A self-learning combustible explosive gas measuring method and system based on virtual mixed samples

Abstract

The embodiments of the present disclosure provide a combustible explosive gas measurement method and system based on virtual mixed samples. The method is applied to the field of gas detection technology, and comprises the following steps: obtaining combustible explosive gas historical analysis results, historical environmental states and real measurement data of a to-be-measured area, collating and associating to obtain a real response sample set, and establishing a basic measurement model based on the real response sample set. A virtual working condition set is derived using low-coverage working condition data, and a virtual mixed sample set is simulated and generated. Consistency evaluation and credibility weighting are performed on the virtual mixed sample set and the real sample set, and an enhanced training sample set is obtained. The basic model is self-learned and trained based on the enhanced sample set, and a self-learning measurement model is formed. Real-time gas measurement and environmental data are input into the model, and accurate measurement and identification of combustible explosive gas are realized. The present scheme can enhance the diversity and representativeness of the training data, improve the accuracy and robustness of the self-learning measurement model, and realize more accurate measurement and identification of combustible explosive gas.

Description

Technical Field

[0001] This disclosure relates to the field of gas detection technology, and in particular to a self-learning method and system for measuring combustible and explosive gases based on virtual mixed samples. Background Technology

[0002] In numerous fields such as industrial production, energy development, and storage, the presence of flammable and explosive gases constantly threatens the safety of personnel and the stable operation of equipment. Accurate and timely measurement of flammable and explosive gas concentrations, and early warning of potential explosion risks, are crucial for ensuring safe production and preventing major accidents. Therefore, obtaining reliable measurement results for flammable and explosive gases has become a critical issue that urgently needs to be addressed in these fields.

[0003] Current combustible and explosive gas measurement technologies first extensively collect historical gas measurement data, corresponding environmental condition data, and actual results for the area to be measured, and then simply integrate them into a basic dataset. Based on this dataset, a basic measurement model is then built using traditional methods. During actual measurements, newly acquired gas response data and real-time environmental data are directly input into the model to obtain the results.

[0004] However, due to the complexity and variability of actual operating conditions, limited historical data collection, and severely insufficient coverage of operating conditions in the basic dataset, the model, lacking sufficient training data, cannot accurately grasp the relationship between gas and the environment, resulting in inaccurate measurements or even errors. Furthermore, the model is completely incapable of handling new operating conditions, exhibiting a large measurement blind spot. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this disclosure provides a self-learning method and system for measuring flammable and explosive gases based on virtual mixed samples. This disclosure solves the problems of existing technologies, such as the complexity and variability of actual operating conditions, limited historical data collection, and severely insufficient coverage of operating conditions in the basic dataset. Furthermore, under low-coverage operating conditions, the model, due to insufficient training data, cannot accurately grasp the relationship between the gas and the environment, resulting in inaccurate measurements or even errors. The model is also completely incapable of handling new operating conditions, exhibiting a large measurement blind zone.

[0006] According to a first aspect of this disclosure, a self-learning method for measuring combustible and explosive gases based on virtual mixed samples is provided, comprising: acquiring historical analysis results of combustible and explosive gases in the area to be measured, as well as corresponding historical environmental state data and corresponding real gas measurement results; and performing data processing and correlation analysis based on the historical analysis results, historical environmental state data and real gas measurement results to obtain a real response sample set.

[0007] Based on the real response sample set, response patterns are modeled to obtain a basic measurement model;

[0008] Obtain low-coverage operating condition data from the basic measurement model, perform operating condition simulation based on the low-coverage operating condition data, and obtain a virtual operating condition set.

[0009] The virtual operating condition set is input into the basic measurement model for response simulation to obtain a virtual hybrid sample set;

[0010] Based on the virtual mixed sample set and the real response sample set, sample consistency evaluation and credibility weighting are performed to obtain the enhanced training sample set;

[0011] The basic measurement model is trained by self-learning based on the enhanced training sample set to obtain a self-learning measurement model.

[0012] The gas measurement response data and the corresponding real-time environmental status data are acquired. The gas measurement response data and the real-time environmental status data are then input into the self-learning measurement model for measurement identification to obtain the measurement results of flammable and explosive gases.

[0013] According to a second aspect of this disclosure, a self-learning combustible and explosive gas measurement system based on virtual mixed samples is provided, for performing the method as described in the first aspect, including: a historical data acquisition module, for acquiring historical analysis results of combustible and explosive gases in the area to be measured, as well as corresponding historical environmental state data and corresponding real gas measurement results, and performing data processing and correlation analysis based on the historical analysis results, historical environmental state data and real gas measurement results to obtain a real response sample set;

[0014] The basic model building module is used to model the response patterns based on the real response sample set to obtain a basic measurement model.

[0015] The working condition simulation module is used to acquire low-coverage working condition data in the basic measurement model, perform working condition simulation based on the low-coverage working condition data, and obtain a virtual working condition set.

[0016] The response simulation module is used to input the virtual operating condition set into the basic measurement model for response simulation to obtain a virtual mixed sample set;

[0017] An enhanced sample construction module is used to perform sample consistency evaluation and credibility weighting based on the virtual mixed sample set and the real response sample set to obtain an enhanced training sample set.

[0018] The self-learning training module is used to perform self-learning training on the basic measurement model based on the enhanced training sample set to obtain a self-learning measurement model.

[0019] The measurement module is used to acquire gas measurement response data and corresponding real-time environmental status data. The gas measurement response data and real-time environmental status data are input into the self-learning measurement model for measurement identification to obtain the measurement results of combustible and explosive gases.

[0020] According to a third aspect of this disclosure, an electronic device is provided, comprising: a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0021] According to a fourth aspect of this disclosure, a readable storage medium is provided on which a program or instructions are stored, which, when executed by a processor, implement the method described above.

[0022] In the self-learning flammable and explosive gas measurement method and system based on virtual hybrid samples provided above, the embodiments of this disclosure can effectively expand the coverage of data samples, especially for extreme working conditions where direct collection is difficult. By combining virtual samples with real samples, the diversity and representativeness of training data are enhanced, thereby improving the accuracy and robustness of the self-learning measurement model, ultimately achieving more accurate measurement and identification of flammable and explosive gases. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this disclosure 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 this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A first flowchart of a self-learning combustible and explosive gas measurement method based on virtual mixed samples according to an embodiment of the present disclosure is shown.

[0025] Figure 2 A second flowchart of a self-learning combustible and explosive gas measurement method based on virtual mixed samples according to an embodiment of the present disclosure is shown.

[0026] Figure 3 A schematic block diagram of a self-learning combustible and explosive gas measurement system based on virtual mixed samples according to an embodiment of the present disclosure is shown.

[0027] Figure 4 A schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0028] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0029] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of this disclosure are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them. It should also be understood that in the embodiments of this disclosure, "multiple" can refer to two or more, and "at least one" can refer to one, two, or more. It should also be understood that any component, data, or structure mentioned in the embodiments of this disclosure can generally be understood as one or more unless explicitly limited or given a contrary indication in the context. Furthermore, the term "and / or" in this disclosure is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this disclosure generally indicates that the related objects before and after are in an "or" relationship. It should also be understood that the descriptions of the various embodiments in this disclosure emphasize the differences between the various embodiments; their similarities or commonalities can be referred to mutually, and for the sake of brevity, they will not be elaborated upon one by one.

[0030] Furthermore, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn to actual scale. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use. Techniques, methods, and apparatus known to those skilled in the art will not be discussed in detail, but where appropriate, such techniques, methods, and apparatus should be considered part of the specification. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0031] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0032] Figure 1 This is a schematic diagram of the first process of a self-learning combustible and explosive gas measurement method based on virtual mixed samples provided in an embodiment of this disclosure.

[0033] S101, acquire the historical analysis results of combustible and explosive gases in the area to be tested, as well as the corresponding historical environmental state data and the corresponding real gas measurement results. Based on the historical analysis results, historical environmental state data and real gas measurement results, perform data processing and correlation analysis to obtain a real response sample set.

[0034] The area to be tested is a specially designated space or location for conducting combustible gas measurement and monitoring. Its designation primarily focuses on potentially hazardous areas, such as the vicinity of combustible gas leak sources.

[0035] Historical analysis results are the analysis of monitoring data related to flammable and explosive gases over a past period. Through statistical analysis, modeling, and anomaly identification of past monitoring data, valuable conclusions are extracted, focusing on revealing the behavioral patterns, concentration trends, and the impact of environmental factors on gas behavior.

[0036] Historical environmental data refers to various environmental parameters that significantly influence the behavior, diffusion rate, and chemical reactions of gases. These include ambient temperature, humidity, air pressure, wind speed, and light intensity, each of which has different effects on the physical properties and diffusion patterns of gases. For example, excessively high temperatures can accelerate the volatilization of some flammable gases and may also exacerbate their explosiveness.

[0037] Real gas measurement results are raw data collected directly by specialized monitoring instruments, reflecting the actual state of the gas. These data record in detail the type of gas detected, its specific concentration, measurement time, measurement location, and environmental conditions during measurement, such as temperature and humidity.

[0038] The real response sample set is a comprehensive collection of samples formed by combining historical analysis results of flammable and explosive gases, historical environmental condition data, and real gas measurement data, after a series of processing steps including data organization, cleaning, and correlation analysis. It is a standardized sample set formed by systematically analyzing actual response conditions under different environmental conditions and gas concentrations and integrating various relevant data.

[0039] The delineation of the test area should focus on potential hazards, primarily covering the vicinity of flammable gas leak sources, flammable and explosive areas, and other key safety control locations. After delineation, the next step is to obtain historical analysis results of flammable and explosive gases within that area. These results come from the systematic processing and in-depth analysis of long-term accumulated historical monitoring data. This historical data comprehensively records the concentration changes, behavior patterns, and correlations between gas states and various environmental factors. The core objective of the analysis is to reveal the trends in gas concentration changes and the intrinsic relationships between concentration changes and various environmental factors using statistical modeling and regression analysis. Temperature, humidity, air pressure, and wind speed are all key environmental factors that directly affect gas behavior and diffusion patterns. For example, high temperatures accelerate gas evaporation rates, thus increasing the risk of explosion; low temperatures may cause gases to linger in the air, accumulating and forming hazardous areas. Time series analysis can clearly reveal the patterns of gas concentration changes over time, identifying periodic fluctuations or sudden abnormal changes in concentration. Simultaneously, regression analysis can be used to construct mathematical models between gas concentration and various environmental parameters. These analyses can yield conclusions with practical value, such as clarifying the correlation between gas concentration and specific environmental conditions, and how various factors work together to affect the propagation speed of gas and the likelihood of explosion.

[0040] While carrying out the above work, acquiring historical environmental data is also a crucial step. Environmental data primarily comes from various specialized monitoring equipment such as weather stations and environmental monitoring sensors. These devices can collect and record environmental parameters closely related to gas diffusion, such as temperature, humidity, air pressure, and wind speed, in real time. This environmental data, combined with gas concentration data, forms a complete analytical dataset. The collected environmental data also needs time alignment and noise reduction. Time alignment ensures that gas concentration data and environmental data remain synchronized over time; noise reduction effectively removes interfering information from the environmental data. Subsequently, real gas measurement results need to be obtained. These results are raw data directly collected by specialized monitoring instruments such as gas detectors and infrared gas sensors, fully including gas type, concentration, measurement time, measurement location, and environmental conditions at the time of measurement. To ensure data accuracy, the collected raw data needs to be verified. For example, the response time and sensitivity of gas sensors may be affected by environmental factors, requiring necessary correction and filtering of the measurement data to eliminate interference from the external environment. In addition, these calibrated gas measurement data need to be combined with historical environmental data to conduct correlation analysis and explore in depth the specific impact of environmental factors on gas concentration changes, so as to gain a more comprehensive understanding of the gas behavior patterns.

[0041] Finally, all historical analysis results, historical environmental data, and real gas measurement data need to be integrated and correlated. The core of data processing is to standardize and format data from different sources, ensuring that all types of data can be compared and integrated within the same analytical framework. Data correlation analysis utilizes multivariate analysis methods such as correlation analysis and regression analysis to uncover the intrinsic relationship between gas concentration changes and environmental factors. This process can identify gas behavior patterns under different gas concentrations and environmental conditions. For example, it can clarify the changing trends of gas concentration under specific temperature and humidity conditions, as well as the diffusion patterns of gases in environments with different wind speeds. Ultimately, through systematic processing and correlation analysis, a real response sample set can be formed.

[0042] Based on the above technical solution, optionally, data processing and correlation analysis can be performed based on the historical analysis results, historical environmental state data, and real gas measurement results to obtain a real response sample set, including:

[0043] Based on the historical analysis results, historical environmental state data, and real gas measurement results, anomalies are removed and missing data are filled to obtain cleaning measurement data.

[0044] Based on the cleaning measurement data, dimensional unification and normalization are performed to obtain standard measurement data;

[0045] Response feature extraction is performed based on the standard measurement data to obtain response feature data including environmental feature data;

[0046] Environmental correlation analysis is performed based on the response feature data to obtain environmental correlation data;

[0047] Based on the response feature data and environmental correlation data, sample mapping and classification are performed to obtain a real response sample set.

[0048] In this scheme, the cleaning measurement data is the original gas measurement data after outlier removal and missing value completion.

[0049] Standard measurement data are gas measurement data that have undergone dimensional unification and normalization. The core function of dimensional unification is to ensure that all types of data are within the same dimensional system, meeting the basic conditions for mutual comparison; normalization converts the data to a fixed standard range, generally from 0 to 1, enabling data from different sources to be analyzed and compared on the same scale.

[0050] Environmental characteristic data refers to various environmental factors closely related to gas behavior and diffusion processes, specifically including indicators such as temperature, humidity, air pressure, and wind speed. It reveals the mechanisms by which different environmental conditions influence the physical properties and diffusion behavior of gases.

[0051] Response characteristic data are feature data extracted from gas measurement data that are related to gas behavior, concentration changes, sensor response, and other dimensions. They can reflect the dynamic response process of gas under different operating conditions, including sensor output values, gas concentration change trends, and the response of other key parameters.

[0052] Environmental correlation data is derived from the analysis of the correlation between environmental characteristic data and gas response characteristics. It reveals the interaction between environmental factors such as temperature and humidity and changes in gas concentration.

[0053] After obtaining historical analysis results, historical environmental condition data, and real gas measurement results, the first step is to clean this data. Cleaning begins with outlier removal, using statistical methods such as box plots, Z-scores, and IQR to accurately identify and remove outliers that fall outside the normal range. These outliers are often related to instrument malfunctions, data entry errors, or other external interference factors; their removal effectively improves the overall quality of the dataset. After outlier removal, the next step is missing value imputation. This typically uses interpolation methods or machine learning-based imputation algorithms such as K-nearest neighbors and regression models to fill in the missing parts of the data, ultimately yielding cleaned measurement data.

[0054] After data cleaning, the cleaned gas measurement data needs to be dimensionally unified and normalized. The core purpose of dimensional unification is to ensure that all data are within the same unit system, which is especially crucial for measurement data from different sensors. Gas concentrations collected by different sensors may use different units, and environmental parameters such as temperature, humidity, and air pressure also have different dimensions. Therefore, all data must be converted to a unified dimension to lay the foundation for subsequent analysis. Normalization is then performed, mapping all data to a fixed standard range, typically between 0 and 1. Through these two steps, standard measurement data are generated, completely eliminating biases caused by dimensional differences in different characteristics.

[0055] After data standardization, the next stage is response feature data extraction. The core objective of this stage is to screen and extract the most valuable features for subsequent prediction and analysis from the gas measurement data. For example, key features reflecting the dynamic behavior of the gas can be extracted from multiple dimensions such as gas concentration changes, sensor output fluctuations, and gas diffusion processes. During extraction, environmental factors such as temperature, humidity, and air pressure need to be organically combined with the gas measurement data to extract response feature data that includes environmental characteristics. This not only obtains information on gas concentration changes but also incorporates the influence of environmental factors, resulting in more comprehensive and complete response feature data. After the response feature data extraction is complete, the next step is environmental correlation analysis. The core of environmental correlation analysis is to explore the intrinsic relationship between gas behavior and environmental factors, focusing on the influence mechanism of environmental states such as temperature, humidity, and air pressure on gas response. During the analysis, statistical analysis methods such as correlation analysis, regression analysis, and principal component analysis are first used to screen out the environmental factors most strongly correlated with gas concentration changes. Then, mathematical models are established or machine learning methods are used to further quantify the relationship between environmental factors and gas response, generating environmental correlation data. This data can clearly reveal the changing patterns of the gas under different environmental conditions.

[0056] Finally, by combining response feature data with environmental correlation data, sample mapping and classification were carried out. This step aims to systematically categorize response samples under different conditions, thereby generating a standardized set of real response samples. During the classification process, each sample is mapped to a corresponding category or numerical range based on its own key information such as gas concentration, sensor output, and environmental factors. This classification method can further optimize the subsequent model training effect, allowing the model to more accurately identify the behavior patterns of gases under different environments and more accurately predict the trend of gas concentration changes. After the above series of complete steps, the final generated set of real response samples provides high-quality training data support for subsequent modeling and analysis.

[0057] In this solution, by integrating historical gas measurements, environmental conditions, and response data, a model is constructed and trained to accurately predict the behavior of flammable and explosive gases, thereby improving the accuracy and real-time performance of the measurement results.

[0058] S102, Based on the real response sample set, model the response pattern to obtain the basic measurement model.

[0059] The basic measurement model is a preliminary mathematical and statistical model constructed through a systematic analysis and modeling process, based on a real response sample set. The core objective of this model is to accurately describe and reliably predict the response patterns of flammable and explosive gases under specific environmental conditions, revealing objective patterns of gas behavior by quantifying the intrinsic correlation between gas concentration changes and various environmental factors.

[0060] The modeling work is based on valid data from the real response sample set. The core task is to uncover the intrinsic relationship between gas concentration and environmental factors, focusing on how to accurately predict the behavioral characteristics and concentration variation patterns of flammable and explosive gases based on environmental factors such as temperature, humidity, and wind speed. In the initial stage of modeling, classical statistical methods such as regression analysis are used to systematically analyze the correlation between gas concentration data and key environmental parameters, deeply clarifying the intrinsic relationship between environmental factors and gas concentration, laying a solid foundation for subsequent predictive modeling work. Simultaneously, the variation patterns of gas concentration and the influence weights of environmental parameters on gas concentration are quantified to ensure that the specific impacts of various environmental conditions on gas behavior are realistically and accurately reflected.

[0061] After clarifying the relationship between environmental factors and gas concentration, a preliminary mathematical model was constructed, with the core objective of making the gas response patterns concrete and quantifiable. Appropriate mathematical methods were selected based on the data characteristics; multiple linear regression, decision trees, support vector machines, and neural networks are commonly used choices. Environmental feature data extracted from real response sample sets were input into the model, allowing it to fully capture the complex patterns of gas concentration changes with environmental factors. Specifically, the model uses key environmental factors such as temperature, humidity, and wind speed as core input variables. Through the analysis and processing of this data, it achieves accurate prediction of the changing trends of flammable and explosive gas concentrations.

[0062] After the initial model is built, validation and optimization are essential to ensure its practicality. To ensure the model can adapt to data under different environmental conditions and achieve accurate predictions, techniques such as cross-validation are introduced to comprehensively and systematically evaluate the model's accuracy and stability. During validation, the real response sample set is divided into a training set and a test set. The training set is used for model training, refinement, and parameter fitting, while the test set is specifically used to test the model's prediction accuracy. This ensures that the model can still accurately predict gas concentrations when faced with new, unseen data, avoiding excessive prediction bias.

[0063] The model refinement and optimization phase focuses on adjusting hyperparameters, such as the learning rate and regularization coefficient, to further improve the model's prediction accuracy. Depending on actual needs, more complex deep learning models may be introduced to enrich the model's fitting capabilities. Simultaneously, by combining model error analysis results, performance shortcomings under certain special environmental conditions are accurately identified, and targeted improvements and optimizations are carried out. Through this series of refinements and optimizations, not only can the model's prediction accuracy be significantly improved, but its adaptability can also be enhanced, enabling it to cope with complex and ever-changing real-world application scenarios. After the complete process of modeling, verification, and optimization described above, the final basic measurement model can accurately predict the concentration change trends of flammable and explosive gases under different environmental conditions. Based on this model, dynamic changes in gas concentration can be quickly inferred from real-time collected environmental data, ensuring timely detection of abnormal changes in flammable and explosive gases in potentially hazardous areas.

[0064] S103, acquire low-coverage operating condition data from the basic measurement model, perform operating condition deduction based on the low-coverage operating condition data, and obtain a virtual operating condition set.

[0065] Low-coverage operating condition data refers to data gaps that arise during actual measurements due to limitations imposed by gas concentration, environmental factors, and other constraints, making it impossible to achieve complete coverage of all operating condition combinations. This type of data corresponds to operating condition information that could not be effectively collected under normal monitoring conditions, and its typical characteristic is the lack of sufficient observational support for gas behavior responses under specific operating conditions.

[0066] A virtual operating condition set is a collection of operating condition data constructed based on existing low-coverage operating condition data through methods such as numerical simulation, model extrapolation, and hypothetical experiments. Addressing the limitation of low-coverage operating condition data in comprehensively covering various operating scenarios, virtual operating condition sets extend and extrapolate from existing data, simulating possible operating conditions under different environmental and gas concentration variations. These virtual operating condition sets not only effectively expand the coverage dimensions of existing data but also reproduce extreme conditions that are difficult to observe in reality.

[0067] To obtain low-coverage operating condition data from the basic measurement model, it is necessary to start by tracing back the construction process of the basic measurement model. The basic measurement model is generally trained based on a real response sample set, combined with machine learning, regression analysis, deep learning, and other statistical methods. It can accurately describe the behavior of flammable and explosive gases under different environmental conditions, while clearly presenting the complex correlation between gas concentration and various environmental factors. Due to the dual constraints of the diversity of operating conditions and the complexity of environmental conditions, the response data produced by the basic measurement model cannot achieve comprehensive coverage of all operating conditions. Especially under certain specific extreme conditions, relevant measurement data are often scarce, or even difficult to collect effectively. Low-coverage operating condition data refers precisely to the deficiencies and missing parts of the model's response data under these scarce or insufficiently monitored operating conditions.

[0068] To obtain low-coverage operating condition data, in-depth analysis of existing basic measurement models is necessary to identify operating condition types that have not been fully observed in historical measurements. Specifically, comparing model predictions with actual collected response data can clarify which combinations of gas concentrations and environmental factors are not adequately covered by actual measurements. Based on the analysis of these gaps in operating condition data, the types of operating conditions missing from the model can be further determined, and the location and type of low-coverage operating condition data can be accurately labeled. After identifying low-coverage operating condition data, operating condition extrapolation needs to be conducted based on this data to generate a virtual operating condition set. The core of this process lies in using the model's own reasoning capabilities to predict and supplement missing data. The basic measurement model has already established a correlation between gas concentration and environmental factors based on historical data. This correlation pattern can be applied to low-coverage operating condition areas, using existing data to conduct extended extrapolations. For example, in scenarios with extremely low or high gas concentrations and significantly abnormal environmental conditions, the model can perform reasonable extrapolation based on internal rules to predict the possible responses of gases under these operating conditions.

[0069] Operating condition simulations are typically conducted using numerical simulation techniques. First, one or more low-coverage operating condition areas are selected. Then, based on a fundamental measurement model, initial conditions are set, and simulations are performed on different combinations of gas concentrations and environmental conditions. During this process, environmental factors such as temperature, humidity, and air pressure can be dynamically adjusted, combined with trends in gas concentration changes, to simulate the specific impact of these factors on gas reactions. The simulation output forms a set of virtual operating condition data. This data effectively fills gaps in previously unobserved operating condition scenarios and maintains a high degree of consistency with the model's response patterns. The core requirement of operating condition simulations is ensuring that the generated virtual operating condition set is sufficiently representative and conforms to reality. For example, the virtual operating condition set can be compared with a real response sample set, or validation can be achieved by combining model predictions under known operating conditions. These virtual operating conditions can fill gaps in low-coverage operating condition data.

[0070] Based on the above technical solution, optionally, operating condition simulation can be performed based on the low-coverage operating condition data to obtain a virtual operating condition set, including:

[0071] Based on the low-coverage operating condition data, the change trend is extracted to obtain the operating condition change trend data;

[0072] Based on the operating condition change trend data, the rate of change is calculated and the direction of change is determined to obtain the operating condition evolution characteristics.

[0073] Based on the aforementioned operating condition evolution characteristics, parameter value boundaries are determined to obtain operating condition boundary data.

[0074] Based on the low-coverage operating condition data and the operating condition boundary data, parameter linkage relationship analysis is performed to obtain operating condition correlation data.

[0075] Based on the aforementioned working condition correlation data, parameter linkage constraints and evolution rules are established, and working condition deduction rules are generated based on the parameter linkage constraints and evolution rules.

[0076] Based on the aforementioned operating condition extrapolation rules, the low-coverage operating condition data is continuously expanded to generate a candidate operating condition set with coverage operating condition boundaries and extrapolation rule constraints.

[0077] The candidate working condition set is compared with the working condition boundary data, and working conditions that exceed the boundary are eliminated to obtain the boundary compliant working condition set.

[0078] Based on the verification and linkage relationship between the boundary compliance working condition set and the working condition associated data, inconsistent working conditions are eliminated to obtain the associated consistent working condition set;

[0079] Based on the aforementioned consistent operating condition set, abnormal operating conditions are eliminated and duplicate operating conditions are merged to obtain a virtual operating condition set.

[0080] In this solution, the operating condition trend data is extracted and generated based on low-coverage operating condition data. The core data reflects the patterns and directions of changes in operating condition parameters over time or other related factors. This data can clearly reveal the dynamic changes in various operating condition parameters, such as gas concentration over time.

[0081] Operating condition evolution characteristics are the core features extracted from data on operating condition change trends to describe the process of operating condition change, including key attributes such as the rate and direction of change. They visually present the evolution of operating conditions over time and clearly reveal the changing patterns of gas behavior under different environmental conditions. For example, by analyzing operating condition evolution characteristics, it can be determined whether the effect of temperature change on gas concentration is increasing or decreasing.

[0082] Operating condition boundary data is calculated based on the characteristics of operating condition evolution and other relevant data. It is used to clearly indicate the acceptable range of operating condition parameters. Its function is to define operating condition attributes, distinguishing which operating conditions fall within the reasonable range of normal operating conditions and which have deviated from the normal range. Operating condition boundary data ensures that during the process of operating condition extrapolation and expansion, the selected operating conditions conform to the established physical or technical limitations, avoiding invalid operating conditions that exceed the actual measurable range.

[0083] Operating condition correlation data is derived from low-coverage operating condition data and other relevant influencing factors by analyzing the interactions between operating condition parameters. These relationships may exhibit linear or non-linear dependencies, such as the correlation between temperature changes and gas concentrations.

[0084] Parameter linkage constraints are constraints used to describe the mutual influence and restriction relationships between operating parameters. The core purpose of these constraints is to ensure that the correlation between different parameters always follows objective laws during the operating condition simulation process. For example, the objective law that gas concentration will likely increase with increasing temperature can be transformed into a linkage constraint between temperature and concentration.

[0085] Evolutionary rules are extracted from the characteristics of operating condition evolution and are used to clearly describe the specific patterns of operating condition changes over time, including the direction and rate of change of the operating condition, as well as the influence of various environmental factors that may affect it. The core application scenario of evolutionary rules is to deduce future trends in operating condition changes based on low-coverage operating condition data.

[0086] Operating condition simulation rules are a set of rules formed by integrating parameter linkage constraints and evolution rules. Their core purpose is to deduce future or uncovered operating condition scenarios from low-coverage operating condition data. These rules are based on historical data and physical models to predict possible operating condition states and thus generate a candidate operating condition set.

[0087] The candidate operating condition set is a group of potential operating conditions obtained by expanding low-coverage operating condition data based on operating condition extrapolation rules. All operating conditions in this set meet the extrapolation rules and boundary constraints, including both routine operating condition data and possible extreme operating conditions as well as operating conditions that could not be directly collected in actual measurements.

[0088] The boundary compliance set is a set of compliant operating conditions obtained by comparing each candidate operating condition with the operating condition boundary data and eliminating operating conditions that exceed the set boundaries. All operating conditions in this set meet the predefined physical or technical boundary conditions.

[0089] The consistent operating condition set is obtained by comparing and verifying the operating condition correlation data with the boundary compliant operating condition set, and eliminating operating condition data with inconsistent parameter correlations. The operating conditions in this set not only meet the boundary requirements, but the correlation between the parameters also highly matches the actual data model.

[0090] The core application of low-coverage operating condition data is trend extraction. Time series analysis of this type of data can effectively identify the changing patterns of operating parameters. The core objective of trend extraction is to clarify the changing characteristics of various operating parameters under different time periods and conditions; typical parameters include temperature, humidity, and gas concentration. This process can incorporate the sliding window technique, dividing the overall data into several continuous data segments and analyzing the changing trends of each segment. Simultaneously, by combining trend detection methods such as regression analysis and exponential smoothing, the data trends are fitted to ultimately obtain the changing trend data of the operating conditions.

[0091] Based on the extracted operating condition trend data, further calculations of the rate of change and determination of the direction of change are needed. By comparing the magnitude of changes in operating condition parameters over different time periods, the rate of change of each parameter can be quantitatively calculated, and its direction of change can be clarified. Taking temperature as an example, if the temperature shows a continuous upward trend over a certain period, the direction of change of this parameter is determined to be positive; otherwise, it is determined to be negative. The rate of change can be calculated using the first-order difference method or gradient descent method, combined with the time interval of parameter changes, to achieve an accurate measurement of the rate of change of operating condition parameters. The core purpose of this step is to capture the dynamic characteristics of the changes in operating condition parameters and, based on the rate and direction of change, further analyze the overall evolution trend of the operating conditions. After completing the analysis of the operating condition evolution characteristics, the parameter value boundaries need to be determined based on the above characteristics. By integrating the analysis results of the rate and direction of change, combined with the physical constraints and actual environmental requirements of the operating condition parameters, the reasonable value range of each operating condition parameter is determined, i.e., the operating condition boundary data. This type of boundary data not only includes the minimum and maximum values ​​of each parameter, but also fully considers the mutual constraints between different operating condition parameters. For example, gas concentrations must be strictly controlled within safe thresholds, and the combination of temperature and humidity parameters must also be maintained within a safe operating range.

[0092] By combining low-coverage operating condition data and operating condition boundary data, parameter linkage analysis is conducted. The core of this step is to analyze the mutual influence mechanism between various operating condition parameters and clarify their correlations. Taking common operating condition parameters as an example, rising temperature may cause a synchronous increase in gas concentration, while rising humidity may affect gas diffusion rate. By constructing mathematical analysis models, the inherent linkage relationships between different operating condition parameters can be deeply revealed. The obtained correlation data can provide crucial support for subsequent operating condition extrapolation and prediction. Parameter linkage analysis can utilize statistical methods such as regression analysis, correlation calculation, and principal component analysis to ensure the scientific validity and accuracy of the analysis results. Based on the acquired operating condition correlation data, further parameter linkage constraints and evolution rules need to be established. These constraints and evolution rules are generated based on the results of the previous parameter linkage analysis and are used to describe the interaction patterns between different operating condition parameters and the evolutionary laws of parameters over time. For example, some operating condition parameters need to maintain synchronous changes within a specific range to maintain the stable operation of the system. These requirements will serve as explicit constraints to guide subsequent operating condition extrapolation. Meanwhile, the evolution rules need to clearly define the specific patterns of operating conditions changing over time. For example, when the temperature is within a certain range, the gas concentration will change according to a preset rate. The construction of the above constraints and evolution rules should be based on historical operating condition data, physical laws, and statistical analysis results.

[0093] After establishing the operating condition extrapolation rules, the low-coverage operating condition data can be continuously expanded according to these rules to generate a candidate operating condition set. Generating the candidate operating condition set involves applying the operating condition extrapolation rules to the existing low-coverage data, using extrapolation calculations to obtain possible future operating condition scenarios, or to supplement operating condition types not covered in actual measurements. These operating conditions include extreme operating conditions, extremely low-frequency operating conditions, and special operating conditions that are difficult to collect in actual measurements. During the operating condition expansion process, the established extrapolation rules must be continuously implemented to ensure that the generated candidate operating conditions meet physical constraints and logical requirements. After the candidate operating condition set is generated, it needs to be comprehensively compared with the operating condition boundary data, and operating condition data exceeding the boundary range should be eliminated. The candidate operating condition set may contain some operating condition parameter values ​​that exceed the preset boundaries; these operating conditions are not feasible or measurable in actual application scenarios and therefore need to be eliminated.

[0094] After screening the boundary compliance work condition set, a second verification is required using work condition correlation data to ensure consistency in the parameter linkage relationships within each work condition. Some work conditions may have unreasonable parameter linkage relationships, leading to logical contradictions in the work condition data; these work conditions also need to be removed. The core objective of this verification process is to ensure that the linkage relationships between parameters within the final work condition set are consistent, avoiding the introduction of logically contradictory work condition scenarios. After obtaining a consistent work condition set, abnormal work condition removal and duplicate work condition merging are carried out to ultimately generate a virtual work condition set. Abnormal work conditions refer to extreme work conditions, erroneous work conditions, or work conditions that do not conform to physical laws generated during the work condition simulation process; all such work conditions must be removed. Duplicate work conditions refer to work condition data with duplicate or redundant content generated during the work condition expansion process; these must be merged. After the above series of screening and optimization operations, the final virtual work condition set includes a set of stable work condition data that is highly consistent with actual work conditions.

[0095] This approach can expand to extreme and scarce operating conditions when samples are insufficient, thereby improving operating condition coverage, parameter linkage integrity, and subsequent model training accuracy.

[0096] S104, input the virtual operating condition set into the basic measurement model for response simulation to obtain a virtual mixed sample set.

[0097] A virtual hybrid dataset is a dataset generated through response simulation after a virtual operating condition set is input into the basic measurement model. It includes multi-dimensional data such as gas concentration, sensor response, and environmental disturbances obtained from the virtual operating conditions. This data, based on the operating conditions derived from the model, fills in the range of operating conditions not covered in actual measurements. Its purpose is to expand the sample library, providing richer sample data support, especially for extreme or unobserved operating conditions, further enhancing the model's generalization ability under various operating conditions.

[0098] Virtual operating condition sets are input into the basic measurement model to conduct response simulations. Each element of the virtual operating condition set corresponds to a hypothetical scenario, carrying information on the gas and environmental states under different conditions. This type of virtual operating condition data originates from low-coverage operating condition data in the basic measurement model and is generated through an operating condition extrapolation process, covering areas that were not collected in the original data, especially for those extreme or difficult-to-measure operating conditions.

[0099] Once the virtual operating condition set is generated, the input process to the basic measurement model can begin. The basic measurement model, trained on a set of real response samples, serves as a mathematical model capable of predicting gas behavior under different operating conditions. It can receive various parameter inputs from the virtual operating condition set. After each operating condition parameter is input into the model, the model will simulate changes in gas concentration, fluctuations in sensor output, and the specific impacts of environmental factors on these changes, based on the learned response patterns. During the response simulation, the model will calculate the corresponding gas concentration, sensor response, and environmental impact results based on the input operating condition data.

[0100] After simulation, each output data point of the virtual operating condition constitutes an independent simulation sample, which is then fused with the actual collected data. The generated simulation samples include multi-dimensional data, including information related to gas concentration, sensor response, and environmental disturbances under each virtual operating condition. Because these simulation samples are generated based on virtual operating conditions, they can expand the sample range. After completing the response simulation, each output data point of the virtual operating condition becomes a simulation sample. Each simulation sample represents a multi-dimensional dataset of gas concentration, sensor response, and environmental factors under a specific virtual operating condition. This data is fused with the real response sample data obtained from actual measurements to form a set including real and virtual samples; this is the virtual hybrid sample set.

[0101] Based on the above technical solution, optionally, the virtual operating condition set is input into the basic measurement model for response simulation to obtain a virtual mixed sample set, including:

[0102] Each virtual operating condition in the virtual operating condition set is input into the basic measurement model to obtain the corresponding initial response result;

[0103] Based on the initial response results, gas concentration response data, sensor output response data, and environmental disturbance response data are extracted. Based on the gas concentration response data, sensor output response data, and environmental disturbance response data, operating condition response characteristic data are generated.

[0104] Based on the operating condition response feature data, time series combination and multi-parameter mapping are performed to obtain the simulated response samples corresponding to each virtual operating condition.

[0105] Based on the association and annotation of the simulated response samples and the corresponding operating parameters of the virtual operating conditions, virtual annotation samples are obtained;

[0106] Based on the virtual labeled samples, a sample aggregation is performed to obtain a virtual hybrid sample set.

[0107] In this solution, the virtual operating conditions are flammable and explosive gas test states generated based on low-coverage operating condition data, operating condition boundary data, and operating condition extrapolation rules. These test states have not been actually collected, but fully meet all constraints. Each virtual operating condition includes a combination of multi-dimensional parameters such as gas composition, concentration of each component, temperature, humidity, pressure, flow rate, background gas ratio, sensor operating voltage, and sampling time.

[0108] The initial response result is the predicted output data obtained after inputting a virtual operating condition into the basic measurement model. This result is used to characterize the theoretical response of the sensor to flammable and explosive gases under the corresponding virtual operating condition, specifically including the predicted concentration value, predicted output voltage, predicted resistance change, predicted current value, and the corresponding time variation curve.

[0109] Gas concentration response data is extracted from the initial response results and is directly related to changes in the concentration of flammable and explosive gases. Its core function is to reflect the response amplitude, response increment, concentration-output correspondence, and concentration change trend of target gases such as methane, hydrogen, carbon monoxide, and acetylene under different concentration conditions.

[0110] Sensor output response data consists of the sensor's own electrical response data recorded in the initial response results. This type of data is used to reflect the changing characteristics of the sensor's output signal under different virtual operating conditions, including key indicators such as output voltage, output current, resistance change rate, peak response, recovery value, response time, and recovery time.

[0111] Environmental disturbance response data are additional response data in the initial response results caused by changes in environmental factors. Environmental factors specifically include temperature, humidity, pressure, airflow velocity, background oxygen content, etc. This type of data is used to describe the degree of influence of environmental disturbances on the detection results of flammable and explosive gases, as well as the resulting output signal offset, drift, or amplification effect.

[0112] Operating condition response characteristic data is a comprehensive feature data formed by integrating gas concentration response data, sensor output response data, and environmental disturbance response data. Its core purpose is to uniformly describe the coupled response relationship between flammable and explosive gases, environmental factors, and sensor outputs under a certain virtual operating condition. Specifically, it includes response peak value, slope of change, fluctuation amplitude, settling time, drift amount, and multi-parameter combination characteristics.

[0113] Simulated response samples are sample data formed by performing time series arrangement and multi-parameter mapping processing on operating condition response characteristic data. These samples are used to simulate the continuous response process of sensors to flammable and explosive gases during real detection processes. Each sample includes the operating condition parameters, sensor output values, and environmental change information at the corresponding time.

[0114] Operating parameters are a set of parameters used to describe the specific state of a virtual operating condition. They include the type of target gas, the concentration of each gas, temperature, humidity, pressure, background gas ratio, flow rate, sensor operating status, sampling duration, and gas mixing ratio.

[0115] Virtual labeled samples are labeled samples formed by associating and matching simulated response samples with corresponding operating condition parameters. The simulated response samples serve as input features, and the operating condition parameters serve as label information. These samples are primarily used for training subsequent flammable and explosive gas identification models or concentration prediction models.

[0116] Each virtual operating condition in the virtual operating condition set is input into the basic measurement model one by one. Each virtual operating condition corresponds to a unique combination of gas concentration and environmental conditions. Before inputting into the model, the data needs to be standardized. The gas concentration in each virtual operating condition is converted into a concentration vector, and environmental parameters such as temperature, humidity, and pressure are normalized to ensure the standardization and comparability of various data. The standardized data sequence after the above processing can be used as valid input into the basic measurement model. After receiving the standardized input data, the basic measurement model performs calculations based on the trained response rules and finally outputs the corresponding initial response results. These initial response results mainly reflect two aspects: first, the sensor output changes caused by changes in gas concentration, such as fluctuations in voltage and current; and second, the impact of environmental disturbances on the sensor response. Throughout the calculation process, the model can accurately simulate the gas behavior and sensor response characteristics under different operating conditions, comprehensively presenting the dynamic relationship between gas concentration, sensor response, and environmental factors.

[0117] Based on the initial response results of each virtual operating condition, three types of core data are further extracted: gas concentration response data, sensor output response data, and environmental disturbance response data. Gas concentration response data includes the response amplitude, response speed, and reaction time caused by concentration changes; sensor output response data includes changes in voltage, current, and resistance, as well as the rate of change of these parameters and sensor recovery time; environmental disturbance response data mainly reflects the impact of temperature, humidity, pressure, and background oxygen content on sensor response, specifically including the output offset and drift caused by changes in environmental factors. After extraction, these three types of data undergo unified time-series processing, are aligned according to time nodes, and the response peak, rate of change, and drift rate at each time node are calculated, ultimately forming operating condition response characteristic data with time-series features.

[0118] For the operating condition response characteristic data, time-series combination is performed according to a preset sampling period to construct a multi-dimensional feature vector corresponding to each virtual operating condition. To fully account for the comprehensive impact of multiple operating conditions on the sensor response, a multi-parameter mapping method is adopted to integrate gas concentration data, sensor output data, and environmental disturbance data from different dimensions into the same feature space. Through this integration method, the sensor response patterns under different operating conditions can be effectively characterized, thereby generating simulated response samples.

[0119] Each simulated response sample is associated with and labeled with parameters of the corresponding virtual operating condition. Specifically, operating parameters such as gas type, gas concentration, temperature, humidity, and pressure in the virtual operating condition are used as labels and labeled into the corresponding simulated response samples, forming virtual labeled samples. In this way, each simulated response sample not only includes complete sensor response data but also comes with closely related operating parameters, achieving a precise correspondence between response data and operating conditions. Finally, all generated virtual labeled samples are uniformly summarized and processed. During this process, duplicate samples and invalid samples with missing labels are removed, retaining valid and diverse samples. The remaining samples are then reorganized according to criteria such as gas type, concentration range, and environmental conditions, ultimately generating a virtual mixed sample set.

[0120] This solution can generate multiple types of response data when real explosion condition samples are lacking, thereby improving the training coverage and prediction accuracy of the combustible explosive gas identification model.

[0121] S105, based on the virtual mixed sample set and the real response sample set, sample consistency evaluation and credibility weighting are performed to obtain the enhanced training sample set.

[0122] The enhanced training sample set is a new sample set formed by integrating the real response sample set and the virtual hybrid sample set, and then performing consistency evaluation and credibility weighting. It includes gas response data and environmental state data obtained from actual measurements, as well as virtual data generated through response simulation. This virtual data fills in the data gaps that may exist in the actual sampling process and covers extreme operating conditions that are difficult to collect directly.

[0123] First, it's crucial to define the evaluation criteria for consistency. Judging sample consistency primarily relies on data matching and similarity. The evaluation is based on whether the response performance of virtual and real samples matches under the same environmental conditions. If a virtual sample can closely approximate the response characteristics of a real sample under specific conditions, it indicates high reliability; conversely, if it fails to do so, its reliability is low. During the evaluation process, Euclidean distance and Manhattan distance are commonly used metrics. By comparing the differences between virtual and real samples in the parameter space, the level of consistency between the two can be accurately determined, thereby identifying virtual samples that deviate from the patterns of real data.

[0124] After the consistency assessment is completed, the reliability weighting stage begins. Based on the results of the consistency assessment, each sample is assigned a corresponding weight. Virtual samples that closely match real samples will be given higher weights, while samples that deviate significantly from real patterns will be given lower weights. This weighting mechanism effectively avoids interference from low-reliability data in model training, ensuring the efficient use of real samples and reliable virtual samples. Various weighting strategies can be used, such as linear weighting, exponential weighting, or distance-based weighting; the specific choice depends on the characteristics of the data and the actual application scenario. After sample consistency assessment and reliability weighting, an enhanced training sample set is finally formed. This sample set integrates real data obtained from actual measurements with weighted virtual data, effectively enhancing the model's generalization ability and enabling it to output more accurate predictions and judgments when dealing with complex, variable, and uncertain conditions.

[0125] Based on the above technical solution, optionally, sample consistency evaluation and credibility weighting are performed on the virtual mixed sample set and the real response sample set to obtain an enhanced training sample set, including:

[0126] Based on the comparison of feature distributions between the virtual mixed sample set and the real response sample set, sample difference data corresponding to each virtual sample is obtained.

[0127] Based on the sample difference data, a sample consistency assessment is performed to obtain a consistency score for each virtual sample.

[0128] Based on the consistency score and the preset confidence threshold, virtual samples with consistency scores lower than the preset confidence threshold are filtered out to obtain an effective set of virtual samples.

[0129] Based on the consistency score, weights are assigned to each virtual sample in the effective virtual sample set to obtain the credibility weight corresponding to each virtual sample.

[0130] The effective virtual sample set and the real response sample set are weighted and fused based on the credibility weight to obtain the enhanced training sample set.

[0131] In this scheme, a virtual sample is a single sample within a virtual mixed sample set. It is generated based on a virtual operating condition of flammable and explosive gases and has been labeled. Essentially, it is a simulated response sample. Its core components include three types of response data: gas concentration response, sensor output response, and environmental disturbance response, as well as corresponding operating condition parameter labels.

[0132] Sample difference data is obtained by comparing the feature distributions of a virtual sample with those of the corresponding samples in the real response sample set. It is used to characterize the degree of deviation between the two types of samples in many aspects, including gas concentration response, sensor output, environmental disturbance, time series change trend, and characteristic statistics. Common deviation indicators include mean difference, variance, peak deviation, response slope deviation, time delay deviation, and distribution distance.

[0133] The consistency score is a numerical indicator calculated based on sample difference data. Its purpose is to measure how closely a virtual sample resembles a real response sample. A higher score indicates a closer match between the virtual and real sample's response patterns, and thus a higher level of sample reliability.

[0134] The preset credibility threshold is a pre-defined minimum standard for consistency scoring, specifically used to determine whether virtual samples have retention value. When the consistency score of a virtual sample is lower than this threshold, it means that it does not conform to the actual response pattern and must be discarded.

[0135] The effective virtual sample set is a collection of virtual samples retained after consistency scoring. This set includes only virtual samples with a consistency score reaching or exceeding a preset confidence threshold, and can accurately reflect the response patterns in the actual detection process of flammable and explosive gases.

[0136] Credibility weights are weights assigned to a sample based on its consistency score in the virtual sample set. They are used to define the importance of the sample in subsequent model training. The higher the consistency score, the greater the corresponding credibility weight, and the stronger the sample's influence on augmenting the training sample set.

[0137] Each virtual sample in the virtual mixed sample set is matched and compared with the corresponding features of the real response sample set. During the comparison, the focus is on the flammable and explosive gas concentration response, sensor output signals, and environmental state parameters. Quantitative calculations are performed from multiple key dimensions, including mean, variance, peak value, and response time, to obtain the sample difference data for each virtual sample. The core function of this data is to accurately quantify the degree of deviation between the virtual sample and the real sample in the feature space. After calculating the sample difference data, it is first normalized to eliminate the influence of different dimensions of data and ensure the reasonableness of the evaluation results. Subsequently, methods such as weighted summation, cosine similarity, or statistical distance are used to comprehensively evaluate the consistency between each virtual sample and the real response sample, generating a corresponding consistency score.

[0138] Each virtual sample's consistency score is compared one by one with a preset credibility threshold. Virtual samples with consistency scores below the threshold are discarded, retaining only those with scores at or above the threshold, thus forming a valid virtual sample set. This ensures the reliability of subsequent model training samples from the outset. Within the valid virtual sample set, credibility weights are assigned to each virtual sample based on its consistency score; samples with higher consistency scores receive greater credibility weights. Finally, the weighted valid virtual sample set is merged with the real response sample set. This integrates the feature distributions of both types of samples while fully preserving the representativeness of each type, ultimately forming an enhanced training sample set.

[0139] In this approach, the training sample set is enhanced to have both real and virtual data characteristics, thereby improving the model's coverage of various operating conditions and prediction accuracy, as well as its generalization ability and reliability.

[0140] S106, The basic measurement model is trained by self-learning based on the enhanced training sample set to obtain a self-learning measurement model.

[0141] Self-learning measurement models are models that can automatically adjust and continuously optimize based on new data and feedback. These models continuously learn from newly added sample data, uncovering and identifying complex relationships between data points, thereby iteratively optimizing the original model. The core of the self-learning process lies in the deep mining and learning of the characteristics and inherent patterns in the sample set, prompting the model to gradually improve accuracy and environmental adaptability in the prediction and decision-making stages. In flammable and explosive gas monitoring scenarios, the model can autonomously adjust relevant parameters based on real-time measurement data and historical response samples, continuously improving measurement accuracy and operational reliability, and better adapting to actual monitoring needs.

[0142] Once the enhanced training sample set is obtained, the self-learning training process of the basic measurement model can be initiated. Training typically begins with model parameter initialization. The basic measurement model itself has a pre-defined structure and algorithm used to capture the intrinsic relationship between gas behavior and environmental changes. However, the initial model is difficult to accurately adapt to all operating conditions and requires continuous training and optimization to gradually improve its performance. The first step in training is data input. After various types of data from the enhanced training sample set are fed into the model, the model will provide preliminary predictions and responses based on features such as input gas concentration, sensor output, and environmental conditions. During the input process, the model needs to preprocess the data according to its format and standardize different types of data to ensure that various data sources participate in training on the same scale. These preprocessed data, through the model's forward propagation process, combined with the calculation of weights and biases, output preliminary measurement results.

[0143] After the data input is complete, the model evaluates the error between the predicted results and the actual measurement results; this is the core step of self-learning training. Error calculation is mainly achieved through loss functions, with mean squared error and cross-entropy loss being commonly used types to quantify the degree of deviation between the model's predictions and the actual results. Based on the backpropagation algorithm, the model adjusts each parameter accordingly based on the calculated error, gradually optimizing the weights and biases. This process requires multiple iterations, with each iteration improving the model's prediction accuracy. As training iterations continue, the model learns, adjusts, and optimizes on augmented training sample sets, and each backpropagation of error makes the model's measurement results within a given data range closer to the true values.

[0144] During training, adaptive adjustment capability is particularly crucial. When new data sources are input, the model can autonomously adjust its internal parameters based on this new data to cope with new measurement scenarios. For example, when the system detects new environmental state parameters, the model will automatically adjust its parameters based on learned patterns, ensuring the reliability and accuracy of the prediction results. To avoid overfitting, regularization techniques may also be introduced during training. Overfitting often manifests as a model performing well on training data but experiencing a significant drop in prediction accuracy on unknown data. Regularization effectively suppresses overfitting by penalizing model complexity, helping the model maintain good generalization ability.

[0145] After self-learning training, the basic measurement model undergoes multiple rounds of comprehensive optimization, ultimately transforming into a self-learning measurement model. This model can not only make accurate predictions for known operating conditions but also effectively handle new and uncertain operating scenarios, significantly improving the accuracy and reliability of flammable and explosive gas measurements. This self-learning measurement model can be further applied to the processing and analysis of real-time measurement data, helping the monitoring system to respond more accurately to changes in environmental conditions. Through this self-learning process, the basic measurement model iterates from an initial preset model into a highly adaptive measurement tool, capable of flexibly handling various complex measurement tasks based on changes in gas concentration and fluctuations in environmental factors, fully adapting to the diverse needs of actual monitoring.

[0146] Based on the above technical solution, optionally, the basic measurement model can be self-learned and trained using an enhanced training sample set to obtain a self-learning measurement model, including:

[0147] Based on the enhanced training sample set, the training samples are divided to obtain a training subset, a validation subset, and a test subset;

[0148] The training subset is input into the basic measurement model for iterative parameter training to obtain an initial self-learning measurement model.

[0149] The validation subset is input into the initial self-learning measurement model for prediction validation to obtain the model prediction result;

[0150] Error calculation is performed based on the model prediction results and the real labels corresponding to the validation subset to obtain model error data;

[0151] Based on the model error data, the model parameters of the initial self-learning measurement model are adaptively adjusted, and parameter iterative training is re-executed until the model error data meets the preset convergence condition, thereby obtaining the optimized self-learning measurement model.

[0152] The test subset is input into the optimized self-learning measurement model to test its generalization ability, and the model test results are obtained.

[0153] If the model test results meet the preset measurement accuracy requirements, the optimized self-learning measurement model is determined to be a self-learning measurement model.

[0154] In this scheme, the training subset is a data set divided from the augmented training sample set for model parameter fitting and iterative training, which carries the main information for the model to learn the gas response law.

[0155] The validation subset is a set of data partitioned from the augmented training sample set to evaluate model performance during training, calculate errors, and guide model parameter tuning.

[0156] The test subset is a set of data that is partitioned from the augmented training sample set to evaluate the model’s generalization ability after training is completed, and to verify the model’s prediction accuracy on unseen data.

[0157] The initial self-learning measurement model is a measurement model formed by iterative training of preliminary parameters based on a training subset, used to predict gas response and perform error assessment.

[0158] The model prediction results are the measurement response of the initial self-learning measurement model to the input data of the validation subset, which are used to compare the error with the true label.

[0159] Real labels are actual measurements of combustible and explosive gases in the validation or test subsets, serving as a standard reference for model training and evaluation.

[0160] Model error data is the difference between the model's predictions and the true labels, used to assess model accuracy and guide parameter adjustments.

[0161] Model parameters are adjustable values ​​used within a self-learning measurement model to describe the gas response, including weights, biases, or other mathematical model coefficients.

[0162] The preset convergence condition is a set standard for terminating model training, such as error data falling below a specific threshold or continuous iteration error change being less than a specified value, used to determine whether model training is complete.

[0163] The model testing results are the prediction outputs and accuracy evaluations of the optimized self-learning measurement model on the test subset, used to measure the model's generalization ability.

[0164] The preset measurement accuracy requirement is a set standard for the final prediction accuracy of the model, used to determine whether the optimized model meets the actual application requirements for measuring combustible and explosive gases.

[0165] The enhanced training sample set is divided into training, validation, and test subsets according to a certain ratio. The training subset is used to train the self-learning measurement model, the validation subset is used for model tuning and performance evaluation, and the test subset is used to test the model's generalization ability. During the partitioning, the data distribution is ensured to be as uniform as possible among the three subsets to avoid data bias affecting model training performance. The training subset is then input into the base measurement model, initiating an iterative parameter training process. During this process, the model parameters are adjusted based on the data from the training subset. The model learning process includes multiple iterative steps, each adjusting the model parameters based on the feedback from the training subset, making the model's predictions as close as possible to the true labels corresponding to the training subset. The goal of this stage is to obtain the initial self-learning measurement model.

[0166] After the initial self-learning measurement model is completed, a validation subset is input into the model for prediction validation, yielding the model's prediction results. Then, the model's prediction results are compared with the true labels corresponding to the validation subset to calculate the model error data. This step is crucial for evaluating the accuracy of the model's predictions, as the model error data reflects the current learning performance of the initial self-learning measurement model. Based on the model error data, the model parameters are adaptively adjusted. Parameter adjustment can employ optimization algorithms such as gradient descent and Adam optimization to minimize model error and improve model performance. The adjusted model parameters are then retrained iteratively, continuously optimized, until the model error data meets a preset convergence condition. The preset convergence condition can be that the error decreases to a specific threshold, or that the error change is very small after multiple consecutive iterations.

[0167] The optimized self-learning measurement model will accept input from a test subset to perform a generalization ability test, yielding model test results. By comparing and analyzing the model test results with the true labels in the test subset, the model's performance on unseen data can be evaluated. If the model test results meet the preset measurement accuracy requirements, the optimized self-learning measurement model is determined as the final self-learning measurement model.

[0168] In this approach, the model training is optimized by using an enhanced training sample set, and adaptive parameter adjustment is achieved by combining validation and testing feedback, thereby improving prediction accuracy and generalization ability.

[0169] S107: Acquire gas measurement response data and corresponding real-time environmental status data, input the gas measurement response data and real-time environmental status data into the self-learning measurement model for measurement identification, and obtain the measurement results of combustible and explosive gases.

[0170] Gas measurement response data is acquired through various gas detection devices such as infrared sensors, electrochemical sensors, and semiconductor sensors, with the core being the dynamic changes in gas concentration. The data comprehensively records key information such as the type of specific gas, concentration value, measurement time, monitoring location, and sensor operating status. This includes not only flammable and explosive gases but also toxic gases, oxygen, and other gases, covering their concentration changes. Real-time environmental status data is a collection of various environmental parameters closely related to gas diffusion, chemical reactions, and physical properties, including temperature, humidity, air pressure, wind speed, and light intensity.

[0171] The measurement results of flammable and explosive gases are the actual concentration information of flammable and explosive gases obtained through various sensors or complete monitoring systems, directly reflecting the current state of such gases in the area under test. The data usually includes core information such as the type of flammable gas, concentration value, measurement time and location, and is a quantitative representation of the gas's real-time state.

[0172] Real-time gas measurement response data acquisition relies on dedicated gas sensors to monitor gas concentration changes in real time. The response data output by these sensors includes electrical signals, frequency response, and rates of change corresponding to gas concentration changes, directly reflecting the state changes when the sensor surface interacts with gas molecules. The raw response data output by the sensors is typically converted from analog to digital, transforming it into a digital signal suitable for subsequent analysis. Simultaneously, environmental condition data must be collected concurrently. This data includes key parameters such as current temperature, humidity, air pressure, and wind speed, which significantly affect gas diffusion rates, concentration distribution, and the response characteristics of gas sensors. Therefore, environmental data must be acquired in real time using meteorological monitoring equipment or dedicated environmental sensors, with the acquisition frequency synchronized with the gas response data to ensure accurate correlation between the two types of data over time.

[0173] After acquiring gas measurement response data and environmental status data, both types of data are input into a self-learning measurement model for processing. This model, having undergone prior training, has mastered the ability to extract gas concentration patterns from gas sensor response signals and changes in environmental factors. Its core advantage lies in its ability to accurately identify gas concentration change characteristics by correlating historical analysis results with environmental data. Specifically, based on the parameters obtained from prior training, combined with the input real-time gas response data and environmental data, the model uses multi-level mathematical operations and pattern recognition to calculate the current gas concentration value.

[0174] The model's internal operating mechanism is based on multidimensional data analysis and machine learning algorithms. After gas response data and environmental state data are input, they undergo key steps such as preprocessing and feature extraction. The inherent correlation between the two types of data is used by the model as the core basis for predicting gas concentration. When the model performs measurement identification, it first uses previously learned knowledge to determine the similarity between the current gas response signal and historical response signals. Then, combining environmental factors and the inherent laws of gas behavior, it uses real-time environmental data to correct the gas response signal, ultimately providing an accurate estimate of the gas concentration. This process is not a simple data stream input and output, but a complete process including real-time calculation and dynamic adjustment. Whenever new gas measurement response data and environmental state data are input, the model performs a series of real-time inferences and dynamically adjusts the prediction results. After this series of coherent calculations, the model finally outputs the predicted concentration of flammable and explosive gases.

[0175] In this embodiment, the coverage of data samples can be effectively expanded, especially for extreme working conditions where direct data collection is difficult. By combining virtual and real samples, the diversity and representativeness of training data are enhanced, thereby improving the accuracy and robustness of the self-learning measurement model, ultimately achieving more accurate measurement and identification of flammable and explosive gases.

[0176] Figure 2 This is a schematic diagram of the second process of a self-learning flammable and explosive gas measurement method based on virtual mixed samples provided in an embodiment of this disclosure. The method may include the following steps:

[0177] S201, acquire the historical analysis results of combustible and explosive gases in the area to be tested, as well as the corresponding historical environmental state data and the corresponding real gas measurement results. Based on the historical analysis results, historical environmental state data and real gas measurement results, perform data processing and correlation analysis to obtain a real response sample set.

[0178] S202, based on the real response sample set, extract and classify the feature quantities of the gas response of flammable and explosive gases and environmental factors to obtain a feature parameter set.

[0179] S203, Based on the set of characteristic parameters, perform statistical correlation analysis between the gas components of the combustible and explosive gas and the response signal to obtain response relationship data.

[0180] S204. Based on the response relationship data, trend fitting and mathematical modeling are performed to obtain the basic measurement model.

[0181] S205, acquire low-coverage operating condition data from the basic measurement model, perform operating condition deduction based on the low-coverage operating condition data, and obtain a virtual operating condition set.

[0182] S206, Input the virtual operating condition set into the basic measurement model for response simulation to obtain a virtual mixed sample set.

[0183] S207. Based on the virtual mixed sample set and the real response sample set, sample consistency evaluation and credibility weighting are performed to obtain the enhanced training sample set.

[0184] S208, The basic measurement model is trained by self-learning based on the enhanced training sample set to obtain a self-learning measurement model.

[0185] S209, acquire gas measurement response data and corresponding real-time environmental status data, input the gas measurement response data and real-time environmental status data into the self-learning measurement model for measurement identification, and obtain the measurement results of combustible and explosive gases.

[0186] In this embodiment, the feature parameter set is a collection of key features extracted from historical gas measurement data and corresponding environmental state data. Its purpose is to describe the response behavior of flammable and explosive gases under different operating conditions, and the correlation between this behavior and environmental factors. This includes fluctuations in flammable gas concentration and gas response signals, as well as changes in various environmental conditions such as temperature, humidity, air pressure, and wind speed. These features can intuitively reflect the physical behavior of flammable and explosive gases under specific environments, providing quantitative input for subsequent modeling work, enabling the model to accurately capture the core laws of gas behavior.

[0187] Response relationship data is a dataset obtained after statistical correlation analysis of a set of characteristic parameters. It is used to reveal the intrinsic correlation between the components of flammable and explosive gases and the response signals detected by sensors, while also clarifying how these response signals are affected by environmental factors such as temperature and humidity. This data presents the reaction mechanisms of gases under different environmental conditions, providing crucial mathematical support for the construction of basic measurement models. It helps the models deeply learn and grasp the changing patterns of gas behavior, ensuring that the models can accurately capture the correlation between gas responses and environmental factors.

[0188] From the real response sample set, the first step is to extract gas response data and environmental factor characteristics of flammable and explosive gases. Flammable and explosive gases are typically composed of a mixture of multiple gaseous components, and their concentrations are strongly influenced by environmental factors. The primary step in data extraction is to collect data on the concentration changes of flammable gases monitored by gas sensors within the target area. This data reflects the concentration response of the gas at different time points and under different environmental conditions. Simultaneously, environmental parameters such as temperature, humidity, wind speed, and air pressure need to be acquired, as these factors directly affect the diffusion behavior of flammable and explosive gases. After cleaning and denoising these raw data to ensure accuracy and completeness, key features such as concentration changes and response amplitudes for each gas component can be extracted. Environmental factor characteristics include the degree of influence of various environmental parameters on gas behavior, such as the effect of temperature increases on gas evaporation rates. Finally, these gas concentration and environmental factor characteristics are categorized to obtain a comprehensive and representative set of characteristic parameters, providing fundamental data for subsequent analysis.

[0189] Next, based on the extracted feature parameter set, a statistical correlation analysis is performed between the gas components and the response signal. The core purpose of this step is to deeply explore the intrinsic relationship between the changing patterns of flammable and explosive gas concentrations and environmental factors. For example, statistical regression analysis can determine how changes in gas concentration are affected by environmental factors such as temperature, humidity, and air pressure, and quantify this influence. Especially in the monitoring of flammable and explosive gases, changes in gas concentration depend not only on the properties of the gas itself but are also greatly affected by changes in the external environment. Therefore, using methods such as multiple regression analysis and chi-square tests to analyze the correlation between gas concentration and environmental factors can reveal the gas's response patterns under different environmental conditions. This analysis will generate response relationship data, reflecting the statistical relationship between gas concentration changes and environmental factors, laying a data foundation for subsequent modeling work.

[0190] Finally, trend fitting and mathematical modeling are performed based on the response relationship data. In this step, the relationship between changes in flammable and explosive gas concentrations and environmental parameters is modeled using trend fitting methods to ensure that the model accurately reflects the dynamic behavior of the gas under different environmental conditions. For example, regression models, support vector machines, or neural networks are used to fit the nonlinear relationship between gas concentration and environmental changes, constructing a basic measurement model capable of predicting gas responses. The purpose of mathematical modeling is to ensure that the model can accurately predict the behavior of flammable and explosive gases under various environmental conditions, thereby providing reliable guidance for practical applications. Through repeated training and optimization of the model, the final basic measurement model will be able to accurately predict gas concentration changes based on environmental parameters such as temperature and humidity, providing strong support for flammable and explosive gas monitoring and early warning systems.

[0191] In this embodiment, the quantitative relationship between flammable and explosive gases and sensor responses can be revealed, thereby improving the accuracy of gas identification, concentration measurement, and cross-interference suppression in complex environments.

[0192] Figure 3 This is a schematic block diagram of a self-learning combustible and explosive gas measurement system based on virtual mixed samples, provided as an embodiment of the present disclosure. The system is characterized by comprising:

[0193] The historical data acquisition module 301 is used to acquire historical analysis results of combustible and explosive gases in the area to be tested, as well as corresponding historical environmental state data and corresponding real gas measurement results. Based on the historical analysis results, historical environmental state data and real gas measurement results, data is organized and correlated to obtain a real response sample set.

[0194] The basic model building module 302 is used to model the response pattern based on the real response sample set to obtain a basic measurement model.

[0195] The working condition simulation module 303 is used to acquire low-coverage working condition data in the basic measurement model, perform working condition simulation based on the low-coverage working condition data, and obtain a virtual working condition set.

[0196] The response simulation module 304 is used to input the virtual operating condition set into the basic measurement model to perform response simulation and obtain a virtual mixed sample set.

[0197] The enhanced sample construction module 305 is used to perform sample consistency evaluation and credibility weighting based on the virtual mixed sample set and the real response sample set to obtain an enhanced training sample set.

[0198] Self-learning training module 306 is used to perform self-learning training on the basic measurement model based on the enhanced training sample set to obtain a self-learning measurement model.

[0199] The measurement module 307 is used to acquire gas measurement response data and corresponding real-time environmental status data, input the gas measurement response data and real-time environmental status data into the self-learning measurement model for measurement identification, and obtain the measurement results of combustible and explosive gases.

[0200] Figure 4 An electronic device 400 provided in an embodiment of this application is shown, including a processor 401, a memory 402, and a program or instructions stored in the memory 402 and executable on the processor 401. When the program or instructions are executed by the processor 401, they implement the various processes of the above-described embodiment of the self-learning combustible and explosive gas measurement method based on virtual mixed samples and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0201] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0202] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described embodiments of the self-learning combustible and explosive gas measurement method based on virtual mixed samples, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0203] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0204] Furthermore, it should be understood that since the various modules are only provided to illustrate the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the processor's software, hardware, or a combination of software and hardware. Therefore, the number of modules shown in the figures is merely illustrative.

[0205] Those skilled in the art will understand that the various modules in the device can be adaptively split or combined. Such splitting or combining of specific modules will not cause the technical solution to deviate from the principles of the present invention; therefore, the technical solutions after splitting or combining will fall within the protection scope of the present invention.

[0206] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A self-learning method for measuring combustible and explosive gases based on virtual mixed samples, characterized in that, The method includes: Historical analysis results of combustible and explosive gases in the area to be tested, as well as corresponding historical environmental state data and corresponding real gas measurement results, are obtained. Based on the historical analysis results, historical environmental state data and real gas measurement results, data are organized and correlation analysis is performed to obtain a real response sample set. Based on the real response sample set, response patterns are modeled to obtain a basic measurement model; Obtain low-coverage operating condition data from the basic measurement model, perform operating condition simulation based on the low-coverage operating condition data, and obtain a virtual operating condition set. The virtual operating condition set is input into the basic measurement model for response simulation to obtain a virtual hybrid sample set; Based on the virtual mixed sample set and the real response sample set, sample consistency evaluation and credibility weighting are performed to obtain the enhanced training sample set; The basic measurement model is trained by self-learning based on the enhanced training sample set to obtain a self-learning measurement model. The gas measurement response data and the corresponding real-time environmental status data are acquired. The gas measurement response data and the real-time environmental status data are then input into the self-learning measurement model for measurement identification to obtain the measurement results of flammable and explosive gases.

2. The method according to claim 1, characterized in that, in, Based on the historical analysis results, historical environmental state data, and real gas measurement results, data processing and correlation analysis are performed to obtain a real response sample set, including: Based on the historical analysis results, historical environmental state data, and real gas measurement results, anomalies are removed and missing data are filled to obtain cleaning measurement data. Based on the cleaning measurement data, dimensional unification and normalization are performed to obtain standard measurement data; Response feature extraction is performed based on the standard measurement data to obtain response feature data including environmental feature data; Environmental correlation analysis is performed based on the response feature data to obtain environmental correlation data; Based on the response feature data and environmental correlation data, sample mapping and classification are performed to obtain a real response sample set.

3. The method according to claim 1, characterized in that, in, Based on the aforementioned real response sample set, response patterns are modeled to obtain a basic measurement model, including: Based on the real response sample set, feature quantities of gas response and environmental factors of combustible and explosive gases are extracted and classified to obtain a feature parameter set; Based on the set of characteristic parameters, a statistical correlation analysis is performed between the gas components of the combustible explosive gas and the response signal to obtain response relationship data; Based on the response relationship data, trend fitting and mathematical modeling are performed to obtain the basic measurement model.

4. The method according to claim 1, characterized in that, in, Based on the low-coverage operating condition data, operating condition simulation is performed to obtain a virtual operating condition set, including: Based on the low-coverage operating condition data, the change trend is extracted to obtain the operating condition change trend data; Based on the operating condition change trend data, the rate of change is calculated and the direction of change is determined to obtain the operating condition evolution characteristics. Based on the aforementioned operating condition evolution characteristics, parameter value boundaries are determined to obtain operating condition boundary data. Based on the low-coverage operating condition data and the operating condition boundary data, parameter linkage relationship analysis is performed to obtain operating condition correlation data. Based on the aforementioned working condition correlation data, parameter linkage constraints and evolution rules are established, and working condition deduction rules are generated based on the parameter linkage constraints and evolution rules. Based on the aforementioned operating condition extrapolation rules, the low-coverage operating condition data is continuously expanded to generate a candidate operating condition set with coverage operating condition boundaries and extrapolation rule constraints. The candidate working condition set is compared with the working condition boundary data, and working conditions that exceed the boundary are eliminated to obtain the boundary compliant working condition set. Based on the verification and linkage relationship between the boundary compliance working condition set and the working condition associated data, inconsistent working conditions are eliminated to obtain the associated consistent working condition set; Based on the aforementioned consistent operating condition set, abnormal operating conditions are eliminated and duplicate operating conditions are merged to obtain a virtual operating condition set.

5. The method according to claim 1, characterized in that, in, The virtual operating condition set is input into the basic measurement model for response simulation to obtain a virtual mixed sample set, including: Each virtual operating condition in the virtual operating condition set is input into the basic measurement model to obtain the corresponding initial response result; Based on the initial response results, gas concentration response data, sensor output response data, and environmental disturbance response data are extracted. Based on the gas concentration response data, sensor output response data, and environmental disturbance response data, operating condition response characteristic data are generated. Based on the operating condition response feature data, time series combination and multi-parameter mapping are performed to obtain the simulated response samples corresponding to each virtual operating condition; Based on the association and annotation of the simulated response samples and the corresponding operating parameters of the virtual operating conditions, virtual annotation samples are obtained; Based on the virtual labeled samples, a sample aggregation is performed to obtain a virtual hybrid sample set.

6. The method according to claim 1, characterized in that, in, Based on the virtual mixed sample set and the real response sample set, sample consistency evaluation and credibility weighting are performed to obtain the enhanced training sample set, including: Based on the comparison of feature distributions between the virtual mixed sample set and the real response sample set, sample difference data corresponding to each virtual sample is obtained. Based on the sample difference data, a sample consistency assessment is performed to obtain a consistency score for each virtual sample. Based on the consistency score and the preset confidence threshold, virtual samples with consistency scores lower than the preset confidence threshold are filtered out to obtain an effective set of virtual samples. Based on the consistency score, weights are assigned to each virtual sample in the effective virtual sample set to obtain the credibility weight corresponding to each virtual sample. The effective virtual sample set and the real response sample set are weighted and fused based on the credibility weight to obtain the enhanced training sample set.

7. The method according to claim 1, characterized in that, in, The basic measurement model is trained using an enhanced training sample set to obtain a self-learning measurement model, which includes: Based on the enhanced training sample set, the training samples are divided to obtain a training subset, a validation subset, and a test subset; The training subset is input into the basic measurement model for iterative parameter training to obtain an initial self-learning measurement model. The validation subset is input into the initial self-learning measurement model for prediction validation to obtain the model prediction result; Error calculation is performed based on the model prediction results and the real labels corresponding to the validation subset to obtain model error data; Based on the model error data, the model parameters of the initial self-learning measurement model are adaptively adjusted, and parameter iterative training is re-executed until the model error data meets the preset convergence condition, thereby obtaining the optimized self-learning measurement model. The test subset is input into the optimized self-learning measurement model to test its generalization ability, and the model test results are obtained. If the model test results meet the preset measurement accuracy requirements, the optimized self-learning measurement model is determined to be a self-learning measurement model.

8. A self-learning combustible and explosive gas measurement system based on virtual mixed samples, characterized in that, The system includes: The historical data acquisition module is used to acquire historical analysis results of combustible and explosive gases in the area to be tested, as well as corresponding historical environmental state data and corresponding real gas measurement results. Based on the historical analysis results, historical environmental state data and real gas measurement results, data is organized and correlation analysis is performed to obtain a real response sample set. The basic model building module is used to model the response patterns based on the real response sample set to obtain a basic measurement model. The working condition simulation module is used to acquire low-coverage working condition data in the basic measurement model, perform working condition simulation based on the low-coverage working condition data, and obtain a virtual working condition set. The response simulation module is used to input the virtual operating condition set into the basic measurement model for response simulation to obtain a virtual mixed sample set; An enhanced sample construction module is used to perform sample consistency evaluation and credibility weighting based on the virtual mixed sample set and the real response sample set to obtain an enhanced training sample set. The self-learning training module is used to perform self-learning training on the basic measurement model based on the enhanced training sample set to obtain a self-learning measurement model. The measurement module is used to acquire gas measurement response data and corresponding real-time environmental status data. The gas measurement response data and real-time environmental status data are input into the self-learning measurement model for measurement identification to obtain the measurement results of combustible and explosive gases.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the self-learning combustible and explosive gas measurement method based on any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the self-learning combustible and explosive gas measurement method based on virtual mixed samples as described in any one of claims 1-7.

Images