A method and system for identifying dangerous areas based on underground liquid level monitoring

Abstract

This invention discloses a method and system for identifying hazardous areas based on downhole liquid level monitoring, relating to the field of downhole liquid level monitoring technology. The method includes: deploying an adaptive sensor network at key downhole locations; dynamically adjusting the sensor distribution density according to environmental changes; employing an intelligent data acquisition strategy; dynamically adjusting the acquisition frequency according to environmental changes; collecting sensor data using sensors; transmitting the sensor data to a central processing unit via low-power wireless communication technology; the central processing unit using deep learning algorithms to extract and fuse features from different types of sensor data to generate a comprehensive feature set; collecting current environmental conditions and historical data; introducing a context-aware mechanism; dynamically adjusting the threshold for hazardous area identification based on current environmental conditions and historical data; and generating intelligent alarm signals.

Description

Technical Field

[0001] The present invention relates to the technical field of underground liquid level monitoring, and in particular to a method and system for identifying a dangerous area based on underground liquid level monitoring. Background Art

[0002] In the fields of mining, oil extraction and underground engineering, underground liquid level monitoring has always been a key link to ensure operational safety and production efficiency. With the rapid development of Internet of Things technology, sensor technology and wireless communication technology, underground liquid level monitoring systems have gradually developed from traditional manual detection and wired transmission to automation and intelligence. Early underground liquid level monitoring systems mainly relied on a single type of sensor, such as liquid level sensors, to transmit data to the ground monitoring center through wired cables. However, this system has problems such as complex wiring, high maintenance costs and unstable data transmission. In recent years, the introduction of wireless sensor network (WSN) technology has improved these problems, enabling nodes to transmit data wirelessly, reducing wiring costs and maintenance difficulties. In addition, with the maturity of low-power wide area network (LPWAN) technology, such as LoRa and NB-IoT, wireless sensor networks are more widely used in underground environments, realizing long-distance, low-power data transmission, and further improving the reliability and stability of the system.

[0003] Although the existing downhole liquid level monitoring system has made significant progress in data acquisition and transmission, there are still many shortcomings in data processing and fault prediction. First, traditional data processing methods often rely on simple statistical analysis and threshold judgment, which is difficult to effectively cope with complex downhole environmental changes. Secondly, the existing hazardous area prediction models are mostly based on static data and lack the ability to adapt to dynamic environmental changes, resulting in low prediction accuracy. In addition, most systems fail to fully consider the context perception mechanism and cannot dynamically adjust the threshold for hazardous area identification according to current environmental conditions and historical data, resulting in false alarms and missed alarms. These shortcomings not only affect the accuracy and reliability of the system, but also increase maintenance costs and safety hazards. Therefore, the development of a downhole liquid level monitoring system that can dynamically adjust the sensor distribution density, intelligently collect data, fuse multiple sensor data and perform fault prediction has become an important direction of current research. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method and system for identifying dangerous areas based on downhole liquid level monitoring to solve the problem that most existing dangerous area prediction models are based on static data and lack the ability to adapt to dynamic environmental changes, resulting in low prediction accuracy.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for identifying a dangerous area based on underground liquid level monitoring, which comprises:

[0008] Deploy adaptive sensor networks at key locations underground, dynamically adjust sensor distribution density according to environmental changes, adopt intelligent data collection strategies, dynamically adjust collection frequency according to environmental changes, use sensors to collect sensor data, and transmit sensor data to the central processing unit through low-power wireless communication technology;

[0009] The central processing unit uses deep learning algorithms to extract and fuse features from different types of sensor data to generate a comprehensive feature set;

[0010] Collect current environmental conditions and historical data, introduce context-aware mechanisms, dynamically adjust the thresholds for identifying hazardous areas based on current environmental conditions and historical data, and generate intelligent alarm signals;

[0011] A dangerous area prediction model is constructed using a deep learning algorithm. The dangerous area data of the downhole liquid level and environmental condition data are extracted from historical data and input into the dangerous area prediction model for training. The comprehensive feature set is input into the trained dangerous area prediction model, and maintenance recommendations are generated based on the output results of the dangerous area prediction model and the intelligent alarm signals.

[0012] As a preferred solution of the method for identifying dangerous areas based on underground liquid level monitoring of the present invention, wherein: the deployment of an adaptive sensor network at a key position underground means:

[0013] Deploy an initial sensor network at wellheads, major pipeline interfaces, and pump stations at downhole level monitoring points;

[0014] Drones are used for auxiliary deployment so that the nodes corresponding to the sensor network are evenly distributed near the liquid level monitoring points. Each node is equipped with a liquid level sensor, a temperature sensor, and a pressure sensor to collect the height of the downhole hydraulic pressure, the temperature of the downhole environment, and the pressure of the downhole environment, respectively, and a unique node ID is marked for each node.

[0015] As a preferred solution of the method for identifying dangerous areas based on underground liquid level monitoring of the present invention, wherein: the distribution density of sensors is dynamically adjusted according to environmental changes, an intelligent data acquisition strategy is adopted, the acquisition frequency is dynamically adjusted according to environmental changes, sensor data is collected by sensors, and the sensor data is transmitted to the central processing unit through low-power wireless communication technology. The specific steps are as follows:

[0016] Each sensor acts as a node to regularly collect the downhole hydraulic pressure height, downhole environment temperature and downhole environment pressure, and pre-process the collected sensor data;

[0017] The pre-processed sensor data is transmitted to the central processing unit via the low-power wireless communication technology LoRa;

[0018] The central processing unit uses the K-means clustering algorithm to perform cluster analysis on the collected sensor data. Specifically, the K-means clustering algorithm divides the data points into several clusters, each cluster represents an area, and identifies the areas where the sensor density needs to be increased and decreased by analyzing the data density and distribution of each cluster. For the areas where sensors need to be increased, corresponding sensors are added, and for the areas where sensors need to be reduced, corresponding sensors are reduced.

[0019] Set the initial acquisition frequency of the sensor to , set according to the maximum acquisition capacity of the equipment and user needs. The current downhole liquid level is The collection time interval is ;

[0020] According to the current downhole liquid level Calculate the change in liquid level , the expression is:

[0021] ;

[0022] in, Indicates the underground liquid level at the previous moment;

[0023] According to the collection time interval Calculate the change in time , the expression is:

[0024] ;

[0025] Each node monitors the temperature, humidity and pressure of the underground environment in real time. The central processing unit dynamically adjusts the acquisition frequency of each node according to the changes in the underground environmental conditions. , the expression is:

[0026] ;

[0027] in, represents the initial acquisition frequency, represents the frequency adjustment coefficient, Indicates the change in liquid level. represents the change in time, Indicates the absolute value of the rate of change of liquid level;

[0028] Select LoRa as the communication protocol;

[0029] The pre-processed sensor data is stored in a database of the central processing unit.

[0030] As a preferred solution of the method for identifying dangerous areas based on underground liquid level monitoring of the present invention, the central processing unit uses a deep learning algorithm to extract and fuse features of different types of sensor data to generate a comprehensive feature set. The specific steps are as follows:

[0031] Select convolutional neural network CNN as the deep learning model;

[0032] The pre-processed downhole hydraulic pressure height, downhole environment temperature and downhole environment pressure are input into the deep learning model, and the convolutional neural network CNN is used to extract the features of the sensor data. The expression is:

[0033] ;

[0034] in, Indicates The feature vector of sensor data, represents the convolution operation, which is used to extract local features of sensor data. After preprocessing, Class sensor data, represents the weight matrix in the convolutional layer, which is used to learn the local features of sensor data. represents the bias term;

[0035] The extracted features are fused to generate a comprehensive feature set, which is expressed as:

[0036] ;

[0037] in, represents the comprehensive feature set, represents the activation function ReLU, Indicates the number of sensor data types, Indicates The weight of the sensor data.

[0038] As a preferred solution of the method for identifying dangerous areas based on underground liquid level monitoring of the present invention, the method collects current environmental conditions and historical data, introduces a context-aware mechanism, dynamically adjusts the threshold of dangerous area identification according to the current environmental conditions and historical data, and generates an intelligent alarm signal. The specific steps are:

[0039] The central processing unit monitors the temperature, humidity and pressure of the underground environment in real time to obtain the current environmental conditions; the nearest neighbor algorithm KNN is used to extract historical data fragments similar to the current environmental conditions;

[0040] According to the current environmental conditions and historical data fragments, the threshold for identifying dangerous areas is dynamically adjusted. The threshold adjustment formula is:

[0041] ;

[0042] in, represents the adjusted threshold, represents the initial threshold, represents the adjustment factor, Indicates the number of historical data fragments, Represents an index variable, Indicates The weight of each historical data segment, Indicates The environmental conditions of each historical data segment, Indicates current environmental conditions;

[0043] Compare the currently collected sensor data with the adjusted threshold Make a comparison;

[0044] When the current sensor data exceeds the threshold When a dangerous area is detected, an intelligent alarm signal is generated to indicate the presence of a dangerous area;

[0045] The generated intelligent alarm signals are stored in a database of the central processing unit.

[0046] As a preferred solution of the method for identifying dangerous areas based on underground liquid level monitoring of the present invention, the method uses a deep learning algorithm to build a dangerous area prediction model, extracts dangerous area data and environmental condition data of underground liquid levels from historical data and inputs them into the dangerous area prediction model for training, and inputs the comprehensive feature set into the trained dangerous area prediction model. The specific steps are as follows:

[0047] Select the long short-term memory network LSTM as the dangerous area prediction model;

[0048] The dangerous area prediction model structure includes an input layer, multiple LSTM layers, a fully connected layer, and an output layer;

[0049] Extract the dangerous area data of downhole liquid level and environmental condition data from historical data and input them into the dangerous area prediction model for training;

[0050] Based on the comprehensive feature set, a convolutional neural network model CNN is constructed using automatic feature extraction. The comprehensive feature set is input into the convolutional neural network model CNN to extract high-level features, which are expressed as:

[0051] ;

[0052] in, Represents high-level features;

[0053] The high-level features are input into the dangerous area prediction model to extract time series features. The expression is:

[0054] ;

[0055] in, Represents time series characteristics;

[0056] The time series features The input is sent to the fully connected layer to predict the probability of the dangerous area, the type of dangerous area and the appearance time of the dangerous area. The expression is: ;

[0057] in, Represents the probability of the dangerous area, with a value range of [0, 1], Indicates the type of hazardous area, Indicates the appearance time of the danger zone, Represents the Sigmoid activation function, which is used to map the output to the [0, 1] interval. represents the Softmax activation function, which is used to convert the output into a probability distribution. , and The weight matrices representing the probability of dangerous areas, the types of dangerous areas, and the appearance time of dangerous areas, respectively, , and They represent the probability of dangerous areas, the type of dangerous areas and the bias term of the appearance time of dangerous areas, respectively. Indicates the dimension of the output;

[0058] As a preferred solution of the method for identifying dangerous areas based on downhole liquid level monitoring of the present invention, the specific steps of generating maintenance suggestions based on the output results of the dangerous area prediction model and the intelligent alarm signal are as follows:

[0059] According to the type of hazardous area The probability distribution of the dangerous area is selected, and the type of dangerous area with the highest probability is selected as the type of dangerous area predicted;

[0060] According to the time when the danger zone appears , determine the specific time when the equipment is expected to fail;

[0061] Setting Thresholds ;

[0062] when ≥ When the device is at a high risk of failure, the type of hazardous area and the time of appearance of the danger zone Notify maintenance personnel to immediately inspect and repair the equipment;

[0063] when < When it is detected, it means that the risk in the hazardous area is within the allowable range. Continue to monitor the underground environment and regularly check the equipment and environmental conditions.

[0064] In a second aspect, the present invention provides a dangerous area identification system based on underground liquid level monitoring, comprising:

[0065] Acquisition module, central processing unit, hazard identification module, hazard area prediction module and maintenance suggestion module,

[0066] The acquisition module is used to deploy an adaptive sensor network at key locations underground, regularly collect the height of the underground hydraulic pressure, the temperature of the underground environment and the pressure of the underground environment, and pre-process the sensor data to reduce the calculation burden of the central processing unit, and transmit the sensor data to the central processing unit through low-power wireless communication technology;

[0067] The central processing unit is used to perform preprocessing, cluster analysis and feature extraction on the collected sensor data, generate a comprehensive feature set, and store it in a database;

[0068] The hazard identification module is used to introduce a context-aware mechanism to dynamically adjust the threshold for identifying hazardous areas based on current environmental conditions and historical data, and generate intelligent alarm signals;

[0069] The dangerous area prediction module is used to build a dangerous area prediction model based on a comprehensive feature set using a deep learning algorithm LSTM, and completes training by extracting dangerous area data of downhole liquid level and environmental condition data from historical data and inputting them into the dangerous area prediction model;

[0070] The maintenance suggestion module is used to generate maintenance suggestions based on the output results of the dangerous area prediction model and the intelligent alarm signal, and notify maintenance personnel to take corresponding measures.

[0071] The beneficial effects of the present invention are as follows: the present invention significantly improves the accuracy and reliability of the system and reduces maintenance costs and operational risks by dynamically adjusting the sensor distribution density and acquisition frequency, multimodal data fusion, context perception mechanism and hazardous area prediction model. The present invention belongs to the technical field of underground liquid level monitoring and hazardous area identification, and provides strong support for the safe and efficient operation of mines, oil extraction and underground engineering through intelligent data processing and fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0073] Figure 1 This is a flow chart of the dangerous area identification method based on downhole liquid level monitoring in Example 1.

[0074] Figure 2 This is a flowchart of decompressing a compressed data packet in Example 1. DETAILED DESCRIPTION

[0075] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0076] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0077] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0078] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for identifying dangerous areas based on underground liquid level monitoring, comprising the following steps:

[0079] S1 deploys an adaptive sensor network at key locations underground, dynamically adjusts the sensor distribution density according to environmental changes, adopts an intelligent data collection strategy, dynamically adjusts the collection frequency according to environmental changes, collects sensor data using sensors, and transmits sensor data to the central processing unit through low-power wireless communication technology;

[0080] Deploy an initial sensor network at wellheads, major pipeline interfaces, and pump stations at downhole level monitoring points;

[0081] Use drones for auxiliary deployment, so that the nodes corresponding to the sensor network are evenly distributed near the liquid level monitoring point. Each node is equipped with a liquid level sensor, a temperature sensor and a pressure sensor to collect the height of the downhole hydraulic pressure, the temperature of the downhole environment and the pressure of the downhole environment, and mark each node with a unique node ID;

[0082] Ensure that all key monitoring points are covered to provide a basis for subsequent dynamic adjustments;

[0083] Each node regularly collects the downhole hydraulic height, downhole environment temperature and downhole environment pressure, and pre-processes the collected sensor data. A lightweight signal pre-processing algorithm is deployed on each node to preliminarily filter and process the collected sensor data to reduce the computational burden of the central processing unit. The pre-processing algorithm is constructed based on adaptive filtering and wavelet transform, and the expression is:

[0084] ;

[0085] in, represents the preprocessed sensor data, represents the raw sensor data, i.e. the initial data collected by each node, Represents the adaptive filter coefficient, which is used to adjust the steepness of the Sigmoid function. represents the base of natural logarithms, represents the mean of the raw sensor data, represents the number of decomposition levels of wavelet transform, Represents an index variable, Indicates The weights of the layer wavelet transform, represents the wavelet basis function, represents the time variable, Indicates Time migration of layer wavelet transform, Indicates The scale parameter of the layer wavelet transform is used to adjust the width of the wavelet basis function and affect the captured frequency range;

[0086] The range of is [0, 1], The closer the value is to 0, the less noise there is in the data. The closer the value is to 1, the stronger the signal component in the data;

[0087] Adaptive filter coefficients , the calculation process is:

[0088] Set the initial adaptive filter coefficients ;

[0089] Using raw sensor data Subtract the preprocessed sensor data Get the error ;

[0090] According to the error Use proportional control method to adjust , the expression is:

[0091] ;

[0092] in, represents the updated filter coefficients, represents the original filter coefficients, represents the learning rate;

[0093] It should be noted that this formula combines adaptive filtering and wavelet transform, which can perform preliminary denoising and feature extraction on the data at the node, reducing the computational burden of the central processing unit while retaining important features in the data;

[0094] Each sensor acts as a node to regularly collect the downhole hydraulic pressure height, downhole environment temperature and downhole environment pressure, and pre-process the collected sensor data;

[0095] The pre-processed sensor data is transmitted to the central processing unit via the low-power wireless communication technology LoRa;

[0096] The central processing unit uses the K-means clustering algorithm to perform cluster analysis on the collected sensor data. Specifically, the K-means clustering algorithm divides the data points into several clusters, each cluster represents an area, and by analyzing the data density and distribution of each cluster, determines which areas need to increase the sensor density and which areas need to reduce the sensor density, identifies the areas where the sensor density needs to be increased and reduced, adds corresponding sensors to the areas where sensors need to be increased, and reduces corresponding sensors to the areas where sensors need to be reduced;

[0097] The central processing unit generates adjustment instructions based on the results of the cluster analysis;

[0098] Specifically, for areas where sensor density needs to be increased, an instruction to add sensors is generated, and the ID of the new node is specified to move to the target location;

[0099] For areas where the sensor density needs to be reduced, an instruction to reduce the sensor is generated, the ID of the node to be removed is specified, and the node is removed;

[0100] A wireless communication module is integrated inside the central processing unit. The communication module of the central processing unit is responsible for sending and receiving data. The central processing unit sends the generated adjustment instructions to the corresponding nodes through the wireless communication module.

[0101] After receiving the command, the node adjusts its position through the built-in micro motor to achieve the purpose of optimizing the distribution density;

[0102] Ensure higher monitoring density in key areas to improve monitoring accuracy and reliability;

[0103] Set the initial acquisition frequency of the sensor to , which can be set according to the maximum acquisition capacity of the equipment and user needs. The current downhole liquid level is The collection time interval is ;

[0104] According to the current downhole liquid level Calculate the change in liquid level , the expression is:

[0105] ;

[0106] in, Indicates the underground liquid level at the previous moment;

[0107] According to the collection time interval Calculate the change in time , the expression is:

[0108] ;

[0109] Each node monitors the temperature, humidity and pressure of the underground environment in real time. The central processing unit dynamically adjusts the acquisition frequency of each node according to the changes in the underground environmental conditions. , the expression is:

[0110] ;

[0111] in, Indicates the default acquisition frequency, that is, the initial acquisition frequency. Represents the frequency adjustment coefficient, which can be obtained from historical data. Indicates the change in liquid level. represents the change in time, Indicates the absolute value of the rate of change of liquid level;

[0112] LoRa was chosen as the communication protocol to ensure low power consumption and long-distance characteristics of data transmission;

[0113] Lightweight compression is performed on the pre-processed sensor data at the node to form a compressed data packet, thus reducing the amount of transmitted data;

[0114] Specifically, the pre-processed sensor data is compressed using the LZ77 compression algorithm, and the compressed data is packaged to form a compressed data packet. Compression can significantly reduce the amount of data transmission, thereby reducing the energy consumption of wireless communication and improving transmission efficiency. In the underground environment, low power consumption and efficient communication are very important because the battery life is limited and the communication distance is long. In the central processing unit, storing a large amount of sensor data may occupy a large amount of storage resources. Compression can effectively reduce storage requirements and improve the overall performance of the system;

[0115] Each node sends the compressed data packet to the central processing unit via the wireless communication module;

[0116] The communication module of the central processing unit receives the compressed data packets transmitted via LoRa and temporarily stores them in the buffer, reads the compressed data packets from the buffer, decompresses the data using the same compression algorithm LZ77 as the node, and restores the decompressed data to the original sensor data format, including the liquid level, temperature and pressure of the downhole environment;

[0117] Decompression is to restore the compressed data to the original sensor data format for further processing and analysis. The decompressed data needs to be consistent with the original data to ensure the accuracy and reliability of subsequent processing;

[0118] Preprocess the decompressed data, including filtering, denoising and standardization. Specifically, use a digital low-pass filter to remove high-frequency noise, apply wavelet transform to further reduce the noise in the data, and standardize the data to between 0 and 1 for subsequent processing;

[0119] The pre-processed sensor data is stored in a database of the central processing unit. Specifically, the pre-processed sensor data is written into the database to ensure the integrity and security of the data and to create an index for the data to facilitate subsequent query and analysis;

[0120] Ensure low power consumption and high efficiency of data transmission, reduce energy consumption and extend system operation time.

[0121] The S2 central processing unit uses deep learning algorithms to extract and fuse features from different types of sensor data to generate a comprehensive feature set;

[0122] Convolutional neural network (CNN) is selected as the deep learning model because CNN performs well in multimodal data fusion;

[0123] The pre-processed downhole hydraulic pressure height, downhole environment temperature and downhole environment pressure are input into the deep learning model, and the convolutional neural network CNN is used to extract the features of the sensor data. The expression is:

[0124] ;

[0125] in, Indicates The feature vector of sensor data, represents the convolution operation, which is used to extract local features of sensor data. After preprocessing, Class sensor data, represents the weight matrix in the convolutional layer, which is used to learn the local features of sensor data. represents the bias term;

[0126] The extracted features are fused to generate a comprehensive feature set, which is expressed as:

[0127] ;

[0128] in, represents the comprehensive feature set, represents the activation function ReLU, Indicates the number of sensor data types, Represents an index variable, Indicates The weight of the sensor data;

[0129] The robustness of the system is improved by generating a comprehensive feature set. Even if a sensor data is abnormal or lost, the features of other sensor data can still provide effective information to ensure the stability and reliability of the system.

[0130] The fused feature vector Stored in the database of the central processing unit to ensure the integrity and security of the data; it is a feature vector Create indexes to facilitate subsequent queries and analysis; regularly update historical data sets, introduce context-aware mechanisms, dynamically adjust the thresholds for hazardous area identification based on current environmental conditions and historical data, and generate intelligent alarm signals.

[0131] S3 collects current environmental conditions and historical data, introduces a context-aware mechanism, dynamically adjusts the threshold for identifying dangerous areas based on current environmental conditions and historical data, and generates intelligent alarm signals;

[0132] The central processing unit monitors the temperature, humidity and pressure of the underground environment in real time to obtain the current environmental conditions; the nearest neighbor algorithm KNN is used. The KNN algorithm is an instance-based learning method suitable for classification and regression tasks, especially for finding the historical data fragments most similar to the current data point and extracting the historical data fragments similar to the current environmental conditions;

[0133] According to the current environmental conditions and historical data fragments, the current environmental conditions include the temperature, humidity and pressure of the current underground environment, and the threshold of dangerous area identification is dynamically adjusted. The threshold adjustment formula is:

[0134] ;

[0135] in, represents the adjusted threshold, represents the initial threshold, represents the adjustment factor, Indicates the number of historical data fragments, Represents an index variable, Indicates The weight of each historical data segment, Indicates The environmental conditions of each historical data segment, Indicates current environmental conditions;

[0136] The value range of is [0, 1]. The closer the value is to 0, the lower the threshold for identifying dangerous areas is under the current environmental conditions, and the easier it is to trigger an alarm. The closer the value is to 1, the higher the threshold for identifying dangerous areas is under the current environmental conditions, and it is more difficult to trigger an alarm.

[0137] Compare the currently collected sensor data with the adjusted threshold Make a comparison;

[0138] When the current sensor data exceeds the threshold When a dangerous area is detected, an intelligent alarm signal is generated to indicate the presence of a dangerous area;

[0139] The generated intelligent alarm signals are stored in the database of the central processing unit for subsequent analysis and processing.

[0140] S4 uses deep learning algorithms to build a dangerous area prediction model, extracts dangerous area data of downhole liquid level and environmental condition data from historical data and inputs them into the dangerous area prediction model for training, inputs the comprehensive feature set into the trained dangerous area prediction model, and generates maintenance suggestions based on the output results of the dangerous area prediction model and intelligent alarm signals;

[0141] The long short-term memory network LSTM is selected as the dangerous area prediction model because LSTM performs well in processing time series data;

[0142] The dangerous area prediction model structure includes an input layer, multiple LSTM layers, a fully connected layer, and an output layer;

[0143] Extract the dangerous area data of downhole liquid level and environmental condition data from historical data and input them into the dangerous area prediction model for training;

[0144] Based on the comprehensive feature set, a convolutional neural network model CNN is constructed using automatic feature extraction. The comprehensive feature set is input into the convolutional neural network model CNN to extract high-level features, which are expressed as:

[0145] ;

[0146] in, Represents high-level features;

[0147] The high-level features are input into the dangerous area prediction model to extract time series features. The expression is:

[0148] ;

[0149] in, Represents time series characteristics;

[0150] The time series features The input is sent to the fully connected layer to predict the probability of the dangerous area, the type of dangerous area and the appearance time of the dangerous area. The expression is:

[0151] ;

[0152] in, Represents the probability of the dangerous area, with a value range of [0, 1], Indicates the type of hazardous area, Indicates the appearance time of the danger zone, Represents the Sigmoid activation function, which is used to map the output to the [0, 1] interval. represents the Softmax activation function, which is used to convert the output into a probability distribution. , and The weight matrices representing the probability of dangerous areas, the types of dangerous areas, and the appearance time of dangerous areas, respectively, , and They represent the probability of dangerous areas, the type of dangerous areas and the bias term of the appearance time of dangerous areas, respectively. Indicates the dimension of the output;

[0153] By mapping the output to the [0, 1] interval through the Sigmoid activation function, the probability of equipment failure can be accurately predicted, which provides an important reference for decision makers and helps them take preventive measures in time to reduce the risk of failure.

[0154] According to the type of hazardous area The probability distribution of the dangerous area is selected, and the type of dangerous area with the highest probability is selected as the type of dangerous area predicted;

[0155] According to the time when the danger zone appears , determine the specific time when the equipment is expected to fail;

[0156] Setting Thresholds ;

[0157] By setting a threshold for the probability of a dangerous area , which can provide clear maintenance recommendations to decision makers. When the probability of a dangerous area exceeds the threshold, the system will automatically notify maintenance personnel to immediately inspect and repair the equipment to avoid production interruptions and economic losses caused by failures;

[0158] when ≥ When the device is at a high risk of failure, the type of hazardous area and the time of appearance of the danger zone Notify maintenance personnel to immediately inspect and repair the equipment;

[0159] when < , it means that the failure risk of the equipment is within the allowable range. Continue to monitor the equipment operating status and check the equipment regularly.

[0160] This embodiment also provides a dangerous area identification system based on underground liquid level monitoring, including: an acquisition module, a central processing unit, a danger identification module, a dangerous area prediction module and a maintenance suggestion module.

[0161] The acquisition module is used to deploy an adaptive sensor network at key locations underground, regularly collect the height of the underground hydraulic pressure, the temperature of the underground environment and the pressure of the underground environment, and pre-process the sensor data to reduce the calculation burden of the central processing unit, and transmit the sensor data to the central processing unit through low-power wireless communication technology;

[0162] The central processing unit is used to perform preprocessing, cluster analysis and feature extraction on the collected sensor data, generate a comprehensive feature set, and store it in a database;

[0163] The hazard identification module is used to introduce a context-aware mechanism to dynamically adjust the threshold for identifying hazardous areas based on current environmental conditions and historical data, and generate intelligent alarm signals;

[0164] The dangerous area prediction module is used to build a dangerous area prediction model based on a comprehensive feature set using a deep learning algorithm LSTM, and completes training by extracting dangerous area data of downhole liquid level and environmental condition data from historical data and inputting them into the dangerous area prediction model;

[0165] The maintenance suggestion module is used to generate maintenance suggestions based on the output results of the dangerous area prediction model and the intelligent alarm signal, and notify maintenance personnel to take corresponding measures.

[0166] In summary, the present invention significantly improves the accuracy and reliability of the system and reduces maintenance costs and operational risks by: dynamically adjusting the sensor distribution density and acquisition frequency, multimodal data fusion, context perception mechanism and hazardous area prediction model. The present invention belongs to the field of underground liquid level monitoring and hazardous area identification technology, and provides strong support for the safe and efficient operation of mines, oil extraction and underground engineering through intelligent data processing and fault prediction.

[0167] Example 2, referring to Table 1, is the second example of the present invention. To further verify the technical solution of the present invention, experimental simulation data of a dangerous area identification method based on downhole liquid level monitoring is provided.

[0168] In order to verify the effectiveness and superiority of the dangerous area identification method based on underground liquid level monitoring, a detailed experiment was designed. The experimental site was selected in the underground environment of a large mine, which has complex geological conditions and variable environmental parameters. The main goal of the experiment was to evaluate the performance of the method in practical applications, especially in fault prediction and dangerous area identification.

[0169] 1. Deployment of sensor networks

[0170] First, the initial sensor network is deployed at the wellhead, main pipeline interface and pump station of the underground liquid level monitoring point; drones are used for auxiliary deployment to ensure that the nodes are evenly distributed near the liquid level monitoring point; each node is equipped with a liquid level sensor, a temperature sensor and a pressure sensor, which are used to collect the height of the underground hydraulic pressure, the temperature of the underground environment and the pressure of the underground environment respectively; each node is marked with a unique node ID;

[0171] 2. Data Collection and Preprocessing

[0172] Each node regularly collects the height of the downhole hydraulic pressure, the temperature of the downhole environment, and the pressure of the downhole environment, and pre-processes the collected data; the pre-processing algorithm is based on adaptive filtering and wavelet transform, and the data is preliminarily denoised and feature extracted through dynamic adjustment of the adaptive filter coefficient and multi-layer decomposition of the wavelet transform; the pre-processed sensor data is transmitted to the central processing unit through the low-power wireless communication technology LoRa; the central processing unit uses the K-means clustering algorithm to perform cluster analysis on the collected sensor data to determine which areas need to increase the sensor density and which areas need to reduce the sensor density; the central processing unit generates adjustment instructions based on the results of the cluster analysis and sends them to the corresponding nodes through the wireless communication module; after receiving the instructions, the node adjusts its position through the built-in micro motor to achieve the purpose of optimizing the distribution density;

[0173] 3. Generation of comprehensive feature sets

[0174] The central processing unit uses a deep learning algorithm to fuse different types of sensor data and generate a comprehensive feature set. The convolutional neural network (CNN) is selected as the deep learning model because CNN performs well in multimodal data fusion. The pre-processed downhole hydraulic pressure height, downhole environment temperature and downhole environment pressure are input into the deep learning model, and CNN is used to extract features from different types of sensor data. The extracted features are fused to generate a comprehensive feature set. The generation of the comprehensive feature set improves the robustness of the system. Even if a sensor data is abnormal or lost, the features of other sensor data can still provide effective information to ensure the stability and reliability of the system.

[0175] 4. Hazardous area identification and failure prediction

[0176] The central processing unit introduces a context-aware mechanism to dynamically adjust the threshold for identifying dangerous areas according to current environmental conditions and historical data, and generate intelligent alarm signals. The central processing unit monitors the temperature, humidity, and pressure of the underground environment in real time to obtain current environmental conditions. It uses the nearest neighbor algorithm KNN to dynamically adjust the threshold for identifying dangerous areas according to current environmental conditions and historical data. It compares the currently collected sensor data with the adjusted threshold. When the current sensor data exceeds the threshold, an intelligent alarm signal is generated to indicate the existence of a dangerous area.

[0177] Based on the comprehensive feature set, a dangerous area prediction model is constructed using a deep learning algorithm. The long short-term memory network LSTM is selected as the dangerous area prediction model because LSTM performs well in processing time series data. The structure of the dangerous area prediction model includes an input layer, multiple LSTM layers, a fully connected layer, and an output layer. The operating status data and fault records of the equipment are extracted from the historical data set, including the temperature, vibration, current and pressure of the equipment, as well as the time and type of the fault. The comprehensive feature set is input into the convolutional neural network model CNN to extract high-level features. The high-level features are input into the dangerous area prediction model to extract time series features. The time series features are input into the fully connected layer to predict the probability of the dangerous area, the type of the dangerous area, and the appearance time of the dangerous area respectively. According to the probability distribution of the type of dangerous area, the type of dangerous area with the highest probability is selected as the predicted type of dangerous area. According to the appearance time of the dangerous area, the specific time point when the equipment is expected to fail is determined. A threshold is set. When the probability of the dangerous area exceeds the threshold, the system will automatically notify the maintenance personnel to immediately check and repair the equipment. When the probability of the dangerous area is lower than the threshold, it means that the failure risk of the equipment is within the allowable range, and the equipment operation status continues to be monitored and the equipment is checked regularly.

[0178] The details are shown in Table 1 below:

[0179] Table 1 Experimental record table

[0180] Test subjects Liquid level(m) Temperature (°C) Pressure(Pa) Probability of Dangerous Area Types of Hazardous Areas Danger zone appearance time (hours) Node 1 10.2 25.3 101325 0.15 Type A 24 Node 2 9.8 26.1 101350 0.20 Type B 36 Node 3 10.5 24.8 101300 0.08 Type A 48 Node 4 10.0 25.5 101330 0.25 Type B 12 Node 5 10.3 25.0 101310 0.18 Type A 30 Node 6 10.1 25.7 101340 0.12 Type B 42

[0181] From the table data, it can be seen that the probability of the dangerous area of ​​node 4 is 0.25, which exceeds the set threshold of 0.2; according to the probability distribution of the types of dangerous areas, the type of dangerous area of ​​node 4 is type B, and the expected occurrence time of the dangerous area is 12 hours; the system immediately generates an intelligent alarm signal to notify maintenance personnel to conduct inspections and repairs; this shows that by setting the threshold of the probability of the dangerous area, the system can detect high-risk equipment in time and avoid production interruptions and economic losses caused by failures;

[0182] By generating a comprehensive feature set, the system can make full use of the features of multiple sensor data to improve the accuracy and reliability of fault prediction. For example, the liquid level of node 2 is 9.8 meters, the temperature is 26.1°C, and the pressure is 101350Pa. These data are extracted and fused by CNN to generate high-level features, which are further input into the LSTM model for time series feature extraction. Finally, the probability of the dangerous area of ​​node 2 is 0.20, the type of dangerous area is type B, and the expected occurrence time of the dangerous area is 36 hours. This shows that multimodal data fusion can more comprehensively reflect the operating status of the equipment and reduce the risk of false alarms and missed alarms.

[0183] The central processing unit introduces a context-aware mechanism to dynamically adjust the threshold for identifying dangerous areas based on current environmental conditions and historical data. For example, the liquid level at node 5 is 10.3 meters, the temperature is 25.0°C, and the pressure is 101310Pa. The current environmental conditions are similar to the historical data fragments, and the system dynamically adjusts the threshold to 0.18. The probability of node 5 being in a dangerous area is 0.18, which is exactly equal to the adjusted threshold. The system generates an intelligent alarm signal to indicate the existence of a dangerous area. This shows that dynamically adjusting the threshold can better adapt to environmental changes and improve the accuracy and reliability of identification.

[0184] The preprocessing algorithm combines adaptive filtering and wavelet transform, which can perform preliminary denoising and feature extraction on the data at the node, reduce the computational burden of the central processing unit, and retain important features in the data. For example, the liquid level at node 3 is 10.5 meters, the temperature is 24.8°C, and the pressure is 101300Pa. The value range of the sensor data after preprocessing is [0,1], where the closer the value is to 0, the less the noise component in the data, and the closer the value is to 1, the stronger the signal component in the data. This shows that the preprocessing algorithm can effectively improve the quality of the data and enhance the robustness and stability of the system.

[0185] Through the above experimental data and analysis, it can be seen that the dangerous area identification method based on downhole liquid level monitoring of the present invention performs well in practical applications; the application of key technologies such as multimodal data fusion, dynamic adjustment of thresholds and preprocessing algorithms significantly improves the accuracy and reliability of fault prediction and reduces the risks of false alarms and missed alarms; compared with the prior art, the present invention has obvious advantages in fault prediction and dangerous area identification, can better adapt to complex downhole environments, and improve the overall performance and safety of the system.

[0186] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for identifying dangerous areas based on underground liquid level monitoring, characterized in that: include, Deploy adaptive sensor networks at key locations underground, dynamically adjust sensor distribution density according to environmental changes, adopt intelligent data collection strategies, dynamically adjust collection frequency according to environmental changes, use sensors to collect sensor data, and transmit sensor data to the central processing unit through low-power wireless communication technology; The central processing unit uses deep learning algorithms to extract and fuse features from different types of sensor data to generate a comprehensive feature set; Collect current environmental conditions and historical data, introduce context-aware mechanisms, dynamically adjust the thresholds for identifying dangerous areas based on current environmental conditions and historical data, and generate intelligent alarm signals; the specific steps are: The central processing unit monitors the temperature, humidity and pressure of the underground environment in real time to obtain the current environmental conditions; the nearest neighbor algorithm KNN is used to extract historical data fragments similar to the current environmental conditions; According to the current environmental conditions and historical data fragments, the threshold for identifying dangerous areas is dynamically adjusted. The threshold adjustment formula is: ； in, represents the adjusted threshold, represents the initial threshold, represents the adjustment factor, Indicates the number of historical data fragments, Represents an index variable, Indicates The weight of each historical data segment, Indicates Environmental conditions for each historical data segment, Indicates current environmental conditions; Compare the currently collected sensor data with the adjusted threshold Make a comparison; When the current sensor data exceeds the threshold When a dangerous area is detected, an intelligent alarm signal is generated to indicate the presence of a dangerous area; storing the generated intelligent alarm signal in a database of a central processing unit; A dangerous area prediction model is constructed using a deep learning algorithm. The dangerous area data of the downhole liquid level and environmental condition data are extracted from historical data and input into the dangerous area prediction model for training. The comprehensive feature set is input into the trained dangerous area prediction model, and maintenance recommendations are generated based on the output results of the dangerous area prediction model and the intelligent alarm signals.

2. The method for identifying dangerous areas based on underground liquid level monitoring according to claim 1, characterized in that: The deployment of an adaptive sensor network at key locations underground means: Deploy an initial sensor network at wellheads, major pipeline interfaces, and pump stations at downhole level monitoring points; Drones are used for auxiliary deployment so that the nodes corresponding to the sensor network are evenly distributed near the liquid level monitoring points. Each node is equipped with a liquid level sensor, a temperature sensor, and a pressure sensor to collect the height of the downhole hydraulic pressure, the temperature of the downhole environment, and the pressure of the downhole environment, respectively, and a unique node ID is marked for each node.

3. The method for identifying dangerous areas based on underground liquid level monitoring according to claim 1, characterized in that: The sensor distribution density is dynamically adjusted according to environmental changes, an intelligent data collection strategy is adopted, the collection frequency is dynamically adjusted according to environmental changes, sensor data is collected by sensors, and the sensor data is transmitted to the central processing unit through low-power wireless communication technology. The specific steps are: Each sensor acts as a node to regularly collect the downhole hydraulic pressure height, downhole environment temperature and downhole environment pressure, and pre-process the collected sensor data; The pre-processed sensor data is transmitted to the central processing unit via the low-power wireless communication technology LoRa; The central processing unit uses the K-means clustering algorithm to perform cluster analysis on the collected sensor data. Specifically, the K-means clustering algorithm divides the data points into several clusters, each cluster represents an area, and identifies the areas where the sensor density needs to be increased and decreased by analyzing the data density and distribution of each cluster. For the areas where sensors need to be increased, corresponding sensors are added, and for the areas where sensors need to be reduced, corresponding sensors are reduced. Set the initial acquisition frequency of the sensor to , set according to the maximum acquisition capacity of the equipment and user needs. The current downhole liquid level is The collection time interval is ; According to the current downhole liquid level Calculate the change in liquid level , the expression is: ； in, Indicates the downhole liquid level at the previous moment; According to the collection time interval Calculate the change in time , the expression is: ； Each node monitors the temperature, humidity and pressure of the underground environment in real time. The central processing unit dynamically adjusts the acquisition frequency of each node according to the changes in the underground environmental conditions. , the expression is: ； in, represents the initial acquisition frequency, represents the frequency adjustment coefficient, Indicates the change in liquid level. represents the change in time, Indicates the absolute value of the rate of change of liquid level; Select LoRa as the communication protocol; The pre-processed sensor data is stored in a database of the central processing unit.

4. The method for identifying dangerous areas based on underground liquid level monitoring according to claim 3, characterized in that: The central processing unit uses a deep learning algorithm to extract and fuse features from different types of sensor data to generate a comprehensive feature set. The specific steps are: Select convolutional neural network CNN as the deep learning model; The pre-processed downhole hydraulic pressure height, downhole environment temperature and downhole environment pressure are input into the deep learning model, and the convolutional neural network CNN is used to extract the features of the sensor data. The expression is: ； in, Indicates The feature vector of sensor data, represents the convolution operation, which is used to extract local features of sensor data. After preprocessing, Class sensor data, represents the weight matrix in the convolutional layer, which is used to learn the local features of sensor data. represents the bias term; The extracted features are fused to generate a comprehensive feature set, which is expressed as: ； in, represents the comprehensive feature set, represents the activation function ReLU, Indicates the number of sensor data types, Indicates The weight of the sensor data.

5. The method for identifying dangerous areas based on underground liquid level monitoring according to claim 1, characterized in that: The method uses a deep learning algorithm to construct a dangerous area prediction model, extracts dangerous area data of downhole liquid level and environmental condition data from historical data and inputs them into the dangerous area prediction model for training, and inputs the comprehensive feature set into the trained dangerous area prediction model. The specific steps are: Select the long short-term memory network LSTM as the dangerous area prediction model; The dangerous area prediction model structure includes an input layer, multiple LSTM layers, a fully connected layer, and an output layer; Extract the dangerous area data of downhole liquid level and environmental condition data from historical data and input them into the dangerous area prediction model for training; Based on the comprehensive feature set, a convolutional neural network model CNN is constructed using automatic feature extraction. The comprehensive feature set is input into the convolutional neural network model CNN to extract high-level features, which are expressed as: ； in, Represents high-level features; The high-level features are input into the dangerous area prediction model to extract time series features. The expression is: ； in, Represents time series characteristics; The time series features The input is sent to the fully connected layer to predict the probability of the dangerous area, the type of dangerous area and the appearance time of the dangerous area. The expression is: ； in, represents the probability of the danger zone, Indicates the type of hazardous area, Indicates the appearance time of the danger zone, represents the Sigmoid activation function, represents the Softmax activation function, , and The weight matrices representing the probability of dangerous areas, the types of dangerous areas, and the appearance time of dangerous areas, respectively, , and They represent the probability of dangerous areas, the type of dangerous areas and the bias term of the appearance time of dangerous areas, respectively. Indicates the dimension of the output.

6. The method for identifying dangerous areas based on underground liquid level monitoring according to claim 5, characterized in that: The specific steps of generating maintenance suggestions based on the output results of the dangerous area prediction model and the intelligent alarm signal are as follows: According to the type of hazardous area The probability distribution of the dangerous area is selected, and the type of dangerous area with the highest probability is selected as the type of dangerous area predicted; According to the time when the danger zone appears , determine the specific time when the equipment is expected to fail; Setting Thresholds ; when ≥ When the device is at a high risk of failure, the type of hazardous area and the time of appearance of the danger zone Notify maintenance personnel to immediately inspect and repair the equipment; when < When it is detected, it means that the risk in the hazardous area is within the allowable range. Continue to monitor the underground environment and regularly check the equipment and environmental conditions.

7. A system for identifying dangerous areas based on downhole liquid level monitoring, applied to the method for identifying dangerous areas based on downhole liquid level monitoring as claimed in any one of claims 1 to 6, characterized in that: It includes acquisition module, central processing unit, hazard identification module, hazard area prediction module and maintenance suggestion module. The acquisition module is used to deploy an adaptive sensor network at key locations underground, regularly collect the height of the underground hydraulic pressure, the temperature of the underground environment and the pressure of the underground environment, and pre-process the sensor data to reduce the calculation burden of the central processing unit, and transmit the sensor data to the central processing unit through low-power wireless communication technology; The central processing unit is used to perform preprocessing, cluster analysis and feature extraction on the collected sensor data, generate a comprehensive feature set, and store it in a database; The hazard identification module is used to introduce a context-aware mechanism to dynamically adjust the threshold for identifying hazardous areas based on current environmental conditions and historical data, and generate intelligent alarm signals; The dangerous area prediction module is used to build a dangerous area prediction model based on a comprehensive feature set using a deep learning algorithm LSTM, and completes training by extracting dangerous area data of downhole liquid level and environmental condition data from historical data and inputting them into the dangerous area prediction model; The maintenance suggestion module is used to generate maintenance suggestions based on the output results of the dangerous area prediction model and the intelligent alarm signal, and notify maintenance personnel to take corresponding measures.

Images