Underground environment dynamic monitoring method and system based on multi-source information fusion

Abstract

This invention relates to the field of underground environmental monitoring technology, and particularly to a method and system for dynamic monitoring of underground environments based on multi-source information fusion. The method includes: delineating the underground area requiring dynamic environmental monitoring; dividing the underground area into several blocks; and collecting sensor data within each block using sensors pre-installed in the blocks. By using environmental data as input to real-world operating conditions, this invention enables on-site verification of the outputs of the fusion model and the environmental risk identification model. This allows for timely detection of model deviations in the current block, achieving dynamic correction of model predictions and improving the accuracy and stability of early warnings. It not only enables localized risk identification in complex underground environments and ensures continuous model evolution and adaptive optimization, but also provides reliable technical support for underground mine safety monitoring and risk early warning.

Description

Technical Field

[0001] This invention relates to the field of downhole environmental monitoring technology, and in particular to a method and system for dynamic monitoring of the downhole environment based on multi-source information fusion. Background Technology

[0002] Underground environmental monitoring is typically achieved by deploying various types of sensors in roadways, mining faces, and key nodes. These sensors include detectors for toxic and harmful gases such as carbon monoxide, oxygen, and hydrogen sulfide, as well as monitoring equipment for temperature, humidity, dust concentration, wind speed and volume, pressure, and roof stress. The real-time data collected by these sensors is uploaded to the nearest substation or edge computing node via a combination of wired and wireless communication, and then transmitted to the ground dispatch center or cloud platform via a ring network or multi-hop routing.

[0003] However, the inability to perform analysis and processing on the data side can easily lead to increased transmission delays and excessive communication bandwidth consumption. Furthermore, it makes it difficult to detect anomalies in a timely manner when the underground network is unstable or partially interrupted, greatly reducing the real-time performance and response speed of the overall monitoring and weakening the ability to warn of sudden risks.

[0004] Therefore, "how to utilize edge devices to achieve near-end processing of sensor data" is the technical problem that this invention needs to solve. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for dynamic monitoring of the downhole environment based on multi-source information fusion, so as to solve the problem of "how to use edge devices to realize near-end processing of sensor data" mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for dynamic monitoring of the downhole environment based on multi-source information fusion, the method comprising: The downhole area that needs to be dynamically monitored for the environment is delineated, and the downhole area is divided into several blocks. Sensing devices pre-installed in the blocks are used to collect sensing data within the blocks. The sensing data includes at least: gas concentration, temperature and humidity and dust concentration. An environmental risk identification model is created and deployed to the edge devices pre-installed in the blocks. Risk events are obtained from historical data, the change process of sensor data is captured, risk characteristics are marked, a training set is built, and the environmental risk identification model is trained. The model parameters of the environmental risk identification model in each edge device are obtained, a federated learning architecture is constructed, the fusion parameter set is calculated, and it is distributed to all edge devices in parallel. Real-time values ​​of sensor data are collected and input into the environmental risk identification model in the corresponding edge device to determine whether risk characteristics exist. Using a Bluetooth receiver pre-integrated in the edge device, the Bluetooth signal of the four-in-one gas detector worn by the staff is collected to locate the staff's real-time position. Several sample points are selected, and environmental data at each sample point is collected using the four-in-one gas detector. The block corresponding to the real-time position is defined as the target block. The environmental risk identification model in the edge device of the target block is initialized and written into the fusion parameter group to obtain the fusion model. The environmental data is input into the fusion model, and the judgment results of the environmental risk identification model are cross-validated.

[0008] Furthermore, the step of collecting sensor data within the segment, wherein the sensor data includes at least: gas concentration, temperature and humidity, and dust concentration, includes: The sensor data is divided into several individual items, and the correlation between the individual items is established; Based on the individual items and relationships, several evaluation rules are edited to form a rule set, which is then sent to the edge device.

[0009] Furthermore, the steps of acquiring risk events from historical data, extracting the change process of sensor data, labeling risk characteristics, and constructing a training set include: Set an alarm level for each risk event, wherein the alarm levels include at least: high, medium and low; Insert labels generated from alarm levels into the training set.

[0010] Furthermore, the steps of constructing a federated learning architecture, calculating a fusion parameter set, distributing it in parallel to all edge devices, collecting real-time values ​​of sensor data, and inputting them into the environmental risk identification model within the corresponding edge device to determine whether risk characteristics exist include: Set attribute data for each block, wherein the attribute data includes at least: job type and historical risk, and cluster the blocks into several groups based on the attribute data; Establish the correspondence between groups and fusion parameter groups, and update the fusion model.

[0011] Furthermore, the step of using a Bluetooth receiver pre-integrated in the edge device to collect the Bluetooth signal from the four-in-one gas detector worn by the worker and to locate the worker's real-time position includes: From the Bluetooth signals, select the test signal, extract the signal list of edge devices, and define the edge devices in the signal list that contain the test signal as signal-associated devices; The strength of the test signal in the signal-correlated device is obtained, and the real-time position of the four-in-one gas detector corresponding to the test signal is located based on the multi-angle positioning algorithm.

[0012] Furthermore, the steps of initializing the environmental risk identification model in the target block edge device, writing it into the fusion parameter group to obtain the fusion model, inputting environmental data into the fusion model, and cross-validating the judgment results of the environmental risk identification model include: Identify the triggering conditions for updating the fusion model, wherein the triggering conditions include at least: event triggering and time triggering; When the triggering condition is met, the fusion model is updated, the change process of the fusion parameter group is recorded, key parameters are extracted, and an optimized version is generated.

[0013] Furthermore, the system includes: The deployment module is used to delineate the underground area that needs to be dynamically monitored for the environment, divide the underground area into several blocks, use the sensing devices pre-installed in the blocks to collect sensing data in the blocks, wherein the sensing data includes at least: gas concentration, temperature and humidity and dust concentration, create an environmental risk identification model, and deploy it to the edge devices pre-installed in the blocks. The judgment module is used to obtain risk events in historical data, extract the change process of sensor data, mark risk features, build a training set, train the environmental risk identification model, obtain the model parameters of the environmental risk identification model in each edge device, construct a federated learning architecture, calculate the fusion parameter set, distribute it in parallel to all edge devices, collect the real-time value of sensor data, input it into the environmental risk identification model in the corresponding edge device, and determine whether there are risk features. The verification module uses a Bluetooth receiver pre-integrated in the edge device to collect the Bluetooth signal of the four-in-one gas detector worn by the staff, locate the staff's real-time position, select several sample points, collect environmental data at each sample point using the four-in-one gas detector, define the block corresponding to the real-time position as the target block, initialize the environmental risk identification model in the edge device of the target block, write it into the fusion parameter group to obtain the fusion model, input the environmental data into the fusion model, and cross-validate the judgment results of the environmental risk identification model.

[0014] Furthermore, the deployment module includes: A unit is established to divide the sensing data into several individual items and establish the relationship between the individual items; The sending unit is used to edit several evaluation rules, form a rule set, and send it to the edge device via the individual items and associations.

[0015] Furthermore, the determination module includes: The setting unit is used to set the alarm level for each risk event, wherein the alarm levels include at least: high, medium and low; An insertion unit is used to insert labels generated by alarm levels into the training set; Clustering unit, used to set attribute data for each block, wherein the attribute data includes at least: job type and historical risk, and based on the attribute data, the blocks are clustered into several groups; The update unit is used to establish the correspondence between the group and the fusion parameter group and update the fusion model.

[0016] Furthermore, the verification module includes: The definition unit is used to select the test signal from the Bluetooth signal, extract the signal list of the edge device, and define the edge device containing the test signal in the signal list as the signal associated device. The acquisition unit is used to acquire the strength of the test signal in the signal association device and locate the real-time position of the four-in-one gas detector corresponding to the test signal based on the multi-angle positioning algorithm. The identification unit is used to identify the triggering conditions for updating the fusion model, wherein the triggering conditions include at least: event triggering and time triggering; The extraction unit is used to update the fusion model when the triggering condition is met, record the change process of the fusion parameter group, extract key parameters, and generate an optimized version.

[0017] Compared with the prior art, the beneficial effects of the present invention are: By creating an environmental risk identification model, separate models can be built for different geological structures, ventilation conditions, and operational intensities corresponding to different blocks. This avoids the over-averaging of complex downhole environments by a uniform model, improving the accuracy and specificity of risk identification. Simultaneously, it reduces the computational burden on a single central model, making data processing more real-time and ensuring rapid response and coordinated early warning for local anomalies. Training the environmental risk identification model allows for a more accurate learning of the inherent patterns and evolutionary characteristics of risks in the downhole environment, enabling dynamic updates and adaptive adjustments to the model, thus enhancing the stability and accuracy of downhole environmental monitoring and early warning. By aggregating model parameters learned from different blocks under various geological conditions, ventilation environments, and operational states, the unified model possesses stronger global generalization capabilities, avoids local biases in single-block models, and improves the overall accuracy of risk identification in complex underground multi-scenario environments. By utilizing environmental data as input for real-world working conditions, the output results of the fusion model and the environmental risk identification model can be verified on-site, thereby promptly identifying model deviations in the current block and achieving dynamic correction of model prediction results. This allows risk assessment to no longer rely solely on historical learning results but instead undergo secondary confirmation based on the current actual environment, improving the accuracy and stability of early warnings. It enables localized processing of risk identification in complex underground environments, ensuring continuous model evolution and adaptive optimization, and providing reliable technical support for underground mine safety monitoring and risk early warning. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the downhole environment dynamic monitoring method based on multi-source information fusion provided in this embodiment of the invention; Figure 2 This is a first sub-flowchart of the downhole environment dynamic monitoring method based on multi-source information fusion provided in an embodiment of the present invention. Figure 3 This is a second sub-flowchart of the downhole environment dynamic monitoring method based on multi-source information fusion provided in an embodiment of the present invention. Figure 4 This is a third sub-flow diagram of the downhole environment dynamic monitoring method based on multi-source information fusion provided in an embodiment of the present invention; Figure 5 This is a block diagram of the composition of a downhole environment dynamic monitoring system based on multi-source information fusion provided in an embodiment of the present invention; Figure 6 This is a block diagram of the deployment modules in the downhole environment dynamic monitoring system based on multi-source information fusion provided in an embodiment of the present invention. Figure 7 This is a block diagram of the judgment module in the downhole environment dynamic monitoring system based on multi-source information fusion provided in an embodiment of the present invention. Figure 8 This is a block diagram of the verification module in the downhole environment dynamic monitoring system based on multi-source information fusion provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In Example 1, Figure 1 The implementation flow of the downhole environment dynamic monitoring method based on multi-source information fusion provided in this embodiment of the invention is illustrated below, and is described in detail below: S100: Delineate the downhole area that needs to be dynamically monitored for the environment, divide the downhole area into several blocks, use sensing devices pre-installed in the blocks to collect sensing data within the blocks, wherein the sensing data includes at least: gas concentration, temperature and humidity and dust concentration, create an environmental risk identification model, and deploy it to the edge devices pre-installed in the blocks.

[0021] Based on actual production needs, the underground areas requiring environmental monitoring are determined. These areas should include: mining faces, return air tunnels, transport tunnels, and ventilation dead zones. According to the tunnel structure, work distribution, ventilation paths, and geological conditions, the underground area is divided into several clearly defined sub-regions. Each sub-region can be simply understood as a portion of the underground area. Based on the risk level and exposure risk of each sub-region, sensing devices are deployed within each sub-region. These devices include: gas concentration sensors, temperature and humidity sensors, dust monitors, and wind speed, direction, and structural stress sensors. Data is periodically collected from these sensors within each sub-region.

[0022] By utilizing deep learning algorithms, an environmental risk identification model is constructed. This model can perform real-time risk assessment on multi-source sensor data from underground. The environmental risk identification model is deployed to edge devices, with each block containing an edge device. These edge devices can be underground distributed edge processing units or edge gateways, etc. The environmental risk identification model in the edge devices can directly receive and process real-time data streams from sensor devices on the local side where the data is generated, enabling on-site inference and rapid risk assessment. This reduces dependence on the central cloud, lowers communication latency and network load, and improves response speed and real-time early warning capabilities for localized sudden anomalies.

[0023] S200: Obtain risk events from historical data, extract the change process of sensor data, label risk features, build a training set, train the environmental risk identification model, obtain the model parameters of the environmental risk identification model in each edge device, construct a federated learning architecture, calculate the fusion parameter set, distribute it in parallel to all edge devices, collect real-time values ​​of sensor data, input them into the environmental risk identification model in the corresponding edge device, and determine whether risk features exist.

[0024] Historical monitoring data refers to records of past safety hazards or anomalies. This process involves temporally aligning and segmenting multi-source sensor data related to risk events to extract changes in the sensor data before and after the risk occurred. These changes include trends in gas concentration fluctuations, temperature and humidity gradients, dust accumulation rates, and abnormal ventilation parameters. Risk characteristics during these changes are manually labeled, including sudden changes in carbon monoxide concentration, decreases in oxygen concentration, and exceedances of toxic gases such as hydrogen sulfide. This process determines the normal state and risk characteristics of multi-source sensor data across different time periods, constructing a training set for training the environmental risk identification model. The advantage of this approach is that it allows the model to gradually learn the evolutionary patterns and discriminative characteristics of different risk types within multi-dimensional sensor data, thereby improving the model's ability to identify and analyze potential risks in complex downhole environments.

[0025] Model parameters for the environmental risk identification model are extracted from edge devices. These parameters include weight parameters, bias parameters, network layer connection parameters, feature mapping parameters, and normalization parameters, with each edge device corresponding to a set of model parameters. A federated learning architecture is constructed, which involves assigning weight coefficients to each edge device after local model training at each edge node, corresponding to its data contribution, sample size, and environmental variability. The locally updated model parameters and weight coefficients are uploaded to the cloud. On the cloud side, a weighted average of the model parameters uploaded from different edge devices is calculated using the weight coefficients to obtain a set of globally optimized parameters, i.e., the fusion parameter set. The fusion parameter set is distributed to all edge devices. When the latest sensor data is collected using the sensing devices, the obtained real-time value is input into the environmental risk identification model to determine whether there is an abnormal evolution trend in the current data sequence. If the output of the environmental risk identification model meets the risk judgment criteria, it is identified as having risk characteristics.

[0026] S300: Using a Bluetooth receiver pre-integrated in the edge device, the system collects the Bluetooth signal from the four-in-one gas detector worn by the staff, locates the staff's real-time position, selects several sample points, and uses the four-in-one gas detector to collect environmental data at each sample point. The block corresponding to the real-time position is defined as the target block. The environmental risk identification model in the edge device of the target block is initialized and written into the fusion parameter group to obtain the fusion model. The environmental data is input into the fusion model, and the judgment results of the environmental risk identification model are cross-validated.

[0027] When working underground, workers wear a four-in-one gas detector. This portable, multi-parameter gas detection device can simultaneously detect oxygen, carbon monoxide, hydrogen sulfide, and combustible gases. "Four-in-one" is a colloquial term; in practice, the specific portable monitoring device should be determined based on the actual usage scenario. Using a Bluetooth receiver integrated into the edge device, the system continuously scans and receives Bluetooth broadcast signals emitted by the four-in-one gas detector worn by the worker. It analyzes the signal strength, arrival time, and differences between multiple Bluetooth receivers. Combining this with the spatial coordinates of the deployed Bluetooth receivers underground, a trilateration algorithm is used to calculate and spatially map the worker's location in real time, determining their relative coordinates within the underground tunnel. Multiple sample points are pre-selected in the underground area. When the worker's real-time location coincides with a sample point, the corresponding environmental data is recorded. This environmental data is the set of specific environmental monitoring parameters directly collected by the four-in-one gas detector at the sample point. The block corresponding to the worker's real-time location is identified and defined as the target block. The environmental risk identification model corresponding to the target block is determined, and its model parameters are initialized. The fusion parameter group is written into the initialized environmental risk identification model to generate a fusion model, which is the unified model obtained after federated learning. Environmental data collected by the four-in-one gas detector is input into the fusion model, and the risk identification result is output. This result is compared and cross-validated with the output result of the environmental risk identification model of the corresponding block. Consistency analysis is used to determine whether the verification results are the same. If they are the same, the corresponding emergency response rules are activated. The emergency response rules include: initiating local ventilation enhancement measures and initiating personnel evacuation and location warning, etc. After the four-in-one gas detector collects environmental data at the sample point, it encapsulates the collected environmental data to generate an environmental feature dataset containing oxygen concentration, carbon monoxide concentration, hydrogen sulfide concentration, and combustible gas concentration as mentioned above. The obtained environmental feature dataset is then input into the fusion model. The fusion model is obtained through federated learning based on the environmental risk identification models corresponding to multiple blocks. The environmental risk identification model can make real-time risk judgments based on downhole gas composition data (i.e., environmental feature dataset). The input data type of the fusion model is consistent with the input data type of each block's environmental risk identification model. Therefore, the environmental feature dataset collected by the four-in-one gas detector can be directly used as the input data of the fusion model.

[0028] In this embodiment, cross-validation is performed using the outputs of the local model (environmental risk identification model) of the segment and the fusion model of the whole domain. If the outputs are the same, it indicates that there is an environmental anomaly in the segment from which the data is sourced, and emergency response rules should be activated. If the outputs are different, it indicates that the data characteristics collected by the segment deviate from the environmental characteristics of the whole domain, and the current abnormal characteristics have not yet reached the risk threshold. There may also be sensor anomalies, data drift, etc., and a secondary verification process should be activated. The secondary verification process includes: on-site inspection of equipment retesting or retesting after sensor health diagnosis.

[0029] In Example 2, Figure 2 The first sub-flowchart of the downhole environment dynamic monitoring method based on multi-source information fusion provided in this embodiment of the invention is shown. The following details the steps for collecting sensor data within the block, wherein the sensor data includes at least gas concentration, temperature and humidity, and dust concentration: S101: Divide the sensing data into several individual items and establish the correlation between the individual items.

[0030] The sensor data is divided into multiple individual items, such as gas concentration, temperature and humidity, and dust concentration. Based on the mutual influence between different individual items, the correlation between the individual items is established. The correlation is the linkage characteristic between the individual items. For example, there is a negative correlation between changes in gas concentration and changes in ventilation volume, and a positive correlation between dust concentration and work intensity.

[0031] S102: Based on the individual items and relationships, edit several evaluation rules, form a rule set, and send it to the edge device.

[0032] Based on the typical fluctuation range and correlation of individual items, evaluation rules are constructed. For example, one evaluation rule is: when the ambient temperature exceeds the upper limit of the typical fluctuation range within a short period of time, and the humidity decreases simultaneously, it is determined to be a thermal environment imbalance. All evaluation rules are summarized to generate a rule set, which is then sent to the edge device for storage.

[0033] In Example 3, Figure 3 The second sub-flowchart of the downhole environment dynamic monitoring method based on multi-source information fusion provided in this embodiment of the invention is shown. The following details the steps of acquiring risk events from historical data, extracting the change process of sensor data, marking risk characteristics, and constructing a training set: S201: Set the alarm level for each risk event, wherein the alarm levels include at least: high, medium and low.

[0034] Risks are classified into levels based on factors such as the severity, scope of impact, and speed of evolution of historical risk events. A corresponding alarm level is set for each risk event in the historical data, which includes high, medium, and low alarm levels.

[0035] S202: Insert the labels generated by the alarm levels into the training set.

[0036] Labels generated by alarm levels are inserted into manually labeled anomaly features in the training set, and a set of corresponding emergency response rules are set for each alarm level in order to achieve graded response and precise handling of downhole anomalies.

[0037] In Example 4, Figure 3 The diagram illustrates the second sub-process flowchart of the downhole environment dynamic monitoring method based on multi-source information fusion provided in this embodiment of the invention. The following details the steps of constructing a federated learning architecture, calculating the fusion parameter set, distributing it in parallel to all edge devices, collecting real-time values ​​of sensor data, inputting them into the environmental risk identification model within the corresponding edge device, and determining whether risk characteristics exist: S203: Set attribute data for each block, wherein the attribute data includes at least: job type and historical risk, and cluster the blocks into several groups based on the attribute data.

[0038] Determine the attribute data for each block, including: operation type and historical risk, etc. Based on the similarity of the attribute data, divide the block into several groups, with each block corresponding to a group. For example, divide the block into: mining operation group, transportation roadway group and return ventilation group.

[0039] S204: Establish the correspondence between the group and the fusion parameter group, and update the fusion model.

[0040] Federated learning is performed within each group, and a fusion model corresponding to each group is generated based on the parameter aggregation results.

[0041] In Example 5, Figure 4 The third sub-flow diagram of the downhole environment dynamic monitoring method based on multi-source information fusion provided by an embodiment of the present invention is shown. The following details the step of using a Bluetooth receiver pre-integrated in the edge device to collect the Bluetooth signal of the four-in-one gas detector worn by the worker and locate the worker's real-time position: S301: Select the test signal from the Bluetooth signal, extract the signal list of edge devices, and define the edge devices containing the test signal in the signal list as signal associated devices.

[0042] Select a test signal and iterate through the signal list of the Bluetooth receiver in the edge device. If the signal list contains the corresponding test signal, then define the edge device as a signal-associated device. It should be noted that the selection of the test signal does not have any practical meaning and is only used to intuitively describe the positioning process.

[0043] S302: Obtain the strength of the test signal in the signal association device, and locate the real-time position of the four-in-one gas detector corresponding to the test signal based on the multi-angle positioning algorithm.

[0044] The strength of the test signal in the signal-associated device is collected sequentially. A mapping relationship between signal strength and propagation distance is established using a multi-point signal strength attenuation model. By calculating the spatial geometric constraints of the test signal in different Bluetooth receivers, the real-time position of the four-in-one gas detector corresponding to the test signal is determined.

[0045] In Example 6, Figure 4 The diagram illustrates the third sub-process flow of the downhole environment dynamic monitoring method based on multi-source information fusion provided in this embodiment of the invention. The following details the steps of initializing the environmental risk identification model in the target block edge device, writing the fusion parameter group to obtain the fusion model, inputting environmental data into the fusion model, and cross-validating the judgment results of the environmental risk identification model: S303: Identify the triggering conditions for the fusion model update, wherein the triggering conditions include at least: event triggering and time triggering.

[0046] Set the trigger conditions for updating the fusion parameter group of the fusion model. The trigger conditions include event triggering and time triggering. Event triggering refers to actively triggering updates based on external changes or changes in the model's own performance. For example, when a sudden change in environmental state or a sensor device failure occurs, the model parameters in the fusion model are readjusted. Time triggering refers to updating the fusion parameter group according to a pre-set cycle.

[0047] S304: When the triggering condition is met, update the fusion model, record the change process of the fusion parameter group, extract key parameters, and generate an optimized version.

[0048] When the fusion parameter group is updated, the change process of the fusion parameter group is recorded. From the change process, the fusion parameter group before and after the update is extracted and defined as key parameters. A corresponding optimized version is generated for each set of key parameters, that is, the key parameters are stored in a versioned manner.

[0049] Figure 5 This diagram illustrates the structural block diagram of a downhole environment dynamic monitoring system based on multi-source information fusion provided in an embodiment of the present invention. The downhole environment dynamic monitoring system 1 based on multi-source information fusion includes: Deployment module 11 is used to delineate the underground area that needs to be dynamically monitored for the environment, divide the underground area into several blocks, use sensing devices pre-installed in the blocks to collect sensing data in the blocks, wherein the sensing data includes at least: gas concentration, temperature and humidity and dust concentration, create an environmental risk identification model, and deploy it to the edge devices pre-installed in the blocks. The judgment module 12 is used to obtain risk events in historical data, extract the change process of sensor data, mark risk features, build a training set, train the environmental risk identification model, obtain the model parameters of the environmental risk identification model in each edge device, construct a federated learning architecture, calculate the fusion parameter set, distribute it in parallel to all edge devices, collect the real-time value of sensor data, input it into the environmental risk identification model in the corresponding edge device, and determine whether there are risk features. The verification module 13 is used to collect the Bluetooth signal of the four-in-one gas detector worn by the staff using a Bluetooth receiver pre-integrated in the edge device, locate the staff's real-time position, select several sample points, collect environmental data at each sample point using the four-in-one gas detector, define the block corresponding to the real-time position as the target block, initialize the environmental risk identification model in the edge device of the target block, write it into the fusion parameter group to obtain the fusion model, input the environmental data into the fusion model, and cross-validate the judgment results of the environmental risk identification model.

[0050] Figure 6 This diagram illustrates the composition of deployment module 11 in a downhole environment dynamic monitoring system based on multi-source information fusion provided in an embodiment of the present invention. Deployment module 11 includes: Establishment unit 111 is used to divide the sensing data into several individual items and establish the correlation between the individual items; The sending unit 112 is used to edit several evaluation rules, form a rule set, and send it to the edge device via the individual items and the association relationship.

[0051] Figure 7 This diagram illustrates the structural composition of the judgment module 12 in a downhole environment dynamic monitoring system based on multi-source information fusion provided in an embodiment of the present invention. The judgment module 12 includes: Setting unit 121 is used to set the alarm level for each risk event, wherein the alarm levels include at least: high, medium and low; Insertion unit 122 is used to insert labels generated by alarm levels into the training set; Clustering unit 123 is used to set attribute data for each block, wherein the attribute data includes at least: job type and historical risk, and the blocks are clustered into several groups based on the attribute data; The update unit 124 is used to establish the correspondence between the group and the fusion parameter group and update the fusion model.

[0052] Figure 8 This diagram illustrates the structural composition of the verification module 13 in the downhole environment dynamic monitoring system based on multi-source information fusion provided in an embodiment of the present invention. The verification module 13 includes: The definition unit 131 is used to select the test signal from the Bluetooth signal, extract the signal list of the edge device, and define the edge device containing the test signal in the signal list as the signal associated device. The acquisition unit 132 is used to acquire the strength of the test signal in the signal association device and locate the real-time position of the four-in-one gas detector corresponding to the test signal based on the multi-angle positioning algorithm. The identification unit 133 is used to identify the triggering conditions for updating the fusion model, wherein the triggering conditions include at least: event triggering and time triggering; Extraction unit 134 is used to update the fusion model when the triggering condition is met, record the change process of the fusion parameter group, extract key parameters, and generate an optimized version.

[0053] The deployment module 11 is mainly used to complete step S100, the judgment module 12 is mainly used to complete step S200, and the verification module 13 is mainly used to complete step S300. The establishment unit 111 is mainly used to complete step S101, and the sending unit 112 is mainly used to complete step S102; The setting unit 121 is mainly used to complete step S201, the insertion unit 122 is mainly used to complete step S202, the clustering unit 123 is mainly used to complete step S203, and the update unit 124 is mainly used to complete step S204. The definition unit 131 is mainly used to complete step S301, the acquisition unit 132 is mainly used to complete step S302, the identification unit 133 is mainly used to complete step S303, and the extraction unit 134 is mainly used to complete step S304.

[0054] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0055] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic monitoring of the downhole environment based on multi-source information fusion, characterized in that, The method includes: The downhole area that needs to be dynamically monitored for the environment is delineated, and the downhole area is divided into several blocks. Sensing devices pre-installed in the blocks are used to collect sensing data within the blocks. The sensing data includes at least: gas concentration, temperature and humidity and dust concentration. An environmental risk identification model is created and deployed to the edge devices pre-installed in the blocks. Risk events are obtained from historical data, the change process of sensor data is captured, risk characteristics are marked, a training set is built, and the environmental risk identification model is trained. The model parameters of the environmental risk identification model in each edge device are obtained, a federated learning architecture is constructed, the fusion parameter set is calculated, and it is distributed to all edge devices in parallel. Real-time values ​​of sensor data are collected and input into the environmental risk identification model in the corresponding edge device to determine whether risk characteristics exist. Using a Bluetooth receiver pre-integrated in the edge device, the Bluetooth signal of the four-in-one gas detector worn by the staff is collected to locate the staff's real-time position. Several sample points are selected, and environmental data at each sample point is collected using the four-in-one gas detector. The block corresponding to the real-time position is defined as the target block. The environmental risk identification model in the edge device of the target block is initialized and written into the fusion parameter group to obtain the fusion model. The environmental data is input into the fusion model, and the judgment results of the environmental risk identification model are cross-validated.

2. The method for dynamic monitoring of the downhole environment based on multi-source information fusion according to claim 1, characterized in that, The step of collecting sensor data within the segment, wherein the sensor data includes at least: gas concentration, temperature and humidity, and dust concentration, includes: The sensor data is divided into several individual items, and the correlation between the individual items is established; Based on the individual items and relationships, several evaluation rules are edited to form a rule set, which is then sent to the edge device.

3. The method for dynamic monitoring of the downhole environment based on multi-source information fusion according to claim 1, characterized in that, The steps of acquiring risk events from historical data, extracting the change process of sensor data, labeling risk characteristics, and constructing a training set include: Set an alarm level for each risk event, wherein the alarm levels include at least: high, medium and low; Insert labels generated from alarm levels into the training set.

4. The method for dynamic monitoring of the downhole environment based on multi-source information fusion according to claim 1, characterized in that, The steps of constructing a federated learning architecture, calculating a fusion parameter set, distributing it in parallel to all edge devices, collecting real-time values ​​of sensor data, inputting them into the environmental risk identification model within the corresponding edge device, and determining whether risk characteristics exist include: Set attribute data for each block, wherein the attribute data includes at least: job type and historical risk, and cluster the blocks into several groups based on the attribute data; Establish the correspondence between groups and fusion parameter groups, and update the fusion model.

5. The method for dynamic monitoring of the downhole environment based on multi-source information fusion according to claim 2, characterized in that, The step of using a Bluetooth receiver pre-integrated in the edge device to collect the Bluetooth signal of the four-in-one gas detector worn by the worker and to locate the worker's real-time position includes: From the Bluetooth signals, select the test signal, extract the signal list of edge devices, and define the edge devices in the signal list that contain the test signal as signal-associated devices; The strength of the test signal in the signal-correlated device is obtained, and the real-time position of the four-in-one gas detector corresponding to the test signal is located based on the multi-angle positioning algorithm.

6. The method for dynamic monitoring of the downhole environment based on multi-source information fusion according to claim 4, characterized in that, The steps of initializing the environmental risk identification model in the target block edge device, writing it into the fusion parameter group to obtain the fusion model, inputting environmental data into the fusion model, and cross-validating the judgment results of the environmental risk identification model include: Identify the triggering conditions for updating the fusion model, wherein the triggering conditions include at least: event triggering and time triggering; When the triggering condition is met, the fusion model is updated, the change process of the fusion parameter group is recorded, key parameters are extracted, and an optimized version is generated.

7. A dynamic monitoring system for the downhole environment based on multi-source information fusion, characterized in that, The system includes: The deployment module is used to delineate the underground area that needs to be dynamically monitored for the environment, divide the underground area into several blocks, use the sensing devices pre-installed in the blocks to collect sensing data in the blocks, wherein the sensing data includes at least: gas concentration, temperature and humidity and dust concentration, create an environmental risk identification model, and deploy it to the edge devices pre-installed in the blocks. The judgment module is used to obtain risk events in historical data, extract the change process of sensor data, mark risk features, build a training set, train the environmental risk identification model, obtain the model parameters of the environmental risk identification model in each edge device, construct a federated learning architecture, calculate the fusion parameter set, distribute it in parallel to all edge devices, collect the real-time value of sensor data, input it into the environmental risk identification model in the corresponding edge device, and determine whether there are risk features. The verification module uses a Bluetooth receiver pre-integrated in the edge device to collect the Bluetooth signal of the four-in-one gas detector worn by the staff, locate the staff's real-time position, select several sample points, collect environmental data at each sample point using the four-in-one gas detector, define the block corresponding to the real-time position as the target block, initialize the environmental risk identification model in the edge device of the target block, write it into the fusion parameter group to obtain the fusion model, input the environmental data into the fusion model, and cross-validate the judgment results of the environmental risk identification model.

8. The downhole environment dynamic monitoring system based on multi-source information fusion according to claim 7, characterized in that, The deployment module includes: A unit is established to divide the sensing data into several individual items and establish the relationship between the individual items; The sending unit is used to edit several evaluation rules, form a rule set, and send it to the edge device via the individual items and associations.

9. The downhole environment dynamic monitoring system based on multi-source information fusion according to claim 7, characterized in that, The judgment module includes: The setting unit is used to set the alarm level for each risk event, wherein the alarm levels include at least: high, medium and low; An insertion unit is used to insert labels generated by alarm levels into the training set; Clustering unit, used to set attribute data for each block, wherein the attribute data includes at least: job type and historical risk, and based on the attribute data, the blocks are clustered into several groups; The update unit is used to establish the correspondence between the group and the fusion parameter group and update the fusion model.

10. The downhole environment dynamic monitoring system based on multi-source information fusion according to claim 8, characterized in that, The verification module includes: The definition unit is used to select the test signal from the Bluetooth signal, extract the signal list of the edge device, and define the edge device containing the test signal in the signal list as the signal associated device. The acquisition unit is used to acquire the strength of the test signal in the signal association device and locate the real-time position of the four-in-one gas detector corresponding to the test signal based on the multi-angle positioning algorithm. The identification unit is used to identify the triggering conditions for updating the fusion model, wherein the triggering conditions include at least: event triggering and time triggering; The extraction unit is used to update the fusion model when the triggering condition is met, record the change process of the fusion parameter group, extract key parameters, and generate an optimized version.

Images