Abnormal data identification method and device for deep coal bed gas
Through the DQN reinforcement learning model combined with LSTM and BP networks, it identifies the abnormal data of deep coalbed methane, solves the problem of abnormal data in deep coalbed methane mining, improves data processing efficiency and reduces costs, and provides guarantees for intelligent operation.
Patent Information
- Application Number
- CN202411089852.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-12-06
AI Technical Summary
During the deep coalbed methane mining process, there are failures in the collection equipment, data storage problems and inaccurate manual entry, resulting in abnormal data in massive data, affecting data processing efficiency and cost control.
The DQN reinforcement learning model is adopted, combined with LSTM and BP networks, abnormal data is identified based on the production data of the well, and the abnormal data identification model is automatically identified and adjusted to improve the recognition efficiency.
It significantly improves the identification efficiency of deep coalbed methane abnormal data, reduces processing costs, and provides a foundation and guarantee for the intelligent operation of coalbed methane fields.
Smart Images

Figure CN119089343A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of coalbed methane physical information processing technology, particularly the technical field related to machine learning in this technology, and specifically to a method and device for identifying abnormal data of deep coalbed methane. Background Art
[0002] Improving the extraction of deep coalbed methane (gas buried at depths exceeding 1,500 meters) is a major concern for relevant technical personnel in the industry. However, deep formations are characterized by strong reservoir sensitivity, high formation stress, and poor permeability, which place higher demands on extraction technology and surface gathering and transportation capabilities. In recent years, with the advancement of deep coalbed methane development technology and the level of extraction, the potential for growth in deep coalbed methane production is considerable. To further increase deep coalbed methane production, data-based and intelligent development are important methods for achieving cost reduction and efficiency improvement in oil and gas field operations. Through various sensors and monitoring equipment, real-time monitoring and data analysis of oil and gas reservoirs are achieved to more accurately understand the reservoir's geological structure, reserve size, and production layer properties. However, during the production process, problems such as failure of acquisition or transmission equipment, damage to data storage servers, and inaccurate manual data entry can lead to the presence of some abnormal data within the massive amount of data. Summary of the Invention
[0003] The present application provides a method and device for identifying abnormal data of deep coalbed methane, the purpose of which is to improve the efficiency of the entire process of processing coalbed methane-related data, reduce processing costs, achieve the purpose of reducing costs and increasing efficiency, and provide a basis and guarantee for the intelligent operation of coalbed methane fields.
[0004] To achieve the above objectives, in a first aspect, this paper provides a method for identifying abnormal data of deep coalbed methane, comprising:
[0005] Determining a production mode of the target well based on pre-acquired production data and production test data of the target well; wherein the production mode is composed of multiple production scenarios and an evolution order of the multiple production scenarios;
[0006] Abnormal data of the target well is identified according to the production pattern and a pre-generated abnormal data identification model corresponding to the production pattern; wherein, the production pattern corresponds one-to-one to the abnormal data identification model; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive the production data and the test production data, and to characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0007] In some embodiments of the present invention, the production data and test production data include: well diameter, dynamic liquid level, daily gas production, daily liquid production, well depth, bottom hole pressure and well casing pressure;
[0008] Determining the production mode of the target well based on pre-acquired production data and production test data of the target well includes:
[0009] determining an average density of produced fluid from the target well according to the dynamic liquid level, the bottom hole pressure, and the casing pressure;
[0010] Determine the average liquid holdup of the target well according to the daily liquid production and the daily gas production;
[0011] Determine the gas superficial velocity of the target well according to the daily gas production and the well diameter;
[0012] determining a well pressure gradient of the target well according to the bottom hole pressure, the well depth, and the well casing pressure;
[0013] The production mode is determined according to the average density of the produced liquid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient.
[0014] In some embodiments of the present invention, determining the production mode according to the average density of the produced fluid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient includes:
[0015] Determining, in chronological order, a plurality of production scenarios corresponding to the target well according to the average density of the produced liquid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient;
[0016] Determine the production mode according to the time sequence corresponding to the production scenario
[0017] In some embodiments of the present invention, the step of generating the abnormal data recognition model includes:
[0018] generating a state of each time step of the abnormal data recognition model according to the average density of the produced fluid, the average liquid holdup, the gas superficial velocity and the well pressure gradient of the target well;
[0019] Determining a reward value of the abnormal data recognition model for the current time step according to the state of each time step and the corresponding real state;
[0020] Perform the following iterative training until the preset training cutoff condition is met:
[0021] Inputting the time series data of the average density of the produced fluid, the time series data of the average liquid holdup, the time series data of the well pressure gradient, and the time series data of the gas phase superficial flow rate into the LSTM network to determine the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial flow rate;
[0022] Inputting the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial velocity into the BP network to output the Q value;
[0023] The abnormal data recognition model of the current training round is adjusted according to the Q value and the reward value.
[0024] In some embodiments of the present invention, the production scenarios include: drainage gas production, production increase, slow production increase, rapid attenuation, slow attenuation, stable production and gas production fluctuations; the production modes include: drainage pressure reduction-production increase-slow attenuation of gas production, drainage pressure reduction-production increase-rapid attenuation-production increase again-slow attenuation, drainage pressure reduction-slow production increase-production increase-stable production-slow attenuation, drainage pressure reduction-slow production increase-stable production-slow attenuation, drainage pressure reduction-slow production increase-production increase-rapid attenuation and drainage pressure reduction-slow production increase-stable production-slow attenuation.
[0025] In some embodiments of the present invention, the abnormal data includes acquisition abnormality, abnormal working condition change, and no abnormality.
[0026] In a second aspect, the present application provides a device for identifying abnormal data of deep coalbed methane, the device comprising:
[0027] A production mode determination module is configured to determine a production mode of a target well based on pre-acquired production data and production test data of the target well; wherein the production mode is composed of a plurality of production scenarios and an evolution order of the plurality of production scenarios;
[0028] An abnormal data identification module is used to identify abnormal data of the target well based on the production pattern and a pre-generated abnormal data identification model corresponding to the production pattern; wherein the production pattern corresponds to the abnormal data identification model one-to-one; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive the production data and test production data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0029] In some embodiments of the present invention, the production data and test production data include: well diameter, dynamic liquid level, daily gas production, daily liquid production, well depth, bottom hole pressure and well casing pressure;
[0030] The production mode determination module includes:
[0031] an average density determining unit, configured to determine an average density of the produced fluid of the target well according to the dynamic liquid level, the bottom hole pressure, and the casing pressure;
[0032] an average liquid holdup determination unit, configured to determine an average liquid holdup of the target well according to the daily liquid production and the daily gas production;
[0033] a gas phase superficial flow rate determination unit, configured to determine the gas phase superficial flow rate of the target well according to the daily gas production and the well diameter;
[0034] a well pressure gradient determining unit, configured to determine the well pressure gradient of the target well according to the bottom hole pressure, the well depth, and the well casing pressure;
[0035] A production mode determination unit is used to determine the production mode according to the average density of the produced liquid, the average liquid holdup, the gas phase superficial velocity and the well pressure gradient.
[0036] In some embodiments of the present invention, the production mode determination unit includes:
[0037] a production scenario determination unit, configured to determine, in chronological order, a plurality of production scenarios corresponding to the target well based on the average density of the produced liquid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient;
[0038] The production mode determination subunit is used to determine the production mode according to the time sequence corresponding to the production scenario.
[0039] In some embodiments of the present invention, a device for identifying abnormal data of deep coalbed methane further includes:
[0040] A recognition model generation module is used to generate the abnormal data recognition model, and the recognition model generation module includes:
[0041] a time-step state generating unit, configured to generate a state of each time step of the abnormal data recognition model according to the average density of the produced fluid, the average liquid holdup, the gas superficial velocity and the well pressure gradient of the target well;
[0042] a reward value determining unit, configured to determine a reward value of the abnormal data recognition model for a current time step according to the state of each time step and the corresponding real state;
[0043] The iterative training unit is used to perform the following iterative training until the preset training cutoff condition is met:
[0044] a change state determining unit, configured to input the time series data of the average density of the produced fluid, the time series data of the average liquid holdup, the time series data of the well pressure gradient, and the time series data of the gas phase superficial flow rate into the LSTM network to determine the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial flow rate;
[0045] A Q value generating unit is configured to input the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial velocity into a BP network to output the Q value;
[0046] A model optimization unit is used to adjust the abnormal data recognition model of the current training round according to the Q value and the reward value.
[0047] In some embodiments of the present invention, the production scenarios include: drainage gas production, production increase, slow production increase, rapid attenuation, slow attenuation, stable production and gas production fluctuations; the production modes include: drainage pressure reduction-production increase-slow attenuation of gas production, drainage pressure reduction-production increase-rapid attenuation-production increase again-slow attenuation, drainage pressure reduction-slow production increase-production increase-stable production-slow attenuation, drainage pressure reduction-slow production increase-stable production-slow attenuation, drainage pressure reduction-slow production increase-production increase-rapid attenuation and drainage pressure reduction-slow production increase-stable production-slow attenuation.
[0048] In some embodiments of the present invention, the abnormal data includes acquisition abnormality, abnormal working condition change, and no abnormality.
[0049] In a third aspect, the present application provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of a method for identifying abnormal data of deep coalbed methane.
[0050] In a fourth aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of a method for identifying abnormal data of deep coalbed methane are implemented.
[0051] In a fifth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for identifying abnormal data of deep coalbed methane.
[0052] From the above description, it can be seen that the embodiment of the present application provides a method and device for identifying abnormal data of deep coalbed methane, and the corresponding method includes: first, determining the production mode of the target well based on the pre-acquired production data and test production data of the target well; wherein, the production mode is composed of multiple production scenarios and the evolution order of multiple production scenarios; then, identifying the abnormal data of the target well according to the production mode and the abnormal data identification model corresponding to the pre-generated production mode; wherein, the production mode corresponds one-to-one to the abnormal data identification model; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive production data and test production data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0053] The corresponding device includes: a production mode determination module, which is used to determine the production mode of the target well based on the pre-acquired production data and test production data of the target well; wherein the production mode is composed of multiple production scenarios and the evolution order of multiple production scenarios; an abnormal data identification module, which is used to identify the abnormal data of the target well based on the production mode and the abnormal data identification model corresponding to the pre-generated production mode; wherein the production mode corresponds to the abnormal data identification model one-to-one; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive production data and test production data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0054] The present invention can improve the efficiency of the entire process of processing coalbed methane-related data, reduce processing costs, achieve the purpose of reducing costs and increasing efficiency, and provide a basis and guarantee for the intelligent operation of deep coalbed methane fields. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0056] Figure 1 A schematic diagram of the process of an abnormal data identification method for deep coalbed methane provided in an embodiment of the present application Figure 1 ;
[0057] Figure 2 This is a flow chart of step 100 of a method for identifying abnormal data of deep coalbed methane provided in an embodiment of the present application;
[0058] Figure 3 This is a flow chart of step 105 of a method for identifying abnormal data of deep coalbed methane provided in an embodiment of the present application;
[0059] Figure 4 A schematic diagram of the process of an abnormal data identification method for deep coalbed methane provided in an embodiment of the present application Figure 2 ;
[0060] Figure 5 This is a flow chart of step 300 of a method for identifying abnormal data of deep coalbed methane provided in an embodiment of the present application;
[0061] Figure 6 This is a schematic diagram of a deep coalbed methane well scenario in a certain production mode provided in an embodiment of the present application;
[0062] Figure 7 A schematic diagram of the classification of production modes of deep coalbed methane wells provided in the embodiments of this application;
[0063] Figure 8 This is a flow chart of a method for identifying abnormal data of deep coalbed methane provided in a specific application example of this application;
[0064] Figure 9 This is a schematic diagram of the production mode identification process of a deep coalbed methane well provided in a specific application example of this application;
[0065] Figure 10 This is a schematic diagram of the Q network input and output structure provided in the specific application example of this application;
[0066] Figure 11 This is a schematic diagram of abnormal data collected from a deep coalbed methane well provided in a specific application example of this application;
[0067] Figure 12 A schematic diagram of a change in production mode of a deep coalbed methane well provided in a specific application example of this application;
[0068] Figure 13 A schematic diagram of the structure of an abnormal data identification device for deep coalbed methane provided in an embodiment of the present application Figure 1 ;
[0069] Figure 14 This is a structural diagram of the production mode determination module 10 provided in an embodiment of the present application;
[0070] Figure 15 This is a schematic structural diagram of a production mode determination unit 10e provided in an embodiment of the present application;
[0071] Figure 16A schematic diagram of the structure of an abnormal data identification device for deep coalbed methane provided in an embodiment of the present application Figure 2 ;
[0072] Figure 17 Schematic diagram of the structure of the recognition model generation module 30 provided in an embodiment of the present application;
[0073] Figure 18 Schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION
[0074] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0075] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
[0076] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products or apparatuses.
[0077] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0078] The embodiment of the present application provides a specific implementation method of a method for identifying abnormal data of deep coalbed methane, see Figure 1 , the method specifically includes the following contents:
[0079] Step 100: Determine a production mode of the target well based on pre-acquired production data and production test data of the target well; wherein the production mode is composed of multiple production scenarios and the evolution order of the multiple production scenarios;
[0080] Step 200: Identify abnormal data of the target well based on the production pattern and a pre-generated abnormal data identification model corresponding to the production pattern; wherein the production pattern corresponds one-to-one to the abnormal data identification model; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive the production data and the production test data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0081] From the above description, it can be seen that the embodiment of the present application provides a method for identifying abnormal data of deep coalbed methane. First, the influencing parameters and corresponding data that can characterize the production and operation status of different deep coalbed methane wells are extracted; the production modes of different deep coalbed methane wells are judged using the extracted influencing parameters; different abnormality recognition models are trained using the drainage data of deep coalbed methane wells with different production modes; the data of wells that need to be identified for abnormality are extracted for production mode judgment to adopt the corresponding abnormality recognition model; the data within a specific time period before the current time point is obtained to judge the abnormal state at the current moment; finally, the data required for the next moment is extracted to continue the execution process to judge whether the deep coalbed methane well is in an abnormal working condition based on the data at each moment.
[0082] The present invention can effectively verify abnormalities in the data in the daily deep coalbed methane production report and reflect the production status of deep coalbed methane wells based on the data in the daily production report. This fundamentally changes the existing practice of requiring manual judgment of the data status of deep coalbed methane wells. In theory, it can determine the production status of each well in a deep coalbed methane well field and verify the historical production status of each well. This solves the problem that the daily production data cannot intuitively reflect the production status of deep coalbed methane wells. It also provides a data foundation for in-depth analysis of the production status of deep coalbed methane wells and provides technical support for intelligent operation and intelligent decision-making in deep coalbed methane well fields.
[0083] In some embodiments of the present invention, the production data and test production data in step 100 include: DCS data and daily data of deep coalbed methane wells, etc. DCS data is the control point data collected by several controllers during the oil and gas production process. Different controllers are connected through a network to collect production data. The collected parameters include wellhead temperature, reinjection pressure, temperature switch, flame probe value, pump voltage and current, etc. Daily data is a supplement to DCS data, and includes stroke, stroke frequency, pump efficiency, dynamic liquid level, daily gas production, daily water production, bottom hole pressure, and casing pressure. In addition, the production data in step 100 has the characteristics of a full life cycle.
[0084] In some embodiments of the present invention, for step 200, since the deep coalbed methane production status is continuous and sequential without external interference, the LSTM network can be applied as part of the Q network in the DQN algorithm. In addition, the Q network needs to input the action at the current time step and the status at the current moment, so it is necessary to design a different Q network in combination with other methods.
[0085] In this application, the Q network is composed of LSTM and BP neural networks, where the role of LSTM is to store the changing patterns of production parameters at the current moment and the previous period, and convert it into the production status at the current moment, and input it into the BP neural network.
[0086] It should be noted that step 200 may also output the Q value of normal data, that is, when the Q value corresponds to the range of normal data, the data output by the BP network is normal data.
[0087] In some embodiments of the present invention, the production data and test production data include: well diameter, dynamic liquid level, daily gas production, daily liquid production, well depth, bottom hole pressure, and well casing pressure;
[0088] In some embodiments of the present invention, in order to eliminate the influence of the drilling depth and well diameter of coalbed methane wells of different depths, these influencing factors are converted into fluid property parameters and gas well production status parameters, including liquid density in the produced fluid, average liquid holdup, gas phase apparent velocity and gas well pressure gradient. Based on this, see Figure 2 , step 100 includes:
[0089] Step 101: determining the average density of the produced fluid of the target well according to the dynamic liquid level, the bottom hole pressure, and the casing pressure;
[0090]
[0091] Step 102: determining an average liquid holdup of the target well according to the daily liquid production and the daily gas production;
[0092]
[0093] Step 103: determining the gas superficial velocity of the target well according to the daily gas production and the well diameter;
[0094]
[0095] Step 104: determining the well pressure gradient of the target well according to the bottom hole pressure, the well depth, and the well casing pressure;
[0096]
[0097] In formulas (1) to (4), ρ l is the average density of liquid phase produced from deep coalbed methane wells, kg / m 3 ;p wf is the bottom hole pressure, Pa; p c is the casing pressure of deep coalbed methane well, Pa; g is the acceleration of gravity, which is taken as 9.81kg / (m·s 2 );h f is the dynamic liquid level depth of deep coalbed methane well, m; H l is the average liquid holdup of the produced fluid in deep coalbed methane wells, V g is the daily gas production of deep coalbed methane wells, m 3 ; V l is the daily liquid production of deep coalbed methane wells, m 3 ;v sg is the gas phase apparent velocity of the deep coalbed methane well, m / s; d is the diameter of the deep coalbed methane well, m, ▽p is the well pressure gradient, Pa / m, and h is the well depth, m.
[0098] Step 105: Determine the production mode according to the average density of the produced fluid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient.
[0099] In some embodiments of the present invention, see Figure 3 , step 105 includes:
[0100] Step 1051: Determine, in chronological order, multiple production scenarios corresponding to the target well based on the average density of the produced liquid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient;
[0101] Step 1052: Determine the production mode according to the time sequence corresponding to the production scenario.
[0102] In steps 1051 and 1052, scenario recognition is performed on the average density of the produced liquid, average liquid holdup, well pressure gradient, and gas superficial velocity of each deep coalbed methane well. The results of the scenario recognition of each deep coalbed methane well, including the scenario type and scenario evolution order, are summarized and analyzed, and the combinations with the most scenario evolution orders of the gas wells are statistically counted as typical production modes. Specifically:
[0103] The extracted influencing factors are used to judge the production modes of different deep coalbed methane wells: feature extraction is performed on the production data and static data in the daily reports of deep coalbed methane wells. The extracted features include well diameter, dynamic liquid level, well depth, bottom hole pressure, casing pressure, daily gas production, and daily water production. The extracted influencing factors are converted into the output liquid density of deep coalbed methane wells, the average liquid holdup of the output fluid, the average gas phase apparent velocity, and the well pressure gradient. These eliminate the influence of different well diameters and well depths, and characterize the production dynamics of deep coalbed methane wells.
[0104] The features extracted from different deep coalbed methane wells are input into the scene recognition model to obtain the production scenarios of each well and the order of their evolution. The scene recognition model is an analytical model that analyzes the production status of deep coalbed methane wells at each time step based on the data characteristics of the wells. Scenarios include increasing production, stable production, and production fluctuations, and are used to evaluate the production and expected production capacity of deep coalbed methane wells.
[0105] The production scenario identification results of existing deep coalbed methane wells, including the scenario types and sequences of each well, were selected. Several scenario types and sequence patterns with the most wells were selected as typical production patterns. Other deep coalbed methane wells with similar production scenarios were grouped into the typical production pattern with the smallest differences. This group of scenario types and sequences were classified as a specific type under this production pattern.
[0106] The current deep coalbed methane well data is input into the scene recognition model, and the production mode to which it belongs is summarized according to the scene recognition results. If the current well scene is the same as that of a typical production mode, the current well is directly classified into this production mode. If the current well scene is different from the scenes of all typical production modes, the most similar typical production mode is found as the production mode of the current well, and the group of scene types and their order are classified as a specific type under this production mode, so as to achieve the effect of continuously updating and improving each typical production mode.
[0107] In some embodiments of the present invention, see Figure 4 , a method for identifying abnormal data of deep coalbed methane also includes:
[0108] Step 300: Generate the abnormal data recognition model, then refer to Figure 5 , step 300 includes:
[0109] Step 301: generating the state of each time step of the abnormal data recognition model according to the average density of the produced fluid, the average liquid holdup, the gas superficial velocity and the well pressure gradient of the target well;
[0110] Step 302: determining a reward value of the abnormal data recognition model for the current time step according to the state of each time step and the corresponding real state;
[0111] Specifically, the average density of produced fluid, average liquid holdup, gas superficial velocity, and well pressure gradient of deep coalbed methane wells every day are used as the state of each time step in DQN, and the correctness of the judgment is used as the reward in DQN, where a correct prediction is +1 point and an incorrect prediction is -1 point.
[0112] Step 303: Perform the following iterative training until a preset training cutoff condition is met:
[0113] Step 304: Inputting the time series data of the average density of the produced fluid, the time series data of the average liquid holdup, the time series data of the well pressure gradient, and the time series data of the gas phase superficial flow rate into the LSTM network to determine the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial flow rate;
[0114] Step 305: Inputting the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial velocity into the BP network to output the Q value;
[0115] Step 306: Adjust the abnormal data recognition model of the current training round according to the Q value and the reward value.
[0116] Furthermore, steps 301 to 306 can be described as follows:
[0117] 1: Randomly select a time step.
[0118] 2: At this time step, based on the current Q network, calculate the corresponding Q values of the data under the three abnormal states, and take the largest Q value as the judgment result.
[0119] 3: Calculate the reward based on the deviation between the judgment result in step 2 and the actual result. If there is no deviation, there will be a reward; if there is a deviation, there will be no reward.
[0120] 4: At the next time step, the current Q network is still used to calculate the maximum Q value among the three judgment results.
[0121] 5: Update the Q network based on the reward in step 3 and the maximum Q value in step 4.
[0122] In steps 303 to 306, the abnormal state of the deep coalbed methane well determined in step 302 is used as the action of the intelligent agent in DQN. The input of the Q network is the current time step state, the state of the previous time steps, and the action determined at the current moment. In this embodiment, the hidden layer of the LSTM part in the Q network is two layers, the number of neurons in the first layer is 16, and the number of neurons in the second layer is 8. The output result is directly input into the BP network layer. The hidden layer of the BP network is one layer, the number of neurons in the first layer is 7, including the 4 state outputs of the LSTM layer and the one-hot input for the judgment of the three abnormal states of the data at the current moment. The number of neurons in the hidden layer is 16, and the output layer is the Q value. The activation function of both networks uses Softplus. The parameters extracted from the daily report generated on the day the deep coalbed methane well was put into production are used as the starting state of the intelligent agent in DQN, and the parameters extracted from the daily report generated on the day before the deep coalbed methane well was shut down are used as the ending state of the intelligent agent in DQN. The goal of DQN is to identify as much abnormal data as possible to obtain more rewards.
[0123] In some embodiments of the present invention, the production scenarios include: drainage gas production, production increase, slow production increase, rapid attenuation, slow attenuation, stable production and gas production fluctuations; the production modes include: drainage pressure reduction-production increase-slow attenuation of gas production, drainage pressure reduction-production increase-rapid attenuation-production increase again-slow attenuation, drainage pressure reduction-slow production increase-production increase-stable production-slow attenuation, drainage pressure reduction-slow production increase-stable production-slow attenuation, drainage pressure reduction-slow production increase-production increase-rapid attenuation and drainage pressure reduction-slow production increase-stable production-slow attenuation.
[0124] Since the order of occurrence of the scenarios of increasing production-stable production and stable production-increasing production is different, the corresponding drainage curve characteristics are also completely different. Therefore, they are regarded as two production modes. The production mode is derived according to the type of production scenario and the evolution order of the production scenario. Figure 6 The type and evolution law of production scenarios under a certain production mode, as well as the relationship with production parameters. In this embodiment, the relationship between the production scenario and production mode of deep coalbed methane wells is as follows: Figure 7 In this embodiment, the production scenarios include: drainage gas production, production increase, slow production increase, rapid decay, slow decay, stable production, and gas production fluctuation, a total of 7 types. The production modes include: drainage and pressure reduction-production increase-gas production slow decay, drainage and pressure reduction-production increase-rapid decay-production again-slow decay, drainage and pressure reduction-slow production increase-production increase-stable production-slow decay, drainage and pressure reduction-slow production increase-stable production-slow decay, drainage and pressure reduction-slow production increase-up production-rapid decay, drainage and pressure reduction-slow production increase-stable production-slow decay. The drainage and production data of the new well are input for scene recognition. If the recognized scene type and sequence are the same as the production scene type and sequence of a certain production mode mentioned above, the new well is classified into the production mode. The production modes of all new wells are classified according to this method.
[0125] In some embodiments of the present invention, the abnormal data includes acquisition abnormality, abnormal working condition change, and no abnormality.
[0126] Specifically, data collection anomaly refers to the situation where a certain feature in the data entry is outlier, but other data items are normal; operating condition change anomaly refers to the situation where all features of the data entry are outliers, but meet the characteristics of other production modes; the rest of the situations are no data anomalies.
[0127] As can be seen from the above description, the embodiment of the present application provides a method for identifying abnormal data of deep coalbed methane. First, the well data that needs to be identified for abnormality is extracted for production scene identification and production mode discrimination, and the corresponding abnormality recognition model is used to identify abnormalities; based on the daily production data, the deep coalbed methane well data that needs to be identified for abnormality is extracted and feature converted to confirm the production mode of the deep coalbed methane well, and the abnormal data recognition model is applied to judge the data status. The data within the current time point and a specific period before is obtained to judge the abnormal state at the current moment. The abnormal state is divided into three types: well data collection abnormality, well operating condition change abnormality, and no state abnormality. When some feature data results are outliers, the production data can be considered to be well data collection abnormality; when the data results of all features are outliers, but meet the production rules of deep coalbed methane wells, the well operating condition change can be considered to be abnormal; the rest of the cases can be considered to be data without abnormalities. Compared with the existing methods, the present invention has the following advantages:
[0128] (1) The recognition efficiency of abnormal data of deep coalbed methane well production is significantly improved. Compared with the previous manual experience judgment method, the present invention has the advantages of automation and high efficiency.
[0129] (2) A production model structure based on production scenarios is proposed for deep coalbed methane wells. Compared with the existing methods for deep coalbed methane drainage data, the present invention is more adaptable to new wells in different production scenarios and has a wider range of applications, which is conducive to unifying the production patterns of new deep coalbed methane wells and existing deep coalbed methane wells.
[0130] (3) Aiming at the characteristics of deep coalbed methane production data, a new Q network structure is proposed. This network uses LSTM, which has the characteristics of time series data memory. Combined with the input time series production data, it analyzes the production status of each production parameter at the current moment and inputs it and the abnormal judgment results into the BP neural network for analysis. Compared with the previous direct calculation of Q value through neural network, this method outputs the current production status of deep coalbed methane wells in the LSTM part and realizes the function of traditional Q network in the BP part, which more clearly reflects the relationship between production status and production parameters.
[0131] To further illustrate this solution, this application also provides a specific application example of a method for identifying abnormal data of deep coalbed methane, see Figure 8 , specifically including the following contents.
[0132] Step S001: Extracting mining and test mining data of deep coalbed methane wells.
[0133] Extract production and test production data of deep coalbed methane wells, including dynamic liquid level, casing pressure, bottom hole pressure, daily gas production and daily water production in daily data; convert the extracted dynamic liquid level, casing pressure, bottom hole pressure, daily gas production and daily water production data of each deep coalbed methane well into average density of produced fluid, liquid holdup of produced fluid, well pressure gradient and gas phase apparent flow rate.
[0134] Specifically, a certain number of deep coalbed methane wells are selected. The daily reports of deep coalbed methane wells include: pump stroke, stroke frequency, frequency, dynamic liquid level, casing pressure, bottom hole pressure, daily gas production, daily water production, and production status. Production parameters representing the production status are extracted from the daily reports, including the dynamic liquid level, daily gas production, daily liquid production, bottom hole pressure, and casing pressure of the deep coalbed methane wells. Based on the extracted parameters, the average density of the produced fluid, the average liquid holdup of the produced fluid, the gas phase superficial velocity of the deep coalbed methane well, and the well pressure gradient of the deep coalbed methane well are calculated according to formulas (1) to (4).
[0135] Step S002: Identify the production mode of the production well.
[0136] The average density of the produced liquid, the liquid holdup of the produced fluid, the well pressure gradient, and the gas phase superficial velocity of each deep coalbed methane well are used for scene recognition. According to the results of scene recognition, the production mode of the deep coalbed methane well is judged and summarized into the aforementioned typical production modes.
[0137] Specifically, the data is converted, and the converted characteristic data of each well is input into the well production scenario recognition model. The production scenarios and production scenario evolution sequence of each deep coalbed methane well are statistically obtained. The process is as follows: Figure 9 shown.
[0138] After summarizing the production modes of several deep coalbed methane wells, the influencing factors, feature conversion, and production scenario identification were also performed on other deep coalbed methane wells. Then, correlation analysis was performed based on the production scenario type and production scenario sequence. Deep coalbed methane wells with higher correlation were classified into the same category. After this processing, all deep coalbed methane wells and their production parameters included in each production mode were obtained. The production and drainage parameters of these gas wells constituted the training set based on the DQN anomaly recognition model.
[0139] For deep coalbed methane wells that require abnormal data identification, their production feature data are extracted and converted for scene identification. If their scene type and combination order meet a typical production mode, they are classified into that production mode; specifically, if their scene type and combination order do not meet any typical production mode, the production mode closest to their scene is also sought for classification. In this way, the production modes of deep coalbed methane wells are divided, and the scene combination method under the production mode is updated, which greatly improves the applicability of each production mode.
[0140] Step S003: Use different models to train wells with different production modes.
[0141] Deep coalbed methane wells with different typical production modes are trained separately using the DQN reinforcement learning model. For deep coalbed methane wells that require abnormal data identification, their production feature data is extracted and converted to perform scene recognition and production mode discrimination. Based on the production mode, the data of the well is input into the corresponding anomaly recognition model for judgment. After the judgment, the next data of the well is extracted.
[0142] Specifically, deep coalbed methane wells are divided according to production modes, and anomaly recognition models based on DQN are established respectively. The Q network structure in this application is as follows: Figure 10 As shown in the figure, the average density of produced liquid, average liquid holdup, gas phase superficial velocity and well pressure gradient of deep coalbed methane wells every day are used as the state of each time step in DQN, and the correctness of the judgment is used as the reward in DQN, where a correct prediction is +1 point and a wrong prediction is -1 point. The abnormal state of the deep coalbed methane well judged is used as the action of the intelligent agent in DQN. The input of the Q network is the current time step state, the state of the previous time steps and the action judged at the current moment. In this embodiment, the hidden layer of the LSTM part in the Q network is two layers, the number of neurons in the first layer is 16, and the number of neurons in the second layer is 8. The output result is directly input into the BP network layer. The hidden layer of the BP network is one layer, and the number of neurons in the first layer is 7, including the 4 state outputs of the LSTM layer and the one-hot input for judging the three abnormal states of the data at the current moment. The number of neurons in the hidden layer is 16, and the output layer is the Q value. The activation function of both networks uses Softplus. The parameters extracted from the daily report generated on the day the deep coalbed methane well was put into production are used as the starting state of the intelligent agent in DQN, and the parameters extracted from the daily report generated on the day before the deep coalbed methane well was shut down are used as the ending state of the intelligent agent in DQN. The goal of DQN is to identify as many abnormal data as possible to obtain more rewards.
[0143] Furthermore, since the production status of deep coalbed methane is continuous and sequential without external interference, the LSTM network can be applied as part of the Q network in the DQN algorithm. In addition, the Q network needs to input the action at the current time step and the status at the current moment, so it is necessary to design a different Q network in combination with other methods. For deep coalbed methane wells under different typical production modes, the data corresponding to the aforementioned influencing factors are extracted to train the DQN reinforcement learning model; wherein in the DQN reinforcement learning model, the Q network consists of two parts: an LSTM part and a BP neural network part. As mentioned above, the Q network in this application consists of an LSTM and a BP neural network, wherein the role of the LSTM is to store the changing patterns of the production parameters at the current moment and the previous period, and convert it into the production status at the current moment, and input it into the BP neural network. Since the input of the LSTM neural network is the time series data of the four influencing parameters, and it can be seen from formulas (1) to (4) that there is a certain correlation between the four parameters, but because the calculation formulas of the four input parameters are different, the four influencing factors input to the LSTM are still independent of each other. Therefore, the number of nodes in the LSTM output part is 4, which represent the change state of the liquid phase density of the produced fluid in the deep coalbed methane well, the change state of the liquid holdup, the change state of the gas phase apparent flow rate, and the change state of the well pressure gradient. The input of the BP neural network in the Q network consists of two parts: the state output of the LSTM neural network and the action judged by the agent at the current time step. The output of the BP neural network is used as the output value of the Q network. Therefore, the input of the Q network is the gas density, average liquid holdup, gas phase apparent flow rate, well pressure gradient, and the result of the judgment in the current and previous period, and the output is the Q value at the current moment. The strategy for calculating the Q value is to use the Q network to calculate all actions in the current state and find its maximum value. The reward for the DQN network is the correctness of the judgment. A correct judgment is a positive reward, and an incorrect judgment is a negative reward. According to the above method, the data is input into DQN for training to obtain the abnormal data recognition model under different production modes.
[0144] The LSTM part receives time series data on the average density of produced liquid, liquid holdup, well pressure gradient, and gas superficial velocity of deep coalbed methane wells, outputs the changing states of these four production characteristics, and inputs them into the BP neural network. This part mainly converts the time series data of deep coalbed methane wells into the changing states of various parameters to represent the production status of the deep coalbed methane wells at the current moment.
[0145] In the BP neural network part, it receives the deep coalbed methane well status output of LSTM and the action of the intelligent agent at that time step, that is, the judgment result of data anomaly, and outputs the Q value at the current time step. This part mainly implements the functions of the previous Q network, that is, input status and action, and output Q value.
[0146] Step S004: extract well data to identify the production mode and select the corresponding anomaly recognition model.
[0147] For deep coalbed methane gas that needs to be identified as abnormal, daily data is extracted, feature conversion is performed, and production mode identification is performed. The identification method has been introduced above and will not be repeated here. Based on the identification results, the corresponding anomaly recognition model is selected.
[0148] Specifically, different anomaly recognition models were trained using deep coalbed methane wells with different production modes. Based on the deep coalbed methane well scenario recognition model, data corresponding to four factors, namely, produced liquid density, average liquid holdup, gas superficial velocity, and well pressure gradient, extracted from each deep coalbed methane well, were input. The results were summarized to classify these wells into typical production modes that correspond to their respective production characteristics. For deep coalbed methane wells with the same production mode, the four parameters, gas density, average liquid holdup, gas superficial velocity, and well pressure gradient, were input into the DQN model. The input data served as the given environment for the DQN, with each day's data corresponding to a state in the DQN. The results of the environmental judgment included three states: abnormal data acquisition for the deep coalbed methane well, abnormal operating condition change, and no abnormality. A data acquisition anomaly refers to a data point where any of the four input parameters are zero or significantly outliers. An operating condition change anomaly refers to a data point where all four input parameters are non-zero but do not conform to the parameter variation pattern for the corresponding production mode.
[0149] Step S005: Obtain data for a period of time before the current time and determine whether the data is abnormal.
[0150] The average density of produced gas, average liquid holdup of produced fluid, gas superficial velocity, and well pressure gradient data of the deep coalbed methane well whose abnormal data is to be identified are input into the production scenario recognition model. According to the type and sequence of scenario recognition, the production mode of the well is confirmed in combination with the deep coalbed methane production mode discrimination method, and the corresponding abnormal data recognition model is called for judgment. The result of the judgment is one of the three abnormal states mentioned above.
[0151] Step S006: When the data collection is in an abnormal state, check the detailed information data in the daily report.
[0152] If the data collection is abnormal, it means that some parameters in the data collected at the current moment are consistent with the current production mode, but there is an outlier in some parameters that does not conform to the production behavior, such as Figure 11 The data corresponding to the negative values of the well pressure gradient and liquid density do not conform to the current production mode. At this time, it is necessary to combine the detailed information data in the daily report to check whether there are estimated values on that day or inaccurate input values.
[0153] Step S007: When the working condition is abnormal, check whether there is any change in the production mode caused by human operations such as pump inspection and production restriction at this moment.
[0154] If an abnormal working condition occurs, it means that all the characteristics of the current production data do not conform to the current production mode, such as Figure 12 For data points where all parameters change significantly, it is necessary to check whether there are any human operations such as pump inspection and production restriction that have caused the production mode to change, or whether there are any fault conditions such as sand injection burial and coal powder pump jam that require maintenance.
[0155] Step S008: extracting the deep coalbed methane well data at the next moment.
[0156] After the abnormal data at the current moment is determined, the data of the deep coalbed methane well at the next moment is input into the abnormality recognition model to continue to determine its abnormal state.
[0157] Based on the same inventive concept, the embodiments of the present application also provide an abnormal data identification device for deep coalbed methane, which can be used to implement the method described in the above embodiments, such as the following embodiments. Since the principle of solving the problem by the abnormal data identification device for deep coalbed methane is similar to that of the abnormal data identification method for deep coalbed methane, the implementation of the abnormal data identification device for deep coalbed methane can refer to the implementation of the abnormal data identification method for deep coalbed methane, and the repeated parts will not be repeated. As used below, the term "unit" or "module" can be a combination of software and / or hardware that implements the predetermined function. Although the system described in the following embodiments is preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and conceived.
[0158] The embodiment of the present application provides a specific implementation of a device for identifying abnormal data of deep coalbed methane that can implement a method for identifying abnormal data of deep coalbed methane. Figure 13 A device for identifying abnormal data of deep coalbed methane specifically includes the following contents:
[0159] The production mode determination module 10 is configured to determine the production mode of the target well based on the pre-acquired production data and production test data of the target well; wherein the production mode is composed of multiple production scenarios and the evolution order of the multiple production scenarios;
[0160] The abnormal data identification module 20 is used to identify the abnormal data of the target well based on the production pattern and a pre-generated abnormal data identification model corresponding to the production pattern; wherein the production pattern corresponds to the abnormal data identification model one-to-one; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive the production data and test production data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0161] In some embodiments of the present invention, the production data and test production data include: well diameter, dynamic liquid level, daily gas production, daily liquid production, well depth, bottom hole pressure and well casing pressure;
[0162] See also Figure 14 , the production mode determination module 10 includes:
[0163] an average density determining unit 10a, configured to determine an average density of the produced fluid of the target well according to the dynamic liquid level, the bottom hole pressure, and the casing pressure;
[0164] an average liquid holdup determining unit 10b, configured to determine an average liquid holdup of the target well according to the daily liquid production and the daily gas production;
[0165] a gas phase superficial flow rate determination unit 10c, configured to determine the gas phase superficial flow rate of the target well according to the daily gas production and the well diameter;
[0166] a well pressure gradient determining unit 10d, configured to determine the well pressure gradient of the target well according to the bottom hole pressure, the well depth, and the well casing pressure;
[0167] The production mode determination unit 10e is configured to determine the production mode according to the average density of the produced liquid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient.
[0168] In some embodiments of the present invention, see Figure 15 , the production mode determination unit 10e includes:
[0169] A production scenario determination unit 10e1 is configured to determine, in chronological order, a plurality of production scenarios corresponding to the target well based on the average density of the produced liquid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient;
[0170] The production mode determination subunit 10e2 is configured to determine the production mode according to the time sequence corresponding to the production scenarios.
[0171] In some embodiments of the present invention, see Figure 16 , a device for identifying abnormal data of deep coalbed methane, further comprising:
[0172] Identification model generation module 30, used to generate the abnormal data identification model, see Figure 17 , the recognition model generation module 30 includes:
[0173] A time-step state generating unit 30a is configured to generate a state of each time step of the abnormal data recognition model according to the average density of the produced fluid, the average liquid holdup, the gas superficial velocity, and the well pressure gradient of the target well;
[0174] A reward value determining unit 30b, configured to determine a reward value of the abnormal data recognition model for the current time step according to the state of each time step and the corresponding real state;
[0175] The iterative training unit 30c is configured to perform the following iterative training until a preset training cutoff condition is met:
[0176] a change state determining unit 30d, configured to input the time series data of the average density of the produced fluid, the time series data of the average liquid holdup, the time series data of the well pressure gradient, and the time series data of the gas phase superficial flow rate into the LSTM network to determine the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial flow rate;
[0177] A Q value generating unit 30e is configured to input the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial velocity into a BP network to output the Q value;
[0178] The model optimization unit 30f is used to adjust the abnormal data recognition model of the current training round according to the Q value and the reward value.
[0179] In some embodiments of the present invention, the production scenarios include: drainage gas production, production increase, slow production increase, rapid attenuation, slow attenuation, stable production and gas production fluctuations; the production modes include: drainage pressure reduction-production increase-slow attenuation of gas production, drainage pressure reduction-production increase-rapid attenuation-production increase again-slow attenuation, drainage pressure reduction-slow production increase-production increase-stable production-slow attenuation, drainage pressure reduction-slow production increase-stable production-slow attenuation, drainage pressure reduction-slow production increase-production increase-rapid attenuation and drainage pressure reduction-slow production increase-stable production-slow attenuation.
[0180] In some embodiments of the present invention, the abnormal data includes acquisition abnormality, abnormal working condition change, and no abnormality.
[0181] From the above description, it can be seen that an embodiment of the present application provides an abnormal data identification device for deep coalbed methane, including: a production mode determination module, used to determine the production mode of a target well based on pre-acquired production data and test production data of the target well; wherein, the production mode is composed of multiple production scenarios and the evolution order of multiple production scenarios; an abnormal data identification module, used to identify abnormal data of the target well based on the production mode and the abnormal data identification model corresponding to the pre-generated production mode; wherein, the production mode corresponds one-to-one to the abnormal data identification model; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive production data and test production data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0182] The present invention can improve the efficiency of the entire process of processing deep coalbed methane related data, reduce processing costs, achieve the purpose of reducing costs and increasing efficiency, and provide a basis and guarantee for the intelligent operation of deep coalbed methane fields.
[0183] The embodiment of the present application also provides a specific implementation of an electronic device capable of implementing all steps in the abnormal data identification method of deep coalbed methane in the above embodiment, see Figure 18 , electronic equipment specifically includes the following:
[0184] Processor 1201, memory 1202, communications interface 1203, and bus 1204;
[0185] The processor 1201 , the memory 1202 , and the communication interface 1203 communicate with each other via the bus 1204 . The communication interface 1203 is used to implement information transmission between relevant devices such as the server-side device, the power measurement device, and the user-side device.
[0186] The processor 1201 is used to call the computer program in the memory 1202. When the processor executes the computer program, all steps of the abnormal data identification method for deep coalbed methane in the above embodiment are implemented. For example, when the processor executes the computer program, the following steps are implemented:
[0187] Step 100: Determine a production mode of the target well based on pre-acquired production data and production test data of the target well; wherein the production mode is composed of multiple production scenarios and the evolution order of the multiple production scenarios;
[0188] Step 200: Identify abnormal data of the target well based on the production pattern and a pre-generated abnormal data identification model corresponding to the production pattern; wherein the production pattern corresponds one-to-one to the abnormal data identification model; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive the production data and the production test data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0189] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all steps of the method for identifying abnormal data of deep coalbed methane in the above-mentioned embodiment. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the computer program implements all steps of the method for identifying abnormal data of deep coalbed methane in the above-mentioned embodiment. For example, when the processor executes the computer program, the following steps are implemented:
[0190] Step 100: Determine a production mode of the target well based on pre-acquired production data and production test data of the target well; wherein the production mode is composed of multiple production scenarios and the evolution order of the multiple production scenarios;
[0191] Step 200: Identify abnormal data of the target well based on the production pattern and a pre-generated abnormal data identification model corresponding to the production pattern; wherein the production pattern corresponds one-to-one to the abnormal data identification model; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive the production data and the production test data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status and output the Q value of the abnormal data.
[0192] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences between the other embodiments. In particular, the hardware + program embodiments are generally similar to the method embodiments, so their description is relatively simple. For relevant portions, refer to the description of the method embodiments.
[0193] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0194] Although the present application provides method operation steps such as embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-creative work. The order of steps listed in the embodiments is only one way of executing the steps among many steps and does not represent the only execution order. When an actual device or client product is executed, it can be executed in the order shown in the embodiments or the drawings or in parallel (for example, in a parallel processor or multi-threaded processing environment).
[0195] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0196] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0197] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0198] Specific embodiments are used in this application to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A method for identifying abnormal data of deep coalbed methane, characterized in that: include: Determine the production mode of the target well according to the pre-acquired production data and test production data of the target well; wherein the production mode is composed of a plurality of production scenarios and the evolution order of the plurality of production scenarios; The abnormal data of the target well is identified according to the production mode and a pre-generated abnormal data identification model corresponding to the production mode; wherein the production mode corresponds to the abnormal data identification model one-to-one; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive the production data and the test production data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status, and output the Q value of the abnormal data.
2. The abnormal data identification method according to claim 1, characterized in that: The production data and test production data include: well diameter, dynamic liquid level, daily gas production, daily liquid production, well depth, bottom hole pressure and well casing pressure; Determining the production mode of the target well according to the pre-acquired production data and test production data of the target well includes: Determining an average density of produced fluid from the target well according to the dynamic liquid level, the bottom hole pressure, and the casing pressure; Determine the average liquid holdup of the target well according to the daily liquid production and the daily gas production; Determine the gas phase superficial velocity of the target well according to the daily gas production and the well diameter; Determining a well pressure gradient of the target well according to the bottom hole pressure, the well depth and the well casing pressure; The production mode is determined according to the average density of the produced fluid, the average liquid holdup, the gas superficial velocity and the well pressure gradient.
3. The abnormal data identification method according to claim 2, characterized in that: Determining the production mode according to the average density of the produced liquid, the average liquid holdup, the gas phase superficial velocity and the well pressure gradient includes: Determine, in chronological order, a plurality of production scenarios corresponding to the target well according to the average density of the produced fluid, the average liquid holdup, the gas phase superficial velocity, and the well pressure gradient; The production mode is determined according to the time sequence corresponding to the production scenarios.
4. The abnormal data identification method according to claim 3, characterized in that: The steps of generating the abnormal data recognition model include: Generate the state of each time step of the abnormal data recognition model according to the average density of the produced fluid, the average liquid holdup, the gas phase superficial velocity and the well pressure gradient of the target well; Determining a reward value of the abnormal data recognition model for the current time step according to the state of each time step and the corresponding real state; Perform the following iterative training until the preset training cutoff condition is met: Inputting the time series data of the average density of the produced liquid, the time series data of the average liquid holdup, the time series data of the well pressure gradient, and the time series data of the gas phase superficial flow rate into the LSTM network to determine the change state of the average density of the produced liquid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial flow rate; Inputting the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient and the change state of the gas phase superficial velocity into the BP network to output the Q value; The abnormal data recognition model of the current training round is adjusted according to the Q value and the reward value.
5. The abnormal data identification method according to any one of claims 1 to 4, characterized in that: The production scenarios include: drainage gas production, increase in production, slow increase in production, rapid attenuation, slow attenuation, stable production and gas production fluctuations; the production modes include: drainage and pressure reduction-increase in production-slow attenuation of gas production, drainage and pressure reduction-increase in production-rapid attenuation-increase in production again-slow attenuation, drainage and pressure reduction-slow increase in production-increase in production-stable production-slow attenuation, drainage and pressure reduction-slow increase in production-stable production-slow attenuation, drainage and pressure reduction-slow increase in production-increase in production-rapid attenuation and drainage and pressure reduction-slow increase in production-stable production-slow attenuation.
6. The abnormal data identification method according to claim 1, characterized in that: The abnormal data includes acquisition abnormality, abnormal working condition change and no abnormality.
7. A device for identifying abnormal data of deep coalbed methane, characterized in that: include: A production mode determination module, used to determine the production mode of the target well according to the pre-acquired production data and test production data of the target well; wherein the production mode is composed of multiple production scenarios and the evolution order of the multiple production scenarios; An abnormal data identification module is used to identify the abnormal data of the target well according to the production mode and a pre-generated abnormal data identification model corresponding to the production mode; wherein the production mode corresponds to the abnormal data identification model one-to-one; the abnormal data identification model is a DQN reinforcement learning model, and the Q network model in the DQN reinforcement learning model includes an LSTM network and a BP network; the LSTM network is used to receive the production data and the test production data, and characterize the production status of the target well at the current moment; the BP network is used to receive the production status, and output the Q value of the abnormal data.
8. The abnormal data identification device according to claim 7, characterized in that: The production data and test production data include: well diameter, dynamic liquid level, daily gas production, daily liquid production, well depth, bottom hole pressure and well casing pressure; The production mode determination module includes: An average density determination unit, used to determine the average density of the produced fluid of the target well according to the dynamic liquid level, the bottom hole pressure and the casing pressure; an average liquid holdup determination unit, configured to determine an average liquid holdup of the target well according to the daily liquid production and the daily gas production; A gas phase superficial flow rate determination unit, used to determine the gas phase superficial flow rate of the target well according to the daily gas production and the well diameter; a well pressure gradient determination unit, configured to determine the well pressure gradient of the target well according to the bottom hole pressure, the well depth and the well casing pressure; A production mode determination unit is used to determine the production mode according to the average density of the produced liquid, the average liquid holdup, the gas phase superficial velocity and the well pressure gradient.
9. The abnormal data identification device according to claim 8, characterized in that: The production mode determination unit comprises: A production scenario determination unit, configured to determine, in chronological order, a plurality of production scenarios corresponding to the target well according to the average density of the produced liquid, the average liquid holdup, the gas phase superficial velocity, and the well pressure gradient; The production mode determination subunit is used to determine the production mode according to the time sequence corresponding to the production scenario.
10. The abnormal data identification device according to claim 9, characterized in that: Also includes: The recognition model generation module is used to generate the abnormal data recognition model, and the recognition model generation module includes: A time-step state generating unit, used for generating the state of each time step of the abnormal data recognition model according to the average density of the produced fluid, the average liquid holdup, the gas phase superficial velocity and the well pressure gradient of the target well; A reward value determination unit, used to determine the reward value of the abnormal data recognition model for the current time step according to the state of each time step and the corresponding real state; The iterative training unit is used to perform the following iterative training until a preset training cutoff condition is met: a change state determination unit, used for inputting the time series data of the average density of the produced liquid, the time series data of the average liquid holdup, the time series data of the well pressure gradient, and the time series data of the gas phase superficial flow rate into the LSTM network to determine the change state of the average density of the produced liquid, the change state of the average liquid holdup, the change state of the well pressure gradient, and the change state of the gas phase superficial flow rate; A Q value generating unit is used to input the change state of the average density of the produced fluid, the change state of the average liquid holdup, the change state of the well pressure gradient and the change state of the gas phase superficial velocity into the BP network to output the Q value; A model optimization unit is used to adjust the abnormal data recognition model of the current training round according to the Q value and the reward value.
11. The abnormal data identification device according to any one of claims 7 to 10, characterized in that: The production scenarios include: drainage gas production, increase in production, slow increase in production, rapid attenuation, slow attenuation, stable production and gas production fluctuations; the production modes include: drainage and pressure reduction-increase in production-slow attenuation of gas production, drainage and pressure reduction-increase in production-rapid attenuation-increase in production again-slow attenuation, drainage and pressure reduction-slow increase in production-increase in production-stable production-slow attenuation, drainage and pressure reduction-slow increase in production-stable production-slow attenuation, drainage and pressure reduction-slow increase in production-increase in production-rapid attenuation and drainage and pressure reduction-slow increase in production-stable production-slow attenuation.
12. The abnormal data identification device according to claim 11, characterized in that: The abnormal data includes acquisition abnormality, abnormal working condition change and no abnormality.
13. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the method for identifying abnormal data of deep coalbed methane described in any one of claims 1 to 6 are implemented.
14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method for identifying abnormal data of deep coalbed methane described in any one of claims 1 to 6 are implemented.
15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for identifying abnormal data of deep coalbed methane described in any one of claims 1 to 6 are implemented.
Citation Information
Patent Citations
Method and device for performing abnormity detecting on coal-bed gas well and storage medium
CN111305812A
Network anomaly intrusion detection model training method, detection method and device
CN118018426A
Field development planning based on deep reinforcement learning
US20220220828A1
