Coal-bed gas well production dynamic analysis method and system based on AI time sequence prediction

Abstract

The invention provides a coal-bed gas well production dynamic analysis method and system based on AI time sequence prediction, and the method comprises the steps: firstly collecting a production dynamic data sequence of a coal-bed gas well, a corresponding well position geological condition data sequence and a surrounding environment data sequence, and constructing a ternary time sequence dynamic association network comprising node dynamic time sequence segments and a connection line real-time association relationship; and then calling a pre-trained AI time sequence coupling deviation evolution mining model to process the ternary time sequence dynamic association network to obtain a ternary time sequence coupling deviation result. And a deviation conduction closed-loop path is determined according to a ternary time sequence coupling deviation result in combination with forward conduction positioning and reverse tracing verification, and a deviation conduction analysis report is generated. And finally, generating a dynamic regulation strategy based on the deviation conduction analysis report in combination with a production dynamic prediction threshold value, outputting the dynamic regulation strategy to a production control terminal, and meanwhile, feeding back regulation data to update a network node association relationship, thereby realizing accurate analysis and regulation of the production dynamic state of the coal-bed gas well.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a method and system for dynamic analysis of coalbed methane well production based on AI time-series prediction. Background Technology

[0002] In the production process of coalbed methane wells, accurate analysis of production dynamics and the formulation of reasonable control strategies are crucial for improving production efficiency and ensuring production safety. Traditional methods for analyzing the production dynamics of coalbed methane wells are mainly based on single data types. For example, they rely solely on production dynamic data, such as changes in gas production rate and bottom hole pressure, to assess the production status. However, these methods neglect the impact of geological conditions and environmental factors on production. Regarding geological conditions, changes in coal seam permeability and porosity directly affect coalbed methane migration and production, but traditional methods fail to effectively correlate these with production dynamic data. Surrounding environmental factors, such as surface temperature and humidity, also indirectly affect coalbed methane well production, and are similarly not adequately considered in traditional analyses. Furthermore, existing technologies are relatively superficial in analyzing the relationships between different data sequences, lacking in-depth exploration of dynamic correlations and the evolution of coupling deviations between data. This makes it difficult to accurately grasp the key factors and transmission paths affecting production dynamics, resulting in a lack of scientific basis for formulating production control strategies and hindering efficient and precise management of coalbed methane well production. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for dynamic analysis of coalbed methane well production based on AI time-series prediction, the method comprising: The production dynamic data sequence of the coalbed methane well, the geological condition data sequence of the corresponding well location, and the surrounding environmental data sequence are collected. The production dynamic data sequence includes information on changes in gas production rate and bottom hole pressure. The geological condition data sequence includes information on changes in coal seam permeability and coal seam porosity. The environmental data sequence includes information on changes in surface temperature and surface humidity. Based on the production dynamic data sequence, the geological condition data sequence, and the environmental data sequence, a ternary time-series dynamic correlation network for coalbed methane wells is constructed. Each node in the ternary time-series dynamic correlation network corresponds to a dynamic time-series segment of a single type of data sequence. The connections between nodes correspond to the real-time correlation relationship of dynamic time-series segments of different types of data sequences, and the connection weights evolve and update over time. The pre-trained AI temporal coupling deviation evolution mining model is invoked to process the ternary temporal dynamic correlation network, mining the coupling deviation evolution information between dynamic temporal segments of different data sequences, and obtaining the ternary temporal coupling deviation result containing the deviation evolution trend. Based on the ternary time-series coupling deviation results, combined with forward transmission location and reverse tracing verification, the deviation transmission closed-loop path affecting production dynamics is determined, and a deviation transmission analysis report containing the deviation transmission direction, correlation strength and tracing verification results is generated. Based on the deviation transmission analysis report and combined with the production dynamic prediction threshold, a dynamic control strategy for coalbed methane well production parameters is generated and output to the production control terminal. At the same time, the controlled data is fed back to the ternary time-series dynamic correlation network to update the node correlation relationship.

[0004] In another aspect, embodiments of the present invention also provide a dynamic analysis system for coalbed methane well production based on AI time-series prediction, 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 code. The processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0005] Based on the above, a ternary time-series dynamic correlation network for coalbed methane wells was constructed by comprehensively collecting production dynamic data sequences, geological condition data sequences of corresponding well locations, and surrounding environmental data sequences. This network achieves organic integration and dynamic correlation analysis of multi-source heterogeneous data. A pre-trained AI time-series coupling deviation evolution mining model is used to process the network, enabling in-depth mining of coupling deviation evolution information between dynamic time-series segments of different data sequences. Based on the obtained deviation evolution results, combined with forward transmission location and reverse tracing verification, the deviation transmission closed-loop path affecting production dynamics can be accurately determined, generating a comprehensive and accurate deviation transmission analysis report. Based on this deviation transmission analysis report and production dynamic prediction thresholds, a dynamic control strategy is generated and fed back to update the control data network, effectively improving the accuracy of coalbed methane well production dynamic analysis. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the execution flow of the coalbed methane well production dynamic analysis method based on AI time-series prediction provided in an embodiment of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of the AI-based time-series prediction-based coalbed methane well production dynamic analysis system provided in an embodiment of the present invention. Detailed Implementation

[0008] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for dynamic analysis of coalbed methane well production based on AI time-series prediction, provided in one embodiment of the present invention. The following is a detailed description of this method for dynamic analysis of coalbed methane well production based on AI time-series prediction.

[0009] Step S110: Collect the production dynamic data sequence of the coalbed methane well, the geological condition data sequence of the corresponding well location, and the surrounding environmental data sequence. The production dynamic data sequence includes information on changes in gas production rate and bottom hole pressure. The geological condition data sequence includes information on changes in coal seam permeability and porosity. The environmental data sequence includes information on changes in surface temperature and surface humidity.

[0010] In this embodiment, a production well in a coalbed methane field is used as the application object. Data acquisition is completed through various sensors and monitoring equipment deployed in the production well. The acquisition of the production dynamic data sequence is achieved through a gas production rate sensor and a bottom hole pressure sensor installed at the wellhead. The gas production rate sensor records the change in gas production volume per unit time in real time, and the bottom hole pressure sensor transmits the pressure change data at the bottom hole through a downhole cable. The two sensors generate gas production rate change information and bottom hole pressure change information containing multiple time data points according to a preset acquisition interval, which together form the production dynamic data sequence.

[0011] The acquisition of geological condition data sequences relies on downhole geological monitoring devices. Coal seam permeability change information is obtained through a permeability tester, which calculates permeability changes by injecting test fluid into the coal seam and monitoring the pressure response. Coal seam porosity change information is acquired through a porosity logging tool, which uses acoustic or neutron logging principles to obtain porosity data. Both types of data are recorded at fixed intervals to form a geological condition data sequence.

[0012] The surrounding environmental data sequence was collected through environmental monitoring stations installed around the well site. Surface temperature changes were recorded in real time by temperature sensors, and surface humidity changes were acquired by humidity sensors. These two types of environmental parameters were used to generate time-series data at the same collection intervals, forming the environmental data sequence. During the collection process, all data were timestamped and transmitted to the data processing center using an encrypted transmission protocol to prevent data leakage and ensure data security.

[0013] Step S120: Based on the production dynamic data sequence, the geological condition data sequence, and the environmental data sequence, construct a ternary time-series dynamic correlation network for coalbed methane wells. Each node in the ternary time-series dynamic correlation network corresponds to a dynamic time-series segment of a single type of data sequence. The connections between nodes correspond to the real-time correlation relationship of dynamic time-series segments of different types of data sequences, and the connection weights evolve and update over time.

[0014] In this embodiment, constructing a ternary temporal dynamic association network requires data preprocessing, temporal segment division, feature extraction, association strength calculation, network construction, and weight update, which are implemented through the following sub-steps.

[0015] Step S121: Perform time sequence continuity detection on the production dynamic data sequence, the geological condition data sequence, and the environmental data sequence, and retain continuous data segments as valid data sequences.

[0016] A temporal continuity detection algorithm is used to detect the three types of data sequences collected. It iterates through the timestamps in each data sequence and determines whether the interval between two adjacent timestamps conforms to the preset collection interval. If the intervals of multiple consecutive timestamps in a data segment are all within the allowable error range and there is no missing data, then the data segment is determined to be a continuous data segment. If there are abnormal timestamp intervals or missing data, the data segment is truncated, and only continuous data segments are retained. For example, in a production dynamics data sequence, if the collection intervals of multiple consecutive times in a data segment meet the requirements and there are no missing values, this data segment is considered a valid production dynamics data sequence. In a geological conditions data sequence, if there are missing data segments due to equipment failure, these missing segments are removed, and the remaining continuous data segments are retained as valid geological conditions data sequences.

[0017] Step S122: Based on the frequency of change and fluctuation of the effective data sequence, determine the dynamic time interval division rule, and divide the effective production dynamic data sequence into multiple production dynamic time series segments, the effective geological condition data sequence into multiple geological condition dynamic time series segments, and the effective environmental data sequence into multiple environmental dynamic time series segments, thereby obtaining the production dynamic time series segment set, the geological condition dynamic time series segment set, and the environmental dynamic time series segment set. The higher the frequency of change and fluctuation, the smaller the time interval.

[0018] Calculate the fluctuation frequency of each valid data sequence by counting the number of changes in data values ​​per unit time. Based on the fluctuation frequency, set dynamic time interval division rules: for data sequences with high fluctuation frequencies, such as production dynamic data sequences with frequent changes in gas production rates, use smaller time intervals; for data sequences with low fluctuation frequencies, such as geological condition data sequences (coal seam permeability and porosity change slowly), use larger time intervals.

[0019] According to the partitioning rules, the effective production dynamic data sequence is divided into multiple production dynamic time series segments based on shorter time intervals. Each segment contains information on changes in gas production rate and bottom hole pressure within that time period. Similarly, the effective geological condition data sequence is divided into multiple geological condition dynamic time series segments based on longer time intervals. Each geological condition dynamic time series segment contains information on changes in coal seam permeability and porosity within the corresponding time period. The effective environmental data sequence is divided into multiple environmental dynamic time series segments based on time intervals determined by its fluctuation frequency. Each environmental dynamic time series segment contains information on changes in surface temperature and surface humidity. After partitioning, all production dynamic time series segments are categorized into a production dynamic time series segment set, and similarly, a geological condition dynamic time series segment set and an environmental dynamic time series segment set are formed.

[0020] Step S123: Extract the gas production rate change trend characteristics and bottom hole pressure change trend characteristics of each production dynamic time segment; extract the coal seam permeability change trend characteristics and coal seam porosity change trend characteristics of each geological condition dynamic time segment; and extract the surface temperature change trend characteristics and surface humidity change trend characteristics of each environmental dynamic time segment.

[0021] For each dynamic time-series segment of production, trend analysis algorithms are used to extract the gas production rate change trend characteristics and bottom hole pressure change trend characteristics. Specifically, the gas production rate data points within the segment are fitted to determine their overall upward, downward, or stable trend, while the inflection point position and magnitude of the trend change are recorded to form the gas production rate change trend characteristics. The same method is used to process the bottom hole pressure data to obtain the bottom hole pressure change trend characteristics.

[0022] For each dynamic time series segment of geological conditions, the coal seam permeability data sequence is analyzed to determine its long-term variation trend and short-term fluctuation characteristics, forming the coal seam permeability variation trend characteristics; the same analysis is performed on the coal seam porosity data to extract the coal seam porosity variation trend characteristics.

[0023] For each environmental dynamic time series segment, the surface temperature data sequence is processed to extract its changing trend over time, including diurnal fluctuation patterns and long-term changing trends, forming surface temperature change trend characteristics; the changing patterns of surface humidity data are analyzed to obtain surface humidity change trend characteristics.

[0024] Step S124: Calculate the real-time correlation strength between the gas production rate change trend feature and the coal seam permeability change trend feature; calculate the real-time correlation strength between the gas production rate change trend feature and the surface temperature change trend feature; calculate the real-time correlation strength between the bottom hole pressure change trend feature and the coal seam porosity change trend feature; calculate the real-time correlation strength between the bottom hole pressure change trend feature and the surface humidity change trend feature; calculate the real-time correlation strength between the coal seam permeability change trend feature and the surface temperature change trend feature; calculate the real-time correlation strength between the coal seam porosity change trend feature and the surface humidity change trend feature.

[0025] The calculation of the real-time correlation strength between various trend features follows a similar analytical logic. Taking the calculation of the real-time correlation strength between the gas production rate change trend feature and the coal seam permeability change trend feature as an example, it is specifically implemented through the following sub-steps.

[0026] Step S1241: Obtain the first dynamic time-series vector corresponding to the gas production rate change trend feature, where each element of the first dynamic time-series vector corresponds to the gas production rate value at different times within the dynamic time-series segment of production.

[0027] Extract the specific gas production rate value for each moment within the dynamic time series segment of production dynamics, arrange them in chronological order to form the first dynamic time series vector, the length of the vector is equal to the number of moments within the segment, and each element corresponds to the gas production rate value at a moment.

[0028] Step S1242: Obtain the second dynamic time series vector corresponding to the coal seam permeability change trend characteristics. Each element of the second dynamic time series vector corresponds to the coal seam permeability value at different times within the dynamic time series segment of geological conditions.

[0029] Extract the specific values ​​of coal seam permeability at each moment within the dynamic time series segment of geological conditions, and arrange them in chronological order to form a second dynamic time series vector. The length of the vector is equal to the number of moments within the segment, and each element corresponds to the coal seam permeability value at a given moment.

[0030] Step S1243: Input the first dynamic time series vector and the second dynamic time series vector into the time series correlation analysis module, and perform time dimension elastic alignment processing on the first dynamic time series vector and the second dynamic time series vector based on the dynamic time warping algorithm to obtain the aligned first dynamic time series vector and the aligned second dynamic time series vector.

[0031] After receiving the first and second dynamic time series vectors, the time series correlation analysis module initiates a dynamic time warping algorithm. This algorithm finds the optimal time mapping relationship and elastically stretches or compresses the time axes of the two vectors to match the corresponding trend moments in the vectors. For example, if a trend in the first dynamic time series vector occurs earlier than the corresponding trend moment in the second dynamic time series vector, the algorithm adjusts the time axis of the second dynamic time series vector to align the corresponding trend moments. Finally, it outputs aligned first and second dynamic time series vectors with the same length and matched trend moments.

[0032] Step S1244: Extract the rising inflection point and falling inflection point of the aligned first dynamic time series vector, and extract the rising inflection point and falling inflection point of the aligned second dynamic time series vector.

[0033] Inflection point detection is performed on the aligned first dynamic time-series vector. When the vector element value changes from rising to falling, the moment is the falling inflection point; when the element value changes from falling to rising, the moment is the rising inflection point. All rising and falling inflection points are recorded. The same inflection point detection method is used to process the aligned second dynamic time-series vector, and its rising and falling inflection points are extracted and recorded.

[0034] Step S1245: Count the number of times the rising inflection point of the first dynamic time series vector coincides with the rising inflection point of the second dynamic time series vector, and count the number of times the falling inflection point of the first dynamic time series vector coincides with the falling inflection point of the second dynamic time series vector. Add the two coincidence counts together to obtain the total number of coincidence inflection points.

[0035] The rising inflection point of the first dynamic time series vector is compared with the rising inflection point of the second dynamic time series vector. The number of times with completely identical timestamps is counted as the number of overlapping rising inflection points. Similarly, the number of overlapping falling inflection points is counted. The number of overlapping rising inflection points and the number of overlapping falling inflection points are added together to obtain the total number of overlapping inflection points.

[0036] Step S1246: Count the total number of inflection points of the first dynamic time series vector. The total number of inflection points of the first dynamic time series vector is the sum of the number of rising inflection points and the number of falling inflection points. Count the total number of inflection points of the second dynamic time series vector. The total number of inflection points of the second dynamic time series vector is the sum of the number of rising inflection points and the number of falling inflection points. Add the two total inflection point counts together to obtain the total number of inflection points.

[0037] Count the number of all rising and falling inflection points in the first dynamic time series vector separately, and the sum of the two is the total number of inflection points in the first dynamic time series vector; calculate the total number of inflection points in the second dynamic time series vector in the same way, and add the two total inflection point counts to get the total number of inflection points.

[0038] Step S1247: Calculate the ratio of the total number of overlapping inflection points to the total number of inflection points, as the inflection point coordination ratio.

[0039] The ratio obtained by dividing the total number of overlapping inflection points by the total number of inflection points is the inflection point coordination ratio. This inflection point coordination ratio reflects the degree of coordination between two dynamic time series vectors at the moment of trend reversal.

[0040] Step S1248: Calculate the mean change amplitude of the aligned first dynamic time-series vector, where the mean change amplitude is the average of the absolute values ​​of the differences between each time step and the previous time step. Calculate the mean change amplitude of the aligned second dynamic time-series vector, where the mean change amplitude is the average of the absolute values ​​of the differences between each time step and the previous time step. Use the ratio of the two mean change amplitudes as the amplitude coordination coefficient.

[0041] For the aligned first dynamic time-series vector, calculate the difference between the value at each time step and the value at the previous time step, take the absolute value of the difference, and then average all the absolute values ​​to obtain the mean amplitude of the first dynamic time-series vector's change. Calculate the mean amplitude of the second dynamic time-series vector's change using the same method. Divide the mean amplitude of the first dynamic time-series vector's change by the mean amplitude of the second dynamic time-series vector's change to obtain the amplitude coordination coefficient, which reflects the degree of matching between the two vectors in terms of amplitude.

[0042] Step S1249: Normalize the inflection point coordination ratio and the amplitude coordination coefficient respectively to convert them into dimensionless score values; set the weight of the inflection point coordination ratio and the weight of the amplitude coordination coefficient, the sum of the weight of the inflection point coordination ratio and the weight of the amplitude coordination coefficient is 1, multiply the normalized inflection point coordination ratio by the weight of the inflection point coordination ratio, and add the normalized amplitude coordination coefficient multiplied by the weight of the amplitude coordination coefficient to obtain the real-time correlation strength between the gas production rate change trend feature and the coal seam permeability change trend feature.

[0043] The min-max normalization method is used to transform the inflection point coordination ratio and the amplitude coordination coefficient into the range of 0 to 1, respectively, to obtain dimensionless scores. Based on the analysis results of historical data, weights are set for the inflection point coordination ratio and the amplitude coordination coefficient, with their sum being 1. The normalized inflection point coordination ratio is multiplied by its weight, and the normalized amplitude coordination coefficient is multiplied by its weight. The sum of the two products is the real-time correlation strength between the gas production rate change trend and the coal seam permeability change trend.

[0044] The real-time correlation strength between other trend characteristics, such as the trend characteristics of gas production rate change and surface temperature change, and the trend characteristics of bottom hole pressure change and coal seam porosity change, are all calculated according to the methods in steps S1241 to S1249 above.

[0045] Step S125: Introduce a time-series correlation strength correction coefficient. The time-series correlation strength correction coefficient is determined based on the correlation strength fluctuation range of the same type of data sequence in the same historical period. Multiply each real-time correlation strength by the corresponding time-series correlation strength correction coefficient to obtain the corrected correlation strength.

[0046] Collect production dynamics data, geological conditions data, and environmental data sequences from the same period over several years for this coalbed methane well. Calculate the correlation strength between similar trend characteristics during the same period, statistically analyze the fluctuation range of these correlation strengths, and determine the time-series correlation strength correction coefficient based on the fluctuation range. For example, if the correlation strength between historical gas production rate and coal seam permeability fluctuates little, the correction coefficient is close to 1; if the fluctuation is large, an appropriate correction coefficient is set according to the fluctuation amplitude.

[0047] A corresponding temporal correlation strength correction coefficient is matched for each real-time correlation strength. The real-time correlation strength is multiplied by the correction coefficient to obtain the corrected correlation strength, thereby eliminating the influence of seasonal, annual and other periodic factors on the correlation strength.

[0048] Step S126: Using each production dynamic time series segment, each geological condition dynamic time series segment, and each environmental dynamic time series segment as network nodes, and using the corrected correlation strength as the initial weight of the connection between nodes, construct an initial ternary time series dynamic correlation network.

[0049] Each segment in the sets of dynamic time series segments of production, geological conditions, and environment is mapped to a node in the network, and a unique node identifier is assigned to each node. For nodes of different categories, such as production dynamic nodes and geological condition nodes, production dynamic nodes and environmental nodes, and geological condition nodes and environmental nodes, connections are established between nodes based on the corresponding corrected association strength, and the corrected association strength is used as the initial weight of the connection.

[0050] For example, the connection weight between a production dynamic time series segment node and a geological condition dynamic time series segment node represents the corrected correlation strength between their trend characteristics; the connection weight between a production dynamic time series segment node and an environmental dynamic time series segment node represents the corresponding corrected correlation strength. An initial ternary time series dynamic correlation network is constructed in this way and stored in a graph database.

[0051] Step S127: Set the weight update cycle, recalculate and correct the real-time association strength between nodes according to the weight update cycle, update the connection weights of the initial ternary time-series dynamic association network, and obtain the final ternary time-series dynamic association network of coalbed methane wells.

[0052] The weight update cycle is set according to the frequency of data changes, and this weight update cycle can be adjusted according to actual production needs. In each weight update cycle, the latest production dynamic data sequence, geological condition data sequence, and environmental data sequence are re-collected, and the real-time correlation strength between various nodes is recalculated and corrected according to the methods in steps S121 to S125 to obtain the updated correlation strength.

[0053] The initial weights of the connections between corresponding nodes in the initial ternary time-series dynamic association network are replaced with the updated association strengths to achieve dynamic updates of the network weights. After multiple weight updates, the final ternary time-series dynamic association network for coalbed methane wells is formed, which can reflect the changes in the association relationships between time-series segments of different categories of data in real time.

[0054] Step S130: Call the pre-trained AI temporal coupling deviation evolution mining model to process the ternary temporal dynamic correlation network, mine the coupling deviation evolution information between dynamic temporal segments of different data sequences, and obtain the ternary temporal coupling deviation result containing the deviation evolution trend.

[0055] In this embodiment, a pre-trained AI model is used to perform in-depth analysis on the ternary temporal dynamic correlation network to mine coupling deviation information. This is achieved through the following sub-steps.

[0056] Step S131: Input the ternary temporal dynamic association network into the network parsing layer of the AI ​​temporal coupling deviation evolution mining model, parse the dynamic temporal segment data corresponding to each node, the dynamic temporal segment data corresponding to each node includes the specific data values ​​at each time within the segment, parse the corrected association strength data and weight update records corresponding to the connections between nodes, and obtain the network parsing data.

[0057] The network parsing layer of the AI ​​temporal coupling deviation evolution mining model receives the ternary temporal dynamic correlation network and then parses the network structure and data. First, it extracts the identifier of each node and the corresponding dynamic temporal segment data. This dynamic temporal segment data contains the specific values ​​at each moment within the segment, such as the gas production rate and bottom hole pressure at each moment within the production dynamic node segment, and the permeability and porosity at each moment within the geological condition node segment.

[0058] Then, the attribute information of the connections between nodes is parsed, including the corrected association strength data and the update records of the weights, which contain the update time and the weight value for each update. The parsed node data, connection data, and weight update records are integrated into network parsing data and passed to the next layer of the model.

[0059] Step S132: Input the network parsing data into the temporal coupling analysis layer of the AI ​​temporal coupling deviation evolution mining model, and analyze the coupling matching degree between the dynamic time series segments of production dynamics and the dynamic time series segments of geological conditions from three dimensions: trend consistency, phase difference synergy, and amplitude synergy. Among them, trend consistency is determined by calculating the proportion of the matching duration of the change trends of the two types of dynamic time series segments, phase difference synergy is determined by calculating the average time difference of the inflection point of the two types of dynamic time series segments, and amplitude synergy is determined by calculating the fluctuation range of the ratio of the change amplitude of the two types of dynamic time series segments.

[0060] After receiving the network parsing data, the temporal coupling analysis layer calculates the coupling matching degree from three dimensions for the combination of dynamic time series segments of production dynamics and dynamic time series segments of geological conditions.

[0061] In terms of trend consistency, the trend curves of the two types of segments are compared, the duration of consistent curve trends is counted, and the ratio obtained by dividing the duration by the total duration of the segments is the trend consistency index.

[0062] In the phase difference synergy dimension, all inflection point moments of the two types of segments are extracted, the time difference of the corresponding inflection point moments is calculated, and the average of all time differences is calculated to obtain the phase difference synergy index. The smaller the mean of the time difference, the higher the phase difference synergy.

[0063] In terms of amplitude coherence, the change amplitudes of the two types of segments are calculated separately to obtain the amplitude ratio sequence. The fluctuation range of this sequence is then statistically analyzed. The smaller the fluctuation range, the higher the amplitude coherence.

[0064] Through calculations in the above three dimensions, basic data on the coupling matching degree between dynamic time series segments of production dynamics and dynamic time series segments of geological conditions are obtained.

[0065] Step S133: Normalize the coupling matching degree of the three dimensions respectively, set weights for the normalized coupling matching degree of each dimension, and calculate the weighted sum of the coupling matching degree scores of the three dimensions according to the set weights, which is used as the comprehensive coupling matching degree of the two types of dynamic time series segments.

[0066] The indicators for the three dimensions of trend consistency, phase difference synergy, and amplitude synergy are normalized and converted into scores between 0 and 1. Based on the influence of each dimension on the coupling matching degree, a weight is assigned to each normalized score, with the sum of the three weights being 1. The normalized trend consistency score, the normalized phase difference synergy score, and the normalized amplitude synergy score are multiplied by their respective weights, and the sum of these three products is the comprehensive coupling matching degree between the production dynamic time series segment and the geological condition dynamic time series segment. Using the same method, the comprehensive coupling matching degree between the production dynamic time series segment and the environmental dynamic time series segment, and between the geological condition dynamic time series segment and the environmental dynamic time series segment, is calculated separately, achieving a quantitative assessment of the coupling matching of all different types of dynamic time series segment combinations.

[0067] Step S134: Identify dynamic time-series segment combinations with a comprehensive coupling matching degree lower than a preset matching threshold, and mark the dynamic time-series segment combinations as deviation candidate combinations.

[0068] Based on historical production data and coupling matching under normal production conditions, a preset matching threshold is set. All combinations of dynamic time series segments of different categories and their corresponding comprehensive coupling matching degrees are traversed, and combinations with comprehensive coupling matching degrees less than the preset matching threshold are filtered out. For example, if the comprehensive coupling matching degree of a certain production dynamic time series segment and a certain geological condition dynamic time series segment is lower than the preset threshold, it indicates that their coupling relationship deviates from the normal level. This combination is marked as a deviation candidate combination, and the node identifiers, categories, and comprehensive coupling matching degree values ​​of the two dynamic time series segments in this combination are recorded.

[0069] Step S135: For each candidate combination of deviations, extract the change data of its comprehensive coupling matching degree in multiple consecutive weight update cycles, and construct a deviation evolution curve. The horizontal axis of the deviation evolution curve is the weight update cycle, and the vertical axis of the deviation evolution curve is the comprehensive coupling matching degree.

[0070] From the weight update records of the ternary time-series dynamic association network, the comprehensive coupling matching degree data of each deviation candidate combination over multiple consecutive weight update cycles is extracted. The data points are plotted on a coordinate system with the weight update cycle as the x-axis and the comprehensive coupling matching degree corresponding to each cycle as the y-axis. Then, a curve fitting algorithm is used to connect these data points into a smooth curve, forming the deviation evolution curve of the deviation candidate combination. This curve allows for a direct observation of the trajectory of the comprehensive coupling matching degree over time.

[0071] Step S136: Analyze the slope change trend of the deviation evolution curve. If the slope is negative and the absolute value gradually increases, the deviation is determined to be expanding. If the slope is negative and the absolute value gradually decreases, the deviation is determined to be slowing down. If the slope is positive, the deviation is determined to be correcting.

[0072] Slope analysis is performed on the deviation evolution curve to calculate the slope of the curve within each weight update cycle interval. If the slope of multiple consecutive intervals is negative and the absolute value of the slope gradually increases, it indicates that the overall coupling matching degree is decreasing at an accelerated rate, and the deviation of the candidate combination is being judged to be expanding. If the slope of multiple consecutive intervals is negative but the absolute value of the slope gradually decreases, it indicates that the overall coupling matching degree is decreasing at a slower rate, and the judgment deviation is showing a mitigating trend. If the slope of multiple consecutive intervals is positive, it means that the overall coupling matching degree is increasing, and the judgment deviation is showing a corrective trend.

[0073] Step S137: Calculate the degree of deviation for the dynamic time segments in the candidate deviation combinations to obtain the deviation value corresponding to each candidate deviation combination.

[0074] The calculation of the degree of deviation requires multi-dimensional comparison with reference to the standard coupling relationship, which is achieved through the following sub-steps.

[0075] Step S1371: Collect production dynamic data sequences, geological condition data sequences, and environmental data sequences of coalbed methane wells under historical normal production conditions, extract coupling data of similar dynamic time series segments, and construct a standard coupling relationship library. The standard coupling relationship library stores the standard trend consistency range, standard phase difference coordination range, and standard amplitude coordination range of different types of dynamic time series segment combinations.

[0076] Collect production dynamic data sequences, geological condition data sequences, and environmental data sequences from the coalbed methane well during its historical period of normal production. Divide the above data sequences into dynamic time series segments according to the division rules in step S122. Extract coupled data of similar dynamic time series segment combinations (such as production-geology, production-environment, geology-environment), including historical normal values ​​of trend consistency, phase difference coordination, and amplitude coordination.

[0077] Statistical analysis was performed on the above historical normal values ​​to determine the normal fluctuation range of different types of dynamic time series segment combinations in three dimensions: standard trend consistency range, standard phase difference coordination range, and standard amplitude coordination range. The above ranges and the corresponding dynamic time series segment combination types were stored in the standard coupling relationship library.

[0078] Step S1372: Based on the type of deviation candidate combination, retrieve the corresponding standard trend consistency range, standard phase difference synergy range, and standard amplitude synergy range from the standard coupling relationship library as the standard coupling relationship of the deviation candidate combination. The types of deviation candidate combinations include production-geological deviation, production-environmental deviation, and geological-environmental deviation.

[0079] The type of each candidate combination of deviations is determined. For example, the combination type composed of production dynamic time series segments and geological condition dynamic time series segments is production-geological deviation, the combination type composed of production dynamic time series segments and environmental dynamic time series segments is production-environmental deviation, and the combination type composed of geological condition dynamic time series segments and environmental dynamic time series segments is geological-environmental deviation.

[0080] Based on the determined type, a matching query is performed in the standard coupling relationship database to retrieve the standard trend consistency range, standard phase difference synergy range, and standard amplitude synergy range corresponding to that type, which serve as the standard coupling relationship for evaluating the degree of deviation of the candidate combination of deviation.

[0081] Step S1373: Extract the actual change curve of the first type of dynamic time series segment in the deviation candidate combination, and extract the actual change curve of the second type of dynamic time series segment in the deviation candidate combination. The horizontal axis of the actual change curve is the time within the segment, and the vertical axis of the actual change curve is the data value.

[0082] Taking two dynamic time series segments from the candidate deviation combinations as objects, specific data values ​​at each time point within each segment are extracted. The actual change curves of the two dynamic time series segments are plotted with the time point on the x-axis and the data values ​​on the y-axis. These are the actual change curves of the first type of dynamic time series segment and the second type of dynamic time series segment. For example, in the candidate deviation combinations of production-geological types, the first type is the production dynamic time series segment, with its actual change curve plotted with time point on the x-axis and gas production rate and bottom hole pressure on the y-axis; the second type is the geological condition dynamic time series segment, with its actual change curve plotted with time point on the x-axis and coal seam permeability and porosity on the y-axis.

[0083] Step S1374: Based on the standard trend consistency range in the standard coupling relationship, generate the standard trend curve range corresponding to the first type of dynamic time series segment, and generate the standard trend curve range corresponding to the second type of dynamic time series segment. The standard trend curve range includes an upper limit trend curve and a lower limit trend curve.

[0084] Based on the standard trend consistency range in the standard coupling relationship, and combined with the basic data of the first type of dynamic time series segment (such as initial data values ​​and average rate of change), the standard trend curve range corresponding to this segment is generated. This standard trend curve range consists of an upper trend curve and a lower trend curve, and the area between the two curves is the data fluctuation range of this segment under normal trend. Using the same method, the standard trend curve range corresponding to the second type of dynamic time series segment is generated based on the standard trend consistency range, which also includes upper and lower trend curves.

[0085] Step S1375: Calculate the area difference between the actual change curve of the first type of dynamic time series segment and the upper limit of the standard trend curve range, calculate the area difference between the actual change curve of the first type of dynamic time series segment and the lower limit of the standard trend curve range, and take the maximum value of the two area differences as the trend deviation area of ​​the first type of segment.

[0086] The actual change curve of the first type of dynamic time series segment is compared with the upper limit curve of the standard trend curve range. The area enclosed by the two curves is calculated, which is the area of ​​difference between the actual change curve and the upper limit curve. Then, the area enclosed by the actual change curve and the lower limit curve is calculated, which is also the area of ​​difference between the actual change curve and the lower limit curve. The two difference areas are compared, and the larger one is selected as the trend deviation area of ​​the first type of segment. The larger this area, the more severe the trend deviation of the first type of dynamic time series segment.

[0087] Step S1376: Calculate the area difference between the actual change curve of the second type of dynamic time series segment and the upper limit of the standard trend curve range, calculate the area difference between the actual change curve of the second type of dynamic time series segment and the lower limit of the standard trend curve range, and take the maximum value of the two area differences as the trend deviation area of ​​the second type of segment.

[0088] Using the same method as in step S1375, calculate the area difference between the actual change curve of the second type of dynamic time series segment and the upper and lower limits of the standard trend curve range. Select the maximum value as the trend deviation area of ​​the second type of segment to quantify the degree of trend deviation of the second type of dynamic time series segment.

[0089] Step S1377: Based on the standard phase difference coordination range in the standard coupling relationship, calculate the absolute value of the difference between the actual phase difference and the upper limit of the standard phase difference coordination range of the two types of dynamic time segments in the deviation candidate combination. The actual phase difference is the difference at the inflection point. Calculate the absolute value of the difference between the actual phase difference and the lower limit of the standard phase difference coordination range, and take the maximum value of the two absolute values ​​of the difference as the phase difference deviation value.

[0090] The actual phase difference (i.e., the difference at the corresponding inflection point) between the two types of dynamic time segments in the candidate deviation combinations is extracted. This actual phase difference is compared with the upper and lower limits of the standard phase difference compatibility range, respectively. The absolute values ​​of the differences between the actual phase difference and the upper and lower limits are calculated. The two absolute values ​​are compared, and the larger value is selected as the phase difference deviation value. The larger the phase difference deviation value, the more significant the deviation from the normal range.

[0091] Step S1378: Based on the standard amplitude coordination range in the standard coupling relationship, calculate the absolute value of the difference between the actual amplitude ratio of the two types of dynamic time segments in the deviation candidate combination and the upper limit of the standard amplitude coordination range. The actual amplitude ratio is the ratio of the average amplitude change. Calculate the absolute value of the difference between the actual amplitude ratio and the lower limit of the standard amplitude coordination range. Take the maximum value of the two absolute values ​​of difference as the amplitude coordination deviation value.

[0092] Calculate the actual amplitude ratio (i.e., the ratio of the average amplitude changes of the two types of dynamic time segments) in the candidate combination of deviations. Then, compare this actual amplitude ratio with the upper and lower limits of the standard amplitude coherence range to obtain the absolute values ​​of the differences between the actual amplitude ratio and the upper and lower limits, respectively. Compare the two absolute values ​​and select the larger value as the amplitude coherence deviation value, reflecting the degree of deviation from amplitude coherence.

[0093] Step S1379: Set the trend deviation weight, phase difference deviation weight, and amplitude coordination deviation weight, and the sum of the trend deviation weight, phase difference deviation weight, and amplitude coordination deviation weight is 1; normalize the average value of the trend deviation area of ​​the first type of segment and the trend deviation area of ​​the second type of segment, the phase difference deviation value, and the amplitude coordination deviation value, respectively; multiply the normalized trend deviation area score by the trend deviation weight, add the normalized phase difference deviation score multiplied by the phase difference deviation weight, and add the normalized amplitude coordination deviation score multiplied by the amplitude coordination deviation weight to obtain the deviation value corresponding to the deviation candidate combination.

[0094] Based on the weights of the three deviation indicators on the overall deviation degree, trend deviation weight, phase difference deviation weight, and amplitude coordination deviation weight are set, and the sum of the three is 1. The average of the trend deviation area of ​​the first type of segment and the trend deviation area of ​​the second type of segment is calculated as the comprehensive trend deviation area.

[0095] The area of ​​the overall trend deviation, the phase difference deviation, and the amplitude coordination deviation are all normalized and converted into a score between 0 and 1. The normalized area of ​​the overall trend deviation score is multiplied by the trend deviation weight, the normalized phase difference deviation score is multiplied by the phase difference deviation weight, and the normalized amplitude coordination deviation score is multiplied by the amplitude coordination deviation weight. The sum of these three products is the deviation value corresponding to the candidate deviation combination. The larger the deviation value, the more severe the overall deviation.

[0096] Step S138: Integrate candidate combinations of deviations, deviation values, and deviation evolution trends to generate a ternary time-series coupled deviation result containing deviation location, deviation type, deviation degree, and deviation evolution trend. The deviation location corresponds to the dynamic time-series segment number, the deviation type includes production-geological deviation, production-environment deviation, and geological-environment deviation, and the deviation degree is the deviation value.

[0097] The information of each candidate deviation combination is integrated, wherein the deviation position is represented by the node number of the two dynamic time segments in the combination, the deviation type is determined according to the combination category as production-geological deviation, production-environmental deviation or geological-environmental deviation, the deviation degree is the deviation value calculated in step S1379, and the deviation evolution trend is the expansion, mitigation or repair trend determined in step S136.

[0098] The above information is organized in a unified format to form a ternary time series coupling bias result, which presents the specific location, type, severity and development trend of each bias.

[0099] Step S140: Based on the ternary time-series coupling deviation results, combined with forward transmission positioning and reverse tracing verification, determine the deviation transmission closed-loop path affecting production dynamics, and generate a deviation transmission analysis report containing the deviation transmission direction, correlation strength and tracing verification results.

[0100] Based on the results of the ternary time-series coupling deviation, the deviation propagation path is sorted out by combining forward localization and reverse verification. Specifically, this is achieved through the following sub-steps.

[0101] Step S141: Extract all candidate combinations of deviations and their corresponding deviation values ​​and deviation evolution trends from the ternary time-series coupling deviation results, screen out candidate combinations of deviations with an expanding trend, and sort the screened candidate combinations of deviations in descending order of deviation values.

[0102] Iterate through all candidate combinations of deviations in the ternary time-series coupling deviation results, extracting the deviation value and evolution trend for each combination. Select combinations with an increasing deviation trend; these combinations indicate the deviation is intensifying and their propagation path should be analyzed first. Sort the selected combinations in descending order of deviation value, placing combinations with larger deviation values ​​at the top for priority propagation path analysis.

[0103] Step S142: Select the top-ranked preset number of candidate deviation combinations as key deviation combinations, extract the timestamp information of the two types of dynamic time series segments in each key deviation combination, and determine the temporal order of the two types of dynamic time series segments.

[0104] Based on actual analysis needs, a preset number is set, and the candidate combinations of deviations with the highest ranking are selected as key deviation combinations. The timestamp information of the two dynamic time series segments in each key deviation combination is extracted, namely the start and end times of the segments. By comparing the start times of the two segments, their temporal order is determined; the segment with the earlier start time is the earlier dynamic time series segment, and the segment with the later start time is the later dynamic time series segment.

[0105] Step S143: Based on the chronological order, mark the first dynamic timing segment as the starting dynamic timing segment and the later dynamic timing segment as the affected dynamic timing segment, and preliminarily determine the positive propagation direction of the deviation from the starting dynamic timing segment to the affected dynamic timing segment.

[0106] Based on the temporal sequence determined in step S142, the first appearing dynamic time series segment is defined as the initial dynamic time series segment, which is considered the starting source of the deviation. The later appearing dynamic time series segment is defined as the affected dynamic time series segment, and the deviation of the affected dynamic time series segment is considered to be propagated from the initial dynamic time series segment. Therefore, the forward propagation direction of the deviation is initially determined to be from the initial dynamic time series segment to the affected dynamic time series segment. For example, in a critical deviation combination, if the geological condition dynamic time series segment appears first and the production dynamic time series segment appears later, then the initial dynamic time series segment is the geological condition dynamic time series segment, the affected dynamic time series segment is the production dynamic time series segment, and the forward propagation direction is from the geological condition dynamic time series segment to the production dynamic time series segment.

[0107] Step S144: Extract the connection strength between the node corresponding to the starting dynamic time segment and the node corresponding to the affected dynamic time segment from the ternary time-series dynamic association network. At the same time, extract the connection strength between the node corresponding to the starting dynamic time segment and other related nodes. Other related nodes are nodes that are associated with the affected dynamic time segment. Extract the connection strength between the node corresponding to the affected dynamic time segment and other related nodes.

[0108] In a ternary temporal dynamic correlation network, the correlation strength of the connections between the initial dynamic temporal segment and the affected dynamic temporal segment is determined based on their node identifiers. This correlation strength reflects the closeness of their direct association. Simultaneously, all other nodes connected to the affected dynamic temporal segment node are identified; these are the other related nodes. The correlation strength between the initial dynamic temporal segment node and each other related node, as well as the correlation strength between the affected dynamic temporal segment node and each other related node, are extracted to prepare for constructing the transmission strength matrix.

[0109] Step S145: Construct a forward conduction strength matrix. The rows of the forward conduction strength matrix represent the starting node and related nodes, the columns of the forward conduction strength matrix represent the affected node and related nodes, and the elements of the forward conduction strength matrix are the correlation strength between the corresponding nodes.

[0110] Constructing the positive propagation strength matrix requires defining the nodes corresponding to the rows and columns of the matrix and filling them with the corresponding correlation strengths. This can be achieved through the following sub-steps.

[0111] Step S1451: Determine the starting node number corresponding to the initial dynamic timing segment, and determine the affected node number corresponding to the affected dynamic timing segment.

[0112] The node identifier corresponding to the initial dynamic time series segment in the ternary dynamic time series association network is determined as the starting node number, and the node identifier corresponding to the affected dynamic time series segment is determined as the affected node number, ensuring that the number is completely consistent with the node identifier in the network.

[0113] Step S1452: From the node association list of the ternary time-series dynamic association network, query all nodes that are connected to the starting node and use them as the starting related nodes; query all nodes that are connected to the affected node and use them as the affected related nodes.

[0114] Access the node association list of the ternary temporal dynamic association network. This list records information about all associated nodes for each node. Based on the starting node number, query all nodes in the list that are connected to that node, and designate these nodes as the starting associated nodes. Based on the affected node number, query all nodes in the list that are connected to that node, and designate these nodes as the affected associated nodes.

[0115] Step S1453: Integrate the starting node and the starting related nodes to form a set of matrix row nodes, and arrange the nodes in the set of matrix row nodes in ascending order of node number; integrate the affected node and the affected related nodes to form a set of matrix column nodes, and arrange the nodes in the set of matrix column nodes in ascending order of node number.

[0116] The starting node and all related nodes are merged to form a set of matrix row nodes. A sorting algorithm is then used to arrange the nodes in this set in ascending order of their node numbers to determine the node order for each row of the matrix. Similarly, the affected nodes and all related affected nodes are merged to form a set of matrix column nodes, which are then arranged in ascending order of their node numbers to determine the node order for each column of the matrix.

[0117] Step S1454: Construct a blank positive conduction intensity matrix. The number of rows in the blank positive conduction intensity matrix is ​​equal to the number of nodes in the matrix row node set, and the number of columns in the blank positive conduction intensity matrix is ​​equal to the number of nodes in the matrix column node set.

[0118] The number of rows in the blank matrix is ​​determined based on the number of nodes in the matrix row node set, and the number of columns in the blank matrix is ​​determined based on the number of nodes in the matrix column node set. A blank positive conduction intensity matrix with matching row and column numbers is created, and all elements in the matrix are initialized to 0.

[0119] Step S1455: For each node in the set of matrix row nodes, denote the node as a row node; for each node in the set of matrix column nodes, denote the node as a column node. Query the corrected association strength between row nodes and column nodes in the ternary temporal dynamic association network. If there is a connection between row nodes and column nodes, fill the corrected association strength into the element positions of the row corresponding to the row node and the column corresponding to the column node in the forward transmission strength matrix. If there is no connection between row nodes and column nodes, set the element value to 0 and fill it into the element positions of the row corresponding to the row node and the column corresponding to the column node in the forward transmission strength matrix to obtain the complete forward transmission strength matrix.

[0120] The algorithm iterates through each row node in the matrix's row node set and each column node in the matrix's column node set. For each pair of row and column nodes, it queries the ternary temporal dynamic association network to determine if a connection exists between them. If a connection exists, the modified association strength is extracted, and this strength value is filled into the intersection element of the row corresponding to the row node and the column corresponding to the column node in the forward propagation strength matrix. If no connection exists, the value of this element is kept at 0. After the iteration is complete, the complete forward propagation strength matrix is ​​obtained. This forward propagation strength matrix comprehensively presents the distribution of association strengths between the starting node, the starting related node and the affected node, and the affected related node. For example, if a row node in the matrix corresponds to the starting node and a column node corresponds to the affected node, the value of the intersection element is the modified association strength between the starting node and the affected node. If the starting related node and the affected related node are associated, the corresponding element is filled with the corresponding association strength; otherwise, it is 0.

[0121] Step S146: Calculate the cumulative association strength of the forward propagation path, and filter the paths with a cumulative association strength higher than the forward propagation threshold as candidate forward propagation paths. The cumulative association strength is the average value of the association strength between all nodes on the path.

[0122] Based on the forward propagation strength matrix, all possible paths starting from the starting node, passing through the initial related node or the affected related node, and finally reaching the affected node are enumerated. For each path, the association strength values ​​between each adjacent node on the path are extracted. The sum of these values ​​is then divided by the number of connections between nodes on the path to obtain the cumulative association strength of the path, which is the average value of all association strengths on the path.

[0123] A forward conduction threshold is set, which is determined based on the statistical results of the cumulative correlation strength of historical normal conduction paths. The cumulative correlation strength of each path is compared with the forward conduction threshold, and paths with a cumulative correlation strength higher than the threshold are selected as candidate forward conduction paths.

[0124] Step S147: For each candidate forward propagation path, initiate reverse tracing verification: Starting from the affected dynamic time segment and ending from the initial dynamic time segment, query the reverse association path from the affected node to the starting node in the ternary time-series dynamic association network, and calculate the cumulative association strength of the reverse association path. The cumulative association strength is the average value of the association strength between all nodes on the path.

[0125] For each candidate forward propagation path, the reverse tracing verification process starts with the affected node corresponding to the affected dynamic time segment and ends with the starting node corresponding to the starting dynamic time segment. It searches for all possible reverse association paths in the ternary time-series dynamic association network, that is, paths where the node connection direction is opposite to the forward path.

[0126] For each reverse association path, the same calculation method as for the forward path is used to extract the association strength values ​​between each adjacent node on the path, and calculate the average of these values ​​as the cumulative association strength of the reverse association path.

[0127] Step S148: If the cumulative association strength of the reverse association path is higher than the reverse tracing threshold, and the reverse association path and the candidate forward propagation path form a closed loop structure with interconnected nodes, then the candidate forward propagation path is determined to be a valid deviation propagation closed loop path; if the cumulative association strength of the reverse association path is lower than the reverse tracing threshold, or cannot form a closed loop structure, then the candidate forward propagation path is eliminated.

[0128] A reverse tracing threshold is set, which is determined with reference to the forward propagation threshold and the characteristics of reverse propagation. The cumulative association strength of the reverse association path is compared with the reverse tracing threshold. At the same time, it is checked whether the reverse association path and the corresponding candidate forward propagation path form a closed loop structure with interconnected nodes. That is, the starting point of the forward path is the ending point of the reverse path, and the ending point of the forward path is the starting point of the reverse path. Moreover, the nodes traversed by the two paths overlap or complement each other, forming a complete closed loop.

[0129] If the cumulative correlation strength of the reverse correlation path is higher than the reverse tracing threshold and forms a closed loop with the candidate forward transmission path, then the candidate forward transmission path is determined to be a valid deviation transmission closed loop path; if the cumulative correlation strength of the reverse correlation path is lower than the reverse tracing threshold, or the two paths cannot form a closed loop, then the candidate forward transmission path is eliminated.

[0130] Step S149: Calculate the forward cumulative correlation strength, reverse cumulative correlation strength, number of nodes involved, and dynamic time series segment type for each effective deviation transmission closed-loop path. Integrate the deviation transmission direction, correlation strength, and traceability verification results of the effective deviation transmission closed-loop path to generate a deviation transmission analysis report. The deviation transmission direction is the forward transmission direction, the correlation strength is the forward cumulative correlation strength, and the traceability verification results include the reverse cumulative correlation strength and the closed-loop verification results.

[0131] For each effective deviation transmission closed-loop path, information statistics are performed, including the positive cumulative correlation strength of the path (i.e., the cumulative correlation strength of candidate positive transmission paths), the negative cumulative correlation strength (i.e., the cumulative correlation strength of the corresponding negative correlation path), the number of all nodes involved in the path, and the dynamic time series segment type (production dynamics, geological conditions, or environment) corresponding to these nodes.

[0132] The deviation transmission direction (i.e., forward transmission direction), correlation strength (i.e., forward cumulative correlation strength), and traceability verification results (including reverse cumulative correlation strength and closed-loop verification results, with the closed-loop verification result being "pass" or "fail") of the effective deviation transmission closed-loop path are integrated and compiled into a deviation transmission analysis report according to a preset report format. This deviation transmission analysis report clearly presents the key information of each effective deviation transmission path.

[0133] Step S150: Based on the deviation transmission analysis report and combined with the production dynamic prediction threshold, generate a dynamic control strategy for the production parameters of coalbed methane wells, and output the dynamic control strategy to the production control terminal. At the same time, feed back the controlled data to the ternary time-series dynamic correlation network to update the node correlation relationship.

[0134] The generation and implementation of dynamic control strategies and the network update process are achieved through the following sub-steps.

[0135] Step S151: Analyze the effective deviation transmission closed-loop paths in the deviation transmission analysis report, determine the initial deviation type corresponding to each effective deviation transmission closed-loop path, including geological condition deviation and environmental factor deviation, and extract the positive cumulative correlation strength and negative cumulative correlation strength of each effective deviation transmission closed-loop path.

[0136] The deviation transmission analysis report is analyzed, and all effective deviation transmission closed-loop paths are extracted. The initial deviation type is determined based on the type of the initial dynamic time series segment for each path: if the initial dynamic time series segment is a geological condition dynamic time series segment, the initial deviation type is geological condition deviation; if the initial dynamic time series segment is an environmental dynamic time series segment, the initial deviation type is environmental factor deviation. Simultaneously, the positive cumulative correlation strength and negative cumulative correlation strength of each path are extracted as a basis for subsequently determining the control priority.

[0137] Step S152: Call the pre-trained production dynamics prediction model, input the current production dynamics data sequence, the current geological conditions data sequence, and the current environmental data sequence, and predict the gas production rate and bottom hole pressure within a preset time period in the future, as the production dynamics prediction result.

[0138] The process of constructing, training, and predicting the production dynamics forecasting model is as follows.

[0139] Step S1521: Collect historical coalbed methane well production dynamic data sequences, geological condition data sequences, environmental data sequences, and corresponding actual gas production rate data and actual well bottom pressure data within a future preset time period to construct a training dataset for the production dynamic prediction model.

[0140] Collect long-term production dynamics data sequences, geological condition data sequences, and environmental data sequences for the coalbed methane well. Simultaneously, collect the actual gas production rate data and actual bottom hole pressure data for the corresponding future preset time period from these historical data. Pair each set of historical input data (production dynamics, geological conditions, and environmental data sequences) with the corresponding future actual output data (gas production rate and bottom hole pressure data) to construct a training dataset for the production dynamics prediction model.

[0141] Step S1522: Divide the training dataset of the production dynamic prediction model into a training set, a validation set, and a test set. The training set is used for model parameter learning, the validation set is used for model hyperparameter tuning, and the test set is used for model performance evaluation.

[0142] The training dataset is randomly divided into training, validation, and test sets according to a certain ratio. The training set is used to learn the latent patterns in the data and determine the basic parameters of the model; the validation set is used to adjust the hyperparameters of the model during training, such as the number of network layers and hidden nodes; the test set is used to evaluate the generalization performance of the model after training, ensuring that the model can maintain good prediction performance on unseen data.

[0143] Step S1523: Construct the network structure of the production dynamic prediction model. The network structure of the production dynamic prediction model includes a temporal feature encoding layer, a multi-source feature fusion layer, and a prediction output layer. The temporal feature encoding layer adopts a bidirectional long short-term memory network structure to extract temporal features from the input production dynamic data sequence, geological condition data sequence, and environmental data sequence, and outputs production dynamic temporal encoding features, geological condition temporal encoding features, and environmental temporal encoding features. The multi-source feature fusion layer adopts an attention mechanism fusion structure to perform weighted fusion of production dynamic temporal encoding features, geological condition temporal encoding features, and environmental temporal encoding features. The weights are dynamically adjusted based on the degree of influence of each feature on the production dynamic prediction result, and output multi-source fused features. The prediction output layer adopts a fully connected network structure to process the multi-source fused features and output the predicted gas production rate and bottom hole pressure at each moment within a preset time period.

[0144] The temporal feature encoding layer consists of a bidirectional long short-term memory network, which contains memory units in both forward and backward directions, enabling it to simultaneously capture past and future information of the input sequence. Production dynamics data sequences, geological conditions data sequences, and environmental data sequences are input into three parallel branches of this layer, respectively. Each branch independently extracts the temporal features of its corresponding sequence, outputting production dynamics temporal encoding features, geological conditions temporal encoding features, and environmental temporal encoding features. These features are all in high-dimensional vector form.

[0145] The multi-source feature fusion layer employs an attention mechanism, assigning different weights to different features by calculating the correlation between each temporal encoded feature and the prediction target. Features with higher correlation have larger weights and contribute more to the fusion result. The three temporal encoded features are then weighted and concatenated according to their respective weights to obtain the multi-source fused feature.

[0146] The prediction output layer consists of multiple fully connected layers. It processes the multi-source fusion features through linear transformation and activation functions, mapping high-dimensional features to a low-dimensional prediction space, and finally outputs the predicted gas production rate and bottom hole pressure for each time point within a preset time period.

[0147] Step S1524: Train the production dynamic prediction model using the training set. After each preset training round, calculate the prediction error of the production dynamic prediction model using the validation set. The prediction error is the mean square error between the predicted and actual gas production rate and the mean square error between the predicted and actual wellbore pressure. If the prediction error is higher than a preset error threshold, adjust the number of hidden layer nodes in the bidirectional long short-term memory network, the weight calculation method of the attention mechanism, and the number of layers in the fully connected network, and retrain the production dynamic prediction model. If the prediction error is lower than the preset error threshold, perform final performance verification using the test set. If the prediction error of the test set is also lower than the preset error threshold, a pre-trained production dynamic prediction model is obtained.

[0148] The input data from the training set is fed into the production dynamics prediction model, and the model's predicted output is calculated through forward propagation. Mean squared error (MSE) is used as the loss function to calculate the MSE between the predicted and actual gas production rate, and the MSE between the predicted and actual bottom hole pressure. The two errors are then summed to obtain the total loss value. Using the backpropagation algorithm, the parameters of each layer of the model are adjusted based on the total loss value, including the weights and biases of the bidirectional long short-term memory network, the parameters of the attention mechanism, and the weights and biases of the fully connected network.

[0149] After each preset number of training rounds, the model is tested using a validation set, and the prediction error on the validation set is calculated. If the prediction error on the validation set is higher than a preset error threshold, it indicates that the model may be overfitting or the parameters may be set inappropriately. In this case, the number of hidden layer nodes in the bidirectional long short-term memory network, the weight calculation method of the attention mechanism (such as changing the correlation calculation function), or the number of layers in the fully connected network need to be adjusted, and then retraining is performed.

[0150] If the prediction error on the validation set is lower than the preset error threshold, the model is finally validated using the test set. If the prediction error on the test set is also lower than the preset error threshold, it indicates that the model performs well. The model parameters at this point are saved to obtain the pre-trained production dynamic prediction model.

[0151] Step S1525: Input the current production dynamic data sequence, current geological condition data sequence, and current environmental data sequence into the pre-trained production dynamic prediction model. Extract the current production dynamic time-series encoded features, current geological condition time-series encoded features, and current environmental time-series encoded features through the time-series feature encoding layer. Perform weighted fusion of the current production dynamic time-series encoded features, current geological condition time-series encoded features, and current environmental time-series encoded features through the multi-source feature fusion layer to obtain the current multi-source fusion features. Process the current multi-source fusion features through the prediction output layer to obtain the predicted gas production rate and bottom hole pressure values ​​at each moment within the future preset time period, which are used as the production dynamic prediction results.

[0152] The collected current production dynamics data sequence, current geological conditions data sequence, and current environmental data sequence are input into the pre-trained production dynamics prediction model. First, the three parallel branches of the temporal feature encoding layer extract features from the three types of current data sequences, respectively, and output the current production dynamics temporal encoding features, current geological conditions temporal encoding features, and current environmental temporal encoding features.

[0153] Subsequently, the multi-source feature fusion layer calculates the attention weights of the three types of current temporal encoded features, performs weighted fusion, and generates the current multi-source fused features. Finally, the prediction output layer processes the current multi-source fused features and outputs the predicted gas production rate and bottom hole pressure for each time point within a preset future time period. These predicted values ​​together constitute the production dynamic prediction results.

[0154] Step S153: Set production dynamic prediction thresholds. The production dynamic prediction thresholds include the gas production rate prediction threshold range and the bottom hole pressure prediction threshold range. The gas production rate prediction threshold range includes an upper limit threshold and a lower limit threshold, and the bottom hole pressure prediction threshold range includes an upper limit threshold and a lower limit threshold.

[0155] Based on the normal production standards of coalbed methane wells, equipment safety operating parameters, and production target requirements, dynamic production prediction thresholds are set. The gas production rate prediction threshold range consists of an upper and a lower threshold. The upper threshold is the maximum permissible gas production rate under normal production conditions, and the lower threshold is the minimum gas production rate value to ensure production efficiency. Similarly, the bottom hole pressure prediction threshold range also includes an upper and a lower threshold. The upper threshold is the maximum permissible bottom hole pressure to prevent equipment damage, and the lower threshold is the minimum bottom hole pressure value to ensure gas production efficiency. These threshold ranges need to be determined in conjunction with field production experience and equipment technical parameters.

[0156] Step S154: Compare the predicted gas production rate in the production dynamic prediction results with the predicted gas production rate threshold range. If the predicted gas production rate is lower than the lower threshold or higher than the upper threshold, it is marked as an abnormal gas production rate. Compare the predicted bottom hole pressure in the production dynamic prediction results with the predicted bottom hole pressure threshold range. If the predicted bottom hole pressure is lower than the lower threshold or higher than the upper threshold, it is marked as an abnormal bottom hole pressure.

[0157] The predicted gas production rate at each moment in the production dynamics forecast is compared with the predicted gas production rate threshold range one by one. If the predicted gas production rate at a certain moment is lower than the lower threshold or higher than the upper threshold, the gas production rate at that moment is marked as abnormal; if it is within the normal range, it is marked as normal.

[0158] Using the same method, the predicted bottomhole pressure value at each moment is compared with the predicted bottomhole pressure threshold range. If it exceeds the range, it is marked as abnormal bottomhole pressure; otherwise, it is marked as normal. The number of all abnormal moments and their corresponding predicted values ​​are counted to identify the abnormal situations.

[0159] Step S155: For the effective deviation transmission closed-loop path with the initial deviation type being geological condition deviation, if there is an abnormal gas production rate, query the preset geological-gas production control rule library and extract the production parameter adjustment direction corresponding to the geological condition deviation and the abnormal gas production rate type; if there is an abnormal bottom hole pressure, extract the production parameter adjustment direction corresponding to the geological condition deviation and the abnormal bottom hole pressure type; for the effective deviation transmission closed-loop path with the initial deviation type being environmental factor deviation, if there is an abnormal gas production rate, query the preset control rule library and extract the production parameter adjustment direction corresponding to the environmental factor deviation and the abnormal gas production rate type; if there is an abnormal bottom hole pressure, extract the production parameter adjustment direction corresponding to the environmental factor deviation and the abnormal bottom hole pressure type.

[0160] A pre-defined geological-gas production control rule base and an environmental-production control rule base are established. The rule base stores the correspondence between different initial deviation types, different anomaly types and production parameter adjustment directions.

[0161] For the effective deviation transmission closed-loop path with the initial deviation type being geological condition deviation, if there is an abnormal gas production rate, the production parameter adjustment direction (such as adjusting the extraction pressure) corresponding to the geological condition deviation (such as low coal seam permeability) and the abnormal gas production rate type (such as below the lower limit) is searched in the geological-gas production control rule base; if there is an abnormal bottom hole pressure, the corresponding bottom hole pressure abnormality adjustment direction (such as adjusting the water injection volume) is searched.

[0162] For effective deviation transmission closed-loop paths where the initial deviation type is environmental factor deviation, if there is an abnormal gas production rate or an abnormal bottom hole pressure, the corresponding production parameter adjustment direction is found in the environment-production control rule base. For example, if the abnormal gas production rate is caused by excessively high surface temperature, the cooling system parameters are adjusted.

[0163] Step S156: Determine the adjustment priority coefficient based on the positive cumulative correlation strength of the effective deviation transmission closed-loop path. The higher the positive cumulative correlation strength, the larger the adjustment priority coefficient.

[0164] The strength of the positive cumulative correlation along the effective deviation transmission closed-loop path is used as the core basis for adjusting the priority coefficient. The strength of the positive cumulative correlation reflects the tightness and impact of deviation transmission; the higher the strength, the greater the impact of the deviation on production dynamics, requiring priority handling. Therefore, the higher the strength of the positive cumulative correlation, the larger the corresponding adjustment priority coefficient should be; conversely, the lower the strength, the smaller the coefficient. The adjustment priority coefficient needs to be normalized to a value between 0 and 1.

[0165] Step S157: Determine the abnormality impact coefficient based on the degree of abnormality in production dynamics. The degree of abnormality in production dynamics is the absolute value of the difference between the predicted gas production rate and the threshold, and the absolute value of the difference between the predicted bottom hole pressure and the threshold. The higher the degree of abnormality, the greater the abnormality impact coefficient.

[0166] The degree of anomaly in production dynamics is calculated as follows: for anomalies in gas production rate, the degree of anomaly is the absolute value of the difference between the predicted gas production rate and the corresponding threshold (lower or upper limit); for anomalies in bottom hole pressure, the degree of anomaly is the absolute value of the difference between the predicted bottom hole pressure and the corresponding threshold. The larger the absolute value of the difference, the more severe the anomaly and the greater its impact on production.

[0167] The anomaly impact coefficient is determined based on the degree of anomaly; the higher the degree of anomaly, the larger the anomaly impact coefficient. The anomaly impact coefficient also needs to be normalized to a value between 0 and 1.

[0168] Step S158: Normalize the adjustment priority coefficient and the abnormal impact coefficient respectively, and add the normalized adjustment priority coefficient and the normalized abnormal impact coefficient to obtain the comprehensive priority score of each production parameter adjustment direction.

[0169] If the adjustment priority coefficient and the abnormal impact coefficient are not yet normalized, the min-max normalization method is first used to convert them into values ​​between 0 and 1. Then, the normalized adjustment priority coefficient and the normalized abnormal impact coefficient are added together, and the sum is the comprehensive priority score for each production parameter adjustment direction. The higher the comprehensive priority score, the more priority the adjustment direction needs to be implemented.

[0170] Step S159: Sort the production parameter adjustment directions in descending order of comprehensive priority score, and mark the effective deviation transmission closed-loop path number, correlation strength data and predicted anomaly type corresponding to each production parameter adjustment direction.

[0171] All production parameter adjustment directions are sorted in descending order of comprehensive priority score, with the highest-scoring adjustment direction ranked first and executed first. During the sorting process, the corresponding effective deviation transmission closed-loop path number is marked for each production parameter adjustment direction to trace its source; the positive cumulative correlation strength and negative cumulative correlation strength of the path are marked; and the corresponding predicted anomaly type is also marked, such as gas production rate below the lower limit or bottom hole pressure above the upper limit.

[0172] Step S1510: Integrate the sorted production parameter adjustment direction, annotation information, and adjustment implementation cycle. The adjustment implementation cycle is determined based on the deviation evolution trend. The faster the deviation expands, the shorter the implementation cycle. Generate a dynamic control strategy that includes adjustment order, adjustment object, adjustment direction, implementation cycle, and related basis. The adjustment object is the specific production parameter.

[0173] Based on the deviation evolution trend of the effective deviation transmission closed-loop path in the deviation transmission analysis report, determine the adjustment implementation cycle for each production parameter adjustment direction. If the deviation evolution trend is an expanding trend and the expansion rate is relatively fast, the implementation cycle should be set to a shorter time to quickly curb the expansion of the deviation; if the deviation expansion rate is relatively slow or shows a slowing trend, the implementation cycle can be appropriately extended.

[0174] The sorted production parameters' adjustment directions, corresponding annotation information (path number, correlation strength data, predicted anomaly type), and adjustment implementation cycles are integrated to generate a dynamic control strategy according to a preset format. This dynamic control strategy clearly defines the adjustment order (sorted by comprehensive priority score), the adjustment object (such as specific production parameters like extraction pressure, water injection volume, and cooling system parameters), the adjustment direction (such as increasing, decreasing, or maintaining), the implementation cycle (such as immediate execution, execution within 24 hours, or phased execution within 72 hours), and the correlation basis (corresponding effective deviation transmission closed-loop path information and production dynamic prediction anomaly information). For example, if a production parameter's adjustment direction has the highest comprehensive priority score, its adjustment object is extraction pressure, its adjustment direction is increasing, its implementation cycle is immediate execution, and its correlation basis is the positive cumulative correlation strength of a certain effective deviation transmission closed-loop path and the anomaly information that the predicted gas production rate is below the lower limit; all of these elements must be clearly reflected in the dynamic control strategy.

[0175] For example, step S1511: convert the dynamic control strategy into an instruction format that can be recognized by the production control terminal, wherein the instruction format includes adjustment parameter identifier, adjustment direction code, and adjustment implementation time node.

[0176] Based on the communication protocol and instruction specifications of the production control terminal, all elements of the dynamic control strategy are converted into an instruction format recognizable by the terminal. Specifically, the adjustment parameter identifier is a unique code for the specific production parameter (such as extraction pressure or injection volume); the adjustment direction code is a preset numeric or character code corresponding to adjustment directions such as increase, decrease, and hold; and the adjustment implementation time node is the specific execution time, such as a specific timestamp for immediate execution or the start and end timestamps for phased execution. The instruction format must meet the parsing requirements of the production control terminal to ensure accurate recognition of the instructions.

[0177] Step S1512: The converted dynamic control strategy instruction is sent to the production control terminal. After receiving the instruction, the production control terminal performs the production parameter adjustment operation according to the adjustment implementation time node in the instruction.

[0178] The converted dynamic control strategy commands are sent to the production control terminal via industrial Ethernet or dedicated communication lines. Upon receiving the commands, the production control terminal first verifies their completeness and validity. If the verification is successful, it retrieves the internal parameter adjustment module and executes the adjustment operation on the corresponding production parameters according to the adjustment implementation time nodes specified in the command. For example, for an immediate command to increase the extraction pressure, the terminal immediately sends a control signal to the extraction pressure regulating device, which then changes the extraction pressure value accordingly. For a phased command to adjust the water injection volume, the terminal sends adjustment signals sequentially according to the preset phase time, gradually completing the water injection volume adjustment.

[0179] Step S1513: After the production parameter adjustment operation is executed, collect the production dynamic data sequence, geological condition data sequence, and environmental data sequence of the coalbed methane well after regulation according to the preset data acquisition frequency.

[0180] A preset data acquisition frequency is set, which is consistent with the data acquisition frequency before the adjustment to ensure data comparability. After the production parameter adjustment operation is executed, the data acquisition device is started to continuously collect the adjusted production dynamic data (gas production rate, bottom hole pressure), geological condition data (coal seam permeability, porosity), and environmental data (surface temperature, humidity) according to the acquisition frequency, forming the adjusted production dynamic data sequence, the adjusted geological condition data sequence, and the adjusted environmental data sequence. All data are accompanied by corresponding timestamps.

[0181] Step S1514: Validate the production dynamics data sequence, geological condition data sequence, and environmental data sequence after regulation, and retain the valid regulated data sequences.

[0182] Using the same time-series continuity detection method as in step S121, the validity of the three types of data sequences after regulation is checked. The timestamps in the data sequences are traversed, and it is checked whether the interval between adjacent timestamps conforms to the preset acquisition frequency and whether the data values ​​are within a reasonable range (e.g., gas production rate is not less than 0, and pressure does not exceed the equipment's maximum withstand value). Data segments that are continuous and have reasonable data values ​​are determined to be valid regulated data sequences; data segments with abnormal timestamps, missing data, or abnormal data values ​​are discarded, retaining only the valid parts.

[0183] Step S1515: Divide the effectively regulated data sequence into new production dynamic time series segments, new geological condition dynamic time series segments, and new environmental dynamic time series segments according to the dynamic time interval division rules.

[0184] The dynamic time interval division rules determined in step S122 are applied to divide the production dynamic data sequence, geological condition data sequence, and environmental data sequence after effective regulation into time series segments. The division process is the same as before regulation: the time interval is determined based on the frequency of data fluctuations; smaller intervals are used for high fluctuation frequencies, and larger intervals are used for low fluctuation frequencies. After division, new production dynamic time series segments, new geological condition dynamic time series segments, and new environmental dynamic time series segments are formed. Each new segment contains complete data within the corresponding time period after regulation.

[0185] Step S1516: Extract the features of the new production dynamic time series segment, the features of the new geological condition dynamic time series segment, and the features of the new environmental dynamic time series segment. The features of the new production dynamic time series segment include the gas production rate change trend and the bottom hole pressure change trend. Recalculate and correct the correlation strength between the new production dynamic time series segment and the original production dynamic time series segment, the correlation strength between the new geological condition dynamic time series segment and the original geological condition dynamic time series segment, and the correlation strength between the new environmental dynamic time series segment and the original environmental dynamic time series segment.

[0186] Using the feature extraction method in step S123, the following features are extracted: the gas production rate change trend and the bottom hole pressure change trend of the new production dynamic time series segment; the coal seam permeability change trend and the coal seam porosity change trend of the new geological condition dynamic time series segment; and the surface temperature change trend and the surface humidity change trend of the new environmental dynamic time series segment.

[0187] According to the correlation strength calculation method in step S124 and the correction method in step S125, the correlation strength between the new production dynamic time series segment and the original production dynamic time series segment (the segment divided before regulation), the correlation strength between the new geological condition dynamic time series segment and the original geological condition dynamic time series segment, and the correlation strength between the new environmental dynamic time series segment and the original environmental dynamic time series segment are calculated respectively. The time series correlation strength correction coefficient is introduced for correction to obtain the corrected new correlation strength.

[0188] Step S1517: Based on the new production dynamic time series segment, the new geological condition dynamic time series segment, the new environmental dynamic time series segment, and the corrected correlation strength, update the node set and the connection weights between nodes of the ternary time series dynamic correlation network. The node set of the ternary time series dynamic correlation network is updated with the nodes corresponding to the new production dynamic time series segment, the new geological condition dynamic time series segment, and the new environmental dynamic time series segment.

[0189] In the node set of the ternary time-series dynamic correlation network, new production dynamic time-series segments, new geological condition dynamic time-series segments, and new environmental dynamic time-series segments are added, and unique identifiers are assigned to the new nodes. Based on the revised new correlation strength, connections are established between the new nodes and existing nodes, with the new correlation strength used as the connection weight. Simultaneously, for correlations existing between new nodes (such as between new production dynamic nodes and new geological condition nodes), their correlation strength is calculated, corrected, and connections and weights are established. Furthermore, based on the correlation status of new nodes, the weights of the connections between existing nodes are updated (if the original correlation strength changes due to regulatory influences), completing the update of the entire network node set and connection weights.

[0190] Step S1518: Simultaneously, store the adjusted production dynamic data sequence, adjusted geological condition data sequence, adjusted environmental data sequence, and ternary time series dynamic correlation network update records to the historical database.

[0191] The three types of data sequences collected after adjustment, along with the update records of the ternary time-series dynamic correlation network (including information on newly added nodes, newly added connections, weight values ​​before and after updates, and update times), are transmitted to a historical database for storage. The historical database employs a distributed storage architecture to ensure data storage security and scalability.

[0192] Step S1519: Generate a network update report and output the network update report and the execution feedback information of the dynamic control strategy to the monitoring terminal. The network update report includes the number of new nodes, the weight update ratio, and the statistical information of the updated correlation strength. The execution feedback information of the dynamic control strategy includes whether the adjustment is completed and the data collection status after the adjustment.

[0193] The update information of the ternary time-series dynamic correlation network is integrated to generate a network update report. The report includes the number of newly added production dynamics, geological conditions, and environmental nodes in this update, the proportion of weight updates (the proportion of updated connections to the total number of connections), and statistical information on the strength of various correlation types after the update (such as average, maximum, minimum, and distribution).

[0194] Simultaneously, dynamic control strategy execution information is collected from the production control terminal, including whether various adjustment operations were completed as planned, whether any abnormalities occurred during the adjustment process, and whether data collection was normal after the control, forming execution feedback information. The network update report and execution feedback information are merged and sent to the monitoring terminal via a data transmission protocol. The monitoring terminal receives and displays the information, allowing maintenance personnel to monitor the control execution status and network update status in real time.

[0195] Based on the same inventive concept, please refer to Figure 2 The diagram shows a schematic block diagram of the structure of a coalbed methane well production dynamic analysis system 100 based on AI time-series prediction provided in an embodiment of this application. The coalbed methane well production dynamic analysis system 100 based on AI time-series prediction may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0196] In this embodiment, both the machine-readable storage medium 120 and the processor 130 are located within the AI-based time-series prediction-based coalbed methane well production dynamic analysis system 100 and are separately configured. However, it should be understood that the machine-readable storage medium 120 may also be independent of the AI-based time-series prediction-based coalbed methane well production dynamic analysis system 100 and may be accessed by the processor 130 via a bus interface. Alternatively, the machine-readable storage medium 120 may be integrated into the processor 130 and may communicate with external systems via the communication unit 110.

[0197] The processor 130 is the control center of the AI-based time-series prediction-based coalbed methane well production dynamic analysis system 100. It connects various parts of the system via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in the machine-readable storage medium 120, and by calling data stored in the machine-readable storage medium 120, thereby providing overall monitoring of the system. Optionally, the processor 130 may include one or more processing cores; for example, the processor 130 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor. The machine-readable storage medium 120 is used to store machine-executable instructions for executing the scheme of this application, and the processor 130 is used to execute the machine-executable instructions stored in the machine-readable storage medium 120 to implement the inspection video stream processing method provided in the aforementioned method embodiments.

[0198] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for dynamic analysis of coalbed methane well production based on AI time-series prediction, characterized in that, The method includes: The production dynamic data sequence of the coalbed methane well, the geological condition data sequence of the corresponding well location, and the surrounding environmental data sequence are collected. The production dynamic data sequence includes information on changes in gas production rate and bottom hole pressure. The geological condition data sequence includes information on changes in coal seam permeability and coal seam porosity. The environmental data sequence includes information on changes in surface temperature and surface humidity. Based on the production dynamic data sequence, the geological condition data sequence, and the environmental data sequence, a ternary time-series dynamic correlation network for coalbed methane wells is constructed. Each node in the ternary time-series dynamic correlation network corresponds to a dynamic time-series segment of a single type of data sequence. The connections between nodes correspond to the real-time correlation relationship of dynamic time-series segments of different types of data sequences, and the connection weights evolve and update over time. The pre-trained AI temporal coupling deviation evolution mining model is invoked to process the ternary temporal dynamic correlation network, mining the coupling deviation evolution information between dynamic temporal segments of different data sequences, and obtaining the ternary temporal coupling deviation result containing the deviation evolution trend. Based on the ternary time-series coupling deviation results, combined with forward transmission location and reverse tracing verification, the deviation transmission closed-loop path affecting production dynamics is determined, and a deviation transmission analysis report containing the deviation transmission direction, correlation strength and tracing verification results is generated. Based on the deviation transmission analysis report and combined with the production dynamic prediction threshold, a dynamic control strategy for coalbed methane well production parameters is generated and output to the production control terminal. At the same time, the controlled data is fed back to the ternary time-series dynamic correlation network to update the node correlation relationship.

2. The method for dynamic analysis of coalbed methane well production based on AI time-series prediction according to claim 1, characterized in that, The construction of a ternary time-series dynamic correlation network for coalbed methane wells based on the production dynamic data sequence, the geological condition data sequence, and the environmental data sequence includes: The production dynamic data sequence, the geological condition data sequence, and the environmental data sequence are subjected to time sequence continuity detection, and continuous data segments are retained as valid data sequences. Based on the frequency of change and fluctuation of the effective data sequence, a dynamic time interval division rule is determined. According to this dynamic time interval division rule, the effective production dynamic data sequence is divided into multiple production dynamic time series segments, the effective geological condition data sequence is divided into multiple geological condition dynamic time series segments, and the effective environmental data sequence is divided into multiple environmental dynamic time series segments, resulting in a set of production dynamic time series segments, a set of geological condition dynamic time series segments, and a set of environmental dynamic time series segments. The higher the frequency of change and fluctuation, the smaller the time interval. Extract the gas production rate and bottom hole pressure trends for each dynamic time segment of production; extract the coal seam permeability and coal seam porosity trends for each dynamic time segment of geological conditions; and extract the surface temperature and surface humidity trends for each dynamic time segment of environmental conditions. Calculate the real-time correlation strength between the gas production rate change trend feature and the coal seam permeability change trend feature; calculate the real-time correlation strength between the gas production rate change trend feature and the surface temperature change trend feature; calculate the real-time correlation strength between the bottom hole pressure change trend feature and the coal seam porosity change trend feature; calculate the real-time correlation strength between the bottom hole pressure change trend feature and the surface humidity change trend feature; calculate the real-time correlation strength between the coal seam permeability change trend feature and the surface temperature change trend feature; calculate the real-time correlation strength between the coal seam porosity change trend feature and the surface humidity change trend feature. A time-series correlation strength correction coefficient is introduced, which is determined based on the correlation strength fluctuation range of the same type of data sequence in the same historical period. Each real-time correlation strength is multiplied by the corresponding time-series correlation strength correction coefficient to obtain the corrected correlation strength. Using each production dynamic time series segment, each geological condition dynamic time series segment, and each environmental dynamic time series segment as network nodes, and using the corrected correlation strength as the initial weight of the connection between nodes, an initial ternary time series dynamic correlation network is constructed. Set a weight update cycle, recalculate and correct the real-time association strength between nodes according to the weight update cycle, update the connection weights of the initial ternary time-series dynamic association network, and obtain the final ternary time-series dynamic association network of coalbed methane wells.

3. The method for dynamic analysis of coalbed methane well production based on AI time-series prediction according to claim 2, characterized in that, The calculation of the real-time correlation strength between the gas production rate change trend feature and the coal seam permeability change trend feature includes: Obtain the first dynamic time-series vector corresponding to the gas production rate change trend feature, where each element of the first dynamic time-series vector corresponds to the gas production rate value at different times within the dynamic time-series segment of production dynamics. Obtain the second dynamic time series vector corresponding to the coal seam permeability change trend characteristics, where each element of the second dynamic time series vector corresponds to the coal seam permeability value at different times within the dynamic time series segment of geological conditions; The first dynamic time series vector and the second dynamic time series vector are input into the time series correlation analysis module. Based on the dynamic time warping algorithm, the first dynamic time series vector and the second dynamic time series vector are elastically aligned in the time dimension to obtain the aligned first dynamic time series vector and the aligned second dynamic time series vector. Extract the rising inflection point and falling inflection point of the aligned first dynamic time series vector; extract the rising inflection point and falling inflection point of the aligned second dynamic time series vector. The number of times the rising inflection point of the first dynamic time series vector coincides with the rising inflection point of the second dynamic time series vector is counted. The number of times the falling inflection point of the first dynamic time series vector coincides with the falling inflection point of the second dynamic time series vector is counted. The two coincidence counts are added together to obtain the total number of coincidence inflection points. The total number of inflection points of the first dynamic time series vector is counted. The total number of inflection points of the first dynamic time series vector is the sum of the number of rising inflection points and the number of falling inflection points. The total number of inflection points of the second dynamic time series vector is counted. The total number of inflection points of the second dynamic time series vector is the sum of the number of rising inflection points and the number of falling inflection points. The two total inflection point counts are added together to obtain the total number of inflection points. Calculate the proportion of the total number of overlapping inflection points to the total number of inflection points, and use this as the inflection point synergy ratio. Calculate the mean change amplitude of the first dynamic time-series vector after alignment, where the mean change amplitude is the average of the absolute values ​​of the differences between each time step and the previous time step. Calculate the mean change amplitude of the second dynamic time-series vector after alignment, where the mean change amplitude is the average of the absolute values ​​of the differences between each time step and the previous time step. Use the ratio of the two mean change amplitudes as the amplitude coordination coefficient. The inflection point coordination ratio and the amplitude coordination coefficient are normalized to convert them into dimensionless scores. Weights for the inflection point coordination ratio and the amplitude coordination coefficient are set, with the sum of the weights being 1. The normalized inflection point coordination ratio is multiplied by the inflection point coordination ratio weight, and then the normalized amplitude coordination coefficient is multiplied by the amplitude coordination coefficient weight to obtain the real-time correlation strength between the gas production rate change trend and the coal seam permeability change trend.

4. The method for dynamic analysis of coalbed methane well production based on AI time-series prediction according to claim 1, characterized in that, The pre-trained AI temporal coupling deviation evolution mining model is invoked to process the ternary temporal dynamic correlation network, mining the coupling deviation evolution information between dynamic temporal segments of different data sequences, and obtaining ternary temporal coupling deviation results containing deviation evolution trends, including: The ternary time-series dynamic correlation network is input into the network parsing layer of the AI ​​time-series coupling deviation evolution mining model to parse the dynamic time-series segment data corresponding to each node, the dynamic time-series segment data corresponding to each node includes the specific data values ​​of each time within the segment, and the corrected correlation strength data and weight update records corresponding to the connection between nodes are parsed to obtain the network parsing data. The network parsing data is input into the temporal coupling analysis layer of the AI ​​temporal coupling deviation evolution mining model. The coupling matching degree between the dynamic time series segments of production dynamics and the dynamic time series segments of geological conditions is analyzed from three dimensions: trend consistency, phase difference synergy, and amplitude synergy. Among them, trend consistency is determined by calculating the proportion of the matching duration of the change trends of the two types of dynamic time series segments; phase difference synergy is determined by calculating the average time difference of the inflection points of the two types of dynamic time series segments; and amplitude synergy is determined by calculating the fluctuation range of the ratio of the change amplitudes of the two types of dynamic time series segments. The coupling matching degree of the three dimensions is normalized respectively. A weight is set for the normalized coupling matching degree of each dimension. The weighted sum of the coupling matching degree scores of the three dimensions is calculated according to the set weights, which is used as the comprehensive coupling matching degree of the two types of dynamic time segments. Identify dynamic time-series segment combinations whose overall coupling matching degree is lower than a preset matching threshold, and mark the dynamic time-series segment combinations as deviation candidate combinations; For each candidate combination of deviations, extract the change data of its comprehensive coupling matching degree in multiple consecutive weight update cycles, and construct a deviation evolution curve. The horizontal axis of the deviation evolution curve is the weight update cycle, and the vertical axis of the deviation evolution curve is the comprehensive coupling matching degree. Analyzing the slope trend of the deviation evolution curve, if the slope is negative and the absolute value gradually increases, the judgment deviation shows an increasing trend; if the slope is negative and the absolute value gradually decreases, the judgment deviation shows a slowing trend; if the slope is positive, the judgment deviation shows a corrective trend. The degree of deviation is calculated for the dynamic time segments in the candidate deviation combinations to obtain the deviation value corresponding to each candidate deviation combination. By integrating candidate combinations of deviations, deviation values, and deviation evolution trends, a ternary time-series coupled deviation result is generated, which includes deviation location, deviation type, deviation degree, and deviation evolution trend. The deviation location corresponds to the dynamic time-series segment number, the deviation type includes production-geological deviation, production-environment deviation, and geological-environment deviation, and the deviation degree is the deviation value.

5. The method for dynamic analysis of coalbed methane well production based on AI time-series prediction according to claim 4, characterized in that, The step of calculating the degree of deviation for the dynamic time segments in the candidate deviation combinations to obtain the deviation value corresponding to each candidate deviation combination includes: Collect production dynamic data sequences, geological condition data sequences, and environmental data sequences of coalbed methane wells under historical normal production conditions, extract coupling data of similar dynamic time series segments, and construct a standard coupling relationship library. The standard coupling relationship library stores the standard trend consistency range, standard phase difference coordination range, and standard amplitude coordination range of different types of dynamic time series segment combinations. Based on the type of deviation candidate combination, the corresponding standard trend consistency range, standard phase difference synergy range, and standard amplitude synergy range are retrieved from the standard coupling relationship library as the standard coupling relationship for the deviation candidate combination. The types of deviation candidate combinations include production-geological deviation, production-environmental deviation, and geological-environmental deviation. Extract the actual change curve of the first type of dynamic time series segment in the candidate combination of deviations, and extract the actual change curve of the second type of dynamic time series segment in the candidate combination of deviations. The horizontal axis of the actual change curve is the time within the segment, and the vertical axis of the actual change curve is the data value. Based on the standard trend consistency range in the standard coupling relationship, a standard trend curve range corresponding to the first type of dynamic time series segment is generated, and a standard trend curve range corresponding to the second type of dynamic time series segment is generated. The standard trend curve range includes an upper trend curve and a lower trend curve. Calculate the area of ​​difference between the actual change curve of the first type of dynamic time series segment and the upper limit of the standard trend curve range, calculate the area of ​​difference between the actual change curve of the first type of dynamic time series segment and the lower limit of the standard trend curve range, and take the maximum value of the two difference areas as the trend deviation area of ​​the first type of segment. Calculate the area of ​​difference between the actual change curve of the second type of dynamic time series segment and the upper limit of the standard trend curve range, calculate the area of ​​difference between the actual change curve of the second type of dynamic time series segment and the lower limit of the standard trend curve range, and take the maximum value of the two difference areas as the trend deviation area of ​​the second type of segment. Based on the standard phase difference coordination range in the standard coupling relationship, calculate the absolute value of the difference between the actual phase difference and the upper limit of the standard phase difference coordination range of the two types of dynamic time segments in the deviation candidate combination. The actual phase difference is the difference at the inflection point. Calculate the absolute value of the difference between the actual phase difference and the lower limit of the standard phase difference coordination range, and take the maximum value of the two absolute values ​​as the phase difference deviation value. Based on the standard amplitude coordination range in the standard coupling relationship, calculate the absolute value of the difference between the actual amplitude ratio of the two types of dynamic time segments in the deviation candidate combination and the upper limit of the standard amplitude coordination range. The actual amplitude ratio is the ratio of the average amplitude change. Calculate the absolute value of the difference between the actual amplitude ratio and the lower limit of the standard amplitude coordination range. Take the maximum value of the two absolute values ​​of difference as the amplitude coordination deviation value. Set the trend deviation weight, phase difference deviation weight, and amplitude coordination deviation weight, and the sum of the trend deviation weight, phase difference deviation weight, and amplitude coordination deviation weight is 1; The average value of the trend deviation area of ​​the first type of segment, the average value of the trend deviation area of ​​the second type of segment, the phase difference deviation value, and the amplitude coordination deviation value are normalized respectively. The normalized trend deviation area score is multiplied by the trend deviation weight, plus the normalized phase difference deviation score multiplied by the phase difference deviation weight, plus the normalized amplitude coordination deviation score multiplied by the amplitude coordination deviation weight, to obtain the deviation value corresponding to the deviation candidate combination.

6. The method for dynamic analysis of coalbed methane well production based on AI time-series prediction according to claim 1, characterized in that, Based on the ternary time-series coupling deviation results, combined with forward propagation location and reverse tracing verification, the deviation propagation closed-loop path affecting production dynamics is determined, and a deviation propagation analysis report containing the deviation propagation direction, correlation strength, and tracing verification results is generated, including: Extract all candidate combinations of deviations and their corresponding deviation values ​​and deviation evolution trends from the ternary time-series coupling deviation results. Select candidate combinations of deviations with an expanding trend of deviation evolution and sort the selected candidate combinations of deviations in descending order of deviation value. Select the top-ranked candidate deviation combinations as key deviation combinations, extract the timestamp information of the two types of dynamic time series segments in each key deviation combination, and determine the temporal order of the two types of dynamic time series segments. Based on the chronological order, the first dynamic timing segment is marked as the initial dynamic timing segment, and the later dynamic timing segment is marked as the affected dynamic timing segment, thus initially determining the positive propagation direction of the deviation from the initial dynamic timing segment to the affected dynamic timing segment. The connection strength between the node corresponding to the starting dynamic time segment and the node corresponding to the affected dynamic time segment is extracted from the ternary time series dynamic correlation network. At the same time, the connection strength between the node corresponding to the starting dynamic time segment and other related nodes is extracted. Other related nodes are nodes that are associated with the affected dynamic time segment. The connection strength between the node corresponding to the affected dynamic time segment and other related nodes is extracted. Construct a forward propagation strength matrix. The rows of the forward propagation strength matrix represent the starting node and related nodes, the columns represent the affected node and related nodes, and the elements of the forward propagation strength matrix represent the correlation strength between the corresponding nodes. Calculate the cumulative association strength of the forward propagation path, and filter the paths with a cumulative association strength higher than the forward propagation threshold as candidate forward propagation paths. The cumulative association strength is the average of the association strengths between all nodes on the path. For each candidate forward propagation path, reverse tracing verification is initiated: starting from the affected dynamic time segment and ending from the initial dynamic time segment, the reverse association path from the affected node to the starting node in the ternary time-series dynamic association network is queried, and the cumulative association strength of the reverse association path is calculated. The cumulative association strength is the average value of the association strength between all nodes on the path. If the cumulative association strength of the reverse association path is higher than the reverse tracing threshold, and the reverse association path and the candidate forward propagation path form a closed loop structure with interconnected nodes, then the candidate forward propagation path is determined to be an effective deviation propagation closed loop path. If the cumulative association strength of the reverse association path is lower than the reverse tracing threshold, or if a closed loop structure cannot be formed, the candidate forward propagation path will be eliminated. The report analyzes the forward cumulative correlation strength, reverse cumulative correlation strength, number of nodes involved, and dynamic time series segment type of each effective deviation transmission closed-loop path. It integrates the deviation transmission direction, correlation strength, and traceability verification results of the effective deviation transmission closed-loop path to generate a deviation transmission analysis report. The deviation transmission direction is the forward transmission direction, the correlation strength is the forward cumulative correlation strength, and the traceability verification results include the reverse cumulative correlation strength and the closed-loop verification results.

7. The method for dynamic analysis of coalbed methane well production based on AI time-series prediction according to claim 6, characterized in that, The construction of the positive transmission intensity matrix includes: Determine the starting node number corresponding to the initial dynamic timing segment, and determine the affected node number corresponding to the affected dynamic timing segment; From the node association list of the ternary time-series dynamic association network, query all nodes that are connected to the starting node and take them as the starting related nodes; query all nodes that are connected to the affected node and take them as the affected related nodes. Integrate the starting node and related nodes to form a set of matrix row nodes, and arrange the nodes in the set of matrix row nodes in ascending order of node number; Integrate the affected nodes and the related affected nodes to form a set of matrix column nodes, and arrange the nodes in the set of matrix column nodes in ascending order of node number; Construct a blank positive conduction intensity matrix. The number of rows in the blank positive conduction intensity matrix is ​​equal to the number of nodes in the set of row nodes of the matrix, and the number of columns in the blank positive conduction intensity matrix is ​​equal to the number of nodes in the set of column nodes of the matrix. For each node in the set of matrix row nodes, the node is designated as a row node; for each node in the set of matrix column nodes, the node is designated as a column node. The corrected correlation strength between row nodes and column nodes in the ternary time-series dynamic correlation network is queried. If there is a connection between a row node and a column node, the corrected connection strength will be filled into the element positions of the row corresponding to the row node and the column corresponding to the column node in the forward transmission strength matrix. If there is no connection between row nodes and column nodes, the element value is set to 0, and the element positions of the row corresponding to the row node and the column corresponding to the column node in the forward transmission intensity matrix are filled in to obtain the complete forward transmission intensity matrix.

8. The method for dynamic analysis of coalbed methane well production based on AI time-series prediction according to claim 1, characterized in that, The dynamic control strategy for coalbed methane well production parameters is generated based on the deviation transmission analysis report and combined with the production dynamic prediction threshold, including: The effective deviation transmission closed-loop paths in the deviation transmission analysis report are analyzed to determine the initial deviation type corresponding to each effective deviation transmission closed-loop path. The initial deviation types include geological condition deviation and environmental factor deviation. The positive cumulative correlation strength and negative cumulative correlation strength of each effective deviation transmission closed-loop path are extracted. Call the pre-trained production dynamics prediction model, input the current production dynamics data sequence, the current geological conditions data sequence, and the current environmental data sequence, and predict the gas production rate and bottom hole pressure within a preset time period in the future, as the production dynamics prediction result; Set production dynamic prediction thresholds, which include the gas production rate prediction threshold range and the bottom hole pressure prediction threshold range. The gas production rate prediction threshold range includes an upper threshold and a lower threshold, and the bottom hole pressure prediction threshold range includes an upper threshold and a lower threshold. Compare the predicted gas production rate in the production dynamics prediction results with the predicted gas production rate threshold range. If the predicted gas production rate is lower than the lower threshold or higher than the upper threshold, it is marked as an abnormal gas production rate. Compare the predicted bottom hole pressure in the production dynamics prediction results with the predicted bottom hole pressure threshold range. If the predicted bottom hole pressure is lower than the lower threshold or higher than the upper threshold, it is marked as an abnormal bottom hole pressure. For the effective deviation transmission closed-loop path with the initial deviation type being geological condition deviation, if there is an abnormal gas production rate, the preset geological-gas production control rule library is queried to extract the production parameter adjustment direction corresponding to the geological condition deviation and the abnormal gas production rate type; if there is an abnormal bottom hole pressure, the production parameter adjustment direction corresponding to the geological condition deviation and the abnormal bottom hole pressure type is extracted. For the effective deviation transmission closed-loop path with the initial deviation type being environmental factor deviation, if there is an abnormal gas production rate, the preset control rule library is queried to extract the production parameter adjustment direction corresponding to the environmental factor deviation and the abnormal gas production rate type; if there is an abnormal bottom hole pressure, the production parameter adjustment direction corresponding to the environmental factor deviation and the abnormal bottom hole pressure type is extracted. The adjustment priority coefficient is determined based on the positive cumulative correlation strength of the effective deviation transmission closed-loop path. The higher the positive cumulative correlation strength, the larger the adjustment priority coefficient. The abnormality impact coefficient is determined by combining the degree of abnormality in production dynamics. The degree of abnormality in production dynamics is the absolute value of the difference between the predicted gas production rate and the threshold, and the absolute value of the difference between the predicted bottom hole pressure and the threshold. The higher the degree of abnormality, the greater the abnormality impact coefficient. The adjustment priority coefficient and the abnormal impact coefficient are normalized respectively. The normalized adjustment priority coefficient and the normalized abnormal impact coefficient are added together to obtain the comprehensive priority score for each production parameter adjustment direction. The production parameter adjustment directions are sorted in descending order of comprehensive priority score, and the effective deviation transmission closed-loop path number, correlation strength data and predicted anomaly type corresponding to each production parameter adjustment direction are marked. The integrated and sorted production parameters include adjustment direction, annotation information, and adjustment implementation cycle. The adjustment implementation cycle is determined based on the deviation evolution trend. The faster the deviation expands, the shorter the implementation cycle. A dynamic control strategy is generated, which includes adjustment sequence, adjustment object, adjustment direction, implementation cycle, and related basis. The adjustment object is the specific production parameter.

9. The method for dynamic analysis of coalbed methane well production based on AI time-series prediction according to claim 8, characterized in that, The process involves calling a pre-trained production dynamics prediction model, inputting the current production dynamics data sequence, the current geological conditions data sequence, and the current environmental data sequence, and predicting the gas production rate and bottom hole pressure within a preset time period as the production dynamics prediction result, including: Collect historical coalbed methane well production dynamic data sequences, geological condition data sequences, environmental data sequences, and corresponding actual gas production rate data and actual well bottom pressure data within a future preset time period to construct a training dataset for the production dynamic prediction model. The training dataset of the production dynamic prediction model is divided into a training set, a validation set, and a test set. The training set is used for model parameter learning, the validation set is used for model hyperparameter tuning, and the test set is used for model performance evaluation. A network structure for a production dynamic prediction model is constructed, comprising a temporal feature encoding layer, a multi-source feature fusion layer, and a prediction output layer. The temporal feature encoding layer employs a bidirectional long short-term memory network structure to extract temporal features from the input production dynamic data sequence, geological condition data sequence, and environmental data sequence, respectively, and outputs production dynamic temporal encoded features, geological condition temporal encoded features, and environmental temporal encoded features. The multi-source feature fusion layer uses an attention mechanism fusion structure to weight and fuse the production dynamic temporal encoded features, geological condition temporal encoded features, and environmental temporal encoded features. The weights are dynamically adjusted based on the influence of each feature on the production dynamic prediction result, and output multi-source fused features. The prediction output layer employs a fully connected network structure to process the multi-source fused features and output the predicted gas production rate and bottom hole pressure values ​​for each moment within a preset future time period. The production dynamic prediction model is trained using the training set. After each preset training round, the prediction error of the production dynamic prediction model is calculated using the validation set. The prediction error is the mean square error between the predicted and actual values ​​of the gas production rate and the mean square error between the predicted and actual values ​​of the bottom hole pressure. If the prediction error exceeds the preset error threshold, the number of hidden layer nodes in the bidirectional long short-term memory network, the weight calculation method of the attention mechanism, and the number of layers in the fully connected network are adjusted, and the production dynamic prediction model is retrained. If the prediction error is lower than the preset error threshold, the test set is used for final performance verification. If the prediction error of the test set is also lower than the preset error threshold, a pre-trained production dynamic prediction model is obtained. Input the current production dynamic data sequence, the current geological condition data sequence, and the current environmental data sequence into the pre-trained production dynamic prediction model, and extract the current production dynamic time-series coding features, the current geological condition time-series coding features, and the current environmental time-series coding features through the time-series feature coding layer; The current production dynamic time-series coding features, current geological condition time-series coding features, and current environmental time-series coding features are weighted and fused through a multi-source feature fusion layer to obtain the current multi-source fused features. By processing the current multi-source fusion features through the prediction output layer, the predicted gas production rate and bottom hole pressure values ​​at each moment within a preset time period are obtained, which serve as the production dynamic prediction results.

10. A coalbed methane well production dynamic analysis system based on AI time-series prediction, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the AI-based time-series prediction method for coalbed methane well production dynamic analysis as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images