A method and system for detecting toxic gases in power tunnels

By integrating multiple sensors and deep learning technologies, combining feature extraction and data fusion, the limitations of traditional toxic gas detection methods in power tunnels are solved, and comprehensive, accurate, real-time detection and early warning of toxic gases in power tunnels is achieved, ensuring that Run safely.

CN118707047BActive Publication Date: 2024-11-12GUANGDONG CHENGYU ENG CONSULTING SUPERVISION CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411183917.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-12
Estimated Expiration
2044-08-27

AI Technical Summary

Technical Problem

Traditional toxic gas detection methods in power tunnels rely on a single type of sensor, which is difficult to comprehensively and accurately reflect the gas conditions in the tunnel, and are susceptible to environmental interference, are costly or insensitive to changes in gas concentration.

Method used

Integrate multiple types of gas sensors and environmental sensors, combine feature extraction, deep learning models and data fusion technology, dynamic weighted fusion of gas features through pre-trained deep learning models, and generate dynamic thresholds using adaptive threshold models to trigger early warnings. mechanism.

Benefits of technology

It realizes comprehensive, accurate and real-time detection and early warning of toxic gases in power tunnels, improves the timeliness and reliability of detection, reduces the probability of false alarms and missed reports, and ensures the safety of personnel and facilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118707047B_ABST
    Figure CN118707047B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for detecting toxic gases in a power tunnel, which relate to the technical field of gas detection, and include obtaining gas data and environmental data through a sensor set deployed in the power tunnel, extracting features of the gas data collected by MOS sensors, electrochemical sensors and infrared gas sensors respectively to obtain gas features, performing dynamic weighted fusion on all extracted gas features using a pre-trained deep learning model to obtain dynamic fusion features, inputting the dynamic fusion features and the environmental data into a pre-trained adaptive threshold model to generate a threshold, outputting a dynamic threshold, comparing a resistivity change value with the dynamic threshold, and triggering an early warning mechanism when the resistivity change value exceeds the dynamic threshold. The present invention integrates multiple types of gas sensors and environmental sensors, and combines feature extraction, deep learning models and data fusion technology to achieve detection and early warning of toxic gases in power tunnels.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of gas detection, and in particular to a method and a system for detecting toxic gases in a power tunnel. Background Art

[0002] During the construction and operation of power tunnels, due to the complex and closed environment inside the tunnels, a variety of toxic and harmful gases are often generated or accumulated, such as methane (CH4), carbon monoxide (CO), hydrogen sulfide (H2S), etc. The presence of these gases not only threatens the life safety of workers in the tunnels, but may also cause damage to power facilities and affect the stable operation of the power system. Therefore, it is particularly important to conduct real-time monitoring and early warning of toxic and harmful gases in power tunnels.

[0003] Traditional methods for detecting toxic gases in power tunnels mainly rely on a single type of sensor, such as electrochemical sensors, MOS sensors or infrared gas sensors. However, each sensor has its own specific detection range and sensitivity, and it is often difficult to fully and accurately reflect the gas conditions in the tunnel when used alone. For example, electrochemical sensors have high detection sensitivity for specific gases, but are easily interfered by environmental factors; MOS sensors have fast response speeds, but poor long-term stability; although infrared gas sensors have a wide detection range, they are expensive and may not be sensitive enough to small changes in gas concentration.

[0004] Therefore, it is necessary to provide a method and system for detecting toxic gases in power tunnels to solve the above technical problems. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method and system for detecting toxic gases in power tunnels. By integrating various types of gas sensors and environmental sensors and combining feature extraction, deep learning models and data fusion technology, comprehensive, accurate and real-time detection and early warning of toxic gases in power tunnels are achieved.

[0006] The present invention provides a method for detecting toxic gases in a power tunnel, the method comprising the following steps:

[0007] S1: Acquire gas data and environmental data through a sensor set deployed in the power tunnel, wherein the sensor set includes a gas sensor array and an environmental sensor array, the gas sensor array is composed of a plurality of sensor nodes, and each sensor node integrates a MOS sensor, an electrochemical sensor and an infrared gas sensor;

[0008] S2: extracting features from the gas data collected by the MOS sensor, the electrochemical sensor and the infrared gas sensor respectively to obtain gas features corresponding to each sensor, wherein the gas feature of the MOS sensor is a resistivity change value, the gas feature of the electrochemical sensor is a current rate change value, and the infrared gas sensor is a spectral intensity change value;

[0009] S3: Use the pre-trained deep learning model to dynamically weight and fuse all the extracted gas features to obtain dynamic fusion features;

[0010] S4: Inputting the dynamic fusion feature and the environmental data into a pre-trained adaptive threshold model to generate a threshold, and outputting a dynamic threshold, wherein the adaptive threshold model is a neural network model trained based on historical data;

[0011] S5: Compare the resistivity change value at the same moment with the dynamic threshold, and trigger an early warning mechanism when the resistivity change value exceeds the dynamic threshold.

[0012] Preferably, in step S1, the environmental sensor array includes a temperature sensor, a humidity sensor and an air pressure sensor, and the environmental data includes temperature, humidity and air pressure.

[0013] Preferably, step S1 specifically includes the following steps:

[0014] S101: Initializing the MOS sensor, electrochemical sensor, infrared gas sensor, temperature sensor, humidity sensor and air pressure sensor, and acquiring respective raw gas data;

[0015] S102: Preprocess the acquired raw gas data to obtain gas data and environmental data.

[0016] Preferably, step S2 specifically includes the following steps:

[0017] S201: Obtaining a resistivity change value of the MOS sensor by smoothing the gas data of the MOS sensor and calculating a change amount based on a reference resistance value;

[0018] S202: Calculating a change amount based on a baseline current according to a current signal of the electrochemical sensor to obtain a current rate change value of the electrochemical sensor;

[0019] S203: Calculating the variation according to the spectrum data of the infrared gas sensor and based on the reference spectrum intensity to obtain a spectrum intensity variation value of the infrared gas sensor.

[0020] Preferably, step S3 specifically includes the following steps:

[0021] S301: normalizing the resistivity change value, the current rate change value, and the spectrum intensity change value;

[0022] S302: Based on the weight allocation mechanism in the pre-trained deep learning model, dynamically calculate the weight coefficient of the gas characteristics of each sensor to obtain a dynamic weight matrix;

[0023] S303: performing weighted summation on the normalized resistivity change value, current rate change value and spectral intensity change value according to the dynamic weight matrix to obtain a dynamic fusion feature.

[0024] Preferably, step S4 specifically includes the following steps:

[0025] S401: merging the dynamic fusion feature and the environmental data to obtain a comprehensive feature vector;

[0026] S402: taking the feature vector as input and performing forward propagation using the input layer of the adaptive threshold model to obtain activation outputs of each hidden layer of the adaptive threshold model;

[0027] S403: performing a linear transformation on the output of the previous hidden layer according to the weight matrix and bias term of each hidden layer to obtain a linear combination of the hidden layer;

[0028] S404: Apply an activation function to perform nonlinear transformation on the linear combination of the hidden layer, and calculate the transformed result to obtain a dynamic threshold.

[0029] Preferably, in step S4, the training step of the adaptive threshold model includes:

[0030] Collecting multiple sets of training data of matching historical gas data and historical environmental data, and each set of historical gas data and historical environmental data is assigned a known historical dynamic threshold;

[0031] Inputting the training data into the adaptive threshold model, and continuously adjusting the weights and biases of the model through forward propagation and back propagation mechanisms to minimize the difference between the predicted dynamic threshold and the known historical dynamic threshold;

[0032] The adaptive threshold model is verified and optimized using independent verification data to obtain a trained adaptive threshold model.

[0033] Preferably, in step S5, the early warning mechanism includes sound and light alarms, remote notifications, and activation of an automatic ventilation system.

[0034] The present invention provides a power tunnel toxic gas detection system for executing the power tunnel toxic gas detection method. The detection system comprises:

[0035] A data acquisition module, used to acquire gas data and environmental data through a sensor set deployed in the power tunnel, wherein the sensor set includes a gas sensor array and an environmental sensor array, the gas sensor array is composed of a plurality of sensor nodes, and each sensor node is integrated with a MOS sensor, an electrochemical sensor and an infrared gas sensor;

[0036] A feature extraction module is used to extract features from the gas data collected by the MOS sensor, the electrochemical sensor and the infrared gas sensor, respectively, to obtain gas features corresponding to each sensor, wherein the gas feature of the MOS sensor is a resistivity change value, the gas feature of the electrochemical sensor is a current rate change value, and the infrared gas sensor is a spectral intensity change value;

[0037] The feature fusion module is used to dynamically weight and fuse all extracted gas features using a pre-trained deep learning model to obtain dynamic fusion features;

[0038] A threshold generation module, used for inputting the dynamic fusion feature and the environmental data into a pre-trained adaptive threshold model to generate a threshold and output a dynamic threshold, wherein the adaptive threshold model is a neural network model trained based on historical data;

[0039] The early warning decision module is used to compare the resistivity change value at the same time with the dynamic threshold, and trigger the early warning mechanism when the resistivity change value exceeds the dynamic threshold.

[0040] Compared with the related art, the method and system for detecting toxic gas in a power tunnel provided by the present invention have the following beneficial effects:

[0041] The present invention acquires gas data and environmental data through a sensor set deployed in a power tunnel, performs feature extraction on gas data collected by MOS sensors, electrochemical sensors and infrared gas sensors respectively, obtains gas features, uses a pre-trained deep learning model to dynamically weighted fuse all extracted gas features, obtains dynamic fusion features, inputs the dynamic fusion features and the environmental data into a pre-trained adaptive threshold model for threshold generation, outputs a dynamic threshold, compares the resistivity change value with the dynamic threshold, and triggers an early warning mechanism when the resistivity change value exceeds the dynamic threshold. The present invention integrates multiple types of gas sensors and environmental sensors, combines feature extraction, deep learning models and data fusion technology, and realizes detection and early warning of toxic gases in power tunnels. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flow chart of a method for detecting toxic gases in a power tunnel provided by the present invention;

[0043] Figure 2 This is a module structure diagram of a power tunnel toxic gas detection system provided by the present invention. DETAILED DESCRIPTION

[0044] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only the parts related to the present invention, rather than all structures, are shown in the accompanying drawings. In addition, the embodiments of the present invention and the features in the embodiments may be combined with each other without conflict.

[0045] It should also be noted that, for ease of description, only the parts related to the present invention, but not all of the contents, are shown in the accompanying drawings. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the various operations (or steps) as being processed sequentially, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the various operations can be rearranged. The processing can be terminated when its operation is completed, but it can also have additional steps not included in the accompanying drawings. The processing can correspond to methods, functions, procedures, subroutines, subprograms, etc.

[0046] In the present application, the sensor set includes a gas sensor array and an environmental sensor array, which are arranged in a power tunnel to continuously monitor the gas composition and environmental conditions in the tunnel.

[0047] The gas sensor array consists of multiple sensor nodes, each of which integrates multiple types of sensors to comprehensively capture the presence and concentration of different types of gases. These sensor types mainly include:

[0048] MOS sensor (Metal Oxide Semiconductor): This type of sensor can detect various gases, such as carbon monoxide, hydrogen sulfide, etc. Its working principle is to reflect the presence of gas by measuring the change in resistance on the sensor surface. When the target gas molecules are adsorbed on the sensor surface, the conductivity of the semiconductor material will change, causing a change in resistivity.

[0049] Electrochemical sensors: They detect gases through changes in electrical current produced by chemical reactions. For example, oxygen sensors produce a current through the reduction reaction of oxygen molecules at the electrode, and the magnitude of the current is proportional to the oxygen concentration.

[0050] Infrared gas sensor: Based on the absorption characteristics of different gases to specific wavelengths of infrared light, infrared gas sensors can accurately identify and quantify the concentration of specific gases. By analyzing the changes in spectral intensity, the type and concentration of the gas can be determined.

[0051] The environmental sensor array collects environmental data within the tunnel, including temperature, humidity, and air pressure, which are critical for interpreting the readings from the gas sensors, as environmental conditions may affect the response of the gas sensors.

[0052] Each sensor node integrates MOS sensors, electrochemical sensors and infrared gas sensors to ensure that the presence of toxic gases can be accurately detected even in complex and changing environments. These sensor nodes are distributed in key locations in the tunnel to cover the entire monitoring area and ensure that any potential gas leaks or anomalies can be detected in a timely manner. Embodiment 1

[0053] The present invention provides a method for detecting toxic gases in a power tunnel, referring to Figure 1 As shown, the detection method comprises the following steps:

[0054] S1: Acquire gas data and environmental data through a sensor set deployed in a power tunnel, wherein the sensor set includes a gas sensor array and an environmental sensor array, the gas sensor array is composed of a plurality of sensor nodes, and each sensor node integrates a MOS sensor, an electrochemical sensor and an infrared gas sensor.

[0055] In this embodiment, a sensor set is deployed inside the power tunnel, which includes a gas sensor array and an environmental sensor array. The gas sensor array consists of multiple sensor nodes, each of which is integrated with a metal oxide semiconductor (MOS) sensor, an electrochemical sensor, and an infrared gas sensor, which are used to fully monitor the gas composition in the power tunnel. At the same time, the environmental sensor array monitors environmental parameters such as temperature, humidity, and air pressure, which are crucial for correctly interpreting the readings of the gas sensor. By acquiring gas data and environmental data in real time, the system can quickly respond to environmental changes in the tunnel, improving the timeliness and accuracy of detection.

[0056] S2: Feature extraction is performed on the gas data collected by the MOS sensor, electrochemical sensor and infrared gas sensor respectively to obtain the gas features corresponding to each sensor, wherein the gas feature of the MOS sensor is the resistivity change value, the gas feature of the electrochemical sensor is the current rate change value, and the infrared gas sensor is the spectral intensity change value.

[0057] In this embodiment, features are extracted from gas data collected by MOS sensors, electrochemical sensors and infrared gas sensors, respectively. The gas features of MOS sensors are reflected as resistivity change values, the gas features of electrochemical sensors are reflected as current rate change values, and the gas features of infrared gas sensors are spectral intensity change values. This feature extraction method of multi-sensor data can more comprehensively reflect the diversity and complexity of gases in power tunnels, and increase the robustness and sensitivity of the detection system.

[0058] S3: Use the pre-trained deep learning model to dynamically weight and fuse all the extracted gas features to obtain dynamic fusion features.

[0059] In this embodiment, a pre-trained deep learning model is used to dynamically weight all extracted gas features to obtain a comprehensive dynamic fusion feature. The deep learning model can automatically adjust the weights of different sensor features to adapt to the changing environmental conditions in the power tunnel by learning patterns in historical data. This method overcomes the limitations that may exist in a single sensor and improves the reliability and accuracy of the detection results.

[0060] S4: Input the dynamic fusion feature and the environmental data into a pre-trained adaptive threshold model to generate a threshold, and output a dynamic threshold, wherein the adaptive threshold model is a neural network model trained based on historical data.

[0061] Since different sensor types have different response mechanisms to the same gas, MOS sensors respond to gas as a change in resistivity, electrochemical sensors detect gas through changes in current, and infrared gas sensors identify the presence of specific gases through absorption spectra. These three sensors each have advantages and limitations when detecting gas. For example, MOS sensors respond quickly to certain gases but have low selectivity, electrochemical sensors have high selectivity and sensitivity, and infrared gas sensors can accurately measure the concentration of specific gases. Therefore, combining their data can complement each other and improve the reliability and accuracy of the overall detection system.

[0062] For example, if at a certain moment, the current change rate and the spectral intensity change show a certain pattern of increasing gas concentration, even if the resistance value change rate has not yet reached an abnormal level, this trend will be reflected in the dynamic fusion feature.

[0063] When receiving the dynamic fusion features, the adaptive threshold model will take into account the changes in these related features and may adjust the threshold of the resistance value change rate in advance to make it more sensitive so that an early warning can be triggered at an early stage.

[0064] Therefore, in this embodiment, the dynamic fusion features and environmental data are input into the pre-trained adaptive threshold model, which is trained based on historical data and can generate dynamic thresholds that adjust as the environment changes. By using the adaptive threshold model, it is possible to intelligently determine when the gas concentration reaches a dangerous level, reducing the possibility of false alarms and missed alarms, and improving the effectiveness and intelligence of the early warning mechanism.

[0065] S5: Compare the resistivity change value at the same moment with the dynamic threshold, and trigger an early warning mechanism when the resistivity change value exceeds the dynamic threshold.

[0066] In this embodiment, when the resistivity change value at any time exceeds the dynamic threshold, the early warning mechanism is activated. The setting of this dynamic threshold avoids the limitations of fixed thresholds in complex environments and ensures that abnormal conditions of harmful gases can be accurately and timely detected under different conditions. The triggering of the early warning mechanism not only protects the lives of workers, but also effectively prevents damage to power facilities caused by gas leakage, reduces the risk of accidents, and ensures the safe operation of power tunnels.

[0067] Specifically, in step S1, the environmental sensor array includes a temperature sensor, a humidity sensor and an air pressure sensor, and the environmental data includes temperature, humidity and air pressure.

[0068] Specifically, step S1 specifically includes the following steps:

[0069] S101: Initialize the MOS sensor, electrochemical sensor, infrared gas sensor, temperature sensor, humidity sensor and air pressure sensor, and obtain their respective raw gas data.

[0070] In this embodiment, the MOS sensor, electrochemical sensor, infrared gas sensor, temperature sensor and humidity sensor are activated and ready to start collecting data. This process may include but is not limited to:

[0071] MOS Sensors: Metal Oxide Semiconductor (MOS) sensors need to be heated to operating temperature to achieve optimal sensitivity, so the initialization phase may include activation of the heating element.

[0072] Electrochemical sensors: These sensors rely on a chemical reaction between an electrolyte and an electrode. Initialization may involve stabilizing the voltage between the electrolyte and the electrode.

[0073] Infrared gas sensor: This type of sensor may require a stable light source and detector, and initialization may include calibration of the light source and detector.

[0074] Temperature and humidity sensors: These sensors are used to monitor environmental conditions and initialization may be as simple as powering on and preparing for data collection.

[0075] Once the sensors are ready, they begin to collect raw data, either continuously or at a certain frequency. This data may include:

[0076] Resistance change of MOS sensor output;

[0077] Changes in the current output by an electrochemical sensor;

[0078] Variations in the spectral intensity of infrared gas sensor outputs;

[0079] Temperature and humidity values ​​output by temperature sensor and humidity sensor;

[0080] The air pressure value output by the air pressure sensor.

[0081] S102: Preprocess the acquired raw gas data to obtain gas data and environmental data.

[0082] In this embodiment, after acquiring the raw data from different sensors, the next step is to preprocess the data to eliminate noise, standardize the data format, and extract useful information from the raw signals. The preprocessing steps include:

[0083] Data cleaning: Removing invalid or erroneous data points, such as outliers.

[0084] Signal filtering: Apply filtering techniques, such as a low-pass filter, to remove high-frequency noise.

[0085] Data normalization: Convert data to the same dimension or scale to facilitate comparison and analysis.

[0086] Specifically, step S2 specifically includes the following steps:

[0087] S201: Obtaining a resistivity change value of the MOS sensor by smoothing the gas data of the MOS sensor and calculating a change amount based on a reference resistance value.

[0088] In this embodiment, the raw data output by the MOS sensor is first smoothed to reduce fluctuations caused by noise and ensure data stability. Smoothing includes but is not limited to moving average, exponential smoothing or digital filter. The smoothed data is closer to the actual gas concentration change trend and reduces the risk of false alarms.

[0089] Next, the resistivity change is calculated based on the baseline resistance of the MOS sensor in clean air. This baseline resistance is measured in the absence of gas and is used as a reference point. The resistivity change is calculated by comparing the difference between the current resistance and the baseline resistance, and is usually expressed as a percentage of resistance change.

[0090] S202: Calculating a change amount based on the current signal of the electrochemical sensor and the baseline current to obtain a current rate change value of the electrochemical sensor.

[0091] In this embodiment, the feature extraction process of the electrochemical sensor also begins with baseline correction. The baseline current is the current reading when there is no gas to be measured, which provides the background value of the sensor response. By comparing the difference between the real-time current and the baseline current, the system can calculate the current rate change value. This change value reflects the intensity of the chemical reaction between the gas and the sensor electrode surface, and thus indicates the concentration of the gas.

[0092] S203: Calculating the variation according to the spectrum data of the infrared gas sensor and based on the reference spectrum intensity to obtain a spectrum intensity variation value of the infrared gas sensor.

[0093] In this embodiment, the infrared gas sensor identifies and quantifies the gas by detecting the absorption characteristics of specific gas molecules to infrared light. In this step, the reference spectral intensity, that is, the spectral intensity when there is no gas absorption, is used as the zero point reference. By analyzing the difference between the real-time spectral data and the reference spectral intensity, the spectral intensity change value is calculated. This change value is directly related to the gas concentration because specific gas molecules absorb infrared light of a specific wavelength, resulting in a decrease in the spectral intensity at that wavelength.

[0094] Specifically, step S3 specifically includes the following steps:

[0095] S301: normalizing the resistivity change value, the current rate change value, and the spectrum intensity change value.

[0096] In this embodiment, the obtained resistivity change value, current rate change value and spectral intensity change value are normalized. The purpose of normalization is to ensure that the data of all sensors are compared on the same scale to avoid a sensor dominating the final fusion result due to its large range. Normalization is to convert all values ​​to a fixed range, such as between 0 and 1. After this processing, the importance of each eigenvalue depends only on its relative change rather than its absolute value.

[0097] S302: Based on the weight allocation mechanism in the pre-trained deep learning model, the weight coefficients of the gas characteristics of each sensor are dynamically calculated to obtain a dynamic weight matrix.

[0098] In this embodiment, the purpose of the deep learning model is to enable the model to learn how to automatically adjust the weights of different sensor features according to the current environment and gas type. To this end, the model structure needs to be able to receive the feature data of all sensors as input and output a weight matrix, in which each element represents the weight coefficient of a sensor feature.

[0099] The training process of the model involves the use of a large amount of historical data, which includes sensor response data corresponding to different types and concentrations of gases under different environmental conditions. The goal of training is to minimize the gap between the predicted weights and the ideal weights (based on domain knowledge or expert judgment), and this is achieved by defining a loss function and using gradient descent to adjust the model parameters to gradually reduce the loss.

[0100] Once the model is trained, it can dynamically calculate the weight matrix based on the input sensor features. This means that in practical applications, whenever new sensor data is received, the model can update the weights in real time to reflect the importance of each sensor feature in the current environment.

[0101] For example, if the signal of the MOS sensor is severely interfered in a certain environment, the model may automatically reduce the weight of the MOS sensor feature and correspondingly increase the weight of the electrochemical sensor or infrared gas sensor feature to ensure the accuracy and reliability of the gas detection results.

[0102] S303: performing weighted summation on the normalized resistivity change value, current rate change value and spectral intensity change value according to the dynamic weight matrix to obtain a dynamic fusion feature.

[0103] In this embodiment, after obtaining the dynamic weight matrix, the normalized eigenvalues ​​are weighted and summed to generate dynamic fusion features. This process integrates the information of all sensors and weights them according to their relative importance, ensuring the comprehensiveness and accuracy of the final decision. The dynamic fusion feature represents a comprehensive assessment of the gas detection situation in the current environment, which can be used to further determine whether there is toxic gas and its possible concentration level.

[0104] Specifically, step S4 specifically includes the following steps:

[0105] S401: Merge the dynamic fusion features and environmental data to obtain a comprehensive feature vector.

[0106] In this embodiment, the previously calculated dynamic fusion feature is merged with the real-time environmental data by connecting the two vectors to form a new longer vector, i.e., a comprehensive feature vector. For example, if the dynamic fusion feature is a vector containing 10 elements and the environmental data is a vector containing 5 elements, the comprehensive feature vector obtained after merging will contain 15 elements.

[0107] S402: Taking the feature vector as input and performing forward propagation using the input layer of the adaptive threshold model, to obtain activation outputs of each hidden layer of the adaptive threshold model.

[0108] In this embodiment, the comprehensive feature vector is input into the adaptive threshold model, which is a deep neural network. The data starts from the input layer and passes through the nodes of each layer until it reaches the output layer. In this process, each layer will calculate according to its weights and biases to generate outputs, which will serve as the input of the next layer.

[0109] S403: According to the weight matrix and bias term of each hidden layer, a linear transformation is performed on the output of the previous hidden layer to obtain a linear combination of the hidden layer.

[0110] In this embodiment, in each hidden layer of the neural network, the output of the previous layer is firstly dot-multiplied with the weight matrix of the layer, and then a bias term is added, which is essentially a linear combination of data in a high-dimensional space.

[0111] S404: Apply an activation function to perform nonlinear transformation on the linear combination of the hidden layer, and calculate the transformed result to obtain a dynamic threshold.

[0112] In this embodiment, after completing the linear transformation, in order to introduce nonlinearity, the model will apply the ReLU activation function to the result of the linear combination. The role of the activation function is to allow the model to learn complex patterns and relationships, which is impossible for a linear model. After applying the activation function, the final output of the hidden layer is obtained, which will be used as the input of the next layer, or as the predicted output of the model in the last layer.

[0113] Specifically, in step S4, the training step of the adaptive threshold model includes:

[0114] A plurality of training data sets of matching historical gas data and historical environmental data are collected, and each set of historical gas data and historical environmental data is assigned a known historical dynamic threshold.

[0115] The training data is input into the adaptive threshold model, and the weights and biases of the model are continuously adjusted through forward propagation and back propagation mechanisms to minimize the difference between the predicted dynamic threshold and the known historical dynamic threshold.

[0116] The adaptive threshold model is verified and optimized using independent verification data to obtain a trained adaptive threshold model.

[0117] In this embodiment, multiple sets of matching historical gas data and historical environmental data are collected, which means that for each piece of historical gas data, there is corresponding environmental data corresponding to it, and each set of data should also have a known historical dynamic threshold, which is usually set based on past experience or expert knowledge.

[0118] The model training process includes:

[0119] Forward propagation: The training data (historical gas data and environmental data) is input into the adaptive threshold model, and the model performs forward propagation based on the current weights and biases to generate a predicted dynamic threshold.

[0120] Calculate the loss: Compare the dynamic threshold predicted by the model with the known historical dynamic threshold and calculate the difference between the two. This difference is usually quantified as a loss function.

[0121] Back propagation: Through the back propagation algorithm, the weights and biases of the model are adjusted according to the gradient of the loss function so that the predicted value is as close to the actual value as possible.

[0122] Iterative optimization: Repeat the forward propagation and backpropagation process until the performance of the model reaches a predetermined stopping condition, such as the loss function converges or reaches a preset number of training rounds.

[0123] Independent validation data: Use an independent validation dataset that was not used in training to evaluate the generalization ability of the model. This helps ensure that the model not only performs well on the training data, but can also accurately predict dynamic thresholds on new data.

[0124] Optimization and selection: Further adjust the model’s hyperparameters based on the performance of the validation data.

[0125] After the above steps, the trained adaptive threshold model can accurately predict the dynamic threshold when receiving new gas data and environmental data, thereby realizing real-time and intelligent monitoring of gas detection.

[0126] Specifically, in step S5, the early warning mechanism includes sound and light alarms, remote notifications, and activation of the automatic ventilation system.

[0127] In this embodiment, once the dynamic threshold is triggered, that is, when the real-time detected gas concentration exceeds the dynamic threshold, the early warning mechanism is immediately activated to ensure the safety of personnel and equipment in the power tunnel. The implementation of the early warning mechanism may include the following key components:

[0128] Siren: Siren installed in power tunnels emits a loud sound and flashing lights to immediately warn site personnel of the presence of potentially hazardous gases. This immediate alert helps to quickly evacuate personnel and prevent further exposure.

[0129] Communication system: The early warning system can be integrated with a communication module, which can send emergency notifications to remote monitoring centers or designated security managers via SMS, email or dedicated applications. This ensures that even personnel who are not on site can understand the situation in a timely manner and take necessary response measures.

[0130] Automatic control: The early warning mechanism can be linked to the ventilation system in the power tunnel. Once the early warning is triggered, the system will automatically start ventilation equipment such as fans and exhaust devices to accelerate the dilution of harmful gases and reduce the concentration to a safe level. This automated response can quickly improve the air quality in the tunnel and reduce the threat to personnel.

[0131] In addition, in this application, the sensor set not only includes the gas sensor array and the environmental sensor array, but also involves the intelligent optimization deployment of the sensor network. In order to improve the efficiency and accuracy of toxic gas detection in the entire power tunnel, the FW-PSO algorithm is also used to optimize the topology of the wireless sensor network.

[0132] Scale-free property: The FW-PSO algorithm uses its scale-free property to create a scientific WSN topology model. This model can automatically adjust the connection between sensor nodes to adapt to the complex and changing working environment in the power tunnel, ensuring that the network can maintain good connectivity and data transmission efficiency even under harsh conditions.

[0133] Network structure optimization: Through the FW-PSO algorithm, the network topology is optimized, ensuring a more reasonable layout of sensor nodes, reducing signal transmission delays and energy consumption, and enhancing the stability and robustness of the system.

[0134] Improve operational efficiency: The algorithm can identify and eliminate redundant paths in the network, which not only reduces the uncertainty of data transmission, but also greatly improves the operational efficiency of the wireless sensor network in the power tunnel. By reducing unnecessary data transmission paths, the system can complete the detection of toxic gases and data aggregation in a shorter time, speeding up the alarm response speed.

[0135] Multi-component mixed toxic gas detection: The integration of the WSN topology optimized by the FW-PSO algorithm and sensor technologies such as MOS sensors, electrochemical sensors, and infrared gas sensors provides an efficient toxic gas detection solution for the power tunnel construction site. This fusion technology can monitor multiple gases in real time, improve detection efficiency and alarm speed, and ensure the safety of construction workers.

[0136] The working principle of a method for detecting toxic gases in a power tunnel provided by the present invention is as follows: The present invention is based on multi-sensor fusion and deep learning technology. By deploying a sensor set in the power tunnel, it integrates various types of gas sensors such as MOS sensors, electrochemical sensors and infrared gas sensors, as well as an environmental sensor array, to achieve comprehensive collection of gas data and environmental data in the tunnel.

[0137] Based on data collection, the present invention further uses feature extraction technology to extract respective gas features from different types of sensor data, including resistivity change values ​​of MOS sensors, current rate change values ​​of electrochemical sensors, and spectral intensity change values ​​of infrared gas sensors. These feature data provide a rich source of information for subsequent data processing and analysis.

[0138] In order to make full use of these feature data and improve the accuracy and robustness of detection, the present invention introduces a pre-trained deep learning model for dynamic weighted fusion. The model can automatically adjust the weights of different sensor data according to the characteristics of real-time data, realize dynamic fusion of features, and thus obtain more comprehensive and accurate dynamic fusion features.

[0139] Finally, the present invention inputs the dynamic fusion features and environmental data into the pre-trained adaptive threshold model for threshold generation. The model is trained based on historical data and can dynamically adjust the warning threshold according to different environmental conditions and gas characteristics to ensure the accuracy and timeliness of the warning. When the detected resistivity change value exceeds the dynamic threshold, the system will automatically trigger the early warning mechanism to remind relevant personnel to take timely measures to ensure the safety of workers in the tunnel and the stable operation of power facilities.

[0140] In summary, the present invention realizes comprehensive, accurate and real-time detection and early warning of toxic gases in power tunnels by integrating various types of gas sensors and environmental sensors, combining feature extraction, deep learning models and data fusion technology. This method not only improves the accuracy and reliability of detection, but also reduces the probability of false alarms and missed alarms, providing a strong technical guarantee for the safe operation of power tunnels. Embodiment 2

[0141] The present invention provides a power tunnel toxic gas detection system for executing the power tunnel toxic gas detection method, referring to Figure 2 As shown, the detection system comprises:

[0142] The data acquisition module 100 is used to obtain gas data and environmental data through a sensor set deployed in the power tunnel, wherein the sensor set includes a gas sensor array and an environmental sensor array, the gas sensor array is composed of a plurality of sensor nodes, and each sensor node integrates a MOS sensor, an electrochemical sensor and an infrared gas sensor;

[0143] The feature extraction module 200 is used to extract features from the gas data collected by the MOS sensor, the electrochemical sensor and the infrared gas sensor, respectively, to obtain gas features corresponding to each sensor, wherein the gas feature of the MOS sensor is a resistivity change value, the gas feature of the electrochemical sensor is a current rate change value, and the infrared gas sensor is a spectral intensity change value;

[0144] The feature fusion module 300 is used to dynamically weight and fuse all extracted gas features using a pre-trained deep learning model to obtain a dynamic fusion feature;

[0145] A threshold generation module 400 is used to input the dynamic fusion feature and the environmental data into a pre-trained adaptive threshold model to generate a threshold and output a dynamic threshold, wherein the adaptive threshold model is a neural network model trained based on historical data;

[0146] The early warning decision module 500 is used to compare the resistivity change value at the same time with the dynamic threshold, and trigger the early warning mechanism when the resistivity change value exceeds the dynamic threshold.

[0147] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0148] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, the storage medium including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically-erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0149] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or includes elements that are inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.

Claims

1. A method for detecting toxic gases in a power tunnel, characterized in that: The detection method comprises the following steps: S1: Acquire gas data and environmental data through a sensor set deployed in the power tunnel, wherein the sensor set includes a gas sensor array and an environmental sensor array, the gas sensor array is composed of a plurality of sensor nodes, and each sensor node integrates a MOS sensor, an electrochemical sensor and an infrared gas sensor; S2: extracting features from the gas data collected by the MOS sensor, the electrochemical sensor and the infrared gas sensor respectively to obtain gas features corresponding to each sensor, wherein the gas feature of the MOS sensor is a resistivity change value, the gas feature of the electrochemical sensor is a current rate change value, and the infrared gas sensor is a spectral intensity change value; Step S2 specifically includes the following steps: S201: Obtaining a resistivity change value of the MOS sensor by smoothing the gas data of the MOS sensor and calculating a change amount based on a reference resistance value; S202: Calculating a change amount based on a baseline current according to a current signal of the electrochemical sensor to obtain a current rate change value of the electrochemical sensor; S203: Calculating the variation according to the spectrum data of the infrared gas sensor and based on the reference spectrum intensity to obtain a spectrum intensity variation value of the infrared gas sensor; S3: Use the pre-trained deep learning model to dynamically weight and fuse all the extracted gas features to obtain dynamic fusion features; Step S3 specifically includes the following steps: S301: normalizing the resistivity change value, the current rate change value, and the spectrum intensity change value; S302: Based on the weight allocation mechanism in the pre-trained deep learning model, dynamically calculate the weight coefficient of the gas characteristics of each sensor to obtain a dynamic weight matrix; S303: performing weighted summation on the normalized resistivity change value, current rate change value and spectral intensity change value according to the dynamic weight matrix to obtain a dynamic fusion feature; S4: Inputting the dynamic fusion feature and the environmental data into a pre-trained adaptive threshold model to generate a threshold, and outputting a dynamic threshold, wherein the adaptive threshold model is a neural network model trained based on historical data; Step S4 specifically includes the following steps: S401: merging the dynamic fusion feature and the environmental data to obtain a comprehensive feature vector; S402: taking the feature vector as input and performing forward propagation using the input layer of the adaptive threshold model to obtain activation outputs of each hidden layer of the adaptive threshold model; S403: performing a linear transformation on the output of the previous hidden layer according to the weight matrix and bias term of each hidden layer to obtain a linear combination of the hidden layer; S404: Applying an activation function to perform nonlinear transformation on the linear combination of the hidden layer, and calculating the transformed result to obtain a dynamic threshold; S5: Compare the resistivity change value at the same moment with the dynamic threshold, and trigger an early warning mechanism when the resistivity change value exceeds the dynamic threshold.

2. A method for detecting toxic gases in a power tunnel according to claim 1, characterized in that: In step S1 , the environmental sensor array includes a temperature sensor, a humidity sensor and an air pressure sensor, and the environmental data includes temperature, humidity and air pressure.

3. A method for detecting toxic gases in a power tunnel according to claim 2, characterized in that: Step S1 specifically includes the following steps: S101: Initializing the MOS sensor, electrochemical sensor, infrared gas sensor, temperature sensor, humidity sensor and air pressure sensor, and acquiring respective raw gas data; S102: Preprocess the acquired raw gas data to obtain gas data and environmental data.

4. A method for detecting toxic gases in a power tunnel according to claim 3, characterized in that: In step S4, the training step of the adaptive threshold model includes: Collecting multiple sets of training data of matching historical gas data and historical environmental data, and each set of historical gas data and historical environmental data is assigned a known historical dynamic threshold; Inputting the training data into the adaptive threshold model, and continuously adjusting the weights and biases of the model through forward propagation and back propagation mechanisms to minimize the difference between the predicted dynamic threshold and the known historical dynamic threshold; The adaptive threshold model is verified and optimized using independent verification data to obtain a trained adaptive threshold model.

5. A method for detecting toxic gases in a power tunnel according to claim 4, characterized in that: In step S5, the early warning mechanism includes sound and light alarm, remote notification and activation of the automatic ventilation system.

6. A power tunnel toxic gas detection system, used to execute a power tunnel toxic gas detection method according to any one of claims 1 to 5, characterized in that: The detection system comprises: A data acquisition module, used to acquire gas data and environmental data through a sensor set deployed in the power tunnel, wherein the sensor set includes a gas sensor array and an environmental sensor array, the gas sensor array is composed of a plurality of sensor nodes, and each sensor node is integrated with a MOS sensor, an electrochemical sensor and an infrared gas sensor; A feature extraction module is used to extract features from the gas data collected by the MOS sensor, the electrochemical sensor and the infrared gas sensor, respectively, to obtain gas features corresponding to each sensor, wherein the gas feature of the MOS sensor is a resistivity change value, the gas feature of the electrochemical sensor is a current rate change value, and the infrared gas sensor is a spectral intensity change value; The feature fusion module is used to dynamically weight and fuse all extracted gas features using a pre-trained deep learning model to obtain dynamic fusion features; A threshold generation module, used for inputting the dynamic fusion feature and the environmental data into a pre-trained adaptive threshold model to generate a threshold and output a dynamic threshold, wherein the adaptive threshold model is a neural network model trained based on historical data; The early warning decision module is used to compare the resistivity change value at the same time with the dynamic threshold, and trigger the early warning mechanism when the resistivity change value exceeds the dynamic threshold.

Citation Information

Patent Citations

  • Tunnel fire early-warning controlling method based on multi-sensor data fusion technology and system using the same

    CN103136893A

  • Gas analyzer

    CN105026915A

  • Aviation four-station guarantee comprehensive detecting box group

    CN110127082A

  • Method, apparatus and system for vaporization and concentration level determination of liquid substances

    CN114577980A

  • Double-light-path infrared thermal conductivity gas sensor

    CN217542849U