Intelligent tunnel gas concentration prediction system and method based on deep learning

Abstract

The application discloses a tunnel gas concentration intelligent prediction system and method based on deep learning, belongs to the technical field of tunnel gas concentration prediction, can effectively capture the dramatic abnormal fluctuation of gas concentration caused by geological structure mutation, and improves the prediction accuracy; the method comprises the following steps: collecting original gas concentration sequences of each monitoring point in a tunnel and geological structure mutation parameters; dynamically adjusting the length of a time window according to the geological structure mutation parameters, adaptively fragmenting the original sequences, and generating variable-length concentration segmented sequences; merging the variable-length concentration segmented sequences into a multi-monitoring-point concentration segmented sequence matrix; constructing a dynamic weight gas concentration spatial relationship fusion graph based on the spatial positions of the monitoring points and the distribution of the geological structure mutation zones; inputting a trained graph neural network to extract dynamic spatial relationship representation vectors of the monitoring points; inputting a trained recurrent neural network encoder again to generate space-time features containing the influence of geological mutations; and finally, outputting the gas concentration in the prediction period through variable-length prediction decoding.

Description

Technical Field

[0001] This invention relates to the field of tunnel gas concentration prediction technology, specifically a tunnel gas concentration intelligent prediction system and method based on deep learning. Background Technology

[0002] In tunnel construction, methane concentration prediction is a core technical aspect of ensuring construction safety. Among related technologies, prediction methods based on time series analysis or traditional machine learning are mainstream. These methods typically employ a fixed time window segmentation mechanism to preprocess monitoring data, dividing the continuous methane concentration sequence into equal-length time intervals, and then using models to learn the concentration variation patterns within each segment. However, in long-distance tunnel construction, when drilling and blasting operations or geological exploration trigger abrupt geological changes, the stress field of the tunnel surrounding rock and the methane occurrence state can undergo nonlinear abrupt changes, leading to drastic and abnormal fluctuations in methane concentration within a short period.

[0003] The gas concentration prediction method using a fixed time window segmentation mechanism cannot effectively capture the abnormal concentration fluctuations caused by sudden changes, resulting in a lag in prediction results.

[0004] The above background information is provided only to assist in understanding the concept and technical solution of this invention. It does not necessarily belong to the prior art of this application. In the absence of clear evidence that the above information was disclosed on the filing date of this application, the above background information should not be used to evaluate the novelty and inventiveness of this application. Summary of the Invention

[0005] This application provides a deep learning-based intelligent prediction system and method for tunnel gas concentration, which can effectively capture drastic and abnormal fluctuations in gas concentration caused by abrupt changes in geological structure, improve prediction accuracy, and ensure the safety of tunnel construction.

[0006] To achieve the above objectives, the embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, embodiments of this application provide a deep learning-based intelligent prediction method for tunnel gas concentration, including:

[0008] The original gas concentration sequence and geological structural abrupt change parameters of each monitoring point in the tunnel were collected.

[0009] The time window length is dynamically adjusted based on geological structural mutation parameters to adaptively segment the original gas concentration sequence and generate a variable-length concentration segment sequence.

[0010] The variable-length concentration segmented sequences were merged into a multi-monitoring-point concentration segmented sequence matrix;

[0011] Based on the spatial location of monitoring points and the distribution of geological structural abrupt change zones, a dynamic weighted spatial relationship fusion map of gas concentration is constructed.

[0012] The dynamic weight fusion graph is input into the trained graph neural network to extract the dynamic spatial relationship representation vector of each monitoring point.

[0013] The dynamic spatial relationship representation vector is input into the trained recurrent neural network encoder to generate spatiotemporal features that include the impact of geological mutations;

[0014] A segmented attention network is used to perform variable-length prediction decoding of spatiotemporal features, and the gas concentration for the predicted time period is output.

[0015] In this embodiment, based on the original gas concentration sequence and geological structural abrupt change parameters, the collected data is adaptively segmented using a dynamically adjusted time window length. The segmentation window is adjusted using the geological abrupt change parameters reflected in the input, resulting in a variable-length concentration segment sequence. This ensures that information on instantaneous and drastic fluctuations during nonlinear abrupt changes caused by drilling and blasting operations or geological exploration during tunnel construction is fully preserved. After matrix construction, the monitoring point data directly reflects the physical distance between monitoring points and the local impact caused by the distribution of geological structural abrupt change zones in the spatial information. The use of a dynamically weighted reconstructed graph gives the spatial relationships a true and accurate physical meaning. The spatial fusion graph is input into a graph neural network, and the network's multi-layer information aggregation operation is used to extract a deep representation of the complex dynamic changes between monitoring points, thereby obtaining a vector expression that finely reflects the spatial characteristics of gas concentration.

[0016] Subsequently, a recurrent neural network was selected to encode spatiotemporal features and generate a temporal expression that incorporates the influence of geological mutations. A segmented attention mechanism was used to perform variable-length prediction of spatiotemporal features in the decoding stage, and the gas concentration data for the predicted time period was output to achieve accurate estimation of short-term drastic fluctuation trends.

[0017] In summary, this scheme accurately captures localized and drastic fluctuations caused by nonlinear mutations during the data preprocessing stage using dynamic slicing, fully reflects the geological anomaly effects between monitoring points during the spatial modeling stage using dynamic weighted reconstruction, and achieves effective fusion of spatiotemporal features during the depth feature extraction and prediction stage, ultimately forming a real-time and accurate gas concentration prediction. This solves the problem that traditional fixed-time-window methods cannot capture abnormal concentration fluctuations caused by mutations in a timely manner during tunnel construction.

[0018] Secondly, embodiments of this application provide a deep learning-based intelligent prediction system for tunnel gas concentration, comprising:

[0019] The data acquisition module is used to collect the original gas concentration sequence and geological structural change parameters of each monitoring point in the tunnel;

[0020] The adaptive slicing module is used to dynamically adjust the time window length according to geological structural change parameters, adaptively slice the original gas concentration sequence, and generate a variable-length concentration segment sequence.

[0021] The sequence matrix merging module is used to merge variable-length concentration segmented sequences into a multi-monitoring-point concentration segmented sequence matrix;

[0022] The spatial relationship fusion map construction module is used to construct a dynamic weighted gas concentration spatial relationship fusion map based on the spatial location of monitoring points and the distribution of geological structural abrupt change zones;

[0023] The spatial feature extraction module is used to input the dynamic weight fusion graph into the trained graph neural network and extract the dynamic spatial relationship representation vector of each monitoring point.

[0024] The spatiotemporal feature generation module is used to input the dynamic spatial relationship representation vector into the trained recurrent neural network encoder to generate spatiotemporal features that include the impact of geological mutations.

[0025] The variable-length prediction decoding module is used to perform variable-length prediction decoding on spatiotemporal features through a segmented attention network, and output the gas concentration for the predicted time period.

[0026] Thirdly, embodiments of this application provide an electronic device, including one or more processors; a storage device having one or more programs stored thereon; when the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in any of the technical solutions of the first aspect.

[0027] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method described in any of the technical solutions of the first aspect.

[0028] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the technical solutions of the first aspect.

[0029] The technical effects of any of the design methods in aspects two through five can be found in the technical effects of different design methods in aspect one, and will not be repeated here. Attached Figure Description

[0030] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0031] Figure 1 A flowchart illustrating a deep learning-based intelligent prediction method for tunnel gas concentration provided in some embodiments of this application;

[0032] Figure 2 A schematic diagram of the structure of a tunnel gas concentration intelligent prediction system based on deep learning provided in some embodiments of this application;

[0033] Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of this application. Detailed Implementation

[0034] Specific embodiments of the invention will now be described in detail. Although the invention is described in conjunction with these specific embodiments, it should be understood that it is not intended to limit the invention to these specific embodiments. Rather, these embodiments are intended to cover alternative, modified, or equivalent embodiments that may be included within the spirit and scope of the invention as defined by the claims. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. The invention may be practiced without some or all of these specific details.

[0035] When used in conjunction with the terms "comprising," "method comprising," or similar language in this specification and appended claims, the singular forms "a," "some," and "the" include plural references unless the context clearly indicates otherwise. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0036] Application Overview: In the field of tunnel construction, methane concentration prediction is a core technical aspect of ensuring construction safety. Among related technologies, prediction methods based on time series analysis or traditional machine learning are the mainstream. These methods typically employ a fixed time window segmentation mechanism to preprocess monitoring data, dividing the continuous methane concentration sequence into equal-length time intervals, and then using models to learn the concentration variation patterns within each segment. However, in long-distance tunnel construction, when drilling and blasting operations or geological exploration trigger abrupt geological structural changes, the stress field of the tunnel surrounding rock and the methane occurrence state can undergo nonlinear abrupt changes, leading to drastic and abnormal fluctuations in methane concentration within a short period.

[0037] The gas concentration prediction method using a fixed-time-window segmentation mechanism cannot effectively capture the abnormal concentration fluctuations caused by sudden changes, resulting in delayed prediction results. Specifically, because the fixed-time-window segmentation mechanism adopts a static partitioning strategy, its window length and sliding step size are preset based on normal geological conditions, making it difficult to adapt to the time-scale changes in concentration fluctuations under sudden change scenarios. This rigid segmentation method leads to two core problems: first, abnormal fluctuation features are segmented into different time windows, causing discontinuity in the feature space; second, the model input data cannot reflect the concentration jumps caused by sudden changes in a timely manner, ultimately resulting in significant lag in the prediction results, making it difficult to meet the real-time requirements of tunnel construction safety early warning.

[0038] To address the aforementioned technical issues, the overall approach of the technical solution provided in this application is as follows: A deep learning-based intelligent prediction method for tunnel gas concentration is provided, comprising: collecting the original gas concentration sequences and geological structural abrupt change parameters from each monitoring point within the tunnel; dynamically adjusting the time window length based on the geological structural abrupt change parameters to adaptively segment the original gas concentration sequence, generating a variable-length concentration segment sequence; merging the variable-length concentration segment sequences into a multi-monitoring-point concentration segment sequence matrix; constructing a dynamic weighted gas concentration spatial relationship fusion map based on the spatial location of the monitoring points and the distribution of geological structural abrupt change zones; inputting the dynamic weighted fusion map into a trained graph neural network to extract the dynamic spatial relationship representation vector of each monitoring point; inputting the dynamic spatial relationship representation vector into a trained recurrent neural network encoder to generate spatiotemporal features containing the influence of geological abrupt changes; and performing variable-length prediction decoding on the spatiotemporal features through a segmented attention network to output the gas concentration for the predicted time period.

[0039] This method, based on the original gas concentration sequence and geological structural abrupt change parameters, adaptively segments the collected data by dynamically adjusting the time window length. The segmentation window is adjusted using the geological abrupt change parameters reflected in the input, resulting in a variable-length concentration segment sequence. This ensures that information on the instantaneous and drastic fluctuations during nonlinear abrupt changes caused by drilling and blasting operations or geological exploration during tunnel construction is fully preserved. After matrix construction, the monitoring point data directly reflects the physical distance between monitoring points and the local impact caused by the distribution of geological structural abrupt change zones in the spatial information. The use of dynamic weighted graph reconstruction gives the spatial relationships a true and accurate physical meaning. The spatial fusion graph is input into a graph neural network, and the network's multi-layer information aggregation operation is used to extract a deep representation of the complex dynamic changes between monitoring points, thereby obtaining a vector expression that finely reflects the spatial characteristics of gas concentration.

[0040] Subsequently, a recurrent neural network was selected to encode spatiotemporal features and generate a temporal expression that incorporates the influence of geological mutations. A segmented attention mechanism was used to perform variable-length prediction of spatiotemporal features in the decoding stage, and the gas concentration data for the predicted time period was output to achieve accurate estimation of short-term drastic fluctuation trends.

[0041] In summary, this scheme accurately captures localized and drastic fluctuations caused by nonlinear mutations during the data preprocessing stage using dynamic slicing, fully reflects the geological anomaly effects between monitoring points during the spatial modeling stage using dynamic weighted reconstruction, and achieves effective fusion of spatiotemporal features during the depth feature extraction and prediction stage, ultimately forming a real-time and accurate gas concentration prediction. This solves the problem that traditional fixed-time-window methods cannot capture abnormal concentration fluctuations caused by mutations in a timely manner during tunnel construction.

[0042] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings. Please refer to... Figure 1 This application provides a method for intelligent prediction of tunnel gas concentration based on deep learning, including:

[0043] S101: Collect the original gas concentration sequence and geological structural change parameters of each monitoring point in the tunnel;

[0044] Specifically, in some embodiments, the original gas concentration sequence of each monitoring point in the tunnel can be collected by sensors; geological structural change parameters can be collected by geological monitoring equipment.

[0045] S102: The time window length is dynamically adjusted according to the geological structural change parameters to adaptively segment the original gas concentration sequence and generate a variable-length concentration segment sequence.

[0046] Specifically, in some embodiments, the original gas concentration sequence can be adaptively segmented to generate a variable-length concentration segment sequence by dynamically adjusting the time window length based on geological structural abrupt change parameters through the following steps:

[0047] The first step is to calculate the dynamic time window length for each monitoring point based on geological structural abrupt change parameters. The specific formula is as follows: ;

[0048] In the formula, Indicates the first The dynamic time window length for each monitoring point; The base value for the preset time window; These are mutation parameter mapping coefficients, used to adjust the effect of mutation intensity on window length; This represents the first parameter directly extracted from geological structural abrupt change parameters. Original values ​​of mutation intensity at each monitoring point;

[0049] More specifically, the dynamic time window length for each monitoring point can be calculated based on geological structural abrupt change parameters using the following steps:

[0050] The first sub-step involves normalizing the original geological structural abrupt change data. The specific formula is as follows:

[0051] ;

[0052] In the formula, Indicates the first Normalized mutation intensity values ​​for each monitoring point; For the first The original geological structural abrupt change parameter values ​​collected at each monitoring point; and These represent the minimum and maximum values ​​of the geological structural abrupt change parameters observed at all monitoring points within the tunnel, respectively.

[0053] The second sub-step is to extract the mutation intensity value from the geological structural mutation parameters after normalization.

[0054] The third sub-step calculates the dynamic time window length based on the mapping relationship between the mutation intensity value and the preset time window base value. This scheme employs geological structural mutation parameter normalization to express input parameters at a unified scale. The normalized mutation intensity value serves as the basis for calculating the dynamic time window length, eliminating inherent scale differences between monitoring data and thus maintaining the objective stability of the model input. Normalization standardizes the expression of mutation intensity, ensuring that the dynamic time window length mapping process is less affected by data fluctuations, improving data continuity and consistency. The mutation intensity obtained after normalization provides a clear parameter reference in dynamic time window adjustment, enhancing the model's ability to capture gas concentrations.

[0055] The second step involves segmenting the original gas concentration sequence using a dynamic time window length to generate a variable-length concentration segmented sequence. This dynamic time window length, determined based on the original abrupt change intensity value and a predetermined mapping coefficient, ensures that the continuous gas concentration sequence accurately reflects the local nonlinear changes caused by geological structural abrupt changes during the segmentation process. The segmented data formed after adaptive segmentation of the continuous gas concentration data has a more defined temporal structure. Its input into subsequent prediction models can fully reflect the key characteristics of gas concentration changes within the tunnel, facilitating the extraction of anomalous fluctuation information and improving the sensitivity and accuracy of the prediction model in responding to anomalous fluctuations.

[0056] S103: Merge the variable-length concentration segmented sequences into a multi-monitoring-point concentration segmented sequence matrix;

[0057] S104: Based on the spatial location of monitoring points and the distribution of geological structural abrupt change zones, a dynamic weighted spatial relationship fusion map of gas concentration is constructed;

[0058] Specifically, in some embodiments, a dynamic weighted spatial relationship fusion map of gas concentration can be constructed based on the spatial location of monitoring points and the distribution of geological structural abrupt change zones through the following steps:

[0059] The first step is to calculate the distance between monitoring points based on their spatial location;

[0060] The second step involves assigning weights to each monitoring point based on the distribution of geological structural abrupt change zones; the specific implementation steps are as follows:

[0061] The first sub-step involves calculating the distance attenuation factor based on the distance from the monitoring point to the nearest geological structural abrupt change zone. The specific formula is as follows:

[0062] ;

[0063] In the formula, Indicates monitoring point Distance attenuation factor; Indicates monitoring point Distance to the nearest geological abrupt change zone; This is the distance attenuation constant, used to adjust the effect of distance on the attenuation factor;

[0064] The second sub-step assigns weights to each monitoring point based on the distance attenuation factor and the intensity of geological structural abrupt changes, using the following formula:

[0065] ;

[0066] In the formula, Indicates allocation to monitoring points The weights; This is a mutation intensity adjustment coefficient, used to further amplify or suppress mutation intensity. Impact on weights; Indicates the first The normalized abrupt change intensity values ​​for each monitoring point. In this scheme, the final weight calculation formula for monitoring points uses a distance attenuation factor multiplied by the normalized abrupt change intensity. This explicitly reflects the moderating effect of physical distance on the degree of influence and the expression effect of geological structural abrupt changes on abnormal fluctuations in the weight values, achieving a precise characterization of the influence of monitoring points in the spatial distribution within the tunnel. The monitoring point weight assignment mechanism expresses the distance effect in an exponential form and eliminates differences in data scale between different regions through normalization processing. This ensures that the monitoring point weights accurately reflect geological anomaly information, thereby guaranteeing the effective capture of dynamic changes in abnormal gas concentration during subsequent data processing, ultimately providing sufficient spatial information support for gas concentration prediction.

[0067] The third step involves constructing a dynamic weighted spatial relationship fusion map of gas concentration based on the distances and weights between monitoring points. The connection weights are calculated using the following formula during the map construction process:

[0068] ;

[0069] In the formula, Indicates monitoring point With monitoring points Connection weights between them; Indicates monitoring point With monitoring points The Euclidean distance between them; This is the distance decay constant, used to control the impact of distance on the weights; For monitoring points The basic weights are assigned based on the distribution of geological structural abrupt change zones.

[0070] This scheme constructs a dynamic weighted gas concentration spatial relationship fusion map that combines the Euclidean distance between monitoring points with the basic weights assigned by the distribution of geological structural abrupt change zones. An exponential decay function quantifies the numerical expression of physical distance on the connection weights, thus organically combining the spatial location of monitoring points with the influence of geological conditions in a mathematical expression, forming clear spatial connection weights. The dynamic weighted gas concentration spatial relationship fusion map provides finely structured spatial information during the data input stage of the deep learning model. Its construction process quantifies both the physical distance between monitoring points and the effects of geological structures, rigorously presenting the dynamic characteristics of gas concentration spatial distribution within the tunnel. This enhances the deep learning model's ability to identify local and global spatial correlation changes within the tunnel, and overall improves the accuracy and stability of gas concentration prediction in capturing abnormal fluctuations.

[0071] S105: Input the dynamic weight fusion graph into the trained graph neural network to extract the dynamic spatial relationship representation vector of each monitoring point;

[0072] Specifically, in some embodiments, the dynamic weight fusion graph can be input into the trained graph neural network through the following steps to extract the dynamic spatial relationship representation vector of each monitoring point:

[0073] The first step is to fuse the dynamic weights into the graph neural network trained with the graph input; the graph neural network includes an input layer, multiple hidden layers, and an output layer; in the multiple hidden layers of the graph neural network, the weights are fused into the graph input layer. The layer performs the following calculations to generate the first... Hidden feature vectors of the layer:

[0074] ;

[0075] In the formula, Indicates monitoring point In the Hidden feature vectors of the layer; Indicates monitoring point In the Hidden feature vectors of the layer; Indication and monitoring points The set of connected adjacent nodes; For monitoring points With monitoring points Connection weights between them; Indicates the first Layer bias terms; This is an activation function used to introduce nonlinear feature transformations;

[0076] The second step involves processing the dynamic weight fusion graph through multiple hidden layers of the trained graph neural network to extract the dynamic spatial relationship representation vector of each monitoring point.

[0077] This scheme employs a trained graph neural network to perform multi-layer nonlinear mapping on the dynamic weight fusion graph, enabling the complex spatial relationships between monitoring points to be fully expressed through layered information aggregation. The trained graph neural network iteratively updates adjacency relationships according to preset weights, and each layer continuously refines the hidden vector representation during the neighborhood feature transmission process, thereby obtaining a high-quality dynamic spatial relationship representation that reflects the true contribution of monitoring points. The multi-layer structure ensures that the local information of adjacent monitoring points is fused with the overall spatial structure layer by layer, and the final output feature vector can fully capture the subtle changes and dynamic interactions in the spatial distribution of the monitoring area. Through this deep feature extraction process, the smooth transmission and effective amplification of complex spatial information are achieved, ensuring that the subsequent spatiotemporal prediction module obtains more accurate and comprehensive input features, thereby improving the overall prediction accuracy and response capability.

[0078] S106: Input the dynamic spatial relationship representation vector into the trained recurrent neural network encoder to generate spatiotemporal features that include the impact of geological mutations;

[0079] Specifically, the obtained dynamic spatial relationship representation vectors of each monitoring point are first arranged in chronological order to form the input sequence describing continuous time steps. The recurrent neural network encoder employs a multi-layer recurrent neural network structure, such as a Long Short-Term Memory (LSTM) network or a gated recurrent unit (GRU) network, to characterize the dependencies between time steps. Within each time step, the recurrent neural network encoder achieves information transfer and updating by fusing the current input vector with the hidden state of the previous time step, and extracts temporal features through the following update formula:

[0080]

[0081] in, Indicates at time step The hidden state, This indicates the hidden state of the previous time step. Indicates at time step The input dynamic spatial relation representation is a vector. The weight matrix represents the hidden states. The weight matrix is ​​the input vector. For bias terms, The activation function is tanh or ReLU. The encoder gradually integrates local spatiotemporal information and features reflecting the impact of geological mutations by stacking multiple layers of the above-mentioned temporal update units, and finally generates a high-dimensional feature representation that integrates spatiotemporal dynamic information. This representation is the spatiotemporal feature containing the impact of geological mutations.

[0082] S107: The spatiotemporal features are predicted and decoded using a segmented attention network to output the gas concentration for the predicted time period.

[0083] Specifically, in some embodiments, the following steps can be used to perform variable-length prediction decoding of spatiotemporal features through a segmented attention network, outputting the gas concentration for the predicted time period:

[0084] The first step is to input the spatiotemporal features into the trained segmented attention network;

[0085] The second step involves using time periods during the segmented attention decoding process after training. attention weights The specific calculation formula is as follows:

[0086] ;

[0087] In the formula, Represents spatiotemporal characteristic segments Attention weights; Representing time-space segment The query vector; Representing time-space segment The key vector; This represents the dimension of the key vector, used for scaling. "" represents vector dot product. This scheme employs a segmented attention network to achieve variable-length prediction output during the spatiotemporal feature decoding process. The segmented attention network calculates and normalizes the dot product of the query vector and key vector between each spatiotemporal segment, ensuring that the importance of local features is fully reflected during decoding, thus effectively strengthening key spatiotemporal information. During computation, the segmented attention network assigns corresponding weights to each spatiotemporal segment and allocates prediction capabilities based on these weights, ensuring that the effective signals contained in the input spatiotemporal features are accurately represented in the prediction output. In this way, spatiotemporal features obtained under conditions of drastic gas concentration fluctuations can adaptively adjust their contribution intensity, thereby improving the overall prediction accuracy and stability of the system.

[0088] Please see Figure 2 Based on the same inventive concept as the deep learning-based intelligent prediction method for tunnel gas concentration in the foregoing embodiments, this application provides a deep learning-based intelligent prediction system for tunnel gas concentration, comprising:

[0089] Data acquisition module 201 is used to collect the original gas concentration sequence and geological structural change parameters of each monitoring point in the tunnel;

[0090] The adaptive segmentation module 202 is used to dynamically adjust the time window length according to geological structural abrupt parameters, adaptively segment the original gas concentration sequence, and generate a variable-length concentration segment sequence.

[0091] The sequence matrix merging module 203 is used to merge variable-length concentration segmented sequences into a multi-monitoring-point concentration segmented sequence matrix.

[0092] The spatial relationship fusion map construction module 204 is used to construct a dynamic weighted gas concentration spatial relationship fusion map based on the spatial location of monitoring points and the distribution of geological structural abrupt change zones.

[0093] The spatial feature extraction module 205 is used to input the dynamic weight fusion graph into the trained graph neural network and extract the dynamic spatial relationship representation vector of each monitoring point.

[0094] The spatiotemporal feature generation module 206 is used to input the dynamic spatial relationship representation vector into the trained recurrent neural network encoder to generate spatiotemporal features that include the impact of geological mutations.

[0095] The variable-length prediction decoding module 207 is used to perform variable-length prediction decoding on spatiotemporal features through a segmented attention network and output the gas concentration for the predicted time period.

[0096] In some embodiments, the adaptive sharding module 202 is specifically used for:

[0097] Based on geological structural abrupt change parameters, the dynamic time window length for each monitoring point is calculated using the following formula:

[0098] ;

[0099] In the formula, Indicates the first The dynamic time window length for each monitoring point; The base value for the preset time window; These are mutation parameter mapping coefficients, used to adjust the effect of mutation intensity on window length; This represents the first parameter directly extracted from geological structural abrupt change parameters. Original values ​​of mutation intensity at each monitoring point;

[0100] The original gas concentration sequence is segmented using a dynamic time window length to generate a variable-length concentration segmented sequence.

[0101] In some embodiments, the spatial relationship fusion graph construction module 204 is specifically used for:

[0102] Calculate the distance between monitoring points based on their spatial location;

[0103] Based on the distribution of geological structural abrupt change zones, a weight is assigned to each monitoring point;

[0104] A dynamic weighted spatial relationship fusion map of gas concentration is constructed based on the distance and weights between monitoring points; the connection weights are calculated using the following formula during the mapping process:

[0105] ;

[0106] In the formula, Indicates monitoring point With monitoring points Connection weights between them; Indicates monitoring point With monitoring points The Euclidean distance between them; This is the distance decay constant, used to control the impact of distance on the weights; For monitoring points The basic weights are assigned based on the distribution of geological structural abrupt change zones.

[0107] In some embodiments, the spatial feature extraction module 205 is specifically used for:

[0108] The graph neural network is trained by fusing dynamic weights with a graph input; the graph neural network includes an input layer, multiple hidden layers, and an output layer; in the multiple hidden layers of the graph neural network, for the first... The layer performs the following calculations to generate the first... Hidden feature vectors of the layer:

[0109] ;

[0110] In the formula, Indicates monitoring point In the Hidden feature vectors of the layer; Indicates monitoring point In the Hidden feature vectors of the layer; Indication and monitoring points The set of connected adjacent nodes; For monitoring points With monitoring points Connection weights between them; Indicates the first Layer bias terms; The activation function is used to introduce a nonlinear feature transformation; for example, the activation function can be ReLU or Sigmoid.

[0111] The dynamic weight fusion graph is processed by multiple hidden layers of the trained graph neural network to extract the dynamic spatial relationship representation vector of each monitoring point.

[0112] In some embodiments, the variable-length prediction decoding module 207 is specifically used for:

[0113] Spatiotemporal features are input into the trained segmented attention network;

[0114] A segmented attention network, trained after training, performs variable-length prediction and decoding of spatiotemporal features, outputting the gas concentration for the predicted time period. During the decoding process, the time period... attention weights The specific calculation formula is as follows:

[0115] ;

[0116] In the formula, Represents spatiotemporal characteristic segments Attention weights; Representing time-space segment The query vector; Representing time-space segment The key vector; This represents the dimension of the key vector, used for scaling. " represents the dot product of vectors, This represents all time-space segments.

[0117] In some embodiments, the data acquisition module 201 is specifically used for:

[0118] The raw methane concentration sequence of each monitoring point inside the tunnel was collected using sensors.

[0119] Geological structural change parameters are collected using geological monitoring equipment.

[0120] In some embodiments, the adaptive sharding module 202 is further configured to:

[0121] The original geological structural abrupt change data are normalized using the following formula:

[0122] ;

[0123] In the formula, Indicates the first Normalized mutation intensity values ​​for each monitoring point; For the first The original geological structural abrupt change parameter values ​​collected at each monitoring point; and These represent the minimum and maximum values ​​of the geological structural abrupt change parameters observed at all monitoring points within the tunnel, respectively.

[0124] Extract mutation intensity values ​​from the normalized geological structural mutation parameters;

[0125] The length of the dynamic time window is calculated based on the mapping relationship between the mutation intensity value and the preset time window base value.

[0126] In some embodiments, the spatial relationship fusion graph construction module 204 is further used for:

[0127] The distance attenuation factor is calculated based on the distance from the monitoring point to the nearest geological abrupt change zone. The specific formula is as follows:

[0128] ;

[0129] In the formula, Indicates monitoring point Distance attenuation factor; Indicates monitoring point Distance to the nearest geological abrupt change zone; This is the distance attenuation constant, used to adjust the effect of distance on the attenuation factor;

[0130] Based on the distance attenuation factor and the intensity of geological structural abrupt changes, a weight is assigned to each monitoring point, as shown in the following formula:

[0131] ;

[0132] In the formula, Indicates allocation to monitoring points The weights; This is a mutation intensity adjustment coefficient, used to further amplify or suppress mutation intensity. Impact on weights; Indicates the first The normalized mutation intensity values ​​of each monitoring point.

[0133] It is understandable that the modules recorded in this deep learning-based intelligent prediction system for tunnel gas concentration are similar to those in the reference system. Figure 1 The steps described correspond to those in the deep learning-based intelligent prediction method for tunnel gas concentration. Therefore, the operations, features, and beneficial effects described above also apply to the deep learning-based intelligent prediction system for tunnel gas concentration and its constituent modules, and will not be repeated here.

[0134] Please see Figure 3 Based on the inventive concept of the deep learning-based intelligent prediction method for tunnel gas concentration in the foregoing embodiments, this application provides an electronic device. This electronic device may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), etc., and fixed terminals such as digital TVs, desktop computers, etc. The electronic device includes a processing unit 301 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM 302 (read-only memory) or a program loaded from storage device 308 into RAM 303 (random access memory). RAM 303 also stores various programs and data required for the operation of the electronic device. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output interface (i.e., I / O interface 305) is also connected to the bus 304.

[0135] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touch screens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data.

[0136] In particular, according to some embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this application.

[0137] It should be noted that, in some embodiments of this application, the computer-readable medium described may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0138] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol, such as HTTP (Hypertext Transfer Protocol), and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and end-to-end networks (e.g., ad hoc end-to-end networks), as well as any currently known or future-developed networks.

[0139] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, enable the electronic device to implement the method steps of any of the aforementioned technical solutions.

[0140] Computer program code for performing operations of some embodiments of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0141] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0142] The modules described in some embodiments of this application can be implemented in software or hardware. The described modules can also be located in a processor. The names of these modules do not, in some cases, constitute a limitation on the module itself.

[0143] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0144] Some embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the above-described deep learning-based intelligent prediction method for tunnel gas concentration.

[0145] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A deep learning-based intelligent prediction method for tunnel gas concentration, characterized in that, include: The original gas concentration sequence and geological structural abrupt change parameters of each monitoring point in the tunnel were collected. The time window length is dynamically adjusted based on geological structural mutation parameters to adaptively segment the original gas concentration sequence and generate a variable-length concentration segment sequence. The variable-length concentration segmented sequences were merged into a multi-monitoring-point concentration segmented sequence matrix; Based on the spatial location of monitoring points and the distribution of geological structural abrupt change zones, a dynamic weighted spatial relationship fusion map of gas concentration is constructed. The dynamic weight fusion graph is input into the trained graph neural network to extract the dynamic spatial relationship representation vector of each monitoring point. The dynamic spatial relationship representation vector is input into the trained recurrent neural network encoder to generate spatiotemporal features that include the impact of geological mutations; A segmented attention network is used to perform variable-length prediction decoding of spatiotemporal features, and the gas concentration for the predicted time period is output.

2. The intelligent prediction method for tunnel gas concentration based on deep learning according to claim 1, characterized in that, The steps involved in dynamically adjusting the time window length based on geological structural abrupt change parameters to adaptively segment the original gas concentration sequence and generate a variable-length concentration segment sequence include: Based on the geological structural abrupt change parameters, the dynamic time window length for each monitoring point is calculated using the following formula: ； In the formula, Indicates the first The dynamic time window length for each monitoring point; The base value for the preset time window; These are mutation parameter mapping coefficients, used to adjust the effect of mutation intensity on window length; This represents the first parameter directly extracted from geological structural abrupt change parameters. Original values ​​of mutation intensity at each monitoring point; The original gas concentration sequence is segmented using the dynamic time window length to generate the variable-length concentration segment sequence.

3. The intelligent prediction method for tunnel gas concentration based on deep learning according to claim 1, characterized in that, The steps for constructing a dynamic weighted spatial relationship fusion map of gas concentration based on the spatial location of monitoring points and the distribution of geological structural abrupt change zones include: Calculate the distance between the monitoring points based on their spatial locations; Based on the distribution of the geological structural abrupt change zones, a weight is assigned to each monitoring point; Based on the distances between the monitoring points and the weights, a dynamic weighted spatial relationship fusion map of gas concentration is constructed; wherein, the connection weights are calculated using the following formula during the mapping process: ； In the formula, Indicates monitoring point With monitoring points Connection weights between them; Indicates monitoring point With monitoring points The Euclidean distance between them; This is the distance decay constant, used to control the impact of distance on the weights; For monitoring points The basic weights are assigned based on the distribution of geological structural abrupt change zones.

4. The intelligent prediction method for tunnel gas concentration based on deep learning according to claim 1, characterized in that, The steps of inputting the dynamic weight fusion graph into the trained graph neural network and extracting the dynamic spatial relationship representation vector of each monitoring point include: The dynamic weight fusion graph is input into the trained graph neural network; the graph neural network includes an input layer, multiple hidden layers, and an output layer; in the multiple hidden layers of the graph neural network, for the first... The layer performs the following calculations to generate the first... Hidden feature vectors of the layer: ； In the formula, Indicates monitoring point In the Hidden feature vectors of the layer; Indicates monitoring point In the Hidden feature vectors of the layer; Indication and monitoring points The set of connected adjacent nodes; For monitoring points With monitoring points Connection weights between them; Indicates the first Layer bias terms; This is an activation function used to introduce nonlinear feature transformations; The dynamic weight fusion graph is processed by multiple hidden layers of the trained graph neural network to extract the dynamic spatial relationship representation vector of each monitoring point.

5. The intelligent prediction method for tunnel gas concentration based on deep learning according to claim 1, characterized in that, The steps for performing variable-length prediction decoding of spatiotemporal features using a segmented attention network to output the gas concentration for the predicted time period include: The spatiotemporal features are input into the trained segmented attention network; The trained segmented attention network performs variable-length prediction decoding on the spatiotemporal features, outputting the gas concentration for the predicted time period. During the decoding process, the time period... attention weights The specific calculation formula is as follows: ； In the formula, Represents spatiotemporal characteristic segments Attention weights; Representing time-space segment The query vector; Representing time-space segment The key vector; This represents the dimension of the key vector, used for scaling. " represents the dot product of vectors.

6. The intelligent prediction method for tunnel gas concentration based on deep learning according to claim 3, characterized in that, Based on the distribution of the geological structural abrupt change zones, the step of assigning weights to each monitoring point includes: The distance attenuation factor is calculated based on the distance from the monitoring point to the nearest geological abrupt change zone. The specific formula is as follows: ； In the formula, Indicates monitoring point Distance attenuation factor; Indicates monitoring point Distance to the nearest geological abrupt change zone; This is the distance attenuation constant, used to adjust the effect of distance on the attenuation factor; Based on the distance attenuation factor and the intensity of geological structural abrupt changes, a weight is assigned to each monitoring point, using the following formula: ； In the formula, Indicates allocation to monitoring points The weights; This is a mutation intensity adjustment coefficient, used to further amplify or suppress mutation intensity. Impact on weights; Indicates the first The normalized mutation intensity values ​​of each monitoring point.

7. A deep learning-based intelligent prediction system for tunnel gas concentration, characterized in that, include: The data acquisition module is used to collect the original gas concentration sequence and geological structural change parameters of each monitoring point in the tunnel; The adaptive slicing module is used to dynamically adjust the time window length according to geological structural change parameters, adaptively slice the original gas concentration sequence, and generate a variable-length concentration segment sequence. The sequence matrix merging module is used to merge variable-length concentration segmented sequences into a multi-monitoring-point concentration segmented sequence matrix; The spatial relationship fusion map construction module is used to construct a dynamic weighted gas concentration spatial relationship fusion map based on the spatial location of monitoring points and the distribution of geological structural abrupt change zones; The spatial feature extraction module is used to input the dynamic weight fusion graph into the trained graph neural network and extract the dynamic spatial relationship representation vector of each monitoring point. The spatiotemporal feature generation module is used to input the dynamic spatial relationship representation vector into the trained recurrent neural network encoder to generate spatiotemporal features that include the impact of geological mutations. The variable-length prediction decoding module is used to perform variable-length prediction decoding on spatiotemporal features through a segmented attention network, and output the gas concentration for the predicted time period.

8. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processing device, implements the method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processing device, it implements the method of any one of claims 1 to 6.

Images