Methods, devices, and systems for handling interference in smoke and fire alarm systems

Abstract

This application provides an anti-interference processing method for a smoke and fire alarm system, relating to the field of smoke and fire alarm technology. The method includes: real-time acquisition and processing of environmental data from multiple sensors to generate a fused feature vector; comparing the fused feature vector with interference mode feature vectors in a dynamic interference feature library to obtain a similarity result; adjusting the decision weights of feature vectors related to the interference mode feature vectors based on the result; inputting the fused feature vector into a decision model, and using the adjusted decision weights to analyze the evolution of the fused feature vector over time to determine whether the fused feature vector is in a continuous or instantaneous state. The anti-interference processing method provided in this application, by employing multi-sensor data fusion, comparison with a dynamic interference feature library, and real-time dynamic weight adjustment, obtains a more comprehensive and accurate environmental state, providing a data foundation for accurate judgment from the source, and greatly improving the anti-interference processing capability and accuracy of the smoke and fire alarm system.

Description

Technical Field

[0001] This invention relates to the field of smoke and fire alarm technology, and in particular to an anti-interference processing method, device, and system for a smoke and fire alarm system. Background Technology

[0002] In most scenarios, such as train transportation, smoke and fire alarm systems are core technological facilities for ensuring transportation safety. These systems deploy sensors at critical locations on the train to monitor environmental parameters in real time, aiming to issue timely warnings in the early stages of a fire.

[0003] However, existing train smoke and fire alarm systems generally suffer from high false alarm rates due to insufficient anti-interference capabilities. The train operating environment is extremely complex and variable, with numerous non-fire-related interference sources, such as smoke from passengers smoking or using e-cigarettes, water vapor from heating food, aerosols generated when the train passes through high-dust tunnels, localized high temperatures caused by direct sunlight through train windows in summer, heat radiation from interior lighting fixtures, and localized overheating caused by friction between electrical equipment or mechanical components. The physical or chemical signals generated by these interference sources (such as particulate matter concentration, temperature changes, and specific spectra) are very similar to the early signs of a real fire.

[0004] Most existing smoke and fire alarm systems rely on one or a few sensors and use simple fixed threshold comparison algorithms to determine the presence of interference sources. This method is difficult to fundamentally distinguish between real fire signals and interference signals, resulting in a high false alarm rate. This leads to unnecessary emergency braking of trains, operational disruptions, and passenger panic, which not only seriously affects transportation efficiency but also greatly weakens the reliability of the alarm system.

[0005] Therefore, how to effectively distinguish between real fires and various interference sources in complex environments, significantly reduce false alarm rates, and at the same time ensure high detection sensitivity for real fires is a technical problem that urgently needs to be solved in the field of train smoke and fire alarm technology. Summary of the Invention

[0006] To address the problems existing in the prior art, the first aspect of this application provides an anti-interference processing method for a smoke alarm system, comprising:

[0007] Environmental data is collected in real time by multiple sensors and processed to generate a fused feature vector.

[0008] The similarity results are obtained by comparing the fused feature vector with the interference pattern feature vector in the dynamic interference feature library.

[0009] Based on the similarity results, adjust the decision weights of feature vectors related to the feature vectors of the interference patterns;

[0010] The fused feature vector is input into the decision model, which uses the adjusted decision weights to analyze the evolution of the fused feature vector over time and determine whether the fused feature vector is in a continuous or instantaneous state.

[0011] Optionally, the step of processing environmental data to generate a fused feature vector specifically includes:

[0012] Environmental data is standardized and time-aligned at edge computing nodes of multiple sensors;

[0013] On edge computing nodes of multiple sensors, feature vectors of processed environmental data are extracted in real time.

[0014] At edge computing nodes of multiple sensors, feature vectors are input into a feature fusion model to generate fused feature vectors.

[0015] Optionally, environmental data may include one or more of image data, gas concentration data, and temperature and humidity data.

[0016] Optionally, the step of extracting features from the processed environmental data in real time specifically includes:

[0017] Real-time extraction of visual features and dynamic texture change features of smoke and flame based on deep learning from image data;

[0018] Real-time extraction of gas concentration change gradients and short-term integral features from gas concentration data;

[0019] Real-time extraction of abnormal temperature rise rates and humidity drop characteristics from temperature and humidity data.

[0020] Optionally, the following steps may be included before performing the similarity comparison:

[0021] Establish a dynamic interference feature library;

[0022] The dynamic interference feature library includes preset typical interference features and historical interference features. Historical interference features include continuously identified interference modes and their corresponding fused feature vectors.

[0023] Optionally, the similarity results satisfy the following formula:

[0024] ;

[0025] Where V is the fused feature vector. Let i be the feature vector of the i-th interference mode in the dynamic interference feature library. The similarity result is obtained by comparing the fused feature vector with the feature vector of the i-th interference pattern, where i is a natural number.

[0026] Optionally, the step of adjusting the decision weights of feature vectors related to the interference pattern feature vectors based on the similarity results specifically includes:

[0027] When the similarity result is greater than the similarity threshold, the decision weight of the feature vector that is strongly correlated with the feature vector of the interference pattern is reduced.

[0028] Optionally, the decision weights satisfy the following adjustment formula:

[0029] ;

[0030] in, For similarity results, The decay coefficient of the decision weight is... The decision weights before adjustment The adjusted decision weights are represented by i, where i is a natural number.

[0031] Optionally, the magnitude of the reduction in decision weights is positively correlated with the similarity results;

[0032] The attenuation coefficient satisfies:

[0033] ;

[0034] in, Here, k is the decay coefficient for the decision weights, and k is the proportionality constant. This is the similarity result.

[0035] Optionally, the step of analyzing the evolution of the fused feature vector over time using the adjusted decision weights in the decision model specifically includes:

[0036] No alarm is triggered when the fused feature vector is in a transient state;

[0037] A primary alert is triggered when the fused feature vector remains in a persistent state within the first time period.

[0038] When the fused feature vector remains in a state for more than a preset threshold in the second time period, and the criterion is met for a number of times greater than or equal to the preset number, a fire confirmation alarm is triggered.

[0039] Optional: the first time is less than 30 seconds; the second time is between 30 and 60 seconds; the preset number of times is 3.

[0040] Optionally, a primary alarm is triggered by displaying an alarm message;

[0041] The fire alarm is triggered by an audible and visual alarm and by sending an alarm message.

[0042] Optionally, the feature fusion model includes an attention-based neural network model, which learns and assigns weights to different feature vectors under different environmental data, and generates a fused feature vector based on the weighted features.

[0043] Optionally, the formula for calculating the fused feature vector satisfies:

[0044] ;

[0045] ;

[0046] in, Let i be the i-th eigenvector. The weight coefficients of the i-th feature vector are calculated using the attention mechanism. Let i be a natural number and represent the fused feature vector.

[0047] Optionally, the method also includes:

[0048] In the absence of an alarm, a preset concentration of test gas is injected into the test chamber of the sensor used to collect gas concentration data;

[0049] Compare the sensor's response value with the expected value;

[0050] Calculate the calibration coefficient of the sensor to compensate for sensor drift.

[0051] Optionally, the step of collecting environmental data in real time using multiple sensors specifically includes:

[0052] The gas concentration data, including smoke particle concentration data, is collected in real time using dual-wavelength differential detection technology.

[0053] By comparing the difference in scattered light intensity between smoke particles and interfering particles, smoke particles and interfering particles can be distinguished.

[0054] The intensity difference of scattered light from smoke particles or interfering particles at different wavelengths satisfies:

[0055] ;

[0056] in, and They are respectively wavelength and The intensity of scattered light at the specified wavelength, where k is a correction factor.

[0057] Optionally, the environmental data may include one or more of the following: smoke particle concentration, visible light image, infrared thermal image, carbon monoxide concentration, carbon dioxide concentration, and ambient temperature and humidity data.

[0058] A second aspect of this application provides an anti-interference processing device for a smoke alarm system, comprising:

[0059] The acquisition module is used to collect environmental data in real time through multiple sensors and process the environmental data to generate a fused feature vector.

[0060] The comparison module is used to compare the fused feature vector with the interference pattern feature vector in the dynamic interference feature library to obtain the similarity result;

[0061] The adjustment module is used to adjust the decision weights of feature vectors related to the feature vectors of the interference pattern based on the similarity results.

[0062] The decision module is used to input the fused feature vector into the decision model. The decision model uses the adjusted decision weights to analyze the evolution of the fused feature vector in the time dimension in order to determine whether the fused feature vector is in a continuous state or an instantaneous state.

[0063] Optionally, the acquisition module is specifically used for:

[0064] Environmental data is standardized and time-aligned at edge computing nodes of multiple sensors;

[0065] On edge computing nodes of multiple sensors, feature vectors of processed environmental data are extracted in real time.

[0066] At edge computing nodes of multiple sensors, feature vectors are input into a feature fusion model to generate fused feature vectors.

[0067] Optionally, environmental data may include one or more of image data, gas concentration data, and temperature and humidity data.

[0068] Optionally, the acquisition module is also specifically used for:

[0069] Real-time extraction of visual features and dynamic texture change features of smoke and flame based on deep learning from image data;

[0070] Real-time extraction of gas concentration change gradients and short-term integral features from gas concentration data;

[0071] Real-time extraction of abnormal temperature rise rates and humidity drop characteristics from temperature and humidity data.

[0072] Optionally, a modeling module may also be included to build a dynamic interference feature library;

[0073] The dynamic interference feature library includes preset typical interference features and historical interference features. Historical interference features include continuously identified interference modes and their corresponding fused feature vectors.

[0074] Optionally, the similarity results satisfy the following formula:

[0075] ;

[0076] Where V is the fused feature vector. Let i be the feature vector of the i-th interference mode in the dynamic interference feature library. The similarity result is obtained by comparing the fused feature vector with the feature vector of the i-th interference pattern, where i is a natural number.

[0077] Optionally, the adjustment module is specifically used for:

[0078] When the similarity result is greater than the similarity threshold, the decision weight of the feature vector that is strongly correlated with the feature vector of the interference pattern is reduced.

[0079] Optionally, the decision weights satisfy the following adjustment formula:

[0080] ;

[0081] in, For similarity results, The decay coefficient of the decision weight is... The decision weights before adjustment The adjusted decision weights are represented by i, where i is a natural number.

[0082] Optionally, the magnitude of the reduction in decision weights is positively correlated with the similarity results;

[0083] The attenuation coefficient satisfies:

[0084] ;

[0085] in, Here, k is the decay coefficient for the decision weights, and k is the proportionality constant. This is the similarity result.

[0086] Optionally, the decision module is specifically used for:

[0087] No alarm is triggered when the fused feature vector is in a transient state;

[0088] A primary alert is triggered when the fused feature vector remains in a persistent state within the first time period.

[0089] When the fused feature vector remains in a state for more than a preset threshold in the second time period, and the criterion is met for a number of times greater than or equal to the preset number, a fire confirmation alarm is triggered.

[0090] Optional: the first time is less than 30 seconds; the second time is between 30 and 60 seconds; the preset number of times is 3.

[0091] Optionally, a primary alarm is triggered by displaying an alarm message;

[0092] The fire alarm is triggered by an audible and visual alarm and by sending an alarm message.

[0093] Optionally, the feature fusion model includes an attention-based neural network model, which learns and assigns weights to different feature vectors under different environmental data, and generates a fused feature vector based on the weighted features.

[0094] Optionally, the formula for calculating the fused feature vector satisfies:

[0095] ;

[0096] ;

[0097] in, Let i be the i-th eigenvector. The weight coefficients of the i-th feature vector are calculated using the attention mechanism. Let i be a natural number and represent the fused feature vector.

[0098] Optionally, a self-test module is also included, for:

[0099] In the absence of an alarm, a preset concentration of test gas is injected into the test chamber of the sensor used to collect gas concentration data;

[0100] Compare the sensor's response value with the expected value;

[0101] Calculate the calibration coefficient of the sensor to compensate for sensor drift.

[0102] Optionally, the acquisition module may also include:

[0103] The gas concentration data, including smoke particle concentration data, is collected in real time using dual-wavelength differential detection technology.

[0104] By comparing the difference in scattered light intensity between smoke particles and interfering particles, smoke particles and interfering particles can be distinguished.

[0105] The intensity difference of scattered light from smoke particles or interfering particles at different wavelengths satisfies:

[0106] ;

[0107] in, and They are respectively wavelength and The intensity of scattered light at the specified wavelength, where k is a correction factor.

[0108] Optionally, the environmental data may include one or more of the following: smoke particle concentration, visible light image, infrared thermal image, carbon monoxide concentration, carbon dioxide concentration, and ambient temperature and humidity data.

[0109] A third aspect of this application provides a smoke alarm system, comprising:

[0110] Multiple sensors are used to collect environmental data in real time;

[0111] The anti-interference processing apparatus provided in the second aspect of this application is used to execute any of the anti-interference processing methods provided in the first aspect of this application.

[0112] The beneficial effects that this application can achieve are:

[0113] Compared with existing technologies, the anti-interference processing method, device, and system of the train smoke and fire alarm system provided by this invention constructs a dynamic interference feature library that can learn and update, enabling the system to actively identify or record various known or unknown interference patterns. When real-time environmental features have a high degree of similarity to known interference patterns, the system can effectively suppress the influence of interference signals on fire situation judgment by dynamically adjusting the decision weights of the features, fundamentally solving the problem of high false alarm rate caused by the inability to distinguish the signal source in existing technologies.

[0114] Furthermore, by utilizing multi-sensor data fusion technology, this invention obtains a more comprehensive environmental status. By analyzing the evolution of the fused feature vector over time, it effectively filters out interference signals while improving the sensitivity and reliability of detecting real fires.

[0115] Meanwhile, by setting up a graded mechanism of primary alarm and final fire confirmation alarm, this invention avoids triggering a large-scale emergency response immediately due to minor anomalies, provides the system with fault tolerance time, further filters out transient false alarms, makes the emergency response more accurate and efficient, and avoids causing unnecessary panic. Attached Figure Description

[0116] Figure 1 A flowchart illustrating an optional anti-interference processing method for a smoke alarm system provided in this application embodiment;

[0117] Figure 2 A flowchart of an anti-interference processing device for an optional smoke alarm system provided in this application embodiment. Detailed Implementation

[0118] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present application.

[0119] The first aspect of this application provides an anti-interference processing method for a smoke and fire alarm system, mainly applicable to complex scenarios such as train transportation that are enclosed, dynamic, and have multiple known and unknown interference sources. It aims to solve the problem of high false alarm rates caused by insufficient anti-interference capabilities in existing technologies. This method is implemented through a train smoke and fire alarm system. The core hardware of the smoke and fire alarm system mainly includes a sensor network composed of multiple sensors distributed in key monitoring locations such as train carriages, equipment compartments, and driver's cab, edge computing nodes, a carriage processor, and a train central processing unit.

[0120] To clearly illustrate the technical solution of this application, the following detailed description will be based on a typical application scenario of a train. It should be noted that the application scope of this invention is not limited to this; any environment with similar characteristics of enclosure and diverse interference sources, such as ship cabins, underground utility tunnels, and large warehouses, can be applied to the anti-interference processing method provided in this application.

[0121] like Figure 1 As shown, the anti-interference processing method for a smoke and fire alarm system provided in this application includes:

[0122] S1. Collect environmental data in real time through multiple sensors and process the environmental data to generate a fused feature vector.

[0123] In this embodiment, a sensor network consisting of multiple sensors distributed across the monitoring area synchronously collects multi-dimensional environmental data in real time and processes the data. Subsequently, the processed environmental data are fused to generate a fused feature vector that comprehensively characterizes the current environmental state.

[0124] It should be noted that this application uses a sensor network composed of multiple sensors to synchronously collect different environmental data in real time. In one embodiment of this application, the environmental data includes one or more of image data, gas concentration data, and temperature and humidity data. These data reflect the environmental state from different dimensions and are highly correlated with fire signs, but are also susceptible to various non-fire interference sources. In a specific embodiment, the environmental data specifically includes one or more of smoke particle concentration, visible light images, infrared thermal imaging images, carbon monoxide concentration, carbon dioxide concentration, and environmental temperature and humidity data. Among them, the image data includes visible light images and infrared thermal imaging images, and the gas concentration data includes smoke particle concentration, carbon monoxide concentration, and carbon dioxide concentration.

[0125] It should be understood that the environmental data types listed above are merely illustrative examples and not a limitation on the scope of protection of this application. Any environmental parameter that can be included in the "fusion feature vector" and participate in the subsequent dynamic interference comparison and weight decision process is within the concept and protection scope of the technical solution of this application.

[0126] S2. Compare the similarity between the fused feature vector and the interference pattern feature vector in the dynamic interference feature library to obtain the similarity result.

[0127] It should be noted that the dynamic interference feature library stores feature vectors for various typical interference patterns caused by non-fire factors, such as passenger smoking, food vapor, e-cigarette vapor, sunlight exposure, heat radiation from lighting fixtures, mechanical friction overheating, and one or more of these in high-dust environments. The similarity result is obtained by calculating the similarity between the fused feature vector and the feature vectors of each interference pattern in the dynamic interference feature library. This similarity result objectively reflects the degree of matching between the current environmental state and various known interference patterns.

[0128] S3. Based on the similarity results, adjust the decision weights of the feature vectors related to the feature vectors of the interference mode.

[0129] After obtaining the similarity results, the feature decision weights in the subsequent fire situation determination process need to be adaptively and dynamically adjusted based on the similarity results. In this embodiment, the decision weights of feature vectors related to the interference mode feature vectors can be increased or decreased based on the similarity results to influence the subsequent fire situation determination. In one specific embodiment, this step can be run on a carriage processor or a train central processing unit.

[0130] S4. Input the fused feature vector into the decision model. The decision model uses the adjusted decision weights to analyze the evolution of the fused feature vector in the time dimension and determine whether the fused feature vector is in a continuous state or an instantaneous state.

[0131] After adjusting the weights of the feature vectors related to the interference mode feature vector in step S3, it is further necessary to analyze the evolution of the fusion feature vector in the continuous time series to determine whether the scene corresponding to the fusion feature vector is a transient interference scene or a real fire.

[0132] It should be noted that the decision model runs on the carriage processor or the train's central processing unit, and is used to analyze the evolution of the fused feature vector over time based on the adjusted decision weights. In other words, the model uses the decision weights adjusted in step S3, and identifies whether the changes in the fused feature vector over time show a continuous, correlated, and deteriorating trend, or an instantaneous, isolated, and recoverable fluctuation, to distinguish between real fires and various transient disturbances, and makes a decision on whether to issue an alarm and the alarm level.

[0133] In summary, the anti-interference processing method for the smoke and fire alarm system provided in this application integrates various environmentally relevant data collected by a sensor network, providing a rich data foundation for accurate judgment. Simultaneously, the fused features are compared with a dynamic interference feature library to adjust the feature decision weights based on the comparison results, thereby determining the interference mode or the actual fire situation. This fundamentally reduces the system's false alarm rate and effectively solves the technical problem of insufficient anti-interference capability of existing smoke and fire alarm systems in complex and enclosed environments, providing an efficient and reliable solution.

[0134] In some embodiments, the step S1, which involves processing the environmental data to generate a fused feature vector, specifically includes:

[0135] S11. On the edge computing nodes of multiple sensors, environmental data is standardized and time-series aligned.

[0136] Specifically, standardization refers to converting raw data from sensors of different types and dimensions, such as image pixel values, gas concentration in ppm values, and temperature in degrees Celsius, into a unified, dimensionless numerical range through normalization or scaling methods, thereby eliminating data inconsistencies caused by differences in sensor characteristics.

[0137] Timing alignment refers to unifying the data collected by all sensors at the same time or within the same time period into a high-precision timestamp, and correcting any delays in the data stream based on this timestamp to ensure the accuracy and consistency of the description of the environmental state at the same moment.

[0138] Through the above standardization and time-series alignment operations, the original data is transformed into a data stream with a uniform format and a uniform timestamp, providing a computable input for subsequent real-time feature extraction and fusion steps.

[0139] S12. On the edge computing nodes of multiple sensors, extract the feature vectors of the processed environmental data in real time to generate multi-dimensional feature vectors.

[0140] Specifically, edge computing nodes can perform corresponding feature extraction algorithms for different types of environmental data, for example:

[0141] For image data, it is necessary to extract visual features of smoke and flames and dynamic texture change features based on deep learning in real time. In one specific embodiment, deep learning can use a convolutional neural network model to extract the texture, contour, and color features of the image, as well as the regions suspected of being smoke / flame and their dynamic change features; the visual features of smoke and flames mainly include visual semantic features related to smoke and flames in the image, such as specific color distribution and shape contours; dynamic texture features are mainly used to reflect smoke diffusion or changes in heat sources. These features can effectively characterize abnormal patterns in the visible light and thermal radiation dimensions.

[0142] For gas concentration data, it is necessary to extract the gas concentration change gradient and short-term integral features in real time. In one specific embodiment, real-time extraction of the gas concentration change gradient means calculating the gradient of the gas concentration value change over a unit of time. This is used to understand the instantaneous rate of change of gas concentration, specifically, when > Time markers are used as anomalous features, where, The preset concentration gradient threshold is used; real-time extraction of short-term integral features is used to calculate the integral value of the gas in a short period of time, which reflects the cumulative exposure of the gas in a short period of time.

[0143] For temperature and humidity data, it is necessary to extract abnormal temperature rise rates and humidity drop characteristics in real time. In one specific embodiment, real-time extraction of abnormal temperature rise rates involves calculating the abnormal rate of temperature increase near the sensor in real time to identify abnormal rapid temperature rise phenomena; real-time extraction of humidity drop characteristics is mainly used to monitor humidity status in certain types of fires that can cause a rapid drop in humidity.

[0144] S13. At the edge computing nodes of multiple sensors, the generated multi-dimensional feature vectors are input into the feature fusion model to generate fused feature vectors.

[0145] In some embodiments, the feature fusion model includes an attention-based neural network model for learning and assigning weights to different feature vectors under different environmental data, and generating a fused feature vector based on the weighted features. In other words, this model can adaptively learn and dynamically assign importance weights to each feature vector in the fusion process based on the different environmental contexts reflected by the current input multi-dimensional feature vectors, thereby highlighting key information and suppressing secondary or unreliable information, ultimately outputting a fused feature vector that represents the current environment. For example, near a train toilet in extremely high humidity, where water vapor interference is likely, the weight of white clump features in the image would be appropriately reduced, while the weight of carbon monoxide gas would be increased.

[0146] Furthermore, after feeding these multi-dimensional feature vectors into a feature fusion model based on an attention mechanism, the model first maps the input multi-dimensional feature vectors into a query vector representing the overall state of the current environment, while treating each feature itself as a key-value pair. The model then generates an attention distribution by calculating the similarity between the query vector and each key-value pair. This distribution is essentially a set of normalized weight coefficients, whose values ​​directly reflect the importance and reliability of the corresponding feature in the current specific environmental context. For example, when the system detects through location sensors or humidity data that the current monitoring point is near a restroom and the ambient humidity is significantly higher than the baseline value, the model infers that the probability of water vapor interference is extremely high. Therefore, it automatically assigns a lower weight to the white wispy smoke feature in the visible light image in the attention distribution, while assigning a higher weight to the gradient of carbon monoxide gas concentration changes. This is because carbon monoxide is a hallmark product of incomplete combustion and is less affected by water vapor, making it a more reliable fire indicator in this environmental context.

[0147] According to the above embodiments, the output fused feature vector The calculation formula satisfies:

[0148] ;

[0149] ;

[0150] in, This is the i-th feature vector input into the feature fusion model; The weight coefficients of the i-th feature vector, calculated through the attention mechanism, reflect the importance or credibility of the i-th feature vector in the final environmental state judgment under a specific environmental context. Let i be the calculated fused feature vector, where i is a natural number. The above formula can intelligently weight and fuse information features from various dimensions.

[0151] It should be understood that the embodiments of this application preprocess the data, extract features, and fuse them through edge computing nodes of multiple sensors. This architecture can not only significantly reduce the data transmission load and alleviate the pressure and transmission latency of the vehicle communication network, but also share the computing pressure of the central processing unit, distributing the computationally intensive feature extraction and fusion operations to each edge node for simultaneous processing, further improving the overall processing efficiency and real-time response capability of the system.

[0152] It should be noted that this application does not impose a unique limitation on the sensor distribution for data processing. The above description of using edge computing nodes as the main computing carrier is a preferred embodiment of this application. Those skilled in the art should understand that as long as the functions of data preprocessing, feature extraction, and fusion can be achieved, using other architectures with corresponding computing capabilities to perform the corresponding steps are all optional implementations of the technical solution of this application and should fall within the protection scope of this application.

[0153] In some embodiments, the step S2, prior to the similarity comparison step, includes the following step:

[0154] Establish a dynamic interference feature library.

[0155] The dynamic interference feature library is established on the carriage processor based on real train operation scenarios and mainly includes preset typical interference features and historical interference features. Specifically, the preset typical interference features include the feature vector of the baseline interference mode simulated in the laboratory during the initial deployment of the system. The historical interference features include historical data recorded during the initial deployment of the system, as well as alarm events that were continuously identified as interference during the normal operation of the system and their corresponding fused feature vectors. In this application, historical interference features need to be confirmed by the system administrator before being added to the dynamic interference feature library as new sample increments.

[0156] According to the above embodiment, during the execution of step S2, the fused feature vector is obtained. The similarity result S between the feature vector D of the interference mode in the dynamic interference feature library and the feature vector D of the interference mode needs to be calculated in real time according to the following cosine similarity formula:

[0157] ;

[0158] Where V is the fused feature vector; This represents the feature vector of the i-th interference mode in the dynamic interference feature library; The similarity result obtained by comparing the fused feature vector with the feature vector of the i-th interference mode is given in the application scenario of this application. The closer the similarity result is to 1, the higher the consistency between the fused feature vector and the feature vector of the interference mode, that is, the higher the feature matching degree between the current environment state and the interference mode. Conversely, the closer the similarity result is to 0, the lower the matching degree between the fused feature vector and the feature vector of the interference mode. i is a natural number.

[0159] In some embodiments, step S3, which involves adjusting the decision weights of feature vectors related to the interference pattern feature vectors based on the similarity results, specifically includes:

[0160] Based on the calculated similarity results, the feature weights in subsequent decision algorithms are activated or adjusted.

[0161] Specifically, when the similarity results When the similarity value exceeds the similarity threshold, it indicates that the current environment is highly similar to the i-th interference pattern, and the system determines that there is a high probability of interference. In this case, it is necessary to reduce the decision weights of feature vectors strongly correlated with the interference pattern's feature vectors, thereby suppressing the influence of the interference pattern in the decision logic. For example, if the current environmental state is highly similar to the "food steam" pattern, the weights of white clump features in the visible light image and humidity sensor-related features may be reduced.

[0162] In some embodiments, the system may also implement a weight enhancement strategy as an alternative. That is, when the similarity result is greater than a similarity threshold, the decision weight of feature vectors that are strongly correlated with the actual fire situation and are not easily affected by interference pattern feature vectors may be increased accordingly. For example, the weight of carbon monoxide concentration change features may be increased. This strategy aims to further target more reliable fire indication information and provide a data foundation for subsequent accurate judgment.

[0163] When the similarity result is less than the similarity threshold, it indicates that there is no interference pattern feature vector in the dynamic interference feature library that is highly similar to the current environment. At this point, it is necessary to further determine whether the current environmental state belongs to an unknown interference pattern or a scenario of a real fire.

[0164] In summary, the decision weight adjustment mechanism provided in this application is not a one-way adjustment, but a dynamic and directional redistribution process based on comparison results. This process is a dynamic feedback, which ensures that the system will not make a judgment directly due to interference from a certain sensor, especially a certain feature vector. At the same time, it ensures that the system can flexibly cope with different scenarios such as known interference modes, unknown interference modes and potential fires.

[0165] According to the above embodiments, when the similarity result is within the similarity threshold, the system adjusts the decision weights using the following formula:

[0166] ;

[0167] in, Similarity results; is the decay coefficient of the decision weight, used to control the rate at which the decision weight decays as the similarity result increases. The larger this value is, the higher the system's ability to suppress the feature vector of the i-th type of interference mode, and the faster the weight of the feature vector of the i-th type of interference mode decays. The decision weights before adjustment; represents the adjusted decision weights; i is a natural number.

[0168] Furthermore, in order to more accurately correlate the attenuation level with the interference matching degree, the attenuation coefficient... It can be designed to be similar to the results Related dynamic variables. In one specific embodiment, the magnitude of the reduction in decision weights. (attenuation coefficient) Similarity results They are positively correlated. The attenuation coefficient satisfies:

[0169] ;

[0170] in, Here, k is the decay coefficient for the decision weights, and k is the proportionality constant. This is the similarity result.

[0171] In some embodiments, the step S4, in which the decision model analyzes the evolution of the fused feature vector over time using the adjusted decision weights, specifically includes:

[0172] S41. When the fused feature vector is in a transient state, no alarm is triggered.

[0173] Specifically, when the judgment model analyzes and determines that the fused feature vector indicates that the current abnormal mode of the environment is a transient state, the system will not trigger any level of alarm.

[0174] It should be noted that the instantaneous state in this application refers to a specific pattern identified by analyzing the evolution of fused feature vectors over a continuous time series. This pattern is typically characterized by one or more feature vectors exhibiting brief, isolated peaks or abrupt changes. However, within one or more preset time periods, the abnormal indicator does not show a sustained and worsening trend, but rather rapidly decays and recovers to normal levels. For example, when a passenger smokes, a large amount of smoke is produced, causing a significant increase in the concentration of smoke particles in a localized area of ​​the carriage within a short period. The smoke particle concentration data collected by the sensor reflects this instantaneous abnormal increase in concentration. However, smoke from a real fire is usually persistent and diffuse. When the passenger finishes smoking, the smoke particle concentration typically does not remain high but shows a rapid decreasing trend, recovering to the pre-smoking concentration level within a short time. This matches the scenario of a sudden change in smoke particle concentration within a short period. In this case, the model determines that the concentration data fails to meet the criterion of sustained deterioration in the subsequent time period, but rather conforms to the interference feature pattern of rapid recovery after an instantaneous change. Therefore, the system will not trigger a fire alarm according to the aforementioned processing strategy, thus effectively avoiding a false alarm caused by a passenger smoking.

[0175] S42. When the fused feature vector is in a continuous state in the first time, a primary alarm is triggered.

[0176] Specifically, when the judgment model analyzes and determines that the abnormal environmental pattern of the fused feature vector is in a state of continuous and worsening in the first time, the system triggers a primary alarm.

[0177] It should be noted that, in this embodiment, the "persistent state" refers to the situation where, through analysis of the evolution of the fused feature vectors over time, at least one feature vector anomaly is identified that fails to recover quickly, instead remaining at an abnormal level for the first time period and showing a gradually worsening trend. Simultaneously, these abnormal indicators may begin to show some correlation. For example, when a train is in the early stages of a fire, it gradually releases carbon monoxide, and the temperature and humidity inside the carriage show a gradual changing trend. At this time, the carbon monoxide concentration data and temperature and humidity data collected by the sensors reflect the persistent state in the first time period. The model then determines that these data have met the criteria for continuous worsening in subsequent time periods. Therefore, the system will trigger a primary alarm according to the aforementioned processing strategy to attract the attention of train operation and maintenance personnel, thereby achieving early warning and risk management of potential real fires.

[0178] In one specific embodiment, the primary alarm is triggered by displaying alarm information. Specifically, the primary alarm is only triggered by logging and displaying information on the maintenance interface of the train control system, and does not activate audible or visual alarms to avoid causing panic.

[0179] S43. When the fused feature vector is in a continuous state for the second time and exceeds the preset threshold, and the criterion is greater than or equal to the preset number of times, a fire confirmation alarm is triggered.

[0180] Specifically, when the judgment model analyzes and determines that the abnormal environmental pattern of the fused feature vector is in a state of continuous deterioration and crosses the preset threshold in the second time period, the system triggers a fire confirmation alarm.

[0181] It should be noted that, in this embodiment, "persistent state exceeding a preset threshold" refers to the trend where, through analysis of the evolution of the fused feature vectors over time, at least one feature vector's outlier fails to recover quickly and instead continues to worsen and exceeds the preset threshold in a second period. For example, when a train is in the middle or later stages of a fire, the abnormal states of carbon monoxide concentration data and temperature and humidity concentration data within the carriage will further deteriorate compared to the initial stage. When such abnormal states are detected more than a preset number of times, the system can immediately confirm the fire and activate a fire confirmation alarm, notifying all passengers and crew members and the operations control center in the most timely and clear manner. This provides maximum time for personnel evacuation and fire fighting, fundamentally ensuring operational safety.

[0182] In one specific embodiment, the fire confirmation alarm is triggered by an audible and visual alarm and the sending of an alarm message. Specifically, the final fire confirmation alarm will activate the audible and visual alarms in the carriage, send the highest priority alarm message to the driver's cab, and prepare the real-time braking system (e.g., to alert the driver or initiate automatic braking preparation).

[0183] Specifically, the first time is less than 30 seconds; the second time is between 30 and 60 seconds; and the preset number of times is 3. In other words, when the abnormal environmental mode persists for 30 seconds, the system triggers a primary alarm. When the abnormal environmental mode persists for 30 to 60 seconds and crosses the preset threshold, and this condition occurs more than 3 times, the system triggers a fire confirmation alarm.

[0184] It should be noted that those skilled in the art will understand that the core of this invention lies in constructing a dynamic, adaptive, and multi-dimensional collaborative anti-interference processing method for a smoke and fire alarm system. In the embodiments of this application, in order to clearly and completely describe the logical flow and implementation principle of the method, the specific values ​​of the preset thresholds mentioned in the specification are adjustable and optimizable exemplary parameters for implementing the method, rather than essential technical features or limiting conditions of this invention.

[0185] In summary, this invention analyzes the persistence and correlation of features over time, rather than relying on thresholds at a single moment. Therefore, it effectively distinguishes between transient behavioral disturbances and genuine fires, significantly reducing the system's false alarm rate. Furthermore, this invention, by activating a tiered alarm mechanism, enhances safety early warning capabilities while preventing operational interruptions caused by transient disturbances directly triggering the highest-level alarm.

[0186] In some embodiments, to ensure long-term reliability, the system performs a self-test periodically at night or during idle periods without alarms; therefore, the method provided in this application further includes the following steps:

[0187] In the absence of an alarm, a preset concentration of test gas is injected into the test chamber of the sensor used to collect gas concentration data.

[0188] Compare the sensor's response value with the expected value.

[0189] Calculate the calibration coefficient of the sensor to compensate for sensor drift.

[0190] The above steps are self-test procedures that the system performs periodically or as needed in the absence of alarms. Specifically, the system injects a preset concentration of standard test gas into the test chamber of the sensor used to collect gas concentration data to detect the sensor's response value. The system then calculates a calibration coefficient to compensate for sensor drift by comparing the response value with the expected value. The concentration of this standard test gas is pre-calibrated and stored as a calibration basis. The main purpose of these steps is to assess the sensor's current performance status by acquiring its response under known ideal input conditions, and to provide a basis for compensating for potential drift, thereby ensuring the accuracy and reliability of sensor data acquisition.

[0191] The self-test procedure described above is mainly for gas sensor testing. For image sensors, the system can control the camera to capture images of a built-in standard color swatch and white swatch, automatically correcting white balance and color parameters to ensure the accuracy of feature extraction.

[0192] In some embodiments, the acquisition of gas concentration data, especially smoke particle concentration data, can be achieved using a photoelectric smoke sensor. At the same time, dual-wavelength differential detection technology is used for real-time acquisition. By comparing the difference in the intensity of scattered light from smoke particles with the difference in the intensity of scattered light from interfering particles (such as water vapor, dust and other interfering aerosols), the differences in optical characteristics between smoke particles and interfering particles can be distinguished, thereby improving the anti-interference capability of the original data.

[0193] In one specific embodiment, dual-wavelength differential detection technology can use a dual-wavelength light source capable of emitting blue and infrared light to detect air, thereby physically distinguishing smoke from water vapor / dust.

[0194] The intensity difference of scattered light from smoke particles or interfering particles at different wavelengths satisfies:

[0195] ;

[0196] in, and They are respectively wavelength and The intensity of scattered light at the specified wavelength, where k is a correction factor.

[0197] In some embodiments, for temperature and humidity data, an infrared thermal imaging camera can be used to collect temperature distribution data, and a temperature and humidity sensor can be used to collect environmental parameters.

[0198] In some embodiments, gas concentration data, especially carbon monoxide and carbon dioxide concentration data, can be collected using appropriate dedicated gas sensors.

[0199] A second aspect of this application provides an anti-interference processing device for a smoke alarm system, comprising:

[0200] The acquisition module 100 is used to collect environmental data in real time through multiple sensors and process the environmental data to generate a fused feature vector.

[0201] The comparison module 200 is used to compare the fused feature vector with the interference mode feature vector in the dynamic interference feature library to obtain the similarity result;

[0202] The adjustment module 300 is used to adjust the decision weights of feature vectors related to the feature vectors of the interference pattern based on the similarity results.

[0203] The decision module 400 is used to input the fused feature vector into the decision model. The decision model uses the adjusted decision weights to analyze the evolution of the fused feature vector in the time dimension in order to determine whether the fused feature vector is in a continuous state or an instantaneous state.

[0204] Optionally, module 100 is specifically used for:

[0205] Environmental data is standardized and time-aligned at edge computing nodes of multiple sensors;

[0206] On edge computing nodes of multiple sensors, feature vectors of processed environmental data are extracted in real time.

[0207] At edge computing nodes of multiple sensors, feature vectors are input into a feature fusion model to generate fused feature vectors.

[0208] Optionally, environmental data may include one or more of image data, gas concentration data, and temperature and humidity data.

[0209] Optionally, the acquisition module 100 is also specifically used for:

[0210] Real-time extraction of visual features and dynamic texture change features of smoke and flame based on deep learning from image data;

[0211] Real-time extraction of gas concentration change gradients and short-term integral features from gas concentration data;

[0212] Real-time extraction of abnormal temperature rise rates and humidity drop characteristics from temperature and humidity data.

[0213] Optionally, a modeling module may also be included to build a dynamic interference feature library;

[0214] The dynamic interference feature library includes preset typical interference features and historical interference features. Historical interference features include continuously identified interference modes and their corresponding fused feature vectors.

[0215] Optionally, the similarity results satisfy the following formula:

[0216] ;

[0217] Where V is the fused feature vector. Let i be the feature vector of the i-th interference mode in the dynamic interference feature library. The similarity result is obtained by comparing the fused feature vector with the feature vector of the i-th interference pattern, where i is a natural number.

[0218] Optionally, the adjustment module 300 is specifically used for:

[0219] When the similarity result is greater than the similarity threshold, the decision weight of the feature vector that is strongly correlated with the feature vector of the interference pattern is reduced.

[0220] Optionally, the decision weights satisfy the following adjustment formula:

[0221] ;

[0222] in, For similarity results, This is the decay coefficient for the decision weights. The decision weights before adjustment The adjusted decision weights are represented by i, where i is a natural number.

[0223] Optionally, the magnitude of the reduction in decision weights is positively correlated with the similarity results;

[0224] The attenuation coefficient satisfies:

[0225] ;

[0226] in, Here, k is the decay coefficient for the decision weights, and k is the proportionality constant. This is the similarity result.

[0227] Optionally, the decision module 400 is specifically used for:

[0228] No alarm is triggered when the fused feature vector is in a transient state;

[0229] A primary alert is triggered when the fused feature vector remains in a persistent state within the first time period.

[0230] When the fused feature vector remains in a state for more than a preset threshold in the second time period, and the criterion is met for a number of times greater than or equal to the preset number, a fire confirmation alarm is triggered.

[0231] Optional: the first time is less than 30 seconds; the second time is between 30 and 60 seconds; the preset number of times is 3.

[0232] Optionally, a primary alarm is triggered by displaying an alarm message;

[0233] The fire alarm is triggered by an audible and visual alarm and by sending an alarm message.

[0234] Optionally, the feature fusion model includes an attention-based neural network model, which learns and assigns weights to different feature vectors under different environmental data, and generates a fused feature vector based on the weighted features.

[0235] Optionally, the formula for calculating the fused feature vector satisfies:

[0236] ;

[0237] ;

[0238] in, Let i be the i-th eigenvector. The weight coefficients of the i-th feature vector are calculated using the attention mechanism. Let i be a natural number and represent the fused feature vector.

[0239] Optionally, a self-test module is also included, for:

[0240] In the absence of an alarm, a preset concentration of test gas is injected into the test chamber of the sensor used to collect gas concentration data;

[0241] Compare the sensor's response value with the expected value;

[0242] Calculate the calibration coefficient of the sensor to compensate for sensor drift.

[0243] Optionally, the acquisition module 100 may also specifically include:

[0244] The gas concentration data, including smoke particle concentration data, is collected in real time using dual-wavelength differential detection technology.

[0245] By comparing the difference in scattered light intensity between smoke particles and interfering particles, smoke particles and interfering particles can be distinguished.

[0246] The intensity difference of scattered light from smoke particles or interfering particles at different wavelengths satisfies:

[0247] ;

[0248] in, and They are respectively wavelength and The intensity of scattered light at the specified wavelength, where k is a correction factor.

[0249] Optionally, the environmental data may include one or more of the following: smoke particle concentration, visible light image, infrared thermal image, carbon monoxide concentration, carbon dioxide concentration, and ambient temperature and humidity data.

[0250] A third aspect of this application provides a smoke alarm system, comprising:

[0251] Multiple sensors are used to collect environmental data in real time;

[0252] The anti-interference processing apparatus provided in the second aspect of this application is used to execute any of the anti-interference processing methods provided in the first aspect of this application.

[0253] To make the technical solutions, technical features and beneficial effects of the present invention clearer and to facilitate understanding and implementation by those skilled in the art, the following will further elaborate on the operating logic, processing procedures and technical effects of the aforementioned methods and systems in conjunction with multiple specific application scenarios and embodiments.

[0254] Example 1

[0255] When a passenger smokes in the carriage, the smoke sensor detects an increase in particulate matter concentration, and the visible light camera captures the characteristics of wisps of smoke. However, the carbon monoxide concentration change gradient detected by the gas sensor is much lower than that of actual open flame combustion. The system compares the current fused feature vector with the "smoking" pattern in the dynamic interference feature library in real time and calculates a highly similar score. Based on this result, the adaptive weight adjustment module automatically reduces the decision weight of the visible light smoke feature and particulate matter concentration data, while maintaining or slightly increasing the weight of the carbon monoxide gas concentration data. The feature vector after weight adjustment is sent to the decision model. The model analysis found that the peak of particulate matter concentration was short and then quickly dissipated, while the carbon monoxide concentration never showed a continuous upward trend. Therefore, the event was judged to be an interference event, and the event was only recorded in the system log without triggering any audible or visual alarms, thus effectively avoiding false alarms.

[0256] Example 2

[0257] In the food cart area, the large amount of water vapor generated by heating food filled the air. Traditional photoelectric smoke sensors, due to their sensitivity to water vapor particles, would trigger an alarm. The photoelectric smoke sensor in this invention first distinguishes the optical characteristics of water vapor particles from smoke particles at the physical level, outputting a preliminary judgment. Simultaneously, the visible light camera captures a large amount of white, rapidly diffusing clumps of mist, while the carbon monoxide sensor reading remains unchanged. Based on the context of "high humidity, no carbon monoxide," the feature fusion model automatically assigns higher reliability weights to the gas readings through an attention mechanism, while reducing the weight of white clump features in the image. Dynamic interference library comparison confirms that this pattern is highly similar to "food steam" interference. The decision model ultimately classifies it as an interference event due to the lack of feature persistence and the absence of deterioration in key indicators, successfully avoiding false alarms caused by steam.

[0258] Example 3

[0259] When smoldering occurs in a corner of the carriage, a small amount of smoke is produced but no obvious flame. The carbon monoxide gas concentration begins to show a slow but continuous upward trend. The infrared thermal imager detects a slight local temperature rise. The feature vector extracted by the system is not highly similar to the patterns in the interference library. Therefore, the weights of each sensor are kept relatively balanced. The feature vector after weight adjustment is input into the decision model. The model identifies that the carbon monoxide concentration and smoke particle concentration show a correlated and irreversible upward trend in multiple consecutive time windows. Although the absolute value is not high, its continuous deterioration pattern is consistent with the characteristics of a fire. The system first triggers the primary alarm log. After confirming the trend continues in subsequent monitoring, it decisively issues the final fire confirmation alarm and initiates the full set of emergency response procedures, thereby achieving early and accurate alarm.

[0260] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for anti-interference processing in a smoke and fire alarm system, characterized in that, include: Environmental data is collected in real time by multiple sensors and processed to generate a fused feature vector. The fused feature vector is compared with the interference pattern feature vector in the dynamic interference feature library to obtain the similarity result; Based on the similarity results, adjust the decision weights of the feature vectors related to the feature vectors of the interference pattern; The fused feature vector is input into the decision model, which uses the adjusted decision weights to analyze the evolution of the fused feature vector over time and determine whether the fused feature vector is in a continuous or instantaneous state.

2. The anti-interference processing method according to claim 1, characterized in that, The steps of processing the environmental data to generate a fused feature vector specifically include: On the edge computing nodes of the multiple sensors, the environmental data is standardized and time-aligned. On the edge computing nodes of the multiple sensors, feature vectors of the processed environmental data are extracted in real time. At the edge computing nodes of the multiple sensors, the feature vectors are input into the feature fusion model to generate the fused feature vectors.

3. The anti-interference processing method according to claim 2, characterized in that, The environmental data includes one or more of the following: image data, gas concentration data, and temperature and humidity data.

4. The anti-interference processing method according to claim 3, characterized in that, The step of extracting features from the processed environmental data in real time specifically includes: The image data is extracted in real time based on deep learning to extract visual features of smoke and flames and dynamic texture change features. The gas concentration change gradient and short-term integral features are extracted in real time from the gas concentration data; The abnormal temperature rise rate and humidity drop characteristics are extracted in real time from the temperature and humidity data.

5. The anti-interference processing method according to claim 1, characterized in that, The step of performing similarity comparison is preceded by: Establish the aforementioned dynamic interference feature library; The dynamic interference feature library includes preset typical interference features and historical interference features. The historical interference features include continuously identified interference modes and their corresponding fused feature vectors.

6. The anti-interference processing method according to claim 1, characterized in that, The similarity results satisfy the following formula: ； Where V is the fused feature vector. Let i be the feature vector of the i-th interference mode in the dynamic interference feature library. The similarity result is obtained by comparing the fused feature vector with the i-th interference mode feature vector, where i is a natural number.

7. The anti-interference processing method according to claim 1, characterized in that, The steps of adjusting the decision weights of feature vectors related to the interference pattern feature vectors based on the similarity results specifically include: When the similarity result is greater than the similarity threshold, the decision weight of the feature vector that is strongly correlated with the feature vector of the interference pattern is reduced.

8. The anti-interference processing method according to claim 7, characterized in that, The decision weights satisfy the following adjustment formula: ； in, For the aforementioned similarity results, The decay coefficient of the decision weight is... The decision weights before adjustment The adjusted decision weights are represented by i, where i is a natural number.

9. The anti-interference processing method according to claim 8, characterized in that, The magnitude of the reduction in the decision weight is positively correlated with the similarity result; Wherein, the attenuation coefficient satisfies: ； in, Here, k is the attenuation coefficient of the decision weight, and k is a proportionality constant. This is the similarity result.

10. The anti-interference processing method according to claim 1, characterized in that, The decision model analyzes the evolution of the fused feature vector over time using the adjusted decision weights, specifically including: When the fused feature vector is in a transient state, no alarm is triggered; When the fused feature vector is in a persistent state within the first time period, a primary alarm is triggered; When the fused feature vector remains in a state for more than a preset threshold within a second time period, and the criterion is greater than or equal to a preset number of times, a fire confirmation alarm is triggered.

11. The anti-interference processing method according to claim 10, characterized in that, The first time is less than 30 seconds; the second time is between 30 seconds and 60 seconds; the preset number of times is 3.

12. The anti-interference processing method according to claim 10, characterized in that, The primary alarm is triggered by displaying alarm information; The fire confirmation alarm is triggered by an audible and visual alarm and by sending an alarm message.

13. The anti-interference processing method according to claim 2, characterized in that, The feature fusion model includes an attention-based neural network model, which learns and assigns weights to different feature vectors under different environmental data, and generates a fused feature vector based on the weighted features.

14. The anti-interference processing method according to claim 13, characterized in that, The formula for calculating the fused feature vector satisfies: in, For the i-th feature vector, The weight coefficients of the i-th feature vector obtained through the attention mechanism are... Let i be the fused feature vector, where i is a natural number.

15. The anti-interference processing method according to claim 10, characterized in that, The method further includes: In the absence of an alarm, a preset concentration of test gas is injected into the test chamber of the sensor used to collect gas concentration data; Compare the sensor's response value with the expected value; Calculate the calibration coefficient of the sensor to compensate for sensor drift.

16. The anti-interference processing method according to claim 3, characterized in that, The step of collecting environmental data in real time using multiple sensors specifically includes: The gas concentration data, including smoke particle concentration data, is acquired in real time using dual-wavelength differential detection technology. The smoke particles and the interfering particles are distinguished by comparing the difference in scattered light intensity between the smoke particles and the interfering particles. Wherein, the intensity difference of scattered light from the smoke particles or interfering particles at different wavelengths satisfies: ； in, and They are respectively wavelength and The intensity of scattered light at the specified wavelength, where k is a correction factor.

17. The anti-interference processing method according to claim 3, characterized in that, The environmental data specifically includes one or more of the following: smoke particle concentration, visible light image, infrared thermal imaging image, carbon monoxide concentration, carbon dioxide concentration, and ambient temperature and humidity data.

18. An anti-interference processing device for a smoke and fire alarm system, characterized in that, include: The acquisition module is used to collect environmental data in real time through multiple sensors and process the environmental data to generate a fused feature vector. The comparison module is used to compare the similarity of the fused feature vector with the interference mode feature vector in the dynamic interference feature library to obtain the similarity result; An adjustment module is used to adjust the decision weights of feature vectors related to the interference pattern feature vectors based on the similarity results. The decision module is used to input the fused feature vector into the decision model. The decision model uses the adjusted decision weights to analyze the evolution of the fused feature vector in the time dimension to determine whether the fused feature vector is in a continuous state or an instantaneous state.

19. A smoke and fire alarm system, characterized in that, include: Multiple sensors are used to collect environmental data in real time; The anti-interference processing apparatus as described in claim 18 is used to perform the anti-interference processing method as described in any one of claims 1 to 17.

Images