Harmful gas detection method based on gas fingerprint spectrum and LSTM algorithm

Abstract

The invention relates to a harmful gas detection method based on a gas fingerprint spectrum and an LSTM algorithm. The harmful gas detection method comprises the following steps: collecting gas concentration data; generating a gas fingerprint spectrum; carrying out pretreatment and feature extraction on the gas fingerprint spectrum; and according to the extracted feature data, analyzing the time sequence data of the gas fingerprint spectrum through an LSTM neural network model, and predicting the type and concentration trend of the harmful gas. The method has the beneficial effects that the time sequence change characteristics in the gas fingerprint spectrum can be captured by combining the time sequence analysis capability of the LSTM algorithm, so that the existence of multiple gases in a complex background is identified, the real-time monitoring and detection are realized, and the adaptability to the multi-gas environment is effectively improved. Moreover, by combining with an LSTM deep learning algorithm, the method has self-learning and self-adaptive capabilities, can learn a dynamic change mode of gas from historical data, and can perform parameter adjustment according to an actual environment.

Description

Technical Field

[0001] The present invention relates to the field of gas detection, and more specifically, to a harmful gas detection method based on gas fingerprint spectrum and LSTM algorithm. Background Art

[0002] Traditional gas detection technology usually relies on a single or a few sensors, which can monitor a limited number of gases, and the detection accuracy of different gases is often limited by the performance of the sensor itself. Sensor aging, changes in external environment such as temperature and humidity can also have a significant impact on the detection results.

[0003] Moreover, in most detection systems, gas detection mainly relies on simple threshold judgment, triggering an alarm when the concentration of a certain gas exceeds the set value. This method is difficult to cope with complex multi-gas mixed environments, and is prone to false alarms or missed alarms, and cannot accurately distinguish different gas types.

[0004] With the acceleration of industrialization, the demand for harmful gas detection in complex environments continues to increase, and traditional detection methods cannot meet the needs of high precision and rapid response. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies in the prior art and propose a harmful gas detection method based on gas fingerprint and LSTM algorithm.

[0006] In the first aspect, a harmful gas detection method based on gas fingerprint spectrum and LSTM algorithm is provided, including: S1. Collect gas concentration data; S2. generating a gas fingerprint according to the gas concentration data; S3, preprocessing and feature extraction of the gas fingerprint; the preprocessing includes data cleaning, denoising and normalization processing; S4: Based on the feature data extracted by S3, the time series data of the gas fingerprint is analyzed through the LSTM neural network model, and the types and concentration trends of harmful gases are predicted.

[0007] As a preference, it also includes: S5. When the concentration of a certain harmful gas in the prediction result reaches the set threshold, an audible and visual alarm or a remote alarm is issued.

[0008] Preferably, in S1, the gas concentration data is collected by an array sensor composed of multiple gas sensors of different types, or the gas concentration data is collected by multiple samplings of a single sensor at different frequencies, temperatures or voltages.

[0009] Preferably, S4 includes: S401, constructing input data of the LSTM neural network model according to the feature data; S402, learning the time series features extracted from the input data through the LSTM neural network model; S403. According to the task type, the LSTM neural network model outputs one or more prediction values; the task type includes regression tasks and classification tasks.

[0010] Preferably, in S4, the LSTM neural network model is also optimized by an optimizer and a loss function; when the task type of the LSTM neural network model is a regression task, the loss function adopts a mean square error loss function, and when the task type of the LSTM neural network model is a classification task, the loss function adopts a cross entropy loss function.

[0011] In a second aspect, a harmful gas detection system based on a gas fingerprint and an LSTM algorithm is provided, which is used to execute any of the methods described in the first aspect, including: Gas collection module, used to collect gas concentration data; A gas fingerprint generation module, used to generate a gas fingerprint according to the gas concentration data; A data processing module, used for preprocessing and feature extraction of the gas fingerprint; the preprocessing includes data cleaning, denoising and normalization processing; The LSTM prediction module is used to analyze the time series data of the gas fingerprint map based on the feature data extracted by the data processing module through the LSTM neural network model, and predict the type and concentration trend of harmful gases.

[0012] As a preference, it also includes: The alarm module is used to issue an audible or visual alarm or a remote alarm when the concentration of a certain harmful gas in the prediction result reaches a set threshold.

[0013] According to a third aspect, a computer storage medium is provided, wherein a computer program is stored in the computer storage medium; when the computer program is executed on a computer, the computer executes any method described in the first aspect.

[0014] In a fourth aspect, an electronic device is provided, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement any method as described in the first aspect.

[0015] The beneficial effects of the present invention are: 1. The present invention combines the time series analysis capability of the LSTM algorithm to capture the time series change characteristics in the gas fingerprint, and then identify the presence of multiple gases in a complex background, realize real-time monitoring and detection, and effectively improve the adaptability to the multi-gas environment. In addition, by combining the LSTM deep learning algorithm, the present invention has the ability of self-learning and self-adaptation, can learn the dynamic change pattern of gas from historical data, and adjust parameters according to the actual environment. This intelligent adaptive capability reduces manual intervention and improves the flexibility and accuracy of the detection system in complex scenarios, so that it can maintain stable detection performance in a variety of environments.

[0016] 2. The present invention uses gas fingerprint technology to integrate the response data of multiple sensors to extract more detailed and comprehensive gas characteristics, significantly improving the accuracy of identifying harmful gases in complex gas environments and providing a reliable basis for efficient gas detection.

[0017] 3. The present invention utilizes the advantages of fast prediction and time series processing of LSTM algorithm to greatly improve the response speed of gas detection system, and can quickly capture the changes of harmful gas concentration and issue alarms in time. At the same time, the multi-dimensional data fusion processing of gas fingerprint technology enhances the anti-interference ability of the system, so that it can still maintain high reliability in the case of external noise and environmental changes, ensuring real-time detection and response. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of a harmful gas detection method based on gas fingerprint and LSTM algorithm provided by the present invention; Figure 2 Another flow chart of a harmful gas detection method based on gas fingerprint and LSTM algorithm provided by the present invention; Figure 3 A flow chart of gas concentration data collection and processing provided by the present invention; Figure 4 LSTM prediction flow chart provided by the present invention; Figure 5 It is a structural diagram of a harmful gas detection system based on gas fingerprint and LSTM algorithm; Figure 6 This is a structural diagram of another harmful gas detection system based on gas fingerprint and LSTM algorithm. DETAILED DESCRIPTION

[0019] The present invention is further described below in conjunction with embodiments. The description of the following embodiments is only used to help understand the present invention. It should be noted that for ordinary persons in the art, without departing from the principle of the present invention, the present invention can also be modified in some ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0020] Embodiment 1: In order to solve the problems of the prior art, this application provides a harmful gas detection method based on gas fingerprint and LSTM algorithm. The gas fingerprint technology is combined with LSTM algorithm, and multi-dimensional fingerprint data is used to perform time series analysis on gas characteristics through deep learning model. It can accurately identify gas types and concentrations, and has higher reliability and adaptability. Specifically, Figure 1 As shown, the method includes: S1. Collect gas concentration data.

[0021] Traditional gas detection methods often rely on a single sensor, which makes it difficult to accurately distinguish the mixed components of multiple gases and is easily affected by environmental factors, resulting in false alarms or missed alarms. In this regard, the present application collects gas concentration data through an array sensor composed of multiple different types of gas sensors. Specifically, Figure 3 As shown, the present application collects concentration signals of various gases in the gas transmission system in real time and outputs multi-dimensional sensor response values ​​as basic data for generating gas fingerprints.

[0022] S2. Generate a gas fingerprint according to the gas concentration data.

[0023] The fingerprint of each gas is based on the multi-dimensional response of the sensor to changes in gas concentration, and these data can be expressed in matrix form or graphically.

[0024] Illustratively, the relative differential impedance response of a conducting polymer in steady state is:

[0025] in, is the impedance of the sensor in its initial state; is the impedance of the sensor when it reaches equilibrium; The ratio of is usually used as a description of the sensor response.

[0026] If the bionic olfactory device is made of N An array of conductive polymer gas sensors is used to M If a mixed gas is analyzed, i Gas sensors and m The concentration relationship of the mixed gas components is:

[0027] in, is the constant term of the sensor; Indicates i The sensor pairs j Response of the gas; is the gas concentration constant.

[0028] So, N Gas sensor array M The concentration response matrix of the gas is:

[0029] By analyzing the recognition results of multiple matrix chips, the most current gas content data can be obtained.

[0030] Since any gas is composed of various complex chemical components, no sensor can be sensitive to only one odor. The bionic olfactory system takes advantage of the cross-sensitivity of the sensor and can identify complex odors by constructing a small-scale array.

[0031] In addition, an input vector is collected every other time step to create training data. For example, a set of outputs is generated every 1 second. , use the data from the past k seconds to predict future data.

[0032] It should be noted that the generated gas fingerprint can represent the concentration and characteristics of different gases in the current environment and provide high-dimensional input data for the subsequent LSTM model.

[0033] S3, preprocessing and feature extraction of the gas fingerprint spectrum; the preprocessing includes data cleaning, denoising and normalization.

[0034] In S3, preprocessing is used to eliminate the influence of environmental interference factors such as temperature, humidity, and noise signals, and feature extraction is used to further analyze the preprocessed data, extract representative gas feature values, and provide standardized input for the LSTM model. The purpose of S3 is to ensure the quality of data input to the LSTM model and reduce the influence of sensor aging or external environmental interference.

[0035] S4: Based on the feature data extracted by S3, the time series data of the gas fingerprint is analyzed through the LSTM neural network model, and the types and concentration trends of harmful gases are predicted.

[0036] like Figure 4 As shown, S4 includes: S401. Construct input data of the LSTM neural network model based on the feature data.

[0037] The LSTM model can be combined through multiple layers to form a deep neural network. This application first constructs the input layer of the LSTM neural network model. The input of the LSTM neural network model is a three-dimensional data: [number of samples, number of time steps, number of features]. Among them, the number of samples is the number of training data, the number of time steps is the size of the sliding window (for example, the past k seconds), and the number of features is the data dimension nnn of each time step. The dimension of the input layer depends on the shape of the data. For example, if the input data is 3 feature data collected every 3 seconds, the input layer shape is (3, 3), which means a matrix composed of 3 groups of 3 eigenvalues ​​each.

[0038] S402. Learn the time series features extracted from the input data through the LSTM neural network model.

[0039] Specifically, this application processes time series through the LSTM layer to capture the time series features in the data. The LSTM layer has two main parameters units and return_sequences that need to be set, where units represents the number of neurons in the LSTM layer and determines the learning ability of the model. If return_sequences is set to True, the LSTM layer will return the output of each time step. If it is set to False, it will only return the output of the last time step.

[0040] S403. According to the task type, the LSTM neural network model outputs one or more prediction values; the task type includes regression tasks and classification tasks.

[0041] The LSTM neural network model also includes a Dense layer, which is used to output the prediction results. Generally speaking, if you are doing a regression task (such as predicting a continuous value), the output layer will have one node. If it is a classification task, the number of nodes in the output layer will be set according to the number of categories.

[0042] In addition, in S4, the LSTM neural network model is optimized by an optimizer and a loss function, such as an Adam optimizer, which updates the network weights according to the result of the loss function. When the task type of the LSTM neural network model is a regression task, the loss function uses a mean square error loss function, and when the task type of the LSTM neural network model is a classification task, the loss function uses a cross entropy loss function.

[0043] The LSTM algorithm is good at processing time series data and can memorize and learn the changing patterns of historical data. Therefore, the application of the LSTM algorithm in gas detection can monitor the changing trend of gas concentration in real time and accurately identify potential harmful gases. This capability not only improves the real-time nature of detection, but also reduces the possibility of false alarms and missed alarms, making the entire detection process more intelligent and efficient.

[0044] Embodiment 2: On the basis of Example 1, Example 2 of the present application provides a more specific harmful gas detection method based on gas fingerprint and LSTM algorithm, such as Figure 2 As shown, including: S1. Collect gas concentration data.

[0045] The embodiment of the present application can adopt a multi-mode working mode of a single sensor, and generate gas response signals under different states through multiple sampling at different frequencies, temperatures or voltages, so as to obtain fingerprint characteristics of different gases. The multi-mode working mode of a single sensor can generate different response signals at a lower cost, reduce the number of sensors, and is suitable for detection scenarios with limited space or low cost.

[0046] S2. Generate a gas fingerprint according to the gas concentration data.

[0047] Gas fingerprints can also be generated using gas chromatography. The introduction of gas chromatography can more accurately analyze the component characteristics of complex gas mixtures, and is suitable for high-precision environmental detection needs with extremely low harmful gas content.

[0048] S3, preprocessing and feature extraction of the gas fingerprint spectrum; the preprocessing includes data cleaning, denoising and normalization.

[0049] S4: Based on the feature data extracted by S3, the time series data of the gas fingerprint is analyzed through the LSTM neural network model, and the types and concentration trends of harmful gases are predicted.

[0050] S5. When the concentration of a certain harmful gas in the prediction result reaches the set threshold, an audible and visual alarm or a remote alarm is issued.

[0051] Once the concentration of harmful gases exceeds the safe value, this application will issue an alarm in time to remind on-site staff to take measures to ensure the safety of the work area. At the same time, the prediction results can be fed back to the control center to achieve remote monitoring and decision support.

[0052] Furthermore, the traditional alarm method mainly relies on the local system for data processing and alarm triggering, while the gas detection data of this application can be uploaded to the cloud server for centralized processing through 5G or Internet of Things (IoT) technology. The system performs data analysis and triggers alarm signals in the cloud, while realizing remote monitoring and alarm push. Cloud-based data processing can realize remote monitoring and management, which is suitable for distributed detection systems, and can share data among multiple monitoring points for global monitoring and alarm management. This solution not only improves the real-time and intelligent level of the detection system, but also enhances the user's sense of security and trust.

[0053] In addition, this application uses AI technology to adaptively adjust the alarm threshold to avoid false alarms or missed alarms that may be caused by fixed thresholds. The system can learn the dynamic changes of the alarm threshold based on historical data and adaptively adjust the alarm trigger point under different environmental conditions. The AI ​​self-adjusting threshold mechanism can improve the sensitivity and accuracy of the alarm system and is suitable for scenarios with large fluctuations in gas concentration and complex environments.

[0054] It should be noted that the parts in this embodiment that are the same or similar to those in Embodiment 1 can be referenced to each other and will not be described in detail in this application.

[0055] Embodiment 3: On the basis of Example 1, Example 3 of the present application provides a harmful gas detection system based on gas fingerprint and LSTM algorithm, such as Figure 5 As shown, including: Gas collection module, used to collect gas concentration data; A gas fingerprint generation module, used to generate a gas fingerprint according to the gas concentration data; A data processing module, used for preprocessing and feature extraction of the gas fingerprint; the preprocessing includes data cleaning, denoising and normalization processing; The LSTM prediction module is used to analyze the time series data of the gas fingerprint map based on the feature data extracted by the data processing module through the LSTM neural network model, and predict the type and concentration trend of harmful gases.

[0056] Specifically, the system provided in this embodiment is a system corresponding to the method provided in Embodiment 1. Therefore, the parts in this embodiment that are the same or similar to those in Embodiment 1 can be referenced to each other and will not be repeated in this application.

[0057] Embodiment 4: Based on Example 2, Example 4 of the present application provides another harmful gas detection system based on gas fingerprint and LSTM algorithm, such as Figure 6 As shown, including: Gas collection module, used to collect gas concentration data; A gas fingerprint generation module, used to generate a gas fingerprint according to the gas concentration data; A data processing module, used for preprocessing and feature extraction of the gas fingerprint; the preprocessing includes data cleaning, denoising and normalization processing; The LSTM prediction module is used to analyze the time series data of the gas fingerprint map based on the feature data extracted by the data processing module through the LSTM neural network model, and predict the type and concentration trend of harmful gases.

[0058] The alarm module is used to issue an audible or visual alarm or a remote alarm when the concentration of a certain harmful gas in the prediction result reaches a set threshold.

[0059] Specifically, the system provided in this embodiment is a system corresponding to the method provided in Example 2. Therefore, the parts in this embodiment that are the same or similar to those in Example 2 can be referenced to each other and will not be repeated in this application.

[0060] In summary, this application achieves high-precision harmful gas detection in a complex environment with multiple gases by combining gas fingerprint generation with LSTM time series analysis technology. This solution can not only accurately and real-time identify the type and concentration changes of gases, but also provide more reliable gas monitoring and early warning functions through intelligent learning and prediction systems, which is suitable for gas monitoring needs in multiple fields such as industry and environmental protection.

Claims

1. A harmful gas detection method based on gas fingerprint and LSTM algorithm, characterized in that: include: S1. Collect gas concentration data; S2. generating a gas fingerprint according to the gas concentration data; S3, preprocessing and feature extraction of the gas fingerprint; the preprocessing includes data cleaning, denoising and normalization processing; S4: Based on the feature data extracted by S3, the time series data of the gas fingerprint is analyzed through the LSTM neural network model, and the types and concentration trends of harmful gases are predicted.

2. The harmful gas detection method based on gas fingerprint and LSTM algorithm according to claim 1 is characterized in that: Also includes: S5. When the concentration of a certain harmful gas in the prediction result reaches the set threshold, an audible and visual alarm or a remote alarm is issued.

3. The harmful gas detection method based on gas fingerprint and LSTM algorithm according to claim 2 is characterized in that: In S1, the gas concentration data is collected by an array sensor composed of multiple gas sensors of different types, or by multiple samplings of a single sensor at different frequencies, temperatures or voltages.

4. The harmful gas detection method based on gas fingerprint and LSTM algorithm according to claim 3 is characterized in that S4 include: S401, constructing input data of the LSTM neural network model according to the feature data; S402, learning the time series features extracted from the input data through the LSTM neural network model; S403. According to the task type, the LSTM neural network model outputs one or more prediction values; the task type includes regression tasks and classification tasks.

5. The harmful gas detection method based on gas fingerprint and LSTM algorithm according to claim 4 is characterized in that: In S4, the LSTM neural network model is also optimized by an optimizer and a loss function; when the task type of the LSTM neural network model is a regression task, the loss function adopts a mean square error loss function, and when the task type of the LSTM neural network model is a classification task, the loss function adopts a cross entropy loss function.

6. A harmful gas detection system based on gas fingerprint and LSTM algorithm, characterized in that: The method for executing any one of claims 1 to 5 comprises: Gas collection module, used to collect gas concentration data; A gas fingerprint generation module, used to generate a gas fingerprint according to the gas concentration data; A data processing module, used for preprocessing and feature extraction of the gas fingerprint; the preprocessing includes data cleaning, denoising and normalization processing; The LSTM prediction module is used to analyze the time series data of the gas fingerprint map based on the feature data extracted by the data processing module through the LSTM neural network model, and predict the type and concentration trend of harmful gases.

7. The harmful gas detection system based on gas fingerprint and LSTM algorithm according to claim 6 is characterized in that: Also includes: The alarm module is used to issue an audible or visual alarm or a remote alarm when the concentration of a certain harmful gas in the prediction result reaches a set threshold.

8. A computer storage medium, characterized in that The computer storage medium stores a computer program; when the computer program is executed on a computer, the computer executes any one of the methods described in claims 1 to 5.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method according to any one of claims 1 to 5.

Images