Gas extraction multi-parameter monitoring method and system based on edge calculation

Abstract

The invention provides a gas extraction multi-parameter monitoring method and system based on edge calculation, and relates to the technical field of coal mine gas extraction. The method comprises the following steps: arranging multi-parameter sensor nodes in drill holes, pipelines and gas gathering stations, and accessing an edge calculation unit; performing time synchronization, zero drift correction, outlier elimination and normalization on the data of the gas concentration, the negative pressure, the flow, the temperature, the humidity and the hydrogen sulfide concentration to generate feature vectors; calling a lightweight neural network model for reasoning, and outputting an extraction efficiency score and a leakage risk degree; when the risk degree exceeds the limit, an acousto-optic alarm is triggered and a speed reduction instruction is issued; when the efficiency is insufficient, the negative pressure is adjusted; compressing and encrypting the result, and reporting the result to a cloud platform through a wireless network at regular time; and the cloud performs model incremental training based on historical data and feedback, and issues an update model through hot replacement. According to the invention, local intelligent analysis, rapid early warning and dynamic adaptive optimization are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine gas extraction, and in particular to a multi-parameter monitoring method and system for gas extraction based on edge computing. Background Art

[0002] Coal mine gas extraction is a key step in reducing the risk of underground gas disasters and achieving green and safe mining. Existing monitoring systems typically deploy single or a small number of parameter sensors, such as gas concentration, negative pressure (hole vacuum), and flow, near extraction pipelines or boreholes, and upload the raw data directly to a ground information center for centralized processing via wired or industrial Wi-Fi. Due to long transmission links and strong underground interference, monitoring systems generally suffer from large data delays, high packet loss rates, and a lack of local early warning, making it difficult to support the real-time needs of integrated dynamic control of "extraction-ventilation-sealing."

[0003] With the gradual deployment of 5G, Industrial Ethernet, and Low-Power Wide Area Networks (LPWANs) in coal mines, multi-parameter sensing, distributed computing, and AI analysis are becoming mainstream trends in the coal mine IoT. Edge computing brings lightweight inference models underground, enabling local data denoising, fusion, and discrimination, significantly reducing bandwidth usage and cloud load. Multimodal sensor fusion and federated learning are driving the evolution of gas extraction monitoring towards "advanced sensing, local autonomy, and global collaboration," laying the foundation for proactive early warning and adaptive extraction strategies.

[0004] Currently, most publicly available gas extraction monitoring solutions adopt an "upload-centralize-download" model, which has the following problems: (1) reliance on ground servers, lack of real-time decision-making capabilities at the edge, and failure once communication is interrupted; (2) limited parameter dimensions, unable to comprehensively evaluate extraction efficiency and leakage risk; (3) inconsistent sensor manufacturers, incompatible data formats and time bases, and difficulty in flexible expansion; (4) lack of online model updates and secure encryption mechanisms, making it difficult for the system to cope with changes in working conditions and network attacks. Therefore, a creative method that can complete multi-parameter fusion analysis at the edge of the well, with model self-update and secure transmission capabilities, is urgently needed. Summary of the Invention

[0005] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a multi-parameter monitoring method and system for gas extraction based on edge computing, which overcomes the defects of the existing technology such as reliance on the cloud, delayed response and untimely model updates, and realizes local intelligent analysis, rapid early warning and dynamic adaptive optimization of multi-parameter monitoring during gas extraction.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A multi-parameter monitoring method for gas extraction based on edge computing, including:

[0008] Deploy a number of sensor nodes in gas extraction boreholes, pipelines, and gas gathering stations in the target area, and utilize the sensor nodes to collect raw monitoring parameters composed of multiple parameters; each of the sensor nodes is connected to an edge computing unit disposed around the sensor node via an industrial field bus;

[0009] In the edge computing unit, the received original monitoring parameters are time synchronized, zero drift corrected, outlier removed, and dimensionally normalized to form a multi-parameter feature vector;

[0010] Calling a pre-deployed lightweight deep neural network model to infer the multi-parameter feature vector and output an extraction efficiency score and leakage risk;

[0011] When the leakage risk is greater than a first threshold, the edge computing unit is used to trigger a local sound and light alarm and issue a speed reduction instruction to the nearest extraction pump inverter; when the extraction efficiency score is lower than a second threshold, the edge computing unit is used to adjust the extraction negative pressure set value;

[0012] In the edge computing unit, an elliptic curve encryption algorithm is used to compress and encrypt the multi-parameter feature vector, the extraction efficiency score, and the leakage risk level, and the data are sent to the central cloud platform via a wireless network according to a set upload cycle;

[0013] In the central cloud platform, incremental training is performed on the lightweight deep neural network model based on historical monitoring data and downhole feedback results to generate an updated model, and the updated model is sent to each edge computing unit through a secure channel to complete hot replacement.

[0014] Preferably, the original monitoring parameters include: gas concentration, extraction negative pressure, gas flow, temperature, humidity and hydrogen sulfide concentration.

[0015] Preferably, the historical monitoring data includes a time series constructed by the multi-parameter feature vector, the extraction efficiency score and the leakage risk; the downhole feedback results include alarm response records and efficiency change trends after extraction adjustment.

[0016] Preferably, in the edge computing unit, the received original monitoring parameters are subjected to time synchronization, zero drift correction, outlier removal and dimension normalization to form a multi-parameter feature vector, including:

[0017] Generating a parameter sequence with consistent timestamps for each type of the original monitoring parameters based on a unified timing mechanism;

[0018] The corresponding zero offset values of the parameter sequences of gas concentration C, extraction negative pressure P, and gas flow rate Q are obtained by looking up the table respectively. Then, the parameter values are corrected item by item in combination with the ambient temperature compensation factor to obtain the corrected parameter sequence C′, P′, Q′.

[0019] The calibrated C′, P′, Q′ and the original temperature T, humidity H, and hydrogen sulfide concentration S parameter sequences were tested and outliers were removed using the sliding window median standard deviation method to obtain the processed six parameter sequences. Each sliding window contains N time points. If the parameter value deviates from the median of the sliding window by more than a preset multiple standard deviation threshold, it is repaired by linear interpolation using adjacent valid data.

[0020] The six types of parameter sequences after the processing are uniformly normalized and mapped to the interval [-1, 1], and the multi-parameter feature vector at time point t is constructed using the six-dimensional data at time point t.

[0021] Preferably, the calculation formula of the sliding window median standard deviation method is:

[0022]

[0023] Among them, x t is the monitoring value at the current moment; μ t =median(W t ) is a sliding window W with length N and centered at t. t The median of the data; σ t =1.4826×median{|w-μ t |:w∈W t} is the robust standard deviation estimate of the absolute deviation of the same window data relative to the median after transformation by the normalization coefficient 1.4826; N is the number of sample points in the sliding window, n is an odd number, with a typical value of 11 or 21); It is an adaptive multiplication factor derived from the universal threshold of Gaussian noise and increases monotonically with the window size.

[0024] Preferably, a pre-deployed lightweight deep neural network model is called to infer the multi-parameter feature vector and output the extraction efficiency score and leakage risk, including:

[0025] Before loading the lightweight deep neural network model M, the edge computing unit calculates the SHA-256 hash value of the model file and compares it with the pre-stored reference value. After verification, the inference process is carried out;

[0026] Encapsulating the normalized multi-parameter feature vector into a 1×6 floating-point tensor and writing it into the input cache of the local inference engine;

[0027] Call the inference engine to execute the shared backbone network and dual-task output head of the lightweight deep neural network model M to obtain the efficiency logarithm z E and leakage logarithm z R ;

[0028] Calculate the extraction efficiency score E t =100×σ(z E ); where σ(·) is the Logistid Sigmoid function, so that E t Linear distribution from 0 to 100;

[0029] Calculate the leakage risk R t =σ(z R / T s ); where T s The temperature calibration coefficient is determined by offline verification and is used to improve the probability calibration degree;

[0030] {E t ,R t} is written into the edge event queue together with the timestamp and model version number for subsequent linkage control and upload.

[0031] Preferably, the lightweight deep neural network model M is composed of an input layer, a batch normalization layer, a channel attention compression-excitation bottleneck block, a one-dimensional depth-separable convolution layer, a global average pooling layer, and an efficiency output fully connected layer and a risk output fully connected layer arranged in parallel after the global average pooling layer; wherein, the input layer is used to receive the six-dimensional multi-parameter feature vector and reshape it into a tensor of 1×6×1 shape; the batch normalization layer is used to perform zero-mean unit variance normalization on the tensor to suppress feature distribution drift; The channel attention compression-excitation bottleneck block adaptively recalibrates the channel weights through global average pooling and a two-level fully connected-ReLU-Sigmoid structure to highlight the monitoring parameters; the one-dimensional depth-separable convolution layer uses a channel-by-channel convolution with a convolution kernel size of 3 followed by a 1×1 point-by-point convolution to extract the nonlinear coupling characteristics between the monitoring parameters and compress the parameter quantity; the global average pooling layer is used to reduce the convolution output to a vector of length 64; the efficiency output fully connected layer receives a vector of length 64 and generates the logarithm of the extraction efficiency z E The risk output fully connected layer synchronously receives a vector of length 64 and generates a leakage risk logarithm z R .

[0032] Preferably, in the edge computing unit, an elliptic curve encryption algorithm is used to compress and encrypt the multi-parameter feature vector, the extraction efficiency score, and the leakage risk, and send them to the central cloud platform via a wireless network according to a set upload cycle, including:

[0033] After completing one inference, the edge computing unit packages the multi-parameter feature vector, the extraction efficiency score, the leakage risk, the device identification code, and the millisecond-level timestamp obtained at the same sampling time in a structured message format to form an original upload message;

[0034] Invoking a lightweight lossless compression algorithm on the original uploaded message to compress the message byte stream to one third of its original length, thereby obtaining a compressed message;

[0035] Using the pre-burned central cloud platform public key, an elliptic curve integrated encryption scheme is executed on the national standard 256-bit elliptic curve to perform symmetric key encapsulation and data encryption on the compressed message, and then an elliptic curve digital signature algorithm is used to generate a digital signature for the ciphertext;

[0036] The encrypted and signed ciphertext is placed in the message queue telemetry transmission protocol message body, a 32-bit cyclic redundancy check code is added as a link-level error detection field, and the service quality level is marked as "level 1" in the message header to ensure reliable delivery at least once;

[0037] When the local timer reaches the set upload period, the message is published to the central cloud platform through the wireless network; if no confirmation response is received from the platform within the preset waiting time, the exponential backoff retransmission mechanism is automatically entered until confirmation is received or the maximum number of retransmissions is reached to complete reliable encrypted uplink transmission.

[0038] Preferably, the wireless network is any one of 5G NR or sixth generation wireless local area network link.

[0039] A multi-parameter gas extraction monitoring system based on edge computing, including:

[0040] A multi-source monitoring and acquisition unit is used to deploy a number of sensor nodes in gas extraction boreholes, pipelines, and gas gathering stations in the target area, and use the sensor nodes to respectively collect raw monitoring parameters composed of multiple parameters; each of the sensor nodes is connected to an edge computing unit arranged around the sensor node via an industrial field bus;

[0041] A raw data preprocessing unit, configured to perform time synchronization, zero drift correction, outlier removal, and dimension normalization on the received raw monitoring parameters within the edge computing unit to form a multi-parameter feature vector;

[0042] An edge reasoning and analysis unit, configured to call a pre-deployed lightweight deep neural network model to infer the multi-parameter feature vector and output an extraction efficiency score and a leakage risk degree;

[0043] An on-site risk response control unit, configured to, when the leakage risk level is greater than a first threshold, trigger a local sound and light alarm using the edge computing unit and issue a speed reduction instruction to a nearby drainage pump inverter; and, when the drainage efficiency score is lower than a second threshold, adjust a drainage negative pressure set value using the edge computing unit;

[0044] A secure communication transmission unit is configured to compress and encrypt the multi-parameter feature vector, the extraction efficiency score, and the leakage risk level using an elliptic curve encryption algorithm within the edge computing unit, and transmit the compressed data to a central cloud platform via a wireless network according to a set upload cycle.

[0045] The model incremental learning and hot update unit is used to perform incremental training on the lightweight deep neural network model based on historical monitoring data and downhole feedback results in the central cloud platform, generate an updated model, and send the updated model to each edge computing unit through a secure channel to complete hot replacement.

[0046] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0047] The present invention addresses the problems of existing gas extraction monitoring systems, such as strong data upload dependency, long response delay, delayed abnormal warning, and untimely model updates. A multi-parameter monitoring method based on edge computing is proposed. By deploying edge computing units on-site, local synchronous processing and feature fusion of multiple parameters such as gas concentration, negative pressure, and flow are achieved, replacing the traditional single-point upload method that relies on cloud computing. Real-time reasoning of extraction efficiency and leakage risk is completed on the edge side through a lightweight neural network model, enabling early warning and control to be executed instantly underground, significantly improving the system's response speed and independent operation capabilities. Elliptic curve encryption is used to ensure the confidentiality and integrity of data transmission, enhancing the security of the system in complex underground communication environments. Combined with the incremental training and model hot replacement mechanism of the central cloud platform, the model can be continuously optimized based on historical data and underground feedback, overcoming the defects of traditional system model updates that are delayed and unable to adapt to dynamic changes in extraction, and overall improving the system's intelligence, robustness, and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0050] Figure 2 A flowchart for constructing a multi-parameter feature vector provided by an embodiment of the present invention;

[0051] Figure 3 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] The purpose of the present invention is to provide a multi-parameter monitoring method and system for gas extraction based on edge computing, which overcomes the defects of the existing technology such as reliance on the cloud, delayed response and untimely model updates, and realizes local intelligent analysis, rapid early warning and dynamic adaptive optimization of multi-parameter monitoring during gas extraction.

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Figure 1 A flow chart of the method provided in the embodiment of the present invention is shown in FIG. Figure 1 As shown, the present invention provides a multi-parameter monitoring method for gas extraction based on edge computing, including:

[0056] Step 100: Deploy a number of sensor nodes in gas extraction boreholes, pipelines, and gas gathering stations in the target area, and use the sensor nodes to collect raw monitoring parameters consisting of multiple parameters. Each sensor node is connected to an edge computing unit located around the sensor node via an industrial field bus.

[0057] Step 200: In the edge computing unit, the received original monitoring parameters are time synchronized, zero drift corrected, outlier removed, and dimensionally normalized to form a multi-parameter feature vector;

[0058] Step 300: Calling a pre-deployed lightweight deep neural network model to infer the multi-parameter feature vector and output the extraction efficiency score and leakage risk;

[0059] Step 400: When the leakage risk is greater than a first threshold, the edge computing unit is used to trigger a local sound and light alarm and issue a speed reduction instruction to the nearest drainage pump inverter; when the drainage efficiency score is lower than a second threshold, the edge computing unit is used to adjust the drainage negative pressure set value;

[0060] Step 500: In the edge computing unit, the multi-parameter feature vector, extraction efficiency score, and leakage risk are compressed and encrypted using the elliptic curve encryption algorithm, and then sent to the central cloud platform via the wireless network according to the set upload cycle;

[0061] Step 600: In the central cloud platform, incremental training is performed on the lightweight deep neural network model based on historical monitoring data and downhole feedback results to generate an updated model, and the updated model is sent to each edge computing unit through a secure channel to complete hot replacement.

[0062] Preferably, the original monitoring parameters include: gas concentration, extraction negative pressure, gas flow, temperature, humidity and hydrogen sulfide concentration.

[0063] In step 100 of this embodiment, first, representative borehole outlets, negative pressure pipeline intersections, and gas collection station main pipes in the gas extraction area are selected as the locations for the sensor nodes; based on the relevant specifications of the "Coal Mine Safety Regulations" and the "General Technical Conditions for Electrical Equipment Used in Coal Mines Underground", gas concentration sensors, negative pressure transmitters, thermal gas flow meters, integrated temperature and humidity probes, and electrochemical hydrogen sulfide detectors with intrinsically safe certification are installed respectively, and at least three of the above six types of sensors are installed at each node to ensure the spatiotemporal representativeness and safety coverage of the parameters collected. Each type of sensor is equipped with a unique device identifier and communication address, and is connected to the edge computing unit set nearby via the RS-485 industrial field bus and the Modbus RTU protocol. No more than 32 devices can be mounted on a single bus to meet the requirements of anti-interference and stable operation of the bus bandwidth. The sensor sampling period defaults to 5 seconds and can be configured to a range of 1 to 10 seconds according to the extraction disturbance frequency.

[0064] In order to achieve accurate collection and unified management of original monitoring parameters, this embodiment calls the device identification service through the edge computing unit to automatically identify and register the category, unit, range and physical interface type of each connected sensor; at the same time, the unified millisecond timestamp provided by the edge timing module is obtained through the time synchronization mechanism to ensure that each parameter has comparability and fusion basis within the same time window. The original monitoring parameters specifically include: gas concentration (unit is volume fraction percentage), extraction negative pressure (unit is Pa), gas flow (unit is standard cubic meters per hour), ambient temperature (unit is Celsius), ambient humidity (unit is percentage) and hydrogen sulfide concentration (unit is one millionth of volume ratio). The collected data is organized into a time series form and stored in the buffer area of the edge computing unit as the input basis for subsequent feature construction and model reasoning. Since traditional systems mostly only realize single-channel collection of gas concentration and negative pressure, lack a unified time base and fusion capabilities, it is difficult to support subsequent intelligent analysis and response control. This embodiment integrates multi-source parameters through the field bus and forms a structured input, which significantly improves the integrity and practicality of monitoring.

[0065] Preferably, the historical monitoring data includes a time series constructed by the multi-parameter feature vector, the extraction efficiency score and the leakage risk; the downhole feedback results include alarm response records and efficiency change trends after extraction adjustment.

[0066] Preferably, if Figure 2 As shown, in the edge computing unit, the received original monitoring parameters are subjected to time synchronization, zero drift correction, outlier removal and dimension normalization to form a multi-parameter feature vector, including:

[0067] Generating a parameter sequence with consistent timestamps for each type of the original monitoring parameters based on a unified timing mechanism;

[0068] The corresponding zero offset values of the parameter sequences of gas concentration C, extraction negative pressure P, and gas flow rate Q are obtained by looking up the table respectively. Then, the parameter values are corrected item by item in combination with the ambient temperature compensation factor to obtain the corrected parameter sequence C′, P′, Q′.

[0069] The calibrated C′, P′, Q′ and the original temperature T, humidity H, and hydrogen sulfide concentration S parameter sequences were tested and outliers were removed using the sliding window median standard deviation method to obtain the processed six parameter sequences. Each sliding window contains N time points. If the parameter value deviates from the median of the sliding window by more than a preset multiple standard deviation threshold, it is repaired by linear interpolation using adjacent valid data.

[0070] The six types of parameter sequences after the processing are uniformly normalized and mapped to the interval [-1, 1], and the multi-parameter feature vector at time point t is constructed using the six-dimensional data at time point t.

[0071] In this embodiment, after receiving the original monitoring parameters from each sensor node, the edge computing unit first calls the local timing service to perform a unified timestamp calibration on the data from different sources. Specifically, the edge computing unit uses the embedded high-precision crystal oscillator module as the main clock source, and refers to the external GPS synchronization signal or the industrial Ethernet PTP protocol to stamp various original parameters (including gas concentration, extraction negative pressure, gas flow, temperature, humidity and hydrogen sulfide concentration) with a millisecond-level unified timestamp, thereby constructing a time-aligned multi-type parameter sequence. Subsequently, for the three types of physical quantities of gas concentration, extraction negative pressure and flow, the calibration value is obtained based on the zero point offset of the sensor at the current operating temperature, and the zero drift correction is performed according to the following formula: Corrected value = measured value - offset value under temperature compensation; where the compensation factor is solidified in the local edge database when the equipment is calibrated at the factory. Through this process, this embodiment obtains the corrected gas concentration sequence, negative pressure sequence and flow sequence, which are recorded as C', P' and Q' respectively, laying an accurate foundation for subsequent abnormal processing and modeling analysis.

[0072] After completing the initial calibration, this embodiment further uses the sliding window median standard deviation method to perform outlier removal on six parameter sequences: C′, P′, Q′, as well as the original temperature T, humidity H, and hydrogen sulfide concentration S. Specifically, a sliding window containing N = 11 consecutive sampling points is constructed with each time point as the center. The median and standard deviation of all valid values in the window are calculated, and an adaptive threshold is set. If a value deviates from the median by more than three standard deviations, it is determined to be an outlier and replaced by the linear interpolation result of two adjacent valid points. In this embodiment, the above algorithm runs on a low-power edge processor and has the characteristics of low computational complexity and strong noise resistance. It is more adaptable to downhole sensor drift and interference scenarios than the static threshold method in the prior art. After completing outlier removal, the edge computing unit performs minimum-maximum normalization on each of the six processed parameter sequences, compressing each value to the range [-1, 1]. Finally, a six-dimensional, multi-parameter feature vector is constructed by sequentially arranging the six normalized values at a specific time point t. This vector serves as the input for the subsequent neural network inference process. This feature vector exhibits temporal synchronization, physical consistency, and scale uniformity, providing high-quality input for downhole edge intelligent inference and overcoming the issues of input dimensional imbalance, error accumulation, and response lag in existing methods.

[0073] Preferably, the calculation formula of the sliding window median standard deviation method is:

[0074]

[0075] Among them, x t is the monitoring value at the current moment; μ t =median(Wt ) is a sliding window W with length N and centered at t. t The median of the data; σ t =1.4826×median{|w-μ t |:w∈W t} is the robust standard deviation estimate of the absolute deviation of the same window data relative to the median after transformation by the normalization coefficient 1.4826; N is the number of sample points in the sliding window, N is an odd number, with a typical value of 11 or 21); It is an adaptive multiplication factor derived from the universal threshold of Gaussian noise and increases monotonically with the window size.

[0076] For example, {|w-μ t |:w∈W t} represents the set consisting of the absolute value of the difference between each sample value w and the median μ in the sliding window W.

[0077] In this embodiment, in order to improve the accuracy of outlier recognition of underground gas monitoring data and avoid misjudgment caused by the traditional method's reliance on fixed thresholds, this embodiment adopts a median standard deviation method based on a sliding window and designs the following adaptive outlier recognition criterion: the monitoring value at a certain moment is recorded as x t , taking this moment as the center, construct a sliding window W of length N t , where N is an odd number, typically 11 or 21. First, calculate the median μ of the data in the window t =median(W t ), as the local reference value at that time point. Next, calculate the set of absolute deviations between each data point in the window and the median, and find the median MAD t =median(|x-μ t |); To obtain a scale estimate consistent with the normal standard deviation, the empirical correction factor 1.4826 is used to scale the value to obtain the robust standard deviation estimate σ t =1.4826×MAD t Finally, determine whether the current value satisfies the following inequality:

[0078]

[0079] If the inequality holds, then x tis an outlier, and a linear interpolation value is constructed using the two nearest non-outlier points to replace it; if this is not true, the original value is retained. The multiplication factor of the above-mentioned "square root logarithm N" is derived from the centralized probability boundary estimation under the Gaussian distribution, and can adaptively adjust the sensitivity according to the window length without manually setting a specific threshold, thereby enhancing the versatility and stability of the algorithm. This embodiment effectively filters out isolated abnormal readings caused by electromagnetic interference, sudden changes in the downhole environment, or short-term drift of the sensor through this method, significantly improving the robustness of the monitoring parameter input and the credibility of the neural network inference results, which is different from the existing processing method that relies on a fixed multiple threshold method and is easily misled by noise.

[0080] Preferably, a pre-deployed lightweight deep neural network model is called to infer the multi-parameter feature vector and output the extraction efficiency score and leakage risk, including:

[0081] Before loading the lightweight deep neural network model M, the edge computing unit calculates the SHA-256 hash value of the model file and compares it with the pre-stored reference value. After verification, the inference process is carried out;

[0082] Encapsulating the normalized multi-parameter feature vector into a 1×6 floating-point tensor and writing it into the input cache of the local inference engine;

[0083] Call the inference engine to execute the shared backbone network and dual-task output head of the lightweight deep neural network model M to obtain the efficiency logarithm z E and leakage logarithm z R ;

[0084] Calculate the extraction efficiency score E t =100×σ(z E ); where σ(·) is the Logistid Sigmoid function, so that E t Linear distribution from 0 to 100;

[0085] Calculate the leakage risk R t =σ(z R / T s ); where T s The temperature calibration coefficient is determined by offline verification and is used to improve the probability calibration degree;

[0086] {E t ,R t} is written into the edge event queue together with the timestamp and model version number for subsequent linkage control and upload.

[0087] In this embodiment, in order to ensure the integrity and credibility of the inference model running on the edge device, the model file is first integrity checked before the model is called. Specifically, before starting the inference engine, the edge computing unit uses the SHA-256 digest algorithm to calculate the hash value of the model binary file stored locally, and compares it character by character with the standard reference value pre-burned in the trusted storage; only when the hash value is completely consistent can the model be loaded into the memory, otherwise the inference execution is blocked and a security alarm log is generated. Compared with the "direct loading without verification" method in the prior art, this processing method significantly enhances the anti-tampering capability of the model operation, and is particularly suitable for underground industrial application scenarios with high requirements for model security.

[0088] After the model is loaded, this embodiment encapsulates the normalized six-dimensional multi-parameter feature vectors into a single-batch, single-channel, one-dimensional floating-point tensor and writes it to the input buffer of an edge inference engine (such as TensorRT or ONNX Runtime). The inference engine is then called to execute a lightweight deep neural network model, which consists of a shared backbone network composed of a set of convolutional activation structures and two parallel output task heads: the first output head generates the logarithm of the extraction efficiency, and the second output head generates the logarithm of the leakage risk. After inference is complete, the logarithm of the efficiency is mapped using a sigmoid function and multiplied by 100, so that the extraction efficiency score is linearly distributed between 0 and 100. Simultaneously, the logarithm of the leakage is divided by the temperature calibration coefficient obtained through offline verification and then input into a sigmoid function to generate a leakage risk value between 0 and 1, thereby improving the accuracy of the output value's probability representation under different operating conditions. Finally, the scoring result, along with the corresponding timestamp and the current model version number, is encapsulated as structured event data and written to the edge event queue for subsequent linkage control module to read in real time or report to the cloud at a set interval. This process breaks through the technical bottlenecks of existing methods, such as models being too large, unable to be deployed, and outputs lacking probabilistic meaning, through lightweight model compression, calibration enhancement, and structured cache management, and achieves stable, verifiable, and reliable intelligent reasoning in resource-constrained environments.

[0089] Preferably, the lightweight deep neural network model M is composed of an input layer, a batch normalization layer, a channel attention compression-excitation bottleneck block, a one-dimensional depth-separable convolution layer, a global average pooling layer, and an efficiency output fully connected layer and a risk output fully connected layer arranged in parallel after the global average pooling layer; wherein, the input layer is used to receive the six-dimensional multi-parameter feature vector and reshape it into a tensor of 1×6×1 shape; the batch normalization layer is used to perform zero-mean unit variance normalization on the tensor to suppress feature distribution drift; The channel attention compression-excitation bottleneck block adaptively recalibrates the channel weights through global average pooling and a two-level fully connected-ReLU-Sigmoid structure to highlight the monitoring parameters; the one-dimensional depth-separable convolution layer uses a channel-by-channel convolution with a convolution kernel size of 3 followed by a 1×1 point-by-point convolution to extract the nonlinear coupling characteristics between the monitoring parameters and compress the parameter quantity; the global average pooling layer is used to reduce the convolution output to a vector of length 64; the efficiency output fully connected layer receives a vector of length 64 and generates the logarithm of the extraction efficiency z E The risk output fully connected layer synchronously receives a vector of length 64 and generates a leakage risk logarithm z R .

[0090] In this embodiment, the lightweight deep neural network model M called is specially designed for edge computing unit deployment scenarios, and has the characteristics of compact structure and efficient reasoning. From input to output, the model includes an input layer, a batch normalization layer, a channel attention compression-excitation bottleneck block, a one-dimensional depth-separable convolution layer, a global average pooling layer, and an efficiency output fully connected layer and a risk output fully connected layer arranged in parallel after the global average pooling layer. The input layer is used to receive the normalized six-dimensional multi-parameter feature vector and reconstruct it into a tensor of size 1×6×1 to adapt to the subsequent convolution calculation structure. The batch normalization layer performs zero mean and unit variance standardization on the channel dimension of the tensor, effectively suppressing the feature distribution offset that may occur during training and inference, and improving the stability of the model. Next, the model learns the weights of the six channels through a channel attention compression-excitation bottleneck block. This module first performs global average pooling on each channel to form a one-dimensional channel description vector, and then passes it through two fully connected layers in sequence. The intermediate activation function is the rectified linear unit (ReLU) and the end is the Sigmoid function. The six channel weights are output and recalibrated by multiplication with the original channels, thereby adaptively improving the responsiveness of important parameters under the current working conditions.

[0091] After adjusting the channel weights, the model feeds this tensor into a one-dimensional depthwise separable convolutional layer for feature extraction. First, a channel-by-channel convolution operation with a kernel size of 3 is performed independently on each channel to extract the temporal correlation pattern of each monitoring parameter. Subsequently, a 1×1 point-by-point convolution is used to achieve inter-channel interaction and compression, significantly reducing the number of parameters and improving nonlinear cross-modeling capabilities. The output tensor of the convolution is fed into a global average pooling layer, where it is averaged across the channel dimension and ultimately compressed into a fixed-length feature vector of 64. This vector is then fed into the efficiency output fully connected layer and the risk output fully connected layer, respectively. The former outputs the logarithm of the extraction efficiency, while the latter outputs the logarithm of the leakage risk. Both outputs are then fed into a sigmoid activation function for normalization mapping to form the subsequent score. The overall structure of the model is compact, the number of parameters is controlled within 50,000, the average inference delay is less than 50 milliseconds, and it is compatible with edge processors with ARM Cortex-A55 architecture and above, meeting the low-power, low-latency real-time operation requirements of mine sites. Different from the technical limitations of traditional deep neural networks that cannot be directly deployed on edge devices, it has the significant advantages of both operability and deployment efficiency.

[0092] Preferably, in the edge computing unit, an elliptic curve encryption algorithm is used to compress and encrypt the multi-parameter feature vector, the extraction efficiency score, and the leakage risk, and send them to the central cloud platform via a wireless network according to a set upload cycle, including:

[0093] After completing one inference, the edge computing unit packages the multi-parameter feature vector, the extraction efficiency score, the leakage risk, the device identification code, and the millisecond-level timestamp obtained at the same sampling time in a structured message format to form an original upload message;

[0094] Invoking a lightweight lossless compression algorithm on the original uploaded message to compress the message byte stream to one third of its original length, thereby obtaining a compressed message;

[0095] Using the pre-burned central cloud platform public key, an elliptic curve integrated encryption scheme is executed on the national standard 256-bit elliptic curve to perform symmetric key encapsulation and data encryption on the compressed message, and then an elliptic curve digital signature algorithm is used to generate a digital signature for the ciphertext;

[0096] The encrypted and signed ciphertext is placed in the message queue telemetry transmission protocol message body, a 32-bit cyclic redundancy check code is added as a link-level error detection field, and the service quality level is marked as "level 1" in the message header to ensure reliable delivery at least once;

[0097] When the local timer reaches the set upload period, the message is published to the central cloud platform through the wireless network; if no confirmation response is received from the platform within the preset waiting time, the exponential backoff retransmission mechanism is automatically entered until confirmation is received or the maximum number of retransmissions is reached to complete reliable encrypted uplink transmission.

[0098] Preferably, the wireless network is any one of 5G NR or sixth generation wireless local area network link.

[0099] In this embodiment, to achieve highly secure, low-redundancy, and highly reliable uplink transmission of edge-side inference results, immediately after completing a model inference, the multi-parameter feature vector, extraction efficiency score, and leakage risk generated by this inference are encapsulated along with the sampling device's unique identification code and millisecond timestamp into a structured key-value pair format. The structure definition library is then used to generate the original upload message that complies with the general protocol. This message is compressed using a lightweight lossless compression algorithm, reducing the bit length to approximately one-third of the original value before and after compression, significantly reducing the bandwidth load for subsequent encryption and transmission. To mitigate potential leaks caused by plaintext uplinks, this embodiment uses a national standard 256-bit elliptic curve (compliant with SM2 or NIST P-256 standards) to encrypt the compressed message. The specific process is as follows: First, the local embedded encryption module is invoked to execute the elliptic curve integrated encryption scheme based on the pre-burned cloud public key, achieving dual protection of key agreement and symmetric encryption. The generated ciphertext is then subjected to the elliptic curve digital signature algorithm, and a signature fragment is generated using the edge-side private key and appended to the end of the ciphertext for platform verification.

[0100] After completing the encryption and signing process, this embodiment packages the ciphertext into a message body that conforms to the message queue telemetry transmission protocol format. A 32-bit cyclic redundancy check field is added to the end of the message for link layer error detection. The message header is marked with a quality of service level of level 1 to ensure that the message is successfully delivered at least once. This embodiment uses either a 5G new air interface or a sixth-generation wireless local area network wireless link for data transmission, and the link type can be dynamically switched based on the underground communication coverage. When the local timer in the edge computing unit reaches the set upload period, the encrypted message is immediately released to the central cloud platform via the above-mentioned wireless link. If no confirmation response is received from the cloud platform within the set response waiting time, an exponential backoff retransmission mechanism is automatically initiated. The initial retry interval is 100 milliseconds, and the retry interval doubles with each retry, with a maximum number of retries of five. If confirmation is still not received after five attempts, the error is logged and the system switches to offline caching mode, waiting for retransmission in the next scheduling cycle. Compared with the existing technology that simply relies on physically isolated networks or plain text uplink methods, this embodiment ensures that the inference results can achieve a high-security, high-integrity and high-fault-tolerance data upload process even in harsh underground environments through the integrated design of compression, encryption, signing and retransmission.

[0101] In step 600 of this embodiment, in order to adapt to the data distribution drift caused by the dynamic changes in underground working conditions, fluctuations in extraction load and environmental disturbances, this embodiment constructs a periodic incremental training mechanism for the neural network model based on the central cloud platform. Specifically, first, the uploaded historical monitoring data is received from the edge computing unit. The historical data is organized in a time series structure and contains multi-parameter feature vectors, corresponding timestamps, extraction efficiency scores and leakage risk prediction values. At the same time, the central platform extracts underground feedback results from the manual duty system, sound and light alarm logs and extraction pump frequency adjustment records, mainly including the alarm response status, linkage action execution time, and the change trend of extraction efficiency within 5 minutes after control, etc., which are used to mark the correctness and timeliness of the model output. This embodiment adopts a multi-task learning framework with an attention calibration mechanism. On the basis of freezing the main network parameters, only the output layer and batch normalization layer parameters are locally fine-tuned, thereby achieving adaptive updates to the latest data features and avoiding overall model drift or catastrophic forgetting.

[0102] After the training is completed, the generated updated model is saved as a binary model file with a unified structure of the platform, and version control and summary signature processing are performed. In order to ensure the transmission security and consistency of the model, this embodiment adopts an end-to-end secure channel for the distribution operation. The secure channel is based on the Transport Layer Security Protocol version 1.3 (TLS1.3), combined with a two-way identity authentication mechanism to ensure that the model can only be received by the authorized edge computing unit. The model file is appended with a summary signature and a model version number before sending; the edge computing unit immediately verifies the signature consistency after receiving it, and without interrupting the edge process, the updated model file is hot-loaded into the inference engine cache, and the original model parameters are replaced through memory mapping without restarting the device or interrupting the task. This hot replacement method effectively guarantees the continuous operation capability of edge intelligence, which is different from the traditional model update method that requires shutdown or manual loading, and realizes the self-evolution of the model and edge autonomy in the field environment.

[0103] Corresponding to the above method, such as Figure 2 As shown, this embodiment also provides a gas extraction multi-parameter monitoring system based on edge computing, including:

[0104] A multi-source monitoring and acquisition unit is used to deploy a number of sensor nodes in gas extraction boreholes, pipelines, and gas gathering stations in the target area, and use the sensor nodes to respectively collect raw monitoring parameters composed of multiple parameters; each of the sensor nodes is connected to an edge computing unit arranged around the sensor node via an industrial field bus;

[0105] A raw data preprocessing unit, configured to perform time synchronization, zero drift correction, outlier removal, and dimension normalization on the received raw monitoring parameters within the edge computing unit to form a multi-parameter feature vector;

[0106] An edge reasoning and analysis unit, configured to call a pre-deployed lightweight deep neural network model to infer the multi-parameter feature vector and output an extraction efficiency score and a leakage risk degree;

[0107] An on-site risk response control unit, configured to, when the leakage risk level is greater than a first threshold, trigger a local sound and light alarm using the edge computing unit and issue a speed reduction instruction to a nearby drainage pump inverter; and, when the drainage efficiency score is lower than a second threshold, adjust a drainage negative pressure set value using the edge computing unit;

[0108] A secure communication transmission unit is configured to compress and encrypt the multi-parameter feature vector, the extraction efficiency score, and the leakage risk level using an elliptic curve encryption algorithm within the edge computing unit, and transmit the compressed data to a central cloud platform via a wireless network according to a set upload cycle.

[0109] The model incremental learning and hot update unit is used to perform incremental training on the lightweight deep neural network model based on historical monitoring data and downhole feedback results in the central cloud platform, generate an updated model, and send the updated model to each edge computing unit through a secure channel to complete hot replacement.

[0110] The beneficial effects of the present invention are as follows:

[0111] (1) By deploying edge computing units at the gas extraction site, this invention enables local processing and real-time reasoning of multiple monitoring parameters, avoiding the high latency and communication interruption risks associated with traditional methods that rely on cloud computing. The edge side can independently complete data cleaning, feature extraction, model reasoning, and early warning response, and can maintain continuous operation even in sudden underground environmental conditions, effectively improving the local intelligent response capabilities of the extraction system.

[0112] (2) This paper proposes a complete preprocessing process for multi-source monitoring parameters such as gas concentration, negative pressure, flow rate, temperature and humidity, and harmful gases, including time synchronization, zero drift correction, sliding window robust outlier removal, and normalization processing to form a structured multi-parameter feature vector. By introducing the median absolute deviation and adaptive threshold mechanism, the system's fault tolerance to interference, drift, and sudden changes in data in complex underground environments is enhanced.

[0113] (3) This invention uses elliptic curve encryption and a digital signature mechanism to provide double protection after data compression. Combined with a signed structured message and a retransmission confirmation mechanism, it ensures the confidentiality, integrity, and reliability of uploaded data transmitted over public networks or underground wireless networks. Compared to traditional plaintext upload methods, this significantly improves the system's data security level under complex mining communication conditions and reduces the risk of data leakage and tampering.

[0114] (4) This invention establishes a cloud-based incremental training and edge model hot replacement mechanism, which can continuously optimize the neural network model structure and weight parameters by combining historical monitoring data and downhole feedback results, and complete model version replacement through a secure channel without interrupting task operation. This mechanism overcomes the problems of model update lag and poor adaptability in existing solutions, enabling the gas extraction monitoring system to have dynamic learning and continuous evolution capabilities, making it suitable for long-term operation scenarios.

[0115] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0116] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A multi-parameter monitoring method for gas extraction based on edge computing, characterized in that: include: Deploy a number of sensor nodes in gas extraction boreholes, pipelines, and gas gathering stations in the target area, and use the sensor nodes to collect original monitoring parameters composed of multiple parameters; Each of the sensor nodes is connected to an edge computing unit arranged around the sensor node via an industrial field bus; In the edge computing unit, the received original monitoring parameters are time synchronized, zero drift corrected, outlier removed, and dimensionally normalized to form a multi-parameter feature vector; Calling a pre-deployed lightweight deep neural network model to infer the multi-parameter feature vector and output an extraction efficiency score and leakage risk; When the leakage risk is greater than a first threshold, the edge computing unit is used to trigger a local sound and light alarm and issue a speed reduction instruction to the nearest extraction pump inverter; when the extraction efficiency score is lower than a second threshold, the edge computing unit is used to adjust the extraction negative pressure set value; In the edge computing unit, an elliptic curve encryption algorithm is used to compress and encrypt the multi-parameter feature vector, the extraction efficiency score, and the leakage risk level, and the data are sent to the central cloud platform via a wireless network according to a set upload cycle; In the central cloud platform, incremental training is performed on the lightweight deep neural network model based on historical monitoring data and downhole feedback results to generate an updated model, and the updated model is sent to each edge computing unit through a secure channel to complete hot replacement.

2. The multi-parameter monitoring method for gas extraction based on edge computing according to claim 1 is characterized in that: The original monitoring parameters include: gas concentration, extraction negative pressure, gas flow, temperature, humidity and hydrogen sulfide concentration.

3. The multi-parameter monitoring method for gas extraction based on edge computing according to claim 1 is characterized in that: The historical monitoring data includes a time series constructed by the multi-parameter feature vector, the extraction efficiency score and the leakage risk; the downhole feedback results include alarm response records and efficiency change trends after extraction adjustment.

4. The multi-parameter monitoring method for gas extraction based on edge computing according to claim 2 is characterized in that: In the edge computing unit, the received original monitoring parameters are subjected to time synchronization, zero drift correction, outlier removal, and dimension normalization to form a multi-parameter feature vector, including: Generating a parameter sequence with consistent timestamps for each type of the original monitoring parameters based on a unified timing mechanism; The corresponding zero offset values of the parameter sequences of gas concentration C, extraction negative pressure P, and gas flow rate Q are obtained by looking up the table respectively. Then, the parameter values are corrected item by item in combination with the ambient temperature compensation factor to obtain the corrected parameter sequence C′, P′, Q′. The calibrated C′, P′, Q′ and the original temperature T, humidity H, and hydrogen sulfide concentration S parameter sequences were tested and outliers were removed using the sliding window median standard deviation method to obtain the processed six parameter sequences. Each sliding window contains N time points. If the parameter value deviates from the median of the sliding window by more than a preset multiple standard deviation threshold, it is repaired by linear interpolation using adjacent valid data. The six types of parameter sequences after the processing are uniformly normalized and mapped to the interval [-1, 1], and the multi-parameter feature vector at time point t is constructed using the six-dimensional data at time point t.

5. The multi-parameter monitoring method for gas extraction based on edge computing according to claim 4 is characterized in that: The calculation formula of the sliding window median standard deviation method is: Among them, x t is the monitoring value at the current moment; μ t =median(W t ) is a sliding window W with length N and centered at t. t The median of the data; σ t =1.4826×median{|w-μ t |:w∈W t } is the robust standard deviation estimate of the absolute deviation of the same window data relative to the median after transformation by the normalization coefficient 1.4826; N is the number of sample points in the sliding window, N is an odd number, with a typical value of 11 or 21); It is an adaptive multiplication factor derived from the universal threshold of Gaussian noise and increases monotonically with the window size.

6. The multi-parameter monitoring method for gas extraction based on edge computing according to claim 1 is characterized in that: The pre-deployed lightweight deep neural network model is called to infer the multi-parameter feature vector and output the extraction efficiency score and leakage risk, including: Before loading the lightweight deep neural network model M, the edge computing unit calculates the SHA-256 hash value of the model file and compares it with the pre-stored reference value. After verification, the inference process is carried out; Encapsulating the normalized multi-parameter feature vector into a 1×6 floating-point tensor and writing it into the input cache of the local inference engine; Call the inference engine to execute the shared backbone network and dual-task output head of the lightweight deep neural network model M to obtain the efficiency logarithm z E and leakage logarithm z R ; Calculate the extraction efficiency score E t =100×σ(z E ); where σ(·) is the Logistid Sigmoid function, so that E t Linear distribution from 0 to 100; Calculate the leakage risk R t =σ(z R / T s ); where T s The temperature calibration coefficient is determined by offline verification and is used to improve the probability calibration degree; [E t ,R t } is written into the edge event queue together with the timestamp and model version number for subsequent linkage control and upload.

7. The multi-parameter monitoring method for gas extraction based on edge computing according to claim 1 is characterized in that: The lightweight deep neural network model M is composed of an input layer, a batch normalization layer, a channel attention compression-excitation bottleneck block, a one-dimensional depth-separable convolution layer, a global average pooling layer, and an efficiency output fully connected layer and a risk output fully connected layer arranged in parallel after the global average pooling layer. The input layer is used to receive the six-dimensional multi-parameter feature vector and reshape it into a tensor of 1×6×1 shape; the batch normalization layer is used to perform zero-mean unit variance normalization on the tensor to suppress feature distribution drift; the channel The attention compression-excitation bottleneck block adaptively recalibrates the channel weights through global average pooling and a two-level fully connected-ReLU-Sigmoid structure to highlight the monitoring parameters; the one-dimensional depth-separable convolution layer uses a channel-by-channel convolution with a convolution kernel size of 3 followed by a 1×1 point-by-point convolution to extract the nonlinear coupling characteristics between the monitoring parameters and compress the parameter quantity; the global average pooling layer is used to reduce the convolution output to a vector of length 64; the efficiency output fully connected layer receives a vector of length 64 and generates the logarithm of the extraction efficiency z E The risk output fully connected layer synchronously receives a vector of length 64 and generates a leakage risk logarithm z R .

8. The multi-parameter monitoring method for gas extraction based on edge computing according to claim 1 is characterized in that: In the edge computing unit, the multi-parameter feature vector, the extraction efficiency score, and the leakage risk are compressed and encrypted using an elliptic curve encryption algorithm, and are sent to the central cloud platform via a wireless network according to a set upload cycle, including: After completing one inference, the edge computing unit packages the multi-parameter feature vector, the extraction efficiency score, the leakage risk, the device identification code, and the millisecond-level timestamp obtained at the same sampling time in a structured message format to form an original upload message; Invoking a lightweight lossless compression algorithm on the original uploaded message to compress the message byte stream to one third of its original length, thereby obtaining a compressed message; Using the pre-burned central cloud platform public key, an elliptic curve integrated encryption scheme is executed on the national standard 256-bit elliptic curve to perform symmetric key encapsulation and data encryption on the compressed message, and then an elliptic curve digital signature algorithm is used to generate a digital signature for the ciphertext; The encrypted and signed ciphertext is placed in the message queue telemetry transmission protocol message body, a 32-bit cyclic redundancy check code is added as a link-level error detection field, and the service quality level is marked as "level 1" in the message header to ensure at least once reliable delivery; When the local timer reaches the set upload period, the message is published to the central cloud platform through the wireless network; if no confirmation response is received from the platform within the preset waiting time, the exponential backoff retransmission mechanism is automatically entered until confirmation is received or the maximum number of retransmissions is reached to complete reliable encrypted uplink transmission.

9. The multi-parameter monitoring method for gas extraction based on edge computing according to claim 1 is characterized in that: The wireless network is either 5G NR or a sixth-generation wireless local area network link.

10. A multi-parameter monitoring system for gas extraction based on edge computing, characterized in that: include: A multi-source monitoring and acquisition unit is used to deploy a number of sensor nodes in gas extraction boreholes, pipelines, and gas gathering stations in the target area, and use the sensor nodes to respectively collect raw monitoring parameters composed of multiple parameters; each of the sensor nodes is connected to an edge computing unit arranged around the sensor node via an industrial field bus; A raw data preprocessing unit, configured to perform time synchronization, zero drift correction, outlier removal, and dimension normalization on the received raw monitoring parameters within the edge computing unit to form a multi-parameter feature vector; An edge reasoning and analysis unit, configured to call a pre-deployed lightweight deep neural network model to infer the multi-parameter feature vector and output an extraction efficiency score and a leakage risk degree; An on-site risk response control unit, configured to, when the leakage risk level is greater than a first threshold, trigger a local sound and light alarm using the edge computing unit and issue a speed reduction instruction to a nearby drainage pump inverter; and, when the drainage efficiency score is lower than a second threshold, adjust a drainage negative pressure set value using the edge computing unit; A secure communication transmission unit is configured to compress and encrypt the multi-parameter feature vector, the extraction efficiency score, and the leakage risk level using an elliptic curve encryption algorithm within the edge computing unit, and transmit the compressed data to a central cloud platform via a wireless network according to a set upload cycle. The model incremental learning and hot update unit is used to perform incremental training on the lightweight deep neural network model based on historical monitoring data and downhole feedback results in the central cloud platform, generate an updated model, and send the updated model to each edge computing unit through a secure channel to complete hot replacement.

Images