Hydrogen leakage and combustion identification method

Abstract

The invention relates to the field of hydrogen leakage and combustion detection, and discloses a hydrogen leakage and combustion identification method, which comprises the following steps: S1, collecting multiple types of detection signals from acoustic, electrochemical and infrared thermal image sensors; s2, preprocessing the collected signals to generate a multi-dimensional feature space; s3, constructing a topological structure in the multi-dimensional feature space, extracting topological features, and generating topological descriptors; s4, analyzing the time dynamic change of the topology descriptor based on a Markov chain model; s5, calculating the topological similarity between the real-time signal and the known mode, and carrying out mode matching; and S6, calculating a signal state through a comprehensive decision function, judging whether a hydrogen leakage or combustion event exists, and triggering a corresponding alarm. According to the invention, through multi-sensor cooperative monitoring, signal preprocessing, topological structure analysis and dynamic modeling, high-precision real-time monitoring and reliable alarm of hydrogen leakage and combustion are realized, and the system has the advantages of high sensitivity, low missing report rate and rapid response.

Description

Technical Field

[0001] The invention relates to the technical field of hydrogen leakage and combustion detection, in particular to a method for identifying hydrogen leakage and combustion. Background Art

[0002] In the field of hydrogen leakage and combustion detection, existing technologies mainly rely on a single type of sensor or detection method, usually based on the physical or chemical properties of hydrogen. Most of these methods rely on point sensors to monitor local spaces and infer the hydrogen concentration of the entire area through sample detection results.

[0003] However, due to the diffusion characteristics of hydrogen and its flammable and explosive dangers, existing local detection methods have obvious limitations. In the early stage of hydrogen leakage, the leakage volume is small, and the local detection point may not be able to perceive the concentration change in time, resulting in a detection blind spot. In the case of hydrogen leakage accompanied by combustion, traditional detection equipment responds slowly to hydrogen signals in a high-temperature combustion state and cannot quickly and effectively capture the characteristics of combustion.

[0004] In addition, existing detection systems often rely on a single signal source and are easily interfered by environmental noise, sensor failure and other factors, increasing the risk of missed reports and false alarms, and failing to provide adequate protection for safe production. These technical defects make it difficult to ensure comprehensive, real-time and reliable hydrogen leakage and combustion monitoring in practical applications. Summary of the invention

[0005] In view of the deficiencies of the prior art, the present invention provides a method for identifying hydrogen leakage and combustion, which solves the problems of incomplete hydrogen leakage and combustion detection, slow response and high false alarm and missed alarm rate in the prior art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for identifying hydrogen leakage and combustion, comprising the following steps: Collect multiple types of detection signals from acoustic, electrochemical and infrared thermal imaging sensors; Preprocess the collected signals to generate a multi-dimensional feature space; Construct a topological structure in a multidimensional feature space, extract topological features, and generate a topological descriptor; Analyze the temporal dynamics of topological descriptors based on the Markov chain model; Calculate the topological similarity between real-time signals and known patterns and perform pattern matching; The signal status is calculated through a comprehensive decision function to determine whether there is a hydrogen leak or combustion event and trigger the corresponding alarm.

[0007] Preferably, in the signal acquisition step, the acquisition frequency range of the acoustic signal is 20kHz to 60kHz, the acquisition range of the electrochemical signal includes hydrogen concentration and temperature data, and the acquisition wavelength range of the infrared thermal imaging signal is 8μm-14μm and 2.8μm.

[0008] Preferably, the step of preprocessing the collected signal includes: Perform bandpass filtering on the collected signal to remove noise outside the target frequency band; Apply wavelet transform and Hilbert-Huang transform to the filtered signal to extract time-frequency features; Map the processed signal to dimensional joint feature space ,in is the total number of features extracted from the signal.

[0009] Preferably, the step of constructing a topological structure in a multidimensional feature space comprises: In multidimensional feature space In the example, based on the scale parameter Building filters ; For each data point , create a radius around it Sphere ; along with As , these spheres gradually overlap to form simplex complexes; Calculating different scales Betti numbers under ,in:

[0010] in, for dimensional boundary operator, represents the kernel space, Represents image space; By analyzing the Betti number as a function of the scale parameter The changes of ,extract the topological features of the signal; Generates continuous bar graphs recording the birth and death scales of topological features.

[0011] Preferably, the generation of the topology descriptor includes: Mapping topological features in persistent bar graphs to multidimensional descriptor space ,in is the dimension of the descriptor space; Each topological descriptor vector Represents the persistence and scale information of a specific topological feature and is defined as:

[0012] in, and Respectively The birth and death scales of topological features, is the corresponding Betti number.

[0013] Preferably, the step of analyzing the temporal dynamic changes of the topological descriptor based on the Markov chain model comprises: Constructing the state space , where each state Corresponds to a topological descriptor vector ; According to the time series data, obtain the topological descriptor sequence , the corresponding state sequence is ; Count the number of state transitions and construct the state transition probability matrix ,in:

[0014] in, From the state Transfer to state The number of times Status Total number of transfers; Verify whether the state sequence satisfies the Markov property, that is, it satisfies:

[0015] For any historical state sequence, ensure that the current state only depends on the state at the previous moment; Solving for the Steady-State Distribution of a Markov Chain ,satisfy:

[0016] Analyze the properties of the state transition matrix through eigenvalue decomposition to determine the change pattern of the topological descriptor; Based on the changes in transition probabilities and deviations from steady-state distributions, anomalous topological descriptor changes are identified as indicators of possible hydrogen leaks or combustion events.

[0017] Preferably, the step of calculating the topological similarity between the real-time signal and the known pattern and performing pattern matching comprises: The real-time topology descriptor set and a set of known pattern topology descriptors Represented as a metric space, define the corresponding metric function and ; Establish and The corresponding relationship between , so that for Each element in ,exist Elements in Make ; calculate and The Gromov-Hausdorff distance between , defined as:

[0018] in, Represents taking all possible correspondences The minimum value in Indicated in Take the maximum value; Metric Function and Defined as the Euclidean distance between topological descriptors:

[0019] in, is the Euclidean norm; Using the calculated Gromov-Hausdorff distance Evaluate the similarity between the real-time signal and the known pattern. The smaller the distance, the higher the similarity. Will Similarity threshold with the preset For comparison, if , then the real-time signal is determined to match the known pattern and the corresponding hydrogen leakage or combustion event pattern is identified; For multiple known patterns , repeat the above steps, calculate the Gromov-Hausdorff distance of each known pattern, and select the one with the smallest distance and less than the threshold The pattern is used as the matching result.

[0020] Preferably, the step of calculating the signal state through a comprehensive decision function, determining whether there is a hydrogen leakage or combustion event, and triggering a corresponding alarm includes: Define the comprehensive decision function :

[0021] in, is the preset weight coefficient; is the calculated Gromov-Hausdorff distance; is the state transition probability matrix The entropy is defined as ; is the maximum value of entropy, , is a random matrix The eigenvalue spectrum deviation of The calculated comprehensive decision function value With preset alarm threshold For comparison: like , it is determined that there is a hydrogen leakage or combustion event, and the corresponding alarm signal is triggered; according to The numerical value can be used to set different alarm levels to indicate the severity of the event; When the judgment conditions are met, the system sends an alarm signal to notify the monitoring center or operator to take corresponding measures.

[0022] Preferably, the random matrix The eigenvalue spectrum deviation The calculation includes the following steps: Constructing the standardized covariance matrix :

[0023] in, is the average value of the topological descriptor, is the time window length; Calculate the eigenvalue set ,in For the matrix Dimensions; according to distribution, calculate the theoretical eigenvalue spectrum :

[0024] in, is the variance of the random matrix elements, ; Calculate the eigenvalue spectrum deviation :

[0025] The larger it is, the greater the deviation between the actual eigenvalue spectrum and the theoretical distribution, and the higher the possibility of abnormal system state.

[0026] The present invention also provides a device for identifying hydrogen leakage and combustion, comprising: Multi-type sensor modules for collecting acoustic signals, electrochemical signals and infrared thermal imaging signals; A processing unit, used for preprocessing the collected signals to generate a multi-dimensional feature space; A topological analysis unit, used for constructing a topological structure, extracting topological features and generating a topological descriptor in a multi-dimensional feature space; A dynamic modeling unit, used to analyze the temporal dynamic changes of topological descriptors based on a Markov chain model; A pattern matching unit, used to calculate the topological similarity between the real-time signal and the known pattern and perform pattern matching; The decision module is used to determine the occurrence of hydrogen leakage and combustion through a comprehensive decision function and issue an alarm signal.

[0027] The present invention provides a method for identifying hydrogen leakage and combustion, which has the following beneficial effects: 1. The present invention can capture the entire process of hydrogen leakage and combustion in real time by combining acoustic, electrochemical and infrared thermal imaging sensors. Different sensors detect signals with different physical properties, ensuring that all stages from trace leakage to combustion can be accurately captured, covering a wide range of working conditions and environments, and achieving comprehensive monitoring.

[0028] 2. The present invention uses a variety of preprocessing techniques such as bandpass filtering, wavelet transform and Hilbert-Huang transform to effectively reduce noise and extract the time-frequency characteristics and instantaneous frequency of the signal, ensuring that key features are extracted from complex original signals. These features provide a high-quality data foundation for subsequent topological analysis and greatly improve the detection accuracy of the system.

[0029] 3. The present invention can extract hidden topological features in the signal and reveal its changing rules at different scales by constructing a topological structure in a multidimensional feature space and calculating the Betti number and continuous bar graph. This topological feature analysis method greatly enhances the robustness and feature extraction capability of the system in a complex signal environment.

[0030] 4. The present invention uses the Markov chain model to model and analyze the time dynamic changes of the topological descriptor, so that the system can track the evolution of the signal over time in real time. By calculating the state transition matrix and steady-state distribution, the state changes of the system can be accurately identified, especially the abnormal state can be captured during hydrogen leakage and combustion, which significantly improves the accuracy of event detection.

[0031] 5. By calculating the Gromov-Hausdorff distance between the real-time signal and the known pattern, the system can quickly evaluate the similarity between the real-time signal and the known hydrogen leakage or combustion pattern, and realize efficient event recognition. This similarity matching method ensures that the system can quickly determine the current event type based on the real-time data, shortening the reaction time.

[0032] 6. The present invention defines a comprehensive decision function by comprehensively considering multiple indicators such as topological similarity, state transition entropy and eigenvalue spectrum deviation, which can quantitatively evaluate the overall state of the system. The decision function ensures accurate judgment of hydrogen leakage and combustion events, and triggers corresponding alarms through the threshold judgment mechanism, providing a reliable alarm mechanism and reducing the possibility of missed alarms and false alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 One of the schematic flow charts of the method of the present invention; Figure 2 This is the second schematic diagram of the method flow of the present invention; Figure 3 It is a schematic diagram of the device structure of the present invention.

[0034] Among them, 100, a multi-type sensor module; 200, a processing unit; 300, a topological analysis unit; 400, a dynamic modeling unit; 500, a pattern matching unit; 600, a decision module. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] Please refer to the attached Figure 1 -Attached Figure 2 The present invention provides a method for identifying hydrogen leakage and combustion, which obtains signals through multiple sensors and combines topological data analysis with random process modeling to monitor hydrogen leakage and combustion in real time.

[0037] The method of the present invention is described in detail below in conjunction with specific implementation steps.

[0038] S1. Collect multiple types of detection signals from acoustic, electrochemical and infrared thermal imaging sensors The identification of hydrogen leakage and combustion depends on the collaborative work of multiple types of sensors, which mainly monitor hydrogen leakage and combustion in real time by collecting signals under different physical characteristics. Various sensors detect the specific characteristics of hydrogen leakage and combustion. Through the combined use of acoustic, electrochemical and infrared thermal imaging sensors, they can comprehensively capture the various states of hydrogen leakage and combustion from different dimensions, ensuring accurate monitoring of hydrogen leakage or combustion in complex environments. Multiple types of sensors provide a way to collect signals under different physical phenomena. These signals can further extract useful information through subsequent processing and analysis techniques, laying the foundation for subsequent steps.

[0039] In this embodiment, step S1 specifically includes collecting the following types of signals: 1. Acoustic signal collection During hydrogen leakage, due to the leakage diameter and gas pressure difference, turbulence and high-frequency ultrasonic signals will be generated at the leakage point. This signal is usually difficult to capture directly by human ears, but can be detected by ultrasonic sensors.

[0040] In this embodiment, an ultrasonic sensor with a frequency range of 20 kHz to 60 kHz is used to capture the ultrasonic signal generated by hydrogen leakage. This frequency range can effectively cover the high-frequency sound waves generated by most trace hydrogen leakage.

[0041] When hydrogen leaks, the principle of ultrasonic wave generation is: when gas leaks from a high-pressure environment through a small hole to a low-pressure environment, turbulence will form at the gas leak due to the high Reynolds number, and this turbulence will generate high-frequency sound waves near the leak hole.

[0042] 2. Collection of electrochemical signals Electrochemical sensors have a sensitive response to gas concentration, especially for small molecule gases such as hydrogen, and have a high detection sensitivity. Electrochemical sensors can accurately detect changes in hydrogen concentration in the air, especially in the case of large-scale leakage. The electrochemical signal is a key indicator for evaluating the degree of leakage.

[0043] In this embodiment, an electrochemical sensor is used to collect hydrogen concentration data. According to national safety standards, the lower explosion limit (LEL) of hydrogen is 4%, so the alarm settings of the electrochemical sensor are divided into a first alarm (1% hydrogen concentration) and a second alarm (2% hydrogen concentration).

[0044] 3. Collection of infrared thermal imaging signals When hydrogen burns, the water vapor generated by its combustion reaction releases infrared radiation at specific wavelengths, making infrared sensors an effective tool for detecting hydrogen combustion. By monitoring thermal radiation in the infrared band, especially the mid-infrared and near-ultraviolet bands generated when hydrogen burns, combustion events can be accurately identified.

[0045] In this embodiment, an infrared thermal imaging sensor with a wavelength of 8 μm to 14 μm is used to detect the infrared radiation signal generated by hydrogen combustion, especially the 2.8 μm infrared band signal released by the combustion product water vapor. The infrared sensor can monitor the high temperature generated during the combustion process in real time, and determine the occurrence of combustion by collecting infrared radiation of a specific wavelength.

[0046] Through the coordinated work of the above three sensors, the present invention can comprehensively cover the whole process monitoring from trace hydrogen leakage, large hydrogen leakage to hydrogen combustion.

[0047] S2. Preprocess the collected signals to generate a multi-dimensional feature space Since the data collected by acoustic, electrochemical and infrared thermal imaging sensors have different noise and complex multidimensional characteristics, it is necessary to reduce noise, perform time-frequency analysis and feature extraction on the data through a series of processing steps to provide high-quality feature information for subsequent topological structure construction. Through preprocessing, the raw data can effectively remove noise, extract valuable features, and map them to a multidimensional feature space to ensure the system's real-time response capabilities to different hydrogen leakage and combustion scenarios.

[0048] In this embodiment, step S2 specifically includes the following signal preprocessing process: 1. Bandpass filtering For the noise and interference signals in the sensor signal, bandpass filtering is first required to remove the signals beyond the target frequency band. For example, the acoustic signal may contain low-frequency noise and high-frequency electromagnetic interference, and bandpass filtering can control the frequency range to 20kHz to 60kHz.

[0049] In this embodiment, the collected acoustic signal is band-pass filtered, and the filtering range is set to 20kHz to 60kHz, so as to ensure that the target information in the signal is not lost and effectively reduce the impact of environmental noise.

[0050] 2. Time-frequency feature extraction: wavelet transform The effective information in the signal often presents instantaneous characteristics, so time-frequency analysis is the key step to further extract useful features. Wavelet transform can realize localized analysis of the signal in time and frequency by decomposing the signal into sub-signals of different scales and positions.

[0051] In this embodiment, the signal after bandpass filtering is subjected to wavelet transformation to extract the time-frequency characteristics of the signal. The formula of wavelet transformation is as follows:

[0052] in, is the filtered signal, is the mother wavelet function, is the scale parameter, is the time shift parameter. By adjusting and , the time-frequency changes of signals can be analyzed at different scales to obtain multi-scale information.

[0053] 3. Instantaneous feature extraction: Hilbert-Huang transform For complex nonlinear signals, the Hilbert-Huang transform (HHT) can be used to further decompose the signal and extract its instantaneous features. HHT consists of empirical mode decomposition (EMD) and Hilbert transform, and is suitable for situations where the frequency changes in the signal are large.

[0054] In this embodiment, the Hilbert-Huang transform is performed on the signal after wavelet transform. First, the signal is decomposed into several intrinsic mode functions (IMFs) using empirical mode decomposition, and then these IMFs are Hilbert transformed to calculate the instantaneous frequency:

[0055] in, is the phase information of the signal, is the instantaneous frequency. Through this transformation, the instantaneous frequency information of the signal at different time points can be extracted, and the subtle changes of the signal during hydrogen leakage or combustion can be captured.

[0056] 4. Constructing a multidimensional feature space After denoising and extracting features from various signals, the extracted features need to be mapped into a high-dimensional space to facilitate subsequent topological structure construction and analysis. The dimension of the feature space depends on the extracted time-frequency features, instantaneous frequency and other signal features.

[0057] In this embodiment, the time-frequency features and instantaneous frequency information are mapped to a multidimensional feature space ,in is the total number of features. This high-dimensional space contains not only the original features of the signal, but also its variation information at different time scales, providing rich input data for subsequent topological structure analysis.

[0058] Through the above steps, the signal preprocessing process in this embodiment can effectively reduce noise and extract key time-frequency features and instantaneous features in the signal, ensuring that the signals collected from different sensors can accurately reflect the status of hydrogen leakage or combustion after processing.

[0059] S3. Construct a topological structure in a multidimensional feature space, extract topological features, and generate a topological descriptor In order to extract the characteristics of hydrogen leakage and combustion from these high-dimensional data, it is necessary to identify topological patterns and persistent features in the data by constructing topological structures. Topological data analysis constructs simplex complexes at different scales and extracts their topological properties to generate topological descriptors that can describe the overall structure of the data. These descriptors reflect the global characteristics of the data and can be effectively used for pattern recognition and further dynamic analysis.

[0060] In this embodiment, step S3 specifically includes the following process: 1. Build topology In step S2, the data after signal preprocessing is mapped to a high-dimensional feature space On this basis, it is necessary to construct the topological structure of the data through topological data analysis.

[0061] In this embodiment, based on different scale parameters , construct the simplex complex of the data. Specifically, in the multidimensional feature space In the example, based on the scale parameter , first build a filter For each data point , construct a radius around it Sphere .along with As the number of spheres increases, they gradually overlap to form complex geometric structures and eventually generate simplex complexes.

[0062] The mathematical description of this process is: for a point set , at each point A sphere is defined on the neighborhood of ,along with As increases, the spheres between adjacent points gradually intersect to form simple shapes (simplexes). When large enough, these simplices form a simplicial complex , which captures the topological structure of the dataset at different scales.

[0063] 2. Betti number calculation Betti number is a key indicator for measuring topological characteristics, reflecting the signal's connectivity and voids in different dimensions. The Betti number under the condition can reveal the topological changes of the signal at different scales.

[0064] In this embodiment, different scales are calculated Betti numbers under , which is defined as follows:

[0065] in, for dimensional boundary operator, represents the kernel space, represents the image space. Specifically, represents the number of connected components, represents the number of one-dimensional rings, Represents the number of two-dimensional holes.

[0066] 3. Topological feature extraction With the scale parameter The connected components and holes in the topological structure will change with the change of With scale parameter The topological characteristics of the signal can be extracted by analyzing the changes in the signal.

[0067] In this embodiment, topological features are extracted through continuous homology analysis. Continuous homology is to analyze the topological structure at different scales. The tool records the birth and death of topological features by generating a persistence barcode. The length of the bar in the persistence barcode reflects the "persistence" of the topological feature. The longer the bar, the more important the topological feature is at multiple scales.

[0068] 4. Topology Descriptor Generation In order to facilitate subsequent pattern matching and dynamic analysis, it is necessary to map the topological features to the topological descriptor space. A topological descriptor is a set of values ​​that represents the birth, death, and persistence information of each topological feature.

[0069] In this embodiment, the topological features in the continuous bar graph are mapped to the multidimensional descriptor space ,in is the dimension of the descriptor space. Each topological descriptor vector Represents the persistence and scale information of a specific topological feature and is defined as:

[0070] in, and Respectively The birth and death scales of topological features, is the corresponding Betti number. This descriptor can effectively quantify the topological features in the signal and provide an important numerical basis for subsequent pattern matching.

[0071] By constructing a topological structure in a multidimensional feature space and analyzing the topological changes of the signal at different scales, this embodiment can effectively extract the topological features of the signal and generate a topological descriptor. The calculation of the Betti number reveals the topological properties of the signal in different dimensions, while the continuous bar graph and topological descriptor provide key numerical inputs for subsequent signal pattern recognition.

[0072] S4. Analyzing the temporal dynamics of topological descriptors based on the Markov chain model By dynamically modeling the topological descriptors generated in the previous step, we can further analyze their evolution in the time series and identify the state changes of hydrogen leakage and combustion. Using the Markov chain model to model and analyze the time dynamics of the topological descriptors helps capture the transfer characteristics of the signal in the time series, so as to determine whether the system is in a stable state or abnormal. By constructing a state transition matrix and analyzing the changes in state distribution, we can provide support for subsequent pattern matching and decision-making.

[0073] In this embodiment, the specific content of step S4 is as follows: 1. Define the state space The topological descriptor evolves continuously in the time series as the hydrogen leakage or combustion state changes. In order to describe these dynamic changes, it is first necessary to define a discrete state space, where each state corresponds to a topological descriptor.

[0074] In this embodiment, the state space is defined as , where each state The topological descriptor vector corresponding to a certain point in time , according to the time series data, obtain the topological descriptor sequence , the corresponding state sequence is These states are used to represent how different topological properties of the signal change over time.

[0075] 2. Construct the state transfer matrix The temporal evolution of the topological descriptor can be described by a state transition matrix, where each element in the matrix represents the probability of transition from one state to another.

[0076] In this embodiment, the state transition probability matrix is ​​constructed by analyzing the changes of topological descriptors in the continuous time series. ,in Indicates from the state Transfer to state The probability is calculated as follows:

[0077] in, From the state Transfer to state The number of times Status The total number of transitions, in this way, the state transfer matrix It reflects the topological dynamic characteristics of the system in time series.

[0078] 3. Verify Markov property In order to ensure the accuracy of the state transition model, it is necessary to verify whether the state sequence satisfies the Markov property. Markov property means that the future state of the system depends only on the current state and has nothing to do with the previous state.

[0079] In this embodiment, by checking the state transition probability matrix , verify whether it satisfies the Markov property, that is:

[0080] This ensures that the state transition process of the system can be reasonably described by the Markov chain model, thereby performing effective dynamic modeling.

[0081] 4. Steady-state distribution calculation The long-term evolution trend of the topological descriptor in the time series can be analyzed through the steady-state distribution of the Markov chain. The steady-state distribution reflects the probability that the topological descriptor stays in certain states after the system runs for a long time.

[0082] In this embodiment, the steady-state distribution of the Markov chain is calculated by solving the following equation: :

[0083] And the normalization condition is satisfied:

[0084] Steady-state distribution It reflects the probability of occupying each state of the system in long-term operation. By analyzing the steady-state distribution, it can be determined whether the topological state of the system in long-term operation tends to be stable.

[0085] 5. Anomaly Detection During hydrogen leakage or combustion, the system may experience abnormal conditions. By analyzing the deviation between the actual state distribution and the steady-state distribution, it is possible to identify whether the system has experienced an abnormality.

[0086] In this embodiment, by comparing the current state distribution of the system with the steady-state distribution , to determine whether the system deviates from the normal state. If the difference between the current state distribution and the steady-state distribution is large, it indicates that the system may be in an abnormal state. The mathematical expression of anomaly detection is:

[0087] in, is the state distribution at the current moment, is a steady-state distribution. If the difference If the set threshold is exceeded, the system will trigger an abnormal alarm, indicating that there may be a hydrogen leak or combustion incident.

[0088] By analyzing the temporal dynamic changes of topological descriptors through the Markov chain model, this embodiment can effectively capture the change pattern of the signal in the time series. The state transition matrix reflects the short-term change trend of the system, while the steady-state distribution reveals the long-term operating state of the system. Combined with the anomaly detection mechanism, the present invention can identify abnormal conditions in hydrogen leakage or combustion in real time, and provide basic data for subsequent pattern matching and comprehensive decision-making.

[0089] S5. Calculate the topological similarity between the real-time signal and the known pattern and perform pattern matching By calculating the topological similarity between the real-time signal and the known pattern, the occurrence of hydrogen leakage or combustion events can be effectively identified. To complete this process, it is first necessary to perform a metric analysis on the topological descriptors of the real-time signal and the known pattern, and measure the similarity by calculating the Gromov-Hausdorff distance between the two. Then, by comparing the calculation result with the preset threshold, it can be determined whether the real-time signal matches a known pattern. This pattern matching technology can identify the specific situation of hydrogen leakage or combustion and provide an accurate basis for event determination.

[0090] In this embodiment, the specific content of step S5 is as follows: 1. Constructing a metric space The topological descriptors of real-time signals and known patterns are represented in two different metric spaces, respectively.

[0091] In this embodiment, the real-time topology descriptor set is represented as , the set of topological descriptors of known patterns is represented as These two sets can be viewed as two independent metric spaces. Define the corresponding metric function and to measure the distance between different descriptors.

[0092] 2. Define the metric function In order to compare the similarity of real-time signals with known patterns, a metric function between topological descriptors is defined.

[0093] In this embodiment, the Euclidean distance is used as the metric function between topological descriptors. The specific formula is as follows:

[0094] in, Represents the Euclidean norm, which is used to calculate the distance between two descriptors in the feature space. By calculating the Euclidean distance between topological descriptors, the topological difference between the real-time signal and the known pattern can be quantified.

[0095] 3. Establish corresponding relationship Real-time Descriptor Set and a set of known pattern descriptors A correspondence is established between them to ensure that each real-time descriptor has a corresponding known mode descriptor for comparison.

[0096] In this embodiment, establish and The corresponding relationship between , so that for Each element in ,exist Elements in ,satisfy This correspondence ensures the matching computation between the two sets of descriptors.

[0097] 4. Calculate the Gromov-Hausdorff distance The Gromov-Hausdorff distance is used to measure the similarity between two metric spaces by considering all possible correspondences and finding the minimum maximum difference between descriptor sets.

[0098] In this embodiment, the real-time signal descriptor set is calculated With a set of known pattern descriptors The Gromov-Hausdorff distance between is as follows:

[0099] in, Represents taking all possible correspondences The minimum value in Indicated in The maximum value is taken, and the distance reflects the maximum and minimum similarity between the two metric spaces.

[0100] 5. Similarity Assessment The Gromov-Hausdorff distance calculated by , it is possible to evaluate the similarity between the real-time signal and the known pattern. The smaller the distance, the higher the similarity between the two.

[0101] In this embodiment, the Gromov-Hausdorff distance is used Evaluate the similarity of the live signal to the known pattern. The similarity evaluation is based on the fact that the smaller the distance, the closer the live signal is to the known pattern.

[0102] 6. Matching determination By comparing the calculated Gromov-Hausdorff distance with the preset similarity threshold, it is determined whether the real-time signal matches a known pattern.

[0103] In this embodiment, Similarity threshold with the preset For comparison, if , the real-time signal is determined to match the known pattern and the corresponding hydrogen leakage or combustion event pattern is identified.

[0104] 7.Multiple pattern matching In practical applications, there may be multiple known patterns for matching. In order to ensure the best matching result, it is necessary to calculate each known pattern and select the best matching result.

[0105] In this embodiment, for multiple known patterns , repeat the above steps to calculate the Gromov-Hausdorff distance of each known pattern Finally, the distance is the smallest and less than the threshold. The pattern is used as the matching result, thereby accurately identifying the current hydrogen leakage or combustion event.

[0106] By calculating the Gromov-Hausdorff distance between the real-time signal and the known pattern, the present embodiment can effectively measure the similarity between the two, thereby achieving pattern matching. The distance calculation takes into account all possible correspondences to ensure the accuracy of the matching. In addition, the multi-pattern matching mechanism provides a reliable basis for identifying different hydrogen leakage and combustion events.

[0107] S6. Calculate the signal status through the comprehensive decision function to determine whether there is a hydrogen leak or combustion event, and trigger the corresponding alarm The final judgment of hydrogen leakage and combustion identification is obtained by integrating the analysis results of multiple information sources, including real-time topological similarity, dynamic characteristics of state transitions, and overall state changes of the system. By defining a comprehensive decision function, this information is summarized and quantified to ensure accurate judgment of hydrogen leakage and combustion events. This step also involves a threshold determination mechanism, which compares the calculated comprehensive decision function value with the preset threshold to determine whether there is a leakage or combustion event and trigger the corresponding alarm signal. The comprehensiveness of this method ensures efficient monitoring and accurate early warning of hydrogen events under complex working conditions.

[0108] In this embodiment, step S6 specifically includes the following contents: 1. Define the comprehensive decision function The comprehensive decision function is used to comprehensively consider the results of multiple different information sources (such as topological similarity, state transition entropy, and characteristic spectrum deviation) to reach an overall judgment on the hydrogen leakage or combustion event.

[0109] In this embodiment, the comprehensive decision function is defined as:

[0110] in: is the Gromov-Hausdorff distance between the real-time signal and the known pattern calculated in step S5; is the state transition probability matrix The entropy is used to reflect the dynamic changes of the system and is defined as:

[0111] is the maximum value of entropy, ,in is the size of the state space; It is the deviation of the eigenvalue spectrum of the random matrix, reflecting the difference between the current state of the system and the normal state.

[0112] It is a preset weight coefficient used to balance the impact of various information in decision-making.

[0113] 2. Calculate the value of the comprehensive decision function By calculating the comprehensive decision function, the state evaluation value of the system at the current moment can be obtained: , which combines the weights and state changes of different information sources.

[0114] In this embodiment, the comprehensive decision function value The calculation formula is:

[0115] in, is the topological similarity distance, is the entropy of the system state, is the eigenvalue spectrum deviation of the system state. Used for subsequent threshold determination.

[0116] 3. Threshold determination and event judgment By comparing the comprehensive decision function value Compared with the preset threshold, it can be determined whether there is a hydrogen leakage or combustion event in the system.

[0117] In this embodiment, With preset alarm threshold For comparison, if , it indicates that there may be hydrogen leakage or combustion in the system. The size of the event can be further used to determine the severity of the event. For example: when The value is much larger than When it is, it indicates that the system status is seriously deviated from the normal state, and a major leakage or combustion incident may occur; when Slightly larger than It may be a minor leak that requires continued monitoring.

[0118] 4. Alarm trigger mechanism When the system determines that a hydrogen leak or combustion event occurs, a corresponding alarm signal will be triggered. The alarm mechanism can be set to different levels according to the severity of the event to issue an early warning to monitoring personnel or related systems.

[0119] In this embodiment, when When the alarm occurs, the system immediately triggers an alarm signal to notify the monitoring center or operator to take emergency measures. The alarm level can be divided into multiple levels, such as: Level 1 alarm: Minor leak or burn, the system recommends continuous monitoring; Second level alarm: significant leakage or burning, immediate response is required; Level 3 alarm: Major leakage or combustion may lead to dangerous incidents and requires emergency treatment.

[0120] By integrating the analysis results of multiple information sources, constructing a comprehensive decision function, and judging hydrogen leakage or combustion events through a threshold judgment mechanism, this embodiment can provide comprehensive and reliable event recognition capabilities. The calculated comprehensive decision function value not only takes into account the topological similarity of real-time signals, but also combines the dynamic changes of system states and the overall deviation, ensuring the accuracy of judgment and the timeliness of response. At the same time, the flexible alarm mechanism provides appropriate response measures for different levels of hydrogen leakage and combustion events.

[0121] The present invention realizes real-time and comprehensive monitoring of the entire process of hydrogen leakage and combustion by combining the collaborative work of multiple types of sensors (acoustic, electrochemical and infrared thermal imaging sensors). Through signal preprocessing, topological structure analysis and time dynamic modeling of Markov chain model, the key features of the signal can be effectively extracted and analyzed, and high-precision event recognition and alarm can be ensured through topological similarity matching and comprehensive decision functions. This method has high sensitivity, low false negative rate and real-time response capability, and can effectively improve the reliability and stability of the safety monitoring system under complex working conditions.

[0122] Further, for the random matrix eigenvalue spectrum deviation described in step S6 , which is used to evaluate the degree of deviation between the current state of the system and the normal state, thereby assisting the calculation of the comprehensive decision function. In practical applications, the deviation of the eigenvalue spectrum reflects the difference between the current system state and the expected normal state, which is critical for detecting hydrogen leakage or possible abnormal conditions during combustion.

[0123] In this embodiment, the random matrix eigenvalue spectrum deviation The calculation steps are as follows: 1. Construct a standardized covariance matrix Random Matrix It is a covariance matrix constructed based on feature space data. By standardizing the covariance matrix, the changes in signal characteristics can be quantified.

[0124] In this embodiment, the multi-dimensional feature data after signal processing is used , construct the standardized covariance matrix , the formula is as follows:

[0125] in, is the average value of the topological descriptor, is the time window length, the matrix It is a covariance matrix constructed based on the data in the current time window, reflecting the covariance relationship of the system characteristic data.

[0126] 2. Calculate the eigenvalue set Covariance matrix The eigenvalues ​​reflect the intensity of changes in signal features in different dimensions. By calculating the eigenvalue set, we can further analyze the changes in the system in the current state.

[0127] In this embodiment, the standardized covariance matrix Perform eigenvalue decomposition to obtain the eigenvalue set ,in For the matrix These eigenvalues ​​reflect the distribution and changes of the signal in different directions. The larger the eigenvalue, the greater the signal change in the corresponding dimension.

[0128] 3. Determine the theoretical eigenvalue spectrum In order to evaluate the abnormality of the current state, it is necessary to compare the calculated eigenvalue set with the theoretical eigenvalue spectrum under normal conditions. The theoretical eigenvalue spectrum can be obtained by Distribution (for random matrices) to be estimated.

[0129] In this embodiment, according to distribution, calculate the theoretical eigenvalue spectrum :

[0130] in, is the variance of the random matrix elements, is the ratio of the matrix dimension to the time window length. This distribution gives the range of the eigenvalue distribution of a random matrix under normal conditions.

[0131] 4. Calculate the eigenvalue spectrum deviation By comparing the actual eigenvalue set and the theoretical eigenvalue spectrum , can evaluate the deviation degree of the current system state. Eigenvalue spectrum deviation It is used to quantify this deviation and reflects the difference between the current state of the system and the normal state.

[0132] In this embodiment, the characteristic value spectrum deviation The calculation formula is:

[0133] This formula calculates each eigenvalue and the corresponding theoretical eigenvalue The absolute difference between the two values ​​is calculated and these differences are accumulated to obtain the overall deviation of the system state. If it is large, it means that the current system status is significantly different from the normal status, which may be a sign of hydrogen leakage or combustion event.

[0134] 5. Abnormal status determination By calculating the eigenvalue spectrum deviation, it can be determined whether the system is currently in an abnormal state. Values ​​below 0.1% indicate a significant change in system characteristics and a possible risk of hydrogen leakage or combustion.

[0135] In this embodiment, if the characteristic value spectrum deviation Exceeds a preset threshold , then the current system state is judged to be abnormal, and there may be hydrogen leakage or combustion events. This result will be used to calculate the comprehensive decision function to help the system make alarm judgments.

[0136] The device for identifying hydrogen leakage and combustion described below and the method for identifying hydrogen leakage and combustion described above may refer to each other.

[0137] Please refer to the attached Figure 3 The present invention also provides a device for identifying hydrogen leakage and combustion, comprising: A multi-type sensor module 100 for collecting acoustic signals, electrochemical signals and infrared thermal imaging signals; The processing unit 200 is used to pre-process the collected signals and generate a multi-dimensional feature space; A topology analysis unit 300, used to construct a topological structure, extract topological features and generate a topological descriptor in a multi-dimensional feature space; A dynamic modeling unit 400, for analyzing the temporal dynamic changes of the topological descriptor based on a Markov chain model; A pattern matching unit 500, used to calculate the topological similarity between the real-time signal and the known pattern and perform pattern matching; The decision module 600 is used to determine the occurrence of hydrogen leakage and combustion through a comprehensive decision function and issue an alarm signal.

[0138] The device of this embodiment can be used to execute the above method embodiment, and its principles and technical effects are similar, which will not be repeated here.

[0139] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying hydrogen leakage and combustion, characterized in that: The following steps are involved: Collect multiple types of detection signals from acoustic, electrochemical and infrared thermal imaging sensors; Preprocess the collected signals to generate a multi-dimensional feature space; Construct a topological structure in a multidimensional feature space, extract topological features, and generate a topological descriptor; Analyze the temporal dynamics of topological descriptors based on the Markov chain model; Calculate the topological similarity between real-time signals and known patterns and perform pattern matching; The signal status is calculated through a comprehensive decision function to determine whether there is a hydrogen leak or combustion event and trigger the corresponding alarm.

2. A method for identifying hydrogen leakage and combustion according to claim 1, characterized in that: In the signal acquisition step, the acquisition frequency range of the acoustic signal is 20kHz to 60kHz, the acquisition range of the electrochemical signal includes hydrogen concentration and temperature data, and the acquisition wavelength range of the infrared thermal imaging signal is 8μm-14μm and 2.8μm.

3. A method for identifying hydrogen leakage and combustion according to claim 1, characterized in that: The step of preprocessing the collected signal comprises: Perform bandpass filtering on the collected signal to remove noise outside the target frequency band; Apply wavelet transform and Hilbert-Huang transform to the filtered signal to extract time-frequency features; Map the processed signal to dimensional joint feature space ,in is the total number of features extracted from the signal.

4. A method for identifying hydrogen leakage and combustion according to claim 1, characterized in that: The step of constructing a topological structure in a multidimensional feature space comprises: In multidimensional feature space In the example, based on the scale parameter Building filters ; For each data point , create a radius around it Sphere ; along with As , these spheres gradually overlap to form simplex complexes; Calculating different scales Betti numbers under ,in: in, for dimensional boundary operator, represents the kernel space, Represents image space; By analyzing the Betti number as a function of the scale parameter The changes of ,extract the topological features of the signal; Generates continuous bar graphs recording the birth and death scales of topological features.

5. A method for identifying hydrogen leakage and combustion according to claim 4, characterized in that: The generation of the topology descriptor includes: Mapping topological features in persistent bar graphs to multidimensional descriptor space ,in is the dimension of the descriptor space; Each topological descriptor vector Represents the persistence and scale information of a specific topological feature and is defined as: in, and Respectively The birth and death scales of topological features, is the corresponding Betti number.

6. A method for identifying hydrogen leakage and combustion according to claim 1, characterized in that: The step of analyzing the temporal dynamic changes of the topological descriptor based on the Markov chain model comprises: Constructing the state space , where each state Corresponds to a topological descriptor vector ; According to the time series data, obtain the topological descriptor sequence , the corresponding state sequence is ; Count the number of state transitions and construct the state transition probability matrix ,in: in, From the state Transfer to state The number of times Status Total number of transfers; Verify whether the state sequence satisfies the Markov property, that is, it satisfies: For any historical state sequence, ensure that the current state only depends on the state at the previous moment; Solving for the Steady-State Distribution of a Markov Chain ,satisfy: Analyze the properties of the state transition matrix through eigenvalue decomposition to determine the change pattern of the topological descriptor; Based on the changes in transition probabilities and deviations from steady-state distributions, anomalous topological descriptor changes are identified as indicators of possible hydrogen leaks or combustion events.

7. A method for identifying hydrogen leakage and combustion according to claim 1, characterized in that: The step of calculating the topological similarity between the real-time signal and the known pattern and performing pattern matching comprises: The real-time topology descriptor set and a set of known pattern topology descriptors Represented as a metric space, define the corresponding metric function and ; Establish and The corresponding relationship between , so that for Each element in ,exist Elements in Make ; calculate and The Gromov-Hausdorff distance between , defined as: in, Represents taking all possible correspondences The minimum value in Indicated in Take the maximum value; Metric Function and Defined as the Euclidean distance between topological descriptors: in, is the Euclidean norm; Using the calculated Gromov-Hausdorff distance Evaluate the similarity between the real-time signal and the known pattern. The smaller the distance, the higher the similarity. Will Similarity threshold with the preset For comparison, if , then the real-time signal is determined to match the known pattern and the corresponding hydrogen leakage or combustion event pattern is identified; For multiple known patterns , repeat the above steps, calculate the Gromov-Hausdorff distance of each known pattern, and select the one with the smallest distance and less than the threshold The pattern is used as the matching result.

8. A method for identifying hydrogen leakage and combustion according to claim 1, characterized in that: The steps of calculating the signal state through a comprehensive decision function, determining whether there is a hydrogen leakage or combustion event, and triggering a corresponding alarm include: Define the comprehensive decision function : in, is the preset weight coefficient; is the calculated Gromov-Hausdorff distance; is the state transition probability matrix The entropy is defined as ; is the maximum value of entropy, , is the size of the state space; is a random matrix The eigenvalue spectrum deviation of The calculated comprehensive decision function value With preset alarm threshold For comparison: like , it is determined that there is a hydrogen leakage or combustion event, and the corresponding alarm signal is triggered; according to The numerical value can be used to set different alarm levels to indicate the severity of the event; When the judgment conditions are met, the system sends an alarm signal to notify the monitoring center or operator to take corresponding measures.

9. A method for identifying hydrogen leakage and combustion according to claim 8, characterized in that: The random matrix The eigenvalue spectrum deviation The calculation includes the following steps: Constructing the standardized covariance matrix : in, is the average value of the topological descriptor, is the time window length; Calculate the eigenvalue set ,in For the matrix Dimensions; according to distribution, calculate the theoretical eigenvalue spectrum : in, is the variance of the random matrix elements, ; Calculate the eigenvalue spectrum deviation : The larger it is, the greater the deviation between the actual eigenvalue spectrum and the theoretical distribution, and the higher the possibility of abnormal system state.

10. A hydrogen leakage and combustion identification device, applied to the method according to any one of claims 1 to 9, characterized in that: include: Multi-type sensor modules for collecting acoustic signals, electrochemical signals and infrared thermal imaging signals; A processing unit, used for preprocessing the collected signals to generate a multi-dimensional feature space; A topological analysis unit, used for constructing a topological structure, extracting topological features and generating a topological descriptor in a multi-dimensional feature space; A dynamic modeling unit, used to analyze the temporal dynamic changes of topological descriptors based on a Markov chain model; A pattern matching unit, used to calculate the topological similarity between the real-time signal and the known pattern and perform pattern matching; The decision module is used to determine the occurrence of hydrogen leakage and combustion through a comprehensive decision function and issue an alarm signal.

Images