Drilling gas extraction on-line metering control system

Abstract

The invention provides a drilling gas extraction online metering control system, and relates to the technical field of coal mine informatization systems, and the system comprises a data acquisition unit which is used for acquiring extraction pump control time sequence data and pipeline gas sensing time sequence data; the flow prediction unit is used for predicting gas flow prediction time sequence data corresponding to a second time period in the future through a gas extraction flow prediction model; a backbone network of the gas extraction flow prediction model adopts an LSTM model; and the extraction control unit is used for determining extraction pump control parameter prediction time sequence data corresponding to a second time period through a gas extraction pressure control model, and the gas extraction pressure control model adopts a Transform model based on an attention mechanism. Therefore, through a comprehensive automatic on-line control system, gas data are monitored in real time and dynamically adjusted by means of a machine learning model, and the frequency of human intervention and the dependence on operators are reduced.

Description

Technical Field

[0001] The present application relates to the technical field of coal mine information systems, and in particular to an online metering control system for drilling gas extraction. Background Art

[0002] With the continuous exploitation of coal resources, gas accidents have become one of the main threats to coal mine safety production. Borehole gas extraction is one of the important measures to prevent and control gas accidents. Its purpose is to extract gas from coal seams, reduce the gas content in coal seams, and thus reduce the occurrence of gas accidents.

[0003] The existing borehole gas extraction control system mainly adopts manual control or simple automatic control. In the manual control mode, the operator needs to pay attention to the changes in the gas extraction parameters at all times and manually adjust the operating status of the extraction equipment, which is not only inefficient, but also prone to human errors, resulting in untimely control. In addition, most of the existing borehole gas extraction metering systems only have simple data collection and storage functions, lack the ability to deeply analyze and process data, and cannot extract valuable information from a large amount of gas extraction data, making it difficult to discover potential problems and rules in the gas extraction process.

[0004] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the invention

[0005] The present application provides an online metering and control system for borehole gas extraction, which is used to at least solve the problem that traditional borehole gas extraction systems rely on manual or simple automated control and cannot support real-time and accurate gas extraction parameter adjustment, and are difficult to cope with sudden or dynamically complex working conditions.

[0006] The embodiment of the present application provides an online metering and control system for borehole gas extraction, including: a data acquisition unit, used to acquire extraction pump control timing data and pipeline gas sensor timing data; the pipeline gas sensor timing data includes a plurality of gas sensor data corresponding to a first adjacent time period of a sampling position in a borehole gas extraction pipeline, and the gas sensor data includes gas flow, gas concentration, pressure and temperature; the extraction pump control timing data includes a plurality of extraction pump control parameters corresponding to the first time period, and the extraction pump control parameters include extraction pump speed and extraction valve opening; a flow prediction unit, used to input the extraction pump control timing data and pipeline gas sensor timing data into a gas extraction flow prediction model to predict the gas flow prediction timing data corresponding to a second time period in the future; the backbone network of the gas extraction flow prediction model adopts LSTM (Long Short-Term Memory, long short-term memory network) model; an extraction control unit, used to input the pipeline gas sensor time series data, the gas flow prediction time series data and the extraction pump control time series data into the gas extraction pressure control model to determine the extraction pump control parameter prediction time series data corresponding to the second time period; the gas extraction pressure control model adopts a Transformer model based on the attention mechanism.

[0007] The online metering and control system for drilling gas extraction provided by the present application can produce at least the following technical effects:

[0008] (1) By introducing the LSTM model and the Transformer model based on the attention mechanism, gas flow prediction and adaptive adjustment of the control parameters of the extraction pump are carried out, realizing data-driven automatic control. The LSTM model is good at processing time series data and can capture the long-term and short-term dependencies of gas flow changes. The introduction of the Transformer model further enhances the system's adaptability to complex control scenarios and the fine capture of feature information during the gas extraction process. Through the synergy of these two models, the system can quickly and accurately adjust the control parameters of the extraction pump, such as speed and valve opening, based on real-time and historical data, avoiding the lag and error that may be caused by manual control, and greatly improving the timeliness and accuracy of control.

[0009] (2) By utilizing a variety of gas sensor data, such as gas flow, gas concentration, pressure and temperature, combined with the control time series data of the extraction pump, a complete time series data input is formed. Then, the LSTM model is used to predict the gas flow. Based on the time series data input, the gas flow change in the future can be predicted more accurately, and the changing trend of the gas flow can be perceived in advance, so as to make targeted adjustments in advance to ensure the efficiency and safety of gas extraction.

[0010] (3) Since parameters such as gas concentration, flow, pressure and temperature are affected by the geological conditions of the mining area and the coal mining process, the complexity of gas extraction makes it difficult for traditional control methods to cope with sudden or dynamic changes. Here, by using the Transformer model to deeply mine a large amount of time series data and relying on the attention mechanism, multiple control factors can be comprehensively considered in a complex gas extraction environment, so that the automatic control system based on deep learning can dynamically adjust the control strategy, effectively solving the problem of insufficient processing capacity caused by multiple control dimensions in the traditional control mode. For example, when the gas concentration changes drastically or the pipeline pressure fluctuates greatly, the stability of gas extraction can still be maintained, reducing the risk of failure of the system in a complex environment.

[0011] Through this technical solution, through a fully automated online control system, relying on machine learning models to monitor gas data in real time and make dynamic adjustments, the frequency of human intervention and dependence on operators are reduced, greatly reducing the safety hazards caused by human errors. In addition, when the system detects abnormal gas concentration or flow in real time, it can automatically control the corresponding emergency parameters of the extraction pump, thereby effectively reducing the accumulation of gas in the coal seam and extraction pipeline, reducing the risk of gas explosion or leakage, and improving the safety of gas extraction and coal mine production operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to 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 paying any creative work.

[0013] Figure 1 A structural block diagram of an example of an online metering control system for drilling gas extraction according to an embodiment of the present application is shown;

[0014] Figure 2 A structural block diagram showing another example of an online metering control system for drilling gas extraction according to an embodiment of the present application is shown;

[0015] Figure 3 A schematic diagram of the structural connection of an example of a gas drainage flow prediction model according to an embodiment of the present application is shown;

[0016] Figure 4 A schematic diagram of structural connection of an example of a gas extraction pressure control model according to an embodiment of the present application is shown;

[0017] Figure 5An operation flow chart of an example of an online metering control method for drilling gas extraction according to an embodiment of the present application is shown;

[0018] Figure 6 An on-site installation diagram showing an example of an installation module for a high-position drilling site according to an embodiment of the present application is shown;

[0019] Figure 7 A schematic diagram of an interface of an example of a terminal to which the online metering control system for drilling gas extraction according to an embodiment of the present application is applied is shown;

[0020] Figure 8 It is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.

[0023] Figure 1 A structural block diagram of an example of an online metering control system for borehole gas extraction according to an embodiment of the present application is shown.

[0024] Regarding the carrier of the system of the embodiment of the present application, it can be any controller or processor with computing or processing capabilities. By jointly using LSTM and Transformer deep learning models, it can respond to potential gas exceeding the standard or abnormal pressure conditions in a timely manner, provide effective intelligent control strategies, improve the automation and intelligence level of the gas extraction control system, enhance the system's prediction and decision-making capabilities, and significantly improve the safety, efficiency and adaptability of the gas extraction process.

[0025] In some examples, it can be integrated and configured in an electronic device or terminal through software, hardware, or a combination of software and hardware, and the type of the terminal or electronic device can be diverse, such as a mobile phone, a tablet computer, or a desktop computer, etc.

[0026] like Figure 1 As shown, the borehole gas extraction online metering control system 100 includes a data acquisition unit 110 , a flow prediction unit 120 and an extraction control unit 130 .

[0027] The data acquisition unit 110 is used to acquire the extraction pump control timing data and the pipeline gas sensor timing data.

[0028] Here, the pipeline gas sensor timing data includes multiple gas sensor data corresponding to the adjacent first time period for the sampling position in the borehole gas extraction pipeline, and the gas sensor data includes gas flow, gas concentration, pressure and temperature; the extraction pump control timing data includes multiple extraction pump control parameters corresponding to the first time period, and the extraction pump control parameters include the extraction pump speed and the extraction valve opening.

[0029] It should be noted that the gas pressure in the coal seam directly affects the gas outflow speed and the difficulty of extraction. Monitoring the gas pressure can help understand the state of gas in the coal seam. In addition, temperature changes will affect the physical properties and fluidity of gas, and during the gas extraction process, temperature changes may be caused by the natural heating of the coal seam, the heat generated by the operation of the extraction equipment, and other reasons.

[0030] In some embodiments, multimodal sensors are arranged in the borehole and its surrounding areas, and these sensors include gas flow sensors, gas concentration sensors, pressure sensors, temperature sensors, etc. Sensors are installed at the sealing position, key points inside the borehole, and on the gas extraction pipeline to achieve comprehensive monitoring of the entire extraction process. Specifically, it is recommended that the gas flow sensor and pressure sensor be installed on the borehole gas extraction pipeline, as close to the extraction hole outlet as possible, so as to accurately measure the gas flow and pressure. In addition, temperature sensors can be installed in a distributed manner at different positions of the extraction pipeline to comprehensively monitor the temperature changes of the entire extraction system.

[0031] The data is then transmitted to the central control system via the optical fiber network, and the frequency and resolution of data acquisition are set according to actual needs to ensure the accuracy and real-time nature of the data. In addition, by connecting to the control system of the extraction pump, the operating parameters of the extraction pump are obtained to monitor the working status of the extraction pump in real time, which plays an important role in controlling the gas extraction flow and pressure. For example, the gas extraction flow can be changed by adjusting the speed of the extraction pump.

[0032] It should be emphasized that the gas flow parameter is a key parameter that directly reflects the gas extraction effect, and the size of the gas flow is affected by many factors, such as coal seam permeability, extraction pump performance, pipeline resistance, etc.

[0033] The flow prediction unit 120 is used to input the extraction pump control time series data and the pipeline gas sensor time series data into the gas extraction flow prediction model to predict the gas flow prediction time series data corresponding to the second time period in the future.

[0034] Here, the backbone network of the gas extraction flow prediction model adopts an LSTM model. In an example of an embodiment of the present application, the gas extraction flow prediction model can adopt a hybrid model including a SARIMA (Seasonal Autoregressive Integrated Moving Average) model module and an LSTM model module. Specifically, first, the SARIMA model is used to make a preliminary prediction of the historical gas flow data to obtain a preliminary prediction value sequence. Then, the prediction value sequence of the SARIMA model and other input features (gas pressure, temperature, extraction pump operating parameters, pipeline parameters, etc.) are used as inputs to the LSTM model. Then, the LSTM model further adjusts and optimizes the prediction results of the SARIMA model by learning the complex relationship between these input features and the gas flow, and obtains the final gas extraction flow prediction value.

[0035] It should be noted that during the gas extraction process, the flow rate may be affected by seasonal factors. For example, the coal seam temperature, gas pressure and other conditions may be different in different seasons, thus affecting the gas outflow and extraction flow rate. In this way, through the SARIMA model module integrated in the hybrid model, the seasonal autoregression and seasonal moving average parts can be used to effectively capture this seasonal variation pattern. In addition, in addition to seasonal factors, the gas extraction flow rate may also have trend changes and non-seasonal fluctuations. The non-seasonal autoregression, difference and moving average parts in the SARIMA model module can handle these fluctuations. The number of differences can be adjusted according to the stationarity requirements of the data to eliminate the trend in the data. The non-seasonal autoregression and moving average coefficients can capture short-term dependencies and random fluctuations in the data.

[0036] Through the embodiments of the present application, SARIMA is combined with a deep learning algorithm (i.e., LSTM) to give full play to the advantages of both. SARIMA can provide a preliminary prediction framework, and use its ability to handle seasonality and trend to provide a better input for the deep learning model. Furthermore, LSTM can further learn the complex nonlinear relationship between input features and gas flow, and adjust and optimize the prediction results of SARIMA. Through the hybrid model structure, the accuracy and robustness of the prediction can be improved, better adapting to the characteristics of gas extraction flow data, and better capturing the seasonal, trend and non-seasonal changes of gas extraction flow, improving the accuracy and reliability of the prediction.

[0037] In another example of an embodiment of the present application, the gas extraction flow prediction model can also adopt a Bi-LSTM model based on the attention mechanism, and more details will be expanded in conjunction with other examples below.

[0038] The extraction control unit 130 is used to input pipeline gas sensor time series data, gas flow prediction time series data and extraction pump control time series data into the gas extraction pressure control model to determine the extraction pump control parameter prediction time series data corresponding to the second time period.

[0039] Here, the gas extraction pressure control model adopts the Transformer model based on the attention mechanism. By using the Transformer model based on the attention mechanism to drive the adaptive control of gas extraction control parameters, it can dynamically focus on the key parts of the input time series data, and intelligently adjust the speed and valve opening of the extraction pump according to the estimated gas flow and historical multimodal data to ensure the stability of the extraction pipeline pressure.

[0040] Therefore, by using the Transformer model, the operating parameters of the extraction pump can be dynamically adjusted in the face of various complex situations, realizing automation and refined control of the extraction process, especially providing rapid response when the gas concentration fluctuates drastically, avoiding a decrease in extraction efficiency or accidents.

[0041] Through the embodiment of this application, an intelligent and efficient borehole gas extraction control system is constructed. Through the collaborative work of LSTM and Transformer models, the system realizes accurate prediction and control of gas flow and pressure, reduces the risk of human intervention, and improves the stability of gas extraction. In addition, the system can automatically adapt to complex gas extraction environments, optimize equipment operating parameters, and thus improve the operating efficiency and safety of the overall gas extraction system.

[0042] Figure 2 A structural block diagram of another example of an online metering control system for borehole gas extraction according to an embodiment of the present application is shown.

[0043] like Figure 2 As shown, the borehole gas extraction online metering control system 200 includes a data acquisition unit 210 , a flow prediction unit 220 , an extraction control unit 230 , an extraction volume monitoring unit 240 and an alarm monitoring unit 250 .

[0044] For specific implementation details of the data acquisition unit 210, the flow prediction unit 220 and the sampling control unit 230, please refer to and combine Figure 1 The description of the relevant operations of the corresponding units in will not be repeated here.

[0045] It should be noted that the flow and concentration of gas will be significantly affected by environmental conditions (such as pressure and temperature). Under high pressure or low temperature conditions, the density and flow rate of gas will change, resulting in the inaccuracy of the directly measured gas flow and concentration, which will affect the overall control of the extraction system. The gas state equation shows that the volume (or flow) of gas is directly related to pressure and temperature. As coal mining progresses, the pressure and temperature of the mine environment will change, which in turn affects the changes in gas flow and concentration.

[0046] In some embodiments, the extraction volume monitoring unit 240 is used to perform the following operations: calculate a correction coefficient based on the pressure and temperature in each gas sensor data, and calculate the actual gas extraction volume sequence corresponding to the first time period based on the correction coefficient and the gas flow and gas concentration in each gas sensor data.

[0047] Through the embodiments of the present application, during the gas extraction process, temperature and pressure often change dynamically over time. By calculating the correction coefficients of pressure and temperature in real time, the system can automatically adjust the calculation method of the extraction volume according to the current mine conditions, thereby achieving accurate calculation results of the gas extraction volume.

[0048] In some examples of the embodiments of the present application, the actual gas extraction volume sequence is calculated by the following formula:

[0049]

[0050] V k (t) = Q (tk·Δt) × C g (tk·Δt)×C(tk·Δt)×Δt , formula (2)

[0051]

[0052] In the formula, β 1 , β 2 and β 3 Respectively represent the corresponding calibration coefficients, which are preset according to the coal seam characteristics and experimental data; P(t) and θ(t) are the pressure and temperature sampled and detected at time step t; m represents the total number of historical time steps in the first time period relative to time step t; P 0 and θ 0 Represent standard pressure and standard temperature respectively; C(t) represents the correction coefficient at time step t; V k (t) represents the actual gas extraction volume corresponding to the kth historical time step relative to time step t; w k represents the weight coefficient of the kth historical time step, w 0 >w 1 >...>w m-1 , where w 0represents the weight coefficient of time step t; V(t) represents the actual gas extraction volume at time step t, Q(tk·Δt) represents the gas flow rate at time step tk·Δt, Δt represents the time interval between adjacent time steps; C(tk·Δt) represents the correction coefficient corresponding to time step tk·Δt, C g (tk·Δt) represents the gas concentration corresponding to the time step tk·Δt.

[0053] In this embodiment, the total number of time steps m and the weight factor w are selected appropriately. k , for each historical time step k, calculate the actual gas extraction volume V k (t), and then apply the weighting factor to all V k (t) is weighted summed to obtain the actual gas extraction volume V(t) at the current time point.

[0054] It should be noted that during the coal mine gas extraction process, sensor data may experience short-term abnormal fluctuations due to equipment interference or changes in the external environment.

[0055] Through the embodiment of the present application, the data of multiple historical time steps are introduced for weighted averaging, which effectively smooths the impact caused by the instantaneous fluctuation of gas flow, concentration and other data in a short period of time, reduces the interference of abnormal data on the calculation results, and improves the stability and accuracy of the actual gas extraction volume sequence. In addition, by dynamically adjusting the weight factor w k , the system can give different importance to data at different time steps. For example, the system can give higher weight to the data at the current moment, so as to more sensitively reflect the current gas extraction status; while for earlier time steps, it can give lower weight, so that the influence of historical data is gradually weakened, so that the system can better capture the changing trend of gas flow and concentration, and improve the accuracy of the determined actual gas extraction volume sequence.

[0056] In some embodiments, the alarm monitoring unit 250 is used to perform the following operations: monitor whether the gas sensor data in the second time period meets the preset safety sensor interval. When the monitored gas sensor data does not meet the safety sensor interval, reduce the speed of the extraction pump of the current borehole gas extraction pipeline and gradually close the extraction valve, and start the extraction pump of the standby extraction pipeline to enter the standby state. At this time, the extraction valve of the standby extraction pipeline is in a closed state, and the sensor of the standby extraction pipeline is controlled to start collecting data. After the extraction valve of the current borehole gas extraction pipeline is completely closed, open the extraction valve of the standby extraction pipeline, and adjust the extraction pump speed and valve opening of the standby extraction pipeline to switch to the standby extraction pipeline to respond to the gas extraction task.

[0057] In the embodiment of the present application, the system can monitor the gas sensor data in real time and respond quickly to situations that exceed the safe sensing range, thereby realizing a real-time monitoring and early warning mechanism. When key parameters such as gas concentration, pressure or temperature deviate from the safe range, the system can automatically reduce the speed of the extraction pump and gradually close the extraction valve to avoid dangers caused by excessive extraction. It can reduce the time of manual intervention and improve the system's response speed to emergencies, thereby ensuring the safety of coal mine production.

[0058] On the other hand, when an abnormality occurs in the main extraction pipeline, the system can automatically switch to the backup extraction pipeline, start the backup extraction pump and adjust the valve opening to ensure the continuity of the gas extraction task. The main and backup pipeline switching mechanism can effectively avoid gas extraction interruptions caused by failures in the main extraction pipeline and ensure the stable operation of the system. In addition, during the pipeline switching process, the system will gradually adjust the speed and valve opening of the extraction pump to ensure a smooth transition of pipeline pressure, avoid pressure fluctuations or pipeline damage caused by sudden switching, and help extend equipment life and reduce maintenance costs.

[0059] Figure 3 A schematic diagram of the structural connection of an example of a gas extraction flow prediction model according to an embodiment of the present application is shown.

[0060] like Figure 3 As shown, the gas extraction flow prediction model 300 adopts a Bi-LSTM model based on an attention mechanism, which includes a cascaded TCN layer 310, a Bi-LSTM layer 320, an attention layer 330 and an output layer 340.

[0061] More specifically, the TCN layer 310 is used to capture short-term and long-term dependencies in the extraction pump control time series data and the pipeline gas sensor time series data at different time scales through multiple convolutional layers.

[0062] Specifically, the TCN (Temporal Convolutional Network) layer extracts features at different time scales through convolution operations to capture short-term and long-term dependencies in gas flow data.

[0063]

[0064] In the formula, Indicates the time step t of historical sampling p The multi-scale feature vector extracted by the TCN layer, TCN represents the feature processing process of the multi-scale temporal convolutional network; Represents the time step t p -n+1 to time step t p The input feature sequence is n, and n represents the time window length.

[0065] Through this embodiment, based on the introduction of a multi-scale time convolutional network, the model can capture short-term fluctuations and long-term trends in gas flow from different time scales. Through a multi-level feature extraction method, the model can more comprehensively understand the dynamic changes of gas flow and avoid prediction errors caused by the limitations of a single time scale.

[0066] The Bi-LSTM layer 320 is used to capture the bidirectional dependencies of multi-scale feature vectors.

[0067] Specifically, after extracting the multi-scale features, the Bi-LSTM layer 320 further processes these features to capture the forward and backward temporal dependencies, enabling deeper modeling of the dependencies of sequence data.

[0068]

[0069] In the formula, Indicates that at time step t p The forward LSTM hidden state of It is a forward LSTM unit, which is used to process the forward dependency of time series; Indicates that at the previous time step t p -1 forward LSTM hidden state; Indicates that at time step t p The backward LSTM hidden state; It is a backward LSTM unit, which is used to process the backward dependency of time series; is at the next time step t p +1 backward LSTM hidden state; Indicates that at time step t p The comprehensive hidden state.

[0070] Through this embodiment, the Bi-LSTM layer 320 further processes the multi-scale features, and through the forward and backward LSTM structures, it can effectively capture the complex time series dependencies in the gas flow data. Therefore, by combining the forward information and the backward information, the model can take into account more historical and future context information when predicting, thereby improving the accuracy of the prediction results.

[0071] The attention layer 330 adopts an adaptive multi-head attention mechanism to dynamically adjust the weight of each time step.

[0072] Specifically, the hidden state of each sampling time step is adaptively weighted through the adaptive multi-head attention mechanism, which can dynamically adjust the attention weight according to the characteristics of the input data.

[0073]

[0074] MultiHead(Q,K,V)=Concat(head 1 ,...,head H )W A +b A , formula (12)

[0075] In the formula, Attention(Q,K,V) represents the output of the attention mechanism, Q, K and V represent The query vector, key vector and value vector obtained by linear transformation, K T represents the vector transpose of K, d k represents the dimension of the key vector; Denotes the corresponding time step t p The adaptive bias term, softmax represents the softmax activation function; W b and b b Represent the weight matrix and bias term of the adaptive weight adjustment module respectively, Denotes the corresponding time step t p The input feature vector of head i represents the output of the i-th attention head, W i Q , W i K and W i V Represents the query transformation matrix, key transformation matrix and value transformation matrix of the i-th attention head respectively; Concat(head 1 ,...,head H ) means concatenating the outputs of all H attention heads by dimension, and MultiHead(Q,K,V) means the corresponding time step t p The final output of the multi-head attention mechanism, W A and b A They represent the linear transformation weight matrix and bias vector of the multi-head attention output respectively.

[0076] Through this embodiment, an adaptive multi-head attention mechanism is introduced, and the model can dynamically adjust the weight of each time step according to the specific characteristics of the gas flow data. Furthermore, by highlighting the time point with the greatest impact on the prediction result, the attention mechanism greatly improves the adaptability and prediction accuracy of the model in complex scenarios.

[0077] It should be noted that the adaptive bias term used in the attention mechanism enables the model to dynamically adjust the weight according to the changes in input features, enhancing the model's adaptability to environmental changes. Whether it is a sharp fluctuation in gas flow or a change in external conditions, the model can make corresponding adjustments through the adaptive mechanism to ensure the stability and accuracy of the prediction results.

[0078] The output layer 340 is used to predict the gas flow at each future time step.

[0079]

[0080] In the formula, represents the future time step t s The predicted value of gas flow, t s =t p +Δt, Δt means t s With t p The time interval between o and b o They represent the weight matrix and bias term of the linear output layer respectively.

[0081] Here, based on the output of the multi-head attention mechanism, the model generates the final gas flow prediction value through linear transformation.

[0082] Through the embodiments of the present application, multi-scale feature extraction and dynamic attention mechanism are integrated, so that the model can still maintain efficient prediction capabilities when facing data noise, sudden changes and complex environmental conditions. In particular, in the case of sudden increases or decreases in gas flow or changes in equipment parameters, the model can quickly adjust the prediction strategy and provide accurate prediction results. By accurately predicting gas flow, potential gas accumulation risks can be identified in a timely manner, providing a reliable basis for the safety control of gas extraction and reducing the possibility of gas accidents.

[0083] Figure 4 A schematic diagram of the structural connection of an example of a gas extraction pressure control model according to an embodiment of the present application is shown.

[0084] like Figure 4 As shown, the gas drainage pressure control model 400 includes an input layer 410 , an encoder layer 420 , a decoder layer 430 and an output layer 440 .

[0085] The input layer 410 is used to receive pipeline gas sensor time series data, gas flow prediction time series data and the extraction pump control time series data, and after linear embedding and position encoding, a comprehensive sequence representation containing time and feature information is formed.

[0086] In the input layer, the calculation process of the position encoding vector is very important because it provides sequence information for the model, allowing the Transformer model to perceive the time order in the input sequence. In the gas extraction pressure control scenario, by adding position encoding to each time step, the model can recognize the order between different time steps. Specifically, the position encoding uses sine and cosine functions, which act on different dimensions respectively to provide position information of different frequencies.

[0087]

[0088] In the formula, Indicates the tth corresponding sampling detection in the input data 1 The input information of time steps has a dimension of d input ; W emb Represents a linear embedding matrix with dimension d model ×d input ; Indicates that at t 1 The input feature representation of time steps is d model ; Indicates the corresponding t 1 The position encoding vector of time steps is used to inject the position information in the time series into the model, with a dimension of d model ; j represents the dimension index of the position encoding.

[0089] The encoder layer 420 is used to process the comprehensive sequence representation to capture the dependencies between each time step through a self-attention mechanism to generate the corresponding context feature vector.

[0090] The encoder layer includes multiple encoding units, each of which includes a cascaded self-attention mechanism layer and a feedforward neural network layer. The feature processing process of the encoding unit is expressed by the following formula:

[0091]

[0092] In the formula, is the coding unit for the tth 1 The output feature vector calculated from the input information of time steps represents the context information processed by the self-attention mechanism and the feedforward neural network; represents the time step t of historical sampling 1 The attention weight at time step τ, z τ represents the input feature representation at the τth time step, T 1 Indicates the total number of time steps corresponding to the input data; W enc1 ,W enc2 represents the weight matrix of the feedforward neural network layer in the encoding unit, b enc1 ,b enc2 Represents the bias term of the feedforward neural network layer in the encoding unit; ReLU represents the RELU nonlinear activation function.

[0093] The deduction process for the above formula (17) is as follows:

[0094] The processing of the coding unit can be expressed by the following formula:

[0095]

[0096] In the formula, Represents the input features of the encoding unit, which is a comprehensive sequence representation from the previous encoding unit or input layer. SelfAtt(·) represents the self-attention mechanism processing process; FFN(·) represents the feedforward neural network processing process.

[0097] The self-attention mechanism calculates the dependency of each time step with other time steps in the sequence, generates attention weights and calculates weighted feature representations. The calculation formula is as follows:

[0098]

[0099] The feedforward neural network is used to perform further nonlinear mapping on the features output by the self-attention mechanism. The calculation formula is as follows:

[0100] FFN(h)=W 2 ReLU(W 1 ·h+b 1 )+b 2 , Formula (20)

[0101] Furthermore, by combining the above equations (18) to (20), we obtain equation (17).

[0102] Through this embodiment, each encoding unit captures the dependencies between time steps in the sequence through a self-attention mechanism, and further processes these features through a feedforward neural network to generate a context feature vector.

[0103] The decoder layer 430 is used to process the context feature vector to generate a control parameter feature sequence corresponding to the prediction of the future time step through a multi-head self-attention mechanism combined with a mask mechanism. The decoder layer includes a plurality of decoding units, each of which includes a cascaded mask-based self-attention mechanism layer and a feedforward neural network layer. The feature processing process of the decoding unit is expressed by the following formula:

[0104]

[0105] In the formula, Denotes the decoding unit at the future time step t 2 The generated control parameter characteristics; h enc,ξ represents the context feature vector calculated by the encoder layer for the input information of the ξ-th time step of the historical sampling; Denotes the decoder at time step t 2 The dependency on the time step ξ, which is the attention weight obtained by the multi-head self-attention mechanism; W dec1 ,W dec2represents the weight matrix of the feedforward neural network layer in the decoding unit, b dec1 ,b dec2 Represents the bias term of the feed-forward neural network layer in the decoding unit.

[0106] Here, the self-attention mechanism and feedforward neural network processing of the decoding unit are symmetrical with those of the encoding unit, and the deduction details of equation (21) are not described here. In addition, a mask mechanism is incorporated into the decoding unit, which only allows the decoder to access the current and previous time steps, preventing the decoder from seeing future data when predicting future time steps. In other words, The maximum time step is t 2 itself, using only time step t 2 In this way, the model can adjust the control parameters in real time according to the current and historical data through the autoregressive generation method, avoiding the leakage of future information in advance and improving the accuracy of the prediction results.

[0107] Therefore, the model has the ability to respond quickly and can adjust the extraction parameters in real time when the gas extraction environment changes, realize dynamic regulation, reduce lag, and improve the operating efficiency and safety of the system.

[0108] The output layer 440 is used to perform linear layer processing on the predicted control parameter feature sequence to output the predicted value of the drainage pump control parameter at each time step in the second time period in the future.

[0109] Here, the decoder features are mapped to specific control parameter prediction values ​​through linear transformation to generate the pump speed and valve opening of the future time step.

[0110]

[0111] In the formula, y dec,t represents the decoder layer for the future time step t 2 generated control parameter characteristics; Respectively represent the corresponding time step t 2 Predicted pump speed and valve opening; W rpm ,W valve represents the linear transformation matrix of the output layer, b rpm ,b valve Represents the bias term of the output layer.

[0112] It should be noted that in the Transformer model architecture adopted in the embodiment of the present application, its encoder unit and decoder unit both adopt a method of integrating multi-head self-attention mechanism and feedforward neural network, which can effectively capture the complex interactive relationship between multi-dimensional features, such as nonlinear correlation between factors such as flow, concentration, temperature, etc., and can identify the impact of different factors on control parameters, thereby generating stable control prediction values, which helps to maintain stable operation of the system during gas extraction.

[0113] Through the embodiments of the present application, a gas extraction pressure control model is used to accurately predict future gas flow and pressure, so that the rotation speed and valve opening of the extraction pump can be finely controlled, and the extraction rate can be adjusted according to actual needs to avoid unnecessary energy consumption or excessive extraction, thereby significantly improving the economic benefits of the overall system.

[0114] The following is a detailed description of the training process for the gas drainage pressure control model:

[0115] First, a data sample set for the gas extraction pressure control model is constructed. The data set should be organized in a time series format, that is, for each time step, the corresponding sensor data and control parameters are collected. In each sample, the input features of each time step include: gas flow, gas concentration, pressure, temperature, extraction pump speed, valve opening, etc. In addition, each future time step t is set g Tags It includes the control parameters (rotation speed and valve opening) of the future extraction pump under the corresponding situation. Then, the training set, validation set and test set are divided.

[0116] Then, initialize the layers of the gas drainage pressure control model and set the optimizer, such as Adam or SGD. Then, batch input the data samples of each time step in the training set into the model for training optimization of the prediction task.

[0117] Then, the model loss is calculated based on the output prediction value and the true value of the model, so as to minimize the model loss through iterative training and achieve better prediction performance.

[0118] It should be noted that in the gas extraction control process, the core goal of the system is to accurately predict and control the future extraction pump speed and valve opening to ensure the stability and safety of the extraction volume. However, the gas flow and concentration are affected by multiple factors such as environmental conditions such as temperature and pressure, resulting in strong system dynamics. Therefore, it is necessary to introduce an effective loss function design to guide model learning, ensure that the system can make predictions in complex scenarios, and maintain the smoothness of control parameters.

[0119] In view of this, in some examples of the embodiments of the present application, the loss function L of the gas extraction pressure control model istotal It is expressed by the following formula:

[0120] L total =λ 1 L MSE +λ 2 L smooth , Formula (24)

[0121]

[0122] Where, L MSE and L smooth They represent the prediction error loss term and the change rate constraint loss term, respectively, 1 and λ 2 They represent the weight coefficients of the corresponding loss items respectively; N represents the total number of samples in the data sample set, T g represents the total number of predicted time steps corresponding to the u-th sample, and Respectively represent the u-th sample in the future time step t g and t g -1 is the predicted value of the pump speed; Represents the u-th sample at the future time step t g The extraction pump speed label; and Respectively represent the u-th sample in the future time step t g and t g -1 is the predicted value of valve opening; Represents the u-th sample at the future time step t g Valve opening label.

[0123] In the loss function provided in this application, on the one hand, by introducing the mean square error (MSE) as the main loss term, the model can learn how to accurately predict the pump speed and valve opening of the future time step, ensuring that the model can generate high-precision prediction values ​​in various gas extraction scenarios and avoid excessive prediction errors. In addition, since different mine conditions may vary greatly, MSE optimizes the prediction error of each sample one by one, so that the model can maintain robustness in different environments and adapt to changes in different gas concentrations, temperatures, pressures and other conditions.

[0124] On the other hand, in order to ensure the smooth operation of the extraction pump, the rate-of-change constraint loss term limits the parameter changes between adjacent time steps. When the gas flow changes suddenly or the environmental conditions fluctuate, the smoothing loss makes the change of the control parameters tend to be a gradual adjustment rather than a sudden drastic change. This enhances the smoothness of the system control parameters and reduces the drastic fluctuations in the control parameters, making the gas extraction system more stable and reliable during operation, avoiding safety risks caused by excessive load of mechanical equipment or sudden increase in gas concentration, while also reducing equipment wear and extending the service life of the equipment.

[0125] Through this embodiment, MSE and smooth loss are calculated by weight λ 1 and λ 2 The comprehensive loss function can prevent drastic changes in control parameters while ensuring prediction accuracy, ensuring that the model can generate accurate prediction results and smoothly adjust the control parameters during the extraction process. The comprehensive loss function helps the model optimize these goals at the same time through multiple constraints to achieve efficient and safe gas extraction. In addition, by adjusting λ 1 and λ 2 The system can flexibly adapt to different gas extraction scenarios by adjusting the value of . In scenarios where high-precision prediction is required, λ can be increased accordingly. 1 In addition, when the system stability requirement is high, λ can be increased accordingly. 2 The proportion of the model allows the model to optimize different objectives for different scenarios.

[0126] Figure 5 An operational flow chart of an example of an online metering control method for borehole gas extraction according to an embodiment of the present application is shown.

[0127] like Figure 5 As shown, in step S510, extraction pump control timing data and pipeline gas sensor timing data are obtained.

[0128] Here, the pipeline gas sensor timing data includes multiple gas sensor data corresponding to the adjacent first time period for the sampling position in the borehole gas extraction pipeline, and the gas sensor data includes gas flow, gas concentration, pressure and temperature; the extraction pump control timing data includes multiple extraction pump control parameters corresponding to the first time period, and the extraction pump control parameters include the extraction pump speed and the extraction valve opening.

[0129] In step S520, the extraction pump control time series data and the pipeline gas sensor time series data are input into the gas extraction flow prediction model to predict the gas flow prediction time series data corresponding to the second time period in the future. The backbone network of the gas extraction flow prediction model adopts the LSTM model.

[0130] In step S530, the pipeline gas sensor time series data, the gas flow prediction time series data and the extraction pump control time series data are input into the gas extraction pressure control model to determine the extraction pump control parameter prediction time series data corresponding to the second time period. The gas extraction pressure control model adopts the Transformer model based on the attention mechanism.

[0131] Figure 6 An on-site installation diagram of an example of an installation module for a high-position drilling site according to an embodiment of the present application is shown.

[0132] like Figure 6 As shown, there are 12 boreholes in the high-position drilling site, and each drilling site is equipped with a manifold. Therefore, each drilling site is equipped with 12 single-hole gas extraction pipeline comprehensive parameter measuring instruments, 12 single-hole gas pipeline electric valves, 1 manifold gas extraction pipeline comprehensive parameter measuring instrument, and 1 manifold electric valve; in order to achieve better data transmission and remote control, each drilling site is equipped with 6 mine-use intrinsically safe wireless gateways, 2 of which are used for wireless transmission of monitoring data, and 4 are used to control electric valves.

[0133] Combination Figure 7 The screenshot of the terminal interface of the borehole gas extraction online metering control system according to the embodiment of the present application shows that by monitoring the CH 4 concentration, CO concentration, O 2 Online measuring instruments for parameters such as concentration, temperature, pipeline pressure, ambient pressure, relative pressure, flow velocity and pipeline flow (mixed flow, pure flow) can realize accurate monitoring of gas drilling holes and high-position drilling sites.

[0134] Here, by deploying borehole pipeline monitoring points at high-level drilling sites, the gas extraction effect of the boreholes is analyzed and evaluated, and intelligent valves are deployed on the extraction pipelines to intelligently control the borehole extraction pipelines. Furthermore, unit measurement points and intelligent valves can be deployed at the high-level drilling site pipeline collection point to intelligently control the evaluation unit extraction pipelines.

[0135] For more details and technical effects of the online metering control method for borehole gas extraction provided in the embodiment of the present application, reference can be made to and combined with the description of the processing details of the online metering control system for borehole gas extraction in other embodiments above, and the corresponding technical effects can be achieved.

[0136] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0137] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, which stores one or more programs including execution instructions, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the steps of any of the above-mentioned online metering control methods for borehole gas extraction in the present application.

[0138] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the steps of any one of the above-mentioned online metering control methods for borehole gas extraction.

[0139] In some embodiments, the embodiments of the present application also provide an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the online metering control method for borehole gas extraction.

[0140] Figure 8 is a schematic diagram of the hardware structure of an electronic device for executing an online metering control method for drilling gas extraction provided by another embodiment of the present application, such as Figure 8 As shown, the device includes:

[0141] One or more processors 810 and memory 820, Figure 8 A processor 810 is taken as an example.

[0142] The device for executing the online metering control method for borehole gas extraction may further include: an input device 830 and an output device 840 .

[0143] The processor 810, the memory 820, the input device 830 and the output device 840 may be connected via a bus or other means. Figure 8The example of connecting through bus is taken in the following.

[0144] The memory 820, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the online metering control method for borehole gas extraction in the embodiment of the present application. The processor 810 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 820, that is, the online metering control method for borehole gas extraction in the above method embodiment is implemented.

[0145] The memory 820 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 820 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 820 may optionally include a memory remotely arranged relative to the processor 810, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0146] The input device 830 may receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 840 may include a display device such as a display screen.

[0147] The one or more modules are stored in the memory 820, and when executed by the one or more processors 810, the online metering control method for drilling gas extraction in any of the above method embodiments is executed.

[0148] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0149] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

[0150] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0151] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

[0152] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0153] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.

[0154] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An online metering control system for drilling gas extraction, comprising: A data acquisition unit, used to acquire the extraction pump control timing data and pipeline gas sensor timing data; The pipeline gas sensor time series data includes a plurality of gas sensor data corresponding to a first adjacent time period of a sampling position in a borehole gas extraction pipeline, and the gas sensor data includes gas flow, gas concentration, pressure and temperature; the extraction pump control time series data includes a plurality of extraction pump control parameters corresponding to the first time period, and the extraction pump control parameters include extraction pump speed and extraction valve opening; A flow prediction unit, used for inputting the extraction pump control time series data and the pipeline gas sensor time series data into the gas extraction flow prediction model to predict the gas flow prediction time series data corresponding to the second time period in the future; the backbone network of the gas extraction flow prediction model adopts an LSTM model; An extraction control unit is used to input the pipeline gas sensor time series data, the gas flow prediction time series data and the extraction pump control time series data into a gas extraction pressure control model to determine the extraction pump control parameter prediction time series data corresponding to the second time period; the gas extraction pressure control model adopts a Transformer model based on an attention mechanism.

2. The system according to claim 1, wherein: The system also includes a drainage volume monitoring unit, which is used to perform the following operations: A correction coefficient is calculated according to the pressure and temperature in each of the gas sensor data, and an actual gas extraction volume sequence corresponding to the first time period is calculated according to the correction coefficient and the gas flow and gas concentration in each of the gas sensor data.

3. The system according to claim 1, wherein: The actual gas extraction volume sequence is calculated by the following formula: V k (t)=Q(t-k·Δt)×C g (t-k·Δt)×C(t-k·Δ)×Δt, Where β1, β2 and β3 represent the corresponding calibration coefficients, which are preset according to the coal seam characteristics and experimental data; P(t) and θ(t) are the pressure and temperature sampled and detected at time step t; m represents the total number of historical time steps in the first time period relative to time step t; P0 and θ0 represent the standard pressure and standard temperature, respectively; C(t) represents the correction coefficient at time step t; V k (t) represents the actual gas extraction volume corresponding to the kth historical time step relative to time step t; w k Represents the weight coefficient of the kth historical time step, w0>w1>...>w m-1 , where w0 represents the weight coefficient of time step t; V(t) represents the actual gas extraction volume at time step t, Q(tk·αt) represents the gas flow rate at time step tk·Δt, and Δt represents the time interval between adjacent time steps; C(tk·Δt) represents the correction coefficient corresponding to time step tk·Δt, and Cg(tk·Δt) represents the gas concentration corresponding to time step tk·Δt.

4. The system according to claim 1, further comprising an alarm monitoring unit configured to: Monitoring whether the gas sensor data in the second time period meets a preset safety sensing range; When the monitored gas sensor data does not conform to the safety sensor interval, the extraction pump speed of the current borehole gas extraction pipeline is reduced and the extraction valve is gradually closed, and the extraction pump of the standby extraction pipeline is started to enter the standby state; at this time, the extraction valve of the standby extraction pipeline is in a closed state, and the sensor of the standby extraction pipeline is controlled to start collecting data; After the extraction valve of the current borehole gas extraction pipeline is completely closed, the extraction valve of the backup extraction pipeline is opened, and the extraction pump speed and valve opening of the backup extraction pipeline are adjusted to switch to the backup extraction pipeline to respond to the gas extraction task.

5. The system according to any one of claims 1 to 4, wherein: The gas extraction flow prediction model adopts a hybrid model including a SARIMA model module and a LSTM model module.

6. The system according to any one of claims 1 to 4, wherein: The gas extraction flow prediction model adopts a Bi-LSTM model based on an attention mechanism, which includes a cascaded TCN layer, a Bi-LSTM layer, an attention layer, and an output layer; The TCN layer is used to capture short-term and long-term dependencies in the extraction pump control time series data and the pipeline gas sensor time series data at different time scales through multiple convolutional layers: In the formula, Indicates the time step t of historical sampling p The multi-scale feature vector extracted by the TCN layer, TCN represents the feature processing process of the multi-scale temporal convolutional network; t p Represents the time step t p -n+1 to time step t p The input feature sequence is n, where n represents the time window length; The Bi-LSTM layer is used to capture the bidirectional dependencies of multi-scale feature vectors: In the formula, Indicates that at time step t p The forward LSTM hidden state of It is a forward LSTM unit, which is used to process the forward dependency of time series; Indicates that at the previous time step t p -1 forward LSTM hidden state; Indicates that at time step t p The backward LSTM hidden state; It is a backward LSTM unit, which is used to process the backward dependency of time series; is at the next time step t p +1 backward LSTM hidden state; Indicates that at time step t p The comprehensive hidden state of The attention layer uses an adaptive multi-head attention mechanism to dynamically adjust the weight of each time step: MultiHead(Q,K,V)=Concat(head1,...,head H )W A +b A , In the formula, Attention(Q,K,V) represents the output of the attention mechanism, Q, K and V represent The query vector, key vector and value vector obtained by linear transformation, d k represents the dimension of the key vector; Denotes the corresponding time step t p The adaptive bias term, softmax represents the softmax activation function; W b and b b Represent the weight matrix and bias term of the adaptive weight adjustment module respectively, Denotes the corresponding time step t p The input feature vector of head i represents the output of the i-th attention head, W i Q , W i K and W i V Represents the query transformation matrix, key transformation matrix and value transformation matrix of the i-th attention head respectively; Concat(head1,...,head H ) means concatenating the outputs of all H attention heads by dimension, and MultiHead(Q,K,V) means the corresponding time step t p The final output of the multi-head attention mechanism, W A and b A Represent the linear transformation weight matrix and bias vector of the multi-head attention output respectively; The output layer is used to predict the gas flow at each future time step: In the formula, represents the future time step t s The predicted value of gas flow, t s =t p +Δt, Δt means t s With t p The time interval between o and b o They represent the weight matrix and bias term of the linear output layer respectively.

7. The system according to claim 6, wherein: The gas extraction pressure control model comprises an input layer, an encoder layer, a decoder layer and an output layer; The input layer is used to receive the pipeline gas sensor time series data, the gas flow prediction time series data and the extraction pump control time series data, and after linear embedding and position encoding, a comprehensive sequence representation containing time and feature information is formed: In the formula, Represents the input information of the t1th time step corresponding to the sampling detection in the input data, and its dimension is d input ; W emb Represents a linear embedding matrix with dimension d model ×d input ; Represents the input feature representation at the t1th time step, with a dimension of d model ; Represents the position encoding vector corresponding to the t1th time step, which is used to inject position information in the time series into the model, with a dimension of d model ; j represents the dimension index of the position encoding; The encoder layer is used to process the comprehensive sequence representation to capture the dependency between each time step through the self-attention mechanism, thereby generating a corresponding context feature vector; wherein the encoder layer includes a plurality of encoding units, each of which includes a cascaded self-attention mechanism layer and a feedforward neural network layer; the feature processing process of the encoding unit is expressed by the following formula: In the formula, is the output feature vector calculated by the encoding unit for the input information at the t1th time step, representing the context information after being processed by the self-attention mechanism and the feedforward neural network; represents the attention weight of the historical sampling time step t1 to the time step τ, z τ represents the input feature representation at the τth time step, T1 represents the total number of time steps corresponding to the input data; W enc1 ,W enc2 represents the weight matrix of the feedforward neural network layer in the encoding unit, b enc1 ,b enc2 Represents the bias term of the feedforward neural network layer in the encoding unit; ReLU represents the RELU nonlinear activation function; The decoder layer is used to process the context feature vector to generate a control parameter feature sequence corresponding to the prediction of the future time step through a multi-head self-attention mechanism combined with a mask mechanism; the decoder layer includes a plurality of decoding units, each of which includes a cascaded self-attention mechanism layer based on a mask mechanism and a feedforward neural network layer; the feature processing process of the decoding unit is expressed by the following formula: In the formula, represents the control parameter features generated by the decoding unit at the future time step t2; h enc,ξ represents the context feature vector calculated by the encoder layer for the input information of the ξ-th time step of the historical sampling; represents the dependency of the decoder at time step t2 on time step ξ, which is the attention weight obtained by the multi-head self-attention mechanism; W dec1 ,W dec2 represents the weight matrix of the feedforward neural network layer in the decoding unit, b dec1 ,b dec2 represents the bias term of the feedforward neural network layer in the decoding unit; The output layer is used to perform linear layer processing on the predicted control parameter feature sequence to output predicted values ​​of the drainage pump control parameters at each time step in the second time period in the future: In the formula, y dec,t represents the control parameter features generated by the decoder layer for the future time step t2; They represent the pump speed and valve opening predicted at the corresponding time step t2 respectively; W rpm ,W valve represents the linear transformation matrix of the output layer, b rpm ,b valve Represents the bias term of the output layer.

8. The system according to claim 7, wherein: The loss function L of the gas extraction pressure control model is total It is expressed by the following formula: L total =λ1L MSE +λ2L smooth , Where, L MSE and L smooth They represent the prediction error loss term and the change rate constraint loss term respectively, λ1 and λ2 represent the weight coefficients of the corresponding loss terms respectively; N represents the total number of samples in the data sample set, T g represents the total number of predicted time steps corresponding to the u-th sample, and Respectively represent the u-th sample in the future time step t g and t g -1 is the predicted value of the pump speed; Represents the u-th sample at the future time step t g The extraction pump speed label; and Respectively represent the u-th sample in the future time step t g and t g -1 is the predicted value of valve opening; Represents the u-th sample at the future time step t g Valve opening label.

Images