A grid-based directional laser methane detection system

Abstract

The present invention relates to the field of gas leak detection, and discloses a gridded directional laser methane detection system, comprising the following steps: collecting regional information, the regional information including methane absorption signals, environmental data, background gas, regional images, and historical leakage data; generating No. 1 two-dimensional structure data from the collected regional information, the No. 1 two-dimensional structure data including a data matrix and a relationship matrix; obtaining No. 1 one-dimensional structure data based on sorting the No. 1 two-dimensional structure data, the No. 1 one-dimensional structure data including n data items sorted by time; inputting the No. 1 two-dimensional structure data into a detection model, and outputting a processing strategy for methane leaks. The present invention can realize methane detection and processing strategies in gas, and can also provide a further distribution strategy for methane laser detection devices.

Description

Technical Field

[0001] The invention relates to the technical field of gas leakage detection, and more specifically, to a grid directional laser methane detection system. Background Art

[0002] With the continuous advancement of industrialization, methane leaks occur frequently, which has a serious impact on people's production and life. Traditional methane leak detection methods, such as manual inspections and fixed-point monitoring, have shortcomings such as low efficiency and limited coverage, and are difficult to meet the growing detection needs.

[0003] In recent years, with the rapid development of technologies such as the Internet of Things, big data, and artificial intelligence, intelligent methane leak detection methods have received widespread attention. The general method achieves real-time monitoring of the target area by deploying a large number of directional laser methane sensors. On the one hand, the installation of laser methane sensors is costly and cannot achieve coverage of all pipeline nodes. On the other hand, leaks are now detected by judging whether the detection value exceeds the threshold. When the laser methane sensor is interfered with, it will produce false alarms. Summary of the invention

[0004] The present invention provides a grid-based directional laser methane detection system, which can comprehensively collect regional information, effectively structure the data in the regional information, and comprehensively characterize the feature information of the target area by collecting multi-source heterogeneous data, thereby improving the accuracy and reliability of detection; by introducing a data matrix and a relationship matrix, the correlation information between different data is fully utilized, and the data representation capability is enhanced; by using a deep learning model, the high-level feature representation of the data is automatically learned, thereby avoiding the limitations of artificial feature design; by introducing a simulation detection device, the performance of the detection model is optimized, and the real-time and accuracy of the detection are improved; by outputting the layout strategy of the laser detection device, the actual detection work is guided, and the efficiency and operability of the detection are improved.

[0005] The present invention provides a grid-based directional laser methane detection system, comprising the following modules:

[0006] Information collection module, which collects regional information, including methane absorption signals, environmental data, background gas, regional images, and historical leakage data;

[0007] The first data module generates a first two-dimensional structure data from the collected regional information. The first two-dimensional structure data includes a data matrix and a relationship matrix. A unit of the data matrix represents the data of an independent object. The independent object includes a methane absorption signal, environmental data, background gas, regional image, and historical leakage data.

[0008] The second data module obtains the first one-dimensional structure data based on the sorting of the first two-dimensional structure data, wherein the first one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the first two-dimensional structure data generated by the regional information collected at the t-th time;

[0009] The methane detection module inputs the one-dimensional structure data No. 1 into the detection model, and the detection model includes a first intermediate layer, a second intermediate layer, and a first output layer, wherein the first intermediate layer inputs the two-dimensional structure data No. 1 and outputs the first intermediate representation data to the second intermediate layer; the second intermediate layer outputs the second intermediate representation data to the first output layer, and outputs the processing strategy for methane leakage;

[0010] The third data module obtains distribution data, which is obtained by splicing the methane leakage processing strategy output by the first output layer and the simulated methane leakage processing strategy obtained by the simulation detection device; and obtains regional image data in the regional information;

[0011] The device distribution module inputs the distribution data and the regional image data into the distribution model. The distribution model includes a third intermediate layer, a fourth intermediate layer, a feature fusion layer, and a second output layer. The third intermediate layer inputs the distribution data and outputs the third intermediate representation data to the feature fusion layer. The fourth intermediate layer inputs the regional image data and outputs the fourth intermediate representation data to the feature fusion layer. The feature fusion layer outputs the first fusion representation data to the second output layer, and outputs the distribution strategy of the laser detection device.

[0012] In a preferred embodiment, the calculation formula of the first intermediate layer is as follows:

[0013] ;

[0014] in The first unit of the vth unit of the data matrix represents the data in the middle, represents the aggregation coefficient of the vth cell of the data matrix, represents the set of cells of the data matrix that are associated with the vth cell of the data matrix, represents the sigmoid activation function, represents the state weight coefficient, Represents the region information of the independent object represented by the u-th unit of the data matrix;

[0015] ;

[0016] ;

[0017] ;

[0018] in and Respectively represent the region information of the independent objects represented by the vth and uth cells of the data matrix, and Respectively represent the linear transformation features of the vth and uth units in the data matrix, represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transposition, represents an exponential function with a natural constant as the base, LeakyRelu represents a modified linear unit function, represents the set of cells of the data matrix that are associated with the vth cell of the data matrix;

[0019] ;

[0020] in represents the first intermediate representation data of the output, represents the vth unit of the data matrix, M represents the set of all units of the data matrix, The first unit of the vth unit of the data matrix represents the data in the middle, is the first weight parameter, is the first bias parameter, Represents the sigmoid activation function.

[0021] In a preferred embodiment, the calculation formula of the second intermediate layer is as follows:

[0022] ;

[0023] in represents the t-th second intermediate representation data, represents the t-1th second intermediate representation data, , n represents the total number of data items of the one-dimensional structure data, It represents the first intermediate representation data output when the t-th sequence unit inputs the first intermediate layer. and are the second and third weight parameters, is the second bias parameter and tanh is the hyperbolic tangent function.

[0024] In a preferred embodiment, the calculation formula of the first output layer is as follows:

[0025] ;

[0026] in represents the first output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy represents a vector. The fth component of the vector contains three sub-vectors. The first sub-vector represents the location coordinates of the methane leak, the second sub-vector represents the methane concentration at the leak, and the third sub-vector represents the gas valve control at the leak. Valve control means opening or closing the valve. represents the nth second intermediate representation data, is the first output weight parameter, is the first output bias parameter, Represents the sigmoid activation function.

[0027] In a preferred embodiment, only the first output layer is used during the detection model training, and the training steps include:

[0028] Step 101, initializing the parameters of the detection model;

[0029] Step 102: Observe the regional information at time i , the strategy executed at time i , regional information at time i+1 , execution strategy Rewards ;

[0030] Step 103, then calculate the policy error:

[0031] ;

[0032] in represents the strategy error at time i, represents the discount factor, , Represents the detection model input The maximum probability value in the first output vector output when Represents the detection model input The first output vector corresponding to the strategy The probability value of

[0033] Reward Function It is expressed as follows:

[0034] ;

[0035] in represents the number of locations where methane leaks were correctly detected, represents the number of locations where no methane leaks were correctly detected, It indicates the number of locations where no methane leaks were misjudged as having methane leaks. Indicates the number of locations where methane leaks were misjudged as non-methane leaks;

[0036] Step 104, update the detection model, the updated formula is as follows:

[0037] ;

[0038] , represents the step size of deep learning, Indicates delivery update;

[0039] Step 105, iterate steps 102-104 until the detection model converges or the number of iterations reaches a set value, the default value of which is 30.

[0040] In a preferred embodiment, the calculation formula of the third intermediate layer is as follows:

[0041] ;

[0042] in represents the third intermediate representation data, represents the first convolution function, Represents distribution data.

[0043] In a preferred embodiment, the calculation formula of the fourth intermediate layer is as follows:

[0044] ;

[0045] in represents the fourth intermediate representation data, represents the second convolution function, Represents area image data.

[0046] In a preferred embodiment, the calculation formula of the feature fusion layer is as follows:

[0047] ;

[0048] in represents the first fusion representation data, represents the concatenation function;

[0049] The calculation formula of the second output layer is as follows:

[0050] ;

[0051] in Represents the second output vector, in which the rth component represents the probability value of the rth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy represents a vector. A vector contains two sub-vectors. The first sub-vector represents the position coordinates of the laser detection device, and the second sub-vector represents the deflection amount of the laser detection device direction. The deflection amount represents the angle of deflection. is the second output weight parameter, is the second output bias parameter, Represents the sigmoid activation function.

[0052] In a preferred embodiment, only the second output layer is used during distribution model training, and the training steps include:

[0053] Step 201, initializing the parameters of the distribution model;

[0054] Step 202: Observe the distribution data at time e , the strategy executed at time e , distribution data at time e+1 , execution strategy Rewards ;

[0055] Step 203, then calculate the policy error:

[0056] ;

[0057] in represents the policy error at time e, represents the discount factor, , Represents the distribution model input The maximum probability value in the second output vector output when Represents the distribution model input The first output vector corresponding to the strategy The probability value of

[0058] Reward Function It is expressed as follows:

[0059] ;

[0060] in and represents the location coordinates of the methane leak predicted by the model, and represents the location coordinates of the methane leak obtained by simulation, represents the methane leakage concentration predicted by the model, represents the simulated methane leakage concentration, and represents the weight parameter, The default value of is 0.5. The default value of is 0.5;

[0061] Step 204, update the distribution model, the updated formula is as follows:

[0062] ;

[0063] , represents the step size of deep learning, Indicates delivery update;

[0064] Step 205, iterate steps 202-204 until the distribution model converges or the number of iterations reaches a set value, the default value of which is 30.

[0065] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the modules in the gridded directional laser methane detection system as described above can be executed.

[0066] The beneficial effects of the present invention are:

[0067] The present invention performs effective structural processing and data fusion on the data in the regional information, and utilizes limited laser detection devices to monitor leakage of the entire pipe network.

[0068] The present invention proposes a multimodal data fusion method, which makes full use of heterogeneous data collected by different sensors and comprehensively describes the characteristic information of methane leakage. By introducing the attention mechanism and spatiotemporal correlation analysis, the method can adaptively adjust the weights between different data, highlight the role of key data, suppress the interference of redundant data, and improve the effectiveness and accuracy of data fusion. Compared with a single data source, the multimodal data fusion method can provide a more comprehensive and reliable decision-making basis, reduce the risk of missed detection and false detection, and thus improve the accuracy and reliability of methane leakage detection.

[0069] The present invention adopts a multi-task learning strategy to simultaneously complete multiple related tasks such as methane concentration prediction, leakage source location and leakage estimation. By sharing feature representations between different tasks, multi-task learning can utilize the correlation between tasks to improve the learning efficiency and generalization ability of the model. At the same time, multi-task learning can provide more comprehensive analysis results, which can not only detect the occurrence of methane leakage, but also locate the location of the leakage source and estimate the magnitude of the leakage, providing important decision support for accident handling and emergency response. In addition, multi-task learning can also reduce the training and deployment costs of the model, simplify the maintenance and upgrade process of the system, and has good engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a module diagram of a grid-based directional laser methane detection system of the present invention. DETAILED DESCRIPTION

[0071] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0072] At least one embodiment of the present invention discloses a grid-based directional laser methane detection system, such as Figure 1 As shown, it includes the following modules:

[0073] The information collection module 100 collects regional information, which includes methane absorption signals, environmental data, background gas, regional images, and historical leakage data;

[0074] The methane absorption signal includes the time series data of the signal's absorption intensity, wavelength, line width, and line type, which is acquired by a gridded laser detection device;

[0075] Environmental data includes temperature, humidity, and air pressure, which are obtained by environmental sensors;

[0076] Background gas includes the concentration and coverage area of ​​background gas, which are obtained by the spectrum analysis module;

[0077] The regional image includes a high-definition image of the detection area and the surrounding environment, obtained by the detection camera;

[0078] Historical leakage data includes the location, cause, and environmental information of methane leakage, which is obtained from historical record information;

[0079] The first data module 200 generates a first two-dimensional structure data from the collected regional information. The first two-dimensional structure data includes a data matrix and a relationship matrix. A unit of the data matrix represents data of an independent object. The independent object includes a methane absorption signal, environmental data, background gas, regional image, and historical leakage data. A unit of the data matrix representing the methane absorption signal only contains information about the methane absorption signal. A unit of the data matrix representing the environmental data only contains information about the environmental data. A unit of the data matrix representing the background gas only contains information about the background gas. A unit of the data matrix representing the regional image only contains information about the regional image. A unit of the data matrix representing the historical leakage data only contains historical leakage description text.

[0080] The element in the sth row and jth column of the relationship matrix represents the association between the sth unit of the data matrix and the independent object represented by the jth unit. If there is an association, the value of this element of the relationship matrix is ​​1, otherwise it is 0;

[0081] The correlation between the methane absorption signal and the environmental data means that: the methane absorption signal is affected by the environmental data;

[0082] The correlation between the methane absorption signal and the background gas means that the methane absorption signal is affected by the background gas;

[0083] The association between historical leakage data and environmental data refers to: historical leakage data contains environmental data information at the time of leakage;

[0084] The association between historical leakage data and background gas means that: historical leakage data contains background gas information at the time of leakage;

[0085] The association between the regional image and the historical leakage data means that: the location in the historical leakage data belongs to the regional image;

[0086] In one embodiment of the present invention, the methane absorption signal, environmental data, background gas, regional image, and historical leakage data are input into the model after feature engineering;

[0087] The methane absorption signal is time series data, and the 1D-CNN method is used as a feature engineering method;

[0088] Environmental data and background gases are structured data, and the MLP method is used as a feature engineering method;

[0089] The regional image is image data, and the CNN method is used as a feature engineering method;

[0090] The historical leak description text is text data, and the WordEmbedding method is used as a feature engineering method;

[0091] The second data module 300 obtains the first one-dimensional structure data based on the sorting of the first two-dimensional structure data, wherein the first one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the first two-dimensional structure data generated by the regional information collected at the t-th time;

[0092] The methane detection module 400 inputs the one-dimensional structure data No. 1 into the detection model, and the detection model includes a first intermediate layer, a second intermediate layer, and a first output layer, wherein the first intermediate layer inputs the two-dimensional structure data No. 1 and outputs the first intermediate representation data to the second intermediate layer; the second intermediate layer outputs the second intermediate representation data to the first output layer, and outputs the processing strategy for methane leakage;

[0093] In one embodiment of the present invention, the calculation formula of the first intermediate layer is as follows:

[0094] ;

[0095] in The first unit of the vth unit of the data matrix represents the data in the middle, represents the aggregation coefficient of the vth cell of the data matrix, represents the set of cells of the data matrix that are associated with the vth cell of the data matrix, represents the sigmoid activation function, represents the state weight coefficient, Represents the region information of the independent object represented by the u-th unit of the data matrix;

[0096] ;

[0097] ;

[0098] ;

[0099] in and Respectively represent the region information of the independent objects represented by the vth and uth cells of the data matrix, and Respectively represent the linear transformation features of the vth and uth units in the data matrix, represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transposition, represents an exponential function with a natural constant as the base, LeakyRelu represents a modified linear unit function, represents the set of cells of the data matrix that are associated with the vth cell of the data matrix;

[0100] ;

[0101] in represents the first intermediate representation data of the output, represents the vth unit of the data matrix, M represents the set of all units of the data matrix, The first unit of the vth unit of the data matrix represents the data in the middle, is the first weight parameter, is the first bias parameter, Represents the sigmoid activation function.

[0102] In one embodiment of the present invention, the calculation formula of the second intermediate layer is as follows:

[0103] ;

[0104] in represents the t-th second intermediate representation data, represents the t-1th second intermediate representation data, , n represents the total number of data items of the one-dimensional structure data, It represents the first intermediate representation data output when the t-th sequence unit inputs the first intermediate layer. and are the second and third weight parameters, is the second bias parameter and tanh is the hyperbolic tangent function.

[0105] In one embodiment of the present invention, the calculation formula of the first output layer is as follows:

[0106] ;

[0107] in represents the first output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy represents a vector. The fth component of the vector contains three sub-vectors. The first sub-vector represents the location coordinates of the methane leak, the second sub-vector represents the methane concentration at the leak, and the third sub-vector represents the gas valve control at the leak. Valve control means opening or closing the valve. represents the nth second intermediate representation data, is the first output weight parameter, is the first output bias parameter, Represents the sigmoid activation function.

[0108] In one embodiment of the present invention, a complete methane detection system is simulated according to current regional information using a methane simulation detection device, and a simulated methane leakage processing strategy is obtained in the simulated methane detection system.

[0109] In one embodiment of the present invention, only the first output layer is used during the detection model training, and the training steps include:

[0110] Step 101, initializing the parameters of the detection model;

[0111] Step 102: Observe the regional information at time i , the strategy executed at time i , regional information at time i+1 , execution strategy Rewards ;

[0112] Step 103, then calculate the policy error:

[0113] ;

[0114] in represents the strategy error at time i, represents the discount factor, , Represents the detection model input The maximum probability value in the first output vector output when Represents the detection model input The first output vector corresponding to the strategy The probability value of

[0115] Reward Function It is expressed as follows:

[0116] ;

[0117] in represents the number of locations where methane leaks were correctly detected, represents the number of locations where no methane leaks were correctly detected, It indicates the number of locations where no methane leaks were misjudged as having methane leaks. Indicates the number of locations where methane leaks were misjudged as non-methane leaks;

[0118] Step 104, update the detection model, the updated formula is as follows:

[0119] ;

[0120] , represents the step size of deep learning, Indicates delivery update;

[0121] Step 105, iterate steps 102-104 until the detection model converges or the number of iterations reaches a set value, the default value of which is 30.

[0122] The third data module 500 obtains distribution data, which is obtained by splicing the methane leakage processing strategy output by the first output layer and the simulated methane leakage processing strategy obtained by the simulation detection device; obtains regional image data in the regional information;

[0123] The device distribution module 600 inputs the distribution data and the regional image data into the distribution model, the distribution model includes a third intermediate layer, a fourth intermediate layer, a feature fusion layer, and a second output layer, the third intermediate layer inputs the distribution data, outputs the third intermediate representation data to the feature fusion layer, the fourth intermediate layer inputs the regional image data, outputs the fourth intermediate representation data to the feature fusion layer, the feature fusion layer outputs the first fusion representation data to the second output layer, and outputs the distribution strategy of the laser detection device;

[0124] In one embodiment of the present invention, the calculation formula of the third intermediate layer is as follows:

[0125] ;

[0126] in represents the third intermediate representation data, represents the first convolution function, Represents distribution data.

[0127] In one embodiment of the present invention, the calculation formula of the fourth intermediate layer is as follows:

[0128] ;

[0129] in represents the fourth intermediate representation data, represents the second convolution function, Represents area image data.

[0130] In one embodiment of the present invention, the calculation formula of the feature fusion layer is as follows:

[0131] ;

[0132] in represents the first fusion representation data, Represents the concatenation function.

[0133] In one embodiment of the present invention, the calculation formula of the second output layer is as follows:

[0134] ;

[0135] in Represents the second output vector, in which the rth component represents the probability value of the rth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy represents a vector. A vector contains two sub-vectors. The first sub-vector represents the position coordinates of the laser detection device, and the second sub-vector represents the deflection amount of the laser detection device direction. The deflection amount represents the angle of deflection. is the second output weight parameter, is the second output bias parameter, Represents the sigmoid activation function.

[0136] In one embodiment of the present invention, only the second output layer is used during distribution model training, and the training steps include:

[0137] Step 201, initializing the parameters of the distribution model;

[0138] Step 202: Observe the distribution data at time e , the strategy executed at time e , distribution data at time e+1 , execution strategy Rewards ;

[0139] Step 203, then calculate the policy error:

[0140] ;

[0141] in represents the policy error at time e, represents the discount factor, , Represents the distribution model input The maximum probability value in the second output vector output when Represents the distribution model input The first output vector corresponding to the strategy The probability value of

[0142] Reward Function It is expressed as follows:

[0143] ;

[0144] in and represents the location coordinates of the methane leak predicted by the model, and represents the location coordinates of the methane leak obtained by simulation, represents the methane leakage concentration predicted by the model, represents the simulated methane leakage concentration, and represents the weight parameter, The default value of is 0.5. The default value of is 0.5;

[0145] Step 204, update the distribution model, the updated formula is as follows:

[0146] ;

[0147] , represents the step size of deep learning, Indicates delivery update;

[0148] Step 205, iterate steps 202-204 until the distribution model converges or the number of iterations reaches a set value, the default value of which is 30.

[0149] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the modules in the gridded directional laser methane detection system as described above can be executed.

[0150] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A grid-based directional laser methane detection system, characterized in that: Includes the following modules: Information collection module, which collects regional information, including methane absorption signals, environmental data, background gas, regional images, and historical leakage data; The first data module generates a first two-dimensional structure data from the collected regional information. The first two-dimensional structure data includes a data matrix and a relationship matrix. A unit of the data matrix represents the data of an independent object. The independent object includes a methane absorption signal, environmental data, background gas, regional image, and historical leakage data. The second data module obtains the first one-dimensional structure data based on the sorting of the first two-dimensional structure data, wherein the first one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the first two-dimensional structure data generated by the regional information collected at the t-th time; The methane detection module inputs the No. 1 one-dimensional structure data into the detection model, the detection model includes a first intermediate layer, a second intermediate layer, and a first output layer, wherein the first intermediate layer inputs the No. 1 two-dimensional structure data and outputs the first intermediate representation data to the second intermediate layer; The second intermediate layer outputs the second intermediate representation data to the first output layer, and outputs the processing strategy for methane leakage; The third data module obtains distribution data, which is obtained by splicing the methane leakage processing strategy output by the first output layer and the simulated methane leakage processing strategy obtained by the simulation detection device; and obtains regional image data in the regional information; The device distribution module inputs the distribution data and the regional image data into the distribution model. The distribution model includes a third intermediate layer, a fourth intermediate layer, a feature fusion layer, and a second output layer. The third intermediate layer inputs the distribution data and outputs the third intermediate representation data to the feature fusion layer. The fourth intermediate layer inputs the regional image data and outputs the fourth intermediate representation data to the feature fusion layer. The feature fusion layer outputs the first fusion representation data to the second output layer, and outputs the distribution strategy of the laser detection device.

2. A grid-based directional laser methane detection system according to claim 1, characterized in that: The calculation formula for the first intermediate layer is as follows: ； in The first unit of the vth unit of the data matrix represents the data in the middle, represents the aggregation coefficient of the vth cell of the data matrix, represents the set of cells of the data matrix that are associated with the vth cell of the data matrix, represents the sigmoid activation function, represents the state weight coefficient, Represents the region information of the independent object represented by the u-th unit of the data matrix; ； ； ； in and Respectively represent the region information of the independent objects represented by the vth and uth cells of the data matrix, and Respectively represent the linear transformation features of the vth and uth units in the data matrix, represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transposition, represents an exponential function with a natural constant as the base, LeakyRelu represents a modified linear unit function, represents the set of cells of the data matrix that are associated with the vth cell of the data matrix; ； in represents the first intermediate representation data of the output, represents the vth unit of the data matrix, M represents the set of all units of the data matrix, The first unit of the vth unit of the data matrix represents the data in the middle, is the first weight parameter, is the first bias parameter, Represents the sigmoid activation function.

3. A grid-based directional laser methane detection system according to claim 2, characterized in that: The calculation formula for the second intermediate layer is as follows: ； in represents the t-th second intermediate representation data, represents the t-1th second intermediate representation data, , n represents the total number of data items of the one-dimensional structure data, It represents the first intermediate representation data output when the t-th sequence unit inputs the first intermediate layer. and are the second and third weight parameters, is the second bias parameter and tanh is the hyperbolic tangent function.

4. A grid-based directional laser methane detection system according to claim 3, characterized in that: The calculation formula of the first output layer is as follows: ； in represents the first output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy represents a vector. The fth component of the vector contains three sub-vectors. The first sub-vector represents the location coordinates of the methane leak, the second sub-vector represents the methane concentration at the leak, and the third sub-vector represents the gas valve control at the leak. Valve control means opening or closing the valve. represents the nth second intermediate representation data, is the first output weight parameter, is the first output bias parameter, Represents the sigmoid activation function.

5. The grid-based directional laser methane detection system according to claim 1, characterized in that: Only the first output layer is used when training the detection model. The training steps include: Step 101, initializing the parameters of the detection model; Step 102: Observe the regional information at time i , the strategy executed at time i , regional information at time i+1 , execution strategy Rewards ; Step 103, then calculate the policy error: ； in represents the strategy error at time i, represents the discount factor, , Represents the detection model input The maximum probability value in the first output vector output when Represents the detection model input The first output vector corresponding to the strategy The probability value of Reward Function It is expressed as follows: ； in represents the number of locations where methane leaks were correctly detected, represents the number of locations where no methane leaks were correctly detected, It indicates the number of locations where no methane leaks were misjudged as having methane leaks. Indicates the number of locations where methane leaks were misjudged as non-methane leaks; Step 104, update the detection model, the updated formula is as follows: ； , represents the step size of deep learning, Indicates delivery update; Step 105, iterate steps 102-104 until the detection model converges or the number of iterations reaches a set value, the default value of which is 30.

6. The grid-based directional laser methane detection system according to claim 1, characterized in that: The calculation formula for the third intermediate layer is as follows: ； in represents the third intermediate representation data, represents the first convolution function, Represents distribution data.

7. The grid-based directional laser methane detection system according to claim 1, characterized in that: The calculation formula for the fourth intermediate layer is as follows: ； in represents the fourth intermediate representation data, represents the second convolution function, Represents area image data.

8. The grid-based directional laser methane detection system according to claim 1, characterized in that: The calculation formula of the feature fusion layer is as follows: ； in represents the first fusion representation data, represents the concatenation function; The calculation formula of the second output layer is as follows: ； in Represents the second output vector, in which the rth component represents the probability value of the rth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy represents a vector. A vector contains two sub-vectors. The first sub-vector represents the position coordinates of the laser detection device, and the second sub-vector represents the deflection amount of the laser detection device direction. The deflection amount represents the angle of deflection. is the second output weight parameter, is the second output bias parameter, Represents the sigmoid activation function.

9. The grid-based directional laser methane detection system according to claim 1, characterized in that: Only the second output layer is used when training the distribution model. The training steps include: Step 201, initializing the parameters of the distribution model; Step 202: Observe the distribution data at time e , the strategy executed at time e , distribution data at time e+1 , execution strategy Rewards ; Step 203, then calculate the policy error: ； in represents the policy error at time e, represents the discount factor, , Represents the distribution model input The maximum probability value in the second output vector output when Represents the distribution model input The first output vector corresponding to the strategy The probability value of Reward Function It is expressed as follows: ； in and represents the location coordinates of the methane leak predicted by the model, and represents the location coordinates of the methane leak obtained by simulation, represents the methane leakage concentration predicted by the model, represents the simulated methane leakage concentration, and represents the weight parameter, The default value of is 0.

5. The default value of is 0.5; Step 204, update the distribution model, the updated formula is as follows: ； , represents the step size of deep learning, Indicates delivery update; Step 205, iterate steps 202-204 until the distribution model converges or the number of iterations reaches a set value, the default value of which is 30.

10. A storage medium storing non-transitory computer-readable instructions, characterized in that: When the non-transitory computer-readable instructions are executed by a computer, the steps in the gridded directional laser methane detection system described in any one of claims 1-9 can be executed.

Images