on the target difference metric distance, the data pairs under each dimension are clustered to obtain clusters under each dimension. Based on the mean of the target difference metric distance and the mean of the lag correlation strength between data pairs within the cluster, the optimal cluster corresponding to each dimension is selected. Based on the lag time and lag correlation strength of the data pairs within the optimal cluster, the predicted lag time and predicted lag correlation strength between pumping well a and injection well b under each dimension are obtained. Based on the predicted lag time and predicted lag correlation strength between pumping wells and injection wells that have hydraulic connections in the same dimension, lag time networks and lag correlation networks in each dimension are constructed respectively. Based on the lag time networks and lag correlation networks, the pumping wells and injection wells in the target area are adaptively linked for hydrodynamic regulation.

[0004] Beneficial Effects: This invention first acquires multi-dimensional processed monitoring data sequences of pumping wells and injection wells in the target area, as well as the feature data and local change intensity of the feature data in each processed monitoring data sequence; then, it combines the feature data in the processed monitoring data sequences of pumping wells and injection wells with hydraulic connections in the same dimension to obtain data pairs in each dimension. Based on the length of the local sub-segments containing the data in each data pair, the alignment result of the local sub-segments, and the local change intensity of the data in the data pair, the hysteresis correlation strength of each data pair is obtained. Based on the difference in hysteresis correlation strength, hysteresis time difference, and time range relationship between any two data pairs in the same dimension, the target difference metric distance between any two data pairs in each dimension is obtained. The target difference metric distance is used to cluster data pairs in each dimension to obtain clusters in each dimension. The optimal clusters for each dimension are selected based on the mean target difference metric distance and the mean lag correlation strength between data pairs within the clusters. Based on the lag time and lag correlation strength of data pairs within the optimal clusters, the predicted lag time and predicted lag correlation strength between hydraulically connected pumping wells and injection wells in each dimension are obtained. Finally, based on the predicted lag time and predicted lag correlation strength between hydraulically connected pumping wells and injection wells in the same dimension, lag time networks and lag correlation networks are constructed for each dimension. Based on the lag time networks and lag correlation networks, the pumping wells and injection wells in the target area are adaptively linked for hydrodynamic regulation. Furthermore, by dynamically predicting future lag states and feeding the prediction results back to the control system, this invention can significantly improve the efficiency, stability, and accuracy of coordinated hydrodynamic regulation of pumping wells and injection wells. In other words, by dynamically predicting the predicted lag time and predicted lag correlation strength between pumping wells and injection wells that have hydraulic connections under different dimensions, this invention can significantly improve the stability and accuracy of coordinated hydrodynamic regulation of pumping wells and injection wells. Attached Figure Description

[0005] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0006] Figure 1 This is a flowchart of an adaptive control method for the linkage of injection and extraction wells based on sensor data, according to the present invention. Detailed Implementation

[0007] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the protection scope of the embodiments of the present invention.

[0008] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.

[0009] This embodiment provides a sensor data-based adaptive hydrodynamic control system for injection and extraction wells, including a processor and a memory. The processor executes a computer program stored in the memory to implement a sensor data-based adaptive hydrodynamic control method for injection and extraction wells. Figure 1 As shown, the adaptive hydrodynamic control method for injection and extraction wells includes the following steps: Step S001: Obtain multi-dimensional processed monitoring data sequences of pumping wells and injection wells in the target area, as well as the characteristic data and the intensity of local changes in the characteristic data in each processed monitoring data sequence.

[0010] Due to factors such as the heterogeneity of the formation medium, differences in permeability conditions, and variations in the spatial distance between wells during actual underground fluid seepage, the time lag characteristics between pumping and injection wells are not constant but dynamically change with the operating status of the pumping and injection wells and formation conditions. In actual operation, the hydrodynamic responses of different wells do not occur synchronously; rather, a change in the operating status of one well is followed by a response from another well after a certain time interval. This response delay exhibits dynamic fluctuations with changes in pressure, flow state, and formation conditions. Furthermore, differences in the spatial location and formation structure between wells also lead to significant differences in the response lag relationships between different pumping and injection wells, resulting in a complex and non-uniform temporal sequence of multi-well responses. The inherent characteristics of hydrodynamics make it difficult for modeling methods based on fixed or preset lag parameters to accurately depict real hydrodynamic propagation relationships or accurately quantify the temporal differences or time lag effects between hydraulically connected pumping and injection wells. Furthermore, it is difficult to effectively characterize the dynamic coupling effects between multiple wells in complex underground environments. In other words, modeling methods based on fixed or preset lag parameters cannot adapt to the aforementioned dynamic, nonlinear, and multi-scale coupling characteristics, thus affecting the stability and accuracy of regulation. To improve regulation stability and accuracy, this embodiment dynamically predicts future lag states and feeds the prediction results back to the control system, thereby driving adaptive regulation of pumping and injection wells to enhance regulation stability and accuracy.

[0011] Furthermore, for ease of understanding, this embodiment will subsequently use the coordinated hydrodynamic control process of pumping and injection wells within any area requiring coordinated hydrodynamic control of pumping and injection wells as an example. For instance, it can be described using the coordinated hydrodynamic control process of pumping and injection wells in any oil and gas field development area, and this area will be designated as the target area. Since this embodiment will subsequently quantify the hysteresis characteristics based on monitoring data from pumping and injection wells, this embodiment will first acquire multidimensional operational data of pumping and injection wells in the target area and form a multidimensional time series, specifically: First, multi-dimensional monitoring data of each water well at different times is collected using sensors or acquisition devices. This data is then recorded as the raw monitoring data. In this embodiment, the dimensions of the collected monitoring data or parameters include, but are not limited to, pressure, flow velocity, water level, and water quality parameters within the well. Water quality parameters include, but are not limited to, water temperature and pH value. Therefore, the dimension set in this embodiment consists at least of the pressure, flow velocity, water level, water temperature, and pH value within the well. In this embodiment, different water wells and different dimensions are collected synchronously. The number and type of dimensions collected from different water wells are consistent. The time interval for collecting monitoring data or parameters needs to be set by the implementer based on engineering experience and other practical considerations. In this embodiment, the data acquisition time interval is set to an empirical value. For any well, the time series constructed from all the monitoring data of the same dimension collected from the start of acquisition to the current moment, according to the order of acquisition, is recorded as the original monitoring data sequence of the well at the current moment. The start acquisition time is generally the initial development time. The data in the same original monitoring data sequence are all data of the same dimension. At the current moment, the original monitoring data sequence of the well under each dimension can be obtained. For example, the original monitoring data sequence of the well under the pressure dimension at the current moment is constructed from all the well pressures collected from the start of acquisition to the current moment, according to the order of acquisition. Other original monitoring data sequences are similar. To avoid the impact of noise on subsequent analysis, the original monitoring data sequence obtained above will be filtered in this embodiment, and the filtered original monitoring data sequence will be recorded as the processed monitoring data sequence. In this embodiment, the Kalman filter algorithm can be selected for filtering. The acquisition time or timestamp of each data in the processed monitoring data sequence is the acquisition time or timestamp of the original monitoring data before the data was filtered. If a certain processed monitoring data sequence is the result of filtering a certain original monitoring data sequence, then the acquisition time or timestamp of the v-th data in the processed monitoring data sequence is the acquisition time or timestamp of the v-th original monitoring data in the original monitoring data sequence.

[0012] Therefore, this embodiment can obtain multi-dimensional processed monitoring data sequences for each pumping well and each injection well in the target area at the current moment through the above process. To reduce computational load and enhance the ability to characterize important changes during the hydrodynamic response process, laying the foundation for subsequent characterization of the dynamic time-varying lag relationship between pumping and injection wells, this embodiment will next analyze each processed monitoring data sequence obtained above, extracting key data points and their local change features. Key data points are also feature data, and local change features are the local change intensity of the feature data. Therefore, this embodiment will next extract the feature data from the processed monitoring data sequences and quantify the local change intensity of the feature data. The specific process is as follows: For any processed monitoring data sequence: perform a first-order difference on the processed monitoring data sequence to obtain the first-order difference sequence. Then, record the monitoring data segments corresponding to the consecutive difference sequences with the same sign in the first-order difference sequence as data subsequences. Use the monitoring data corresponding to the sign change positions in the first-order difference sequence as feature data. For example, if the processed monitoring data sequence is {1,2,3,4,5,4,3,2,1}, then the first-order difference sequence is {2-1,3-2,4-3,5-4,4-5,3-4,2-3,1-2}, which is {1,1,1,1,-1, Given the sequence {1,1,1,1,}, the monitoring data segments {1,2,3,4,5} and {5,4,3,2,1} corresponding to the difference numerator sequence {1,1,1,1,} are each a data subsequence. Since the data subsequences {1,2,3,4,5} and {5,4,3,2,1} are adjacent in the processed monitoring data sequence, and the monitoring data corresponding to the sign change position between these two data subsequences is either the last data of the earlier data subsequence or the first data of the later data subsequence, that is, 5 is the monitoring data corresponding to the sign change position between these two data subsequences, and therefore 5 is the feature data. Then, linear fitting is performed on each data subsequence to obtain the fitted data of each data in each data subsequence. In each data subsequence, the data with the largest fitting residual is selected as the feature data. The fitting residual of each data in the data subsequence is the difference between the data and the fitted data of the data. In this embodiment, the least squares method is selected to perform linear fitting on the subsequence, which is a well-known technique. If the fitting residuals of the data in the data subsequence are equal, then no feature data is selected in that data subsequence.

[0013] The following description uses the process of obtaining the local change intensity of the c-th feature data in any processed monitoring data sequence R as an example. Specifically, the process of obtaining the local change intensity of the c-th feature data in the processed monitoring data sequence R is as follows: First, the data segment from the (c-1)th feature data to the cth feature data in the processed monitoring data sequence R is denoted as the left local data segment of the cth feature data. The data segment from the cth feature data to the (c+1)th feature data in the processed monitoring data sequence R is denoted as the right local data segment of the cth feature data. There may be cases where c is 1 or c is N, where N is the total number of data in the processed monitoring data sequence R. When c is 1, it indicates that there are no other feature data to the left of the cth feature data in the processed monitoring data sequence R. In this case, the processed monitoring data is then... The data segment from the first monitoring data to the c-th feature data in the data sequence R is denoted as the left local data segment of the c-th feature data. If c is N, it indicates that there are no other feature data to the right of the c-th feature data in the processed monitoring data sequence R. In this case, the data segment from the c-th feature data to the last monitoring data in the processed monitoring data sequence is denoted as the right local data segment of the c-th feature data. Then, the first-order difference sequences of the left and right local data segments of the c-th feature data are obtained respectively, and denoted as the first difference sequence and the second difference sequence respectively. For each sequence, the absolute value of the difference between the means of the first and second difference sequences is linearly normalized and denoted as the representative value of the degree of change. Then, the mean absolute deviation of the first and second difference sequences is calculated respectively. The mean absolute deviation of any sequence refers to the mean of the absolute deviations of all data in that sequence, and the absolute deviation of any data point in any sequence is the absolute difference between that data point and the mean of that sequence. The maximum mean absolute deviation between the first and second difference sequences is reverse normalized, and the result of the reverse normalization is denoted as the confidence coefficient of the degree of change. Here, reverse normalization is the complement of the result obtained from linear normalization. Finally, the linear normalization result obtained by multiplying the representative value of the degree of change and the confidence coefficient of the degree of change is calculated and denoted as the local change intensity of the c-th feature data. The purpose of normalization here is to eliminate the influence of differences in the overall change amplitude between different wells or different time periods on subsequent analysis, thereby improving the comparability and stability of the estimation of lag relationships between different sequences. The expression for the local change intensity of the c-th feature data is: in, Let represent the local variation intensity of the c-th feature data, and norm() be the linear normalization function. Let be the mean of the first-order difference sequence of the left local data segment of the c-th feature data. Let be the mean of the first-order difference sequence of the local data segment to the right of the c-th feature data, and max() be the function to find the maximum value. Let be the mean absolute deviation of the first-order difference sequence of the left local data segment of the c-th feature data. The mean absolute deviation of the first-order difference sequence of the local data segment to the right of the c-th feature data; Used to measure the degree of difference in trend changes on both sides of the c-th feature data. The larger the value, the more significant the trend change at the c-th feature data point or the greater the degree of change at the c-th feature data point; Reflecting the stability of local changes on both sides of the c-th feature data, smaller or The larger the value, the more stable the local changes on both sides of the c-th feature data. And the more stable the local changes on both sides of the c-th feature data, the more stable it indicates... The more credible or The more reliable and trustworthy the local variation features of the c-th feature data, the better. The larger and The smaller, The larger, therefore A larger value indicates a stronger degree of local variation in the c-th feature data, and vice versa. The smaller the value, the weaker the local variation of the c-th feature data.

[0014] Therefore, this embodiment can obtain multi-dimensional processed monitoring data sequences of each pumping well and each injection well in the target area at the current moment through the above process, as well as the feature data and the local change intensity of the feature data in each processed monitoring data sequence.

[0015] Step S002: For pumping well a and injection well b that are hydraulically connected, the feature data in the processed monitoring data sequences of pumping well a and injection well b under the same dimension are combined pairwise to obtain data pairs under each dimension. Based on the length of the local sub-segments containing the data in each data pair, the alignment result of the local sub-segments, and the intensity of local changes in the data in the data pair, the lag correlation strength of each data pair is obtained. Based on the difference in lag correlation strength, the difference in lag time, and the time range relationship between any two data pairs under the same dimension, the target difference metric distance between any two data pairs under each dimension is obtained. Based on the target difference metric distance, the data pairs under each dimension are clustered to obtain clusters under each dimension. Based on the mean of the target difference metric distance and the mean of the lag correlation strength between data pairs within the cluster, the optimal cluster corresponding to each dimension is selected. Based on the lag time and lag correlation strength of the data pairs within the optimal cluster, the predicted lag time and predicted lag correlation strength between pumping well a and injection well b under each dimension are obtained.

[0016] This embodiment, after obtaining the feature data and the intensity of local changes in the feature data, combines the feature data and the intensity of local changes to predict the lag correlation strength and lag time between pumping wells and injection wells with hydraulic connections. Subsequently, based on the prediction results, coordinated hydrodynamic control is implemented between the pumping wells and injection wells to improve control stability and accuracy. Since the method for predicting the lag correlation strength and lag time between each hydraulically connected pumping well and injection well is the same in this embodiment, for ease of understanding, the following will use the method described below. The process of predicting the lag correlation strength and lag time between pumping well a and injection well b, which are hydraulically connected in the target area, is described using an example. Furthermore, a hydraulic connection between a pumping well and an injection well refers to the mutual influence of groundwater flow between them through the same aquifer or hydrogeological unit. This manifests as the injection behavior of the injection well triggering a water level response in the pumping well, or the pumping behavior of the pumping well causing a drop in the water level of the injection well, forming a dynamically coupled seepage field. The specific process for obtaining the predicted lag time and predicted lag correlation strength between pumping well a and injection well b is as follows: First, the feature data in the processed monitoring data sequences of pumping well a and injection well b under the same dimension are combined in pairs to obtain data pairs under each dimension. The processed monitoring data sequence of any well under any dimension is the result of filtering the original data sequence of that well under that dimension. For example, for the f-th dimension in the dimension set, in all the processed monitoring data sequences of pumping well a at the current time, the processed monitoring data sequences belonging to the f-th dimension are recorded as the first sequence to be combined. In all the processed monitoring data sequences of injection well b at the current time, the processed monitoring data sequences belonging to the f-th dimension are recorded as the second sequence to be combined. Each feature data in the first sequence to be combined is combined with each feature data in the second sequence to be combined in pairs without repetition. The combination results are all recorded as data pairs under the f-th dimension. That is, the two data in each data pair come from one processed monitoring data sequence of pumping well a and one processed monitoring data sequence of injection well b.

[0017] Then, based on the length of the local segments containing the data in each data pair, the alignment result of the local segments, and the intensity of local changes in the data in the data pair, the hysteresis correlation strength of each data pair is obtained. The data segment from the preceding feature data to the following feature data in any processed monitoring data sequence is considered a local segment of that feature data. That is, for the c-th feature data in the processed monitoring data sequence R, the local segment of the c-th feature data is the data segment from the (c-1)-th to the (c+1)-th feature data in the processed monitoring data sequence R. If c is 1, in this embodiment, the data segment from the first monitoring data to the (c+1)-th feature data in the processed monitoring data sequence R is taken as the local segment of the c-th feature data. If c is N, in this embodiment, the data segment from the (c-1)-th feature data to the last monitoring data in the processed monitoring data sequence R is taken as the local segment of the c-th feature data. This embodiment describes the process of obtaining the hysteresis correlation strength of any data pair y as an example. The process of obtaining the hysteresis correlation strength of data pair y is as follows: Let the two feature data in the data pair y be denoted as data 1 and data 2 respectively; let the local segments of data 1 and data 2 be denoted as local segment 1 and local segment 2 respectively; and let the local segment of the c-th feature data in the processed monitoring data sequence R be the data segment from the (c-1)-th feature data to the (c+1)-th feature data in the processed monitoring data sequence R. If c is 1, the local segment of the c-th feature data in the processed monitoring data sequence R is the data segment from the first running data to the (c+1)-th feature data in the processed monitoring data sequence R. If c is N, the local segment of the c-th feature data in the processed monitoring data sequence R is the data segment from the first running data to the (c+1)-th feature data in the processed monitoring data sequence R. The process of determining the local sub-segments of other feature data in column R, from the (c-1)th feature data to the last monitored data, is similar. Using data 1 in local sub-segment 1 and data 2 in local sub-segment 2 as alignment reference points, local sub-segment 1 and local sub-segment 2 are aligned on the time axis. The overlapping data segments of local sub-segment 1 and local sub-segment 2 on the time axis during alignment are extracted and denoted as the first overlapping sub-segment and the second overlapping sub-segment, respectively. An example illustrating the alignment of local sub-segment 1 and local sub-segment 2 and the extraction of overlapping sub-segments is as follows: If local sub-segment 1 is {a1, a2, a3, a4, a5}, and a3 is data 1, and local sub-segment 2 is {b1, b2, b...}... Given the data segments {3, b4, b5}, where b4 is data 2, and since the time interval between adjacent data is consistent in this embodiment, this embodiment uses data 1 in local segment 1 and data 2 in local segment 2 as alignment reference points to align local segment 1 and local segment 2 on the time axis. Essentially, this means moving local segment 1 backward or local segment 2 forward by one data point. That is, when local segment 1 and local segment 2 are aligned on the time axis, a1 aligns with b2, a2 ​​aligns with b3, a3 aligns with b4, and a4 aligns with b5. Therefore, the overlapping data segments of local segment 1 and local segment 2 on the time axis during alignment are {a1, a2, a3, a4}. And {b2,b3,b4,b5}, that is, at this time {a1,a2,a3,a4} is the first overlapping sub-segment, and {b2,b3,b4,b5} is the second overlapping sub-segment; then, based on the correlation coefficient between the first and second overlapping sub-segments, the length difference between local sub-segment 1 and local sub-segment 2, and the difference in local change intensity between data 1 and data 2, the lagged correlation strength of data with y is obtained; and the specific process of obtaining the lagged correlation strength of data with y based on the correlation coefficient between the first and second overlapping sub-segments, the length difference between local sub-segment 1 and local sub-segment 2, and the difference in local change intensity between data 1 and data 2 is as follows: Calculate the Spearman rank correlation coefficient between the first and second overlapping sub-segments. The absolute value of the Spearman rank correlation coefficient is denoted as the first characterization value. The inverse normalized result of the absolute value of the difference between the lengths of local sub-segments 1 and 2 is denoted as the second characterization value. The inverse normalization result of the absolute value of the difference in local variation intensity between data 1 and data 2 is denoted as the third characterization value. The product of the first, second, and third characterization values ​​is denoted as the lag correlation strength of the data pair y. The inverse normalization method here is the same as described above. Additionally, it should be noted that if data 1 in local sub-segment 1 and data 2 in local sub-segment 2 are used as alignment reference points, and local sub-segments 1 and 2 are aligned on the time axis, and there are no overlapping data segments between local sub-segments 1 and 2 on the time axis, then in this embodiment, the first characterization value is denoted as 0. The expression for the lag correlation strength of the data pair y is: in, The lagged correlation strength between the data and y. This is a local sub-segment of data 1. This is a local sub-segment of data 2. is the Spearman rank correlation coefficient between the first and second overlapping segments, and also the Spearman rank correlation coefficient of the overlapping portion when data 1 in a local segment of data 1 and data 2 in a local segment of data 2 are aligned. norm() is the linear normalization function. The length of a local sub-segment of data 1. The length of a local sub-segment of data 2. The intensity of local variation in data 1, This represents the intensity of local changes in data 2. This represents the absolute value of the Spearman rank correlation coefficient of the overlapping portions of local sub-segments of data 1 and data 2 under alignment conditions. It is used to characterize the consistency or morphological similarity between the two in terms of their changing trends. The larger the value, the stronger the lagged correlation between data 1 and data 2; , These are used to reflect the duration characteristics of local sub-segments of data 1 and data 2, respectively. The smaller the value, the more similar the durations of the local segments of data 1 and data 2 are, which in turn indicates a higher lagged correlation strength between data 1 and data 2. Used to measure the consistency of the magnitude of change between data 1 and data 2. The smaller the value, the more similar the change intensity of data 1 and data 2, and thus the stronger the lag correlation between data 1 and data 2; while The bigger, smaller and The smaller, The larger, therefore The larger the value, the stronger the lagged association between data 1 and data 2, meaning the stronger the lagged association between data 1 and y. Conversely, the smaller the value, the stronger the lagged association between data 1 and y. The smaller the value, the weaker the lagged correlation between the data and y.

[0018] Therefore, this embodiment can obtain the lag correlation strength of each data pair in each dimension at the current moment through the above process. While the actual lag relationship between pumping wells and injection wells changes over time, its changes usually exhibit a certain continuity and structural stability, and are not completely disordered. In order to filter out lag relationships that conform to actual physical laws and possess consistency and stability from a large number of potential data pair associations, this embodiment will next combine the differences in lag correlation strength, lag time, and time range between data pairs in the same dimension to measure the difference characteristics. Subsequently, based on the quantified difference characteristics, stable and consistent lag relationships will be identified. The lag pattern is used to eliminate unreasonable associations. Therefore, this embodiment will next obtain the target difference metric distance between any two data pairs in each dimension based on the difference in lag association strength, lag time, and time range relationship between any two data pairs in the same dimension. For ease of understanding, this embodiment will subsequently describe the specific process of obtaining the target difference metric distance between the i-th data pair and the j-th data pair in the f-th dimension as an example, where the i-th data pair and the j-th data pair are not the same data pair in the f-th dimension. That is, the specific process of obtaining the target difference metric distance between the i-th data pair and the j-th data pair in the f-th dimension is as follows: First, calculate the normalized result of the absolute value of the difference between the lag correlation strength of the i-th data pair and the lag correlation strength of the j-th data pair, and record it as the first distance value between the i-th and j-th data pairs. Then, calculate the normalized result of the absolute value of the difference between the lag time of the i-th data pair and the lag time of the j-th data pair, and record it as the second distance value between the i-th and j-th data pairs. The lag time of any data pair is the time interval between the acquisition of the two data points in the corresponding data pair. Then, obtain the time range of the data pair. The time range of any data pair refers to the continuous time interval formed by the acquisition times of the two data points in the data pair. For example, for any data pair, if the two data points in the data pair are... Let's consider the data collected at times t0 and t1 after filtering. Since t0 precedes t1 on the time axis, the lag time of this data pair is the time interval between t0 and t1, and the time range of this data pair is the continuous time interval from t0 to t1. Next, we determine whether the time range of the i-th data pair intersects or crosses with the time range of the j-th data pair on the time axis. If they intersect, it means that the causal relationship represented by these two data pairs is contradictory and cannot be true simultaneously. Subsequently, we need to increase the difference measurement distance between these two data pairs. To facilitate the subsequent identification of such data pairs, this embodiment uses -1 as the distance between the i-th and j-th data pairs. If the constraint factors between the two data pairs are disjoint, it means that the causal relationship represented by these two data pairs is not contradictory and can be simultaneously established. There is no need to amplify the difference metric distance between these two data pairs. To distinguish this from the data pairs requiring amplification, this embodiment uses 1 as the constraint factor between the i-th and j-th data pairs. Then, the product of the first distance value, the second distance value, and the constraint factor between the i-th and j-th data pairs is calculated and recorded as the original difference metric distance between the i-th and j-th data pairs. Next, it is determined whether the original difference metric distance between the i-th and j-th data pairs is less than 0. Since the first and second distance values ​​cannot be less than 0,... Therefore, when the original difference measurement distance between the i-th data pair and the j-th data pair is less than 0, it means that the causal relationship represented by these two data pairs is contradictory and cannot be true at the same time. It is necessary to increase the difference measurement distance between these two data pairs. In this embodiment, the preset distance value is used as the target difference measurement distance between the i-th data pair and the j-th data pair. If the original difference measurement distance between the i-th data pair and the j-th data pair is not less than 0, it means that it is not necessary to increase the difference measurement distance between these two data pairs. Therefore, the original difference measurement distance between the i-th data pair and the j-th data pair can be directly recorded as the target difference measurement distance between the i-th data pair and the j-th data pair.Furthermore, in practical applications, implementers need to set a preset distance value according to the actual situation. However, the preset distance value must be much larger than the range of values ​​when the original difference metric distance is not less than 0, or much larger than the range of values ​​of the absolute value of the original difference metric distance. This is to avoid two data pairs with contradictory or impossible causal relationships being assigned to the same cluster during subsequent clustering. For example, if the range of values ​​when the original difference metric distance is not less than 0 is 0 to 1, then in this embodiment, 2 can be set as the preset distance value. In this embodiment, the expression for the target difference metric distance between the i-th data pair and the j-th data pair under the f-th dimension is: ; in, Measure the distance between the target difference between the i-th data pair and the j-th data pair in the f-th dimension. The distance is the original difference metric between the i-th data pair and the j-th data pair in the f-th dimension, where D0 is a preset distance value. Let be the lag association strength of the i-th data pair under the f-th dimension. Let be the lag association strength of the j-th data pair in the f-th dimension. Let be the lag time of the i-th data pair in the f-th dimension. Let be the lag time of the j-th data pair in the f-th dimension. It is the constraint factor between the i-th data pair and the j-th data pair. Used to measure the consistency in association strength between the i-th data pair and the j-th data pair. The smaller the value, the closer the correlation strength between the i-th data pair and the j-th data pair, indicating higher consistency or more similar lag patterns; Used to measure the consistency between the i-th data pair and the j-th data pair over the lag time. The smaller the value, the closer the lag time is to that of the i-th data pair and the j-th data pair; It is a term used to constrain the rationality of temporal logic; if the temporal relationship between two data pairs is reasonable, If the time sequence relationship between the two data pairs is unreasonable, then the result is a positive number. The value is negative to avoid grouping data pairs with unreasonable temporal relationships into the same category during subsequent clustering. The smaller smaller and When it is a positive number, smaller or The smaller, The smaller the value, the more likely the two data pairs belong to the same physical lag pattern, and the more likely they are to be grouped into the same cluster in subsequent clustering; conversely, the larger the value, the more likely they are to be grouped into the same cluster. The larger the value, the more likely the two data pairs represent different physical lag patterns, and the more likely they are to be separated into different clusters in subsequent clustering. This serves as the basis for subsequent clustering.

[0019] Therefore, this embodiment can obtain the target difference metric distance between any two data pairs in each dimension at the current time through the above-described method. After obtaining the target difference metric distance, this embodiment performs HDBSCAN density clustering on the data pairs in the same dimension based on the target difference metric distance between any two data pairs in each dimension, thereby obtaining clusters in each dimension. For example, for the f-th dimension, based on the target difference metric distance between any two data pairs in the f-th dimension, HDBSCAN density clustering is performed on all data pairs in the f-th dimension, and the resulting clusters are all clusters in the f-th dimension. Furthermore, the process of clustering data pairs using the HDBSCAN density clustering algorithm, given the target difference metric distance between them, is a well-known technique. The purpose of clustering in this embodiment is to select the data pairs that best conform to actual physical laws from a large number of potential lagging correlation data pairs. The set of consistent and stable lag relationships, i.e., the lag relationship between pumping wells and injection wells, is not fixed but dynamically changes with factors such as formation conditions and operating status. The core purpose of clustering is to find groups of data point pairs that show consistency in lag time and lag correlation strength from these complex data pairs. These groups represent the current relatively stable lag response pattern. The direct output of clustering is the set of optimal lag relationships (i.e., the cluster with the highest confidence). The data pairs in this set are considered to most reliably reflect the true lag correspondence between the two processed monitoring data sequences of pumping well a and injection well b under various dimensions. Based on the data pairs of the optimal clusters, the lag correlation strength and lag time for subsequent modeling can be extracted relatively reliably, thereby improving the accuracy and robustness of lag characteristic estimation to ensure the accuracy of subsequent dynamic prediction and adaptive control.

[0020] Based on the above description, after obtaining the clusters under each dimension, this embodiment selects the optimal clusters corresponding to each dimension based on the mean distance and the mean lag association strength of the target difference measure between the data pairs within each cluster under each dimension. Taking the specific process of obtaining the optimal cluster corresponding to the f-th dimension as an example, the specific process of obtaining the optimal cluster corresponding to the f-th dimension is as follows: For the s-th cluster under the f-th dimension, within the s-th cluster, the nearest data pair to each data pair is obtained and denoted as the intra-cluster nearest neighbor data pair. The linearly normalized result of the target difference metric distance between each data pair in the s-th cluster and its intra-cluster nearest neighbor data pair is denoted as the intra-cluster nearest neighbor distance of the corresponding data pair. The reflection result of the mean intra-cluster nearest neighbor distance of all data pairs in the s-th cluster is compared with the linear correlation strength of the mean lag association strength of all data pairs in the s-th cluster. The product of the normalization results is denoted as the selection confidence of the s-th cluster. Here, reflection refers to the complement of the mean intra-cluster nearest neighbor distance of all data pairs in the s-th cluster. The other reflection processes are similar. The mean intra-cluster nearest neighbor distance of all data pairs in the s-th cluster ranges from 0 to 1. The intra-cluster nearest neighbor data pair of any data pair in the s-th cluster is the data pair with the smallest target difference metric distance among the other data pairs in the s-th cluster. The expression for the selection confidence of the s-th cluster is: in, Let be the confidence level for selecting the s-th cluster. Let be the number of data pairs within the s-th cluster. Let be the intra-cluster nearest neighbor distance of the u-th data pair within the s-th cluster, and also the target difference metric distance between the u-th data pair and its intra-cluster nearest neighbor data pair. Let be the mean of the lagged association strength of all data pairs within the s-th cluster; The smaller the value, the higher the consistency, similarity, and intra-cluster compactness of the data pairs within the s-th cluster. This also indicates that the data pairs within the cluster exhibit greater consistency and stability in terms of lag time and lag correlation strength, thus better reflecting the characteristics of true lag relationships. Consequently, it also suggests that the lag characteristics of the data pairs within the s-th cluster better characterize the lag characteristics or lag relationship between pumping well a and injection well b in the f-th dimension. The smaller the value, the more reliably the s-th cluster can reflect the true lag correspondence between pumping well a and injection well b under the current f-th dimension. It also indicates that the confidence of the s-th cluster is higher or the probability that the s-th cluster is the optimal cluster is higher. The larger the value, the stronger the lag association of the data pairs in the s-th cluster, and the more it conforms to the characteristics of the true lag relationship. This also indicates that the lag characteristics of the data pairs within the s-th cluster better characterize the lag characteristics or lag relationship between pumping well a and injection well b in the f-th dimension. The larger the value, the more reliably the s-th cluster reflects the true lag correspondence between pumping well a and injection well b under the current f-th dimension, indicating a higher confidence level for the s-th cluster or a higher probability that the s-th cluster is the optimal cluster; because The greater the sum The smaller, The larger, therefore The larger the confidence level, the more reliably the s-th cluster can reflect the true lag correspondence between pumping well a and injection well b under the current f-th dimension. A higher confidence level for the s-th cluster or a higher probability that the s-th cluster is the optimal cluster increases, and vice versa. The smaller the value, the less the s-th cluster can reflect the true lag correspondence between pumping well a and injection well b under the current f-th dimension. The lower the confidence of the s-th cluster or the lower the probability that the s-th cluster is the optimal cluster.

[0021] In this embodiment, after obtaining the selection confidence of each cluster under the f-th dimension, the cluster with the highest selection confidence is selected as the optimal cluster corresponding to the f-th dimension from all the clusters under the f-th dimension; that is, the optimal cluster corresponding to the f-th dimension can most reliably reflect the true lag correspondence between pumping well a and injection well b under the current f-th dimension compared to other clusters.

[0022] In this embodiment, after selecting the optimal clusters corresponding to each dimension, the predicted lag time and predicted lag association strength between pumping well a and injection well b at future times are estimated based on the lag time and lag association strength of each data pair within the optimal clusters corresponding to each dimension. The process of obtaining the predicted lag time and predicted lag association strength between pumping well a and injection well b at the f-th dimension is illustrated using this example. The specific process of obtaining the predicted lag time and predicted lag association strength between pumping well a and injection well b at the f-th dimension at future times is as follows: Sort all data pairs within the optimal cluster corresponding to the f-th dimension according to their chronological order or the order of data collection time within the data pairs. This yields the data pair sequence for the optimal cluster corresponding to the f-th dimension. In this sequence, data pairs ranked earlier in the order have data collected earlier overall. Then, the sequences consisting of the lag times and lag correlation strengths of the data pairs within this sequence are denoted as the lag time sequence and lag correlation strength sequence for the current time in the f-th dimension, respectively. The positions of the lag time and lag association strength of each data pair in the optimal cluster in the lag time series and lag association strength series corresponding to the f-th dimension correspond to the positions of the data pairs in the data pair series. Then, the lag time series and lag association strength series corresponding to the f-th dimension are input into the LSTM prediction model corresponding to the f-th dimension. The model outputs the predicted lag time and predicted lag association strength between pumping well a and injection well b in the f-th dimension at future times. That is, in this embodiment, one LSTM prediction model is constructed for each dimension. The construction and training process of the LSTM prediction model is a well-known technology, so it will not be described further. Furthermore, since the optimal cluster can most reliably reflect the true lag relationship between pumping well a and injection well b under the current dimensions compared to other clusters, the lag time and lag correlation strength predicted based on the lag time and lag correlation strength of each data pair within the optimal cluster are more credible and reliable. The purpose of prediction is to achieve advanced perception and proactive control, or to achieve early perception and trend analysis of the groundwater dynamics propagation process, that is, to make decisions and actions in advance, thereby achieving precise linkage hydrodynamic control of pumping and injection wells, which can also significantly improve the stability, real-time performance and overall control accuracy of resource extraction or allocation systems in complex underground environments. In this embodiment, the future time is a continuous time interval after the current time, for example, the future 1 hour starting from the current time can be set.

[0023] Therefore, this embodiment can obtain the predicted lag time and predicted lag correlation strength between pumping wells and injection wells with hydraulic connection in the target area at future time through the above process. The predicted lag time and predicted lag correlation strength are the key to the subsequent construction of the lag time network and lag correlation network in various dimensions at future time. The lag time network and lag correlation network are the key to realizing adaptive regulation.

[0024] Step S003: Based on the predicted lag time and predicted lag correlation strength between pumping wells and injection wells that have hydraulic connections in the same dimension, lag time networks and lag correlation networks in each dimension are constructed respectively. Based on the lag time networks and lag correlation networks, the pumping wells and injection wells in the target area are adaptively linked and hydrodynamically regulated.

[0025] In this embodiment, after obtaining the predicted lag time and predicted lag correlation strength between hydraulically connected pumping wells and injection wells within the target area at future times, the lag time network and lag correlation network for each dimension at future times are constructed based on the predicted lag time and predicted lag correlation strength between hydraulically connected pumping wells and injection wells in the same dimension at future times. The specific construction process of the lag time network and lag correlation network for the f-th dimension at future times is as follows: first, each pumping well and each injection well in the target area is treated as a node, or all pumping wells and injection wells in the target area are numbered as nodes, and each well is treated as a network node. The system first manages all points uniformly; then, directed edges are established between pumping wells and injection wells with hydraulic connections, pointing from injection wells to pumping wells, thus obtaining a pumping-injection well relationship network. Nodes in this network are either pumping wells or injection wells, and edges exist between hydraulically connected pumping and injection wells. The predicted lag time between hydraulically connected pumping and injection wells in the f-th dimension at future time is then used as the edge weight of the corresponding edge in the pumping-injection well relationship network, forming a lag time network in the f-th dimension at future time. The predicted lag correlation strength between hydraulically connected pumping and injection wells in the f-th dimension at future time is used as the pumping-injection well relationship network. The edge weights of corresponding edges in the network form a lag-connected network in the f-th dimension at a future time. The edge weights in the lag-time network reflect the temporal order and response speed of underground pressure propagation, while the edge weights in the lag-connected network characterize the degree of influence and coupling strength of groundwater dynamic propagation between different wells, thus forming a dynamic coupling topology among multiple pumping and injection wells in the target area. Both the lag-time network and the lag-connected network are constructed by establishing directed edges between pumping and injection wells with hydraulic connections, using pumping and injection wells in the target area as nodes. The lag-time network in the f-th dimension at a future time contains pumping and injection wells with hydraulic connections... The edge weights between injection wells represent the predicted lag time between the corresponding pumping well and injection well. In the future, at the f-th dimension, the edge weights between pumping wells and injection wells with hydraulic connections on the lag correlation network represent the predicted lag correlation strength between the corresponding pumping well and injection well. For example, in the future, at the f-th dimension, the edge weights between pumping well a and injection well b on the lag time network represent the predicted lag time between pumping well a and injection well b in the f-th dimension, and the edge weights between pumping well a and injection well b on the lag correlation network in the f-th dimension represent the predicted lag correlation strength between pumping well a and injection well b in the f-th dimension.

[0026] Therefore, this embodiment can obtain the lag time network and lag correlation network under each dimension at different times through the above process. The method for obtaining the lag time network and lag correlation network under each dimension at future times and the lag time network and lag correlation network under each dimension at historical times is the same as the method for obtaining the lag correlation network and lag correlation network under the f-th dimension at the future time. Therefore, it will not be described in detail.

[0027] After obtaining the time lag network and the lag correlation network, adaptive linkage hydrodynamic control is performed on the pumping wells and injection wells in the target area. Specifically, this embodiment first inputs the multi-dimensional processed monitoring data sequence of each well in the target area at the current moment, along with the time lag network and lag correlation network at various dimensions in historical and future times, into the spatiotemporal linkage neural network model. The spatiotemporal linkage neural network model includes a graph neural network (GNN) and an LSTM network. That is, the multi-dimensional processed monitoring data sequence of each well obtained at the current moment, along with the time lag network and lag correlation network at various dimensions in historical and future times, are input into the graph neural network and LSTM network. The graph neural network outputs spatial propagation correlation features, and the LSTM network outputs... The TM network outputs temporal evolution features; then, based on the spatial propagation correlation features and temporal evolution features extracted by the graph neural network and LSTM network, model predictive control (MPC) or reinforcement learning control algorithms are used to make coordinated decisions on pumping wells and injection wells at future times. That is, based on the results output by the spatiotemporal linkage neural network model, coordinated hydrodynamic regulation is carried out on pumping wells and injection wells in the target area; and for the f-th dimension, the information input to the graph neural network and LSTM network includes the processed monitoring data sequence of the f-th dimension in the multi-dimensional processed monitoring data sequence of each well in the target area at the current time, the lag time network and lag correlation network in the f-th dimension at each historical time, and the lag time network and lag correlation network in the f-th dimension at future times, and the same applies to other dimensions.

[0028] In this embodiment, the specific process of coordinating the hydrodynamic control of pumping wells and injection wells in the target area is as follows: In this embodiment, after obtaining the multi-dimensional processed monitoring data sequence of each well at the current moment, the lag time network and lag correlation network under each dimension at historical moments, and the lag time network and lag correlation network under each dimension at future moments, the multi-dimensional processed monitoring data sequence of each well at the current moment, the lag time network and lag correlation network under each dimension at historical moments, and the lag time network and lag correlation network under each dimension at future moments are first used as the node features and edge features input to the spatiotemporal linkage neural network model. The model outputs the spatial propagation correlation features between wells and the temporal evolution features of the monitoring data in each dimension. That is, the multi-dimensional processed monitoring data sequence of each well at the current moment, the lag time network and lag correlation network under each dimension at historical moments, and the lag time network and lag correlation network under each dimension at future moments are jointly input into the spatiotemporal linkage neural network model. The system is trained and learned to extract the dynamic linkage relationships between different wells, different monitoring data, and different propagation paths. The spatiotemporal linkage neural network model includes a graph neural network (GNN) and an LSTM network. The graph neural network outputs the spatial propagation correlation features between corresponding well nodes to characterize the coupling propagation strength and influence path between different wells. The LSTM network outputs the temporal evolution features of monitoring parameters in each dimension to characterize the dynamic trend features of monitoring parameters such as pressure and flow rate over time. In other words, the graph neural network is used to learn the spatial propagation relationship, inter-well influence path, and coupling strength features between different pumping wells and injection wells to extract the spatial propagation correlation features between different wells. The LSTM network is used to learn the dynamic evolution process of time-series monitoring data such as pressure and flow rate over time to extract the temporal change features between monitoring parameters in different dimensions. Subsequently, the spatial propagation correlation features extracted by the graph neural network and the temporal evolution features extracted by the LSTM are fused to obtain fused features, or the spatial propagation correlation features extracted by the graph neural network and the temporal evolution features extracted by the LSTM are fused to form spatiotemporal joint features. The fused features or spatiotemporal joint features are used to characterize the dynamic linkage relationship between different wells, between monitoring parameters of different dimensions, and between different propagation paths, and output the change status and trend of monitoring parameters such as pressure and flow rate at future time and the results of inter-well linkage impact. Specifically, the fusion here refers to the spatiotemporal alignment processing of the spatial propagation correlation features extracted by the graph neural network and the temporal evolution features extracted by the LSTM, that is, to uniformly map the spatial propagation correlation features and temporal evolution features corresponding to different well nodes at the same time, and to perform feature correspondence according to the corresponding well nodes and time series. The feature splicing fusion method is used to splice the spatial propagation correlation features and temporal evolution features under the corresponding well nodes to form a spatiotemporal joint feature vector, which is used to characterize the spatiotemporal dynamic linkage relationship in the groundwater dynamic propagation process between different wells.

[0029] After completing the learning of multi-dimensional linkage relationships, model predictive control (MPC) or reinforcement learning control algorithms can be further combined to make linkage decisions on the operating status between pumping wells and injection wells at future times. That is, this embodiment is based on the fusion features or spatiotemporal joint features obtained above, and combined with model predictive control (MPC) or reinforcement learning control algorithms to adaptively link and regulate the hydrodynamics of pumping wells and injection wells in the target area at future times. Moreover, the regulation in this embodiment is executed at the current time, but its goal is to deal with the predicted future risks. For example, when a pumping well is predicted to experience a future pressure decline, the system automatically increases the injection volume of the corresponding injection well, adjusts the injection frequency, or reduces the pumping rate of the pumping well based on the hysteresis propagation relationship between the pumping well and other injection wells, the strength of the correlation, and the coupling relationship between other dimensional parameters. This is done to compensate for changes in underground pressure in advance. When a pumping well is predicted to have excessively high pressure, abnormal flow, or the risk of multiple wells propagating and overlapping, the system dynamically adjusts the injection ratio among multiple injection wells and coordinates with neighboring pumping wells to release pressure in a coordinated manner. This reduces the control lag problem caused by the time lag of groundwater dynamic propagation. After completing the coordinated control, this embodiment can continue to collect feedback data from each well in real time and compare and analyze the actual operating results with the model prediction results. Then, Bayesian parameter updates or online incremental learning methods can be used to dynamically correct the prediction error and the propagation relationship between wells. When changes are found in the actual propagation time, coupling strength, or multi-dimensional parameter linkage relationship between wells, the corresponding hysteresis parameters, network edge weights, and prediction model parameters are automatically updated. This enables the system to continuously optimize the coordinated control strategy according to the changes in the seepage state of the underground medium, thereby forming a closed-loop hydrodynamic adaptive control system of sensor acquisition, hysteresis analysis, network construction, future trend prediction, coordinated control, and feedback correction. This improves the stability, real-time performance, and control accuracy of the coordinated operation of multiple injection and extraction wells in complex underground environments.

[0030] Thus, this embodiment completes the adaptive control of the linkage between injection and extraction wells' hydrodynamics. This embodiment first analyzes the time-series data, such as pressure, collected by sensors. It extracts feature data representing key trends in the sequence using methods such as first-order difference and piecewise fitting, and calculates the local change intensity of each feature data point to quantify the significance and stability of the change, achieving precise location and quantitative evaluation of key events in the dynamic response. Then, based on these feature data and their change intensities, it calculates the lag correlation strength of feature data combinations between different wells of the same dimension, and constructs a difference metric distance that comprehensively reflects the temporal logic, correlation strength, and consistency of lag time. Finally, it uses density clustering to screen out the combinations with the highest lag characteristics from all potential combinations. The system uses the most stable and consistent set of data pairs to reliably extract the time-varying lag correlation strength sequence and lag time sequence between pumping and injection wells. Finally, these key time-series features characterizing the dynamic lag relationship are used as input to construct an LSTM prediction model to achieve dynamic prediction of future lag states. The prediction results are then fed back to the control system to drive adaptive adjustment of parameters such as pumping intensity and injection rate. Furthermore, the control strategy provided in this embodiment can effectively characterize and track time-varying lag characteristics under complex conditions, achieving a leap from fixed time delay to dynamic adaptive control, and significantly improving the collaborative operation efficiency and control accuracy of multi-well systems.

[0031] In summary, this embodiment first acquires multi-dimensional processed monitoring data sequences of pumping wells and injection wells in the target area, as well as the feature data and local change intensity of the feature data in each processed monitoring data sequence; then, it combines the feature data in the processed monitoring data sequences of pumping wells and injection wells that have hydraulic connections in the same dimension pairwise to obtain data pairs in each dimension. Based on the length of the local sub-segments containing the data in each data pair, the alignment result of the local sub-segments, and the local change intensity of the data in the data pair, the hysteresis correlation strength of each data pair is obtained. Based on the difference in hysteresis correlation strength, hysteresis time difference, and time range relationship between any two data pairs in the same dimension, the target difference metric distance between any two data pairs in each dimension is obtained. Clustering is performed on data pairs in each dimension using the target difference metric distance to obtain clusters for each dimension. The optimal clusters for each dimension are selected based on the mean target difference metric distance and the mean lag correlation strength between data pairs within the clusters. Based on the lag time and lag correlation strength of data pairs within the optimal clusters, the predicted lag time and predicted lag correlation strength between hydraulically connected pumping wells and injection wells are obtained in each dimension. Finally, based on the predicted lag time and predicted lag correlation strength between hydraulically connected pumping wells and injection wells in the same dimension, lag time networks and lag correlation networks are constructed for each dimension. Based on the lag time networks and lag correlation networks, the pumping wells and injection wells in the target area are adaptively linked for hydrodynamic regulation. Furthermore, this embodiment can significantly improve the efficiency, stability, and accuracy of coordinated hydrodynamic regulation of pumping wells and injection wells by dynamically predicting future lag states and feeding the prediction results back to the control system. In other words, by dynamically predicting the predicted lag time and predicted lag correlation strength between pumping wells and injection wells that have hydraulic connections under different dimensions, it can significantly improve the stability and accuracy of coordinated hydrodynamic regulation of pumping wells and injection wells.

[0032] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 this application, and should all be included within the protection scope of this application.

Claims

1. A sensor-data-based integrated hydrodynamic adaptive control system for injection and extraction wells, comprising a processor and a memory, characterized in that, The processor executes the computer program stored in the memory to perform the following steps: The system acquires multi-dimensional processed monitoring data sequences of pumping wells and injection wells in the target area, as well as characteristic data and local change intensity of characteristic data in each processed monitoring data sequence. The dimensions include at least the pressure, flow velocity, water level, and water quality parameters in the well. For pumping well a and injection well b that are hydraulically connected, the feature data in the processed monitoring data sequences of pumping well a and injection well b under the same dimension are combined in pairs to obtain data pairs under each dimension. Based on the length of the local sub-segment containing the data in each data pair, the alignment result of the local sub-segment, and the intensity of local change in the data in the data pair, the lag correlation strength of each data pair is obtained. Based on the difference in lag correlation strength, lag time difference, and time range relationship between any two data pairs under the same dimension, the target difference metric distance between any two data pairs under each dimension is obtained. Based on the target difference metric distance, the data pairs under each dimension are clustered to obtain clusters under each dimension. Based on the mean of the target difference metric distance and the mean of the lag correlation strength between data pairs within the cluster, the optimal cluster corresponding to each dimension is selected. Based on the lag time and lag correlation strength of the data pairs within the optimal cluster, the predicted lag time and predicted lag correlation strength between pumping well a and injection well b under each dimension are obtained. Based on the predicted lag time and predicted lag correlation strength between pumping wells and injection wells that have hydraulic connections in the same dimension, lag time networks and lag correlation networks in each dimension are constructed respectively. Based on the lag time networks and lag correlation networks, the pumping wells and injection wells in the target area are adaptively linked for hydrodynamic regulation.

2. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 1, characterized in that, Methods for obtaining feature data include: For any processed monitoring data sequence, the data segments corresponding to the consecutive difference numerator sequences with the same sign in the first-order difference sequence of the processed monitoring data sequence are all recorded as data subsequences; the monitoring data corresponding to the sign change position in the first-order difference sequence are used as feature data; linear fitting is performed on each data subsequence, and the monitoring data with the largest fitting residual in each data subsequence is selected as feature data.

3. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 1, characterized in that, Methods for obtaining the intensity of local changes in feature data include: For any feature data in any processed monitoring data sequence, the data segment from the preceding feature data to the feature data and the data segment from the feature data to the following feature data are respectively denoted as the left local data segment and the right local data segment of the feature data. The first-order difference sequences of the left local data segment and the right local data segment are respectively denoted as the first difference sequence and the second difference sequence. The normalized result of the absolute difference between the mean of the first difference sequence and the mean of the second difference sequence is denoted as the representative value of the degree of change. The inverse normalized result of the maximum value of the mean absolute deviation of the first difference sequence and the mean absolute deviation of the second difference sequence is denoted as the confidence coefficient of the degree of change. The normalized result of the product of the representative value of the degree of change and the confidence coefficient of the degree of change is denoted as the local change intensity of the feature data.

4. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 1, characterized in that, Methods for obtaining the lagged correlation strength of any data pair include: The two feature data in the data pair are denoted as data 1 and data 2, respectively. The local segments of data 1 and data 2 are denoted as local segment 1 and local segment 2, respectively. Using data 1 in local segment 1 and data 2 in local segment 2 as alignment reference points, local segment 1 and local segment 2 are aligned on the time axis. The overlapping data segments of local segment 1 and local segment 2 on the time axis during alignment are extracted and denoted as the first overlapping segment and the second overlapping segment, respectively. Based on the correlation coefficient between the first overlapping segment and the second overlapping segment, the length difference between local segment 1 and local segment 2, and the difference in local change intensity between data 1 and data 2, the hysteresis correlation strength of the data pair is obtained. The data segment from the preceding feature data to the following feature data of any feature data is the local segment of the feature data.

5. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 4, characterized in that, A method for obtaining the hysteresis correlation strength of the data pair based on the correlation coefficient between the first overlapping segment and the second overlapping segment, the length difference between local segment 1 and local segment 2, and the difference in local change intensity between data 1 and data 2 includes: The absolute value of the Spearman rank correlation coefficient between the first overlapping segment and the second overlapping segment is denoted as the first characterization value; the inverse normalized result of the absolute difference in length between local segment 1 and local segment 2 is denoted as the second characterization value; the inverse normalized result of the absolute difference in local change intensity between data 1 and data 2 is denoted as the third characterization value; and the product of the first characterization value, the second characterization value and the third characterization value is denoted as the hysteresis correlation strength of the data pair.

6. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 1, characterized in that, The method for obtaining the target difference metric distance between the i-th data pair and the j-th data pair in the f-th dimension includes: The normalized result of the absolute difference in lag correlation strength between the i-th data pair and the j-th data pair is recorded as the first distance value; the normalized result of the absolute difference in lag time between the i-th data pair and the j-th data pair is recorded as the second distance value; it is determined whether the time range of the i-th data pair intersects with the time range of the j-th data pair on the time axis. If they intersect, -1 is used as the constraint factor; if they do not intersect, 1 is used as the constraint factor; the product of the first distance value, the second distance value, and the constraint factor is recorded as the original difference metric distance between the i-th data pair and the j-th data pair. If the original difference metric distance is not less than 0, it is recorded as the target difference metric distance between the i-th data pair and the j-th data pair. If the original difference metric distance is less than 0, a preset distance value is used as the target difference metric distance between the i-th data pair and the j-th data pair.

7. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 6, characterized in that, The lag time of any data pair is the time interval between the acquisition of the two data points in the corresponding data pair, and the time range of any data pair refers to the continuous time interval formed by the acquisition times of the two data points in the data pair.

8. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 1, characterized in that, The methods for obtaining the optimal cluster corresponding to the f-th dimension include: Among all clusters in the f-th dimension, the cluster with the highest selection confidence is denoted as the optimal cluster corresponding to the f-th dimension. The selection confidence of any cluster is the product of the reflection result of the mean intra-cluster nearest neighbor distance of all data pairs in the cluster and the normalized result of the mean lag association strength of all data pairs in the cluster. The intra-cluster nearest neighbor distance of any data pair in any cluster is the normalized result of the target difference metric distance between the data pair and the nearest neighbor data pair in the cluster.

9. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 7, characterized in that, Methods for obtaining predicted lag time and predicted lag correlation strength include: The lag time and lag association strength of the data pairs within the optimal cluster corresponding to each dimension are sorted according to the time sequence of the data pairs to obtain the lag time series and lag association strength series corresponding to each dimension. The lag time series and lag association strength series corresponding to each dimension are then input into the LSTM prediction model to output the predicted lag time and predicted lag association strength between pumping well a and injection well b under each dimension.

10. The sensor data-based injection-extraction well linkage hydrodynamic adaptive control system as described in claim 1, characterized in that, Both the time lag network and the lag association network are constructed by using pumping wells and injection wells in the target area as nodes and establishing directed edges between pumping wells and injection wells that have hydraulic connections. The edge weights between pumping wells and injection wells that have hydraulic connections in the time lag network are the predicted lag times between the corresponding pumping wells and injection wells, while the edge weights between pumping wells and injection wells that have hydraulic connections in the lag association network are the predicted lag association strength between the corresponding pumping wells and injection wells.

Images