Low-concentration gas leakage detection method based on adaptive digital phase-locked amplification algorithm

Abstract

The application relates to the technical field of micro gas leakage detection, and particularly discloses a low-concentration gas leakage detection method based on an adaptive digital phase-locked amplification algorithm, which comprises the following steps: performing multi-source data synchronous acquisition to obtain original detection electric signals containing gas leakage information, pipeline infrared images, and measured pressure values and measured temperature values in a cavity; and performing integral time adaptive control process to generate an integral time control instruction based on the pipeline infrared images and the measured pressure values. By using a prediction-feedback double-loop control mechanism, the application breaks the inherent game of sensitivity and response speed of a traditional phase-locked amplifier, can dynamically adjust the integral time according to the diffusion trend of the infrared image and the pressure change characteristics, can ensure a very high signal-to-noise ratio through long integration in a stable period, and can realize millisecond-level fast response in a leakage burst period.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of micro gas leakage detection, in particular to a low-concentration gas leakage detection method based on an adaptive digital phase-locked amplification algorithm. BACKGROUND

[0002] Currently, gas leakage detection mainly relies on catalytic combustion sensors or standard infrared imaging technology, which are effective in dealing with large-scale leakage, but have significant deficiencies in detecting ppb-level micro leakage in complex industrial or outdoor deployment environments. The electrical signal generated by micro leakage is extremely weak and is easily overwhelmed by environmental noise, resulting in extremely low signal-to-noise ratio. Traditional digital phase-locked amplifiers usually use fixed integration time parameters when processing such signals, which in practical applications causes a conflict between sensitivity and response speed that is difficult to reconcile: if a longer integration time is set to suppress noise, the system's tracking ability for rapidly diffusing gas clouds or transient pressure changes will be significantly reduced, causing detection lag; if the integration time is shortened to improve response speed, noise cannot be effectively filtered out, resulting in a sharp increase in false positives.

[0003] In addition, there is a serious conflict between environmental interference and signal characteristics in actual operating conditions:

[0004] On the one hand, power frequency interference is often non-stationary, and when the power grid frequency drifts cause the interference frequency to coincide with the target carrier frequency, traditional fixed notch filters will face the dilemma of either not filtering out noise or damaging the signal, directly leading to signal loss;

[0005] On the other hand, external environmental thermal shock, such as rapid warming caused by direct sunlight, can cause gas thermal expansion effects in a sealed cavity. This physical pressure process can offset or even cover up the pressure drop characteristics caused by small leaks, making existing detection logic based on a single pressure threshold or rate completely ineffective in a high temperature rise environment, resulting in a highly concealed missed detection risk. SUMMARY

[0006] The present application aims to at least partially solve one of the technical problems in the related art. To this end, the purpose of the present application is to propose a low-concentration gas leakage detection method based on an adaptive digital phase-locked amplification algorithm to improve the detection accuracy of low-concentration gas.

[0007] To achieve the above-mentioned purpose, the first aspect of the present application proposes a low-concentration gas leakage detection method based on an adaptive digital phase-locked amplification algorithm, comprising the following steps:

[0008] Performing multi-source data synchronous acquisition to obtain original detection electrical signals containing gas leakage information, pipeline infrared images, and measured pressure values and measured temperature values in the cavity;

[0009] An integral time adaptive control process is performed to generate an integral time control instruction based on the pipeline infrared image and the measured pressure value, and the original detection electric signal is adaptively digitally phase-locked amplified using the integral time control instruction to extract a target demodulation signal;

[0010] Based on the target demodulation signal, a gas leakage detection result is determined by combining a multi-source data fusion strategy;

[0011] The integral time adaptive control process includes:

[0012] A prediction control stage: in response to a feature change rate of the pipeline infrared image and a second derivative feature of the measured pressure value, a concentration change trend label is determined, and an initial integral time of the adaptive digital phase-locked amplification processing is set according to the concentration change trend label;

[0013] A feedback correction stage: in response to a signal signal-to-noise ratio and a frequency phase error output by the adaptive digital phase-locked amplification processing, a step adjustment instruction for the initial integral time is generated to dynamically update the integral time control instruction until the signal signal-to-noise ratio and the frequency phase error meet a preset convergence condition.

[0014] To achieve the above purpose, the second aspect embodiment of the present application proposes a low-concentration gas leakage detection system based on an adaptive digital phase-locked amplification algorithm, which includes:

[0015] A multi-source perception module is configured to perform multi-source data synchronous acquisition to obtain an original detection electric signal containing gas leakage information, a pipeline infrared image, and a measured pressure value and a measured temperature value in a cavity;

[0016] A dual-loop control and signal processing module is configured to perform an integral time adaptive control process, set an initial integral time based on the pipeline infrared image and the measured pressure value through a prediction control stage, perform step adjustment of the integral time based on output signal quality through a feedback correction stage, and perform adaptive digital phase-locked amplification processing on the original detection electric signal using the finally determined integral time to extract a target demodulation signal;

[0017] A fusion decision module is configured to determine a gas leakage detection result based on the target demodulation signal, combine data credibility scores of multi-source data, and use a conflict coordination strategy;

[0018] A cooperative communication module is configured to extract a multi-dimensional scene feature vector and perform model parameter incremental update interaction with a cloud server.

[0019] To achieve the above object, the third aspect of the present application provides an electronic device, comprising a memory, a processor and a computer program stored in the memory, wherein the computer program is executed by the processor to realize the low-concentration gas leakage detection method based on adaptive digital phase-locked amplification algorithm.

[0020] Compared with the prior art, the present application has the following beneficial effects:

[0021] The low-concentration gas leakage detection method based on adaptive digital phase-locked amplification algorithm of the present application breaks the inherent game between sensitivity and response speed of the traditional phase-locked amplifier by using the prediction-feedback double-loop control mechanism, and can dynamically adjust the integral time according to the diffusion trend and pressure change characteristics of the infrared image, so as to ensure a high signal-to-noise ratio through long integration in the stable period, and realize a millisecond-level fast response in the leakage burst period.

[0022] Secondly, through the spectrum conflict monitoring and active carrier migration strategy, and the thermal masking compensation logic based on the isochromatic differential residual, the system has strong environmental adaptability, can actively avoid signal false suppression caused by frequency drift, and accurately separates the tiny leakage pressure characteristics from the thermal expansion background, so as to maintain high reliability detection of the gas trace leakage under complex electromagnetic environment and severe temperature fluctuation, and significantly improve the safety monitoring level of the gas facilities. BRIEF DESCRIPTION OF DRAWINGS

[0023] The disclosure of the present application will be described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present application. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0024] Figure 1 is a flowchart of the low-concentration gas leakage detection method based on adaptive digital phase-locked amplification algorithm provided by the present application;

[0025] Figure 2 is a step response simulation comparison diagram of integral time adaptive adjustment and fixed integral time in the low-concentration gas leakage detection method based on adaptive digital phase-locked amplification algorithm provided by the present application;

[0026] Figure 3 is a three-dimensional mapping distribution diagram of leakage state classification based on image diffusion rate and pressure second-order derivative in the low-concentration gas leakage detection method based on adaptive digital phase-locked amplification algorithm provided by the present application;

[0027] Figure 4 is a power spectral density comparison diagram before and after multi-order adaptive notch filtering under strong power frequency interference in the low-concentration gas leakage detection method based on adaptive digital phase-locked amplification algorithm provided by the present application;

[0028] Figure 5 is a time-frequency spectrogram of the spectrum conflict monitoring and active carrier migration process in the low-concentration gas leakage detection method based on the adaptive digital phase-locked amplification algorithm provided by the application;

[0029] Figure 6 is a theoretical isochoric pressure, actual measured pressure and differential residual error comparison graph under thermal shock working conditions in the low-concentration gas leakage detection method based on the adaptive digital phase-locked amplification algorithm provided by the application;

[0030] Figure 7 is an implementation execution schematic diagram of the low-concentration gas leakage detection system based on the adaptive digital phase-locked amplification algorithm provided by the application;

[0031] Figure 8 is a structural schematic diagram of the electronic device provided by the application. DETAILED DESCRIPTION

[0032] The embodiments of the application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the application, and cannot be understood as a limitation of the application.

[0033] The low-concentration gas leakage detection method, system and electronic device based on the adaptive digital phase-locked amplification algorithm of the embodiments of the application are described below with reference to the accompanying drawings.

[0034] Embodiment one

[0035] The embodiment provides a low-concentration gas leakage detection method based on an adaptive digital phase-locked amplification algorithm. The method is mainly applied to the precise monitoring scene of trace gas, especially ppb-level low-concentration gas leakage. The embodiment solves the contradiction between sensitivity and response speed of the prior art and the missed detection problem under environmental thermal shock by constructing a multi-source perception and double-loop control system.

[0036] Referring to the flow schematic diagram shown in Figure 1 , the method of the embodiment mainly includes the following steps:

[0037] S1: Multi-source data synchronous acquisition and preprocessing.

[0038] The core of this step is to establish a high-precision, time axis strictly aligned multi-dimensional perception environment, to provide a reliable data basis for subsequent algorithm processing.

[0039] For example, the system first performs a multi-source data synchronous acquisition operation, which is not a simple parallel reading, but is based on a unified hardware clock trigger mechanism. The system is configured with a high-sensitivity gas sensor, a pipeline infrared image acquisition module, and a pressure and temperature sensor deployed in the gas meter box or pipeline cavity.

[0040] The gas sensor is used to acquire the original detection electrical signal containing gas leakage information, which is usually submerged in environmental noise and appears as a weak nanovolt or microvolt voltage change. In order to capture the transient leakage signal, the sampling frequency of the gas sensor is set to a value sufficient to cover the target modulation frequency, for example, between 1 kHz and 10 kHz.

[0041] The pipeline infrared image acquisition module is used to obtain the pipeline infrared image covering the monitoring area. The module is configured to be able to sense the thermal radiation difference or infrared absorption characteristics of the gas, so as to present the diffusion pattern of the gas cloud on the image. The acquisition frame rate of the image and the acquisition period of the electrical signal are strictly aligned by hardware timestamps, ensuring that each frame of image corresponds to a specific electrical signal data, and the timestamp alignment error is controlled within a very small range, for example, less than or equal to 1 millisecond, to eliminate the spatio-temporal misalignment between multi-source data.

[0042] The measured pressure value and the measured temperature value in the cavity are synchronously acquired by a high-precision pressure and temperature sensor. Since the changes in pressure and temperature are relatively slow compared to the electrical signal, but they are crucial for determining the leakage, the system ensures that the absolute pressure value and the ambient temperature value at the current time are recorded while the electrical signal is being acquired.

[0043] Optionally, after obtaining the above-mentioned original data, the system pre-processes the original detection electrical signal, including removing the DC offset component and filtering out high-frequency clutter through a low-pass filter to retain the baseband signal and modulation signal components. For the pipeline infrared image, the system performs grayscale conversion and Gaussian filtering to enhance image contrast and suppress speckle noise. For the measured pressure value and the measured temperature value, a sliding average filtering algorithm is used to smooth random fluctuations and eliminate outliers caused by sensor electronic noise.

[0044] S2: Integral time adaptive control process.

[0045] This step aims to solve the technical drawbacks caused by fixed integral time. The system executes an integral time adaptive control process, which is not a single parameter adjustment, but a closed-loop control system. The system generates an integral time control instruction based on the pipeline infrared image and the measured pressure value, and uses the integral time control instruction to perform adaptive digital lock-in amplification processing on the original detection electrical signal, thereby extracting the target demodulation signal.

[0046] The integral time adaptive control flow is logically divided into two closely coupled stages: a prediction control stage and a feedback correction stage.

[0047] 1. Prediction control stage:

[0048] The role of the prediction control stage is to feed forward, that is, before signal demodulation, to pre-judge the possibility and type of leakage occurrence according to macro-physical changes in the external environment, namely images and pressure, so as to coarsely adjust the integral time.

[0049] For example, in the prediction control stage, the system determines a concentration change trend label in response to the feature change rate of the pipeline infrared image and the second derivative feature of the measured pressure value. Subsequently, the system sets the initial integral time of the adaptive digital lock-in amplification processing according to the concentration change trend label.

[0050] Specifically, the process of determining the concentration change trend label includes:

[0051] First, the system calculates the area change rate of the leakage region in the pipeline infrared image. This process involves differential processing or optical flow analysis of consecutive frames of infrared images, extracting the pixel area of the suspected gas cloud region, and calculating the gradient of the area change over time. The area change rate reflects the speed of gas diffusion.

[0052] At the same time, the system calculates the pressure second derivative value of the measured pressure value over time. The first derivative of pressure represents the speed of pressure change, and the second derivative of pressure value represents the acceleration of pressure change. The purpose of introducing the second derivative is to discover the nonlinear mutation trend of pressure earlier, because in the instant of leakage occurrence, the pressure drop often presents an acceleration feature.

[0053] As Figure 2 the step response simulation comparison chart, the horizontal axis represents the sampling time process, and the vertical axis represents the normalized amplitude of the demodulation signal after digital lock-in amplification processing. The chart shows the dynamic changes of the system output waveform in the simulation of the instant of simulated gas leakage burst.

[0054] Figure 2The blue dashed line represents the output curve of the traditional fixed long integration time strategy. It can be seen that the rising edge of the curve is very flat at the beginning of the leakage, showing obvious hysteresis characteristics. This is because the fixed long integration time smooths the effective abrupt signal in order to suppress noise. In contrast, the red solid line represents the output curve after applying the integral time adaptive control process of the present application. The curve shows an almost vertical steep rising edge at the initial moment of the leakage, which corresponds to the effect of the system rapidly reducing the initial integration time from fifty milliseconds in the stable period to five milliseconds after detecting that the rate of change of the pipeline infrared image feature and the second derivative of the measured pressure value exceed the threshold and generating an abrupt label, thereby achieving millisecond-level fast tracking of the transient leakage signal.

[0055] Immediately after the rising edge, the red solid line quickly converges and becomes smooth after reaching the steady-state value, with a waveform jitter amplitude comparable to or even smaller than the blue dashed line. This corresponds to the process of gradually increasing the integration time until it reaches the maximum value to maximize the suppression of environmental noise in the feedback correction stage after monitoring that the signal tends to be stable.

[0056] Through the significant comparison of the two curves, it is intuitively proved that the present method can significantly improve the dynamic response speed to trace the trace gas leakage without sacrificing the steady-state signal-to-noise ratio, effectively solving the technical problem that sensitivity and response speed are difficult to balance.

[0057] Based on the above two physical quantities, the system divides the current working condition into three states and generates corresponding labels:

[0058] In the first case, if the area change rate is greater than or equal to the first preset rate threshold, and the pressure second derivative value is less than the negative first preset pressure change threshold, the system determines that it is currently in the explosive leakage period, and generates an abrupt label. At this time, in order to quickly track this transient change, the system sets an initial integration time of a first duration. The first duration is a short time value, which aims to improve the dynamic response ability of the system and ensure that the rapidly changing signal peak is not missed.

[0059] In the second case, if the area change rate is less than the first preset rate threshold and greater than or equal to the second preset rate threshold, the system determines that it is currently in the slow leakage or diffusion initial stage, and generates a slow change label. At this time, the system sets an initial integration time of a second duration. The second duration is between short integration and long integration, which is a compromise between ensuring a certain response speed and providing a certain noise suppression ability.

[0060] In the third case, if the area change rate is less than the second preset rate threshold, and the absolute value of the pressure second derivative value is less than or equal to the first preset pressure change threshold, the system determines that the current environment is relatively static or in a stable diffusion period of a small leakage, and generates a smooth label. At this time, the signal-to-noise ratio becomes the primary contradiction, and the system correspondingly sets an initial integration time of a third time length. The third time length is a relatively long time value, and through long-time integration and accumulation, random noise is suppressed to the greatest extent, and weak signals submerged in noise are extracted.

[0061] In this logic, the first time length is less than the second time length, and the second time length is less than the third time length, thereby constructing an adaptive strategy of short time for mutation, medium time for slow change, and long time for smoothness.

[0062] 2. Feedback correction stage:

[0063] The function of the feedback correction stage is fine tuning, that is, on the basis of the initially set initial integration time, the signal quality actually demodulated is dynamically fine tuned according to the signal quality actually demodulated, so as to reach an optimal working point.

[0064] In the feedback correction stage, the system generates a step adjustment instruction for the initial integration time in response to the signal-to-noise ratio and the frequency phase error of the output signal of the adaptive digital phase-locked amplification processing. The instruction is used to dynamically update the integration time control instruction, and the updating process is an iterative loop until the signal-to-noise ratio and the frequency phase error meet a preset convergence condition.

[0065] Specifically, the process of generating the step adjustment instruction for the initial integration time includes the following logic:

[0066] The system monitors the signal-to-noise ratio and the frequency phase error of the target demodulation signal in real time. The signal-to-noise ratio reflects the clarity of the signal under the current integration time, and the frequency phase error reflects the locking accuracy of the phase-locked amplifier to the target signal.

[0067] If the monitored signal-to-noise ratio is less than a preset signal-to-noise ratio threshold, it indicates that the current integration time is insufficient to suppress noise, and the system generates a step adjustment instruction to increase the integration time. By increasing the integration time, the equivalent noise bandwidth is reduced, thereby improving the signal-to-noise ratio.

[0068] If the monitored frequency phase error is greater than a preset error threshold, it indicates that the current integration time is too long, causing the system to fail to follow the frequency or phase fluctuations of the signal in time, that is, tailing or lock loss occurs, and the system generates a step adjustment instruction to reduce the integration time. By reducing the integration time, the tracking bandwidth of the loop is improved, thereby reducing the error.

[0069] The step adjustment instruction is used to iteratively update based on the initial integration time according to a preset time adjustment step. For example, increase or decrease 1 millisecond or 5 milliseconds each time until the system finds a balance point that can ensure that the signal-to-noise ratio meets the standard and the phase error is controlled within the allowed range.

[0070] As Figure 3 The leakage state classification three-dimensional mapping distribution intuitively shows the core decision logic of the system prediction control stage, and a three-dimensional control space is constructed with the pipeline infrared image diffusion rate as the horizontal axis, the pressure second derivative in the cavity as the vertical axis, and the initial integration time of the adaptive digital lock-in amplifier as the vertical axis.

[0071] Figure 3 The three clear stepped platforms in the figure correspond to three typical gas leakage conditions. The low platform in the lower left corner of the figure represents the sudden state, which corresponds to the case that the image diffusion rate exceeds 20% and the pressure second derivative presents a significant negative value, indicating that the leakage is occurring and the pressure is accelerating. At this time, the system will map the integration time to a very low value of 5 milliseconds to ensure real-time capture of the sudden signal.

[0072] The large-area platform in the middle layer of the figure represents the slow-changing state, which covers the condition that the image diffusion rate is between 5% and 20%. At this time, the system sets the medium integration time of 20 milliseconds to balance the response speed and noise resistance.

[0073] The high platform in the upper right corner of the figure represents the stable state, which corresponds to the case that the image diffusion rate is less than 5% and the pressure change is extremely weak. At this time, the system will increase the integration time to a maximum value of 50 milliseconds to extract extremely weak leakage features from strong background noise using the long-time integration accumulation effect.

[0074] This three-dimensional mapping mechanism ensures that the system can automatically transition between different integration time steps according to the real-time change trajectory of multi-source perception data, thereby achieving adaptive and accurate matching of complex leakage scenarios.

[0075] S3: Specific execution of adaptive digital lock-in amplification processing.

[0076] After determining the integration time, the system uses the parameter to perform core demodulation processing on the original signal. The digital lock-in amplifier (DLIA) in this embodiment uses an advanced algorithm architecture to cope with complex interference environments.

[0077] For example, the specific execution process of the adaptive digital lock-in amplification processing includes three key technical links:

[0078] Link 1: The traditional phase-locked loop often uses a fixed step size, resulting in a contradiction between the locking speed and the steady-state accuracy. This embodiment uses a variable step size frequency update strategy to dynamically adjust the frequency update step size according to the frequency error of the input signal. The system calculates the frequency deviation between the reference signal and the input signal in real time.

[0079] When the frequency error is greater than the preset decision value, it indicates that the system has not been locked or the signal frequency has undergone a large range jump, at which time the system uses a first frequency step size. The first frequency step size is set to a larger value, so that the local oscillator can quickly approach the target frequency to achieve fast capture.

[0080] When the frequency error is less than or equal to the preset decision value, it indicates that the system has approached the locked state, at which time the system uses a second frequency step size. The second frequency step size is set to a smaller value for fine adjustment near the target frequency, reducing steady-state jitter and improving locking accuracy.

[0081] Wherein, the first frequency step size is greater than the second frequency step size, and this coarse-to-fine strategy ensures that the algorithm has both speed and accuracy.

[0082] Link 2: In weak signal detection, the signal is often accompanied by random phase drift or distortion. This embodiment uses a Kalman filter phase estimation strategy to compensate for the phase mismatch caused by modulation waveform distortion. The system establishes a state equation, taking the phase and frequency of the signal as state variables and the quadrature demodulated components as observation variables. Through the prediction and update steps of Kalman filtering, the best estimate value at the last time is used to predict the phase at the current time, and the predicted value is corrected according to the current observation value. This method can effectively estimate the true phase information from the signal containing noise, prevent demodulation amplitude attenuation caused by phase jitter, and ensure that the amplitude of the output signal truly reflects the gas concentration.

[0083] Link 3: There are a large number of power frequency interference in industrial environments, such as 50 Hz and its harmonics. This embodiment uses a multi-stage adaptive notch strategy to dynamically lock and suppress the power frequency interference and harmonic interference in the environment based on the noise spectrum analysis results of the fast Fourier transform. The system first performs fast Fourier transform on the input signal to analyze the noise spectrum distribution and identify the strongest interference frequency point. Then, the system adaptively adjusts the center frequency of the notch filter to accurately align the interference frequency, and configures multiple cascaded notch filters to simultaneously suppress the fundamental wave and harmonics. The signal after notch processing enters the phase-sensitive detector, and finally outputs the pure target demodulation signal.

[0084] As Figure 4The power spectrum density contrast chart intuitively demonstrates the system's ability to extract weak signals in a strong power frequency interference environment. The horizontal axis represents the signal frequency, and the vertical axis represents the signal energy intensity.

[0085] Figure 4 The middle red dotted line represents the original detection electric signal spectrum without processing. It can be clearly observed that there are extremely high energy peaks at fifty hertz and its third harmonic one hundred fifty hertz, which corresponds to the strong electromagnetic interference commonly found in industrial sites. The amplitude of these interference signals is much higher than that of the target modulation signal near two hundred fifteen hertz, severely masking the effective gas leakage characteristics.

[0086] The blue solid line represents the output signal spectrum after processing by the multi-stage adaptive notch strategy of the present application. By comparison, it can be seen that the interference peaks at fifty hertz and one hundred fifty hertz are significantly reduced, with an energy attenuation degree of several tenths of a decibel, forming a clear deep valley, proving the precise locking and suppression ability of the notch filter at specific frequencies.

[0087] At the same time, the energy amplitude of the spectral characteristics in other frequency bands, especially the target modulation signal, remains highly consistent before and after processing, without significant attenuation or distortion. This result powerfully proves that the algorithm described in the present application can dynamically adjust the notch parameters based on the noise spectrum analysis results of the fast Fourier transform, completely filtering out power frequency and harmonic interference while preserving the weak signals carrying gas concentration information, thereby ensuring the detection signal-to-noise ratio and measurement accuracy of the system in a high noise background.

[0088] S4: Multi-source data fusion and leakage judgment.

[0089] After extracting high-quality target demodulation signals, the system enters the decision-making stage. Since a single data source may have false positives, the present embodiment uses a multi-source data fusion strategy.

[0090] For example, the system determines the gas leakage detection result based on the target demodulation signal and in combination with the multi-source data fusion strategy. This process includes three sub-steps: scoring, sorting, and weighting.

[0091] Sub-step 1: Credibility scoring. The system calculates the data credibility score of the original detection electric signal, the pipeline infrared image, and the measured pressure value, respectively. The scoring criteria are as follows:

[0092] For the original detection electric signal, the system divides it into three credibility levels based on its signal-to-noise ratio value: high, medium, and low. For example, when the signal-to-noise ratio is greater than a certain high threshold, it is considered high credibility; between two thresholds, it is considered medium credibility; and below a low threshold, it is considered low credibility.

[0093] For the pipeline infrared image, the system divides the high, medium and low three levels of confidence level based on the intersection-over-union value of the leakage region segmentation. The intersection-over-union reflects the coincidence degree of the suspected leakage region extracted by the algorithm and the theoretical or historical leakage model. The higher the coincidence degree, the higher the confidence of the image data.

[0094] For the measured pressure value, the system divides the high, medium and low three levels of confidence level based on the fluctuation value of the change rate of the continuous sampling points. If the pressure change curve is smooth and conforms to the physical law, the confidence is high; if there is chaotic and violent jumping, the confidence is low.

[0095] The first and second classification thresholds are used to distinguish between the confidence levels, ensuring the objectivity of the score.

[0096] Substep 2: priority sorting and conflict resolution. The system sorts the multi-source data according to the data confidence score. When making the final decision, if there is a conflict in the decision result, for example, the electric signal shows leakage, but the image shows no leakage, the decision is made according to the dominance of high-priority data. This means that the electric signal with high signal-to-noise ratio or the image with high definition will have a veto or a pass, thereby avoiding the interference of low-quality data on the decision.

[0097] Substep 3: phased weighting. If the data confidence scores of different sources are at the same level, for example, both are medium confidence, the system cannot simply vote, but according to the leakage concentration stage, which is designed based on the sensitivity difference of different sensors at different stages.

[0098] In the early stage of leakage, the gas concentration is extremely low and the pressure change is not obvious, but the electric signal is extremely sensitive to concentration change, and the infrared image may capture weak thermal plume. Therefore, at this time, the original detection electric signal and the pipeline infrared image are the dominant decision, and they are given higher weight.

[0099] In the stable stage of leakage, the gas concentration reaches a certain level, and the pressure in the pipeline begins to drop substantially. At this time, the original detection electric signal and the measured pressure value are the dominant decision, and the continuous decline of the pressure is used as the confirmation of the leakage to prevent false positives due to image background interference.

[0100] S5: pressure threshold decision logic.

[0101] In order to solve the heat masking effect, that is, the environmental warming causes the gas to expand, thereby masking the pressure drop caused by leakage, the embodiment introduces a pressure threshold decision logic containing a physical model correction in the process of determining the gas leakage detection result.

[0102] For example, the pressure threshold decision logic includes the following steps:

[0103] Firstly, the system calculates the theoretical predicted pressure value under the no-leakage working condition based on the ideal gas state equation, using the measured temperature value at the current time and the initial pressure value and the initial temperature value under the no-leakage working condition:

[0104] The ideal gas state equation is Under the assumption of no leakage, the volume and the amount of gas are constant, the pressure is proportional to the temperature . Here, the initial time pressure value under the no-leakage working condition is defined as , the initial time temperature value is (unit: Kelvin), and the measured temperature value at the current time is (unit: Kelvin). Under the assumption of airtightness and no leakage, the theoretical predicted pressure value at the current time can be expressed as:

[0105] ;

[0106] Next, the system calculates the product term of the theoretical predicted pressure value and the preset pressure residual statistical coefficient, and subtracts the product term from the theoretical predicted pressure value to obtain the dynamic pressure threshold.

[0107] The preset pressure residual statistical coefficient reflects the random fluctuation level of the system under normal operation, such as 3 times the standard deviation. Therefore, the calculation formula of the dynamic pressure threshold is as follows, where is the confidence coefficient:

[0108] ;

[0109] The above formula indicates that if the current actual measured pressure value is lower than the theoretical pressure value considering the temperature rise by a statistically significant amount, i.e. , then this low pressure can only be caused by gas leakage.

[0110] Finally, the system performs a judgment: when the concentration value reversed from the original detection electric signal reaches the preset detection lower limit, and the measured pressure values of multiple consecutive sampling points are less than the dynamic pressure threshold and show a numerical downward trend, it is determined that there is a gas leakage confirmation. This logic effectively eliminates the missed detection caused by the pressure gauge reading rising instead of falling due to temperature rise, and realizes accurate detection in a thermal shock environment.

[0111] S6: Collaborative optimization of edge and cloud.

[0112] To achieve the continuous evolution of the algorithm and adaptation to new scenarios, the embodiment also includes a collaborative optimization step between the edge and the cloud.

[0113] By way of example, at the edge terminal side, i.e. locally at the detection device, the system extracts a multi-dimensional scene feature vector during the detection process. The multi-dimensional scene feature vector is a digital fingerprint that describes the current working environment, including hardware parameter features (such as sensor model, pipe diameter), environmental features (such as light intensity, temperature and humidity background), and signal features (such as noise floor, frequency spectrum distribution).

[0114] In response to detecting a new scene identifier, for example, the feature vector has a low matching degree with the historical library or a certain leakage event occurs, the edge terminal triggers a data upload process to upload the reduced dimension key data to the cloud server. Here, all raw data is not uploaded in order to save bandwidth.

[0115] The cloud server uses the uploaded data to build a comprehensive data set that covers various complex working conditions. The cloud uses a meta-learning (Meta-Learning) architecture to incrementally train a lightweight meta-learning model. During training, the system updates only the scene adapter parameters, i.e. the classification layer or the regression layer for a specific scene, while keeping the base parameters of the fixed model (universal feature extraction layer) unchanged.

[0116] After training is complete, the cloud distributes the optimized parameters to the edge terminal for parameter hot update. The edge terminal dynamically loads new parameters without stopping the detection service, so that the device becomes smarter and can adapt to changing monitoring environments.

[0117] Embodiment Two

[0118] This embodiment further describes the preferred implementation of the anti-interference mechanism based on embodiment one. In particular, for the working conditions where the electromagnetic environment in the industrial site is complex and the interference frequency is unstable, this embodiment describes in detail the implementation details of a spectrum conflict monitoring and active carrier migration process.

[0119] To ensure that the system can still maintain high detection sensitivity in a strong interference environment, the method also includes performing a spectrum conflict monitoring and active carrier migration process. The core logic of this process is to change the traditional passive filtering approach and instead adopt an active avoidance strategy. When the interference frequency in the environment drifts and approaches the working frequency of the system, instead of forcibly increasing the filtering depth, the system intelligently migrates its working frequency to a clean frequency spectrum region, thereby fundamentally eliminating the risk of signal suppression caused by spectral aliasing.

[0120] For example, the first stage of the flow is interference identification and state awareness. The system identifies the current dominant interference frequency in real time based on the noise spectrum analysis result of the fast Fourier transform. In a specific implementation, this process is performed by a dedicated spectrum analysis module inside the FPGA. After digitizing the original detection signal, the module does not directly enter the phase-locked amplification link, but instead shunts a data stream for windowing processing. In order to reduce the impact of spectral leakage on frequency estimation accuracy, the system usually uses a Hanning window or a Blackman window to truncate the time-domain signal. Subsequently, the system performs a fast Fourier transform operation with a point number of not less than 1024 points to obtain the noise power spectrum density distribution of the full frequency band.

[0121] After obtaining the noise spectrum analysis result, the system starts the peak search algorithm. This algorithm searches for the spectral peak with the largest amplitude in the preset frequency band of interest, such as the power frequency of 50 Hz or 60 Hz and its odd harmonic near the frequency. The frequency point corresponding to this spectral peak is defined as the dominant interference frequency, denoted as .

[0122] It should be noted that, considering the volatility of the power grid frequency, this dominant interference frequency is not a fixed value, but a variable that changes dynamically over time. At the same time, the system reads the current target carrier frequency set for driving the gas sensor light source, denoted as . This target carrier frequency is the frequency of the driving signal currently used by the system to modulate the laser or infrared light source, and is also the reference benchmark for demodulation by the digital phase-locked amplifier.

[0123] For example, the second stage of the flow is conflict determination and threshold comparison. The system calculates the absolute difference between the dominant interference frequency and the target carrier frequency , denoted as , whose calculation logic is shown in the following formula:

[0124] ;

[0125] This absolute difference in frequency intuitively reflects the proximity of the interference signal to the target signal in the frequency domain. In order to quantitatively evaluate whether this proximity poses a threat to signal extraction, the system compares the absolute difference in frequency with a preset safety protection bandwidth threshold, denoted as .

[0126] Optionally, the safety protection bandwidth threshold greater than the stopband cutoff bandwidth in the multi-stage adaptive notch filter. The significance of this setting is that the multi-stage adaptive notch filter, although capable of filtering out the interference, has a certain transition band near the center frequency in its amplitude-frequency response curve. If the frequency of the target signal falls within the transition band or even the stopband of the notch filter, the notch filter will inevitably attenuate the target signal carrying the gas concentration information while attenuating the noise. Therefore, a wide enough safety distance, i.e., a safety protection bandwidth threshold, must be set to ensure that the system can timely warn when the frequency difference is less than the threshold. Usually, the threshold is set to 1.5 to 2 times the 3 decibel bandwidth of the notch filter to ensure that the target signal is always in the passband flat area of the notch filter.

[0127] For example, the third stage of the flow is active migration and synchronous switching. In response to the absolute difference in frequency being less than the safety protection bandwidth threshold , the system determines that it is currently in a spectrum conflict state. This state means that the interference frequency has approached the target carrier frequency, and continuing to maintain the current working frequency will cause the detection signal-to-noise ratio to deteriorate sharply. At this time, the system immediately generates a carrier frequency switching instruction.

[0128] The carrier frequency switching instruction contains two core actions: frequency selection and frequency switching. First, the system selects a new carrier frequency without conflict from the preset available frequency list through an intelligent search algorithm, denoted as . The selection criteria for the new carrier frequency must satisfy that the adjusted absolute difference in frequency, i.e., , must be greater than the safety protection bandwidth threshold , and preferably located in the region with the lowest spectrum noise floor.

[0129] After the new carrier frequency is determined, the system performs a physical layer switching operation to adjust the target carrier frequency to the new carrier frequency without conflict. This process involves direct control of the light source driving module to change the frequency of the modulation signal. At the same time, and most importantly, the system synchronously updates the reference frequency in the adaptive digital lock-in amplification process.

[0130] The time-frequency spectrogram of the spectrum conflict monitoring and active carrier migration process as shown in Figure 5 demonstrates the intelligent obstacle avoidance behavior of the system in a dynamic electromagnetic environment. The horizontal axis represents the time process, the vertical axis represents the signal frequency, and the coldness and warmth of the color represents the strength of the signal energy, where the red or yellow bright band indicates the frequency trajectory of the high-energy signal.

[0131] Figure 5A diagonal bright band in the lower left corner, representing an unstable interference signal in the environment, gradually increases in frequency over time and approaches the working frequency band of the system. The horizontal bright band adjacent to it represents the target carrier frequency set by the system initially.

[0132] As time goes on, when the frequency trajectory of the interference signal approaches the carrier frequency and causes the frequency difference between the two to be less than the preset safety protection bandwidth threshold, a significant breakpoint feature appears in the figure. At this moment, the originally horizontally extended carrier bright band suddenly breaks off and reappears in a higher frequency band in an instant, forming a new horizontal bright band. This vertical frequency jump intuitively reflects the process of the system executing the active carrier migration instruction, that is, when the risk of spectrum collision is detected, the system does not passively wait for interference to occur, but decisively controls the light source driving module to switch the working frequency to an interference-free clean interval.

[0133] This mechanism ensures that the target signal always maintains a sufficient safety frequency domain distance from the strong interference source, thereby completely eliminating the signal masking problem caused by spectral overlap and ensuring the continuity and high signal-to-noise ratio of the detection process.

[0134] Specifically, the synchronous update here requires extremely high time consistency. In the logic design of the FPGA, the sine wave generator of the light source drive (DDS module) and the reference signal generator of the digital lock-in amplifier (Ref-DDS module) must complete the frequency word (Frequency Tuning Word) jump within the same system clock cycle. If they are not synchronized, for example, the light source has already switched to a new frequency, but the demodulation end is still using the old frequency, it will cause the demodulation output to be zeroed or produce severe phase noise, seriously interfering with subsequent concentration calculation. Therefore, this embodiment adopts an atomic level switching mechanism based on a global trigger signal to ensure that the phase continuity of the light source driving signal and the demodulation reference signal is maximally maintained at the moment of frequency jump, or a fast phase recapture is performed immediately after the jump.

[0135] In order to prevent system oscillation caused by frequent frequency switching, the process also introduces a hysteresis comparison mechanism. Only when the new interference frequency stays in the conflict area for more than a certain time window, such as 500 milliseconds, does the carrier migration process trigger; or after switching to a new frequency, a lock protection time is set, during which switching is prohibited again.

[0136] Through the above steps, the embodiment constructs an anti-interference system with high self-protection ability. It is no longer passively subjected to environmental interference test, but has stress avoidance ability like living organisms. No matter how the power grid frequency fluctuates, or there is any complex frequency conversion device interference in the field, the system can always find a quiet frequency spectrum window for high-precision micro gas detection, thereby greatly improving the robustness and reliability of the equipment.

[0137] Embodiment three

[0138] Because in outdoor or industrial site applications, the detection device often faces direct sunlight, large day and night temperature difference or adjacent equipment heat radiation, etc., resulting in rapid temperature rise of the measured cavity. According to the physics Charles law, the gas pressure in the closed container will increase with the temperature rise. This physical pressure rise effect easily masks the pressure drop characteristics caused by small leakage, i.e. the so-called thermal masking effect. In order to solve this safety hazard that will lead to serious missed detection, the embodiment describes in detail a dynamic compensation logic based on equal volume differential residual.

[0139] The dynamic compensation logic based on equal volume differential residual is not a system default resident logic, but an emergency takeover mechanism triggered under certain working conditions. The system performs a thermal shock monitoring step in parallel during operation. In this step, the system uses a high-precision temperature sensor to collect the measured temperature value inside the cavity in real time, and calculates the time rate of change of the measured temperature value in real time. The time rate of change reflects the degree of temperature change, which is usually obtained by differencing or fitting the derivative of the temperature data in the continuous sliding window.

[0140] For example, a preset temperature rise rate threshold is stored in the system memory. The threshold is the watershed for determining whether the environment is in a thermal shock state, and its value is calibrated according to the thermal conductivity and volume size of the measured cavity, for example, set to 0.5 degrees Celsius per minute. When the calculated time rate of change is greater than the preset temperature rise rate threshold, the system determines that there is a severe temperature rise disturbance in the current environment. At this time, if the static pressure threshold determination logic in embodiment one or embodiment two is continued to be used, it is very likely that the real leakage signal will be masked due to the physical expansion of the pressure. Therefore, the system immediately triggers the operation of suspending the pressure threshold determination logic, and instead performs a residual analysis operation. This logic switching mechanism ensures that the system will not produce false positives due to incorrect algorithm models.

[0141] After entering the residual analysis operation, the system first needs to construct a virtual no-leakage reference model. This process is the residual calculation step. The system locks a specific sampling time window, and the starting point of the window is usually the time when the temperature starts to change rapidly. The system obtains the initial temperature value and the initial pressure value and the current measured temperature value at the current time within the window .

[0142] Optionally, to quantify the pure physical pressure increment caused by temperature rise, the system calculates the theoretical isochoric pressure increment under the assumption of airtightness and no leakage by using the ideal gas state equation. The ideal gas state equation describes the relationship between pressure, volume, temperature, and the amount of substance. Under the premise that the gas meter box or the pipeline cavity structure is fixed, the volume is considered as a constant amount; under the assumption of no leakage, the amount of substance of the gas in the cavity is considered as a constant amount. Therefore, the change of pressure is only proportional to the change of temperature.

[0143] Specifically, in the ideal isochoric process without leakage, the pressure value that should theoretically be reached at the current time should satisfy the following relationship: equals times and the ratio of. Further, the theoretical isochoric pressure increment, denoted as , is defined as the difference between the theoretical pressure value and the initial pressure value. Its calculation formula can be expressed as:

[0144] ;

[0145] The result calculated by the above formula represents how much the pressure should rise if there is no leakage at all and only because of thermal expansion and cold contraction.

[0146] At the same time when the theoretical value is calculated, the system obtains the actual pressure increment within the same sampling time window. The system reads the pressure value actually measured by the pressure sensor at the current time, denoted as , and subtracts the initial pressure value , thereby obtaining the actual pressure increment, denoted as . This value represents how much the pressure actually rises, and its calculation formula can be expressed as:

[0147] ;

[0148] The system then performs the core difference operation, subtracts the theoretical isochoric pressure increment from the actual pressure increment, and obtains the pressure difference residual at the current time, denoted as . Its calculation logic is as follows:

[0149] ;

[0150] The pressure difference residual has the meaning that it eliminates the physical effect of temperature change and only retains the pressure fluctuation component caused by the change of the amount of substance, i.e. the leakage. If the system is completely sealed and has no leakage, no matter how high the temperature rises, the actual increment should be strictly equal to the theoretical increment, and the residual should fluctuate in a small range near zero caused by measurement noise. If there is leakage, the loss of gas will cause the actual pressure to rise to follow the temperature, i.e. the actual increment is less than the theoretical increment, so that the pressure difference residual presents a negative value.

[0151] As Figure 6 The theoretical isochoric pressure, the actual measured pressure and the difference residual under the thermal shock working condition of the present application are compared in the graph, which reveals the effectiveness of the dynamic compensation logic based on the isochoric difference residual of the system in dealing with high temperature rise environment. The horizontal axis of the graph represents the time process, and the graph is divided into two related parts.

[0152] Figure 6 The upper graph of the present application directly shows the absolute change trend of the physical quantity, in which the black dotted line represents the measured temperature value in the cavity, which presents a rapid rising trend over time, simulating a typical thermal shock scene. Driven by this thermal effect, the theoretical isochoric pressure value calculated according to the ideal gas state equation, i.e. the blue dashed line, also rises significantly. The key is the actual measured pressure value represented by the red solid line, although the amount of substance of the gas is reduced due to the existence of a small amount of leakage, but due to the dominant role of thermal expansion, the curve still presents an upward trend in the macroscopic view, and if only relying on the traditional absolute pressure drop threshold, it is easy to draw the wrong conclusion that there is no leakage at this time.

[0153] Figure 6 The lower graph of the present application shows the change of the pressure difference residual after algorithm processing, i.e. the purple solid line. The curve is calculated by subtracting the actual measured pressure from the theoretical isochoric pressure, which clearly eliminates the physical pressure increase component caused by temperature rise. It can be seen that with the passage of time, the difference residual representing the pure leakage characteristics presents a clear negative cumulative trend, and at a certain moment it breaks through the preset leakage judgment threshold, i.e. the black dashed line, thereby triggering the system leakage alarm shown in the red area.

[0154] This visualization result powerfully proves that the method described in the present application can accurately separate the small leakage signal from the complex background of thermal expansion, ensuring the detection reliability in extreme environment.

[0155] For example, the instantaneous residual caused by a small amount of leakage is very small, which is easily submerged by the quantization noise or circuit thermal noise of the sensor, and the residual judgment of a single point is often unreliable. Therefore, the present embodiment introduces a cumulative judgment step. The system time integrates the pressure difference residual of a plurality of consecutive sampling periods to obtain a pressure residual cumulative value.

[0156] Specifically, the system establishes an integral buffer, which will calculate every pressure difference residual from the beginning of the thermal shock The cumulative sum is carried out. The effect of time integration is to make a little become a lot, that is, a small negative deviation will form a significant negative value after a period of accumulation, while the mean value of random noise tends to zero and will be offset during the integration process. This processing greatly improves the system's ability to identify small trend deviations in a strong noise background.

[0157] The system monitors the change of the cumulative value in real time. In response to the pressure difference residual being continuously negative and the absolute value of the pressure residual cumulative value exceeding the preset leakage residual threshold, the system determines that there is a gas leakage confirmation. The determination condition here contains two dimensional constraints:

[0158] 1. Continuous negative ensures the directionality of change, eliminating interference caused by occasional sensor jumps;

[0159] 2. The absolute value of the cumulative value exceeding the threshold ensures the significance of the change, eliminating the interference of calculation errors.

[0160] The preset leakage residual threshold is calibrated according to a large amount of experimental data, which represents the minimum leakage cumulative amount that the system can distinguish under a certain confidence level.

[0161] Optionally, once the above determination condition is met, the system will not only issue an alarm locally, but also generate a detailed detection report. In order to facilitate the back-end operation and maintenance personnel or the cloud monitoring platform to distinguish the type of this alarm, the system outputs the detection result containing the thermal masking correction identifier. For example, this identifier explicitly tells the user that this is a leakage event that occurred during a period of temperature change, and the system has removed the influence of thermal expansion through algorithm. This has important reference value for subsequent accident review and maintenance, because it means that the leakage point may be very hidden and can only be found after removing thermal interference.

[0162] The embodiment effectively solves the technical problem that the traditional detection method mistakenly considers no leakage due to pressure rise under environmental thermal shock. It uses a combination of physical models and mathematical statistics to accurately separate the microscopic leakage characteristics from the macroscopic thermal expansion background, achieving high-reliability detection in all-weather and all-working conditions, and greatly expanding the application boundary and safety protection capability of the technology.

[0163] Embodiment Four

[0164] The embodiment provides a low-concentration gas leakage detection system based on an adaptive digital phase-locked amplification algorithm. The system is a hardware and software architecture carrier for implementing the detection methods in embodiments one to three, and aims to solve the technical problems such as low signal-to-noise ratio, poor environmental adaptability and missed detection caused by thermal masking effect in micro gas detection through modular collaborative design.

[0165] Referring to Figure 7 The detection system described in the embodiment is logically divided into four core functional modules, namely a multi-source perception module, a double-loop control and signal processing module, a fusion decision module and a collaborative communication module, according to the system architecture diagram shown in the figure. The modules are connected through a high-speed data bus to realize real-time data flow and synchronous issuance of instructions.

[0166] By way of example, the multi-source perception module is the perception front end of the entire system, mainly used for performing multi-source data synchronous acquisition tasks. The module integrates a variety of heterogeneous sensors, including a high-sensitivity gas sensor for detecting weak leakage signals, an infrared imaging component for capturing gas diffusion patterns, and a high-precision pressure sensor and temperature sensor deployed inside a sealed cavity. In order to overcome the data space dislocation problem mentioned in the background art, the multi-source perception module is built-in with a hardware-level time synchronization unit to ensure that all data frames are stamped with a uniform nanosecond-level timestamp when obtaining the original detection electrical signals containing gas leakage information, pipeline infrared images and measured pressure and temperature values inside the cavity. This strict synchronization mechanism provides a reliable timing reference for subsequent signal fusion, enabling the system to accurately associate electrical signal fluctuations and physical environment changes at the same time.

[0167] By way of example, the double-loop control and signal processing module is the operation core of the system, which carries the most critical integral time adaptive control process and digital signal processing algorithm of the invention. The module is further subdivided into a control unit and a processing unit in terms of function.

[0168] At the control level, the module is used to execute the integral time adaptive control process, which breaks the game between the sensitivity and response speed of traditional devices through the closed-loop cooperation of the prediction control stage and the feedback correction stage. Specifically, in the prediction control stage, the logic unit inside the module will analyze the feature change rate of the pipeline infrared image and the second derivative feature of the measured pressure value in real time, and determine whether the current is in the leakage burst period or the stable diffusion period, so as to set the initial integral time based on the pipeline infrared image and the measured pressure value. Subsequently, in the feedback correction stage, the module will continuously monitor the signal quality output by the signal processing chain, and based on the output signal quality, i.e. signal-to-noise ratio and phase error, generate a step adjustment instruction to perform step adjustment of the integral time.

[0169] At the signal processing level, the module performs adaptive digital lock-in amplification processing on the original detection electrical signal using the final determined integration time to extract the target demodulation signal. In order to deal with the spectrum conflict problem mentioned in embodiment two, the module also integrates spectrum analysis and notch logic. When the environmental interference frequency drift is monitored and approaches the target carrier frequency, the module triggers the active carrier migration logic and synchronously updates the demodulation reference frequency. In addition, the module also runs a variable step size frequency update algorithm, a Kalman filter phase estimation algorithm, and a multi-order adaptive notch algorithm inside, ensuring that in a complex electromagnetic interference environment, the target demodulation signal with high signal-to-noise ratio and accurate phase can still be output.

[0170] By way of example, the fusion decision module is the intelligent hub of the system, mainly used to execute the final leakage judgment logic. Instead of simply relying on a single threshold alarm, the module is used to determine the gas leakage detection result based on the target demodulation signal, combined with the data reliability score of multi-source data and conflict coordination strategy. The module maintains a set of multi-dimensional reliability evaluation model inside, which can score the quality of electrical signal, image data and pressure data respectively, and arbitrate according to high priority data when data conflict occurs.

[0171] In particular, for the thermal masking effect described in detail in embodiment three, the fusion decision module embeds dynamic compensation logic based on equal tolerance differential residual. When the module monitors a sharp rise in measured temperature value, it will automatically take over the regular pressure threshold judgment logic and calculate the differential residual between the actual pressure increment and the theoretical isochoric pressure increment. By accumulating the analysis of the residual, the module can accurately identify the pressure loss component caused by micro-leakage from the background of pressure rise caused by thermal expansion, thereby outputting a detection result containing a thermal masking correction identifier, completely solving the problem of missed detection in high temperature rise environment.

[0172] By way of example, the cooperative communication module is a bridge for the system to interact with the external cloud platform, mainly used to realize the continuous evolution of the algorithm model. The module is not only responsible for transmitting alarm information, but more importantly, it is used to extract multi-dimensional scene feature vectors. These feature vectors cover hardware parameters, environmental features and signal features, forming the digital fingerprint of the current monitoring scene. When encountering a new scene or complex leakage event, the cooperative communication module will trigger the data upload process and perform model parameter incremental update interaction with the cloud server. By receiving the optimized parameters generated by the cloud based on meta-learning training, the module can perform hot updates on the local algorithm model, so that the detection system can adapt to new interference environments and leakage characteristics as time goes on, realizing self-iteration and improvement of device performance.

[0173] In summary, the low-concentration gas leakage detection system provided by the embodiment has the advantages that through the close cooperation of the four modules, multi-source sensing, double-loop control, intelligent decision-making and cloud-edge collaborative deep fusion are achieved.

[0174] Embodiment five

[0175] Corresponding to the above-mentioned embodiments, the application further provides an electronic device.

[0176] As Figure 8 The structure of the electronic device is shown in the figure, and the electronic device 100 comprises a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, through a bus 102. Optionally, the electronic device 100 can further comprise a transceiver 104. It should be noted that the transceiver 104 is not limited to one in actual application, and the structure of the electronic device 100 does not constitute a limitation on the embodiments of the application.

[0177] The processor 101 can be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can realize or execute various exemplary logical blocks, modules and circuits described in combination with the disclosure. The processor 101 can also be a combination of computing functions, such as one or more microprocessor combinations, combinations of DSP and microprocessor, etc.

[0178] The bus 102 can comprise a channel for transmitting information between the above-mentioned components. The bus 102 can be a PCI bus or an EISA bus, etc. The bus 102 can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 8 Only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0179] The memory 103 is used for storing a computer program corresponding to the low-concentration gas leakage detection method based on the adaptive digital phase-locked amplification algorithm of the above-mentioned embodiments of the application, and the computer program is executed by the processor 101. The processor 101 is used for executing the computer program stored in the memory 103 to realize the content shown in the foregoing method embodiments.

[0180] The electronic device 100 includes but is not limited to mobile terminals such as notebook computers, PADs (tablet computers) and the like, and fixed terminals such as desktop computers and the like. Figure 8The electronic device 100 shown is merely an example and should not impose any limitation on the functions and application scope of the embodiments of the present application.

[0181] Although the embodiments of the present application have been shown and described above, it should be understood that the above-described embodiments are exemplary, and should not be construed as limiting the present application, and those ordinarily skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments within the scope of the present application.

Claims

1. A method for detecting low-concentration gas leaks based on an adaptive digital lock-in amplification algorithm, characterized in that, Includes the following steps: Perform multi-source data synchronous acquisition to obtain raw detection electrical signals containing gas leak information, pipeline infrared images, and measured pressure and temperature values ​​inside the cavity; The integral time adaptive control process is executed, an integral time control command is generated based on the pipeline infrared image and the measured pressure value, and the original detection electrical signal is subjected to adaptive digital phase-locked amplification processing using the integral time control command to extract the target demodulated signal; Based on the target demodulated signal, and combined with a multi-source data fusion strategy, the gas leak detection result is determined; The integral time adaptive control process includes: In response to the characteristic rate of change of the pipeline infrared image and the second derivative characteristics of the measured pressure value, a concentration change trend label is determined, and the initial integration time of the adaptive digital lock-in amplification process is set according to the concentration change trend label. In response to the signal-to-noise ratio and frequency phase error of the output of the adaptive digital lock-in amplifier, a step adjustment command for the initial integration time is generated to dynamically update the integration time control command until the signal-to-noise ratio and the frequency phase error meet the preset convergence condition.

2. The method according to claim 1, characterized in that, The process of determining the concentration change trend label includes: Calculate the rate of change of the area of ​​the leak region in the infrared image of the pipeline, and the second derivative of the measured pressure value with time; If the area change rate is greater than or equal to the first preset rate threshold, and the second derivative of the pressure is less than the negative first preset pressure change threshold, then a mutation label is generated, and the initial integration time of the first duration is set accordingly. If the area change rate is less than the first preset rate threshold and greater than or equal to the second preset rate threshold, a slow change label is generated, and the initial integration time of the second duration is set accordingly. If the rate of change of the area is less than the second preset rate threshold, and the absolute value of the second derivative of the pressure is less than or equal to the first preset pressure change threshold, then a stable label is generated, and the initial integration time of the third duration is set accordingly. Wherein, the first duration is less than the second duration, and the second duration is less than the third duration.

3. The method according to claim 1, characterized in that, The process of generating step adjustment instructions for the initial integration time includes: Real-time monitoring of the signal-to-noise ratio and frequency phase error of the target demodulated signal; If the signal-to-noise ratio is less than a preset signal-to-noise ratio threshold, a step adjustment instruction to increase the integration time is generated; If the frequency phase error is greater than a preset error threshold, a step adjustment command to reduce the integration time is generated. The step adjustment command is used to iteratively update the time step based on the initial integration time, according to a preset time adjustment step size.

4. The method according to claim 1, characterized in that, The specific execution process of the adaptive digital phase-locked amplification process includes: Using a variable step size frequency update strategy, the frequency update step size is dynamically adjusted according to the frequency error of the input signal. When the frequency error is greater than a preset judgment value, a first frequency step size is used, and when the frequency error is less than or equal to the preset judgment value, a second frequency step size is used, wherein the first frequency step size is greater than the second frequency step size. A Kalman filter phase estimation strategy is used to compensate for the phase mismatch caused by modulation waveform distortion. By employing a multi-order adaptive notch filtering strategy and based on the noise spectrum analysis results of the fast Fourier transform, the system dynamically locks onto and suppresses power frequency interference and harmonic interference in the environment, and outputs the target demodulated signal.

5. The method according to claim 1, characterized in that, The determination of the gas leak detection result based on the target demodulated signal and in conjunction with a multi-source data fusion strategy includes: Calculate the data reliability scores for the original detection electrical signal, the pipeline infrared image, and the measured pressure value, respectively. The data from multiple sources are prioritized based on the data credibility score. If there is a conflict in the judgment results, the decision is made based on the dominant position of the higher-priority data. If the credibility scores of data from different sources are at the same level, then weighting is applied based on the leakage concentration stage: In the initial stage of leakage, the original detection electrical signal and the pipeline infrared image are the primary indicators for judgment. During the stabilization phase of the leak, the original detection electrical signal and the measured pressure value are used as the primary criteria for judgment.

6. The method according to claim 5, characterized in that, The evaluation criteria for the data credibility score include: The original detected electrical signal is classified into three confidence levels: high, medium, and low, based on its signal-to-noise ratio. For the infrared images of the pipeline, the cross-union ratio (CUI) values ​​based on the segmentation of the leakage area are divided into three confidence levels: high, medium, and low. For the measured pressure values, three confidence levels—high, medium, and low—are determined based on the rate of change fluctuation values ​​of continuous sampling points. The various credibility levels are defined by a preset first grading threshold and a second grading threshold.

7. The method according to claim 1, characterized in that, The process of determining the gas leak detection result includes pressure threshold determination logic, which includes: Based on the ideal gas law, using the measured temperature value at the current moment and the initial pressure and initial temperature values ​​under leak-free conditions, the theoretical predicted pressure value under leak-free conditions is calculated. Calculate the product term of the theoretically predicted pressure value and the preset pressure residual statistical coefficient, and subtract the product term from the theoretically predicted pressure value to obtain the dynamic pressure threshold; When the concentration value retrieved from the original detection electrical signal reaches the preset detection lower limit, and the measured pressure values ​​at multiple consecutive sampling points are less than the dynamic pressure threshold and show a decreasing trend, a gas leak is confirmed.

8. The method according to claim 1, characterized in that, It also includes collaborative optimization steps between the edge and the cloud: A multi-dimensional scene feature vector is extracted during the detection process at the edge terminal side. The multi-dimensional scene feature vector includes hardware parameter features, environmental features, and signal features. In response to the detection of a new scene identifier or a leakage event, a data upload process is triggered to upload the reduced key data to the cloud server; The cloud server uses the uploaded data to construct a comprehensive dataset, performs incremental training on the lightweight meta-learning model, updates the scene adapter parameters while keeping the basic parameters fixed, and distributes the optimized parameters to the edge terminal for hot parameter updates.

9. The method according to claim 4, characterized in that, The method further includes performing a spectrum conflict detection and active carrier migration process, the process comprising: Based on the noise spectrum analysis results of the fast Fourier transform, the current dominant interference frequency is identified in real time, and the target carrier frequency currently set by the system for driving the gas sensor light source is read. The absolute frequency difference between the interference-dominant frequency and the target carrier frequency is calculated in real time, and the absolute frequency difference is compared with a preset security protection bandwidth threshold, wherein the security protection bandwidth threshold is greater than the stopband cutoff bandwidth in the multi-order adaptive notch strategy. In response to the absolute frequency difference being less than the safety protection bandwidth threshold, a carrier frequency switching command is generated to adjust the target carrier frequency to a new, conflict-free carrier frequency, such that the adjusted absolute frequency difference is greater than the safety protection bandwidth threshold, and the reference frequency in the adaptive digital phase-locked amplification process is updated synchronously.

10. The method according to claim 7, characterized in that, The method further includes executing dynamic compensation logic based on equal tolerance residuals, the dynamic compensation logic including: The time change rate of the measured temperature value is calculated in real time. When the time change rate is greater than the preset heating rate threshold, the operation of pausing the pressure threshold determination logic is triggered, and the residual analysis operation is performed instead. Based on the initial temperature value and the current measured temperature value within the sampling time window, as well as the initial pressure value, the theoretical isochoric pressure increment under the assumption of a closed and leak-free environment is calculated using the ideal gas law; the actual pressure increment within the same sampling time window is obtained, and the theoretical isochoric pressure increment is subtracted from the actual pressure increment to obtain the pressure difference residual at the current moment; The pressure differential residuals of multiple consecutive sampling periods are integrated over time to obtain the cumulative pressure residual value. In response to the pressure differential residuals being continuously negative and the absolute value of the cumulative pressure residual value exceeding a preset leakage residual threshold, a gas leak is confirmed, and a detection result including a heat masking correction flag is output.

Images