Method and device for detecting leakage of hydrogen storage spherical tank

Abstract

This invention discloses a method and apparatus for detecting leaks in hydrogen storage spherical tanks. The method acquires detection data from a multimodal sensor network deployed around the target hydrogen storage spherical tank; initializes a particle set within a pre-defined potential leak area of ​​the target tank, with each particle representing a candidate leak point location and having a corresponding weight; iteratively updates the particle filtering process, predicting the particle set based on a hydrogen diffusion model in each iteration, simulating sensor prediction readings using sensor response models, calculating particle likelihood based on the matching degree between the predicted readings and the detection data, updating particle weights, and resampling the particles to generate a new particle set; until the iteration termination condition is met, a weighted average of the particle positions is calculated to obtain the suspected leak point location. This method can accurately locate leak points in hydrogen storage spherical tanks, improving the safety monitoring level of hydrogen storage equipment.

Description

Technical Field

[0001] This invention relates to the field of hydrogen leak detection technology, and more specifically, to a method and apparatus for detecting leaks in hydrogen storage spherical tanks. Background Technology

[0002] Large-capacity hydrogen storage spherical tanks are crucial for hydrogen energy storage, and their safety is paramount. Hydrogen is highly flammable and explosive; even a small leak can trigger a serious accident. Traditional leak detection methods often suffer from low detection accuracy, slow response times, and inability to accurately pinpoint the leak location, making them unsuitable for meeting high safety requirements.

[0003] Therefore, how to propose a detection device and method that integrates multiple sensors, can perform data fusion, and achieve accurate leak location, thereby improving the safety of large-capacity hydrogen storage tanks, is a technical problem that urgently needs to be solved by existing technologies. Summary of the Invention

[0004] In order to solve at least one of the technical problems in the background art, the present invention proposes a method and apparatus for detecting leaks in hydrogen storage spherical tanks.

[0005] One aspect of the present invention provides a method for detecting leaks in a hydrogen storage spherical tank, the method comprising: Acquire detection data from a multimodal sensor network deployed around the target hydrogen storage sphere, the multimodal sensor network including hydrogen sensors, acoustic sensors, and temperature sensors; A particle set is initialized within a preset possible leakage area of ​​the target hydrogen storage spherical tank. Each particle in the particle set represents a candidate leakage point location and has a corresponding particle weight. The particle filtering iterative update process is executed repeatedly until the iteration termination condition is met. Based on the particle position and corresponding particle weight of each particle in the final iteration round, the particle positions of each particle are weighted and averaged to obtain the suspected leak point location corresponding to the target hydrogen storage tank. In each iteration round: first, the particle set obtained in the previous iteration is predicted based on the hydrogen diffusion model to obtain the predicted position of each particle; based on the predicted position of each particle, the sensor prediction readings corresponding to each particle are simulated using the hydrogen sensor response model, acoustic sensor response model, and temperature sensor response model, respectively; the particle likelihood of each particle is calculated based on the degree of matching between the sensor prediction readings and the detection data; the particle weights of each particle are updated according to the particle likelihood, and the updated particle weights are normalized; the particles are resampled based on the normalized particle weights to generate a new particle set for the next iteration round.

[0006] Optionally, the multimodal sensor network may also include a vibration sensor; The method further includes: The detection data collected by the multimodal sensor network is preprocessed, and a normal operation baseline is established based on the detection data collected by the target hydrogen storage sphere under normal and stable operation. The normal operation baseline includes the statistics of each sensor under normal operating conditions. The corresponding feature data are extracted from the detection data of different types of sensors, and the feature data is normalized. The initial weights of various sensors are determined, and the initial weights are dynamically updated based on the confidence level of each sensor to obtain the dynamic weights of each sensor. The confidence level is used to characterize the stability and anomaly of the detection data of the corresponding sensor, and the anomaly is characterized at least by the degree of deviation of the current detection data from the normal operating baseline. Based on the dynamic weights, the normalized feature data is weighted and fused to obtain fused feature values; Based on the normal operation baseline, preset thresholds are set for each risk level, and the risk level of the target hydrogen storage spherical tank is determined according to the relationship between the fusion feature value and the preset threshold.

[0007] Optionally, the method for detecting leaks in the hydrogen storage spherical tank further includes: When the risk level reaches the preset alarm level, and the positioning result of the particle filter iterative update process shows a high-confidence leakage area, the suspected leakage point location is determined as the final leakage point location and output. The high-confidence leakage area is characterized by at least the particle set converging to a preset small area and the total particle weight in the small area reaching a preset threshold. When the risk level indicates low risk or no abnormality, the particle filter localization process is paused or its execution priority is reduced.

[0008] Optionally, the hydrogen sensors are arranged in a ring at the bottom of the spherical tank, the surrounding ground, and the diffusion channel; the acoustic sensors and the vibration sensors are arranged on the surface of the spherical tank, the support structure, the pipeline connections, and the valves; and the temperature sensors are arranged on the surface of the spherical tank and the surrounding area.

[0009] Optionally, the preprocessing of the detection data acquired by the multimodal sensor network includes: The detection data from different sensors are timestamped and resampled under different sampling rate scenarios. Wavelet transform is used to denoise the detection data from the acoustic sensor. The detection data from the vibration sensor is filtered and then synchronized by combining cross-correlation. The temperature sensor's detection data is smoothed and filtered to smooth the temperature field. The detection data from the hydrogen sensor is filtered to highlight the characteristics of the rate of change in hydrogen concentration.

[0010] Optionally, the step of extracting corresponding feature data from the detection data of different types of sensors includes: Extract signal energy characteristics within a preset time window from the detection data of the acoustic sensor; Extracting root mean square (RMS) value features from the detection data of vibration sensors; Extract the difference between the measured temperature and the average normal operating temperature from the temperature sensor's detection data; Extract the hydrogen concentration change rate feature from the detection data of the hydrogen sensor.

[0011] Optionally, the step of predicting the particle set obtained from the previous iteration based on the hydrogen diffusion model to obtain the predicted position of each particle includes: Based on the current position of the particle, the drift effect caused by the environmental wind speed vector and the random diffusion effect are superimposed to obtain the predicted position of each particle in the current iteration round. The random diffusion effect is characterized by a random vector sampled from a standard multidimensional normal distribution and an effective diffusion coefficient. The time step of the particle filtering is matched with the time resolution of the data fusion. The step of calculating the particle likelihood of each particle based on the degree of matching between the sensor's predicted reading and the detection data, and updating the particle weight of each particle based on the particle likelihood, includes: The particle likelihood terms are calculated based on the actual detection data of the hydrogen sensor, acoustic sensor, and temperature sensor and the corresponding predicted readings of the sensors, respectively. The particle likelihood terms are multiplied together to obtain the total particle likelihood function. The total particle likelihood function is then multiplied by the particle weights from the previous iteration to obtain the updated particle weights. The resampling of particles based on normalized particle weights to generate a new particle set for the next iteration includes: Based on particle weights, particles are copied and eliminated to form a new particle set, and after resampling, the particle weights of the new particle set are reset to equal weights. The step of weighted averaging the particle positions of each particle to obtain the location of the suspected leak point corresponding to the target hydrogen storage tank includes: The suspected leak point location is obtained by weighting the particle positions of each particle in the current iteration round with the resampled particle weights.

[0012] In another aspect, the present invention provides a hydrogen storage spherical tank leakage detection device, the device comprising: The detection data acquisition unit is used to acquire detection data from a multimodal sensor network deployed around the target hydrogen storage sphere, wherein the multimodal sensor network includes a hydrogen sensor, an acoustic sensor, and a temperature sensor. The particle set initialization unit is used to initialize a particle set within a preset possible leakage area of ​​the target hydrogen storage spherical tank. Each particle in the particle set represents a candidate leakage point location and has a corresponding particle weight. The particle filtering iteration unit is used to repeatedly execute the particle filtering iteration update process until the iteration termination condition is met. Based on the particle position and corresponding particle weight of each particle in the final iteration round, the particle positions of each particle are weighted and averaged to obtain the suspected leak point location corresponding to the target hydrogen storage sphere. In each iteration round: first, the particle set obtained in the previous iteration is predicted based on the hydrogen diffusion model to obtain the predicted position of each particle; based on the predicted position of each particle, the sensor prediction readings corresponding to each particle are simulated using the hydrogen sensor response model, acoustic sensor response model, and temperature sensor response model, respectively; the particle likelihood of each particle is calculated based on the degree of matching between the sensor prediction readings and the detection data; the particle weights of each particle are updated according to the particle likelihood, and the updated particle weights are normalized; the particles are resampled based on the normalized particle weights to generate a new particle set for the next iteration round.

[0013] To achieve the above objectives, according to another aspect of the present invention, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described hydrogen storage tank leakage detection method.

[0014] To achieve the above objectives, according to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program / instructions are stored, which, when executed by a processor, implement the steps of the above-described hydrogen storage tank leakage detection method.

[0015] The beneficial effects of this invention are as follows: This invention employs a multimodal sensor network deployed around a hydrogen storage tank to acquire multi-source detection data. A particle filtering iterative update mechanism is introduced within a pre-defined potential leak area. Combined with a hydrogen diffusion model, the location of candidate leak points is dynamically predicted and corrected. The particle weights are continuously updated based on the matching degree between sensor predictions and actual detection data. After iterative convergence, the particle positions are weighted and averaged. This process suppresses the influence of noise and uncertainty under complex environmental conditions, enabling reliable determination of the leak location in the hydrogen storage tank. It can accurately locate the leak point and improve the safety monitoring level of hydrogen storage equipment. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart of the leakage point localization based on particle filtering in an embodiment of the present invention; Figure 2 This is a flowchart of the risk detection process according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the output of leak detection results in an embodiment of the present invention; Figure 4 This is a structural block diagram of the hydrogen storage spherical tank leakage detection device according to an embodiment of the present invention; Figure 5 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0019] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] The present invention aims to design a detection method that integrates multiple sensors, performs data fusion, and achieves accurate leak location, thereby significantly improving the safety of large-capacity hydrogen storage spherical tanks.

[0022] Figure 1 This is a flowchart of the leakage point localization based on particle filtering according to an embodiment of the present invention, as follows: Figure 1 As shown, in one embodiment of the present invention, the method for detecting leakage in a hydrogen storage spherical tank includes steps S101 to S103.

[0023] Step S101: Obtain detection data from a multimodal sensor network deployed around the target hydrogen storage sphere, wherein the multimodal sensor network includes a hydrogen sensor, an acoustic sensor, and a temperature sensor.

[0024] Step S102: Initialize a particle set within a preset possible leakage area of ​​the target hydrogen storage sphere. Each particle in the particle set represents a candidate leakage point location and has a corresponding particle weight.

[0025] Step S103: The particle filtering iterative update process is executed repeatedly until the iteration termination condition is met. Based on the particle position and corresponding particle weight of each particle in the final iteration round, the particle position of each particle is weighted and averaged to obtain the suspected leak point position corresponding to the target hydrogen storage tank. In each iteration round: First, the particle set obtained in the previous iteration is predicted based on the hydrogen diffusion model to obtain the predicted position of each particle; based on the predicted position of each particle, the sensor prediction readings corresponding to each particle are simulated using the hydrogen sensor response model, acoustic sensor response model, and temperature sensor response model, respectively; the particle likelihood of each particle is calculated based on the degree of matching between the sensor prediction readings and the detection data; the particle weights of each particle are updated according to the particle likelihood, and the updated particle weights are normalized; the particles are resampled based on the normalized particle weights to generate a new particle set for the next iteration round.

[0026] In one embodiment of the present invention, the hydrogen sensor is arranged in a ring at the bottom of the spherical tank, the surrounding ground, and the diffusion channel; the acoustic sensor and the vibration sensor are arranged on the surface of the spherical tank, the support structure, the pipeline connection, and the valve; and the temperature sensor is arranged on the surface of the spherical tank and the surrounding area.

[0027] In one embodiment of the present invention, step S103 above, which involves predicting the particle set obtained from the previous iteration based on the hydrogen diffusion model to obtain the predicted position of each particle, includes: Based on the current position of the particle, the drift effect caused by the environmental wind speed vector and the random diffusion effect are superimposed to obtain the predicted position of each particle in the current iteration round. The random diffusion effect is characterized by a random vector sampled from a standard multidimensional normal distribution and an effective diffusion coefficient. The time step of the particle filtering is matched with the time resolution of the data fusion. In one embodiment of the present invention, step S103 above, which calculates the particle likelihood of each particle based on the degree of matching between the sensor's predicted reading and the detection data, and updates the particle weight of each particle according to the particle likelihood, includes: The particle likelihood terms are calculated based on the actual detection data of the hydrogen sensor, acoustic sensor, and temperature sensor and the corresponding predicted readings of the sensors, respectively. The particle likelihood terms are multiplied together to obtain the total particle likelihood function. The total particle likelihood function is then multiplied by the particle weights from the previous iteration to obtain the updated particle weights. In one embodiment of the present invention, the step S103 above, which involves resampling the particles based on the normalized particle weights to generate a new particle set for the next iteration, includes: Based on particle weights, particles are copied and eliminated to form a new particle set, and after resampling, the particle weights of the new particle set are reset to equal weights. In one embodiment of the present invention, the step S103 above, which involves weighted averaging the particle positions of each particle to obtain the location of the suspected leak point corresponding to the target hydrogen storage tank, includes: The suspected leak point location is obtained by weighting the particle positions of each particle in the current iteration round with the resampled particle weights.

[0028] In this embodiment of the invention, to achieve precise location of the leak point in the hydrogen storage tank, a multimodal sensor network is constructed around the tank. The multimodal sensor network includes at least a ring-shaped distribution of hydrogen sensors, distributed acoustic sensors, and vibration sensors, as well as temperature sensors covering the surface of the tank and its surrounding area. Through the rational spatial arrangement of different types of sensors, the system can acquire information related to the leak location from multiple physical dimensions, such as gas concentration distribution, acoustic characteristics, structural vibration characteristics, and abnormal temperature distribution.

[0029] Hydrogen sensors can be densely deployed in a ring array at the bottom of the spherical tank, the surrounding ground, and in possible diffusion channels, forming a high-spatial-resolution hydrogen concentration monitoring network to reflect changes in hydrogen concentration gradients, thereby indicating the direction and extent of a potential leak. Acoustic and vibration sensors can be distributed across the surface of the spherical tank, supporting structures, pipeline connections, and valves, capturing directional acoustic signals generated during the leak and local structural vibration characteristics caused by the gas leak. Temperature sensors (such as DFTS or infrared thermal imagers) can cover the surface of the spherical tank and the surrounding area to monitor localized temperature anomalies caused by the leak, providing further spatial clues to the leak source location.

[0030] The core idea of ​​the hybrid sensor triangulation method based on particle filtering is to model the leak point localization problem as a state estimation problem within a Bayesian filtering framework, and then use particle filtering to estimate the location of the leak point in three-dimensional space. Specifically, a particle set is initialized within a pre-defined possible leak area, with each particle representing a potential leak point location and assigned a corresponding weight. As the multimodal sensor detection data is continuously updated, the particle filter dynamically adjusts the particle weights based on the degree of matching between the sensor's predicted readings and the actual detection data during iterative processes such as prediction, weight updates, and resampling. This causes high-weight particles to gradually converge towards the true leak point location, thereby achieving a progressively convergent estimation of the leak point location.

[0031] By incorporating multi-source information from hydrogen sensors, acoustic sensors, vibration sensors, and temperature sensors into the particle filtering localization process, the impact of single sensor uncertainty on the localization results can be effectively reduced, thereby improving the stability and accuracy of leak point localization under complex environmental conditions.

[0032] Initialize the particle set. Randomly generate N particles in the hypothetical potential leak area of ​​the spherical tank. The initial particle set should cover all areas of the spherical tank where leaks may occur, including surfaces, flanges, valves, pipe connections, etc. Each particle k contains its position p in three-dimensional space. k (t) = (x k (t),y k (t),z k (t) and a weight w k (t) (representing the probability that the particle is a real leak point), the initial weights can be set to equal 1 / N or estimated based on historical data of known leak points, and:

[0033] Based on the hydrogen diffusion model, predict the position pk(t+1) of each particle at the next time step. Assume that at time t, the position of particle k is pk(t) = (xk(t), yk(t), zk(t)). Add the drift caused by the environmental wind speed vector m(t) = (mx, my, mz) and the random diffusion term to predict the position at the next time step. , where Δt is the time step of particle filtering, which needs to be matched with the time resolution of data fusion; A random vector sampled from the standard multidimensional normal distribution N(0, I) is used to simulate the random diffusion motion of particles; Deff is the effective diffusion coefficient, the value of which can be estimated based on CFD simulations that take into account factors such as gas molecule diffusion coefficients and eddy diffusion, or by fitting experimental data.

[0034] To simulate the predicted readings of all sensors for each particle k, it is necessary to build separate sensor response models.

[0035] Hydrogen sensor model: Based on a Gaussian diffusion model or empirical model, considering the distance between the particle position and the sensor position, as well as the possible influence of wind direction and speed, according to the position p of particle k. k (t), calculate the value located at P sensor,H The hydrogen sensor readings. Assuming the leaked gas diffuses in a Gaussian distribution under the influence of wind, the model simplifies to:

[0036] Where u is the wind speed, (x, y, z) are the relative coordinates from the leak point to the sensor, and σ y , σ z is the diffusion coefficient (standard deviation), which increases with distance x and can be obtained based on a CFD diffusion model; H is the height of the leakage source; Q is the leakage source velocity, which is estimated by inversion based on a CFD model that considers factors such as gas molecule diffusion coefficient and eddy diffusion.

[0037] Acoustic sensor: Assuming the leak point is the sound source, based on the position P of particle k. k (t), calculate the value located at P sensor,A The sound intensity is measured by an acoustic sensor. A simplified model based on sound wave propagation attenuation is as follows:

[0038] Where S0 is the initial acoustic power of the leakage source, which is correlated with the leakage rate Q or estimated from historical data; r is the particle position P. k (t) to acoustic sensor P sensor,A The distance; β is the exponent of sound attenuation as it propagates through the air, an empirical parameter affected by frequency, temperature, and humidity.

[0039] Temperature sensor: Similarly, based on the position P of particle k k (t), calculate the value located at P sensor, The temperature change is measured by a temperature sensor at position T. The model simplifies to:

[0040] Among them, -T leak The maximum cooling rate at the leak point is negative; σ T It refers to the size of the cooling area; It is the particle position P k (t) to the temperature sensor P sensor,T The distance.

[0041] Update particle weights based on actual sensor readings C for hydrogen, acoustics, temperature, etc. actual And the sensor readings C predicted for each particle obtained through calculation. pred We calculate the likelihood term of the particles to update their weights.

[0042] Likelihood term for hydrogen sensor:

[0043] Where, σ H It is the standard deviation of the measurement noise of the hydrogen sensor i, which is monitored in real time and estimated by the sensor in real time; N H This is the total number of hydrogen sensors.

[0044] Likelihood term for acoustic sensor:

[0045] Where, σ A N is the standard deviation of the measurement noise of acoustic sensor i, which is monitored in real time and estimated by the sensor in real time; A This is the total number of acoustic sensors.

[0046] Temperature sensor likelihood term:

[0047] Where, σ T N is the standard deviation of the measurement noise of the temperature sensor i, which is monitored in real time and estimated by the sensor in real time; T This is the total number of acoustic sensors.

[0048] total likelihood function The larger the likelihood function value, the closer the particle prediction is to the actual measurement value, and the higher the particle weight.

[0049] The weight update formula multiplies the particle's old weights by the likelihood function. .

[0050] To ensure that the sum of the weights of all particles is 1 for resampling, the weights are normalized to obtain the final weights w of the four sensors. i,更新 w i,更新 Take w A,更新 w T,更新 and w H,更新 .

[0051]

[0052] Resampling. To avoid particle weight degradation (i.e., a few particles have extremely high weights while most particles have weights approaching zero, causing the estimation to lose its representativeness), resampling is necessary. Based on the current particle weights, high-weight particles are replicated multiple times, while low-weight particles are discarded, generating a new set of particles. Specifically: 1. Calculate the current cumulative weight ; 2. Generate N random numbers a i Starting from the interval [0, 1 / N], the intervals are [0, 1 / N], [1 / N, 2 / N], ..., [N-1 / N, N]. 3. For each a i Find the first j such that a i Less than C j Copy it to the new particle set; 4. The particle weights after resampling are reset to 1 / N.

[0053] Leakage point location estimation. The estimated location P of the leak point. leak (t+1) is the weighted average of the positions of all particles, and the formula is:

[0054] Among them, w i,final (t+1) is the particle weight after resampling, p k (t+1) is the predicted position of the particle.

[0055] If the stopping condition is not met, return and re-enter the position prediction, likelihood calculation, weight update, and resampling process at time t+1 until the following stopping condition is met: Location convergence: When the location of the leak point changes very little after multiple consecutive estimations.

[0056] Confidence level assessment: When the particle set converges to a very small region and the total weight of the particles in that region is very high, the localization result is considered to have high confidence.

[0057] Time limit: The preset maximum positioning time must be reached.

[0058] In one embodiment of the present invention, the multimodal sensor network further includes a vibration sensor.

[0059] Figure 2 This is a flowchart of the risk detection process according to an embodiment of the present invention, such as... Figure 2 As shown, in one embodiment of the present invention, the method for detecting leakage in a hydrogen storage spherical tank further includes steps S201 to S105.

[0060] Step S201: Preprocess the detection data collected by the multimodal sensor network, and establish a normal operation baseline based on the detection data collected by the target hydrogen storage sphere under normal and stable operating conditions. The normal operation baseline includes the statistics of each sensor under normal operating conditions.

[0061] Step S202: Extract corresponding feature data from the detection data of different types of sensors, and normalize the feature data.

[0062] Step S203: Determine the initial weights of various sensors, and dynamically update the initial weights based on the confidence level of each type of sensor to obtain the dynamic weights of each type of sensor. The confidence level is used to characterize the stability and anomaly degree of the detection data of the corresponding sensor, and the anomaly degree is characterized at least by the degree of deviation of the current detection data from the normal operating baseline.

[0063] Step S204: Based on the dynamic weights, the normalized feature data is weighted and fused to obtain fused feature values.

[0064] Step S205: Based on the normal operation baseline, set preset thresholds corresponding to each risk level, and determine the risk level of the target hydrogen storage tank according to the relationship between the fusion feature value and the preset thresholds.

[0065] In one embodiment of the present invention, the preprocessing of the detection data collected by the multimodal sensor network in step S201 includes: The detection data from different sensors are timestamped and resampled under different sampling rate scenarios. Wavelet transform is used to denoise the detection data from the acoustic sensor. The detection data from the vibration sensor is filtered and then synchronized by combining cross-correlation. The temperature sensor's detection data is smoothed and filtered to smooth the temperature field. The detection data from the hydrogen sensor is filtered to highlight the characteristics of the rate of change in hydrogen concentration.

[0066] In one embodiment of the present invention, step S202 above, which involves extracting corresponding feature data from the detection data of different types of sensors, includes: Extract signal energy characteristics within a preset time window from the detection data of the acoustic sensor; Extracting root mean square (RMS) value features from the detection data of vibration sensors; Extract the difference between the measured temperature and the average normal operating temperature from the temperature sensor's detection data; Extract the hydrogen concentration change rate feature from the detection data of the hydrogen sensor.

[0067] This invention employs multimodal sensor fusion in data detection. The basic principle of multimodal sensor fusion is to uniformly process and comprehensively analyze detection data from a multimodal sensor network with different physical mechanisms, spatial locations, and operating characteristics to obtain more comprehensive, stable, and reliable leak detection information compared to a single sensor. In the scenario of hydrogen storage tank leak detection, the hydrogen leakage process is affected by various factors such as environmental wind fields, temperature changes, and equipment operating status. Single-type sensors are easily affected by noise interference or have monitoring blind spots. By fusing multi-source detection data, the overall detection system's adaptability to complex operating conditions can be effectively improved.

[0068] Specifically, different types of sensors in a multimodal sensor network exhibit varying response mechanisms and sensitivity characteristics to leak events. For example, hydrogen sensors reflect hydrogen concentration and its changes, acoustic sensors detect acoustic anomalies caused by leaking gas flow, vibration sensors reflect abnormal vibrations in structures or pipelines due to leaks, and temperature sensors sense potential localized temperature changes during leaks. By extracting relevant feature data from the detection data of various sensors and combining this with sensor confidence levels for weighted fusion, the ability to identify genuine leak characteristics can be enhanced while suppressing misjudgments from single sensors, thereby improving the accuracy and stability of leak detection results.

[0069] Furthermore, by integrating multi-dimensional detection data and forming a unified fusion feature value, the system can comprehensively analyze the operating status of the target hydrogen storage tank and determine the corresponding risk level accordingly. Compared to relying on a single detection indicator, multi-modal data fusion can cross-validate leakage risks from multiple physical dimensions such as gas concentration changes, acoustic anomalies, vibration characteristics, and temperature changes, which helps reduce the probability of false alarms and false negatives, and achieves more reasonable risk assessment and graded early warning.

[0070] Furthermore, the results obtained from multimodal sensor fusion can provide high-confidence information support for subsequent leak point location, such as indicating the area where a leak may occur or the direction of risk concentration, thereby providing effective constraints for the leak point location process based on particle filtering and improving the efficiency and stability of the location calculation.

[0071] In this embodiment of the invention, the data fusion process of the hydrogen storage spherical tank leakage detection method based on a multimodal sensor network mainly includes stages such as detection data acquisition, data preprocessing, establishment of normal operating baseline, feature data extraction, sensor data fusion, and decision-making and result output. These stages cooperate to transform detection data from different types of sensors into fused information that can be used for risk assessment and leak location.

[0072] During the data acquisition phase, a multimodal sensor network deployed around the target hydrogen storage tank is used to collect detection data. The acoustic sensors in this network can be distributed high-sensitivity MEMS microphone arrays, spatially arranged to cover the entire circumference of the tank, as well as potentially leak-prone areas such as welds, valves, pipeline interfaces, and support structures. These arrays are used to collect acoustic signals generated by leaking gas flow. The monitoring frequency range of the acoustic signals can cover 20 Hz to 20 kHz, and can be further divided into low-frequency, mid-frequency, and high-frequency bands as needed to distinguish the acoustic characteristics under different leakage conditions, while simultaneously acquiring information such as spectral characteristics and short-time energy related to the gas flow sound.

[0073] Vibration sensors, such as high-precision piezoelectric or MEMS accelerometers, are deployed to cover the tank body, supporting structures, and pipeline connections to collect localized resonance or abnormal vibration signals caused by leaks. The frequency range of the vibration signals can cover everything from low-frequency structural response to high-frequency noise components, thereby obtaining vibration characteristics such as root mean square (RMS) values, peak factor, and energy spectrum distribution.

[0074] The temperature sensor can be a high-precision distributed fiber optic temperature sensor to continuously monitor the temperature field on the surface of the spherical tank. It is used to collect temperature change information, including the instantaneous rate of temperature change, the temperature difference with the surrounding environment, and the spatial distribution characteristics of the temperature field, thereby reflecting the local thermal anomalies that may be caused during the leakage process.

[0075] Hydrogen sensors can be of high sensitivity and fast response characteristics, and catalytic combustion, electrochemical, or semiconductor sensors can be selected according to different concentration detection requirements. Hydrogen sensors can be densely deployed in the potential leak area around the spherical tank to form a spatial monitoring network. Combined with environmental wind direction and diffusion direction, the hydrogen concentration in the environment surrounding the spherical tank can be monitored in real time, and the changes in hydrogen concentration over time can be recorded.

[0076] In the data preprocessing stage, to ensure the comparability and fusion of detection data from different types of sensors, the detection data needs to be processed uniformly. Specifically, the detection data collected by different sensors is first time-stamp aligned. Especially when the sensor sampling rates are inconsistent, the original detection data can be interpolated or downsampled to achieve synchronization in the time dimension.

[0077] For acoustic sensor detection data, wavelet transform can be used for noise reduction. This involves decomposing the acoustic signal into different frequency scales, thresholding the high-frequency wavelet coefficients, and then performing an inverse wavelet transform to obtain a noise-suppressed acoustic signal. For vibration sensor detection data, low-pass or high-pass filtering can be used to remove high-frequency instrument noise or low-frequency drift, combined with cross-correlation analysis to achieve signal synchronization and noise suppression. For temperature field data acquired by temperature sensors, two-dimensional Gaussian filtering or median filtering can be used to smooth the temperature field and remove isolated outliers. For hydrogen sensor detection data, moving least squares filtering can be used to smooth the signal while preserving the concentration change trend, thus highlighting the hydrogen concentration change rate characteristics.

[0078] In this embodiment of the invention, to characterize the operating state of the hydrogen storage tank under normal operating conditions, after data preprocessing, a normal operating baseline needs to be established based on the detection data collected under normal and stable operating conditions of the hydrogen storage tank. Specifically, when the hydrogen storage tank is in normal operating condition, continuous and long-term data collection can be performed on various sensors in the multimodal sensor network, for example, continuous collection for no less than 24 hours, to cover diurnal temperature variations and possible slight fluctuations in operating load. Based on this, statistical analysis is performed on the detection data of each sensor under normal operating conditions to calculate the corresponding average, standard deviation, root mean square deviation, and other statistical quantities, which are used to characterize the typical behavioral characteristics of the sensors under normal operating conditions.

[0079] For the temperature data collected by the temperature sensor, a model of the surface temperature field of the spherical tank can be further established based on historical data under normal operating conditions. For example, a three-dimensional Gaussian process regression model or other machine learning models can be used to predict the temperature at any location on the surface of the spherical tank under normal operating conditions, so as to improve the ability to identify local temperature anomalies.

[0080] Considering that environmental conditions and equipment status may change slowly, the normal operating baseline is not absolutely fixed. In practical applications, methods such as exponentially weighted moving averages can be used to adaptively update the baseline, allowing it to smoothly adjust with long-term changes in operating conditions. Based on the established normal operating baseline and its statistical characteristics, corresponding judgment thresholds can be set for different types of sensors and different risk levels, providing a reference for subsequent risk assessments.

[0081] After establishing a normal operating baseline, feature data related to the leak event are extracted from the preprocessed sensor detection data. The extracted feature data can have multi-scale, temporal, and spatial correlations, and can be used to reflect the changes in different physical quantities when a leak occurs.

[0082] For the detection data of acoustic sensors, the preprocessed acoustic signal s a An adaptive time window Δt is selected in (t), and the acoustic signal energy A(t) within this time window Δt is calculated, with particular attention paid to the changes in acoustic energy in the high-frequency band. A significant increase in acoustic signal energy A(t) usually indicates the presence of airflow noise caused by possible gas leakage.

[0083]

[0084] For the detection data of vibration sensors, the root mean square value V(t) of the vibration signal can be calculated within the same or similar time window to reflect the abnormal vibration intensity changes that may occur in the structure or pipeline under leakage conditions.

[0085]

[0086] in, (τ) is the preprocessed vibration signal.

[0087] For the detection data of temperature sensors, the measured temperature T of each temperature sensor can be extracted. k The average temperature T under normal operating conditions avg The difference characteristic T(t) between them is used to reflect the temperature anomaly caused by gas expansion, phase change or changes in local heat exchange conditions during the leakage process.

[0088]

[0089] Where NT represents the number of temperature sensors.

[0090] For the detection data of hydrogen sensors, the rate of change characteristic H(t) of hydrogen concentration over time can be calculated to reflect the rapid change of hydrogen concentration over time, thereby improving the sensitivity to early leaks.

[0091]

[0092] Where C(t) is the hydrogen concentration measured by the hydrogen sensor at time t.

[0093] Through the above feature extraction steps, the raw detection data of different types of sensors are transformed into feature data that can be used for fusion and judgment.

[0094] After feature extraction, in order to eliminate the differences in dimensions and numerical ranges of feature data from different types of sensors, it is necessary to normalize the various feature data. Specifically, acoustic energy features, vibration root mean square features, temperature difference features, and hydrogen concentration change rate features can be mapped to a unified numerical range [0, 1] to ensure that the features of different sensors are comparable during the fusion process.

[0095] During normalization, statistical results obtained from long-term monitoring under normal operating conditions can be used to determine the minimum and maximum values ​​of various feature data under normal operating conditions, and then the current feature data can be normalized accordingly. This method effectively eliminates the influence of differences in the dimensions and magnitudes of different sensors on the fusion results. Normalization formula:

[0096] Where X(t) represents the feature information A(t), V(t), T(t), and H(t) extracted by the sensor, and Xmin and Xmax are the statistical minimum and maximum values ​​obtained by the four sensors under normal operating conditions after long-term monitoring.

[0097] After normalization, the normalized feature data is weighted and fused by combining the dynamic weights of various sensors to obtain a fused feature value that characterizes the current operating status of the hydrogen storage tank. This fused feature value comprehensively reflects information from multiple physical dimensions, including gas concentration, acoustic characteristics, vibration characteristics, and temperature characteristics, providing a unified data foundation for subsequent risk level assessment and leak detection decisions.

[0098] In this embodiment of the invention, to reflect the relative importance and reliability of different types of sensors in the detection of leaks in hydrogen storage tanks, it is necessary to determine the corresponding weights for each type of sensor in the multimodal sensor network. These weights include initial weights and dynamic weights that are dynamically adjusted based on the sensor status during operation.

[0099] The initial weights can be set based on factors such as the inherent sensitivity of the sensor type, the impact of environmental noise on the sensor measurement results, and the sensor's ability to indicate abnormal states in historical operating data. For example, a hydrogen sensor can directly reflect changes in the concentration of leaked gas and has high sensitivity to leak events, so its initial weight can be set relatively high; an acoustic sensor is easily affected by environmental noise, so its initial weight can be relatively low; if historical data shows that a certain type of sensor can provide significant and stable abnormal characteristics when a leak occurs, its initial weight can be increased accordingly. In one embodiment, initial weights of 0.4, 0.2, 0.2, and 0.2 can be set for the acoustic sensor, vibration sensor, temperature sensor, and hydrogen sensor, respectively, so that the sum of the initial weights is 1.

[0100] Considering factors such as changes in environmental conditions, sensor aging, and fluctuations in equipment operating status, the reliability of different sensors will change over time. Therefore, a sensor confidence level C is introduced. i (t) is used to dynamically adjust the initial weights. The confidence level C i (t) is used to characterize the stability and anomaly level of sensor detection data at the current moment. Specifically, the confidence level can be calculated based on the statistical characteristics of the sensor's detection data within a recent time window. The standard deviation of the sensor's detection data reflects data stability; a smaller standard deviation indicates more stable sensor readings and a lower likelihood of being affected by transient noise. The deviation of the current sensor detection data from the normal operating baseline reflects the degree of anomaly; a larger deviation indicates a higher degree of anomaly. The confidence level can be calculated using a sliding time window method, with its calculation period matching the time resolution of data fusion. When the sensor detection data is stable and deviates little from the normal operating baseline, the confidence level tends to be high; when the detection data fluctuates significantly or deviates significantly from the baseline, the confidence level decreases accordingly.

[0101]

[0102] in, It is the standard deviation of sensor i's recent readings. The smaller the standard deviation, the more stable the sensor readings are, the less likely they are to be affected by transient noise, and the higher the confidence level. This is the deviation between the current reading of sensor i and the long-term average value (baseline) during normal operation; a larger deviation indicates an anomaly. i The calculation of (t) can be performed using a sliding window method, and the calculation period should be matched with the time resolution of the data fusion. When the sensor readings stabilize ( Small) and not much deviation from the baseline ( (hours) C i (t) approaching 1 indicates high confidence; when sensor readings fluctuate greatly or deviate significantly from the baseline, C i (t) Decrease.

[0103] After obtaining the confidence levels of various sensors, the sensor weights are dynamically updated based on the initial weights and confidence levels. Dynamic weights are obtained by combining the initial weights of each sensor with their corresponding confidence levels and then normalizing the results, ensuring that the sum of all dynamic weights is one. This method allows sensors that are stable and reliable at the current moment to have higher weights in the data fusion process, while sensors that are subject to interference or significant anomalies have lower weights, thereby improving the robustness and adaptability of the fusion results. Dynamic weights of the four sensors. The calculation formula is:

[0104] After completing the dynamic weight calculation, the dynamic weights of various sensors are weighted and fused with the corresponding normalized feature data to obtain the feature value S(t) of the multimodal sensor fusion. The fused feature value S(t) comprehensively reflects the overall degree of anomaly of acoustic features, vibration features, temperature features, and hydrogen concentration change features at the current moment, and its value range is within the preset interval [0, 1].

[0105] Based on the relationship between the fused characteristic value and the preset threshold, the operating status of the target hydrogen storage tank is determined. When the fused characteristic value exceeds the first preset threshold, a level one warning can be triggered; when the fused characteristic value continues to be higher than the preset threshold, and the leak point location results indicate the existence of a high-probability leak area, or when some sensor readings are detected to significantly exceed safety limits, a higher-level alarm can be triggered. Different risk levels correspond to different threshold ranges, used to distinguish multi-level risk states from normal operation to emergency leak.

[0106] During the output phase, the system can output multi-dimensional information related to the current detection results, including the risk level corresponding to the target hydrogen storage tank, scoring information used to quantify the degree of anomaly, different sensors and their contribution to the current risk assessment, and potentially affected areas or equipment components inferred based on sensor trigger locations and leak point location results. Through the above decision-making and output mechanisms, an intuitive and reliable basis is provided for the safety monitoring and operation and maintenance decisions of hydrogen storage tanks.

[0107] In one embodiment of the present invention, when the fused feature value S(t) exceeds a preset threshold, a first-level warning is triggered; when S(t) remains above the preset threshold, and the leak point location result shows a high-probability leak area, or there are sensor readings far exceeding safety limits, a second-level alarm is triggered, and the following specific information is output: Risk level: Normal: S(t) <T1; Focus on: T1≤S(t) <T2; Low risk: T2 ≤ S(t) <T3; Medium risk: T3≤S(t) <T4; High risk: T4 ≤ S(t) <T5; Urgent: S(t)≥T5; Where Ti represents a set threshold, which can be adaptively adjusted based on expert experience, simulation results, or machine learning models, and T1 <Ti<T5。

[0108] Confidence / Anomaly Score: Quantifies the degree of anomaly in the current state.

[0109] Contribution analysis: Indicates which sensors and features led to the current risk assessment.

[0110] Potentially affected areas / equipment components: Based on the trigger sensor and location results, the approximate location of the leak is preliminarily determined.

[0111] like Figure 3 As shown, in one embodiment of the present invention, the method for detecting leakage in a hydrogen storage spherical tank further includes steps S301 and S302.

[0112] Step S301: When the risk level reaches the preset alarm level and the positioning result of the particle filter iterative update process shows a high-confidence leakage area, the suspected leakage point location is determined as the final leakage point location and output. The high-confidence leakage area is characterized by at least the particle set converging to a preset small area and the sum of particle weights in the small area reaching a preset threshold.

[0113] Step S302: When the risk level indicates low risk or no abnormality, suspend the particle filter positioning process or reduce the execution priority of the particle filter positioning process.

[0114] In this embodiment of the invention, the leak point location process is linked with the risk level result obtained by fusing multimodal sensor data for control, so as to improve the reliability of the location result and the overall system operating efficiency.

[0115] When the risk level output by the data fusion module reaches the preset alarm level, and the positioning result obtained based on the particle filter iterative update process shows the existence of a high-confidence leakage area, the particle filter positioning result is considered to have high reliability. At this time, the suspected leakage point location can be determined as the final leakage point location and output. The high-confidence leakage area can be characterized by conditions such as the particle set spatially converging to a preset small region and the particle weight ratio being relatively high within that region.

[0116] When the data fusion results indicate that the target hydrogen storage sphere is in a low-risk or normal state, the particle filter positioning process can be paused or its execution priority reduced, even if the particle filter positioning process is still in progress, in order to reduce unnecessary consumption of computing resources.

[0117] When there is an inconsistency between the particle filter localization result and the data fusion result, for example, the data fusion result indicates a high risk level, but the particle filter localization process fails to converge to a clear high-confidence leakage area, it can be determined that the uncertainty of the current localization result is high. In this case, it is necessary to further analyze and diagnose the multimodal sensor detection data, data fusion results, or particle filter model parameters to help determine the cause of the anomaly and improve the accuracy of subsequent localization and assessment.

[0118] As can be seen from the above embodiments, the hydrogen storage spherical tank leakage detection method of the present invention achieves at least the following beneficial effects: 1. This invention constructs a multimodal sensor network that includes hydrogen sensors, acoustic sensors, vibration sensors, and temperature sensors to monitor the operating status of hydrogen storage spherical tanks from multiple dimensions. It can obtain detection information related to leakage from different physical mechanism levels, which improves the comprehensiveness and robustness of leakage detection compared to single sensor detection methods.

[0119] 2. This invention preprocesses, extracts, and normalizes the detection data collected by multimodal sensors, and introduces a confidence-based dynamic weight update mechanism to perform weighted fusion of features from different sensors. This allows the fusion result to adaptively adjust as the stability and anomaly degree of the sensors change, thereby improving the reliability and stability of the risk assessment results.

[0120] 3. This invention establishes a normal operation baseline and combines the relationship between fusion feature values ​​and preset thresholds to classify and determine the operational risks of hydrogen storage spherical tanks, thereby achieving a quantitative assessment of leakage risks. This helps to reduce false alarms and missed alarms and improve the rationality of early warning judgments.

[0121] 4. Based on risk assessment, this invention introduces a leak point location method based on particle filtering, which transforms the leak point location problem into a state estimation problem. It uses multi-source detection information to iteratively optimize candidate leak points, thereby achieving accurate estimation of the leak point location.

[0122] 5. This invention links the risk level obtained by data fusion with the particle filter positioning result for control. In high-risk situations, the final leak point location is output, and in low-risk or no-abnormal situations, the execution priority of the positioning process is reduced, thereby improving the overall operating efficiency of the system while ensuring positioning accuracy.

[0123] 6. Through the synergistic combination of the above technical solutions, the present invention can accurately locate the leak point of the hydrogen storage tank and improve the safety monitoring level of hydrogen storage equipment.

[0124] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0125] Based on the same inventive concept, embodiments of the present invention also provide a hydrogen storage tank leakage detection device, which can be used to implement the hydrogen storage tank leakage detection method described in the above embodiments, as described in the following embodiments. Since the principle of the hydrogen storage tank leakage detection device in solving the problem is similar to that of the hydrogen storage tank leakage detection method, embodiments of the hydrogen storage tank leakage detection device can refer to embodiments of the hydrogen storage tank leakage detection method, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0126] Figure 4 This is a structural block diagram of the hydrogen storage spherical tank leakage detection device according to an embodiment of the present invention, as shown below. Figure 4 As shown, in one embodiment of the present invention, the hydrogen storage spherical tank leakage detection device of the present invention includes: The detection data acquisition unit 1 is used to acquire detection data from a multimodal sensor network deployed around the target hydrogen storage sphere, wherein the multimodal sensor network includes a hydrogen sensor, an acoustic sensor, and a temperature sensor. Particle set initialization unit 2 is used to initialize a particle set within a preset possible leakage area of ​​the target hydrogen storage spherical tank. Each particle in the particle set represents a candidate leakage point location and has a corresponding particle weight. The particle filtering iteration unit 3 is used to repeatedly execute the particle filtering iteration update process until the iteration termination condition is met. Based on the particle position and corresponding particle weight of each particle in the final iteration round, the particle position of each particle is weighted and averaged to obtain the suspected leak point position corresponding to the target hydrogen storage sphere. In each iteration round: first, the particle set obtained in the previous iteration is predicted based on the hydrogen diffusion model to obtain the predicted position of each particle; based on the predicted position of each particle, the sensor prediction reading corresponding to each particle is simulated using the hydrogen sensor response model, acoustic sensor response model and temperature sensor response model respectively; the particle likelihood of each particle is calculated based on the degree of matching between the sensor prediction reading and the detection data; the particle weight of each particle is updated according to the particle likelihood, and the updated particle weight is normalized; the particles are resampled based on the normalized particle weight to generate a new particle set for the next iteration round.

[0127] In one embodiment of the present invention, the multimodal sensor network further includes a vibration sensor; The hydrogen storage spherical tank leakage detection device also includes: The data preprocessing unit is used to preprocess the detection data collected by the multimodal sensor network and establish a normal operation baseline based on the detection data collected by the target hydrogen storage sphere tank under normal and stable operating conditions. The normal operation baseline includes the statistics of each sensor under normal operating conditions. The feature data extraction unit is used to extract corresponding feature data from the detection data of different types of sensors, and to normalize the feature data. A dynamic weight determination unit is used to determine the initial weights of various sensors and dynamically update the initial weights based on the confidence level of various sensors to obtain the dynamic weights of various sensors. The confidence level is used to characterize the stability and abnormality of the detection data of the corresponding sensor, and the abnormality is characterized at least by the degree of deviation of the current detection data from the normal operating baseline. The feature value determination unit is used to perform weighted fusion of the normalized feature data based on the dynamic weights to obtain the fused feature value. The risk level determination unit is used to set preset thresholds corresponding to each risk level based on the normal operation baseline, and to determine the risk level of the target hydrogen storage spherical tank according to the relationship between the fusion feature value and the preset threshold.

[0128] In one embodiment of the present invention, the hydrogen storage spherical tank leakage detection device of the present invention further includes: The leakage detection result output unit is used to determine the location of the suspected leakage point as the final leakage point location and output it when the risk level reaches the preset alarm level and the positioning result of the particle filter iterative update process shows a high-confidence leakage area. The high-confidence leakage area is characterized by at least the particle set converging to a preset small area and the sum of particle weights in the small area reaching a preset threshold. The filter positioning process unit is used to pause the particle filter positioning process or reduce the execution priority of the particle filter positioning process when the risk level indicates low risk or no abnormality.

[0129] In one embodiment of the present invention, the data preprocessing unit includes: The resampling processing module is used to align the timestamps of the detection data from different sensors and to resample the original sensor detection data under different sampling rate scenarios. The wavelet transform denoising module is used to perform wavelet transform denoising on the detection data of the acoustic sensor. The first filtering module is used to filter the detection data of the vibration sensor and perform synchronous processing in combination with cross-correlation. The second filtering module is used to smooth the detection data of the temperature sensor in order to smooth the temperature field. The third filtering module is used to filter the detection data from the hydrogen sensor to highlight the characteristics of the hydrogen concentration change rate.

[0130] In one embodiment of the present invention, the feature data extraction unit includes: The signal energy feature extraction module is used to extract signal energy features within a preset time window from the detection data of the acoustic sensor. The root mean square (RMS) feature extraction module is used to extract RMS features from the detection data of the vibration sensor. The temperature difference feature extraction module is used to extract the difference feature between the measured temperature and the average normal operating temperature from the detection data of the temperature sensor. The hydrogen concentration change rate feature extraction module is used to extract hydrogen concentration change rate features from the detection data of the hydrogen sensor.

[0131] In one embodiment of the present invention, the particle filter iteration unit 3 includes: The particle position prediction module is used to superimpose the drift effect caused by the environmental wind speed vector and the random diffusion effect on the current position of the particle to obtain the predicted position of each particle in the current iteration round. The random diffusion effect is characterized by a random vector sampled from a standard multidimensional normal distribution and an effective diffusion coefficient, and the time step of the particle filtering is matched with the time resolution of the data fusion.

[0132] In one embodiment of the present invention, the particle filter iteration unit 3 further includes: The particle weight update module is used to calculate the corresponding particle likelihood terms based on the actual detection data of the hydrogen sensor, acoustic sensor and temperature sensor and the corresponding sensor predicted readings, respectively; and multiply the particle likelihood terms to obtain the total particle likelihood function, and multiply the total particle likelihood function with the particle weight of the previous iteration to obtain the updated particle weight. In one embodiment of the present invention, the particle filter iteration unit 3 further includes: The particle resampling module is used to copy and eliminate particles according to their weights to form a new particle set, and reset the particle weights of the new particle set to equal weights after resampling. In one embodiment of the present invention, the particle filter iteration unit 3 further includes: The suspected leak point location determination module is used to perform a weighted average of the particle positions corresponding to each particle in the current iteration round using the resampled particle weights to obtain the suspected leak point location.

[0133] To achieve the above objectives, according to another aspect of this application, a computer device is also provided. For example... Figure 5 As shown, the computer device includes a memory, a processor, a communication interface, and a communication bus. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps in the method of the above embodiments.

[0134] The processor can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0135] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and units, such as the program units corresponding to the above-described method embodiments of the present invention. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the methods described in the above-described method embodiments.

[0136] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0137] The one or more units are stored in the memory and, when executed by the processor, perform the methods described in the above embodiments.

[0138] The specific details of the aforementioned computer equipment can be understood by referring to the relevant descriptions and effects in the above embodiments, and will not be repeated here.

[0139] To achieve the above objectives, according to another aspect of this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed in a computer processor, implements the steps in the above-described hydrogen storage tank leakage detection method. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0140] To achieve the above objectives, according to another aspect of this application, a computer program product is also provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described hydrogen storage tank leakage detection method.

[0141] Obviously, those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0142] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting leaks in a hydrogen storage spherical tank, characterized in that, include: Acquire detection data from a multimodal sensor network deployed around the target hydrogen storage sphere, the multimodal sensor network including hydrogen sensors, acoustic sensors, and temperature sensors; A particle set is initialized within a preset possible leakage area of ​​the target hydrogen storage spherical tank. Each particle in the particle set represents a candidate leakage point location and has a corresponding particle weight. The particle filtering iterative update process is executed repeatedly until the iteration termination condition is met. Based on the particle position and corresponding particle weight of each particle in the final iteration round, the particle positions of each particle are weighted and averaged to obtain the suspected leak point location corresponding to the target hydrogen storage tank. In each iteration round: first, the particle set obtained in the previous iteration is predicted based on the hydrogen diffusion model to obtain the predicted position of each particle; based on the predicted position of each particle, the sensor prediction readings corresponding to each particle are simulated using the hydrogen sensor response model, acoustic sensor response model, and temperature sensor response model, respectively; the particle likelihood of each particle is calculated based on the degree of matching between the sensor prediction readings and the detection data; the particle weights of each particle are updated according to the particle likelihood, and the updated particle weights are normalized; the particles are resampled based on the normalized particle weights to generate a new particle set for the next iteration round.

2. The method for detecting leaks in a hydrogen storage spherical tank according to claim 1, characterized in that, The multimodal sensor network also includes vibration sensors; The method further includes: The detection data collected by the multimodal sensor network is preprocessed, and a normal operation baseline is established based on the detection data collected by the target hydrogen storage sphere under normal and stable operation. The normal operation baseline includes the statistics of each sensor under normal operating conditions. The corresponding feature data are extracted from the detection data of different types of sensors, and the feature data is normalized. The initial weights of various sensors are determined, and the initial weights are dynamically updated based on the confidence level of each sensor to obtain the dynamic weights of each sensor. The confidence level is used to characterize the stability and anomaly of the detection data of the corresponding sensor, and the anomaly is characterized at least by the degree of deviation of the current detection data from the normal operating baseline. Based on the dynamic weights, the normalized feature data is weighted and fused to obtain fused feature values; Based on the normal operation baseline, preset thresholds are set for each risk level, and the risk level of the target hydrogen storage spherical tank is determined according to the relationship between the fusion feature value and the preset threshold.

3. The method for detecting leaks in a hydrogen storage spherical tank according to claim 2, characterized in that, The method further includes: When the risk level reaches the preset alarm level, and the positioning result of the particle filter iterative update process shows a high-confidence leakage area, the suspected leakage point location is determined as the final leakage point location and output. The high-confidence leakage area is characterized by at least the particle set converging to a preset small area and the total particle weight in the small area reaching a preset threshold. When the risk level indicates low risk or no abnormality, the particle filter localization process is paused or its execution priority is reduced.

4. The method for detecting leaks in a hydrogen storage spherical tank according to claim 2, characterized in that, The hydrogen sensors are arranged in a ring at the bottom of the spherical tank, the surrounding ground, and the diffusion channel; the acoustic sensors and the vibration sensors are arranged on the surface of the spherical tank, the support structure, the pipeline connections, and the valves; the temperature sensors are arranged on the surface of the spherical tank and the surrounding area.

5. The method for detecting leaks in a hydrogen storage spherical tank according to claim 2, characterized in that, The preprocessing of the detection data acquired by the multimodal sensor network includes: The detection data from different sensors are timestamped and resampled under different sampling rate scenarios. Wavelet transform is used to denoise the detection data from the acoustic sensor. The detection data from the vibration sensor is filtered and then synchronized by combining cross-correlation. The temperature sensor's detection data is smoothed and filtered to smooth the temperature field. The detection data from the hydrogen sensor is filtered to highlight the characteristics of the rate of change in hydrogen concentration.

6. The method for detecting leaks in a hydrogen storage spherical tank according to claim 2, characterized in that, The extraction of corresponding feature data from detection data of different types of sensors includes: Extract signal energy characteristics within a preset time window from the detection data of the acoustic sensor; Extracting root mean square (RMS) value features from the detection data of vibration sensors; Extract the difference between the measured temperature and the average normal operating temperature from the temperature sensor's detection data; Extract the hydrogen concentration change rate feature from the detection data of the hydrogen sensor.

7. The method for detecting leaks in a hydrogen storage spherical tank according to claim 1, characterized in that, The prediction of the particle set obtained from the previous iteration based on the hydrogen diffusion model, to obtain the predicted position of each particle, includes: Based on the current position of the particle, the drift effect caused by the environmental wind speed vector and the random diffusion effect are superimposed to obtain the predicted position of each particle in the current iteration round. The random diffusion effect is characterized by a random vector sampled from a standard multidimensional normal distribution and an effective diffusion coefficient. The time step of the particle filtering is matched with the time resolution of the data fusion. The step of calculating the particle likelihood of each particle based on the degree of matching between the sensor's predicted reading and the detection data, and updating the particle weight of each particle based on the particle likelihood, includes: The particle likelihood terms are calculated based on the actual detection data of the hydrogen sensor, acoustic sensor, and temperature sensor and the corresponding predicted readings of the sensors, respectively. The particle likelihood terms are multiplied together to obtain the total particle likelihood function. The total particle likelihood function is then multiplied by the particle weights from the previous iteration to obtain the updated particle weights. The resampling of particles based on normalized particle weights to generate a new particle set for the next iteration includes: Based on particle weights, particles are copied and eliminated to form a new particle set, and after resampling, the particle weights of the new particle set are reset to equal weights. The step of weighted averaging the particle positions of each particle to obtain the location of the suspected leak point corresponding to the target hydrogen storage tank includes: The suspected leak point location is obtained by weighting the particle positions of each particle in the current iteration round with the resampled particle weights.

8. A leak detection device for a hydrogen storage spherical tank, characterized in that, include: The detection data acquisition unit is used to acquire detection data from a multimodal sensor network deployed around the target hydrogen storage sphere, wherein the multimodal sensor network includes a hydrogen sensor, an acoustic sensor, and a temperature sensor. The particle set initialization unit is used to initialize a particle set within a preset possible leakage area of ​​the target hydrogen storage spherical tank. Each particle in the particle set represents a candidate leakage point location and has a corresponding particle weight. The particle filtering iteration unit is used to repeatedly execute the particle filtering iteration update process until the iteration termination condition is met. Based on the particle position and corresponding particle weight of each particle in the final iteration round, the particle positions of each particle are weighted and averaged to obtain the suspected leak point location corresponding to the target hydrogen storage sphere. In each iteration round: first, the particle set obtained in the previous iteration is predicted based on the hydrogen diffusion model to obtain the predicted position of each particle; based on the predicted position of each particle, the sensor prediction readings corresponding to each particle are simulated using the hydrogen sensor response model, acoustic sensor response model, and temperature sensor response model, respectively; the particle likelihood of each particle is calculated based on the degree of matching between the sensor prediction readings and the detection data; the particle weights of each particle are updated according to the particle likelihood, and the updated particle weights are normalized; the particles are resampled based on the normalized particle weights to generate a new particle set for the next iteration round.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images