Real-time monitoring system for intelligent comprehensive management and control platform of coal mine

Abstract

The present application belongs to the technical field of real-time monitoring system, and particularly relates to a real-time monitoring system for a coal mine intelligent comprehensive management and control platform, comprising a sensor health management module, which is used for identifying whether a sensor is abnormal by dynamic baseline modeling and multi-source data checking on monitoring parameters, and triggering a closed-loop self-calibration process when an abnormality is identified, and triggering a data fusion and compensation module when the closed-loop self-calibration process is verified as invalid; the data fusion and compensation module is connected to the sensor health management module; a system fault tolerance and reconstruction module is connected to the data fusion and compensation module. The three-level progressive fault tolerance architecture formed by the sensor health management module, the data fusion and compensation module and the system fault tolerance and reconstruction module realizes self-calibration, data fusion and virtual reconstruction, and ensures that the system can maintain uninterrupted monitoring functions at different stages from abnormality to complete failure of the sensor.

Description

Technical Field

[0001] This invention belongs to the field of real-time monitoring system technology, specifically relating to a real-time monitoring system for an intelligent integrated management and control platform for coal mines. Background Technology

[0002] The underground coal mine environment is extremely complex. The reliability and stability of the numerous sensors deployed in this environment to monitor key parameters such as gas concentration, carbon monoxide, wind speed, temperature, pressure, and equipment status face severe challenges in long-term operation. Most existing coal mine safety monitoring systems employ alarm mechanisms based on fixed thresholds and simple redundancy backup strategies, which have the following significant drawbacks:

[0003] In sensor health management, traditional methods cannot effectively distinguish between real environmental parameter anomalies and data anomalies caused by sensor malfunctions. Systems typically only trigger alarms when data severely exceeds limits, but cannot determine whether this stems from sensor failure. Furthermore, there is a lack of effective early warning mechanisms for slow sensor performance degradation (such as gradual drift).

[0004] In terms of data anomaly handling, existing technologies are limited. When a sensor's data is abnormal, the common practice is to directly discard it and simply replace it with data from adjacent sensors or historical averages. This crude approach completely ignores the spatiotemporal correlation of monitoring parameters downhole and the complex physical laws (such as the mechanisms of gas outbursts and airflow). The reconstructed data has poor reliability, unclear physical meaning, and may mask the true danger or trigger false alarms, failing to provide high-quality and reliable data support for intelligent management and control platforms.

[0005] In terms of system-level fault tolerance, existing systems typically only have hardware redundancy at the device level (such as backup sensors). However, downhole space is limited and wiring is difficult, making it impossible to deploy hardware backups for every sensor. When the primary sensor completely fails, its monitoring function is completely lost, creating a monitoring blind spot until maintenance personnel replace the equipment.

[0006] Therefore, there is an urgent need for a new real-time monitoring system that can intelligently sense the health status of sensors, automatically reconstruct reliable data when anomalies occur, and maintain its monitoring function even after the sensors have completely failed, in order to overcome the above-mentioned technical bottlenecks and comprehensively improve the safety assurance capabilities and intelligent management level of coal mines. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a real-time monitoring system for an intelligent integrated management and control platform for coal mines. This system aims to solve the problem of distorted or interrupted monitoring data caused by faulty sensors in underground coal mines. Through a three-level fault-tolerant mechanism consisting of intelligent diagnosis, data reconstruction, and functional reconstruction, the system ensures the data reliability and functional continuity of the monitoring system.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A real-time monitoring system for an intelligent integrated management and control platform in coal mines includes:

[0010] The sensor health management module is used to identify whether the sensor is abnormal by performing dynamic baseline modeling and multi-source data verification on the monitoring parameters. When an abnormality is detected, a closed-loop self-calibration process is triggered. When the closed-loop self-calibration process is invalid, the data fusion and compensation module is triggered.

[0011] The data fusion and compensation module is connected to the sensor health management module and is used to reconstruct a reliable monitoring data stream based on spatiotemporal feature correlation and multi-source information fusion methods, combined with digital twin simulation data, after being triggered.

[0012] The system fault tolerance and reconstruction module is connected to the data fusion and compensation module. When the data fusion and compensation module cannot effectively reconstruct the data, it determines that the sensor has completely failed and generates alternative monitoring data through virtual sensing reconstruction to maintain the continuity of the monitoring function.

[0013] The sensor health management module includes:

[0014] The dynamic baseline construction unit is used to build a sensor degradation trend model based on the device's historical operating data and dynamically update the health reference threshold;

[0015] An anomaly diagnosis and verification unit, connected to the dynamic baseline construction unit, performs consistency checks and physical rule violation detection on multi-source monitoring data based on the health reference threshold and through the multi-source data verification mechanism to identify abnormal sensors.

[0016] The closed-loop calibration and verification unit, connected to the anomaly diagnosis and verification unit, is used to trigger a calibration operation when an abnormal sensor is detected, and to verify the validity of the calibrated data. If the verification is valid, the sensor is determined to have returned to normal monitoring status. If the verification is invalid, the sensor is determined to be in a severely abnormal state, and the data fusion and compensation module is triggered to reconstruct the data.

[0017] The data fusion and compensation module includes:

[0018] The spatiotemporal feature modeling unit is used to extract the spatiotemporal correlation features of monitoring parameters and fuse digital twin simulation data to construct a physical relationship model between sensors.

[0019] The multi-source weighted fusion unit, connected to the spatiotemporal feature modeling unit, is used to acquire multi-source data from spatially associated sensors, digital twins, and related historical databases, and dynamically allocate fusion weights according to the reliability of each data source and the physical relationship model, and perform weighted fusion calculations accordingly to output preliminary fusion results.

[0020] The error correction unit, connected to the multi-source weight fusion unit, performs residual learning and physical consistency correction on the preliminary fusion result based on the simulation output of the digital twin, and outputs the reconstructed reliable monitoring data stream.

[0021] The system fault tolerance and reconfiguration module includes:

[0022] The failure determination unit is used to respond to the signal that the data fusion and compensation module cannot effectively reconstruct the data, determine that the sensor has completely failed and issue a failure command.

[0023] The fault isolation unit, connected to the failure determination unit, is used to locate the fault node of the sensor according to the failure command and the sensor network topology, and analyze the potential propagation path of the fault.

[0024] The dynamic reconfiguration unit, connected to the fault isolation unit, is used to select the optimal redundant communication path based on the analysis results of the fault node's position and potential propagation path, and to dynamically reconfigure the monitoring network based on sensor reliability records.

[0025] The virtual sensing unit, connected to the failure determination unit and the dynamic reconstruction unit, is used to generate alternative virtual monitoring data based on the failure command and the data access points provided by the reconstructed monitoring network, through digital twin and machine learning methods, in order to maintain the continuity of monitoring functions.

[0026] The dynamic baseline construction unit uses generative adversarial networks to model normal operating conditions and detect anomalies, thereby establishing the sensor degradation trend model and dynamically updating the health reference threshold.

[0027] To construct a physical relationship model, the spatiotemporal feature modeling unit embeds the device operation physical equations as neural network constraints.

[0028] The dynamic reconfiguration unit integrates path planning functions to select the optimal redundant communication path and integrates computing resource allocation functions to ensure the operational efficiency of the reconfigured monitoring network.

[0029] Compared with the prior art, the beneficial effects of this invention are:

[0030] Through a three-tiered progressive fault-tolerant architecture consisting of a sensor health management module, a data fusion and compensation module, and a system fault tolerance and reconstruction module, self-calibration, data fusion, and virtual reconstruction are achieved, ensuring that the system can maintain uninterrupted monitoring functions at different stages of sensor failure, from abnormality to complete failure.

[0031] Dynamic baseline modeling enables adaptive sensor performance degradation, accurate anomaly identification, and avoids false alarms or missed alarms caused by fixed thresholds. Combined with a multi-source data verification mechanism, cross-validation is performed from both data consistency and physical rules perspectives, improving the accuracy of anomaly diagnosis.

[0032] By using an error correction unit to perform residual learning and physical rule correction, the reconstructed data is ensured to conform to the inherent physical laws of equipment operation, thus avoiding the distortion of physical meaning that may occur from pure mathematical fusion. Attached Figure Description

[0033] Figure 1 This is a connection block diagram of the present invention. Detailed Implementation

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0035] like Figure 1 As shown, a real-time monitoring system for an intelligent integrated management and control platform for coal mines includes three core modules: a sensor health management module, a data fusion and compensation module, and a system fault tolerance and reconstruction module.

[0036] The sensor health management module monitors the raw monitoring parameters (such as concentration, Celsius, and kPa) collected by various sensors (such as gas, temperature, and pressure sensors) in real time.

[0037] The module's built-in algorithm performs dynamic baseline modeling on these parameter sequences, specifically using time series analysis algorithms to establish normal value ranges that vary with the production cycle. It also performs multi-source data verification, such as cross-comparing readings from sensors based on different principles at the same location, to intelligently determine if the sensor itself is malfunctioning. Once an anomaly is detected, a closed-loop self-calibration process is initiated, such as triggering the built-in calibration program or sending a command to the smart sensor with self-calibration capabilities. If self-calibration fails (verification invalid), the module generates a trigger signal and invokes the data fusion and compensation module.

[0038] The data fusion and compensation module is activated upon receiving a trigger signal from the health management module. It no longer relies on data from sensors deemed abnormal; instead, it utilizes its built-in algorithm, based on spatiotemporal feature correlation and multi-source information fusion methods, combined with digital twin simulation data, to reconstruct missing or unreliable data streams, outputting a reliable monitoring data stream for use by upper-layer applications.

[0039] Spatiotemporal correlation refers to analyzing the spatiotemporal correlation of data from different sensors on the same hydraulic support assembly. Multi-source information fusion methods can employ techniques such as Kalman filtering and DS evidence theory. Digital twin simulation data can be obtained from the digital twin model of a coal mine scenario, representing theoretical simulation values ​​under the current operating conditions.

[0040] The system fault tolerance and reconstruction module is the last resort mechanism. Its trigger condition is that the data fusion and compensation module also fails to effectively reconstruct the data (e.g., the sensor is completely damaged and there is insufficient data from multiple sources for fusion). In this case, the module determines that the sensor has completely failed. Subsequently, it will no longer attempt to repair the data, but will instead activate the virtual sensing reconstruction function. This function uses other relevant data to generate alternative monitoring data through mechanistic models or AI algorithms, thus maintaining the continuity of monitoring functions and ensuring uninterrupted system operation even in the event of hardware failure.

[0041] Furthermore, the sensor health management module includes a dynamic baseline construction unit, an anomaly diagnosis and verification unit, and a closed-loop calibration and verification unit.

[0042] Among them, the dynamic baseline construction unit continuously reads the historical operating data of the device (long-term data stored in a real-time database), trains and continuously updates a sensor degradation trend model online through machine learning algorithms (such as regression models and LSTM networks), and predicts and dynamically updates the health reference threshold (such as the allowable drift range) of each sensor. This threshold is no longer a fixed value, but adapts to change as the device ages.

[0043] The anomaly diagnosis and verification unit interacts directly with the dynamic baseline construction unit, receiving multi-source monitoring data streams in real time and using the latest health reference threshold calculated by the dynamic baseline construction unit as the judgment benchmark. It performs a two-step diagnosis through a multi-source data verification mechanism:

[0044] Consistency check: Compare whether the data from sensors in similar spatial locations exceed the reasonable range of difference;

[0045] Physical rule violation detection: Determine whether the data violates known physical laws, such as a sudden drop in pressure while the temperature remains unchanged; thereby identifying abnormal sensors.

[0046] The closed-loop calibration and verification unit works in conjunction with the anomaly diagnosis unit. When an anomaly is identified, this unit sends calibration commands to the sensor via the communication interface (such as activating the self-cleaning function or resetting the zero point). After completion, it verifies the validity of the calibrated data (e.g., checking if the data has returned to its normal fluctuation range). If the verification is valid, the sensor is determined to have returned to normal monitoring status; if the verification is invalid, the sensor is determined to be in an unreliable, severely abnormal state, and a command is immediately sent to the data fusion and compensation module to initiate the data reconstruction process.

[0047] Furthermore, the dynamic baseline construction unit implements its functionality using a Generative Adversarial Network (GAN). Specifically, the GAN is trained using normal operating condition data from a large amount of historical equipment operation data. Its generator learns the distribution of normal data for normal operating condition modeling; its discriminator learns to distinguish between real normal data and generated data for anomaly detection. That is, if the discriminator judges real-time data as generated with a high probability, it is considered an anomaly. Through the adversarial training process of the GAN, a more refined sensor degradation trend model can be established, and the health reference threshold can be dynamically updated more sensitively, thereby achieving higher-precision health management.

[0048] Furthermore, the data fusion and compensation module includes a spatiotemporal feature modeling unit, a multi-source weight fusion unit, and an error correction unit.

[0049] The spatiotemporal feature modeling unit extracts spatiotemporal correlation features from monitoring parameters. Specifically, it can use graph neural networks to model the spatial graph and temporal dependencies between sensors. By integrating digital twin simulation data, such as fluid dynamics simulation results, as prior knowledge, a high-fidelity physical relationship model between sensors can be constructed.

[0050] The multi-source weighted fusion unit is connected to the spatiotemporal feature modeling unit, acquiring data from three sources: data from spatially correlated sensors (physically close); real-time simulation data from the digital twin; and data from historical databases under similar operating conditions. Then, based on the reliability of each data source and the aforementioned physical relationship model, fusion weights are dynamically allocated. Finally, these weights are used to perform weighted fusion calculations, outputting a preliminary fusion result.

[0051] Specifically, the reliability assessment criteria for each data source are as follows:

[0052] For spatially correlated sensors, their reliability weight is directly calculated and output in real time by the signal quality score of the anomaly diagnosis and verification unit in the sensor health management module.

[0053] For digital twin simulation data and historical data, their reliability (historical accuracy) is maintained by a built-in offline reputation assessment subunit. This subunit calculates the long-term statistical accuracy and confidence interval by periodically comparing the historical outputs of these data sources with rigorously validated benchmark truth values ​​(such as data recorded by high-precision instruments when the system is stable or data that has been ultimately manually verified).

[0054] The error correction unit receives the preliminary fusion results and uses the simulation output of the digital twin as a golden reference to perform residual learning on the preliminary results. It can predict and eliminate system errors through a correction model; and perform physical consistency correction to ensure that the results comply with physical rules such as mass conservation and energy conservation; and finally outputs the reconstructed reliable monitoring data stream.

[0055] Furthermore, to construct a more accurate physical relationship model, the spatiotemporal feature modeling unit embeds the physical equations of device operation as constraints for the neural network when extracting spatiotemporal features using a neural network. For example, a physical consistency penalty term is added to the loss function of the training graph neural network, requiring that the network's output predictions must approximately satisfy the relevant gas seepage equations or thermodynamic equations. This physics-guided machine learning approach ensures that the learned model not only fits the data but also strictly adheres to physical laws, thereby improving its generalization ability and reliability under extreme operating conditions.

[0056] Furthermore, the system fault tolerance and reconfiguration module includes a failure determination unit, a fault isolation unit, a dynamic reconfiguration unit, and a virtual sensing unit.

[0057] The failure determination unit is used to respond to signals that the data fusion and compensation module cannot effectively reconstruct the data, determine that the sensor has completely failed, and issue a failure command.

[0058] The fault isolation unit receives a failure command and uses pre-set sensor network topology information to locate the faulty node position of the failed sensor in the cluster, such as determining that it is a critical node in a certain measurement area; it uses graph theory algorithms to analyze the potential propagation path of the fault and predict which related sensors or subsystems may be affected.

[0059] The dynamic reconfiguration unit selects the optimal redundant communication path based on the analysis results of the fault isolation unit. Specifically, based on the location of the faulty node and its potential propagation path, it selects the optimal path from a pre-set redundant communication path library, such as choosing the backup network path with the lowest load and least latency. Simultaneously, it calls upon sensor reliability records, prioritizing the selection of highly reliable sensors to form new data links, dynamically reorganizing the monitoring network, and bypassing the faulty point.

[0060] The virtual sensing unit receives a failure command and identifies which sensor needs to be replaced. Simultaneously, it obtains new data access points from the reorganized monitoring network, determining which working sensors are receiving input data. Then, using this data, and through digital twin and machine learning methods such as a trained neural network model, it generates alternative virtual monitoring data and injects it into the data stream to maintain continuous monitoring functionality.

[0061] Furthermore, the dynamic reconfiguration unit integrates path planning functionality. Its core algorithms (such as Dijkstra's algorithm) are specifically designed to select the optimal redundant communication path in complex network topologies to ensure real-time data transmission. Simultaneously, it integrates computing resource allocation functionality, dynamically allocating computing resources from edge computing nodes or cloud platforms based on the complexity of the reconfiguration task to ensure the smooth operation of new data stream processing links, thereby guaranteeing the operational performance of the reconfigured monitoring network.

[0062] The above description only illustrates preferred embodiments of the present invention, but the present invention is not limited to the above embodiments.

Claims

1. A real-time monitoring system for an intelligent integrated management and control platform in coal mines, characterized in that, include: The sensor health management module is used to identify whether the sensor is abnormal by performing dynamic baseline modeling and multi-source data verification on the monitoring parameters. When an abnormality is detected, a closed-loop self-calibration process is triggered. When the closed-loop self-calibration process is invalid, the data fusion and compensation module is triggered. The data fusion and compensation module is connected to the sensor health management module and is used to reconstruct a reliable monitoring data stream based on spatiotemporal feature correlation and multi-source information fusion methods, combined with digital twin simulation data, after being triggered. The data fusion and compensation module includes: a spatiotemporal feature modeling unit, used to extract the spatiotemporal correlation features of monitoring parameters and fuse digital twin simulation data to construct a physical relationship model between sensors; The multi-source weighted fusion unit, connected to the spatiotemporal feature modeling unit, is used to acquire multi-source data from spatially associated sensors, digital twins, and related historical databases, and dynamically allocate fusion weights according to the reliability of each data source and the physical relationship model, and perform weighted fusion calculations accordingly to output preliminary fusion results. An error correction unit, connected to the multi-source weight fusion unit, performs residual learning and physical consistency correction on the preliminary fusion result based on the simulation output of the digital twin, and outputs a reconstructed reliable monitoring data stream. The system fault tolerance and reconstruction module is connected to the data fusion and compensation module. When the data fusion and compensation module cannot effectively reconstruct the data, it determines that the sensor has completely failed and generates alternative monitoring data through virtual sensing reconstruction to maintain the continuity of the monitoring function.

2. The real-time monitoring system for an intelligent integrated management and control platform for coal mines according to claim 1, characterized in that, The sensor health management module includes: a dynamic baseline construction unit, used to establish a sensor degradation trend model based on the device's historical operating data and dynamically update the health reference threshold; An anomaly diagnosis and verification unit, connected to the dynamic baseline construction unit, performs consistency checks and physical rule violation detection on multi-source monitoring data based on the health reference threshold and through the multi-source data verification mechanism to identify abnormal sensors. The closed-loop calibration and verification unit, connected to the anomaly diagnosis and verification unit, is used to trigger a calibration operation when an abnormal sensor is detected, and to verify the validity of the calibrated data. If the verification is valid, the sensor is determined to have returned to normal monitoring status. If the verification is invalid, the sensor is determined to be in a severely abnormal state, and the data fusion and compensation module is triggered to reconstruct the data.

3. The real-time monitoring system for an intelligent integrated management and control platform for coal mines according to claim 1, characterized in that, The system fault tolerance and reconstruction module includes: a failure determination unit, which is used to respond to the signal that the data fusion and compensation module cannot effectively reconstruct the data, determine that the sensor has completely failed and issue a failure command; The fault isolation unit, connected to the failure determination unit, is used to locate the fault node of the sensor according to the failure command and the sensor network topology, and analyze the potential propagation path of the fault. The dynamic reconfiguration unit, connected to the fault isolation unit, is used to select the optimal redundant communication path based on the analysis results of the fault node's position and potential propagation path, and to dynamically reconfigure the monitoring network based on sensor reliability records. The virtual sensing unit, connected to the failure determination unit and the dynamic reconstruction unit, is used to generate alternative virtual monitoring data based on the failure command and the data access points provided by the reconstructed monitoring network, through digital twin and machine learning methods, in order to maintain the continuity of monitoring functions.

4. The real-time monitoring system for an intelligent integrated management and control platform for coal mines according to claim 2, characterized in that, The dynamic baseline construction unit uses generative adversarial networks to model normal operating conditions and detect anomalies, thereby establishing the sensor degradation trend model and dynamically updating the health reference threshold.

5. The real-time monitoring system for an intelligent integrated management and control platform for coal mines according to claim 1, characterized in that, To construct a physical relationship model, the spatiotemporal feature modeling unit embeds the device operation physical equations as neural network constraints.

6. The real-time monitoring system for an intelligent integrated management and control platform for coal mines according to claim 3, characterized in that, The dynamic reconfiguration unit integrates path planning functions to select the optimal redundant communication path and integrates computing resource allocation functions to ensure the operational efficiency of the reconfigured monitoring network.

Images