Electronic nose navigation method based on LSTM concentration gradient time sequence prediction assistance
Through the LSTM-based concentration gradient timing prediction auxiliary method, combined with the metal oxide gas sensor array, the problem that traditional sensors cannot meet the needs of special operations is solved, high-precision and efficient gas navigation is achieved, and navigation needs in special operations environments are met. .
Patent Information
- Application Number
- CN202411203121.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-06
AI Technical Summary
Traditional lidar sensors, millimeter wave sensors and camera sensors cannot meet the special operation needs in special gas environments. The olfactory analyzer is large in size and has low accuracy, cannot be portable, and has low efficiency in navigation.
Using an electronic nasal navigation method based on LSTM based concentration gradient timing prediction, data acquisition is performed using the MQ series metal oxide gas sensor array, data is processed through median filtering and Kriging interpolation method, and the gas concentration gradient field changes are predicted in combination with the LSTM model. , realize path planning and real-time navigation.
It improves the prediction accuracy and response speed of the gas navigation system, ensures the continuity and accuracy of navigation, can effectively locate and avoid obstacles, and meets the navigation needs in special operation environments.
Smart Images

Figure CN119085632A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of robot positioning and navigation, and in particular relates to an electronic nose navigation method assisted by LSTM concentration gradient time series prediction. Background Art
[0002] In special operation environments involving special gases, traditional conventional sensors such as lidar sensors, millimeter wave sensors, camera sensors, etc. cannot meet the needs of special operations. With the continuous innovation and application of related olfactory analysis technologies, new sensors such as "electronic noses" that are mainly based on smell are beginning to have a broader application space in the field of robot automatic navigation.
[0003] Olfactory analysis spectrometer belongs to the field of instrument technology. Its main application objects include important high-end equipment in the fields of chemical plants, chemical laboratories, military environment gas composition detection, and drug detection. Olfactory analysis spectrometer can effectively identify multiple composite gases at the same time by analyzing the composition and concentration of the gas. Its main principle is to measure the composition and concentration of the gas through the corresponding chemical gas sensor or infrared gas sensor, and convert it into a corresponding electrical signal to read from the pin, and process and analyze it in a high-performance single-chip microcomputer. However, the main disadvantage of the current olfactory gas analyzer is that the high-precision olfactory analyzer is large in size and cannot be widely used in a portable manner. Although the small olfactory analyzer is convenient, it has poor accuracy and low efficiency. With the development of artificial intelligence, people use data more efficiently and accurately. On the basis of obtaining a certain amount of data sets, advanced deep learning methods such as LSTM can be used to predict future concentration changes and apply them to the navigation of gas vehicles in advance. Summary of the invention
[0004] The present invention proposes an electronic nose navigation method assisted by LSTM concentration gradient time series prediction. Olfactory analysis is performed on special gases in special environments, and an MQ series metal oxide gas sensor array is used for data collection. Each sensor in the sensor array has high sensitivity and fast response time, and can monitor and capture small changes in gas concentration in real time. The collected signal is transmitted to the embedded central processing unit through AD conversion. The central processing unit uses a median filter algorithm to remove noise and perform data cleaning, and the obtained voltage data is converted into the real gas concentration by the voltage concentration conversion formula fitted by the sensor technical document. Further, the Kriging interpolation method is used to perform spatial modeling on the instantaneous spatial gas concentration to create a continuous odor concentration gradient field. And the high concentration area is identified according to the size of the response slope value. Since the concentration gradient of the gas in the gas navigation process has a time-varying characteristic, the present invention uses a concentration gradient time series prediction method based on LSTM to predict future concentration gradient field changes on the basis of collecting a certain amount of time-varying concentration series. Specifically, the LSTM model can capture the time series characteristics of the concentration gradient, thereby predicting the concentration distribution at future moments. Based on the prediction results, the system can plan the path in advance and achieve continuous movement. In addition, during the movement, the system parallel calculates the concentration gradient field after the movement to perform a posteriori correction. This real-time a posteriori correction mechanism ensures the accuracy of path planning and the continuity of navigation, thereby effectively achieving the continuity of positioning, navigation and obstacle avoidance of the leaking gas source.
[0005] Beneficial effects of the present invention:
[0006] 1. In special operating environments with special gases, traditional conventional sensors such as lidar sensors, millimeter wave sensors, camera sensors, etc. cannot meet the needs of special operations. With the continuous innovation and application of related olfactory analysis technologies, gas sensor arrays are applied to navigation, which broadens the dimensions of navigation methods.
[0007] 2. The present invention effectively improves the prediction accuracy and response speed of the gas navigation system by introducing the LSTM concentration gradient time series prediction method. Compared with the traditional method, the present invention can more accurately predict future concentration changes, plan the path in advance, and avoid frequent calculations and pauses during the navigation process. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] Figure 1 This is a flow chart of an electronic nose navigation method based on LSTM concentration gradient time series prediction assistance;
[0009] Figure 2 A diagram of the gas sensor array in an electronic nose navigation method assisted by LSTM concentration gradient time series prediction;
[0010] Figure 3This is the fitting curve for the conversion between voltage (resistance) and actual gas concentration in the electronic nose navigation method assisted by LSTM concentration gradient time series prediction.
[0011] Figure 4 Schematic diagram of the prediction of odor concentration in the electronic nose navigation method assisted by LSTM concentration gradient time series prediction. DETAILED DESCRIPTION
[0012] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the accompanying drawings, and the words "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.
[0013] The electronic nose navigation method of this embodiment is based on LSTM concentration gradient time series prediction assistance, as shown in the attached Figure 1 As shown, it mainly includes the following steps:
[0014] Step 1: Use a metal oxide gas sensor array for data acquisition. Each sensor in the sensor array monitors and captures changes in gas concentration in real time; the collected signals are transmitted to the embedded central processor through AD conversion.
[0015] Specifically, the hardware structure of the navigation part is as follows Figure 2 As shown, it is composed of MQ series metal oxide semiconductor gas sensors, which respectively constitute 9 gas sensor response signals. Furthermore, the central processing unit, A / D conversion module, power supply module, storage module, communication module and other modules are included.
[0016] Step 2: The central processor uses a median filter algorithm to remove noise and perform data cleaning. The obtained voltage data is converted into the actual gas concentration through the voltage-concentration conversion formula fitted in the sensor technical documentation.
[0017] Specifically, for the MQ-3 series of gas sensors, the gas concentration voltage digital signal collected in the central processor reflects the relative size of the gas concentration value. The conversion formula between the relative concentration and the true concentration requires the sensitivity characteristic curve in the sensor's technical documentation. The double exponential model was obtained by matlab fitting to obtain the relationship between ppm and gas sensor resistance (R s )relation As attached Figure 3 As shown, the voltage digital signal (V RL ) and the actual gas concentration (ppm).
[0018] Step 3: Use Kriging interpolation or inverse distance weighted interpolation to perform spatial modeling of instantaneous spatial gas concentrations, create a continuous odor concentration gradient field, and identify high-concentration areas based on the size of the response slope value;
[0019] Specific details will be attached Figure 2 The known gas concentration point in is denoted as z i , where i = 1-9, the gas concentration z′0 at an unknown point (x0, y0) in space is the weighted sum of some known gas concentration value points, and its calculation formula is Where n is a certain number of known points around the unknown concentration point, z i represents the attribute value of the i-th known point, w i represents the weight of the i-th known point. The semivariance of the known points in the i-th, j-th neighborhood is given by Solve, semi-variance data of unknown points select Gaussian model To fit, where σ is the total variance (variability) of the data, and a (range) is the interpolation parameter. The weight value is calculated according to the following formula
[0020]
[0021] In the formula represents the semivariance in the i,jth neighborhood, λ is a constant, and the 0th neighborhood point represents the point to be found itself.
[0022] Step 4: Use the LSTM-based concentration gradient time series prediction method to predict future concentration gradient field changes.
[0023] The encoder and decoder parts of the seq2seq model are constructed based on LSTM. Normalization is used to scale the data to the range of [0, 1]. The data structure is adjusted to create a sliding window prediction array X[n]. An n-layer LSTM encoder is built and the corresponding embedding layer is configured. The embedding layer converts the sample sequence into an embedding matrix. For each embedding vector at a moment, the encoder continuously reads the input sequence and updates its hidden state one by one.
[0024] The decoder is also composed of LSTM layers. The final hidden state H0 and cell state C0 of the encoder are used as the initial hidden state and initial hidden state of the decoder. The purpose of this is to transfer the context information learned by the encoder to the decoder so that the decoder can generate an output sequence based on a certain time series.
[0025] Finally, add the necessary fully connected layer as the final output, use MSE as the loss function, and use Adam as the optimizer for iterative training to continuously update the weights and biases in the LSTM model. The final effect is shown in the attached figure. Figure 4 As shown in the figure, the trained Lstm model can better obtain the time correlation information of the data and predict the future changes in odor concentration based on the temporal relationship in the input sequence.
[0026] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical schemes composed of any combination of the above technical features.
Claims
1. An electronic nose navigation method based on LSTM concentration gradient time series prediction assistance, Features: The specific steps include: Step 1: Use the MQ series metal oxide gas sensor array for data acquisition; each sensor in the sensor array monitors and captures changes in gas concentration in real time; the collected signal is transmitted to the embedded central processor through AD conversion; Step 2: The CPU uses the median filter algorithm to remove noise and perform data cleaning. The obtained voltage data is converted into the actual gas concentration through the voltage concentration conversion formula fitted in the sensor technical documentation; Step 3: Use Kriging interpolation method to model the instantaneous gas concentration in space and create a continuous odor concentration gradient field; and identify high-concentration areas based on the size of the response slope value; Step 4: Use the LSTM-based concentration gradient time series prediction method to predict future concentration gradient field changes.
2. According to claim 1, an electronic nose navigation method based on LSTM concentration gradient time series prediction assistance, Features: A single gas sensor mainly converts gas concentration information into electrical signals. The corresponding electrical signals are converted into digital signals through the A / D conversion module and transmitted to the central processing unit. The communication module is mainly used for communication with the QT host computer for real-time detection and feedback. The storage module is used to store the corresponding data. The power supply module is used to supply power to the other modules to ensure normal operation.
3. An electronic nose navigation method based on LSTM concentration gradient time series prediction assistance according to claim 1 or 2, Features: The gas concentration voltage digital signal collected in the central processor reflects the relative size of the gas concentration value. For the MQ series sensors, the conversion formula between the relative concentration and the true concentration is combined with the sensitivity characteristic curve. Select several fitting data points and use Matlab to fit the relationship between ppm and gas sensor resistance (R s ) relationship to obtain the double exponential model formula 1, and then combine the hardware circuit schematic to construct the voltage digital signal (V RL ) and the true gas concentration (ppm); V in formula 1 and 2 C , R L , R 0 is a constant value, which depends on the sensor model. a, b, c, and d are the constant parameters of the fitted model.
4. According to claim 1, an electronic nose navigation method based on LSTM concentration gradient time series prediction assistance, Features: Use the Kriging interpolation method described in step 3 to model the instantaneous spatial gas concentration field. 0 ,y 0 ) gas concentration z′ 0 It is the weighted sum of some known gas concentration value points: Where n is a certain number of known points around the unknown concentration point, z i represents the attribute value of the i-th known point, w i Represents the weight of the i-th known point; The semivariance of the known points in the i,jth neighborhood is given by Solution; Select the Gaussian model to fit the relationship between distance and semivariance to obtain the semivariance data of unknown points; Where σ is the total variance of the data, reflecting the overall range of the data, and a is called the range, which is the interpolation parameter. In order to avoid the interpolation being too smooth and difficult to identify high-concentration areas, the range should be gradually adjusted and reduced through the cross-validation method to improve the accuracy of gas concentration field modeling; The weight value is calculated as follows In the formula represents the semivariance in the i,jth neighborhood, λ is a constant, and the 0th neighborhood point represents the point to be found itself.
5. According to claim 1, an electronic nose navigation method based on LSTM concentration gradient time series prediction assistance, Features: Step 4: Based on LSTM, build the encoder and decoder parts of the seq2seq model, use normalization to scale the data to the range of [0, 1], adjust the data structure to create a sliding window prediction array X[n], build an n-layer LSTM encoder and configure the corresponding embedding layer. The embedding layer converts the sample sequence into an embedding matrix. For each embedding vector at a moment, the encoder continuously reads the input sequence and updates its hidden state one by one. The decoder also consists of LSTM layers, and the final hidden state H of the encoder is 0 and cell status C 0 It is used as the initial hidden state and initial hidden state of the decoder. The purpose of this is to transfer the context information learned by the encoder to the decoder so that the decoder can generate an output sequence based on a certain time series; Finally, add the necessary fully connected layer as the final output, use MSE as the loss function, and use Adam as the optimizer for iterative training to continuously update the weights and biases in the LSTM model. The trained LSTM model can better obtain the time correlation information of the data and predict the future changes in odor concentration based on the temporal relationship in the input sequence.
Citation Information
Cited By
Gas concentration prediction method based on static spectrum characteristics
CN120703178A
Gas concentration prediction method based on static spectral characteristics
CN120703178B
