False alarm feature library-based gas alarm method, device and medium

Abstract

The application provides a gas alarm method and device based on a false alarm feature library, and a medium. First, multi-source gas monitoring data is acquired. Then, according to the multi-source gas monitoring data, a target feature vector of gas monitoring is extracted. Finally, according to the target feature vector and a false alarm feature library established in advance, a gas monitoring result is determined. The false alarm feature library is used to record feature changes caused by non-gas substances. By using the false alarm feature library, the application records feature information of various non-gas substances, realizes accurate differentiation between real gas leakage and non-gas interference, avoids invalid alarms caused by non-gas factors such as oil smoke, alcohol evaporation and environmental interference, and improves alarm reliability.

Description

Technical Field

[0001] This invention relates to the field of gas monitoring technology, and in particular to a gas alarm method, device and medium based on a false alarm feature library. Background Technology

[0002] In residential and commercial gas safety monitoring scenarios, gas-sensitive resistor sensors and infrared spectral sensors are commonly used to achieve real-time detection of gas concentration. Gas-sensitive resistor sensors are low-cost and have a fast response speed, but they are broad-spectrum sensitive detection devices. They not only respond to combustible gases such as methane, but are also highly susceptible to interference from non-gas substances such as kitchen fumes, alcohol evaporation, disinfectant vapors, and high-temperature humidity, easily triggering a large number of invalid false alarms. Infrared gas sensors have specific methane spectral recognition capabilities and strong resistance to single interferences, but they are greatly affected by the installation space environment, ventilation conditions, spatial obstruction, and multi-point monitoring layout. When used alone, they are difficult to adapt to complex indoor space multi-condition monitoring scenarios.

[0003] Current conventional gas alarm solutions mostly use a single sensor threshold comparison method and rely solely on concentration thresholds to trigger alarms. This method cannot distinguish between actual gas leaks and sensor signal fluctuations caused by non-gas substances such as fumes and alcohol. It cannot identify interference sources at the feature level, leading to frequent false alarms and reducing users' trust in alarm devices. Summary of the Invention

[0004] This invention provides a gas alarm method, device, and medium based on a false alarm feature library, addressing the problem that existing technologies rely on a single sensor threshold comparison method and depend solely on concentration thresholds to trigger alarms, leading to frequent false alarms.

[0005] A first aspect of the present invention provides a gas alarm method based on a false alarm feature library, comprising: Acquire multi-source gas monitoring data; Based on multi-source gas monitoring data, the target feature vector of gas monitoring is extracted; The gas monitoring results are determined based on the target feature vector and the pre-established false alarm feature library; The false alarm feature library is used to record feature changes caused by non-gas substances.

[0006] In one possible implementation, the multi-source gas monitoring data includes resistive gas monitoring data and infrared gas monitoring data; based on the multi-source gas monitoring data, a target feature vector for gas monitoring is extracted, including: Feature analysis was performed on the resistive gas monitoring data to obtain the change amplitude of the gas-sensitive resistor, the rise time response slope, the fluctuation variance, and the recovery descent coefficient. A subset of resistance features is generated based on the change amplitude, rising edge response slope, fluctuation variance, and recovery descent coefficient of the gas-sensitive resistor. Infrared gas monitoring data is analyzed to obtain infrared characteristic absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability. Infrared feature subsets are generated based on infrared feature absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability. The target feature vector is determined based on the resistive feature subset and the infrared feature subset.

[0007] In one possible implementation, the target feature vector is determined based on a subset of resistive features and a subset of infrared features, including: A unified mapping relationship is established using the standard feature dimensions of the false alarm feature library as the mapping target; The resistance feature subset is evaluated to obtain the first evaluation result, and the infrared feature subset is evaluated to obtain the second evaluation result. Based on the first and second evaluation results, adjust the mapping coefficients of the unified mapping relationship; The target feature vector is obtained by jointly mapping the resistive feature subset and the infrared feature subset according to the unified mapping relationship.

[0008] In one possible implementation, the gas monitoring result is determined based on the target feature vector and a pre-established false alarm feature library, including: The target feature vector is matched with each false alarm feature vector in the pre-established false alarm feature library to obtain multiple matching results; The gas monitoring results are determined based on multiple matching results.

[0009] In one possible implementation, the gas monitoring result is determined based on multiple matching results, including: A time decay strategy is used to decay and correct the confidence of the matching results, and a spatial decay strategy is used to decay and correct the weight of the matching results. Based on the corrected matching results, the gas monitoring results are determined.

[0010] In one possible implementation, a time decay strategy is used to attenuate and correct the confidence level of the matching results, including: Record the entry time of each false alarm feature vector and calculate the time difference between the current time and the entry time; The confidence level of the matching results is adjusted based on the difference and the time decay function. The time decay function is determined by the effective period, basic decay coefficient, and maximum decay amplitude of each false alarm feature vector.

[0011] In one possible implementation, a spatial decay strategy is used to attenuate and correct the weights of the matching results, including: Acquire spatial features and sensor installation locations; Based on spatial characteristics, determine the interference propagation coefficient for each spatial point; The weights of the matching results are attenuated based on the interference propagation coefficient of each spatial point and the sensor installation location.

[0012] In one possible implementation, before determining the gas monitoring result based on the target feature vector and a pre-established false alarm feature library, the method further includes: Acquire monitoring data on interference generated by various non-flammable substances; Extract interference feature vectors from the interference monitoring data that correspond to the resistive monitoring data and the infrared monitoring data; The extracted interference feature vectors are standardized, encoded, and stored to form a false alarm feature library.

[0013] A second aspect of the present invention provides a gas alarm terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the gas alarm method based on a false alarm feature library as described in the first aspect above.

[0014] A third aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a controller, implements the gas alarm method based on a false alarm feature library as described in the first aspect above.

[0015] Compared to traditional technologies, this invention provides a gas alarm method, device, and medium based on a false alarm feature library. First, multi-source gas monitoring data is acquired. Then, based on the multi-source gas monitoring data, a target feature vector for gas monitoring is extracted. Finally, the gas monitoring result is determined based on the target feature vector and a pre-established false alarm feature library. The false alarm feature library records characteristic changes caused by non-gas substances. This invention, by utilizing the false alarm feature library to record the characteristic information of various non-gas substances, achieves accurate differentiation between actual gas leaks and non-gas interference, avoiding invalid alarms caused by non-gas factors such as oil fumes, alcohol evaporation, and environmental interference, thus improving alarm reliability. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the implementation of the gas alarm method based on a false alarm feature library provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of the gas alarm terminal provided in an embodiment of the present invention. Detailed Implementation

[0017] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart illustrating the implementation of the gas alarm method based on a false alarm feature library provided in this embodiment of the invention. Figure 1 As shown, the gas alarm method based on a false alarm feature library includes: S110, acquire multi-source gas monitoring data; S120, Based on multi-source gas monitoring data, extract the target feature vector for gas monitoring; S130, determine the gas monitoring result based on the target feature vector and the pre-established false alarm feature library; The false alarm feature library is used to record feature changes caused by non-gas substances.

[0019] This invention relates to a gas alarm method based on a false alarm feature library, which is suitable for various enclosed, semi-enclosed, and open space scenarios where gas pipelines and gas equipment are present and where false alarms are easily caused by interference from non-gas substances.

[0020] Residential scenarios mainly include enclosed kitchens, open-plan kitchen and dining areas, balconies with gas water heaters, partitions for wall-mounted boilers and gas heating equipment, residential gas pipeline shafts, and basement storage spaces. These scenarios are prone to generating non-gas-related interference substances such as kitchen fumes, cooking steam, alcohol evaporation, and odors from cleaning agents and disinfectants, making traditional gas alarms prone to frequent false alarms. Commercial scenarios include restaurant kitchens, school and company canteen kitchens, hotel and guesthouse gas equipment rooms, public gas pipeline areas in concentrated restaurant districts, and supermarket gas equipment rooms. These scenarios have complex spatial structures, varying ventilation conditions, and accumulate a lot of oil fumes and volatile interference substances, requiring the deployment of monitoring equipment at multiple locations.

[0021] In this embodiment of the invention, the detection device is a gas monitoring terminal integrating dual sensor probes. The device integrates a resistive gas sensor and an infrared gas sensor, and can simultaneously collect two multi-source gas monitoring data. The hardware structure supports multiple installation methods such as wall-mounted, ceiling-mounted, and pipe-embedded, adapting to different spatial layouts and sensor installation location requirements. The terminal has a built-in microprocessor controller, which is responsible for completing multi-source data acquisition, feature parsing, feature subset generation, target feature vector construction, and matching and judgment with the local false alarm feature library.

[0022] In some embodiments, the multi-source gas monitoring data includes resistive gas monitoring data and infrared gas monitoring data. Based on the multi-source gas monitoring data, a target feature vector for gas monitoring is extracted, including: performing feature analysis on the resistive gas monitoring data to obtain the amplitude of the gas-sensitive resistor change, the rising edge response slope, the fluctuation variance, and the recovery slack coefficient; generating a resistance feature subset based on the gas-sensitive resistor change amplitude, rising edge response slope, fluctuation variance, and recovery slack coefficient; performing feature analysis on the infrared gas monitoring data to obtain the infrared characteristic absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability; generating an infrared feature subset based on the infrared characteristic absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability; and determining the target feature vector based on the resistance feature subset and the infrared feature subset.

[0023] In this embodiment of the invention, the collected resistive gas monitoring data needs to be subjected to targeted feature analysis processing. The focus is on mining the signal change patterns in the data that can characterize the effects of gas and various non-gas interference substances. Finally, four core feature parameters are extracted, namely the change amplitude of the gas-sensitive resistor, the rise edge response slope, the fluctuation variance, and the recovery descent coefficient.

[0024] Among them, the amplitude of the gas-sensitive resistor change directly reflects the change in its resistance value relative to the initial reference value after contact with flammable or non-flammable interfering substances. The magnitude of the change amplitude is directly related to the concentration and properties of the interfering substance, and can preliminarily distinguish the strength of the substance's effect. The rising edge response slope describes the rate of change of the gas-sensitive resistor's resistance value from the initial stable value to the new stable value after contact with the target substance. Different types of substances (such as flammable gas, cooking fumes, and alcohol) cause significant differences in response rates, which is an important characteristic for distinguishing substance types. The fluctuation variance measures the degree of fluctuation and dispersion of the resistance signal after reaching a stable state. Generally, the resistance signal fluctuation caused by non-flammable interfering substances is more violent, and the variance value is larger, while the signal fluctuation caused by actual gas leaks is relatively gentle, and the variance value is smaller. The recovery descent coefficient characterizes the rate at which the resistance value of the gas-sensitive resistor slowly recovers from the stable state to the initial reference value after being removed from the target substance. The recovery rate varies significantly for different substances.

[0025] Based on the preset feature importance levels in the false alarm feature library, initial weights were assigned to the four effective feature parameters. Among them, the amplitude of the gas sensor change and the fluctuation variance are directly related to the substance concentration and interference intensity, and have the highest weight (30% each). The rising edge response slope and the recovery descent coefficient focus on assisting in the identification of substance types, and have a relatively lower weight (20% each). During the weight allocation process, the interference level of the current monitoring environment was taken into account (e.g., in a kitchen scene with a high interference level, the weight of the fluctuation variance was appropriately increased) to ensure the targeted nature of the weight allocation. Subsequently, the min-max normalization algorithm was used to standardize the four weighted feature parameters, uniformly mapping the values ​​of all parameters to the 0-1 range.

[0026] Following a preset feature sorting rule (pre-defined as gas-sensitive resistor change amplitude → rising edge response slope → fluctuation variance → recovery slack coefficient), the four standardized feature parameters are arranged in an orderly manner. This sorting rule strictly corresponds to the resistance feature dimension order in the false alarm feature library, ensuring consistency in the feature order of each generated resistance feature subset. After sorting, the four core feature parameters—gas-sensitive resistor change amplitude, rising edge response slope, fluctuation variance, and recovery slack coefficient—are arranged sequentially according to the aforementioned fixed sorting rule, forming an initial feature sequence. Subsequently, based on a preset feature weight allocation rule, corresponding weights are assigned to each of the four feature parameters: gas-sensitive resistor change amplitude and fluctuation variance are assigned 30% weight each as core features, while rising edge response slope and recovery slack coefficient are assigned 20% weight each as auxiliary features. This highlights the differentiated importance of different features and avoids judgment bias caused by a single dominant feature. On this basis, the sorted feature sequence is standardized and encapsulated, integrating the values ​​of each feature parameter, their corresponding weights, normalization parameters, and other information to form a complete resistance feature subset.

[0027] For the infrared gas monitoring data collected simultaneously, key parameters that can reflect the spectral characteristics of gas and the influence of environmental interference are extracted, namely infrared characteristic absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability.

[0028] Among them, infrared characteristic absorption intensity is used to characterize the intensity of infrared light being absorbed by gas after passing through the monitoring space. The absorption intensity is positively correlated with the gas concentration and is the core spectral feature for determining whether there is a gas leak in the monitoring space. The light intensity attenuation coefficient is used to describe the degree of light intensity attenuation after infrared light is affected by interference factors such as dust, water vapor, and oil fumes in the space during propagation. It can effectively quantify the degree of influence of environmental interference on infrared monitoring signals. The absorption peak bandwidth shift is used to characterize the bandwidth of the gas characteristic absorption peak detected by the infrared sensor. The shift relative to the standard gas absorption peak bandwidth will not cause a significant shift in this characteristic peak. Only real gas leaks will cause a fixed and regular shift in the absorption peak bandwidth. It is a key spectral feature for distinguishing between gas and non-gas interference. The signal-to-noise ratio stability is used to measure the stability of the ratio of effective spectral signal to noise signal in the infrared monitoring signal. The signal-to-noise ratio of infrared signals caused by real gas leaks is relatively stable with small fluctuations, while the signal-to-noise ratio of signals caused by non-gas interference fluctuates violently and has poor stability.

[0029] Based on the preset infrared feature standards in the false alarm feature library, infrared monitoring data is analyzed in a targeted manner to extract infrared feature parameters (infrared absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability). This is then used to generate an infrared feature subset. By adhering to the standard requirements of infrared features in the false alarm feature library, the feature parameters of the infrared monitoring data are analyzed to clarify the core characteristics of infrared monitoring data. Compared to the broad-spectrum response of resistive sensors, infrared monitoring data exhibits extremely high specificity, responding specifically only to target combustion gases such as methane, and showing very low responsiveness to non-combustion gas interference such as oil fumes and steam. This characteristic determines that the extraction logic of infrared feature parameters differs fundamentally from that of resistive sensor data.

[0030] By combining the infrared feature standards of the false alarm feature library, infrared monitoring data is accurately analyzed to extract four core feature parameters: infrared absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability. These parameters are used to construct an infrared feature subset. Based on the standard dimensions of infrared features in the false alarm feature library, these four core parameters are standardized to ensure that each feature parameter accurately corresponds to the characteristic differences between non-gas interference and actual gas leaks, providing standardized support for the subsequent construction of target feature vectors.

[0031] Based on the infrared feature standards of the false alarm feature library, the extracted infrared monitoring data features are weighted and standardized. Infrared absorption intensity and absorption peak bandwidth offset are designated as core features, each with a weight of 35%. Light intensity attenuation coefficient and signal-to-noise ratio stability are designated as auxiliary features, each with a weight of 15%. The weights of each feature are dynamically adjusted based on the ventilation conditions and dust concentration of the current monitoring environment to ensure the accuracy and discriminative power of the infrared features. During the generation of the infrared feature subset, the infrared dimension requirements of the false alarm feature library are strictly followed. Instead of simple parameter splicing, the infrared monitoring data is first denoised to remove invalid interference signals. Then, the four core parameters—infrared absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth offset, and signal-to-noise ratio stability—are standardized individually to ensure that each parameter accurately reflects the true state of infrared monitoring and avoids confusion with resistive monitoring data. Finally, according to the infrared standard dimensions of the false alarm feature library, the standardized infrared feature parameters are arranged in an orderly manner and combined with the weighting rules to form a complete infrared feature subset.

[0032] In some embodiments, determining the target feature vector based on the resistance feature subset and the infrared feature subset includes: establishing a unified mapping relationship with the standard feature dimension of the false alarm feature library as the mapping target; evaluating the resistance feature subset to obtain a first evaluation result, and simultaneously evaluating the infrared feature subset to obtain a second evaluation result; adjusting the mapping coefficients of the unified mapping relationship based on the first evaluation result and the second evaluation result; and jointly mapping the resistance feature subset and the infrared feature subset according to the unified mapping relationship to obtain the target feature vector.

[0033] In this embodiment of the invention, the standard feature dimension of the false alarm feature library is an 8-dimensional standardized feature space with a fixed structure, fixed order, and fixed physical meaning. The dimensional order and physical meaning strictly correspond to the predefined rules of the two feature subsets. Specifically, the first to fourth dimensions correspond to resistance features, with a fixed order of "gas-sensitive resistor change amplitude → rising edge response slope → fluctuation variance → recovery slack coefficient"; the fifth to eighth dimensions correspond to infrared features, with a fixed order of "infrared feature absorption intensity → light intensity attenuation coefficient → absorption peak bandwidth offset → signal-to-noise ratio stability". The numerical range of all dimensions is uniformly 0-1, which is completely consistent with the normalized range of the two feature subsets, ensuring that the feature semantics do not shift before and after mapping.

[0034] Based on the above standard dimensions, a structured initial unified mapping relationship with two inputs and one output is established. The core rules and composition of this mapping relationship are as follows: The mapping input consists of two sets of structured data: one is a subset of resistive features (a 4-dimensional ordered, weighted normalized feature sequence), and the other is a subset of infrared features (a 4-dimensional ordered, weighted normalized feature sequence). The mapping output is an 8-dimensional target feature vector that is perfectly aligned with the standard dimensions of the false positive feature library; The mapping relationship has two types of adjustable coefficients, both with an initial value of 1: one type is the subset-level global mapping coefficient, which is divided into the resistance subset global coefficient α and the infrared subset global coefficient β, used to adjust the contribution of the two subsets as a whole to the final vector; the other type is the dimension-level mapping coefficient, which has 8 (k1-k8) and corresponds one-to-one with the standard 8-dimensional features, used to adjust the contribution of a single feature dimension to the final vector. The core constraints of the initial mapping are: strict alignment of dimensions; the four features of the resistance feature subset are mapped to the first to fourth positions of the standard dimension in their original fixed order; the four features of the infrared feature subset are mapped to the fifth to eighth positions of the standard dimension in their original fixed order, without disrupting the original feature order or merging independent feature dimensions, thus fully preserving the feature information of both subsets.

[0035] After establishing the initial unified mapping relationship based on the standard feature dimensions of the false alarm feature library, a full-dimensional quantitative evaluation of the resistance feature subset and the infrared feature subset was carried out simultaneously. The evaluation of the resistance feature subset reused its built-in 4-dimensional ordered normalized feature values, the exclusive weights bound to each dimension, fixed feature sorting rules, and normalization attributes throughout the process. Three core evaluations were conducted sequentially: feature validity, feature reliability, and weight suitability. Feature validity was verified to ensure that all four feature values ​​within the subset were within the 0-1 normalization range, that the sum of the four dimension weights was 100%, and that the feature sorting was perfectly aligned with the first to fourth digits of the standard dimensions of the false alarm feature library. No anomalies resulted in full marks for this item; problems were penalized according to their severity. Feature reliability was based on the four dimensions within the subset. The system verifies whether the change amplitude of the gas-sensitive resistor is within the effective response range of the sensor, whether the fluctuation variance matches the interference level of the current monitoring environment, and whether the rise edge response slope and recovery descent coefficient conform to the response law of the corresponding substance. The higher the degree to which the feature conforms to the real physical rules, the higher the score for this item. The system also verifies whether the dimensional weight allocation within the subset is reasonable, whether the core features, the change amplitude of the gas-sensitive resistor and the fluctuation variance, match the preset high weight ratio, and whether the weight of the fluctuation variance is increased in high interference scenarios. The higher the degree of matching between the weight allocation and the scenario and the importance of the feature, the higher the score for this item. After the three evaluations are weighted and summarized according to the preset weights, a comprehensive evaluation score of 0-100 points and a single-dimensional evaluation detail are output, forming the first evaluation result corresponding to the resistance feature subset.

[0036] The concurrent evaluation of the infrared feature subset, based on its built-in 4-dimensional ordered normalized spectral feature values, the exclusive weights bound to each dimension, fixed feature sorting rules, and spectral specific attributes, employs an evaluation framework symmetrical to that of the resistance feature subset. Feature validity is verified by checking whether all four feature values ​​within the subset fall within the 0-1 normalization range, whether the sum of the four dimension weights is 100%, and whether the feature sorting is perfectly aligned with the 5th to 8th positions of the standard dimensions in the false alarm feature library. A perfect score is awarded for this item if no anomalies are found. Feature reliability is verified based on the 4-dimensional spectral feature values ​​within the subset, checking whether the infrared feature absorption intensity and absorption peak bandwidth shift match the standard spectral response law of methane, and whether the light intensity attenuation coefficient is... The score is determined by the degree of dust and water vapor interference in the current environment, whether the signal-to-noise ratio stability meets the minimum threshold for spectral recognition, and the degree to which the feature conforms to the standard spectral characteristics of methane. The score is also determined by whether the dimensional weight allocation within the weight adaptation verification subset is reasonable, whether the core specific feature infrared feature absorption intensity and absorption peak bandwidth offset match the preset high weight ratio, and whether the weight of the light intensity attenuation coefficient is increased in high dust and water vapor scenarios. The score is also determined by the degree of matching between the weight allocation and the recognition advantages of the infrared sensor. After the three evaluations are weighted and summarized, a comprehensive evaluation score of 0-100 points and a detailed single-dimensional evaluation are output, forming the second evaluation result corresponding to the infrared feature subset.

[0037] Based on this, and using the quantitative scores from the first and second evaluation results, the mapping coefficients built into the initial unified mapping relationship are dynamically adjusted. First, the subset-level global mapping coefficients are adjusted. The core rule is that the higher the comprehensive evaluation score of a feature subset, the greater its contribution to the final target feature vector, and the larger its corresponding global coefficient. Simultaneously, the sum of the global coefficients α for the resistance subset and β for the infrared subset is strictly constrained to be 2, maintaining consistency with the initial coefficient sum to avoid shifts in the overall numerical range. If the comprehensive score of a subset is below 60, it is judged as having low confidence, and its global coefficient is lowered to below 0.5, significantly reducing... To mitigate the interference of unreliable data with the final result, the dimensional mapping coefficients are adjusted. The adjustment magnitude is positively correlated with the weight percentage and single-dimensional evaluation score of the dimension in the corresponding subset. The built-in dimensional weights of the two feature subsets are fully reused. The mapping coefficients of core feature dimensions with high weight percentage and high single-dimensional evaluation scores are increased, while the mapping coefficients of auxiliary feature dimensions with low weight percentage and low single-dimensional evaluation scores are decreased. At the same time, the upper limit of the adjustment magnitude of low-weight feature dimensions is strictly controlled to avoid excessive influence of auxiliary features on the core recognition logic. After the full-dimensional coefficient adjustment is completed, a unified mapping relationship that adapts to the actual quality of the two feature subsets is formed. Subsequently, based on the calibrated unified mapping relationship, a structured joint mapping is performed on the resistance feature subset and the infrared feature subset. First, dimension alignment and padding are performed: the four ordered normalized features of the resistance feature subset are filled into the 1st-4th positions of the standard feature dimension of the false alarm feature library in their original fixed order; the four ordered normalized features of the infrared feature subset are filled into the 5th-8th positions of the standard feature dimension in their original fixed order, forming an initial 8-dimensional feature sequence to be corrected that retains all feature information and sorting rules of both subsets. Then, a subset-level global weighted correction is performed: the resistance features in the 1st-4th positions of the initial sequence are multiplied by the adjusted global coefficient α of the resistance subset; the features in the 5th-8th positions are multiplied by the global coefficient α of the resistance subset. The 8-bit infrared features are multiplied by the adjusted global coefficient β of the infrared subset to obtain a globally corrected sequence that reflects the difference in overall credibility between the two subsets. Then, a dimension-level fine-grained weighted correction is performed. For each dimension of the globally corrected 8-dimensional sequence, the corresponding adjusted dimension-level mapping coefficients k1 to k8 are multiplied to obtain a dimension-corrected sequence that reflects the difference in importance and credibility of a single feature dimension. Finally, the corrected 8-dimensional sequence is interval-verified to ensure that all dimension values ​​are within the standard range of 0-1. At the same time, each dimension is bound to its corresponding physical meaning, the source of the original feature subset, and the details of the mapping coefficients to generate a structured 8-dimensional target feature vector.

[0038] In some embodiments, determining the gas monitoring result based on the target feature vector and a pre-established false alarm feature library includes: matching the target feature vector with each false alarm feature vector in the pre-established false alarm feature library to obtain multiple matching results; and determining the gas monitoring result based on the multiple matching results.

[0039] In this embodiment of the invention, the false alarm feature vectors in the false alarm feature library are all 8-dimensional structured feature vectors with the same dimension, feature order, 0-1 value range, and weight system as the target feature vector. Each vector uniquely corresponds to a non-gas interference type that has been verified in actual scenarios, including cooking fumes, medical alcohol volatilization, chlorine disinfection water vapor, high temperature and high humidity environment interference, daily chemical product volatilization and other daily high-frequency interference scenarios. Each vector is labeled with the corresponding interference type, feature dimension weight, effective matching threshold, and scenario adaptation description. At the same time, the library is classified and archived according to interference type and sensor installation location.

[0040] The target feature vector is matched against each false alarm feature vector stored in the false alarm feature database using a weighted cosine similarity algorithm. The matching calculation employs the mapping weights bound to each dimension of the target feature vector and the dimensional weights of the corresponding vectors in the false alarm feature database, serving as a dual weighting basis. Finally, a similarity score is calculated for each false alarm feature vector participating in the matching. The score range is fixed at 0-1; the closer the score is to 1, the higher the feature overlap between the target feature vector and the false alarm feature vector, indicating that the current monitoring signal is caused by this type of non-gas-related interference. The higher the probability, the more likely the similarity results are to be. After the full database matching calculation is completed, all matching results are filtered. Using the preset effective matching threshold of 0.75 as the dividing line, results with similarity scores greater than or equal to the effective matching threshold are marked as effective matching results, and results with similarity scores less than the effective matching threshold are marked as invalid matching results. After removing all invalid matching results, all effective matching results are sorted from high to low according to similarity scores. Each effective matching result fully includes the corresponding interference type, similarity score, matching details of each dimension, weighted record, and scene adaptation information.

[0041] In some embodiments, determining the gas monitoring result based on multiple matching results includes: applying a time decay strategy to attenuate and correct the confidence level of the matching results, and applying a spatial decay strategy to attenuate and correct the weight of the matching results; and determining the gas monitoring result based on the corrected matching results.

[0042] In this embodiment of the invention, after completing the full-database matching of the target feature vector with the false alarm feature library and obtaining multiple valid matching results sorted by initial similarity, the matching results are double-corrected based on time decay strategy and spatial decay strategy. After the double decay correction is completed, the comprehensive matching degree is calculated for all valid matching results. The formula is: comprehensive matching degree = decayed confidence degree × decayed weight. Then, all results are re-sorted from high to low according to the comprehensive matching degree to form a corrected set of valid matching results. Based on this set, the final gas monitoring result is determined through a four-level progressive logic. The first level is the determination of no valid matching results. If the corrected set is empty, or the comprehensive matching degree of all results is lower than the minimum effective threshold of 0.3, it means that the currently monitored feature change does not effectively overlap with all known non-gas interference features, and there is no historical interference or cross-regional interference reference. It is directly determined to be a real gas leak, triggering a full-process safety handling command such as audible and visual alarm, gas emergency shut-off valve linkage, and remote alarm push. The second level is the high-confidence false alarm judgment. If the highest comprehensive matching degree in the corrected set is ≥0.85, which is the strong false alarm judgment threshold, and the single-dimensional matching degree of the resistance and infrared core feature dimensions is ≥0.80, it means that the current monitoring signal highly overlaps with the corresponding non-gas interference feature, and still has high confidence after double correction. It is directly judged as a non-gas interference false alarm and is excluded. No alarm command is triggered. Only the interference type, data before and after correction, and spatiotemporal attribute information are recorded. At the same time, the effective data is updated to the false alarm feature library self-learning sample set to complete the feature library iterative optimization. The third level is a comprehensive judgment with medium confidence. If the highest comprehensive matching degree is in the medium confidence range of 0.50-0.85, it enters the multi-dimensional analysis: If there are two or more interference matching results of the same type with a comprehensive matching degree ≥0.50, and they all come from sensors in the core monitoring zone, it is judged as a false alarm due to non-gas interference and eliminated; if there is only one valid matching result, and the matching degree of the infrared core feature dimension representing methane specific identification is <0.70, it means that only the resistive broad spectrum response feature overlaps with the interference part, and the infrared specific feature does not match. The risk of leakage cannot be completely eliminated, and a pending confirmation warning status is output. No emergency response is triggered, only a warning prompt is pushed, and high-frequency continuous monitoring is started. Data is collected every 5 seconds to generate a new target feature vector, and the matching and double correction process is repeated. The fourth level is the closed-loop judgment of the pending warning. For the pending warning status, if the overall matching degree does not improve after three consecutive high-frequency monitoring corrections and the matching degree of the infrared core feature dimension remains below 0.70, it is upgraded to a real gas leak and triggers the full-process safety handling; if the overall matching degree improves to above 0.85 after two consecutive monitoring, it is judged as a false alarm due to non-gas interference, the warning is lifted and the regular monitoring frequency is restored.

[0043] In some embodiments, a time decay strategy is used to decay and correct the confidence of the matching results, including: recording the entry time of each false alarm feature vector and calculating the time difference between the current time and the entry time; correcting the confidence of the matching results based on the difference and the time decay function; wherein, the time decay function is determined by the effective period, basic decay coefficient and maximum decay amplitude of each false alarm feature vector.

[0044] In this embodiment of the invention, non-gas-related interfering substances possess the physical characteristics of naturally dissipating over time and rapidly decreasing in concentration. The resulting fluctuations in sensor signals are instantaneous. The confidence level of the matching results corresponding to the same interference event gradually decreases over time as the interfering substance dissipates. This strategy avoids misjudging the current real-time monitoring status based on historically expired interference matching results, while also adapting to the dissipation characteristics of different interference types for differentiated adjustments. The specific execution process is as follows: First, a 30-minute effective decay time window is set. The confidence level of historical matching results exceeding this window is directly reset to zero and no longer participates in the current monitoring result determination. Within the window, an exponential decay function is used as the time decay function, with the formula: Attenuated confidence score = initial similarity score × e (-λ×Δt) Where Δt is the time difference (in minutes) between the acquisition time corresponding to the matching result and the current time, and λ is the reference time decay coefficient, with the initial reference value set at 0.05; subsequently, based on the interference dissipation characteristics archived in the false alarm feature library, the λ value is differentially adapted. For rapidly dissipating interference such as cooking fumes, alcohol evaporation, and daily chemical sprays, the λ value is increased to 0.08-0.12 to accelerate confidence decay; for slowly dissipating environmental interference caused by high temperature and humidity and poor ventilation, the λ value is decreased to 0.02-0.04 to reduce the likelihood of false alarms. To mitigate the decay rate, the baseline λ value remains unchanged for medium-speed dissipation interference such as disinfectant water vapor. At the same time, decay boundary constraints are set, with the lowest confidence threshold after decay being 0.2. If the value is lower than this threshold, it is uniformly set to 0.2 to retain the reference value of the basic interference characteristics. For valid matching results of the same type generated in three consecutive monitoring cycles, the corresponding λ value is reduced by 30% to adapt to the continuous interference scenario. Finally, the confidence after decay is calculated for each valid matching result, and the original initial similarity score is replaced to complete the confidence correction in the time dimension.

[0045] In some embodiments, a spatial attenuation strategy is used to attenuate and correct the weights of the matching results, including: obtaining spatial features and sensor installation locations; determining the interference propagation coefficient of each spatial point based on the spatial features; and attenuating and correcting the weights of the matching results based on the interference propagation coefficient of each spatial point and the sensor installation location.

[0046] In this embodiment of the invention, the influence of non-gas interference has spatial limitations. The intensity of the influence gradually decreases with the increase of spatial distance, physical isolation, and changes in ventilation conditions. The matching results generated by the same interference event on sensors at different installation locations have significantly different judgment weights on the current monitoring area. This strategy can quantify the influence of matching results at different spatial locations and avoid misjudging the current core monitoring area by interference results from non-related areas.

[0047] The monitoring space is divided into several standardized three-dimensional grid units, and four core spatial features are collected: spatial geometry, physical partition distribution, ventilation environment parameters, and high-frequency interference source reference points. The installation three-dimensional coordinates, grid units, and functional zoning attributes of all gas monitoring sensors are simultaneously calibrated, and the sensor locations are uniquely bound to the matching results generated by the device.

[0048] Based on the acquired global spatial features and the propagation characteristics of each non-gas interference type in the false alarm feature library, the corresponding interference propagation coefficient is calculated for each spatial point (3D grid) after gridding. This coefficient is used to quantify the intensity retention ratio of non-gas interference when it propagates from the reference source point of the corresponding type to the spatial point. The value range is fixed at 0-1. The closer the coefficient is to 1, the easier it is for the interference to propagate to the spatial point and the smaller the interference intensity attenuation. The closer the coefficient is to 0, the lower the probability of the interference propagating to the point and the more significant the intensity attenuation. The specific calculation process is as follows: First, for each type of non-gas interference in the false alarm feature library, a reference propagation parameter adapted to its physical characteristics is set, including the reference diffusion coefficient, the blocking attenuation ratio, and the ventilation attenuation ratio. For example, for oil fume interference, the particle size is large and the diffusion ability is weak, so a higher reference diffusion attenuation ratio is set; for alcohol vapor interference, the molecular diffusion ability is strong, so a lower reference diffusion attenuation ratio is set; for high temperature steam interference, the ventilation attenuation ratio is set. Subsequently, four sub-attenuation coefficients are calculated for each spatial point. These are then multiplied to obtain the interference propagation coefficient for that spatial point relative to a specific interference type: The first is the geometric diffusion attenuation coefficient, calculated using a spherical diffusion attenuation model based on the straight-line distance between the spatial point and the corresponding interference source reference point. The formula is as follows: Geometric diffusion attenuation coefficient = 1 / (1+Kd×D) Wherein, D is the straight-line distance between the spatial point and the reference point of the interference source, and Kd is the reference diffusion coefficient for this type of interference. The greater the distance, the smaller the geometric diffusion attenuation coefficient, and the greater the intensity attenuation of interference propagation. Secondly, there is the physical barrier attenuation coefficient, calculated based on the number and level of physical barriers between the spatial point and the reference point of the interference source. The barrier attenuation coefficient is 1 when there are no physical barriers, 0.7 when there is a single door / lightweight partition, and 0.3 when there is a complete solid wall partition. For multi-layered partitions, the coefficient is calculated by multiplying by a factor; the more barriers, the smaller the barrier attenuation coefficient. Thirdly, there is the ventilation convection attenuation coefficient, based on the ventilation direction and airflow between the spatial point and the reference point of the interference source. The ventilation rate is calculated by multiplying the ventilation rate by the corresponding ventilation attenuation factor if the ventilation direction is the same as the interference propagation direction (e.g., the exhaust fan faces outdoors, and the interference is discharged with the airflow). The faster the airflow rate, the smaller the ventilation attenuation coefficient, and the more difficult it is for the interference to propagate to that spatial point. If the ventilation direction is opposite to the interference propagation direction, the ventilation attenuation coefficient remains at 1, with no additional attenuation. Fourthly, the zoning adaptation coefficient is set based on the attribute settings of the functional zone to which the spatial point belongs. The zoning adaptation coefficient for high-incidence interference areas is 1, for the core monitoring area of ​​gas equipment it is 0.9, for non-core auxiliary areas it is 0.6, and for completely isolated sealed zones it is 0.2, further matching the probability of interference occurrence in different zones. Finally, the four sub-coefficients are multiplied to obtain the interference propagation coefficient for that spatial point for a specific type of interference. At the same time, coefficient boundary constraints are set: the minimum value is not less than 0.1, and the maximum value is not more than 1. This completes the calculation of the interference propagation coefficient for all spatial points in the entire domain, forming an interference propagation coefficient matrix that corresponds one-to-one with the spatial grid.

[0049] For each valid match result, the corresponding non-gas interference type and the sensor device code that generated the result are first extracted. Based on the device code, the 3D spatial grid code corresponding to the sensor installation location is retrieved to complete the unique binding of the match result, sensor location, and grid cell. Then, from the coefficient matrix, the exclusive sub-matrix corresponding to the interference type (covering the propagation coefficients of this type of interference in all grids of the entire domain) is retrieved. On the one hand, the exclusive propagation coefficient of the interference type corresponding to the grid where the sensor is located is extracted. On the other hand, along the propagation path from the reference point of the interference source to the sensor location, the propagation coefficient sequence of all continuous grids on the path is retrieved to complete the spatial adaptability pre-verification of the match result: If the propagation coefficient of the sensor location and the attenuation law of the path sequence are continuous and consistent, and the propagation coefficient is positively correlated with the initial similarity of the match result, it is determined that the spatial adaptability is valid and the result is retained for subsequent correction. If the propagation coefficient on the path has decayed to close to 0, or the point propagation coefficient deviates significantly from the initial similarity (e.g., the point propagation coefficient is 0.2 but the initial similarity is 0.9), it is determined that the spatial adaptability is invalid, and the initial weight of the result is directly reduced to below 0.2 for auxiliary reference only.

[0050] Based on the global distribution characteristics of the coefficient matrix, the initial weights of the matching results are dynamically set. Abandoning the static initial weight rules of fixed partitions, the initial weights of each matching result are dynamically set based entirely on the global distribution of the coefficient matrix and the monitoring priority of the partition where the sensor is located: First, the propagation coefficient of the submatrix of the interference type in the coefficient matrix is ​​used as the benchmark to normalize the propagation coefficient of the grid where the sensor is located, obtaining the spatial contribution benchmark value for that point; then, the partition monitoring priority coefficient is superimposed (1.0 for the core monitoring area of ​​gas equipment, 0.6 for non-core auxiliary areas, and 0.3 for closed isolation areas), and the initial weight is dynamically set using the formula: Initial Weight = Spatial Contribution Benchmark Value × Partition Priority Coefficient. For example, at a sensor location within the core monitoring area, the propagation coefficient of the oil fume interference is only 0.15 (far lower than the maximum value of 0.9 across the entire area). Therefore, its initial weight is (0.15 / 0.9)×1.0≈0.17, rather than a fixed 1.0, because the oil fume interference can hardly propagate to this location, resulting in extremely low reliability of the matching result. On the other hand, at sensor locations outside the core cooking area, the propagation coefficient of the oil fume interference is 0.85, and its initial weight is (0.85 / 0.9)×0.6≈0.57, achieving a precise match with the actual propagation probability of the interference.

[0051] After initial dynamic weight setting, based on the spatial neighborhood characteristics of the coefficient matrix and combined with multi-sensor linkage data, a second-order fine-grained attenuation correction is applied to the weights of the matching results. For matching results of the same interference type, the propagation coefficients of the sensor's grid and its eight surrounding grids, as well as the similarity data of matching results of the same type from other sensors within the neighborhood, are retrieved to verify the positive correlation between the propagation coefficient and the matching similarity within the neighborhood: if the propagation coefficient within the neighborhood exhibits a continuous attenuation pattern, and the matching similarity attenuates synchronously, it indicates that the matching result conforms to the spatial diffusion characteristics of interference, and the correction formula is: Corrected weight = Initial weight × Grid-specific propagation coefficient × Neighborhood continuity coefficient In this process, the continuity coefficient of the neighborhood is set to 1.0 if the continuity meets the standard, and adjusted to 0.3-0.7 if it does not. If the propagation coefficient in the neighborhood deviates significantly from the matching similarity, it indicates that the matching result is a single-point anomalous signal, and the corrected weight is directly reduced to below 0.1. At the same time, for scenarios where the same matching result corresponds to multiple types of interference, the sub-matrices corresponding to each type of interference in the coefficient matrix are retrieved respectively. After completing the full-process weight correction, the result with the highest corrected weight is retained as the final weight of the matching result.

[0052] After completing the weight decay correction for all matching results, the final corrected weight of each result is multiplied by the confidence level corrected by the time decay strategy, and the global propagation proportion coefficient of the corresponding interference type in the coefficient matrix is ​​added to obtain the final comprehensive matching degree of each matching result. Then, all matching results are sorted from high to low according to the comprehensive matching degree to form a set of effective matching results after space-time dual correction.

[0053] In some embodiments, before determining the gas monitoring result based on the target feature vector and a pre-established false alarm feature library, the method further includes: acquiring interference monitoring data generated by various non-gas substances; extracting interference feature vectors corresponding to resistive monitoring data and infrared monitoring data from the interference monitoring data; standardizing and encoding the extracted interference feature vectors and storing them to form a false alarm feature library.

[0054] In this embodiment of the invention, firstly, full-volume interference monitoring data of various typical non-flammable substances such as cooking fumes, alcohol evaporation, disinfectant water vapor, and daily chemical product evaporation are acquired under different spatial scenarios, ventilation conditions, and concentration gradients. Each set of interference monitoring data simultaneously includes broad-spectrum response monitoring data collected by a resistive gas sensor and methane-specific spectral monitoring data collected by an infrared gas sensor. Then, from each pair of interference monitoring data, four-dimensional core features corresponding to the resistive monitoring data are extracted: the amplitude of gas-sensitive resistor change, rising edge response slope, fluctuation variance, and recovery descent coefficient. Additionally, infrared characteristic absorption intensity, light intensity attenuation coefficient, and absorption peak bandwidth shift corresponding to the infrared monitoring data are also extracted. The system uses four core features—quantity, signal-to-noise ratio stability, etc.—to generate interference feature vectors that are identical in dimension, feature order, and physical semantics to the target feature vectors to be matched. Then, it performs min-max normalization encoding on all generated interference feature vectors, which is completely consistent with the target feature vectors. Simultaneously, it binds auxiliary information such as the corresponding non-gas substance type, applicable scenario, feature dimension weight, effective matching threshold, and interference dissipation characteristics to each standardized interference feature vector. Finally, it classifies and archives all standardized and information-bound interference feature vectors according to interference type and stores them persistently, forming a structured false alarm feature library that can adapt to subsequent full-process matching and judgment and supports dynamic self-learning and iterative updates.

[0055] Figure 2 This is a schematic diagram of the structure of the gas alarm terminal provided in an embodiment of the present invention. Figure 2 As shown, in an embodiment of the present invention, a gas alarm terminal includes a memory 21, a processor 20, and a computer program 22 stored in the memory 21 and executable on the processor 20. When the processor executes the computer program, it implements the steps of the gas alarm method based on the false alarm feature library as described in the above embodiment.

[0056] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a controller, implements the gas alarm method based on a false alarm feature library as described in the above embodiments.

[0057] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0058] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A gas alarm method based on a false alarm feature library, characterized in that, include: Acquire multi-source gas monitoring data; Based on the multi-source gas monitoring data, the target feature vector of gas monitoring is extracted; The gas monitoring results are determined based on the target feature vector and the pre-established false alarm feature library. The false alarm feature library is used to record feature changes caused by non-gas substances.

2. The gas alarm method based on a false alarm feature library according to claim 1, characterized in that, The multi-source gas monitoring data includes resistive gas monitoring data and infrared gas monitoring data; based on the multi-source gas monitoring data, a target feature vector for gas monitoring is extracted, including: The resistive gas monitoring data is analyzed to obtain the change amplitude of the gas-sensitive resistor, the rise edge response slope, the fluctuation variance, and the recovery descent coefficient. A subset of resistance features is generated based on the change amplitude, rising edge response slope, fluctuation variance, and recovery descent coefficient of the gas-sensitive resistor. The infrared gas monitoring data is analyzed to obtain the infrared characteristic absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability. Infrared feature subsets are generated based on infrared feature absorption intensity, light intensity attenuation coefficient, absorption peak bandwidth shift, and signal-to-noise ratio stability. The target feature vector is determined based on the resistive feature subset and the infrared feature subset.

3. The gas alarm method based on a false alarm feature library according to claim 2, characterized in that, Based on the resistivity feature subset and the infrared feature subset, the target feature vector is determined, including: A unified mapping relationship is established using the standard feature dimensions of the false alarm feature library as the mapping target; The resistive feature subset is evaluated to obtain a first evaluation result, and the infrared feature subset is evaluated to obtain a second evaluation result. Based on the first evaluation result and the second evaluation result, adjust the mapping coefficient of the unified mapping relationship; The target feature vector is obtained by jointly mapping the resistive feature subset and the infrared feature subset according to the unified mapping relationship.

4. The gas alarm method based on a false alarm feature library according to claim 1, characterized in that, Based on the target feature vector and a pre-established false alarm feature library, the gas monitoring results are determined, including: The target feature vector is matched with each false alarm feature vector in the pre-established false alarm feature library to obtain multiple matching results; Based on the multiple matching results, the gas monitoring results are determined.

5. The gas alarm method based on a false alarm feature library according to claim 4, characterized in that, Based on the multiple matching results, the gas monitoring results are determined, including: A time decay strategy is used to decay and correct the confidence of the matching results, and a spatial decay strategy is used to decay and correct the weight of the matching results. Based on the corrected matching results, the gas monitoring results are determined.

6. The gas alarm method based on a false alarm feature library according to claim 5, characterized in that, A time decay strategy is used to attenuate and correct the confidence level of the matching results, including: Record the entry time of each false alarm feature vector and calculate the time difference between the current time and the entry time; The confidence level of the matching results is corrected based on the difference and the time decay function. The time decay function is determined by the effective period, basic decay coefficient, and maximum decay amplitude of each false alarm feature vector.

7. The gas alarm method based on a false alarm feature library according to claim 5, characterized in that, A spatial decay strategy is used to attenuate and correct the weights of the matching results, including: Acquire spatial features and sensor installation locations; Based on spatial characteristics, determine the interference propagation coefficient for each spatial point; The weights of the matching results are attenuated based on the interference propagation coefficient of each spatial point and the sensor installation location.

8. The gas alarm method based on a false alarm feature library according to claim 1, characterized in that, Before determining the gas monitoring result based on the target feature vector and a pre-established false alarm feature library, the method further includes: Acquire monitoring data on interference generated by various non-flammable substances; Extract interference feature vectors from the interference monitoring data that correspond to the resistive monitoring data and the infrared monitoring data; The extracted interference feature vectors are standardized, encoded, and stored to form the false alarm feature library.

9. A gas alarm terminal, 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 gas alarm method based on a false alarm feature library as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the controller, it implements the steps of the gas alarm method based on the false alarm feature library as described in any one of claims 1 to 8.

Images