Combustible gas wireless monitoring system and method based on machine learning algorithm

Abstract

The invention discloses a combustible gas wireless monitoring system and method based on a machine learning algorithm, and the method employs a machine learning method to carry out the quick analysis of a small part of data during the preheating period of a sensor, and predicts the stable output of the sensor after preheating, thereby avoiding the waiting of a sensor preheating process as long as tens of seconds. Thus, most of power consumption of the sensor in the preheating process is saved, the limitation that the time for the sensor to enter the measurement state cannot be compressed due to the influence of preheating time in a traditional intermittent power supply method is broken through, and the working time of the measurement state in the working period of the catalytic combustion sensor is greatly shortened. The main power consumption of the catalytic combustion type combustible gas detection sensor or system is in the measurement state of the sensor, so that the power consumption of the catalytic combustion type sensor or system can be correspondingly and greatly reduced.

Description

Technical Field

[0001] This application relates to the technical field of sensors, and in particular, to a combustible gas wireless monitoring system and method based on machine learning algorithms. Background Art

[0002] In many fields such as industrial production and domestic gas supply, combustible gases are ubiquitous. Once leaked, they are extremely likely to trigger catastrophic accidents such as explosions and fires. Monitoring and leakage warning of combustible gases can effectively prevent the occurrence of accidents such as fires and explosions, reduce environmental pollution and property losses, and provide protection for the production and living environment.

[0003] The combustible gas sensors used in combustible gas monitoring systems mainly include catalytic combustion type, infrared spectroscopy type, laser absorption spectroscopy type, etc. Among them, the catalytic combustion type sensor has the advantages of low cost and simple design structure, and is often used for large-scale deployment. However, due to the catalytic combustion characteristics of the sensor, the power consumption of the sensor is relatively high, and it is not suitable for application scenarios with high measurement frequencies and long-term operation without mains power.

[0004] In the prior art, a method of optimizing the power consumption of the catalytic combustion type sensor by optimizing the processing technology of the sensor itself has been proposed, which effectively reduces the power consumption of the sensor. However, in the application of combustible gas monitoring systems that require operation without mains power and long battery life, the power consumption optimization effect is limited and still cannot meet the application requirements. Another prior art also proposes a power supply scheme through long cables to supply power to each methane gas monitoring node to support the high power consumption requirements of the monitoring nodes. However, due to the excessive length of the cables, it is severely restricted by the site in the application. In addition, there is also a scheme in the prior art that controls the system to alternately enter the "measurement" and "sleep" states through intermittent power supply. After the measurement of the combustible gas sensor is completed in the "measurement" state, the power supply of the sensor is turned off to enter the "sleep" state to reduce the power consumption waste during the interval between two measurements. However, the output of the catalytic combustion type sensor is unstable at the initial power-on, and it takes dozens of seconds to preheat the sensor to stabilize the sensor output, and then the signal of the sensor can be collected and measured. Based on such characteristics, the time to enter the "measurement" state (i.e., the time to supply power to the catalytic combustion type sensor) must be greater than the preheating time, and accurate and stable data can be collected only after the preheating is completed. The power consumption of the sensor during the preheating period is still wasted to a large extent, and the power consumption optimization effect is limited and cannot be applied to the application requirements of long battery life, high measurement frequency, and operation without mains power. Summary of the Invention

[0005] By providing a combustible gas wireless monitoring system and method based on machine learning algorithms, embodiments of this application can accurately monitor the concentration of combustible gases, give real-time alarms for exceeding the standard, and remotely transmit data while significantly reducing the system power consumption and cost, so as to adapt to diverse application scenarios.

[0006] To achieve the above object, the technical solution of the embodiment of the present invention is as follows:

[0007] In a first aspect, an embodiment of the present invention provides a combustible gas wireless monitoring system based on a machine learning algorithm, including: a signal acquisition module, a data processing and control module, a machine learning prediction module, an alarm module, a data transmission module, and a power management module; the signal acquisition module is configured to obtain an analog signal output by a catalytic combustion type methane sensor within a preset duration after power-on, and perform filtering, amplification, and analog-to-digital conversion processing on the acquired signal; wherein, the preset duration is less than the duration of the preheating stage of the sensor; the data processing and control module is configured to obtain the processed signal and transmit it to the machine learning prediction module, and at the same time control the sensor to switch between a measurement state and a sleep state according to a preset period; after the machine learning prediction module determines the actual concentration of the combustible gas, it determines whether the actual concentration of the combustible gas is within a preset safety range, and controls the alarm module and the data transmission module to work according to the determination result; the machine learning prediction module is configured to receive the processed signal, and analyze the characteristics of the processed signal according to a pre-trained machine learning algorithm, and predict the actual concentration of the combustible gas corresponding to the processed signal; wherein, the pre-trained machine learning algorithm is obtained by training with multiple groups of output data of the sensor during the preheating stage under different combustible gas concentration conditions; the alarm module is configured to perform a sound and light alarm after the data processing and control module determines that the combustible gas concentration exceeds the preset safety range; the data transmission module is configured to be connected to the data processing and control module through a communication interface, and is configured to remotely transmit the combustible gas concentration data collected by the system regularly under the control command of the data processing and control module; the power management module is configured to step down the battery voltage and supply power to all modules.

[0008] In some possible implementation manners, the signal acquisition module includes a filtering unit for filtering high-frequency noise, an amplification and impedance matching unit for amplifying and impedance matching the denoised signal, and an analog-to-digital conversion unit for converting the analog signal into a digital signal.

[0009] In some possible implementation manners, the filtering unit includes a first-order active low-pass filter, and its cut-off frequency is expressed as:

[0010]

[0011] wherein, R 1 is the resistance value of the resistor in the filter, and C 1 is the capacitance value of the capacitor in the filter.

[0012] In some possible implementation manners, the machine learning algorithm adopted by the machine learning prediction module is the Gaussian process regression algorithm; wherein, the training data set is expressed as:

[0013] D = (x, y) = {(x i , y i ) | i = 1, …, N};

[0014] wherein, x is an input vector, corresponding to the output sequence of the sensor in the warm-up data, y is the corresponding target output, corresponding to the true concentration value where the sensor is located, and N is the number of features of the training samples, that is, the length of the selected warm-up data;

[0015] For a new sensor output sequence X * , its predicted value Y * is expressed as:

[0016] Y * = K(X, X * )K(X, X) -1 y;

[0017] wherein, K is a kernel function, expressed as:

[0018]

[0019] wherein, A and B are arbitrary multi-dimensional random variables, and i and j are the row numbers and column numbers of the covariance matrix determined by the kernel function K.

[0020] In some possible implementation manners, the data processing control module controls the duration of the sensor in the measurement state to be the same as the preset duration, and the duration of the sleep state is the remaining duration of the preset period.

[0021] In some possible implementation manners, the alarm module includes a buzzer and an indicator light, and the high and low levels are output through the control pin of the data processing control module to control the buzzing of the buzzer and the lighting and extinguishing of the indicator light respectively.

[0022] In some possible implementation manners, the data transmission module includes a wired and / or wireless network communication module; the data from the data processing control module is uploaded to the Internet cloud or data is remotely transmitted through wired or wireless network communication technologies, and the data transmission module is connected to the data processing control module through a bidirectional communication interface.

[0023] In some possible implementation manners, the power management module includes step-down voltage regulation with a large voltage difference using a buck DCDC chip with low quiescent current and low shutdown current, and step-down voltage regulation with a small voltage difference using an LDO chip with low quiescent current and low shutdown current.

[0024] Second aspect, an embodiment of the present invention provides a combustible gas wireless monitoring method based on a machine learning algorithm, which is applied to the combustible gas wireless monitoring system described in the first aspect, and includes: collecting, filtering, and amplifying the analog signal output by the combustible gas sensor within a preset duration after power-on by the signal acquisition module, and transmitting the processed signal to the data processing and control module; the data processing and control module obtains the processed signal and transmits it to the machine learning prediction module, and at the same time controls the sensor to switch from the measurement state to the sleep state, and switches between the measurement state and the sleep state according to a preset period; the machine learning prediction module learns and analyzes the processed signal, and predicts the actual concentration of the combustible gas corresponding to the processed signal; the data processing and control module determines whether the actual concentration of the combustible gas is within the preset safety range, and when the combustible gas concentration exceeds the preset safety range, controls the alarm module to give an audible and visual alarm; at the same time, the data processing and control module regularly transmits the combustible gas concentration data collected by the system through the data transmission module, and controls the power management module to turn on and off the power of each module as needed.

[0025] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages:

[0026] In the embodiment of the present invention, aiming at the problem of excessive power consumption of the sensor during the preheating period caused by the need to wait for the sensor to complete preheating and stabilize before calculating the combustible gas concentration in the traditional intermittent power supply method for catalytic combustion sensors, a machine learning method is used to quickly analyze a small part of the data during the sensor preheating period and predict the stable output after the sensor preheating, so as to avoid waiting for the sensor preheating process that lasts for dozens of seconds. In this way, most of the power consumption of the sensor during the preheating process is saved, breaking through the limitation that the time for the sensor to enter the "measurement" state cannot be compressed due to the influence of the preheating time in the traditional intermittent power supply method, and greatly shortening the working time of the "measurement" state in the working cycle of the catalytic combustion sensor. Since the main power consumption of the catalytic combustion type combustible gas detection sensor or system is in the "measurement" state of the sensor, the power consumption of the catalytic combustion type sensor or system can be correspondingly reduced significantly. At the same time, because the machine learning algorithm can be embedded in a general embedded MCU without introducing new devices or circuits and introducing extremely little new cost and power consumption, it is applicable to almost all miniaturized and low-cost combustible gas monitoring systems, taking into account the low cost and low power consumption of the combustible gas monitoring system, and can be applied to high-frequency, long-term battery-powered applications that require measurement intervals of several minutes or even dozens of seconds, and large-scale low-cost application scenarios. Description of the Drawings

[0027] To more clearly illustrate the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for use in the embodiments of the present invention. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0028] Figure 1 It is a schematic structural diagram of a combustible gas wireless monitoring system provided for the embodiments of the present invention based on a machine learning algorithm;

[0029] Figure 2 It is a schematic diagram of the principle of the signal acquisition module in the embodiments of the present invention;

[0030] Figure 3 It is a schematic diagram of the system operation process under the control of the data processing and control module in the embodiments of the present invention;

[0031] Figure 4 It is the data output of the sensor in the first five seconds under different combustible gas concentrations;

[0032] Figure 5 It is a schematic diagram of the power consumption comparison curve between the traditional intermittent power supply method and the machine learning optimization method;

[0033] Figure 6 It is a schematic structural diagram of the alarm module in the embodiments of the present invention;

[0034] Figure 7 It is a schematic structural diagram of the data transmission module in the embodiments of the present invention;

[0035] Figure 8 It is a schematic structural diagram of the power management module in the embodiments of the present invention;

[0036] Figure 9 It is a schematic diagram of the embodiment process of a combustible gas wireless monitoring method based on a machine learning algorithm in the embodiments of the present invention. Specific embodiments

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0038] In the relevant descriptions of this embodiment, terms such as "including", "containing", "having", etc. are all open terms, and are generally preferably understood as including but not limited to; the term "at least one" is generally preferably understood as one or more, where "a plurality" means two or more; the term "at least one (item)" or its similar expression refers to any combination of these items, including any combination of single item(s) or plural item(s). For example, "at least one (item) among a, b or c", or, "at least one (item) among a, b and c" can all represent: a, b, c, a - b (i.e., a and b), a - c, b - c, or a - b - c, where a, b, c can be single or multiple respectively; the symbol "A / B" is used to describe the selection relationship of associated objects, and generally represents an "or" relationship between the front and the back.

[0039] In the following descriptions of this embodiment, the terms used in the embodiments of this application are only for the purpose of describing specific embodiments, and are not intended to limit this application. The singular forms "a" and "the" used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0040] Those skilled in the art should understand that in the following descriptions of the embodiments of this application, the sequence numbers do not mean the sequence of execution. Some or all of the steps can be executed in parallel or sequentially. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of this application.

[0041] Those skilled in the art should understand that the numerical ranges in the embodiments of this application should be understood as specifically disclosing each intermediate value between the upper and lower limits of the range. The intermediate value within any stated value or stated range, as well as each smaller range between any other stated value or intermediate value within the range, are also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded from the range.

[0042] Unless otherwise specified, the technical / scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs. Although this application only describes preferred methods and materials, any methods and materials similar or equivalent to those described herein can also be used in the implementation or testing of this application. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials related to the documents. In case of conflict with any incorporated document, the content of this specification shall prevail.

[0043] In order to illustrate the technical solution of the present invention, specific embodiments are used for illustration below.

[0044] In many fields such as industrial production and domestic gas supply, combustible gases are everywhere. Once leaked, they are extremely likely to trigger catastrophic accidents such as explosions and fires. Monitoring of combustible gases and leakage early warning can effectively prevent the occurrence of accidents such as fires and explosions, reduce environmental pollution and property losses, and provide protection for the production and living environment.

[0045] The combustible gas sensors used in combustible gas monitoring systems mainly include catalytic combustion type, infrared spectroscopy type, laser absorption spectroscopy type, etc. Among them, catalytic combustion type sensors have the advantages of low cost and simple design structure, and are commonly used for large-scale deployment. However, due to the catalytic combustion characteristics of the sensors, the power consumption of the sensors is relatively high, and they are not suitable for application scenarios with high measurement frequencies and long-term operation without mains power.

[0046] In the prior art, a method of optimizing the power consumption of catalytic combustion type sensors by optimizing the processing technology of the sensors themselves has been proposed, which effectively reduces the power consumption of the sensors. However, in the application of combustible gas monitoring systems that need to operate without mains power and have long battery life, the power consumption optimization effect is limited and still cannot meet the application requirements. In addition, in the prior art, a power supply scheme through long cables has been proposed to supply power to each methane gas monitoring node to support the high power consumption requirements of the monitoring nodes. However, due to the excessive length of the cables, it is severely restricted by the site in the application. In addition, there is also a scheme in the prior art that controls the system to alternately enter the "measurement" and "sleep" states through intermittent power supply. After the measurement of the combustible gas sensor is completed in the "measurement" state, the power supply to the sensor is turned off and it enters the "sleep" state to reduce the power consumption waste during the interval between two measurements. However, the output of the catalytic combustion type sensor is unstable at the initial power-on, and it takes dozens of seconds to preheat the sensor to stabilize the sensor output before the signal of the sensor can be collected and measured. Based on such characteristics, the time to enter the "measurement" state (i.e., the time to supply power to the catalytic combustion type sensor) must be greater than the preheating time, and accurate and stable data can be collected only after the preheating is completed. The power consumption of the sensor during the preheating period is still wasted to a large extent, and the power consumption optimization effect is limited and cannot be applied to the application requirements of long battery life, high measurement frequency and operation without mains power.

[0047] Based on this, the embodiments of the present invention provide a wireless combustible gas monitoring system and method based on machine learning algorithms, which can accurately monitor the concentration of combustible gases, give real-time alarms for exceeding the standard, and remotely transmit data while greatly reducing the system power consumption and cost, so as to adapt to diverse application scenarios.

[0048] Figure 1 The structural schematic diagram of a wireless combustible gas monitoring system based on machine learning algorithms provided for the embodiments of the present invention is shown in Figure 1As shown in the figure, the above-mentioned combustible gas wireless monitoring system based on machine learning algorithms may include: a signal acquisition module 11, a data processing and control module 12, a machine learning prediction module 13, an alarm module 14, a data transmission module 15, and a power management module 16;

[0049] The signal acquisition module 11 is configured to obtain the analog signal output by the catalytic combustion type methane sensor within a preset time period after power-on, and perform filtering, amplification, and analog-to-digital conversion processing on the acquired signal; wherein, the preset time period is less than the time period of the sensor preheating stage;

[0050] In some embodiments, the signal acquisition module 11 includes a filtering unit for filtering high-frequency noise, an amplification and impedance matching unit for amplifying and impedance-matching the denoised signal, and an analog-to-digital conversion unit for converting the analog signal into a digital signal.

[0051] Specifically, the signal acquisition module 11 realizes the filtering and amplification of the analog signal output by the sensor. A low-pass filter can be used to filter out the high-frequency noise above the cut-off frequency of the filter in the input signal to obtain a stable output signal, thereby realizing the denoising of the sensor signal. An in-phase proportional amplifier is formed by using an operational amplifier to amplify the denoised sensor signal. At the same time, the high input impedance and low output impedance of the operational amplifier are utilized to realize the impedance matching between the sensor output signal and the subsequent signal acquisition circuit. A high-precision analog-to-digital conversion chip can be selected to sample the denoised and amplified analog signal at the sampling frequency preset by the data processing and control module 12 and convert it into a digital signal, which is transmitted to the data processing and control module 12 through the data communication interface for data processing.

[0052] Exemplarily, Figure 2 is the schematic diagram of the signal acquisition module 11 in the embodiment of the present invention. Refer to Figure 2 As shown, the signal acquisition module 11 may specifically include a sensor interface terminal P1, an operational amplifier U1 (such as model: OPA4322AQPWRQ1), an analog-to-digital conversion chip U2 (ADS1115IDGSR), a capacitor C1 (10uF), a resistor R1 (10KΩ), a resistor R2 (1KΩ), a resistor R3 (1KΩ), a resistor R4 (4.7KΩ), and a resistor R5 (4.7KΩ). The input signal of the module is the output signal V_sensor of the sensor. The interface for the module to communicate with the data processing and control module 12 is the IIC communication interface (ADC_SCL and ADC_SDA). R4 and R5 are connected to 3.3V to pull up the IIC communication interface signal line to meet the requirement of high level when the communication interface is idle.

[0053] In some embodiments, the filtering unit includes a first-order active low-pass filter, which plays a role in signal amplification and impedance matching while filtering out high-frequency noise. The cut-off frequency of the filter is fs, and its cut-off frequency is expressed as:

[0054]

[0055] where R 1 is the resistance value of the resistor in the filter, and C 1 is the capacitance value of the capacitor in the filter.

[0056] High-frequency noise above fs will be filtered out to achieve signal denoising, which can be obtained using the above cut-off frequency calculation formula. For example, the cut-off frequency of the embodiment of the present invention is 1.6 HZ, which can effectively filter out high-frequency noise from the DC signal output by the sensor.

[0057] Furthermore, the high input impedance and low output impedance of the operational amplifier are used to achieve impedance matching between the module input signal (sensor output signal) and the signal acquisition circuit. At the same time, the operational amplifier, R2, and R3 form a non-inverting proportional amplifier. According to the "virtual short" principle of the operational amplifier, the relationship between the output signal V_adc and the input signal V_sensor of the operational amplifier can be expressed as:

[0058]

[0059] The data processing and control module 12 is used to obtain the processed signal and transmit it to the machine learning prediction module 13, and at the same time control the sensor to switch between the measurement state and the sleep state according to a preset period; after the machine learning prediction module determines the actual concentration of the combustible gas, it judges whether the actual concentration of the combustible gas is within the preset safety range, and controls the alarm module 14 and the data transmission module 15 to work according to the judgment result;

[0060] In some embodiments, the data processing control module 12 can control the sensor to periodically enter the "measurement" and "sleep" states to achieve low-power measurement of the combustible gas concentration. During normal operation, it controls the sensor and the system to enter the "measurement" state, acquires the sensor signals transmitted by the signal acquisition module 11, and transmits them to the machine learning prediction module 13 for analyzing the output signals of the sensor, and predicts the actual combustible gas concentration within a range much smaller than the sensor preheating time. After obtaining the prediction data from the machine learning prediction module 13, it determines whether the combustible gas concentration is within the preset safe range. If so, it issues a control instruction to turn off the power of all modules, and the entire system enters the "sleep" state, waiting for the next measurement cycle to come with extremely low power consumption. If the combustible gas concentration exceeds the preset safe range, it issues a control signal to the alarm module 14 for audible and visual alarms. At the same time, the data processing control module 12 remotely transmits the measured combustible gas concentration through the communication module according to a preset cycle.

[0061] Exemplarily, in the embodiments of the present invention, the data processing control module 12 can be a microcontroller unit (MCU) capable of implementing the above functions. For example, in the embodiments of the present invention, a low-power series STM32 single-chip microcomputer can be selected (for example, the model can be STM32L4P5CET6), whose CPU main frequency is 120MHZ, and it has 512KB of on-chip FLASH and 320KB of on-chip SRAM, which are used to store and run machine learning algorithms with relatively large volume and high computing power requirements. The module is responsible for controlling the sensor and the system to periodically enter the "measurement" and "sleep" states, acquiring the sensor signals transmitted by the signal acquisition module 11, transmitting them to the machine learning prediction module 13 for analyzing the sensor signals, predicting the actual combustible gas concentration, and at the same time controlling the alarm module 14 and the data transmission module 15 to work according to a preset logic.

[0062] Figure 3 It is a schematic diagram of the system operation process controlled by the data processing control module 12 in the embodiments of the present invention. Refer to Figure 3 As shown, the operation process of the above combustible gas wireless monitoring system can specifically include:

[0063] Step 0: Initialize the system clock and peripherals such as GPIO, RTC clock, IIC communication interface, and UART communication interface.

[0064] Step 1: Enable the power enable pins of the sensor and the signal acquisition module to enter the "measurement" state.

[0065] Step 2: Start the signal acquisition module, acquire the sensor signals from the signal acquisition module, and transmit them to the data processing control module.

[0066] Step 3: The data processing and control module transmits the sensor signal to the machine learning prediction module to analyze the sensor signal and predict the actual combustible gas concentration.

[0067] Step 4: The data processing and control module determines whether the combustible gas concentration exceeds the preset alarm threshold. If it exceeds the alarm threshold, it controls the alarm module to give an audible and visual alarm. At the same time, it enables the power supply enable pin of the data transmission module, sends an alarm message to the remote monitoring center, and re-executes Step 4. If it does not exceed the alarm threshold, it proceeds to Step 5.

[0068] Step 5: The data processing and control module determines whether the system timer has reached the preset time threshold for remote data transmission. If it exceeds the threshold, it enables the power supply enable pin of the data transmission module, controls the data transmission module to send the combustible gas concentration monitoring data to the remote monitoring center, and enters Step 6. If it does not exceed the threshold, it directly enters Step 6.

[0069] Step 6: Disable the power supply enable pins of the sensor, the signal acquisition module, and the data transmission module, and enter the "sleep" state.

[0070] Step 7: Determine whether the "sleep" state timing of the RTC timer has ended. If the timing has not ended, re-execute Step 7. If the timing has ended, return to Step 1 for a new measurement cycle.

[0071] It should be noted that the durations of the "measurement" state and the "sleep" state are adjustable. The duration of the "measurement" state can be set according to the duration of the warm-up period data trained and analyzed by the machine learning algorithm prediction module, while the duration of the "sleep" state occupies the remaining time of the measurement cycle. The shorter the "measurement" state time, the lower the system power consumption, but the shorter the data length that the machine learning algorithm can analyze, and the corresponding prediction accuracy will decrease. The durations of the system entering the two states can be adjusted according to the actual requirements for power consumption and accuracy.

[0072] The machine learning prediction module 13 is used to receive the processed signal and analyze the characteristics of the processed signal according to the pre-trained machine learning algorithm to predict the actual combustible gas concentration corresponding to the processed signal. Among them, the pre-trained machine learning algorithm is obtained by training with multiple groups of output data of the sensor during the warm-up stage under different combustible gas concentration conditions;

[0073] Specifically, the machine learning prediction module 13 completes the prediction of new data by learning a large amount of existing sensor data.

[0074] It is understandable that the output of a catalytic combustion type combustible gas sensor is unstable during the preheating stage, and this entire process usually lasts for several seconds or even dozens of seconds. In the embodiments of the present invention, through a large number of experiments and observations, it is found that there is a certain pattern during this unstable process, that is: at different combustible gas concentrations, the unstable output during the preheating stage of the sensor shows a regularity related to the concentration. Accordingly, in the embodiments of the present invention, a large number of experiments are conducted to obtain multiple groups of output data during the preheating stage of the sensor under different combustible gas concentration conditions, and the data within a short period of time at the beginning of power-on is intercepted to form a training set, and a machine learning algorithm is trained and learned. The trained algorithm can predict the actual concentration of the combustible gas based on a short segment of preheating data of the sensor under unknown combustible gas concentration conditions, without waiting for the completion of the preheating stage. Therefore, the system can turn off the power supply of the sensor and enter the "sleep" state before the preheating is completed, thereby greatly reducing the time to enter the "measurement" stage in the intermittent power supply method of traditional catalytic combustion type combustible gas sensors, and further reducing the system power consumption.

[0075] In some embodiments, for the duration of the preheating data intercepted for the training set, it can be appropriately adjusted. The shorter the intercepted time, the shorter the time for the sensor to enter the "measurement" stage, and the lower the system power consumption. However, the corresponding prediction accuracy of the algorithm will also decrease accordingly, and it can be adjusted as needed according to the actual requirements in actual applications.

[0076] In some embodiments, in the embodiments of the present invention, the machine learning prediction module 13 selects an appropriate machine learning algorithm according to the output data characteristics of the sensor. For example, in the embodiments of the present invention, a Gaussian process regression algorithm applicable to processing non-linear relationships and multi-variable regression can be selected. In order to save local computing resources, the algorithm model can be trained on a PC and transplanted to an MCU for operation.

[0077] Exemplarily, for the Gaussian process regression algorithm, the training data set D can be expressed as:

[0078] D = (x, y) = {(x i , y i ) | i = 1, …, N};

[0079] Among them, x is the input vector, corresponding to the output sequence of the sensor in the preheating data, y is the corresponding target output, corresponding to the true concentration value where the sensor is located, and N is the number of features of the training samples, that is, the length of the selected preheating data;

[0080] According to the calculation and derivation of the Gaussian process regression, for a new sensor output sequence X * , its predicted value Y * is expressed as:

[0081] Y * = K(X, X* )K(X,X) -1 y;

[0082] Among them, K is a kernel function, expressed as:

[0083]

[0084] Among them, A and B are arbitrary multi-dimensional random variables, and i and j are the row numbers and column numbers of the covariance matrix determined by the kernel function K.

[0085] In some embodiments, the data for algorithm training and prediction comes from the data in the preheating stage of a catalytic combustion type sensor. As shown in the appendix Figure 4 shown Figure 4 is the data output of the sensor in the first five seconds at different combustible gas concentrations. Figure 4 Shows the data output of the sensor in the 5s preheating stage at different combustible gas concentrations (0% LEL, 3.8% LEL, 11.5% LEL, 18.6% LEL, 25.2% LEL, 31.8% LEL, 40.0% LEL). The sampling time interval of the signal acquisition module 11 is 10 ms, so there are a total of 500 data points within 5s. The sensor is unstable at the beginning of preheating and requires at least 5s of preheating time to stabilize. For the traditional intermittent power supply method, it is necessary to wait for a complete preheating time to complete the measurement, and a large amount of power consumption is consumed by the sensor during the process. The Gaussian process regression machine learning method applied in the embodiments of the present invention can only analyze the data in the first small period of time in the preheating stage and predict the output of the sensor after preheating is completed.

[0086] Exemplarily, in the embodiments of the present invention, the data in the first 1s before the sensor preheating data can be intercepted for the training and prediction test of the algorithm model. Among them, the number of training samples is 100, and the number of test samples is 50. The prediction accuracy of the machine learning algorithm on the test samples and the measurement accuracy of the traditional intermittent power supply method can both reach 1% LEL, while the power consumption required for the machine learning method to complete the measurement is 20% of that of the traditional intermittent power supply method. As shown in the appendix Figure 5 shown Figure 5 is a schematic diagram of the power consumption comparison curve between the traditional intermittent power supply method and the machine learning optimization method. Figure 5 Shows the power consumption comparison curve between the traditional intermittent power supply method and the method of this embodiment within one sensor measurement cycle (60s). The method of this embodiment has a shorter duration of the high-power "measurement" state compared to the traditional intermittent power supply method, and the average power consumption of the entire measurement cycle is lower.

[0087] It is understandable that the longer the warm-up data duration trained and analyzed by the system, the higher the prediction accuracy through the above-mentioned machine learning algorithm, but the corresponding power consumption will also be higher. During the operation of the system, the sensor needs to work continuously to collect warm-up data. The longer the data collection time, the more power consumed by the sensor and the related signal processing module. Moreover, a longer warm-up data duration also means that the time for the system to be in the "measurement" state is correspondingly extended, which further increases the overall power consumption. Therefore, in the embodiments of the present invention, according to the requirements for power consumption and prediction accuracy in the actual application scenario, an appropriate preset duration for analysis and prediction can be comprehensively considered and selected.

[0088] Exemplarily, in some places with extremely high safety requirements and extremely sensitive to changes in combustible gas concentration, such as underground coal mines, chemical production workshops, etc., it is crucial to accurately and timely grasp the combustible gas concentration. In these scenarios, even if it will increase a certain amount of power consumption, it is necessary to ensure a high prediction accuracy. Therefore, a relatively long warm-up data duration can be selected to ensure that the machine learning algorithm can predict the combustible gas concentration as accurately as possible, timely discover potential safety hazards, and avoid dangerous accidents.

[0089] In some other scenarios that are more sensitive to power consumption and have relatively loose requirements for prediction accuracy, such as ordinary warehouses, some small office places, etc., more attention is paid to the long-term stable operation and low maintenance cost of the system. At this time, in order to reduce power consumption and extend the battery life, the warm-up data duration can be appropriately shortened.

[0090] For example, in this embodiment, in order to obtain lower power consumption and longer battery life, the warm-up data duration of the sensor trained and analyzed by the machine learning algorithm is reduced to 0.5 s, and the time for the system to enter the "measurement" state, that is, the power supply time of the catalytic combustion type sensor, is also reduced to 0.5 s.

[0091] The alarm module 14 is used to perform audible and visual alarms after the data processing and control module 12 determines that the combustible gas concentration exceeds the preset safety range;

[0092] In some embodiments, the alarm module 14 drives a buzzer and an indicator light to implement the function of audible and visual alarms. Among them, the buzzer alarm controls the cut-off and conduction of the bipolar transistor by outputting high and low levels through the control pin of the data processing and control module 12, and then controls the high and low levels of the gate of the field effect transistor connected to the collector of the bipolar transistor, and further controls the on and off between the buzzer connected to the drain of the field effect transistor and the power supply connected to the source of the field effect transistor, so as to control the buzzing of the buzzer. The anode of the light-emitting diode is connected to the power supply, and the indicator light alarm outputs high and low levels to the cathode of the light-emitting diode through the control pin of the data processing and control module 12, so as to control the lighting and extinguishing of the light-emitting diode.

[0093] Exemplarily, Figure 6 is a schematic structural diagram of the alarm module 14 in the embodiments of the present invention. Refer to Figure 6 As shown, the alarm module 14 includes a buzzer, a bipolar transistor Q1 (model: S8050), a field effect transistor Q2 (AO3401A), a switching diode D1 (1N4148WS), a light emitting diode D2 (NCD0805R1), resistors R6 (1KΩ), R7 (5KΩ), R8 (10KΩ) and R9 (4.7KΩ). The module input control signals are BUZZER and LED from the MCU. The high and low levels of the BUZZER control signal control the cut-off and conduction of the bipolar transistor Q1, and then control the high and low levels of the gate of the field effect transistor, further controlling the on and off of the buzzer power supply to achieve the control of the buzzer sound. D1 is used as a freewheeling diode and is connected in parallel at both ends of the buzzer to eliminate the voltage spike generated by the buzzer at the moment of switching. The LED control signal is connected to the cathode of the light emitting diode D2 to control the lighting and extinguishing of D2. The MCU data processing and control module 10 realizes the control of the sound and light alarm by controlling the BUZZER and LED signals.

[0094] The data transmission module 15 is used to connect to the data processing and control module 12 through a communication interface, and is used to remotely transmit the combustible gas concentration data collected by the system regularly under the control instruction of the data processing and control module 12;

[0095] In some embodiments, the data transmission module 15 adopts a wireless network communication module to remotely transmit the data from the MCU through wireless network communication technology. The data transmission module 15 is connected to the data processing and control module 12 through a bidirectional communication interface, obtains the data from the data processing and control module 12 through this set of communication interfaces and performs data remote transmission, and similarly transmits the received remote control signal to the data processing and control module 12 through this set of communication interfaces.

[0096] Exemplarily, Figure 7 is a schematic structural diagram of the data transmission module 15 in the embodiments of the present invention. Refer to Figure 7 As shown, the data transmission module 15 adopts a CAT1 communication module to upload the data from the MCU to the Internet cloud through 4G communication technology. It includes a 4G communication module (model: USR-LTE-7S4), capacitors C2 (0.1uF) and C3 (22uF). The bidirectional communication interface for the module to communicate with the MCU is the DATA_TX and DATA_RX serial communication interfaces, and the data from the MCU is obtained through this set of communication interfaces and data remote transmission is performed. Capacitors C2 and C3 are used to stabilize the 5V power supply voltage of the 4G communication module.

[0097] The power management module 16 is used to step down the battery voltage and supply power to all modules.

[0098] In some embodiments, the power management module 16 steps down the output voltage of the lithium battery to generate the supply voltages required by each module in the system. Among them, the high-voltage differential step-down is generated by a step-down DCDC chip with low static current and low shutdown current, and the high-efficiency characteristic of the DCDC chip in high-voltage differential conversion applications is used to reduce the system power consumption. The low-voltage differential step-down is generated by an LDO chip with low static current and low shutdown current, and the voltage generated by the DCDC chip is further stepped down for use by specific modules. The output stability characteristic of the LDO chip in low-voltage differential conversion applications is used to improve the system operation stability and measurement accuracy.

[0099] Exemplarily, Figure 8 is a schematic structural diagram of the power management module 16 in the embodiment of the present invention. Refer to Figure 8As shown, the power management module 16 steps down the output voltage of the 12V lithium battery to generate 5V, 3.3V, and 2.8V to power each module of the system. Among them, the high-voltage difference step-down from 12V to 5V is achieved by a buck-type DCDC chip, and the low-voltage difference step-down from 5V to 3.3V and 2.8V is achieved by LDO chips. The 5V power supply for the data communication module and the 3.3V power supply for the MCU are provided by independent power supply circuits. The P2 terminal is used to connect the 12V lithium battery. The DCDC chip U4 (model: TPS54202DDCR) and its peripheral resistor, capacitor, and inductor components form a buck topology network to step down the 12V output of the lithium battery to 5V. The enable pin of the chip is left floating, and the chip remains in the working state. The source of the field-effect transistor Q3 (model: AO3401A) is connected to the 5V power supply of the system, and the drain is connected to the power supply pin of the data communication module. The high and low levels of the 4G5V_EN control signal from the MCU data processing and control module 10 control the cut-off and conduction of the bipolar transistor Q4 (model: S8050), and then control the high and low levels of the gate of Q3, further controlling the conduction state between the source and drain of Q3, and then controlling the power switch of the data communication module. The LDO chips U5 (model: TPS7A0228PDBVR), U6 (model: TPS7A0233PDBVR), and U7 (model: TPS7A0233PDBVR) and their peripheral capacitor components form a linear voltage regulator buck circuit to step down the system power supply of 5V to 2.8V and 3.3V. The enable pin of U5 is connected to the control pin 2V8_EN of the MCU, and the MCU controls the 2.8V power supply for the sensor by controlling this control pin. The enable pin of U6 is connected to the control pin 3V3_EN of the MCU, and the MCU controls the power supply for other 3.3V power-consuming modules on the system by controlling this control pin. The enable pin of U6 is connected to the 5V input pin of U6, and the chip remains in the working state, providing continuous 3.3V power supply to the MCU to ensure that the RTC clock can still run when the MCU enters the sleep state.

[0100] When the system is in the "measurement" state, the MCU controls the control pins 2V8_EN, 3V3_EN, and 4G5V_EN to turn on the power supply for each module. After the measurement is completed, the MCU controls the control pins 2V8_EN, 3V3_EN, and 4G5V_EN to turn off the power supply for each module. At the same time, the MCU enters the low-power state, turns on the RTC clock for timing, and the system enters the "sleep" state. When the RTC timing is completed, it wakes up the MCU and ends the sleep state. The MCU controls the system to re-enter the "measurement" state and starts the next measurement cycle.

[0101] In some embodiments, for the above-mentioned solution of the combustible gas monitoring system, it can be further extended. For example, the single combustible gas measurement system in the above solution can be used as a basic monitoring node, and the data communication module of this monitoring node can be specifically replaced. A communication module based on LORA is used to replace the original communication method, while other modules maintain the same design and functions as those in the above embodiments.

[0102] If it is necessary to monitor combustible gas in a relatively large working area, multiple modified monitoring nodes can be reasonably and densely arranged in this area. These monitoring nodes operate independently at their respective positions and collect combustible gas data in the areas where they are located in real time. At the same time, a LORA data receiver can be specifically set at the center of the entire working area. The receiver undertakes the key task of data aggregation. It can receive the data transmitted from each monitoring node through the LORA communication module, and after integrating and processing these data, it serves as the data source for the monitoring center, thus successfully building a local combustible gas monitoring network.

[0103] From the perspective of power consumption and cost, the LORA communication module has lower power consumption and cost compared with the communication module used in the above embodiments. This enables, when arranging monitoring nodes on a large scale, not only to reduce the energy consumption cost of long-term operation, but also to effectively control the procurement and deployment costs of hardware devices, greatly improving the economic benefits and making this solution more suitable for cost-sensitive application scenarios. In terms of data security and privacy, since the data does not need to be uploaded to the cloud and all data interactions are only carried out within the local area network, the risk of data leakage that may occur during cloud storage and transmission of data is avoided, effectively protecting the privacy of the data, and it is particularly suitable for industrial production environments with high requirements for data security or other places sensitive to data privacy.

[0104] In the embodiments of the present invention, for the problem of excessive power consumption of the sensor during preheating caused by the need to wait for the sensor to complete preheating and stabilize before calculating the concentration of combustible gas in the traditional intermittent power supply method for catalytic combustion sensors, a machine learning method is used to quickly analyze a small part of the data during the sensor preheating period and predict the stable output of the sensor after preheating, so as to avoid waiting for the sensor preheating process that lasts for dozens of seconds. In this way, most of the power consumption of the sensor during the preheating process is saved, breaking through the limitation that the time for the sensor to enter the "measurement" state cannot be compressed due to the influence of the preheating time in the traditional intermittent power supply method, and greatly shortening the working time of the "measurement" state in the working cycle of the catalytic combustion sensor. Since the main power consumption of the combustible gas detection sensor or system of the catalytic combustion type is in the "measurement" state of the sensor, the power consumption of the catalytic combustion type sensor or system can also be correspondingly reduced significantly. At the same time, because the machine learning algorithm can be embedded in a general embedded MCU without introducing new devices or circuits and introducing very little new cost and power consumption, it is applicable to almost all miniaturized and low-cost combustible gas monitoring systems, taking into account the low cost and low power consumption of the combustible gas monitoring system, and can be applied to high-frequency, long-term battery-powered and large-scale low-cost application scenarios that require measurement intervals of several minutes or even dozens of seconds and are disconnected from the mains power supply.

[0105] Based on the same inventive concept, an embodiment of the present application also provides a wireless monitoring method for combustible gas based on a machine learning algorithm. Figure 9 It is a schematic flowchart of the implementation of a wireless monitoring method for combustible gas based on a machine learning algorithm in the embodiments of the present invention. Refer to Figure 9 As shown, the method may include:

[0106] S901, collect, filter, and amplify the analog signal output by the combustible gas sensor within a preset duration after power-on through the signal acquisition module, and transmit the processed signal to the data processing and control module;

[0107] S902, the data processing and control module obtains the processed signal and transmits it to the machine learning prediction module, and at the same time controls the sensor to switch from the measurement state to the sleep state, and switches between the measurement state and the sleep state according to a preset period;

[0108] S903, the machine learning prediction module learns and analyzes the processed signal, and predicts the actual concentration of combustible gas corresponding to the processed signal;

[0109] S904, the data processing and control module determines whether the actual concentration of combustible gas is within the preset safety range. When the concentration of combustible gas exceeds the preset safety range, it controls the alarm module to give an audible and visual alarm. At the same time, the data processing and control module regularly transmits the combustible gas concentration data collected by the system remotely through the data transmission module, and controls the power management module to turn on and off the power of each module as needed.

[0110] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, reference can be made to each other. The key point of each embodiment is to illustrate the differences from other embodiments.

[0111] The above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present application.

Claims

1. A combustible gas wireless monitoring system based on machine learning algorithm, characterized in that: include: Signal acquisition module, data processing and control module, machine learning prediction module, alarm module, data transmission module and power management module; The signal acquisition module is used to obtain the analog signal output by the catalytic combustion methane sensor within a preset time after power-on, and filter, amplify and perform analog-to-digital conversion on the obtained signal; wherein the preset time is less than the duration of the sensor preheating stage; The data processing control module is used to obtain the processed signal and transmit it to the machine learning prediction module, and at the same time control the sensor to switch between the measurement state and the sleep state according to a preset cycle; after the machine learning prediction module determines the actual concentration of the combustible gas, it determines whether the actual concentration of the combustible gas is within a preset safety range, and controls the alarm module and the data transmission module to work according to the judgment result; The machine learning prediction module is used to receive the processed signal, and analyze the characteristics of the processed signal according to a pre-trained machine learning algorithm to predict the actual concentration of the combustible gas corresponding to the processed signal; wherein the pre-trained machine learning algorithm is obtained by training with a plurality of sets of preheating stage output data of the sensor under different combustible gas concentration conditions acquired in advance; The alarm module is used to give an audible and visual alarm after the data processing and control module determines that the concentration of the combustible gas exceeds the preset safety range; The data transmission module is used to connect to the data processing control module through a communication interface, and is used to periodically transmit the combustible gas concentration data collected by the system under the control instruction of the data processing control module; The power management module is used to step down the battery voltage and then supply power to all modules.

2. The combustible gas wireless monitoring system according to claim 1, characterized in that: The signal acquisition module includes a filtering unit for filtering high-frequency noise, an amplifying and impedance matching unit for amplifying and impedance matching the denoised signal, and an analog-to-digital conversion unit for converting an analog signal into a digital signal.

3. The combustible gas wireless monitoring system according to claim 2, characterized in that: The filtering unit includes a first-order active low-pass filter, whose cut-off frequency is expressed as: Among them, R1 is the resistance value of the resistor in the filter, and C1 is the capacitance value of the capacitor in the filter.

4. The combustible gas wireless monitoring system according to claim 1, characterized in that: The machine learning algorithm used by the machine learning prediction module is the Gaussian process regression algorithm; wherein the training data set is represented as: D=(x,y)={(x i ,y i )|i=1,…,N}; Among them, x is the input vector, corresponding to the output sequence of the sensor in the preheating data, y is the corresponding target output, corresponding to the true concentration value of the sensor, and N is the number of features of the training sample, that is, the selected preheating data length; For a new sensor output sequence X * , its predicted value Y * It is expressed as: Y * =K(X,X * )K(X,X) -1 y; Among them, K is the kernel function, expressed as: Among them, A and B are multidimensional random variables, i and j are the row and column numbers of the covariance matrix determined by the kernel function K.

5. The combustible gas wireless monitoring system according to claim 1, characterized in that: The data processing control module controls the duration of the sensor in the measuring state to be the same as the preset duration, and the duration of the dormant state is the remaining duration of the preset cycle.

6. The combustible gas wireless monitoring system according to claim 1, characterized in that: The alarm module includes a buzzer and an indicator light, and the sounding of the buzzer and the on and off of the indicator light are respectively controlled by outputting high and low levels through the control pin of the data processing control module.

7. The combustible gas wireless monitoring system according to claim 1, characterized in that: The data transmission module includes a wired and / or wireless network communication module; the data from the data processing control module is uploaded to the Internet cloud or the data is transmitted remotely through a wired or wireless network communication technology, and the data transmission module is connected to the data processing control module through a two-way communication interface.

8. The combustible gas wireless monitoring system according to claim 1, characterized in that: The power management module includes a step-down DCDC chip with low quiescent current and low shutdown current for high-voltage difference step-down, and an LDO chip with low quiescent current and low shutdown current for low-voltage difference step-down.

9. A combustible gas wireless monitoring method based on a machine learning algorithm, applied to the combustible gas wireless monitoring system according to any one of claims 1 to 7, characterized in that: include: The signal acquisition module collects, filters and amplifies the analog signal output by the combustible gas sensor within a preset time after power-on, and transmits the processed signal to the data processing control module; The data processing control module acquires the processed signal and transmits it to the machine learning prediction module, and controls the sensor to switch from the measuring state to the sleeping state, and switches between the measuring state and the sleeping state according to a preset period; The machine learning prediction module learns and analyzes the processed signal and predicts the actual concentration of combustible gas corresponding to the processed signal; The data processing control module determines whether the actual concentration of combustible gas is within the preset safety range. When the concentration of combustible gas exceeds the preset safety range, the alarm module is controlled to sound and light alarm. At the same time, the data processing control module regularly transmits the combustible gas concentration data collected by the system through the data transmission module, and controls the power management module to turn on and off the power of each module as needed.

Images