Coal bed gas productivity prediction method and system combined with artificial intelligence

Abstract

The invention provides a coal bed gas productivity prediction method and system combined with artificial intelligence, and the method comprises the steps: firstly obtaining a multi-source coal bed gas productivity data set, which comprises geological structure data, reservoir physical property data and historical production dynamic data, of a target block, and then carrying out the time-space alignment processing of the multi-source coal bed gas productivity data set, the method comprises the following steps: obtaining a standardized coal bed gas productivity characteristic set, then performing characteristic extraction on the standardized coal bed gas productivity characteristic set to generate a prediction characteristic set containing geological attribute characteristic and dynamic attribute characteristic dual codes, and then performing iterative prediction on the prediction characteristic set based on a reinforcement learning strategy network. Finally, an optimization strategy set is generated according to the target coalbed methane productivity prediction result and error distribution of historical production dynamic data and fed back to a coalbed methane production control system to trigger parameter adjustment operation, and therefore the accuracy of coalbed methane productivity prediction can be improved, and optimization control over the production process is achieved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more particularly, to a coalbed methane production capacity prediction method and system combining artificial intelligence. Background Art

[0002] As an important unconventional natural gas resource, the prediction of coalbed methane production capacity is of crucial significance for the efficient development and rational utilization of coalbed methane. In the traditional field of coalbed methane production capacity prediction, early methods did not fully consider the spatio-temporal characteristics of data during the data processing process. The inconsistency of data from different sources in time and space makes it difficult to effectively associate and integrate the data, thus affecting the subsequent analysis and prediction effects.

[0003] In addition, most traditional methods are based on simple statistical analysis or fixed mathematical models and cannot dynamically adapt to the continuously changing situations in the coalbed methane production process. These models lack the ability to mine potential complex relationships in the data and cannot well handle the interaction between geological attribute features and dynamic attribute features, resulting in difficulty in accurately predicting the coalbed methane production capacity in practical applications. Moreover, when there is a deviation between the prediction result and the actual production situation, traditional methods lack an effective feedback mechanism to timely adjust production parameters and cannot achieve optimized control of the coalbed methane production process. Therefore, the existing coalbed methane production capacity prediction methods have deficiencies in terms of accuracy, adaptability, and optimized control, and there is an urgent need for a more advanced and efficient prediction method to meet the requirements of actual production. Summary of the Invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, embodiments of the present invention provide a coalbed methane production capacity prediction method combining artificial intelligence, and the method includes: Obtain a multi-source coalbed methane production capacity data set of a target block, where the multi-source coalbed methane production capacity data set includes geological structure data, reservoir physical property data, and historical production dynamic data; Perform spatio-temporal alignment processing on the multi-source coalbed methane production capacity data set to generate a standardized coalbed methane production capacity feature set; Extract features from the standardized coalbed methane production capacity feature set to generate a prediction feature set, where each feature unit in the prediction feature set includes dual coding of geological attribute features and dynamic attribute features; Based on a reinforcement learning policy network, iteratively predict the geological attribute features and dynamic attribute features to generate a target coalbed methane production capacity prediction result; Generate an optimization strategy set according to the error distribution between the target coalbed methane production capacity prediction result and the historical production dynamic data, and feedback the optimization strategy set to the coalbed methane production control system to trigger parameter adjustment operations.

[0005] On the other hand, an embodiment of the present invention further provides a coalbed methane production capacity prediction system combined with artificial intelligence, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0006] Based on the above aspects, the embodiment of the present invention obtains a multi-source coalbed methane production capacity data set including geological structure, reservoir physical properties and historical production dynamic data, and performs spatio-temporal alignment processing to generate a standardized coalbed methane production capacity feature set, effectively solving the problem that it is difficult to integrate multi-source data with inconsistent spatio-temporal. Feature extraction is performed on the standardized coalbed methane production capacity feature set to generate a prediction feature set including dual coding of geological attribute features and dynamic attribute features, which can comprehensively and deeply explore the potential complex relationships between geological attributes and dynamic attribute features in the data. Based on the reinforcement learning policy network, iterative prediction is performed on the prediction feature set, which can dynamically adapt to the changes in the coalbed methane production process. Compared with traditional fixed models, it can generate more accurate target coalbed methane production capacity prediction results. According to the error distribution between the target coalbed methane production capacity prediction result and the historical production dynamic data, an optimization strategy set is generated and fed back to the coalbed methane production control system to trigger parameter adjustment operations, realizing the improvement of the accuracy of coalbed methane production capacity prediction and the optimization control of the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a schematic execution flow diagram of a coalbed methane production capacity prediction method combined with artificial intelligence provided by an embodiment of the present invention.

[0008] Figure 2 is a schematic diagram of exemplary hardware and software components of a coalbed methane production capacity prediction system combined with artificial intelligence provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0009] The present invention will be specifically described below with reference to the accompanying drawings of the specification. Figure 1 is a schematic flow diagram of a coalbed methane production capacity prediction method combined with artificial intelligence provided by an embodiment of the present invention. The coalbed methane production capacity prediction method combined with artificial intelligence will be introduced in detail below.

[0010] Step S110: Obtain a multi-source coalbed methane production capacity data set of a target block, where the multi-source coalbed methane production capacity data set includes geological structure data, reservoir physical property data and historical production dynamic data.

[0011] In this embodiment, geological structure data is the key information reflecting the geological environment where the coal seam is located. For example, through seismic exploration technology, by utilizing the propagation characteristics of seismic waves in underground media, analyzing reflected waves, refracted waves, etc., information such as the structure and morphology of the strata can be obtained, and these information can be used to determine parameters such as formation dip angle and fault location; drilling is a method to directly obtain underground core samples. By observing and analyzing the cores, the rock types, fracture development conditions, etc. of the strata can be understood more accurately.

[0012] Reservoir physical property data is related to the physical properties of the coal seam, such as porosity, permeability, etc. These properties directly affect the storage and flow capacity of coalbed methane. Obtaining reservoir physical property data usually involves professional analysis of the collected core samples in the laboratory, and relevant parameters are determined by measuring indicators such as the pore volume and fluid passing ability of the cores.

[0013] Historical production dynamic data records the past production status of coalbed methane, including gas production volume sequences, pressure change sequences, and production cycle stage labels, etc. These data can be collected from various monitoring devices deployed in the coalbed methane production process, such as gas flow sensors, pressure sensors, etc. They record various data in real time during the production process and store them in the database for subsequent analysis and use.

[0014] By collecting and integrating the above multi-source data, a multi-source coalbed methane production capacity data set containing geological structure data, reservoir physical property data, and historical production dynamic data is formed.

[0015] Step S120: Perform spatio-temporal alignment processing on the multi-source coalbed methane production capacity data set to generate a standardized coalbed methane production capacity feature set.

[0016] In this embodiment, since the data in the multi-source coalbed methane production capacity data set has diverse sources, different acquisition time and space ranges, and there are also differences in data formats and precisions, in order to ensure the consistency and comparability of the data, spatio-temporal alignment processing is required, and then a standardized coalbed methane production capacity feature set is generated. The purpose of spatio-temporal alignment processing is to unify data from different sources into the same time and space framework, enabling the data to be correlated and analyzed with each other.

[0017] Exemplarily, the specific spatio-temporal alignment processing process is as follows: Step S121: Extract the formation dip angle parameter, fault distribution parameter, and fracture development parameter in the geological structure data, and construct an initial geological feature set.

[0018] In this embodiment, key parameters are extracted from the geological structure data to construct an initial geological feature set. The formation dip angle parameter α can reflect the inclination degree of the formation, which has an important influence on the migration and accumulation of coalbed methane. The dip angle values of the formation at different positions can be calculated by analyzing seismic exploration data and combining with a geological model. The fault distribution parameter β describes the distribution of faults in the target block, including information such as the position, strike, and length of the faults. For example, it can be determined by comprehensively analyzing seismic exploration data, geological mapping, and drilling results. The fracture development parameter γ reflects the degree of fracture development in the coal seam, such as fracture density, aperture, etc. Relevant information can be obtained by microscopic observation of core samples, imaging logging, and other methods.

[0019] By combining these extracted formation dip angle parameter α, fault distribution parameter β, and fracture development parameter γ together, an initial geological feature set G = {α, β, γ} is formed.

[0020] Step S122: Extract the porosity parameter, permeability parameter, and gas content parameter from the reservoir physical property data to construct an initial reservoir feature set.

[0021] In this embodiment, key parameters are extracted from the reservoir physical property data to construct an initial reservoir feature set. The porosity parameter φ represents the proportion of the pore volume in the coal seam to the total volume, which is an important indicator for measuring the coalbed methane storage capacity of the coal seam. The specific value can be determined through laboratory porosity measurement experiments on core samples. The permeability parameter k reflects the ability of the coal seam to allow fluid to pass through, which is closely related to the flow of coalbed methane. The measurement of permeability usually adopts the steady-state method or the non-steady-state method, and the flow of fluid in the core is simulated in the laboratory to obtain it. The gas content parameter C represents the amount of coalbed methane contained in the coal seam, which can be obtained by directly measuring the gas content in the core sample or by estimating in combination with geological conditions and reservoir physical properties. The extracted porosity parameter φ, permeability parameter k, and gas content parameter C are combined to construct an initial reservoir feature set R = {φ, k, C}.

[0022] Step S123: Extract the gas production sequence, pressure change sequence, and production cycle stage label from the historical production dynamic data to construct an initial dynamic feature set.

[0023] In this embodiment, key information is extracted from historical production dynamic data to construct an initial dynamic feature set. The gas production rate sequence Q(t) records the production of coalbed methane at different times t, which reflects the time-varying law of coalbed methane production. The gas production rate data at different time points can be obtained from the production monitoring system to form a time series. The pressure change sequence P(t) describes the change of pressure with time during the production of coalbed methane, and the change of pressure will affect the desorption and flow of coalbed methane. Similarly, the pressure values at different times are recorded by pressure sensors to construct the pressure change sequence. The production cycle stage label L is used to identify different stages of coalbed methane production, such as the initial gas production stage, the stable gas production stage, the declining gas production stage, etc. According to the characteristics of the gas production rate sequence and the pressure change sequence, combined with production experience and professional knowledge, different production time periods can be classified into stages and corresponding labels are assigned. The gas production rate sequence Q(t), the pressure change sequence P(t) and the production cycle stage label L are combined to construct the initial dynamic feature set D = {Q(t), P(t), L}.

[0024] Step S124: Perform data preprocessing on the initial geological feature set, the initial reservoir feature set, and the initial dynamic feature set. The data preprocessing includes missing value filling, outlier correction, and dimension normalization.

[0025] In this embodiment, since there may be missing values and outliers in the data of the initial geological feature set G, the initial reservoir feature set R, and the initial dynamic feature set D, and the dimensions of each feature may be different, data preprocessing is required.

[0026] For missing value filling, various methods can be used. For example, for continuous data (such as the formation dip angle parameter α, the porosity parameter φ, etc.), the mean filling method can be used, that is, calculate the mean of all non-missing values of this feature and use this mean to fill the missing value; for discrete data (such as the production cycle stage label L), the mode filling method can be used, that is, find the label value that appears most frequently to fill the missing value.

[0027] For outlier correction, a reasonable threshold range can be set to determine whether the data is an outlier. For example, for the gas production rate sequence Q(t), if the gas production rate at a certain time point far exceeds the normal fluctuation range, then this value is considered an outlier. The outlier can be corrected by using a smoothing method, such as the moving average method, and replace the outlier with the average gas production rate of several time points before and after this time point.

[0028] Dimensional normalization is to eliminate the influence of dimensions between different features and make each feature have the same scale. The min-max normalization method can be used. For each feature, calculate its minimum value min and maximum value max, and then normalize each feature value x through the formula (x - min) / (max - min), mapping it to the interval [0, 1].

[0029] After the above data preprocessing operations, the initial geological feature set G', initial reservoir feature set R', and initial dynamic feature set D' after preprocessing are obtained.

[0030] Step S125: Align the time window and spatial grid of the initial geological feature set, initial reservoir feature set, and initial dynamic feature set after preprocessing to generate a standardized coalbed methane production capacity feature set, where each feature unit in the standardized coalbed methane production capacity feature set contains a unified timestamp identifier and spatial coordinate identifier.

[0031] In this embodiment, in order to achieve the unification of different feature sets in space and time, it is necessary to perform time window alignment and spatial grid alignment on the initial geological feature set G', initial reservoir feature set R', and initial dynamic feature set D' after preprocessing. Time window alignment is to divide the time data in different feature sets according to the same time interval to form a unified time window. For example, divide the gas production sequence Q(t), pressure change sequence P(t), and geological and reservoir feature data related to time into time windows of one day, so that the data within each time window is comparable. Spatial grid alignment is to divide the target block into several spatial grids, each grid having a unique spatial coordinate identifier. Match the spatial information in the geological structure data and reservoir physical property data with these grids, and map different feature data to the corresponding spatial grids. For example, according to the spatial positions of geological features such as the formation dip parameter α and fault distribution parameter β, assign them to the corresponding spatial grids. After time window alignment and spatial grid alignment, integrate the three preprocessed feature sets to generate a standardized coalbed methane production capacity feature set F. In this coalbed methane production capacity feature set, each feature unit contains a unified timestamp identifier t and spatial coordinate identifier (x, y), that is, F = {F(t, x, y)}, where F(t, x, y) represents the feature vector at time t and spatial coordinate (x, y), and this feature vector contains multi-faceted feature information such as geology, reservoir, and dynamics.

[0032] Step S130: Extract features from the standardized coalbed methane production capacity feature set to generate a prediction feature set, where each feature unit in the prediction feature set contains dual coding of geological attribute features and dynamic attribute features.

[0033] In this embodiment, in order to more effectively utilize the standardized coalbed methane production capacity feature set for production capacity prediction, it is necessary to perform feature extraction on it to generate a prediction feature set that includes dual coding of geological attribute features and dynamic attribute features. The purpose of feature extraction is to extract features that have an important impact on production capacity prediction from the original feature set, reduce the dimensionality of the data, and improve the accuracy and efficiency of prediction.

[0034] Exemplarily, the specific feature extraction process is as follows: Step S131: Perform regional segmentation processing on the initial geological feature set to generate multiple geological sub-region units.

[0035] In this embodiment, perform regional segmentation processing on the initial geological feature set G', and divide the target block into multiple geological sub-region units. It can be divided according to the similarity of geological structures. For example, regions with similar formation dip angles and similar fault distribution characteristics are divided into the same sub-region. Specifically, when operating, the clustering analysis method can be used, and geological features such as the formation dip angle parameter α and the fault distribution parameter β are used as clustering indicators to divide the space of the target block into different clustering clusters, and each clustering cluster corresponds to a geological sub-region unit. Exemplarily, let the set of geological sub-region units obtained by division be S = {S1, S2,..., Sn}, where n is the number of sub-regions.

[0036] Step S132: Synchronously call reservoir physical property correlation analysis within each geological sub-region unit, and combine the porosity parameter, permeability parameter, and gas content parameter in the initial reservoir feature set to perform local feature extraction operations to generate a geological sub-region feature set. The local feature extraction operations include formation continuity analysis, fault density calculation, and fracture network topology modeling.

[0037] In this embodiment, reservoir physical property correlation analysis is carried out within each geological sub-region unit Si (i = 1, 2,..., n), and local feature extraction operations are performed. First, in combination with the porosity parameter φ, permeability parameter k, and gas content parameter C in the initial reservoir feature set R', their correlation relationships with geological features are analyzed. For example, the correlation between porosity and formation dip, and the relationship between permeability and fault distribution are studied. Formation continuity analysis is to analyze the formation information within the geological sub-region to judge the formation continuity. The formation continuity can be evaluated based on factors such as the change trend of formation dip and the change in formation thickness. Fault density calculation is to calculate the ratio of the number of faults in the geological sub-region to the regional area, reflecting the density of faults in this area. Fracture network topology modeling is to model the fractures within the geological sub-region and analyze the connection relationships and distribution laws among the fractures. Through these local feature extraction operations, a set of feature vectors is generated for each geological sub-region unit, and these feature vectors are combined to form a geological sub-region feature set S' = {S1', S2',..., Sn'}, where Si' represents the feature vector of the i-th geological sub-region unit.

[0038] Step S133: Perform cross-region feature aggregation processing on the geological sub-region feature set to generate a global geological feature set integrating reservoir physical property parameters.

[0039] In this embodiment, cross-region feature aggregation processing is performed on the geological sub-region feature set S' to generate a global geological feature set integrating reservoir physical property parameters. The aggregation can be carried out by using the weighted average method, and the weights are determined according to factors such as the area and geological importance of each geological sub-region unit.

[0040] For example, for a sub-region unit with a larger area and a greater impact of geological features on production capacity, a higher weight is assigned. Let the feature vector of the i-th geological sub-region unit be Si', and its weight be wi (i = 1, 2,..., n), and ∑wi = 1. Then the global geological feature set G'' can be calculated by the following formula: G'' = ∑(wi * Si'). Through this cross-region feature aggregation processing, the feature information of each geological sub-region is integrated to obtain a global geological feature set that can reflect the geological features of the entire target block and simultaneously incorporates the influence of reservoir physical property parameters.

[0041] Step S134: Perform time series decomposition processing on the initial dynamic feature set to generate a trend dynamic feature set and a periodic dynamic feature set, and associate the production cycle stage label as an auxiliary feature to the time series decomposition result.

[0042] In this embodiment, the initial dynamic feature set D' is subjected to time series decomposition to separate the trend and periodic features therein. For the gas production rate sequence Q(t) and the pressure change sequence P(t), time series analysis methods such as the Seasonal Decomposition of Time Series can be used. This method decomposes the time series into a trend component, a seasonal component, and a residual component. The trend component reflects the long-term change trend of the data over time, and the periodic component reflects the repetitive change pattern of the data within a certain time period. Through time series decomposition, the trend dynamic feature set Qt and the periodic dynamic feature set Qp (for the gas production rate sequence) and Pt and Pp (for the pressure change sequence) are obtained. At the same time, the production cycle stage label L is associated with the time series decomposition result as an auxiliary feature. For example, the production cycle stage label at each time point is combined with the corresponding trend and periodic features to form the trend dynamic feature set Qt', the periodic dynamic feature set Qp', and Pt' and Pp' with stage label information.

[0043] Step S135: Generate a plurality of feature units according to the global geological feature set, the trend dynamic feature set, and the periodic dynamic feature set, wherein the geological attribute features in each feature unit are synchronously associated with the initial reservoir feature set.

[0044] In this embodiment, a plurality of feature units are generated according to the global geological feature set G'', the trend dynamic feature sets (Qt', Pt'), and the periodic dynamic feature sets (Qp', Pp'). Each feature unit contains a dual encoding of geological attribute features and dynamic attribute features. When generating the feature units, the geological attribute features in the global geological feature set are associated with the porosity parameter φ, the permeability parameter k, and the gas content parameter C in the initial reservoir feature set R' to ensure that the geological attribute features in each feature unit contain reservoir physical property information. At the same time, the trend dynamic features and the periodic dynamic features are added to the feature units as dynamic attribute features. For example, for a certain time point t and spatial coordinates (x, y), the generated feature unit F(t, x, y) contains the geological attribute features (associated with reservoir physical properties) at this location and the corresponding trend and periodic dynamic features. These feature units are combined to form the prediction feature set P = {P(t, x, y)}, where each feature unit has a dual encoding of geological attribute features and dynamic attribute features and can be used for subsequent production capacity prediction.

[0045] Step S140: Iteratively predict the geological attribute features and the dynamic attribute features based on the reinforcement learning policy network to generate the target coalbed methane production capacity prediction result.

[0046] In this embodiment, a reinforcement learning policy network is used to iteratively predict the geological attribute features and dynamic attribute features in the prediction feature set P to generate a target coalbed methane production prediction result. The reinforcement learning policy network can continuously adjust its own policy according to the feedback of the environment, thereby improving the accuracy of prediction.

[0047] Exemplarily, the specific iterative prediction process is as follows: Step S141: Input the prediction feature set into a pre-trained reinforcement learning policy network, which includes a feature selection module and a weight assignment module.

[0048] In this embodiment, the prediction feature set P is input into a pre-trained reinforcement learning policy network. This network consists of a feature selection module and a weight assignment module. The role of the feature selection module is to screen out the most important features for production prediction from the prediction feature set, reducing the interference of unnecessary features on the prediction result. The weight assignment module assigns corresponding weights to each feature according to the importance of the feature to highlight the role of important features. The pre-trained reinforcement learning policy network is trained on a large amount of historical data and has learned the relationship between geological attribute features, dynamic attribute features, and coalbed methane production.

[0049] Step S142: Through the feature selection module, perform a relevance score on the geological attribute features and dynamic attribute features to generate a feature importance score set that incorporates the influence of reservoir physical properties.

[0050] In this embodiment, the feature selection module performs a relevance score on the geological attribute features and dynamic attribute features to generate a feature importance score set that incorporates the influence of reservoir physical properties. The specific process is as follows: For example, step S1421: Perform dimension alignment processing on each feature unit in the geological attribute features and the feature unit with the same timestamp identifier and spatial coordinate identifier in the dynamic attribute features, and calculate the mutual information through the decomposed univariate sequences to generate an initial correlation degree set.

[0051] In this embodiment, first, dimension alignment processing is performed on the geological attribute features and dynamic attribute features to ensure that the feature units with the same timestamp identifier t and spatial coordinate identifier (x, y) can correspond one by one. Then, each feature unit is decomposed into univariate sequences. For example, multi-dimensional geological attribute feature vectors and dynamic attribute feature vectors are decomposed into multiple one-dimensional univariate sequences. Next, the mutual information between these univariate sequences is calculated. Mutual information is an index to measure the correlation between two variables, which reflects the degree to which one variable contains the information of another variable. By calculating the mutual information between the univariate sequences of geological attribute features and dynamic attribute features, an initial correlation degree set I is obtained.

[0052] Step S1422: Based on the dual coding relationship between the geological attribute features and the dynamic attribute features, use the attention mechanism to assign weights to the time dimension and the spatial dimension of each feature unit in the initial association degree set, and generate a spatio-temporal weighted scoring set.

[0053] In this embodiment, the attention mechanism is used to assign weights to the time dimension and the spatial dimension of each feature unit in the initial association degree set I. The attention mechanism can automatically adjust the weights according to the importance of the features, highlighting the key time and spatial information. Under the dual coding relationship between the geological attribute features and the dynamic attribute features, considering that the influence of different time and spatial positions on the coalbed methane production capacity may be different, different weights are assigned to the time dimension and the spatial dimension of each feature unit through the attention mechanism. For example, for the feature units in the key production stage and important geological areas, higher weights are assigned. After the weight assignment, a spatio-temporal weighted scoring set W is generated.

[0054] Step S1423: Extract the reservoir physical property influence factors corresponding to each feature unit in the spatio-temporal weighted scoring set, where the reservoir physical property influence factors are generated from the dynamic reservoir physical property parameters pre-associated in the geological attribute features, and the dynamic reservoir physical property parameters are updated based on the reservoir physical property change data during the historical production cycle.

[0055] In this embodiment, the reservoir physical property influence factors reflect the influence degree of the reservoir physical property parameters on the prediction of the coalbed methane production capacity. In the geological attribute features, the dynamic reservoir physical property parameters have been pre-associated, and these parameters will be updated based on the reservoir physical property change data during the historical production cycle. For example, parameters such as porosity, permeability, and gas content will change with the exploitation of coalbed methane. By analyzing the historical production data, the change laws of these parameters can be obtained, and the dynamic reservoir physical property parameters can be updated. For each feature unit in the spatio-temporal weighted scoring set W, the corresponding influence factor f is extracted from the pre-associated dynamic reservoir physical property parameters. The calculation of the influence factor f can be based on some empirical models or machine learning algorithms, comprehensively considering the influence of the changes in parameters such as porosity, permeability, and gas content on the production capacity. For example, a multiple linear regression model can be established, with the production capacity as the dependent variable and parameters such as porosity, permeability, and gas content as the independent variables, to obtain the regression coefficients of each parameter, and these regression coefficients can be used as part of the influence factor. At the same time, the change trend and interaction relationship of the parameters can also be considered to further adjust the influence factor.

[0056] Step S1424: Perform a multiplication operation on each scoring value in the spatio-temporal weighted scoring set and the corresponding reservoir physical property influence factor to generate an intermediate scoring set corrected by physical properties.

[0057] In this embodiment, each score value in the spatio-temporal weighted score set W is multiplied by the corresponding reservoir physical property influence factor f. Let the i-th score value in the spatio-temporal weighted score set W be Wi, and the corresponding reservoir physical property influence factor be fi. Then, the i-th score value Mi in the intermediate score set M after physical property correction can be calculated by the formula Mi = Wi * fi. Through this multiplication operation, the influence of reservoir physical properties is incorporated into the scores, making the scores more accurately reflect the importance of features for production capacity prediction.

[0058] Step S1425: Perform global normalization processing on the score values of all feature units in the intermediate score set, map them to a preset score dimension interval, and generate the feature importance score set.

[0059] In this embodiment, in order to make the feature importance scores comparable, it is necessary to perform global normalization processing on the intermediate score set M. A preset score dimension interval is set, for example, [0, 1]. First, find the minimum value minM and the maximum value maxM in the intermediate score set M. Then, for each score value Mi in the intermediate score set M, perform normalization processing through the formula Si = (Mi - minM) / (maxM - minM), and map it to the preset score dimension interval [0, 1]. After normalization processing, the feature importance score set S is obtained, where each score value Si represents the importance degree of the corresponding feature unit.

[0060] Step S143: The weight assignment module dynamically assigns weights to each feature unit in the prediction feature set based on the feature importance score set, and generates a weighted feature set.

[0061] In this embodiment, the weight assignment module dynamically assigns weights to each feature unit in the prediction feature set P according to the feature importance score set S. Let the j-th feature unit in the prediction feature set P be Pj, and the score value corresponding to Pj in the feature importance score set S be Sj. Then, the weight wj assigned to Pj can directly take Sj, that is, wj = Sj. Then, multiply each feature unit Pj by the corresponding weight wj to obtain the weighted feature unit Pj' = wj * Pj. Combine all the weighted feature units to generate the weighted feature set P' = {Pj'}.

[0062] Step S144: Call the regression prediction model to perform non-linear mapping processing on the weighted feature set, and generate an initial production capacity prediction result.

[0063] In this embodiment, a regression prediction model is called to perform a non-linear mapping process on the weighted feature set P' to generate an initial production capacity prediction result. The regression prediction model can be a neural network model, such as a multi-layer perceptron (MLP). The multi-layer perceptron consists of an input layer, a hidden layer, and an output layer. The input layer receives the feature vectors in the weighted feature set P'. The hidden layer performs non-linear transformation on the input features, and the output layer outputs the production capacity prediction value. In the training stage, a large amount of historical data is used to train the multi-layer perceptron, and the weights and biases of the model are adjusted so that the model can learn the non-linear relationship between the weighted features and the production capacity. In the prediction stage, the weighted feature set P' is input into the trained multi-layer perceptron, and through the calculation and output of the model, the initial production capacity prediction result Y0 is obtained.

[0064] Step S145: Synchronously adjust the parameter configurations of the feature selection module and the weight allocation module according to the error gradient between the initial production capacity prediction result and the actual gas production sequence, and generate an optimized reinforcement learning policy network.

[0065] In this embodiment, the error gradient between the initial production capacity prediction result Y0 and the actual gas production sequence Q(t) is calculated. The error gradient reflects the degree of difference and the changing trend of the difference between the prediction result and the actual result. The mean square error (MSE) can be used as the error metric index to calculate the average value of the squared errors of the initial production capacity prediction result Y0 and the actual gas production sequence Q(t) at each time point. Then, the parameter configurations of the feature selection module and the weight allocation module are synchronously adjusted according to the error gradient. For example, for the feature selection module, the threshold for screening features can be adjusted so that the selected features can more accurately reflect the importance of the production capacity prediction; for the weight allocation module, the weight allocation strategy can be adjusted so that the weights of the important features are more reasonable. By continuously adjusting the parameter configurations, the error is gradually reduced until the preset convergence condition is met. After adjustment, an optimized reinforcement learning policy network is generated.

[0066] Step S146: Re-weight the prediction feature set through the optimized reinforcement learning policy network to generate the target coalbed methane production capacity prediction result.

[0067] In this embodiment, the prediction feature set P is input into the optimized reinforcement learning policy network again. The optimized feature selection module will re-screen the features that are more important for the production capacity prediction, and the weight allocation module will re-weight the feature units according to the updated feature importance scores. After re-weighting, a new weighted feature set P'' is obtained. Then, the regression prediction model is called to perform a non-linear mapping process on the weighted feature set P'' to generate the target coalbed methane production capacity prediction result Y.

[0068] Step S150: Generate an optimization strategy set based on the error distribution between the target coalbed methane production capacity prediction result and the historical production dynamic data, and feedback the optimization strategy set to the coalbed methane production control system to trigger parameter adjustment operations.

[0069] In this embodiment, in order to improve the accuracy of coalbed methane production capacity prediction and production efficiency, it is necessary to generate an optimization strategy set based on the error distribution between the target coalbed methane production capacity prediction result Y and the historical production dynamic data (mainly the actual gas production sequence Q(t)), and feedback it to the coalbed methane production control system to trigger corresponding parameter adjustment operations.

[0070] Exemplarily, the specific process of generating the optimization strategy set is as follows: Step S151: Evaluate the confidence level of the target coalbed methane production capacity prediction result based on the historical production dynamic data, and generate a prediction confidence score set.

[0071] In this embodiment, evaluate the confidence level of the target coalbed methane production capacity prediction result Y to generate a prediction confidence score set. The specific process is as follows: Step S1511: Extract the prediction value fluctuation range of the target coalbed methane production capacity prediction result within different time windows, and generate a time series fluctuation feature set.

[0072] In this embodiment, divide the target coalbed methane production capacity prediction result Y according to different time windows, for example, one week as a time window. For each time window, calculate the maximum and minimum values of the prediction value, and the difference between the two is the prediction value fluctuation range within this time window. Combine the prediction value fluctuation ranges of all time windows to generate a time series fluctuation feature set Vt.

[0073] Step S1512: Calculate the prediction value difference degree of the target coalbed methane production capacity prediction result within the spatial grid unit, and generate a spatial difference feature set.

[0074] In this embodiment, divide the target block into multiple spatial grid units. For each spatial grid unit, calculate the difference degree between the prediction values at different positions within the unit. The standard deviation can be used to measure the difference degree of the prediction values. Combine the prediction value difference degrees of all spatial grid units to generate a spatial difference feature set Vs.

[0075] Step S1513: Conduct a joint analysis of the time series fluctuation feature set and the spatial difference feature set to determine the prediction stability score.

[0076] In this embodiment, a joint analysis is performed on the temporal fluctuation feature set Vt and the spatial difference feature set Vs. A weighted average method can be used to assign different weights to the two according to the influence degrees of temporal fluctuation and spatial difference on the prediction stability. For example, let the weight of the temporal fluctuation feature be wt and the weight of the spatial difference feature be ws, and wt + ws = 1. For each time point and spatial position, the corresponding temporal fluctuation feature value and spatial difference feature value are weighted and averaged to obtain the prediction stability score Sd.

[0077] Step S1514: Generate a prediction confidence score set according to the matching degree between the prediction stability score and the actual gas production fluctuation range in the historical production dynamic data, where each scoring unit in the prediction confidence score set contains a dual index of a time stamp identifier and a spatial coordinate identifier.

[0078] In this embodiment, the matching degree between the prediction stability score Sd and the actual gas production fluctuation range in the historical production dynamic data is calculated. The matching degree can be measured by calculating the similarity between the two, such as using cosine similarity. For each time point and spatial position, the prediction stability score is compared with the actual gas production fluctuation range to obtain a matching degree value. These matching degree values are used as prediction confidence scores to generate a prediction confidence score set C. Each scoring unit in this set contains a dual index of a time stamp identifier t and a spatial coordinate identifier (x, y), that is, C = {C(t, x, y)}.

[0079] Step S152: Determine a significant error feature subset according to the correlation analysis result between the prediction confidence score set and the error distribution.

[0080] In this embodiment, the correlation between the prediction confidence score set C and the error distribution is analyzed to determine a significant error feature subset. The specific process is as follows: Step S1521: Perform a binary clustering analysis on the error distribution based on a preset error threshold to generate a set of target error regions where the error values are greater than the preset error threshold.

[0081] In this embodiment, a preset error threshold T is set. For each error value in the error distribution, it is judged whether it is greater than the preset error threshold T. If it is greater than T, the region corresponding to this error is marked as a target error region; otherwise, it is marked as a normal region. Through this binary clustering analysis, the error distribution is divided into a target error region and a normal region to generate a set of target error regions E.

[0082] Step S1522: Extract the original spatial coordinate identifiers corresponding to the set of target error regions, and match the standardized spatial coordinate identifiers associated in the prediction confidence score set based on the spatio-temporal alignment mapping relationship.

[0083] In this embodiment, the original spatial coordinate identifiers corresponding to each target error region are extracted from the set E of target error regions. Then, according to the previously performed spatio-temporal alignment mapping relationship, these original spatial coordinate identifiers are converted into the standardized spatial coordinate identifiers associated in the set C of prediction confidence scores. In this way, the target error regions can be corresponding to the scoring units in the set of prediction confidence scores.

[0084] Step S1523: According to the standardized spatial coordinate identifiers with confidence scores lower than the preset threshold, trace back to the feature units after spatial grid alignment in the set of predicted features to generate a candidate feature subset.

[0085] In this embodiment, a preset confidence threshold Tc is set. For each scoring unit in the set C of prediction confidence scores, determine whether its score value is lower than the preset threshold Tc. If it is lower than Tc, extract the corresponding standardized spatial coordinate identifier of this scoring unit. Then, according to these standardized spatial coordinate identifiers, trace back to the feature units after spatial grid alignment in the set P of predicted features, and combine these feature units to generate a candidate feature subset Fc.

[0086] Step S1524: Sort the candidate feature subset according to the importance of the feature contribution degree to generate a feature priority sequence.

[0087] In this embodiment, the feature contribution degree of each feature unit in the candidate feature subset Fc is calculated. The feature contribution degree can be measured by the feature importance score in the feature selection module. Sort the feature units in the candidate feature subset Fc from largest to smallest according to the feature contribution degree to generate a feature priority sequence Fp.

[0088] Step S1525: Generate a significant error feature subset according to the top N feature units in the feature priority sequence, where N is an integer dynamically adjusted according to the severity of the error distribution.

[0089] In this embodiment, the value of N is dynamically adjusted according to the severity of the error distribution. The more severe the error distribution, the larger the value of N, and the more feature units are selected. Select the top N feature units from the feature priority sequence Fp, and combine these feature units to generate a significant error feature subset Fs.

[0090] Step S153: Extract the geological attribute features and dynamic attribute features corresponding to the significant error feature subset to generate a feature optimization priority list.

[0091] In this embodiment, the corresponding geological attribute features and dynamic attribute features are extracted from the significant error feature subset Fs. These features are sorted according to their importance in the significant error feature subset to generate a feature optimization priority list Lp. The features in this list are the features that have a greater impact on the production capacity prediction error and need to be optimized first.

[0092] Step S154: Based on the feature optimization priority list, adjust the data acquisition device deployment plan to generate a first optimization strategy.

[0093] In this embodiment, the data acquisition device deployment plan is adjusted according to the feature optimization priority list Lp to generate a first optimization strategy. The specific process is as follows: Step S1541: Analyze the geological attribute features and dynamic attribute features in the feature optimization priority list to determine the key monitoring parameter set.

[0094] In this embodiment, the geological attribute features and dynamic attribute features in the feature optimization priority list Lp are analyzed. Parameters closely related to data acquisition, such as formation dip angle, porosity, gas production, etc., are found from these features. These parameters are combined to determine the key monitoring parameter set M.

[0095] Step S1542: Adjust the deployment locations of the data acquisition devices according to the spatial distribution density of the key monitoring parameter set to generate a device location optimization plan.

[0096] In this embodiment, the spatial distribution density of each parameter in the key monitoring parameter set M is analyzed. For regions with a relatively high spatial distribution density, increase the number of deployed data acquisition devices to obtain more detailed parameter information; for regions with a relatively low spatial distribution density, the number of data acquisition devices can be appropriately reduced. According to this adjustment principle, determine the new deployment locations of the data acquisition devices to generate a device location optimization plan Dp.

[0097] Step S1543: Adjust the sampling frequency of the data acquisition devices according to the time variation frequency of the key monitoring parameter set to generate a sampling frequency optimization plan.

[0098] In this embodiment, the time variation frequency of each parameter in the key monitoring parameter set M is analyzed. For parameters with a relatively fast time variation frequency, increase the sampling frequency of the data acquisition devices to capture the rapid changes of the parameters; for parameters with a relatively slow time variation frequency, the sampling frequency can be reduced. According to this adjustment principle, determine the new sampling frequency of the data acquisition devices to generate a sampling frequency optimization plan Sp.

[0099] Step S1544: Integrate the device location optimization scheme and the sampling frequency optimization scheme to generate a first optimization strategy, where the first optimization strategy includes a set of configuration instructions for device identification, location coordinates, sampling period, and parameter type.

[0100] In this embodiment, integrate the device location optimization scheme Dp and the sampling frequency optimization scheme Sp. Assign a unique device identification to each data acquisition device, record its new location coordinates, sampling period, and the parameter types to be monitored. Combine this information into a set of configuration instructions to generate the first optimization strategy O1.

[0101] Step S155: Dynamically configure the training frequency of the reinforcement learning policy network according to the spatio-temporal change pattern of the error distribution to generate a second optimization strategy.

[0102] In this embodiment, dynamically configure the training frequency of the reinforcement learning policy network according to the spatio-temporal change pattern of the error distribution to generate a second optimization strategy. The specific process is as follows: Step S1551: Decompose the error distribution in the time dimension to generate long-term error trend features and short-term error fluctuation features.

[0103] In this embodiment, use time series analysis methods to decompose the error distribution in the time dimension. The moving average method can be used to decompose the error distribution into a long-term trend component and a short-term fluctuation component. The long-term trend component reflects the change trend of the error over a long period of time, that is, the long-term error trend feature Et; the short-term fluctuation component reflects the fluctuation of the error over a short period of time, that is, the short-term error fluctuation feature Es.

[0104] Step S1552: Decompose the error distribution in the space dimension to generate regional error aggregation features and discrete error dispersion features.

[0105] In this embodiment, divide the target block into multiple spatial regions and analyze the distribution of errors in each region. For the regions where errors are concentratedly distributed, define them as regional error aggregation features Ea; for the regions where errors are dispersedly distributed, define them as discrete error dispersion features Ed.

[0106] Step S1553: Set a basic training frequency according to the historical change rates of the long-term error trend features and the regional error aggregation features.

[0107] In this embodiment, calculate the historical change rates of the long-term error trend feature Et and the regional error aggregation feature Ea. The historical change rate reflects the change speed of the error over a historical time period. According to these historical change rates, set the basic training frequency Fb of the reinforcement learning policy network. The greater the historical change rate, the higher the basic training frequency, so as to adjust the model parameters faster to adapt to the change of the error.

[0108] Step S1554: Calculate the dynamic adjustment amplitude according to the amplitude threshold of the short-term error fluctuation feature and the spatial density threshold of the discrete error dispersion feature.

[0109] In this embodiment, an amplitude threshold Ts of the short-term error fluctuation feature Es and a spatial density threshold Td of the discrete error dispersion feature Ed are set. When the amplitude of the short-term error fluctuation feature exceeds the amplitude threshold Ts, or the spatial density of the discrete error dispersion feature exceeds the spatial density threshold Td, it is necessary to dynamically adjust the training frequency. Calculate the dynamic adjustment amplitude ΔF, and the dynamic adjustment amplitude can be calculated linearly or non-linearly according to the degree of exceeding the threshold.

[0110] Step S1555: Generate a training frequency configuration curve based on the linear superposition relationship between the basic training frequency and the dynamic adjustment amplitude.

[0111] In this embodiment, linearly superpose the basic training frequency Fb and the dynamic adjustment amplitude ΔF to obtain the training frequency F = Fb + ΔF. According to the error distribution at different time points and spatial positions, calculate the corresponding training frequencies, and connect these training frequencies to generate the training frequency configuration curve Fc.

[0112] Step S1556: Convert the training frequency configuration curve into a set of parameter update instructions for the reinforcement learning policy network to generate a second optimized policy.

[0113] In this embodiment, convert the training frequency configuration curve Fc into a set of parameter update instructions for the reinforcement learning policy network. Each parameter update instruction includes a timestamp, a training frequency, and parameter information to be updated. Combine these parameter update instructions to generate the second optimized policy O2.

[0114] Step S156: Combine the first optimized policy and the second optimized policy to generate an optimized policy set.

[0115] In this embodiment, combine the first optimized policy O1 and the second optimized policy O2 to generate an optimized policy set. The specific process is as follows: For example, step S1561: Obtain the device position optimization scheme and the sampling frequency optimization scheme in the first optimized policy, and extract the set of spatial coordinate identifiers in the device position optimization scheme and the set of time window identifiers in the sampling frequency optimization scheme.

[0116] In this embodiment, extract the device position optimization scheme Dp and the sampling frequency optimization scheme Sp from the first optimized policy O1. Extract the set of spatial coordinate identifiers X of all data acquisition devices from the device position optimization scheme Dp; extract the set of sampling time window identifiers T of each device from the sampling frequency optimization scheme Sp.

[0117] Step S1562: Obtain the training frequency configuration curve in the second optimization strategy, and extract the time series parameters and the corresponding dynamic adjustment amplitude set in the training frequency configuration curve.

[0118] In this embodiment, obtain the training frequency configuration curve Fc from the second optimization strategy O2. Extract the time series parameter t and the corresponding dynamic adjustment amplitude set ΔF from the training frequency configuration curve Fc.

[0119] Step S1563: Perform a time dimension matching analysis on the time window identification set and the time series parameters to generate an association mapping table of the device sampling period and the model training period after time alignment.

[0120] In this embodiment, perform a matching analysis on the time window identification set T and the time series parameter t in the time dimension. First, clarify the time ranges and resolutions of the two sets and consider them on the same time scale. For each time window in the time window identification set T, find the corresponding time point or time period in the time series parameter t. If there is an overlapping part between the time window and the time series parameter, it is considered that they are matched. Combine the information of the matched device sampling period and model training period to generate an association mapping table M. This association mapping table records the corresponding relationship between device sampling and model training at each time point or time period. For example, if there is a device sampling operation within a certain time window and it corresponds to a certain training step in the model training period, record this corresponding relationship in the association mapping table.

[0121] Step S1564: Calculate the influence factor between the sampling interval in the sampling frequency optimization scheme and the frequency change rate in the dynamic adjustment amplitude set according to the period matching degree in the association mapping table.

[0122] In this embodiment, calculate the influence factor between the sampling interval in the sampling frequency optimization scheme Sp and the frequency change rate in the dynamic adjustment amplitude set ΔF according to the period matching degree in the association mapping table M. First, determine the measurement method of the period matching degree. For example, it is represented by the proportion of the overlapping time between the sampling period and the training period in the total time. For each group of matching relationships in the association mapping table M, analyze the relationship between the sampling interval and the frequency change rate. If there is a certain functional relationship between the sampling interval and the frequency change rate, determine this functional relationship through statistical analysis of historical data or machine learning algorithms. The influence factor can be obtained by quantifying this functional relationship. For example, if it is found that the smaller the sampling interval, the larger the frequency change rate, an influence factor can be calculated according to this relationship to describe the interaction strength between them.

[0123] Step S1565: Based on the influence factor, perform spatial density correction on the set of spatial coordinate identifiers in the device location optimization plan to generate an optimized parameter configuration table in which the spatial coverage density is synchronously adapted to the time sampling frequency.

[0124] In this embodiment, the influence factor calculated previously is used to perform spatial density correction on the set of spatial coordinate identifiers X in the device location optimization plan Dp. For example, first, clarify the principle of spatial density correction, that is, adjust the distribution density of devices at different spatial positions according to the influence factor. If the influence factor indicates that the interaction between the sampling interval and the frequency change rate makes some areas require denser sampling, increase the number of data acquisition devices in these areas; otherwise, reduce the number of devices. According to the adjusted device distribution, combined with the sampling frequency optimization plan Sp, generate an optimized parameter configuration table P in which the spatial coverage density is synchronously adapted to the time sampling frequency. This optimized parameter configuration table records the spatial coordinates, sampling frequencies of each device, and the associated information with model training, ensuring that spatial coverage and time sampling can be optimized synergistically.

[0125] Step S1566: According to the parameter matching relationship in the optimized parameter configuration table, jointly encode the device location optimization plan, the sampling frequency optimization plan, and the training frequency configuration curve to generate an optimized strategy set containing spatio-temporal unified dimension parameters, where the adjustment amount of each parameter unit in the optimized strategy set is updated synchronously based on the normalized time step and the spatial grid ratio.

[0126] In this embodiment, according to the parameter matching relationship in the optimized parameter configuration table P, jointly encode the device location optimization plan Dp, the sampling frequency optimization plan Sp, and the training frequency configuration curve Fc. The process of joint encoding is to integrate the information in these three plans into a unified representation, ensuring that each parameter has a clear spatio-temporal identifier. First, normalize the time step and the spatial grid so that time and space have a unified dimension. For example, convert the time step to a standard time unit and divide the spatial grid according to a unified scale. Then, according to the matching relationship in the optimized parameter configuration table P, determine the adjustment amount for each parameter unit. The determination of the adjustment amount should be based on the normalized time step and the spatial grid ratio to ensure that the adjustment of parameters is reasonable and consistent under different spatio-temporal conditions. Combine the information of all parameter units to generate an optimized strategy set O containing spatio-temporal unified dimension parameters. Each parameter unit in this optimized strategy set has a clear spatio-temporal identifier and adjustment amount, which can accurately guide the parameter adjustment operation of the coalbed methane production control system.

[0127] Feed the generated set of optimization strategies \(O\) back to the coalbed methane production control system. After receiving the set of optimization strategies, the coalbed methane production control system will adjust the positions of the data acquisition devices, the sampling frequencies, and the training frequencies of the reinforcement learning policy network according to the configuration instructions therein. For example, adjust the installation positions of the data acquisition devices according to the device position optimization plan, set the sampling periods of the devices according to the sampling frequency optimization plan, and update the training parameters of the reinforcement learning policy network according to the training frequency configuration curve. Through these adjustments, the accuracy of coalbed methane production capacity prediction can be improved, the production process of coalbed methane can be optimized, and the production efficiency and economic benefits can be increased.

[0128] Throughout the process, various technical means are adopted for data privacy protection and anti-disclosure. In the data acquisition stage, the acquisition devices are encrypted to ensure the security of data during transmission. For example, use symmetric encryption algorithms to encrypt the acquired data, and only authorized devices and systems can decrypt and process this data. In terms of data storage, a secure database system is adopted to perform access control and permission management on the data. Only authorized personnel can access and operate the data in the database. At the same time, the data is regularly backed up to prevent data loss. In the data processing and analysis stage, the data is anonymized to remove sensitive information in the data, such as specific geographical locations, personnel identities, etc. For the training and use of artificial intelligence models, technologies such as federated learning are adopted to avoid the centralized storage and processing of data and reduce the risk of data leakage. Through these privacy protection and anti-disclosure technical means, the security and privacy of data in the coalbed methane production capacity prediction process are ensured.

[0129] Through the above embodiments, by obtaining a multi-source coalbed methane production capacity data set, performing spatio-temporal alignment processing and feature extraction, using a reinforcement learning policy network for production capacity prediction, generating a set of optimization strategies based on the error distribution between the prediction results and historical data, feeding it back to the production control system for parameter adjustment, and at the same time adopting effective privacy protection and anti-disclosure technical means, the accuracy of coalbed methane production capacity prediction and production efficiency can be improved.

[0130] Figure 2 FIG. shows a schematic diagram of exemplary hardware and software components of a coalbed methane production capacity prediction system 100 incorporating artificial intelligence that can implement the ideas of the present application. For example, the processor 120 can be used on the coalbed methane production capacity prediction system 100 incorporating artificial intelligence and is used to execute the functions in the present application.

[0131] The coalbed methane production capacity prediction system 100 integrated with artificial intelligence can be a general-purpose server or a special-purpose server, both of which can be used to implement the coalbed methane production capacity prediction method integrated with artificial intelligence of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0132] For example, the coalbed methane production capacity prediction system 100 integrated with artificial intelligence may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as disks, ROM, or RAM, or any combination thereof. Exemplarily, the coalbed methane production capacity prediction system 100 integrated with artificial intelligence may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of this application can be implemented according to these program instructions. The coalbed methane production capacity prediction system 100 integrated with artificial intelligence further includes an I / O interface 150 between the computer and other input / output devices.

[0133] For ease of explanation, only one processor is described in the coalbed methane production capacity prediction system 100 integrated with artificial intelligence. However, it should be noted that the coalbed methane production capacity prediction system 100 in this application may also include multiple processors. Therefore, the steps executed by one processor described in this application can also be jointly executed or separately executed by multiple processors. For example, if the processor of the coalbed methane production capacity prediction system 100 executes step A and step B, it should be understood that step A and step B can also be jointly executed by two different processors or separately executed in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0134] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the coalbed methane production capacity prediction method integrated with artificial intelligence as described above is implemented.

[0135] It should be noted that, in order to simplify the expression of the disclosure of the present invention and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. A method for predicting coalbed methane production capacity combined with artificial intelligence, characterized in that, The method includes: Obtaining a multi-source coalbed methane production capacity data set of a target block, where the multi-source coalbed methane production capacity data set includes geological structure data, reservoir physical property data, and historical production dynamic data; Performing spatio-temporal alignment processing on the multi-source coalbed methane production capacity data set to generate a standardized coalbed methane production capacity feature set; Performing feature extraction on the standardized coalbed methane production capacity feature set to generate a prediction feature set, where each feature unit in the prediction feature set includes dual coding of geological attribute features and dynamic attribute features; Performing iterative prediction on the geological attribute features and dynamic attribute features based on a reinforcement learning policy network to generate a target coalbed methane production capacity prediction result; Generating an optimization policy set according to the error distribution between the target coalbed methane production capacity prediction result and the historical production dynamic data, and feeding back the optimization policy set to the coalbed methane production control system to trigger parameter adjustment operations.

2. The coalbed methane production capacity prediction method combined with artificial intelligence according to claim 1, wherein The performing spatio-temporal alignment processing on the multi-source coalbed methane production capacity data set to generate a standardized coalbed methane production capacity feature set includes: Extracting formation dip parameters, fault distribution parameters, and fracture development parameters in the geological structure data to construct an initial geological feature set; Extracting porosity parameters, permeability parameters, and gas content parameters in the reservoir physical property data to construct an initial reservoir feature set; Extracting gas production sequences, pressure change sequences, and production cycle stage labels in the historical production dynamic data to construct an initial dynamic feature set; Performing data preprocessing on the initial geological feature set, initial reservoir feature set, and initial dynamic feature set, where the data preprocessing includes missing value filling, outlier correction, and dimensionality normalization; Performing time window alignment and space grid alignment on the preprocessed initial geological feature set, initial reservoir feature set, and initial dynamic feature set to generate a standardized coalbed methane production capacity feature set, where each feature unit in the standardized coalbed methane production capacity feature set includes a unified time stamp identifier and a space coordinate identifier.

3. The coalbed methane production capacity prediction method combined with artificial intelligence according to claim 2, wherein The performing feature extraction on the standardized coalbed methane production capacity feature set to generate a prediction feature set including geological attribute features and dynamic attribute features includes: Performing regional segmentation processing on the initial geological feature set to generate multiple geological sub-region units; Synchronously invoking reservoir physical property correlation analysis within each geological sub-region unit, and combining the porosity parameters, permeability parameters, and gas content parameters in the initial reservoir feature set to perform local feature extraction operations to generate a geological sub-region feature set, where the local feature extraction operations include formation continuity analysis, fault density calculation, and fracture network topology modeling; Performing cross-region feature aggregation processing on the geological sub-region feature set to generate a global geological feature set integrating reservoir physical property parameters; Performing time series decomposition processing on the initial dynamic feature set to generate a trend dynamic feature set and a periodic dynamic feature set, and associating the production cycle stage label as an auxiliary feature to the time series decomposition result; Generate multiple feature units according to the global geological feature set, the trend dynamic feature set, and the periodic dynamic feature set, where the geological attribute features in each feature unit are synchronously associated with the initial reservoir feature set.

4. The coalbed methane production capacity prediction method combined with artificial intelligence according to claim 1, wherein The iterative prediction of the geological attribute features and the dynamic attribute features based on the reinforcement learning policy network to generate the target coalbed methane production capacity prediction result includes: Input the prediction feature set into a pre-trained reinforcement learning policy network, which includes a feature selection module and a weight allocation module; Generate a feature importance score set that incorporates the influence of reservoir physical properties by using the feature selection module to perform a correlation score on the geological attribute features and the dynamic attribute features; Dynamically assign weights to each feature unit in the prediction feature set based on the feature importance score set through the weight allocation module to generate a weighted feature set; Call a regression prediction model to perform a non-linear mapping process on the weighted feature set to generate an initial production capacity prediction result; Synchronously adjust the parameter configurations of the feature selection module and the weight allocation module according to the error gradient between the initial production capacity prediction result and the actual gas production sequence to generate an optimized reinforcement learning policy network; Re-weight the prediction feature set through the optimized reinforcement learning policy network to generate the target coalbed methane production capacity prediction result.

5. The coalbed methane production capacity prediction method combined with artificial intelligence according to claim 1, characterized in that The generation of the optimization strategy set according to the error distribution between the target coalbed methane production capacity prediction result and the historical production dynamic data includes: Perform a confidence evaluation on the target coalbed methane production capacity prediction result according to the historical production dynamic data to generate a prediction confidence score set; Determine a significant error feature subset according to the correlation analysis result between the prediction confidence score set and the error distribution; Extract the geological attribute features and the dynamic attribute features corresponding to the significant error feature subset to generate a feature optimization priority list; Adjust the data acquisition device deployment plan based on the feature optimization priority list to generate a first optimization strategy; Dynamically configure the training frequency of the reinforcement learning policy network according to the spatio-temporal change pattern of the error distribution to generate a second optimization strategy; Merge the first optimization strategy and the second optimization strategy to generate an optimization strategy set.

6. The coalbed methane production capacity prediction method combined with artificial intelligence according to claim 5, characterized in that The performance of the confidence evaluation on the target coalbed methane production capacity prediction result according to the historical production dynamic data to generate a prediction confidence score set includes: Extract the prediction value fluctuation range of the target coalbed methane production capacity prediction result within different time windows to generate a time series fluctuation feature set; Calculate the prediction value difference degree of the target coalbed methane production capacity prediction result within the spatial grid unit to generate a spatial difference feature set; Perform a joint analysis on the time series fluctuation feature set and the spatial difference feature set to determine the prediction stability score; Generate a prediction confidence score set according to the matching degree between the prediction stability score and the actual gas production fluctuation range in the historical production dynamic data, where each scoring unit in the prediction confidence score set contains a dual index of a timestamp identifier and a spatial coordinate identifier.

7. The coalbed methane production capacity prediction method combined with artificial intelligence according to claim 5, wherein Determining a significant error feature subset according to the correlation analysis result between the predicted confidence score set and the error distribution includes: Performing binary clustering analysis on the error distribution based on a preset error threshold to generate a set of target error regions with error values greater than the preset error threshold; Extracting the original spatial coordinate identifiers corresponding to the set of target error regions, and matching the associated standardized spatial coordinate identifiers in the predicted confidence score set based on the spatio-temporal alignment mapping relationship; According to the standardized spatial coordinate identifiers with confidence scores lower than the preset threshold, tracing back to the feature units after spatial grid alignment in the predicted feature set to generate a candidate feature subset; Performing importance ranking on the candidate feature subset based on the feature contribution degree to generate a feature priority sequence; Generating a significant error feature subset according to the top N feature units in the feature priority sequence, where N is an integer dynamically adjusted according to the severity of the error distribution.

8. The coalbed methane production capacity prediction method combined with artificial intelligence according to claim 5, wherein Adjusting the data acquisition device deployment plan based on the feature optimization priority list to generate a first optimization strategy, including: Analyzing the geological attribute features and dynamic attribute features in the feature optimization priority list to determine a set of key monitoring parameters; Adjusting the deployment locations of the data acquisition devices according to the spatial distribution density of the set of key monitoring parameters to generate a device location optimization plan; Adjusting the sampling frequencies of the data acquisition devices according to the time change frequencies of the set of key monitoring parameters to generate a sampling frequency optimization plan; Integrating the device location optimization plan and the sampling frequency optimization plan to generate a first optimization strategy, where the first optimization strategy includes a set of configuration instructions for device identifiers, location coordinates, sampling periods, and parameter types.

9. The coalbed methane production capacity prediction method combined with artificial intelligence according to claim 5, characterized in that Dynamically configuring the training frequency of the reinforcement learning policy network according to the spatio-temporal change pattern of the error distribution to generate a second optimization strategy, including: Decomposing the error distribution in the time dimension to generate long-term error trend features and short-term error fluctuation features; Decomposing the error distribution in the spatial dimension to generate regional error aggregation features and discrete error dispersion features; Setting a basic training frequency according to the historical change rates of the long-term error trend features and the regional error aggregation features; Calculating the dynamic adjustment amplitude according to the amplitude threshold of the short-term error fluctuation features and the spatial density threshold of the discrete error dispersion features; Generating a training frequency configuration curve based on the linear superposition relationship between the basic training frequency and the dynamic adjustment amplitude; Converting the training frequency configuration curve into a set of parameter update instructions for the reinforcement learning policy network to generate a second optimization strategy.

10. A coalbed methane production capacity prediction system combined with artificial intelligence, characterized in that, Including a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the coalbed methane production capacity prediction method combined with artificial intelligence according to any one of claims 1-9 above.

Images