A resident-oriented internet of things linkage disposal method

Abstract

This invention relates to an IoT-based linkage response method for smoke detectors in residential settings. The method includes: acquiring alarm information reported by multiple smoke detector terminals located at the residential end, with all terminals establishing a communication connection via the Internet of Things (IoT); determining whether a fire has occurred based on the alarm information; and if a fire has occurred, generating and issuing a processing work order based on the smoke detector terminals. This method, by determining whether a fire has occurred based on alarm information and generating and issuing a processing work order after confirming a fire, improves the responsiveness of alarm information and solves the problem of "no one responding when the alarm sounds."

Description

Technical Field

[0001] This invention relates to the field of smoke control technology, and in particular to an IoT-linked smoke detection method for residential users. Background Technology

[0002] In recent years, the demand for fire prevention and control in residential buildings has been continuously increasing. Urban communities, older residential areas, rental housing, and mixed-use commercial and residential buildings along streets are generally characterized by high electrical loads, frequent open flames in kitchens, and complex resident demographics (such as elderly people living alone, people going out during the day, and people sleeping at night). This makes early fire detection and rapid response a crucial aspect of grassroots safety governance. Smoke detectors, due to their low cost, ease of installation, and rapid response to initial fires, are widely used in residential homes and community grid-based safety management.

[0003] In residential settings, smoke detectors are susceptible to interference from non-fire sources, such as cooking fumes, steam, smoke, dust, and burnt food. These factors often cause a short-term increase in smoke concentration, triggering the alarm. Relying solely on residents for self-response presents several challenges. First, residents may not be present when the alarm occurs, may not hear it, or may not be able to confirm it in time, resulting in a lack of action despite the alarm being triggered. Second, long-term operation of smoke detectors can lead to battery depletion, sensor contamination, communication breakdowns, and equipment damage, resulting in devices that are "installed but not used effectively, and eventually fail," ultimately leading to the risk of "no one responding and no one taking responsibility." Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides an Internet of Things-based linkage method for handling smoke detectors at the residential end.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0008] An IoT-based method for coordinated response to smoke detectors in residential settings, comprising:

[0009] Obtain alarm information reported by smoke detector terminals; there are multiple smoke detector terminals, all located at the residential end, and all smoke detector terminals establish communication connections based on the Internet of Things;

[0010] Determine whether a fire has occurred based on alarm information;

[0011] In the event of a fire, a processing work order will be generated based on the smoke detector terminal and then issued.

[0012] (III) Beneficial Effects

[0013] The beneficial effects of this invention are as follows: This invention relates to an IoT-based linkage response method for smoke detectors in residential settings. The method includes: acquiring alarm information reported by multiple smoke detector terminals located at the residential end, and all smoke detector terminals establishing a communication connection based on the Internet of Things (IoT); determining whether a fire has occurred based on the alarm information; and if a fire has occurred, generating and issuing a processing work order based on the smoke detector terminals. This method, by determining whether a fire has occurred based on alarm information and generating and issuing a processing work order after confirming a fire, improves the responsiveness of alarm information and solves the problem of "no one responding when the alarm sounds." Attached Figure Description

[0014] Figure 1 A flowchart illustrating an IoT-based smoke detector linkage method for residential smoke detection provided by this invention;

[0015] Figure 2 This is a schematic diagram of the structure of a smoke recognition system based on multi-feature fusion and a lightweight gradient booster provided by the present invention. Detailed Implementation

[0016] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] In residential settings, smoke detectors are susceptible to interference from non-fire sources, such as cooking fumes, steam, smoke, dust, and burnt food. These factors often cause a short-term increase in smoke concentration, triggering the alarm. Relying solely on residents for self-response presents several challenges. First, residents may not be present when the alarm occurs, may not hear it, or may not be able to confirm it in time, resulting in a lack of action despite the alarm being triggered. Second, long-term operation of smoke detectors can lead to battery depletion, sensor contamination, communication breakdowns, and equipment damage, resulting in devices that are "installed but not used effectively, and eventually fail," ultimately leading to the risk of "no one responding and no one taking responsibility."

[0018] To address this problem, this invention relates to an IoT-based coordinated response method for smoke detectors in residential settings. The method includes: acquiring alarm information reported by multiple smoke detector terminals located at the residential end, with all terminals establishing a communication connection via the Internet of Things (IoT); determining whether a fire has occurred based on the alarm information; and if a fire has occurred, generating and issuing a processing work order based on the smoke detector terminals. This invention improves the responsiveness of alarm information by determining whether a fire has occurred based on alarm information, and then generating and issuing a processing work order after confirming a fire, thus solving the problem of "no one responding when the alarm sounds."

[0019] The IoT-based smoke detector linkage method for residential users provided in this embodiment can be implemented by an internet-based outsourced maintenance closed-loop system. In this system, outsourced maintenance involves the platform or a third-party maintenance team performing operational and maintenance tasks on behalf of the user, such as alarm verification, work order dispatch, on-site inspection, and repair and restoration.

[0020] The Internet-based outsourced maintenance closed-loop system includes: a sensing terminal layer, a platform access layer, an early warning analysis and push layer, a work order dispatch and coordinated handling layer, and a closed-loop feedback and archiving layer. This Internet-based outsourced maintenance closed-loop system implements the IoT-based coordinated handling method for residential smoke detectors provided in this embodiment. Based on networked smoke detector terminals, it realizes full-process management from alarm information reception, alarm information processing, fire handling, closed-loop feedback and archiving layers, forming a closed-loop maintenance system of "problem discovery - problem handling - result feedback - continuous optimization". This solves problems such as "no alarm response, no equipment maintenance, and high cost of false alarm handling" in residential smoke detector applications.

[0021] 1. Perception Terminal Layer

[0022] This layer consists of smoke detector terminals.

[0023] There are multiple smoke detector terminals installed in residential areas (such as homes or key locations) to collect alarm information and monitor equipment operation status. For example, when a smoke detector terminal detects an alarm event, it uploads the alarm information to the platform access layer.

[0024] In addition to reporting alarm events, smoke detector terminals can also report device health information such as online status, power status, and fault status.

[0025] All smoke detector terminals establish communication connections based on the Internet of Things (IoT).

[0026] For example, each smoke detector terminal is a smoke alarm with communication capabilities, capable of uploading alarm information and device status to the platform. All smoke detector terminals constitute an Internet of Things (IoT), through which data, commands, and other information are transmitted.

[0027] The smoke detector terminal transmits alarm information, equipment operating status, and equipment health information to the platform access layer via the Internet of Things.

[0028] In addition, a composite sensor group will be installed around the smoke detector terminal installation location. This composite sensor group will periodically collect the raw electrical signals in the monitoring space corresponding to each smoke detector terminal.

[0029] The acquisition period of the composite sensor group and the acquisition period of the smoke detector terminal can be the same or different. This embodiment does not limit the relationship between the two.

[0030] 2. Platform Access Layer

[0031] It is used to receive data uploaded by terminals (such as alarm information, device operating status, device health information, etc.), and to complete protocol parsing, data verification and standardization processing to generate a unified event object and form unified event information.

[0032] 3. Early warning analysis and push notification layer

[0033] It is used to make a preliminary assessment of alarm events and to notify residents, duty personnel or relevant personnel of alarm information through telephone, SMS, push notifications or other means.

[0034] 4. Work order dispatch and coordinated handling layer

[0035] It is used to automatically generate work orders based on alarm information and dispatch them to response forces such as dispatchers, patrol personnel, property management, grid workers or fire stations to complete on-site verification and handling.

[0036] 5. Closed-loop feedback and archiving layer

[0037] It is used to collect on-site inspection results, evidence materials and handling records, and complete the event archiving; if equipment failure is found, it will further generate maintenance work orders until the equipment is restored to normal operation.

[0038] See Figure 1 This embodiment provides an IoT-based linkage response method for smoke detectors used in residential applications. The implementation process of this method is as follows:

[0039] 101, retrieve alarm information reported by the smoke detector terminal.

[0040] The system consists of multiple smoke detectors located at the residential end, and all smoke detectors establish communication connections based on the Internet of Things (IoT).

[0041] Step 101 can be executed by the platform access layer. When the smoke detector terminal detects an alarm event, it will upload the alarm information to the platform access layer. In step 101, the platform access layer obtains the alarm information reported by the smoke detector terminal.

[0042] After the platform access layer receives the alarm information reported by the smoke detector terminal, it will automatically generate an event record and simultaneously obtain the location of the smoke detector terminal that reported the alarm information, the resident's contact information, and the device status information.

[0043] The platform access layer then performs preliminary processing of the alarm information, such as contacting residents for verification via outbound phone calls, SMS notifications, or push notifications. If the resident can confirm that there is nothing abnormal at the scene, the alarm information can directly enter the low-risk closed-loop process; if the resident cannot answer or cannot confirm, subsequent steps 102 and 103 are executed.

[0044] In addition, if the smoke detector terminal also reports device operation data, the platform access layer will also obtain the device operation data reported by the smoke detector terminal, and issue an alert when an anomaly is determined based on the device operation data.

[0045] 102. Determine whether a fire has occurred based on alarm information.

[0046] This step is executed by the early warning analysis and push layer.

[0047] In practical implementation, the early warning and push layer can achieve step 102 through a smoke recognition system based on multi-feature fusion and a lightweight gradient booster.

[0048] A smoke detection system based on multi-feature fusion and a lightweight gradient booster can, for example, Figure 2 As shown, it includes: a hardware perception layer, a data preprocessing module, a multi-dimensional feature construction module, and a classification decision module.

[0049] The hardware sensing layer is used to periodically collect the raw electrical signals within the monitoring space corresponding to each smoke detector terminal.

[0050] The data preprocessing module is used to perform analog-to-digital conversion and standardization on the raw electrical signals acquired by the hardware sensing layer to obtain processed data.

[0051] The multidimensional feature construction module is used to extract feature operators from the processed data obtained from the data preprocessing module.

[0052] The classification decision module uses a lightweight gradient booster as a classifier to classify the feature operators extracted by the multi-dimensional feature construction module and obtain the smoke recognition results.

[0053] The acquisition of raw electrical signals by the hardware sensing layer is independent of whether step 102 is executed. That is, before, during, and after step 102, the hardware sensing layer periodically acquires raw electrical signals within the monitoring space corresponding to each smoke detector terminal. Whenever a raw electrical signal is acquired, the data preprocessing module performs analog-to-digital conversion and standardization to obtain processed data. Therefore, the processing behavior of the data preprocessing module is independent of whether step 102 is executed, and only depends on whether the hardware sensing layer acquires new raw electrical signals. Whenever processed data is obtained, the multi-dimensional feature construction module extracts feature operators from it. That is, the processing behavior of the multi-dimensional feature construction module is independent of whether step 102 is executed, and only depends on whether the data preprocessing module obtains processed data.

[0054] Thus, before executing step 102, the original electrical signals in the monitoring space corresponding to each smoke detector terminal will be periodically collected; the original electrical signals will be converted from analog to digital and standardized to obtain processed data; and feature operators will be extracted from the processed data.

[0055] In step 102, the feature operators corresponding to the target monitoring space are determined based on the alarm information; a lightweight gradient booster is used as a classifier to classify the feature operators corresponding to the target monitoring space to determine whether a fire has occurred.

[0056] For example, the early warning analysis and push layer determines the location of the smoke detector terminal that sent the alarm information based on the alarm information, and sends this location to the classification decision module. The classification decision module determines the feature operator corresponding to the composite sensor group at that location (i.e., determines the feature operator corresponding to the target monitoring space). The classification decision module then uses a lightweight gradient booster as a classifier to classify the feature operator corresponding to the target monitoring space to obtain the smoke identification result. The early warning analysis and push layer uses the smoke identification result obtained by the classification decision module as the result of whether a fire has occurred.

[0057] The original electrical signals include: humidity signal, temperature signal, first red light signal, second red light signal, first blue light signal, and second blue light signal.

[0058] The process of performing analog-to-digital conversion and standardization on the original electrical signal to obtain the processed data is as follows:

[0059] The humidity value is obtained by performing analog-to-digital conversion and standardization on the electrical signal of humidity.

[0060] The electrical signal of temperature is converted from analog to digital and standardized to obtain the temperature value.

[0061] The electrical signal of the first red light is subjected to analog-to-digital conversion and standardization to obtain the value of the first red light.

[0062] The electrical signal of the second red light is subjected to analog-to-digital conversion and standardization to obtain the value of the second red light.

[0063] The electrical signal of the first blue light is subjected to analog-to-digital conversion and standardization to obtain the first blue light value.

[0064] The electrical signal of the second blue light is subjected to analog-to-digital conversion and standardization to obtain the second blue light value.

[0065] Among them, the humidity value, temperature value, first red light value, second red light value, first blue light value, and second blue light value are all identified by the collection timestamp.

[0066] The process of extracting feature operators from the processed data is as follows:

[0067] After acquiring a raw electrical signal, instantaneous feature operators, time window feature operators, mode change feature operators, and state memory feature operators are extracted from the processed data acquired in that acquisition.

[0068] Among them, the instantaneous feature operator, the time window feature operator, the mode change feature operator, and the state memory feature operator all use the collection timestamp as an identifier.

[0069] Instantaneous feature operators are operators that reflect the data collected in a given time.

[0070] The time window feature operator is an operator that reflects continuous changes.

[0071] The pattern change feature operator is an operator that reflects jump changes.

[0072] The state memory feature operator is an operator that reflects the normal threshold.

[0073] If the processed data includes humidity value, temperature value, first red light value, second red light value, first blue light value, and second blue light value, then the instantaneous feature operators include: humidity value, temperature value, first red light value, second red light value, first blue light value, second blue light value, blue-red light difference value, blue-red light ratio value, and saturation flag.

[0074] Wherein, the difference between blue and red light = first blue light value - first red light value.

[0075] Blue-to-red light ratio = first blue light value / (first red light value + epsilon). Where epsilon is a preset minimum positive number.

[0076] The saturation flag includes a red light flag and a blue light flag. When the second red light value is greater than a preset full-scale offset threshold, the red light flag is 1; when the second red light value is not greater than the preset full-scale offset threshold, the red light flag is 0. When the second blue light value is greater than a preset full-scale offset threshold, the blue light flag is 1; when the second blue light value is not greater than the preset full-scale offset threshold, the blue light flag is 0.

[0077] In addition, the time window feature operators include: the mean of the processed data for each time window, the maximum value of the processed data for each time window, the minimum value of the processed data for each time window, the standard deviation of the processed data for each time window, the rate of change of the processed data, and the duration for which the processed data meets the preset rules.

[0078] The time window is a sliding window that starts from the current acquisition time and moves forward according to a preset time step.

[0079] The rate of change of the processed data is the difference between the mean of the processed data in the second time window and the mean of the processed data in the first time window, divided by 2 times the preset time step.

[0080] If the preset rule is that the saturation flag is 1, then the duration for which the processed data satisfies the preset rule is the duration between the first moment and the second moment. Here, the first moment is the first acquisition moment when both the red and blue flags are 1, starting from the current acquisition moment, and the second moment is the last acquisition moment when both the red and blue flags are 1, starting from the first moment.

[0081] 103. In the event of a fire, a processing work order will be generated based on the smoke detector terminal and then issued.

[0082] This step is executed by the work order dispatch and linkage handling layer. The work order dispatch and linkage handling layer generates a processing work order according to preset rules and dispatches it to the corresponding handling force.

[0083] The work orders include at least two categories: alarm handling work orders and equipment maintenance work orders.

[0084] The content of a work order should include at least the following: equipment number, installation address, contact person, alarm time, incident level, handling requirements, on-site arrival time limit, on-site feedback requirements, and final handling result.

[0085] After completing the on-site verification, the personnel handling the incident need to send back the handling evidence and conclusion labels. Conclusion labels may include: actual fire, oil fumes, water vapor, smoke extraction, dust, equipment malfunction, test alarm, etc. The closed-loop feedback and archiving layer determines whether the event is closed based on the returned results and archives the relevant information.

[0086] After executing step 103, the execution status of the work order will be confirmed, such as by sending relevant personnel to the site for verification in the closed-loop feedback and archiving layer, and uploading on-site photos, videos, text descriptions and final conclusions via mobile devices.

[0087] The closed-loop feedback and archiving layer completes closed-loop archiving based on on-site feedback.

[0088] If the closed-loop feedback and archiving layer discovers other problems such as equipment being offline, damaged, or having insufficient power based on on-site feedback, it will automatically switch to the equipment maintenance closed loop until the equipment returns to normal.

[0089] The Internet-based maintenance closed-loop system provided in this embodiment implements the IoT-based linkage response method for residential smoke detectors, achieving a closed loop throughout the entire process from alarm information reporting, alarm information processing, work order dispatch, on-site verification to result feedback and archiving, ensuring that every alarm information is handled and results are provided.

[0090] The Internet-based maintenance closed-loop system provided in this embodiment can continuously monitor the offline, faulty, and abnormal states of equipment by executing the IoT-based linkage handling method for residential smoke detectors provided in this embodiment. It can also automatically trigger maintenance work orders after a problem is detected, ensuring the long-term effective operation of the equipment. This not only solves the problem of "what to do after an alarm" but also the problem of "whether the equipment is reliable in the long term".

[0091] The Internet-based maintenance closed-loop system provided in this embodiment can achieve unified access, timely response, rapid handling, and full-process traceability of smoke detector alarm events by executing the IoT-based linkage handling method for residential smoke detectors provided in this embodiment. This improves alarm response efficiency and reduces the risk of missed handling. At the same time, through the device health closed loop, it improves the long-term online rate and reliability of the devices, providing a stable foundation for the optimization of hazard classification and hierarchical scheduling.

[0092] In practice, to make smoke detectors more effective, they can be installed in the locations where fires are most likely to occur, thus avoiding the waste caused by installing them in ineffective locations.

[0093] Because electricity consumption data reflects residential conditions, the location of smoke detectors at residential sites can be determined by analyzing their electricity consumption. For example, the electricity consumption of each meter at the residential site can be collected at regular intervals. Whenever a new electricity consumption data is collected for any meter, steps 201-203 are executed to determine the location of the smoke detector. If a smoke detector is already installed at the determined location, no action is taken. If no smoke detector is installed at the determined location, it is installed there. If an already installed smoke detector is no longer identified as a suitable location after steps 201-203, relevant personnel are notified to determine whether to cancel the installation.

[0094] The electricity collection cycle is preset. For example, if the electricity collection cycle is one month, the electricity consumption of each household meter in the most recent month will be collected once a month.

[0095] Taking any electricity meter z as an example, the implementation process of steps 201-203 is as follows:

[0096] 201. If the newly collected electricity consumption is greater than the first consumption threshold, then the household corresponding to any electricity meter is determined as the location of the smoke detector terminal at the resident end.

[0097] The first consumption threshold is a threshold that reflects high power consumption. The higher the power consumption, the more people living in the household corresponding to the meter and / or the more high-power appliances (such as electric heaters) are used. Either way, it will increase the risk of fire.

[0098] The first consumption threshold can be determined based on the actual electricity consumption of residents. Taking a one-month electricity collection period as an example, if residents consume more than 500 kWh of high-power appliances that are prone to fire, then the first consumption threshold can be set at 500 kWh. Alternatively, the first consumption threshold can be dynamically determined based on seasonal conditions. For example, in winter, high-power appliances prone to fire (such as electric heaters) are frequently used, and the environment is conducive to fire, so the first consumption threshold can be slightly lower, such as 400 kWh. In spring, the first consumption threshold can be slightly higher, such as 500 kWh. This embodiment does not limit the method for determining the first consumption threshold.

[0099] If the energy consumption of the newly collected electricity meter z is greater than the first consumption threshold (e.g., 500 kWh), it means that the household corresponding to electricity meter z is using high-power electrical appliances that are prone to fire (e.g., electric heaters), which may cause a fire. Continuous monitoring is required, and the household corresponding to electricity meter z can be identified as the location of the smoke detector terminal at the resident end.

[0100] 202. If the newly collected energy consumption is less than the second consumption threshold, then:

[0101] 1. If the energy consumption collected by any meter in a first number of consecutive energy collection periods is less than the second consumption threshold, determine whether the location of the smoke detector terminal is the residential end based on the neighboring meters of any meter.

[0102] 2. If the energy consumption collected by any meter in a first number of consecutive energy consumption collection periods is not all less than the second consumption threshold, determine whether the location of the smoke detector terminal is the residential end based on the historical energy consumption of any meter.

[0103] The second consumption threshold is less than the first consumption threshold.

[0104] The second consumption threshold is a threshold that reflects low power consumption. The lower the power consumption, the fewer people living in the household corresponding to the meter (e.g., no one lives there) and / or the less high-power appliances (e.g., electric heaters) are used. Either way, it will reduce the risk of fire.

[0105] The second consumption threshold can be determined based on actual residential electricity consumption. For example, taking a one-month electricity collection period, if residents who are not at home for extended periods consume less than 50 kWh, then the second consumption threshold can be set at 50 kWh. Alternatively, the second consumption threshold can be dynamically determined based on seasonal conditions. For instance, in winter, high-power electrical appliances (such as electric heaters) are frequently used, and the environment is conducive to fires, so the second consumption threshold can be slightly lower, such as 10 kWh. In spring, the second consumption threshold can be slightly higher, such as 50 kWh. This embodiment does not limit the method for determining the second consumption threshold.

[0106] Furthermore, the first quantity reflects that the low power consumption is a continuous phenomenon, not an occasional occurrence. The first quantity is also determined empirically. For example, a first quantity of 3 indicates that the power consumption collected in three different time periods is not less than the second consumption threshold, suggesting that the resident has consistently low power consumption. This implies that the resident may be unoccupied or rarely uses appliances. Otherwise, it is considered an occasional instance of low power consumption, and it cannot be assumed that the resident may be unoccupied or rarely uses appliances. For another example, the first quantity can be dynamically determined based on seasonal conditions. In winter, high-power appliances (such as electric heaters) are frequently used, and the environment is conducive to fires. In this case, the first quantity can be slightly larger, such as 6, requiring a longer period of consistently low power consumption to indicate that the resident is unoccupied or rarely uses appliances. In spring, the first quantity can be slightly smaller, such as 3. This embodiment does not limit the determination of the first quantity.

[0107] Taking the first quantity = 3 as an example, if the energy consumption of the newly collected meter z is less than the second consumption threshold (e.g., 50 kWh), then:

[0108] 1. If the energy consumption recorded by meter z for three consecutive data collection periods is less than the second consumption threshold (including the current data collection period, i.e., the energy consumption recorded in the current data collection period is less than 50 kWh, the energy consumption recorded in the previous data collection period is less than 50 kWh, and the energy consumption recorded in the second data collection period before the current data collection period is also less than 50 kWh), it indicates that the household corresponding to meter z may be unoccupied or rarely uses electrical appliances. In this case, the location of the smoke detector terminal can be determined by checking the neighboring meters of meter z.

[0109] 2. If the energy consumption collected by meter z in three consecutive energy consumption collection periods is not consistently less than the second consumption threshold (including the current collection period; for example, if the energy consumption collected in the current collection period is less than 50 kWh, but the energy consumption collected in the previous collection period and / or the energy consumption collected in the second collection period before the current collection period is not less than 50 kWh), it indicates that meter z only has occasional low energy consumption, and it cannot be determined whether the corresponding household is unoccupied or uses appliances very little. In this case, the location of the smoke detector terminal can be determined based on the historical energy consumption of meter z.

[0110] 203. If the newly collected energy consumption is not greater than the first consumption threshold and not less than the second consumption threshold, then determine whether it is the location of the residential end where the smoke detector terminal is located based on the historical energy consumption of any meter.

[0111] If the energy consumption of the newly collected electricity meter z is not greater than the first consumption threshold (e.g., 500 kWh) and not less than the second consumption threshold (e.g., 50 kWh), that is, the energy consumption of the newly collected electricity meter z is between the second consumption threshold and the first consumption threshold, then the location of the smoke detector terminal in the residential area will be determined based on the historical energy consumption of electricity meter z.

[0112] The process of determining whether the location of the smoke detector terminal is within the residential area based on the neighboring meters of any given meter (i.e., the scheme executed when the newly collected energy consumption in step 202 is less than the second consumption threshold, and the energy consumption collected by any given meter in a consecutive first number of energy collection time periods is less than the second consumption threshold) is as follows:

[0113] 601. Based on the location of each meter, determine the neighboring meters of any given meter.

[0114] Neighboring meters can be determined based on location. Taking any meter as meter z and its coordinates as coordinate z, a circle is drawn with coordinate z as the center and a preset distance (e.g., 500 meters) as the radius. All meters within the area covered by this circle are considered neighboring meters. Alternatively, all meters in the unit containing coordinate z can be considered neighboring meters. This embodiment does not limit the method for determining neighboring meters.

[0115] 602. If a first target meter exists in the neighboring meters, or if the proportion of second target meters in the neighboring meters is greater than a preset proportion, then the household corresponding to any meter is determined as the location of the smoke detector terminal at the resident end.

[0116] Among them, the latest energy consumption collected by the first target meter is greater than the first consumption threshold, that is, the energy consumption of the first target meter is relatively high in the most recent energy collection period.

[0117] The latest energy consumption collected by the second target meter is not greater than the first consumption threshold and not less than the second consumption threshold. However, the energy consumption collected in the second consecutive number of energy consumption collection periods (including the current collection period) from the current collection period is not less than the first consumption threshold - (first consumption threshold - second consumption threshold) / 3.

[0118] The second quantity reflects the meter's recent energy consumption. This second quantity is also determined empirically. For example, a second quantity of 3 indicates that the energy consumption collected over three data collection periods is not less than the first consumption threshold - (first consumption threshold - second consumption threshold) / 3. This suggests that while the meter's energy consumption is not particularly high, its recent energy consumption has been relatively high. Alternatively, the second quantity can be dynamically determined based on seasonal conditions. For instance, in winter, high-power electrical appliances (such as electric heaters) are frequently used, and the environment is conducive to fires. In such cases, the second quantity can be slightly lower, such as 3, allowing for a shorter period of sustained high energy consumption. In spring, the second quantity can be slightly higher, such as 6. This embodiment does not limit the determination of the second quantity.

[0119] Taking a first consumption threshold of 500 kWh and a second consumption threshold of 50 kWh as an example, and any meter as meter z, and the neighboring meters of meter z as meters w1, w2, w3, w4, and w5, the first consumption threshold - (first consumption threshold - second consumption threshold) / 3 = 500 kWh - (500 kWh - 50 kWh) / 3 = 500 kWh - 450 kWh / 3 = 500 kWh - 150 kWh = 350 kWh.

[0120] If the newly collected electricity consumption of meter w1 is greater than 500 kWh, meaning meter w1 is the first target meter with high power consumption, and meter w1 corresponds to the residential location where the smoke detector is located, then even though the newly collected electricity consumption of meter z is normal, because it is close to meter w1, it is also identified as the residential location where the smoke detector is located and is monitored.

[0121] Alternatively, although the energy consumption of newly collected meters w1, w2, w3, w4, and w5 is all greater than 500 kWh, meaning that the recent energy consumption of meters w1, w2, w3, w4, and w5 is not particularly high, meter w1 is the second target meter (i.e., the energy consumption collected by meter w1 in the current data collection period is not less than 350 kWh, the energy consumption collected in the previous data collection period is not less than 350 kWh, and the energy consumption collected in the second data collection period before the current data collection period is not less than 350 kWh). The energy consumption of meter w2 is not less than 350 kWh, the energy consumption of meter w2 in the current data collection period is not less than 350 kWh, the energy consumption of meter w2 in the previous data collection period is not less than 350 kWh, and the energy consumption of meter w2 in the second data collection period before the current data collection period is not less than 350 kWh. Meter w3 is the second target meter (i.e., the energy consumption of meter w3 in the current data collection period is not less than 350 kWh, and the energy consumption of meter w3 in the second data collection period before the current data collection period is not less than 350 kWh). The collected electricity consumption is not less than 350 kWh, and the electricity consumption collected in the second collection period before the current collection period is also not less than 350 kWh. Meter W4 is neither the first target meter nor the second target meter (i.e., the electricity consumption collected by meter W4 in at least one of the following periods: the current collection period, the collection period before the current collection period, and the second collection period before the current collection period is less than 350 kWh). Meter W5 is neither the first target meter nor the second target meter (i.e., the electricity consumption collected by meter W5 in the current collection period is less than 350 kWh). If the energy consumption collected in at least one of the following time periods is less than 350 kWh (e.g., the energy consumption of the first time period before the current energy collection time period and the second time period before the current energy collection time period), then the proportion of the second target meter in the neighboring meters is 3 / 5 = 0.6. If the preset proportion is 0.5, it means that when more than half of the neighboring meters reflect high energy consumption, even if the newly collected energy consumption of meter Z is normal, the high energy consumption reflected by the surrounding meters poses a fire risk. Therefore, it is also identified as the location of the smoke detector terminal at the residential end and monitored.

[0122] The preset ratio is also a pre-set empirical value used to determine the proportion of neighboring meters that show excessively high electricity consumption. For example, if the preset ratio is 0.5, then if more than half of the neighboring meters show excessively high electricity consumption, it is considered that the resident has consistently low electricity consumption, and it can be assumed that the resident is either unoccupied or rarely uses appliances. Otherwise, it is considered that the low electricity consumption is occasional, and it cannot be assumed that the resident is unoccupied or rarely uses appliances. Furthermore, the preset ratio can be dynamically determined based on seasonal conditions. For example, in winter, high-power electrical appliances (such as electric heaters) are frequently used, and the environment is conducive to fire, so the preset ratio can be slightly smaller, such as 0.2, thus a smaller proportion of the second target meters is sufficient. In spring, the preset ratio can be slightly larger, such as 0.5. This embodiment does not limit the determination method of the preset ratio.

[0123] Furthermore, the process for determining whether the location of the smoke detector terminal is the residential end based on the historical energy consumption of any electricity meter (i.e., the scheme executed when the newly collected energy consumption in step 202 is less than the second consumption threshold, and the energy consumption collected by any electricity meter in a consecutive first number of energy collection time periods is not all less than the second consumption threshold, or the scheme executed when the newly collected energy consumption in step 203 is not greater than the first consumption threshold and not less than the second consumption threshold) is as follows:

[0124] 701. Determine the mean and standard deviation of the energy consumption collected by any meter during each energy collection period.

[0125] For example, if the energy consumption of meter z has been collected for 10 time periods, then in step 701, the mean and standard deviation of these 10 energy consumption values ​​are determined.

[0126] If a large number of energy consumption data have been collected, energy consumption data from too long ago will not be of much reference value to the current energy consumption of any meter. In step 701, a maximum number (e.g., 5) will be set. In this way, even if the energy consumption of meter z for 10 data collection periods has been collected, only the mean and standard deviation of the 5 most recently collected energy consumption data will be determined in step 701.

[0127] The mean and standard deviation of the energy consumption collected by any meter during each data collection period reflect the recent average power consumption and fluctuations of that meter.

[0128] 702. If the energy consumption collected in the third consecutive energy consumption collection period (including the current collection period) is greater than the sum of the average energy consumption and the standard deviation of the energy consumption, and the energy consumption collected in the third consecutive energy consumption collection period (including the current collection period) shows an increasing trend over time, then the household corresponding to any meter is determined as the location of the smoke detector terminal at the residential end.

[0129] The third quantity reflects whether the power consumption is abnormal. This third quantity is also determined empirically, for example, it might be 3. Alternatively, the third quantity can be dynamically determined based on seasonal conditions. For instance, in winter, high-power electrical appliances (such as electric heaters) are frequently used, and the environment is conducive to fire, so the third quantity can be slightly smaller, such as 3, requiring a shorter duration. In spring, the first quantity can be slightly larger, such as 6. This embodiment does not limit the method for determining the third quantity.

[0130] Taking any electricity meter as meter z, the third quantity as 3, the mean of electricity consumption of meter z as the mean z, and the standard deviation of electricity consumption of meter z as the standard deviation z, if the electricity consumption collected by meter z in the current time period is greater than the mean z + standard deviation z, the electricity consumption collected in the previous time period is greater than the mean z + standard deviation z, and the electricity consumption collected in the second time period before the current time period is also greater than the mean z + standard deviation z, then it means that the electricity consumption of meter z recently is higher than the recent average electricity consumption of meter z considering fluctuations, and it means that meter z is abnormal recently. If, at this time, the energy consumption collected by meter z in the current data collection period, the energy consumption collected in the previous data collection period, and the energy consumption collected in the second data collection period before the current data collection period all show an increasing trend over time, and this abnormality shows an upward trend, then meter z is exhibiting a clear abnormal phenomenon, and a fire may occur. In this case, the household corresponding to meter z is determined as the location of the smoke detector terminal in the residential area.

[0131] Through the above steps 201-203, the electricity meter that may be involved in a fire can be obtained. The location of this electricity meter is the same as the location of the smoke detector terminal at the resident's end, ensuring the accurate installation of the smoke detector terminal and guaranteeing the accuracy of monitoring while avoiding waste.

[0132] It should be noted that determining the location of the smoke detector terminal at the resident's end through steps 201-203 is only a coarse-grained location, such as a house number, not a specific location. In other words, steps 201-203 can determine that a household using a certain electricity meter needs to install a smoke detector terminal, but it cannot determine the exact location within that household or the number of terminals to be installed. Therefore, after determining the location of the smoke detector terminal at the resident's end through steps 201-203, relevant personnel can go to the resident's home to confirm the required number of smoke detector terminals and their final locations (e.g., the final location of the smoke detector terminals is the resident's kitchen, living room, or hallway), and then install the smoke detector terminals at those final locations. Simultaneously, a composite sensor array should be installed around that final location.

[0133] The implementation details of the smoke recognition system based on multi-feature fusion and lightweight gradient booster are described below.

[0134] like Figure 2 As shown, the smoke recognition system based on multi-feature fusion and lightweight gradient booster includes: a hardware perception layer, a data preprocessing module, a multi-dimensional feature construction module, and a classification decision module.

[0135] 1. Hardware Perception Layer

[0136] The hardware sensing layer comprises a composite sensor array. The installation location of the composite sensor array is around the installation location of the smoke detector terminal; the specific location can be determined by relevant personnel based on the site conditions.

[0137] The composite sensor array is used to periodically collect the raw electrical signals within the monitoring space corresponding to each smoke detector terminal.

[0138] The composite sensor group can be a single group or multiple groups arranged at different locations and / or different heights around the installation site.

[0139] Any composite sensor group shall include at least: a humidity sensor (for real-time monitoring of ambient humidity), a temperature sensor (for real-time monitoring of ambient temperature), and a photoelectric sensor group.

[0140] Each photoelectric sensor group includes: a first red light sensor, a second red light sensor, a first blue light sensor, and a second blue light sensor.

[0141] The first and second red light sensors have different acquisition wavelengths.

[0142] The first and second blue light sensors have the same acquisition wavelength, but different optical path angles and scattering reception directions.

[0143] The first red light sensor, the second red light sensor, the first blue light sensor, and the second blue light sensor are arranged in a compact and concentrated manner.

[0144] In addition, it may include multiple third red light sensors and third blue light sensors. This embodiment does not limit whether to include third red light sensors and third and blue light sensors, or the number of third red light sensors and third and blue light sensors included. However, it is necessary to ensure that the number of third red light sensors and third blue light sensors is the same, and that each red light sensor collects a different wavelength, and each blue light sensor collects the same wavelength, but the optical path angle and scattering reception direction of each blue light sensor are different.

[0145] In addition, the terms "first," "second," and "third" used in this embodiment are used to distinguish between different red light sensors and different blue light sensors.

[0146] Taking the first and second red light sensors as examples, both are red light sensors, differing only in the wavelength of the red light they collect. For instance, the first red light sensor collects red light in the 4.3–4.4 μm (micrometer) range, thereby acquiring the raw electrical signal. This sensor can detect the infrared radiation of hydrocarbon flames (wood, oil, gas, etc.) to obtain flame characteristics. The first red light sensor exhibits a strong signal when flames are present and a weak signal when encountering ordinary heat sources (light bulbs, heaters). The second red light sensor collects the raw electrical signal at 3.8 μm (or 5.0 μm). This sensor can detect ambient background infrared radiation (sunlight, lamplight, high-temperature equipment) to obtain environmental interference, which is then compared with the raw electrical signal obtained by the first red light sensor to determine if it is a real fire. The second red light sensor exhibits a strong signal from non-flame heat sources and a relatively weak signal from flames.

[0147] Taking the first and second blue light sensors as examples, both are blue light sensors that collect the same wavelength, differing only in their optical path angle and scattering / receiving direction. For instance, both sensors collect raw electrical signals at 460nm (nanometers). One sensor detects forward at 20°–60°, while the other detects backward at 120°–160°, resulting in different optical path angles and scattering / receiving directions. The first blue light sensor collects the flame-specific raw electrical signal to monitor the light radiation and flickering changes of open flame combustion, serving as the main channel for fire detection. The second blue light sensor collects background raw electrical signals from ambient natural light, artificial light, etc., as a compensation channel.

[0148] 2. Data Preprocessing Module

[0149] The data preprocessing module is used to perform analog-to-digital conversion and standardization on the raw electrical signals acquired by the hardware sensing layer to obtain processed data.

[0150] The data preprocessing module performs analog-to-digital conversion and standardization on the raw electrical signals collected by each group of composite sensors to obtain the corresponding processed data.

[0151] Taking any composite sensor group as an example, which includes at least a humidity sensor, a temperature sensor, and a photoelectric sensor group; and each photoelectric sensor group including a first red light sensor, a second red light sensor, a first blue light sensor, and a second blue light sensor, the data preprocessing module performs analog-to-digital conversion and standardization on the raw electrical signals collected by any composite sensor group to obtain the corresponding processed data.

[0152] The data preprocessing module performs analog-to-digital conversion and standardization on the raw electrical signals collected by the humidity sensor to obtain the humidity value (H). It also performs analog-to-digital conversion and standardization on the raw electrical signals collected by the temperature sensor to obtain the temperature value (T). Furthermore, it performs analog-to-digital conversion and standardization on the raw electrical signals collected by the first red light sensor to obtain the first red light value (IR1). The same applies to the raw electrical signals collected by the second red light sensor, yielding the second red light value (IR2). Finally, it performs analog-to-digital conversion and standardization on the raw electrical signals collected by the first blue light sensor to obtain the first blue light value (BL1). Finally, it performs analog-to-digital conversion and standardization on the raw electrical signals collected by the second blue light sensor to obtain the second blue light value (BL2).

[0153] The first red light value (IR1) and the second red light value (IR2) are both red light sensing signals, but the original electrical signals they correspond to are acquired by different devices. The first blue light value (BL1) and the second blue light value (BL2) are both blue light sensing signals, but the original electrical signals they correspond to are acquired by different devices.

[0154] Humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) are all identified by the collection timestamp.

[0155] If there are multiple sets of composite sensor groups, since the positions and / or heights of each set of composite sensor groups are different, the humidity value (H), temperature value (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) are all identified by the three-dimensional coordinates and acquisition timestamp of the corresponding composite sensor group. In this way, the processed data of a unique set of composite sensor groups (such as humidity value (H), temperature value (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2)) can be obtained through the identification.

[0156] 3. Multidimensional Feature Construction Module

[0157] The multidimensional feature construction module is used to extract feature operators from the processed data obtained from the data preprocessing module. It is responsible for extracting feature operators from four dimensions: instantaneous, temporal, mode switching, and state memory.

[0158] Taking a hardware sensing layer comprising at least one set of composite sensor groups, where each composite sensor group includes at least: a humidity sensor, a temperature sensor, and a photoelectric sensor group; and each photoelectric sensor group includes: a first red light sensor, a second red light sensor, a first blue light sensor, and a second blue light sensor, as an example, since the hardware sensing layer periodically collects raw electrical signals within the space, that is, each time the humidity sensor, temperature sensor, and photoelectric sensor group collects a raw electrical signal, the data preprocessing module processes the raw electrical signal to obtain the corresponding humidity value (H), temperature value (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2). Thus, the multi-dimensional feature construction module is used to extract instantaneous feature operators, time window feature operators, mode change feature operators, and state memory feature operators from the processed data obtained by the data preprocessing module after the hardware sensing layer collects a raw electrical signal.

[0159] Among them, the instantaneous feature operator, the time window feature operator, the mode change feature operator, and the state memory feature operator all use the collection timestamp as an identifier.

[0160] If there are multiple composite sensor groups, since the positions and / or heights of each composite sensor group are different, the instantaneous feature operator, time window feature operator, mode change feature operator, and state memory feature operator are all identified by the three-dimensional coordinates and acquisition timestamps of the corresponding composite sensor group. In this way, the instantaneous feature operator, time window feature operator, mode change feature operator, and state memory feature operator of a unique composite sensor group can be obtained through the identification.

[0161] The following section uses any set of composite sensor groups as an example to illustrate the process of determining its instantaneous feature operator, time window feature operator, mode change feature operator, and state memory feature operator.

[0162] 1) Instantaneous feature operator

[0163] Instantaneous feature operators are operators that reflect the data collected in a given time.

[0164] Instantaneous feature operators are obtained based on a single acquisition device (such as a humidity sensor, temperature sensor, first red light sensor, second red light sensor, first blue light sensor, and second blue light sensor). Instantaneous feature operators reflect the absolute physical state of the acquisition device at the current acquisition moment.

[0165] Taking the processed data including humidity value (H), temperature value (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) as an example, the instantaneous feature operator includes: humidity value (H), temperature value (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), second blue light value (BL2), blue-red light difference, blue-red light ratio, and saturation flag.

[0166] (1) Humidity value (H), temperature value (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) are the processed data obtained by the data preprocessing module, which can be read directly here.

[0167] (2) The blue-red light difference and blue-red light ratio are spectral scattering ratio characteristics. This characteristic utilizes the difference in scattering rate of particles of different sizes to light of different wavelengths (for example, the scattering ratio of oil fume particles and fire smoke particles is significantly different).

[0168] Wherein, the blue-red light difference = first blue light value - first red light value, such as blue-red light difference = BL1 - IR1.

[0169] Blue-to-red light ratio = first blue light value / (first red light value + epsilon).

[0170] Here, epsilon is a preset very small positive number used to prevent the denominator from being zero.

[0171] For example, the blue-to-red light ratio = BL1 / (IR1+epsilon).

[0172] (3) The saturation flag is used to reflect whether the second red light value (IR2) and the second blue light value (BL2) are close to full offset, and to determine the high concentration of smoke environment.

[0173] The saturation flag is a Boolean flag, such as the red light flag and the blue light flag.

[0174] When the second red light value (IR2) is greater than the preset full-scale threshold (e.g., 810), the red light flag is 1; when the second red light value (IR2) is not greater than the preset full-scale threshold (e.g., 810), the red light flag is 0.

[0175] When the second blue light value (BL2) is greater than the preset full-scale threshold (e.g., 810), the blue light flag is 1; when the second blue light value (BL2) is not greater than the preset full-scale threshold (e.g., 810), the blue light flag is 0.

[0176] 2) Time window feature operator

[0177] The time window feature operator is an operator that reflects continuous changes.

[0178] The instantaneous feature operator is an operator based on a multi-data-point sliding window that can capture the dynamic characteristics of smoke diffusion.

[0179] The time window feature operators include: the mean of the processed data for each time window, the maximum value of the processed data for each time window, the minimum value of the processed data for each time window, the standard deviation of the processed data for each time window, and the rate of change of the processed data. V), the duration for which the processed data meets the preset rules ( (time).

[0180] The time window is a sliding window that starts from the current acquisition time and moves backward with a preset time step (such as 6 seconds, 10 seconds, or 15 seconds). The preset time step is longer than the acquisition cycle of the hardware sensing layer.

[0181] Taking a preset time step of 15 seconds as an example, the sliding window time length is 15 seconds. Then, the processed data obtained from each acquisition is acquired. Starting from the current acquisition time (e.g., t0), all processed data within the acquisition time range of t0 to t0-15 seconds (inclusive of t0, exclusive of t0-15) are formed into one time window of processed data; all processed data within the acquisition time range of t0-15 seconds to t0-30 seconds (inclusive of t0-15, exclusive of t0-30) are formed into another time window of processed data; ..., this process is repeated to obtain processed data for multiple time windows.

[0182] Each time window is sorted from the most recent to the oldest collection time to obtain a time window sequence, such as the time window sequence {time window 1, time window 2, time window 3}.

[0183] (1) Mean of the processed data for each time window, maximum value of the processed data for each time window, minimum value of the processed data for each time window, and standard deviation of the processed data for each time window.

[0184] Taking any time window as time window 1, and the processed data including humidity value (H), temperature value (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) as an example, the mean of each humidity value (H) in time window 1 is determined as mean1(H) of humidity in time window 1, the maximum value of each humidity value (H) in time window 1 is determined as maximum value of humidity in time window 1 max1(H), the minimum value of each humidity value (H) in time window 1 is determined as minimum value of humidity in time window 1 min1(H), and the standard deviation of each humidity value (H) in time window 1 is determined as standard deviation of humidity in time window 1 std1(H).

[0185] The mean of each temperature value (T) in time window 1 is defined as mean1(T) of time window 1, the maximum value of each temperature value (T) in time window 1 is defined as maximum value of temperature in time window 1 max1(T), the minimum value of each temperature value (T) in time window 1 is defined as minimum value of temperature in time window 1 min1(T), and the standard deviation of each temperature value (T) in time window 1 is defined as standard deviation of temperature in time window 1 std1(T).

[0186] The mean of each first red light value (IR1) in time window 1 is defined as mean1(IR1) of the first red light in time window 1. The maximum value of each first red light value (IR1) in time window 1 is defined as max1(IR1) of the first red light in time window 1. The minimum value of each first red light value (IR1) in time window 1 is defined as min1(IR1) of the first red light in time window 1. The standard deviation of each first red light value (IR1) in time window 1 is defined as std1(IR1) of the first red light in time window 1.

[0187] The mean of each second red light value (IR2) in time window 1 is defined as mean1(IR2) of the second red light in time window 1. The maximum value of each second red light value (IR2) in time window 1 is defined as max1(IR2) of the second red light in time window 1. The minimum value of each second red light value (IR2) in time window 1 is defined as min1(IR2) of the second red light in time window 1. The standard deviation of each second red light value (IR2) in time window 1 is defined as std1(IR2) of the second red light in time window 1.

[0188] The mean of each first blue light value (BL1) in time window 1 is defined as the first blue light mean value mean1(BL1) of time window 1. The maximum value of each first blue light value (BL1) in time window 1 is defined as the first blue light maximum value max1(BL1) of time window 1. The minimum value of each first blue light value (BL1) in time window 1 is defined as the first blue light minimum value min1(BL1) of time window 1. The standard deviation of each first blue light value (BL1) in time window 1 is defined as the first blue light standard deviation std1(BL1) of time window 1.

[0189] The mean of each second blue light value (BL2) in time window 1 is defined as mean1(BL2) of the second blue light in time window 1. The maximum value of each second blue light value (BL2) in time window 1 is defined as max1(BL2) of the second blue light in time window 1. The minimum value of each second blue light value (BL2) in time window 1 is defined as min1(BL2) of the second blue light in time window 1. The standard deviation of each second blue light value (BL2) in time window 1 is defined as std1(BL2) of the second blue light in time window 1.

[0190] (2) The rate of change of the processed data ( V)

[0191] The rate of change of the processed data is the difference between the mean of the processed data in the second time window and the mean of the processed data in the first time window, divided by 2 times the preset time step.

[0192] Taking the time window sequence {time window 1, time window 2, time window 3} as an example, two adjacent time windows are time window 1 and time window 2, with time window 2 being the next time window and time window 1 being the previous time window; and time window 2 and time window 3, with time window 3 being the next time window and time window 2 being the previous time window.

[0193] The rate of change for time window 1 and time window 2 is the difference between the mean of the processed data in time window 2 and the mean of time window 1, multiplied by 2 by the preset time step.

[0194] The rate of change for time window 2 and time window 3 is the difference between the mean of the processed data in time window 3 and the mean of time window 2, multiplied by 2, which is the preset time step.

[0195] This yields the rate of change of the processed data, including the rates of change within time window 1 and time window 2, and the rates of change within time window 2 and time window 3. In other words, the final rate of change of the processed data ( V) is a sequence formed by multiple rates of change (this sequence may have only one element or multiple elements, the number of elements is related to the number of time windows in the time window sequence), and is not a single value.

[0196] Taking a time window sequence of {time window 1, time window 2, time window 3}, with a preset time step of 15 seconds, and the processed data including humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) as an example, the rate of change of humidity values ​​in time window 1 and time window 2 ( V1(H) is the difference between the mean humidity value of time window 2 (mean2(H)) and the mean humidity value of time window 1 (mean1(H)) divided by 2 times the preset time step, i.e. V1(H) = (mean2(H) - mean1(H)) / 30 seconds. The rate of change of humidity values ​​in time window 2 and time window 3 ( V2(H) is the difference between the mean humidity value of time window 3 (mean3(H)) and the mean humidity value of time window 2 (mean2(H)) divided by 2 times the preset time step, i.e. V2(H) = (mean3(H) - mean2(H)) / 30 seconds. This gives the rate of change of the humidity value (…). V(H)) is { V1(H), V2(H)}.

[0197] The rate of temperature change in time window 1 and time window 2 ( V1(T) is the difference between the mean temperature of time window 2 (mean2(T)) and the mean temperature of time window 1 (mean1(T)) divided by 2 times the preset time step, i.e. V1(T) = (mean2(T) - mean1(T)) / 30 seconds. The rate of temperature change in time window 2 and time window 3 ( V2(T) is the difference between the mean temperature of time window 3 (mean3(T)) and the mean temperature of time window 2 (mean2(T)) divided by 2 times the preset time step, i.e. V2(T) = (mean3(T) - mean2(T)) / 30 seconds. This gives the rate of change of temperature (…). V(T)) is { V1(T), V2(T)}.

[0198] The rate of change of the first red light value in time window 1 and time window 2 ( V1(IR1)) is the difference between the first red light mean value (mean2(IR1)) of time window 2 and the first red light mean value (mean1(IR1)) of time window 1, multiplied by 2 by the preset time step. V1(IR1) = (mean2(IR1) - mean1(IR1)) / 30 seconds. The rate of change of the first red light value in time window 2 and time window 3 ( V2(IR1)) is the difference between the first red light mean value (mean3(IR1)) of time window 3 and the first red light mean value (mean2(IR1)) of time window 2, multiplied by 2 by the preset time step. V2(IR1) = (mean3(IR1) - mean2(IR1)) / 30 seconds. This gives the rate of change of the first red light value ( V(IR1)) is { V1(IR1), V2(IR1)}.

[0199] The rate of change of the second red light value in time window 1 and time window 2 ( V1(IR2)) is the difference between the second red light mean value (mean2(IR2)) of time window 2 and the second red light mean value (mean1(IR2)) of time window 1, multiplied by 2 by the preset time step. V1(IR2) = (mean2(IR2) - mean1(IR2)) / 30 seconds. The rate of change of the second red light value in time window 2 and time window 3 ( V2(IR2)) is the difference between the second red light mean value (mean3(IR2)) of time window 3 and the second red light mean value (mean2(IR2)) of time window 2, multiplied by 2 by the preset time step. V2(IR2) = (mean3(IR2) - mean2(IR2)) / 30 seconds. Thus, the rate of change of the second red light value is ( V(IR2)) is { V1(IR2), V2(IR2)}.

[0200] The rate of change of the first blue light value in time window 1 and time window 2 ( V1(BL1)) is the difference between the first blue light mean (mean2(BL1)) of time window 2 and the first blue light mean (mean1(BL1)) of time window 1, multiplied by 2 by the preset time step. V1(BL1) = (mean2(BL1) - mean1(BL1)) / 30 seconds. The rate of change of the first blue light value in time window 2 and time window 3 ( V2(BL1)) is the difference between the first blue light mean (mean3(BL1)) of time window 3 and the first blue light mean (mean2(BL1)) of time window 2, multiplied by 2 by the preset time step. V2(BL1) = (mean3(BL1) - mean2(BL1)) / 30 seconds. This gives the rate of change of the first blue light value (…). V(BL1)) is { V1(BL1), V2(BL1)}.

[0201] The rate of change of the second blue light value in time window 1 and time window 2 ( V1(BL2)) is the difference between the second blue light mean (mean2(BL2)) of time window 2 and the second blue light mean (mean1(BL2)) of time window 1, multiplied by 2 by the preset time step. V1(BL2) = (mean2(BL2) - mean1(BL2)) / 30 seconds. The rate of change of the second blue light value in time window 2 and time window 3 ( V2(BL2)) is the difference between the second blue light mean (mean3(BL2)) of time window 3 and the second blue light mean (mean2(BL2)) of time window 2, multiplied by 2 by the preset time step. V2(BL2) = (mean3(BL2) - mean2(BL2)) / 30 seconds. This gives the rate of change of the second blue light value (…). V(BL2)) is { V1(BL2), V2(BL2)}.

[0202] The rate of change of the processed data ( V) reflects the rate of change of the processed data per second. The rate of change of the processed data (V) is used to measure this change. V) can distinguish sudden water vapor (such as (Higher V(H)) and slowly accumulating dust.

[0203] (3) The duration for which the processed data meets the preset rules ( time)

[0204] If the preset rule is that the saturation flag is 1, then the duration for which the processed data satisfies the preset rule is the duration between the first moment and the second moment.

[0205] The first moment is the acquisition moment when both the red and blue light flags are 1, starting from the current acquisition moment. The second moment is the acquisition moment when both the red and blue light flags are 1, starting from the first moment.

[0206] Taking the processed data sorted from most recent to oldest as {processed data 1, processed data 2, processed data 4, processed data 5, processed data 6, processed data 7, processed data 8, processed data 9, processed data 10} as an example, the saturation flag bits of each processed data are shown in Table 1:

[0207] Table 1

[0208]

[0209] Wherein, time 1 is the current time, time 2 is earlier than time 1, time 3 is earlier than time 2, time 4 is earlier than time 3, time 5 is earlier than time 4, time 6 is earlier than time 5, time 7 is earlier than time 6, time 8 is earlier than time 7, time 9 is earlier than time 8, and time 10 is earlier than time 9.

[0210] The first acquisition time (time 1) preceding the current acquisition time when both the red and blue flag bits are 1 is time 3, therefore the first time is time 3. The last consecutive acquisition time preceding the first time (time 3) when both the red and blue flag bits are 1 is time 4, therefore the second time is time 4. The duration for which the processed data meets the preset rules ( (time) is the duration between the first time point and the second time point, that is, the duration from time point 3 to time point 4 (inclusive of time point 3 and time point 4).

[0211] The duration for which the processed data meets the preset rules ( The time indicates the duration during which both the second red light value (IR2) and the second blue light value (BL2) are close to full-scale.

[0212] The duration for which the processed data meets the preset rules ( `time` is used to trigger specific long-term Boolean characteristics, such as the duration for which processed data meets preset rules. If the time is long (e.g., more than 10 seconds), a specific long-cycle Boolean feature is triggered.

[0213] In practice, the preset rules can be other rules as well. This embodiment does not limit the specific content of the preset rules. The preset rules can be determined according to the actual situation.

[0214] 3) State memory feature operator

[0215] The state memory feature operator is an operator that reflects the normal threshold.

[0216] The state memory feature operator can introduce a feedback mechanism into the smoke identification system based on multi-feature fusion and lightweight gradient booster provided in this embodiment, locking the physical quantity (i.e., the normal threshold) at the moment of entering the warning as a memory variable.

[0217] Such as state memory feature operators, including: the normal threshold of the processed data and the deviation value of the processed data.

[0218] (1) Normal threshold of processed data

[0219] The normal threshold for the processed data is determined based on the processed data at all acquisition times.

[0220] The normal threshold of the processed data can lock the initial alarm value. The normal threshold of the processed data reflects the threshold at which the state transitions from normal to warning.

[0221] The normal threshold for processed data can be determined through the following steps:

[0222] 301. Obtain the processed data within a preset number of time windows preceding the current acquisition time. Then, define the preset number of time windows preceding the current acquisition time as candidate windows, and select all acquired processed data as candidate data.

[0223] The time window here is the same as the time window in "2) Time Window Feature Operator", so it will not be explained again here. Please refer to "2) Time Window Feature Operator" for details.

[0224] For example, if the time window sequence is {time window 1, time window 2, time window 3}, time window 1 includes processed data 1 collected at time 1 and processed data 2 collected at time 2; time window 2 includes processed data 3 collected at time 3 and processed data 4 collected at time 4; time window 3 includes processed data 5 collected at time 5 and processed data 6 collected at time 6. If the preset quantity is 2, then in step 301, the processed data (such as processed data 1, processed data 2, processed data 3, and processed data 4) obtained in the two time windows preceding the current acquisition time (such as time window 1 and time window 2) will be acquired. The two time windows preceding the current acquisition time will be determined as candidate windows (i.e., candidate windows are time window 1 and time window 2), and all acquired processed data will be determined as candidate data (i.e., candidate data are processed data 1, processed data 2, processed data 3, and processed data 4).

[0225] 302. Determine the mean and standard deviation of all candidate data.

[0226] Taking the processed data including humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2), and the candidate data being processed data 1, processed data 2, processed data 3, and processed data 4 as an example, in step 302, the mean (denoted as mean'(H)) and standard deviation (denoted as std'(H)) of the humidity values ​​in processed data 1, processed data 2, processed data 3, and processed data 4 are determined. The mean (denoted as mean'(T)) and standard deviation (denoted as std'(T)) of the temperature values ​​in processed data 1, processed data 2, processed data 3, and processed data 4 are also determined. Determine the mean (denoted as mean'(IR1)) and standard deviation (denoted as std'(IR1)) of the first red light value in processed data 1, processed data 2, processed data 3, and processed data 4. Determine the mean (denoted as mean'(IR2)) and standard deviation (denoted as std'(IR2)) of the second red light value in processed data 1, processed data 2, processed data 3, and processed data 4. Determine the mean (denoted as mean'(BL1)) and standard deviation (denoted as std'(BL1)) of the first blue light value in processed data 1, processed data 2, processed data 3, and processed data 4. Determine the mean (e.g., mean'(BL2)) and standard deviation (e.g., std'(BL2)) of the second blue light values ​​in processed data 1, processed data 2, processed data 3, and processed data 4.

[0227] The mean and standard deviation of all candidate data reflect the actual situation of the candidate data in the recent period and can be used as a comparison standard.

[0228] 303, Determine the baseline value for each time window.

[0229] Wherein, if the mean of the processed data in any time window is less than the mean of all candidate data, then the benchmark value of any time window is (4 times the mean of the processed data in any time window + the maximum value of the processed data in any time window + the minimum value of the processed data in any time window) / 6 + the maximum value between the standard deviation of the processed data in any time window and the standard deviation of all candidate data.

[0230] If the mean of the processed data in any time window is not less than the mean of all candidate data, then the benchmark value of any time window is (4 times the mean of all candidate data + the maximum value of the processed data in any time window + the minimum value of the processed data in any time window) / 6 + the minimum value between the standard deviation of the processed data in any time window and the standard deviation of all candidate data.

[0231] (4 times the mean of all candidate data + the maximum value of processed data in any time window + the minimum value of processed data in any time window) / 6 represents the average value of processed data in any time window, with a slight upward fluctuation as the baseline value for that time window. Values ​​above the baseline value may indicate anomalies.

[0232] When the data is fluctuating, if the mean of the processed data in any time window is less than the mean of all candidate data, it means that the processed data in that time window is lower than the most recent data. The lower the processed data, the more normal it is, and the lower the probability of a fire. Therefore, when determining the baseline value for any time window, it can be fluctuated slightly more without affecting the identification of smoke. The standard deviation of the processed data in any time window reflects the fluctuation between the processed data in any time window, and the standard deviation of all candidate data reflects the recent fluctuation. A larger fluctuation can be selected, that is, the maximum value between the standard deviation of the processed data in any time window and the standard deviation of all candidate data is added. If the mean of the processed data in any time window is not less than the mean of all candidate data, it means that the processed data in that time window is higher or equal to the recent data, indicating that it is similar to or more abnormal than the current situation, and the probability of a fire is the same or increased. Therefore, the baseline value for any time window can be slightly fluctuated to avoid affecting the accuracy of subsequent smoke identification. The standard deviation of the processed data in any time window reflects the fluctuation between the processed data in any time window, and the standard deviation of all candidate data reflects the recent fluctuation. The smaller fluctuation can be selected, that is, the minimum value between the standard deviation of the processed data in any time window and the standard deviation of all candidate data.

[0233] The mean, maximum, and minimum values ​​of the processed data for any time window are determined in “2) Time Window Feature Operator”, and the mean of all candidate data is determined in step 302.

[0234] If the time window sequence is {time window 1, time window 2, time window 3}, then in step 303, the reference value of time window 1, the reference value of time window 2, and the reference value of time window 3 will be determined.

[0235] Taking any time window (such as time window 1) as an example, the process of determining its baseline value is as follows:

[0236] If the processed data includes humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2), then:

[0237] If the mean humidity of time window 1 is mean1(H) < the mean humidity of all candidate data is mean'(H), then the baseline humidity value (H_BV1) of time window 1 is (4 times the mean humidity of time window 1 mean1(H) + the maximum humidity of time window 1 max1(H) + the minimum humidity of time window 1 min1(H)) / 6 + max{the standard deviation of humidity of time window 1 std1(H), the standard deviation of humidity of all candidate data std'(H)}, that is, H_BV1 = (4 × mean1(H) + max1(H) + min1(H)) / 6 + max{std1(H), std'(H)}. If the mean humidity of time window 1 is mean1(H) ≥ the mean humidity of all candidate data is mean'(H), then the baseline humidity value (H_BV1) of time window 1 is (4 times the mean humidity of time window 1 mean1(H) + the maximum humidity of time window 1 max1(H) + the minimum humidity of time window 1 min1(H)) / 6 + min{the standard deviation of humidity of time window 1 std1(H), the standard deviation of humidity of all candidate data std'(H)}, that is, H_BV1 = (4 × mean1(H) + max1(H) + min1(H)) / 6 + min{std1(H), std'(H)}.

[0238] Here, max{} is the function to find the maximum value, and min{} is the function to find the minimum value.

[0239] If the mean temperature of time window 1, mean1(T), is less than the mean temperature of all candidate data, mean'(T), then the baseline temperature value (T_BV1) of time window 1 is (4 times the mean temperature of time window 1, mean1(T) + the maximum temperature of time window 1, max1(T) + the minimum temperature of time window 1, min1(T)) / 6 + max{the standard deviation of temperature of time window 1, std1(T), the standard deviation of temperature values ​​among all candidate data, std'(T)}, that is, T_BV1 = (4 × mean1(T) + max1(T) + min1(T)) / 6 + max{std1(T), std'(T)}. If the mean temperature of time window 1, mean1(T), is greater than or equal to the mean temperature of all candidate data, mean'(T), then the baseline temperature value (T_BV1) of time window 1 is (4 times the mean temperature of time window 1, mean1(T) + the maximum temperature of time window 1, max1(T) + the minimum temperature of time window 1, min1(T)) / 6 + min{the standard deviation of temperature of time window 1, std1(T), the standard deviation of temperature values ​​among all candidate data, std'(T)}, that is, T_BV1 = (4 × mean1(T) + max1(T) + min1(T)) / 6 + min{std1(T), std'(T)}.

[0240] If the mean value of the first red light in time window 1 is mean1(IR1) < the mean value of the first red light in all candidate data is mean'(IR1), then the baseline value of the first red light in time window 1 (IR1_BV1) = (4 times the mean value of the first red light in time window 1 mean1(IR1) + the maximum value of the first red light in time window 1 max1(IR1) + the minimum value of the first red light in time window 1 min1(IR1)) / 6 + max{the standard deviation of the first red light in time window 1 std1(IR1), the standard deviation of the first red light in all candidate data std'(IR1)}, that is, IR1_BV1 = (4 × mean1(IR1) + max1(IR1) + min1(IR1)) / 6 + max{std1(IR1), std'(IR1)}. If the mean value of the first red light in time window 1, mean1(IR1), is greater than or equal to the mean value of the first red light in all candidate data, mean'(IR1), then the baseline value of the first red light in time window 1 (IR1_BV1) = (4 times the mean value of the first red light in time window 1, mean1(IR1) + the maximum value of the first red light in time window 1, max1(IR1) + the minimum value of the first red light in time window 1, min1(IR1)) / 6 + min{the standard deviation of the first red light in time window 1, std1(IR1), the standard deviation of the first red light in all candidate data, std'(IR1)}, that is, IR1_BV1 = (4 × mean1(IR1_BV) + max1(IR1_BV) + min1(IR1_BV)) / 6 + min{std1(IR1_BV), std'(IR1_BV)}.

[0241] If the mean value of the second red light in time window 1, mean1(IR2), is less than the mean value of the second red light in all candidate data, mean'(IR2), then the baseline value of the second red light in time window 1 (IR2_BV1) = (4 times the mean value of the second red light in time window 1, mean1(IR2) + the maximum value of the second red light in time window 1, max1(IR2) + the minimum value of the second red light in time window 1, min1(IR2)) / 6 + max{the standard deviation of the second red light in time window 1, std1(IR2), the standard deviation of the second red light in all candidate data, std'(IR2)}, that is, IR2_BV1 = (4 × mean1(IR2) + max1(IR2) + min1(IR2)) / 6 + max{std1(IR2), std'(IR2)}. If the mean of the second red light in time window 1, mean1(IR2), is greater than or equal to the mean of the second red light values ​​of all candidate data, mean'(IR2), then the baseline value of the second red light in time window 1 (IR2_BV1) = (4 times the mean of the second red light in time window 1, mean1(IR2) + the maximum value of the second red light in time window 1, max1(IR2) + the minimum value of the second red light in time window 1, min1(IR2)) / 6 + min{the standard deviation of the second red light in time window 1, std1(IR2), and the standard deviation of the second red light values ​​among all candidate data, std'(IR2)}, that is, IR2_BV1 = (4 × mean1(IR2) + max1(IR2) + min1(IR2)) / 6 + min{std1(IR2), std'(IR2)}.

[0242] If the first blue light mean value of time window 1, mean1(BL1), is less than the mean value of the first blue light value of all candidate data, mean'(BL1), then the first blue light baseline value (BL1_BV1) of time window 1 is (4 times the first blue light mean value of time window 1, mean1(BL1) + the first blue light maximum value of time window 1, max1(BL1) + the first blue light minimum value of time window 1, min1(BL1)) / 6 + max{the first blue light standard deviation of time window 1, std1(BL1), the standard deviation of the first blue light value of all candidate data, std'(BL1)}, that is, BL1_BV1 = (4 × mean1(BL1) + max1(BL1) + min1(BL1)) / 6 + max{std1(BL1), std'(BL1)}. If the mean value of the first blue light in time window 1 is mean1(BL1) ≥ the mean value of the first blue light in all candidate data is mean'(BL1), then the baseline value of the first blue light in time window 1 (BL1_BV1) = (4 times the mean value of the first blue light in time window 1 mean1(BL1) + the maximum value of the first blue light in time window 1 max1(BL1) + the minimum value of the first blue light in time window 1 min1(BL1)) / 6 + min{the standard deviation of the first blue light in time window 1 std1(BL1), the standard deviation of the first blue light in all candidate data std'(BL1)}, that is, BL1_BV1 = (4×mean1(BL1)+max1(BL1)+min1(BL1)) / 6+min{std1(BL1), std'(BL1)}.

[0243] If the second blue light mean value of time window 1, mean1(BL2), is less than the second blue light mean value of all candidate data, mean'(BL2), then the second blue light baseline value of time window 1 (BL2_BV1) = (4 times the second blue light mean value of time window 1, mean1(BL2) + the second blue light maximum value of time window 1, max1(BL2) + the second blue light minimum value of time window 1, min1(BL2)) / 6 + max{the second blue light standard deviation of time window 1, std1(BL2), the standard deviation of the second blue light value among all candidate data, std'(BL2)}, that is, BL2_BV1 = (4 × mean1(BL2) + max1(BL2) + min1(BL2)) / 6 + max{std1(BL2), std'(BL2)}. If the second blue light mean value of time window 1, mean1(BL2), is greater than or equal to the second blue light mean value of all candidate data, mean'(BL2), then the second blue light baseline value (BL2_BV1) of time window 1 is (4 times the second blue light mean value of time window 1, mean1(BL2) + the second blue light maximum value of time window 1, max1(BL2) + the second blue light minimum value of time window 1, min1(BL2)) / 6 + min{the second blue light standard deviation of time window 1, std1(BL2), the standard deviation of the second blue light value among all candidate data, std'(BL2)}, that is, BL2_BV1 = (4 × mean1(BL2) + max1(BL2) + min1(BL2)) / 6 + min{std1(BL2), std'(BL2)}.

[0244] 304. Determine the mean and standard deviation of the baseline values ​​for each time window.

[0245] Taking a time window sequence of {time window 1, time window 2, time window 3}, and the processed data including humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) as an example, in step 304, the mean (denoted as mean(H)) and standard deviation (denoted as std(H)) of the humidity baseline value (H_BV1) of time window 1, the humidity baseline value (H_BV2) of time window 2, and the humidity baseline value (H_BV3) of time window 3 are determined. The mean (denoted as mean(T)) and standard deviation (denoted as std(T)) of the temperature baseline value (T_BV1) of time window 1, the temperature baseline value (T_BV2) of time window 2, and the temperature baseline value (T_BV3) of time window 3 are also determined. Determine the mean (denoted as mean"(IR1)) and standard deviation (denoted as std"(IR1)) of the first red light reference value (IR1_BV1) of time window 1, the first red light reference value (IR1_BV2) of time window 2, and the first red light reference value (IR1_BV3) of time window 3. Determine the mean (denoted as mean"(IR2)) and standard deviation (denoted as std"(IR2) of the second red light reference value (IR2_BV1) of time window 1, the second red light reference value (IR2_BV2) of time window 2, and the second red light reference value (IR2_BV3) of time window 3. Determine the mean (denoted as mean"(BL1)) and standard deviation (denoted as std"(BL1) of the first blue light reference value (BL1_BV1) of time window 1, the first blue light reference value (BL1_BV2) of time window 2, and the first blue light reference value (BL1_BV3) of time window 3. Determine the mean (denoted as mean" (BL2) and standard deviation (denoted as std" (BL2) of the second blue light reference value (BL2_BV1) of time window 1, the second blue light reference value (BL2_BV2) of time window 2, and the second blue light reference value (BL2_BV3) of time window 3.

[0246] The mean of the baseline values ​​for each time window reflects the average situation of the baseline values ​​for each time window, while the standard deviation of the baseline values ​​for each time window reflects the fluctuation of the baseline values ​​for each time window.

[0247] It should be noted that, regardless of whose standard deviation it is, the larger the standard deviation, the greater the volatility. Taking the standard deviation of the benchmark value for each time window as an example, the larger the standard deviation, the greater the volatility of the benchmark value for each time window, and the smaller the standard deviation, the smaller the volatility of the benchmark value for each time window.

[0248] 305. The normal threshold for the processed data is determined to be the mean of the baseline values ​​of each time window + the standard deviation of the processed data of each time window, and the minimum value between the standard deviation of all candidate data and the standard deviation of the baseline values ​​of each time window.

[0249] The mean and standard deviation of the baseline values ​​for each time window are determined in step 304, the standard deviation of the processed data for each time window is determined in “2) Time Window Feature Operator”, and the standard deviation of all candidate data is determined in step 302.

[0250] Taking the time window sequence as {time window 1, time window 2, time window 3}, and the processed data including humidity value (H), temperature value (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) as an example, in step 305, the normal threshold (H_AlarmValue) of humidity value (H) will be determined as the mean (mean) "H" + min{humidity standard deviation std1(H) of time window 1, humidity standard deviation std2(H) of time window 2, humidity standard deviation std3(H) of time window 3, humidity standard deviation std'(H) of all candidate data, and standard deviation of humidity standard value std""H"}, that is, H_AlarmValue=mean""H"+min{std1(H), std2(H), std3(H), std'(H), std""H"}.

[0251] The normal threshold (T_AlarmValue) for temperature value (T) is determined as the mean (T) + min{temperature standard deviation std1(T) for time window 1, temperature standard deviation std2(T) for time window 2, temperature standard deviation std3(T) for time window 3, temperature standard deviation std'(T) for all candidate data, and standard deviation std"(T) for temperature standard deviation of each time window}, i.e., T_AlarmValue = mean"(T) + min{std1(T), std2(T), std3(T), std'(T), std"(T)}.

[0252] The normal threshold (IR1_AlarmValue) of the first red light value (IR1) is determined as the mean (IR1) + min{the standard deviation of the first red light in time window 1 std1(IR1), the standard deviation of the first red light in time window 2 std2(IR1), the standard deviation of the first red light in time window 3 std3(IR1), the standard deviation of the first red light value in all candidate data std'(IR1), and the standard deviation of the first red light reference value in each time window std"(IR1)}, that is, IR1_AlarmValue = mean"(IR1) + min{std1(IR1), std2(IR1), std3(IR1), std'(IR1), std"(IR1)}.

[0253] The normal threshold (IR2_AlarmValue) of the second red light value (IR2) is determined as the mean (IR2) + min{the standard deviation of the second red light in time window 1, std1(IR2), the standard deviation of the second red light in time window 2, the standard deviation of the second red light in time window 3, std3(IR2), the standard deviation of the second red light value in all candidate data, std'(IR2), and the standard deviation of the second red light reference value in each time window, std(IR2)}, that is, IR2_AlarmValue = mean (IR2) + min{std1(IR2), std2(IR2), std3(IR2), std'(IR2), std(IR2)}.

[0254] The normal threshold (BL1_AlarmValue) of the first blue light value (BL1) is determined as the mean (BL1) + min{the first blue light standard deviation std1(BL1) of time window 1, the first blue light standard deviation std2(BL1) of time window 2, the first blue light standard deviation std3(BL1) of time window 3, the standard deviation of the first blue light value in all candidate data std'(BL1), and the standard deviation of the first blue light standard value of each time window std""BL1)}, that is, BL1_AlarmValue=mean""BL1)+min{std1(BL1), std2(BL1), std3(BL1), std'(BL1), std""BL1)}.

[0255] The normal threshold (BL2_AlarmValue) for the second blue light value (BL2) is determined as the mean (BL2) + min{the standard deviation of the second blue light value (BL2_BV) for each time window: std1(BL2), std2(BL2), std3(BL2), std3(BL2), std'(BL2), std'(BL2), std'(BL2), std'(BL2), std'(BL2), std'(BL2), std'(BL2)}. That is, BL2_AlarmValue = mean (BL2) + min{std1(BL2), std2(BL2), std3(BL2), std'(BL2), std'(BL2)}.

[0256] The mean of the baseline values ​​for each time window represents the average value of the baseline values ​​for each time window. Based on this, the mean value will be adjusted upwards to serve as the normal threshold for the processed data.

[0257] The fluctuation range is determined by the minimum of the standard deviations of the processed data, the standard deviations of all candidate data, and the standard deviations of the baseline values ​​for each time window. In other words, it is determined by the minimum fluctuation of the processed data, the fluctuations of all candidate data, and the fluctuations of the baseline values ​​for each time window.

[0258] The normal threshold for processed data is determined based on the processed data over a recent period and the sum of processed data for each time window. It can accurately reflect the threshold at which processed data transitions from a normal state to a warning state.

[0259] (2) Deviation value of the processed data

[0260] The deviation value of the processed data is the difference between the processed data and the normal threshold at each acquisition time.

[0261] The deviation value of the processed data reflects the deviation correlation characteristics. This value can effectively characterize the trend of smoke concentration falling or continuing to rise after the warning is triggered, thereby distinguishing between temporary smoking interference and continuous real fire.

[0262] The processed data at each acquisition time is obtained by the data preprocessing module, and the normal threshold of the processed data is obtained in step 305.

[0263] Taking all collected processed data, including processed data 1, processed data 2, processed data 3, and processed data 4, and including humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2), as an example, the deviation value of humidity is {humidity value in processed data 1 - normal threshold value (H_AlarmValue), humidity value in processed data 2 - normal threshold value (H_AlarmValue), humidity value in processed data 3 - normal threshold value (H_AlarmValue), humidity value in processed data 4 - normal threshold value (H_AlarmValue)}. The deviation value of temperature is {temperature value in processed data 1 - normal threshold value (T_AlarmValue), temperature value in processed data 2 - normal threshold value (T_AlarmValue), temperature value in processed data 3 - normal threshold value (T_AlarmValue), temperature value in processed data 4 - normal threshold value (T_AlarmValue)}. The deviation value of the first red light value is {the first red light value in processed data 1 minus the normal threshold value of the first red light value (IR1_AlarmValue), the first red light value in processed data 2 minus the normal threshold value of the first red light value (IR1_AlarmValue), the first red light value in processed data 3 minus the normal threshold value of the first red light value (IR1_AlarmValue), and the first red light value in processed data 4 minus the normal threshold value of the first red light value (IR1_AlarmValue)}. The deviation value of the second red light value is {the second red light value in processed data 1 minus the normal threshold value of the second red light value (IR2_AlarmValue), the second red light value in processed data 2 minus the normal threshold value of the second red light value (IR2_AlarmValue), the second red light value in processed data 3 minus the normal threshold value of the second red light value (IR2_AlarmValue), and the second red light value in processed data 4 minus the normal threshold value of the second red light value (IR2_AlarmValue)}. The deviation value of the first blue light value is {first blue light value in processed data 1 - normal threshold value of the first blue light value (BL1_AlarmValue), first blue light value in processed data 2 - normal threshold value of the first blue light value (BL1_AlarmValue), first blue light value in processed data 3 - normal threshold value of the first blue light value (BL1_AlarmValue), first blue light value in processed data 4 - normal threshold value of the first blue light value (BL1_AlarmValue)}.The deviation value of the second blue light value is {the second blue light value in processed data 1 minus the normal threshold value of the second blue light value (BL2_AlarmValue), the second blue light value in processed data 2 minus the normal threshold value of the second blue light value (BL2_AlarmValue), the second blue light value in processed data 3 minus the normal threshold value of the second blue light value (BL2_AlarmValue), and the second blue light value in processed data 4 minus the normal threshold value of the second blue light value (BL2_AlarmValue)}.

[0264] The resulting deviation value of the processed data includes the deviation values ​​of each processed data point. In other words, the deviation value of the processed data is a sequence formed by the deviation values ​​of the processed data obtained at multiple acquisition times (this sequence may have only one element or multiple elements, the number of elements being related to the number of acquisition cycles that have passed), and is not a single value.

[0265] 4) Pattern change feature operator

[0266] The pattern change feature operator is an operator that reflects jump changes.

[0267] The pattern change feature operator reflects the abrupt changes in the processed data, which can help determine whether there is an instantaneous switch from "normal state (0)" to "warning state (1)" or "high-risk state (2)". The pattern change feature operator helps to identify interference sources with obvious abrupt changes, such as "burnt pot".

[0268] Among them, the pattern change feature operator includes: the jump value of the processed data.

[0269] The jump value of the processed data is the maximum value within the target processing time.

[0270] All processed data within the target processing time are greater than the normal threshold of processed data, and there is at least one target data in the processed data within the target processing time.

[0271] Within the target processing time, the processed data before the target data shows an upward trend or a constant trend over time, while the processed data after the target data shows a downward trend or a constant trend over time.

[0272] If the processing time is T2, then the processed data can be arranged from the nearest to the oldest collection time to obtain the processed data sequence. Starting from the first element of the processed data sequence, select an element sequentially (e.g., select element i, whose collection time is time i); determine all processed data (including time i and time i-T2) between time i and time i-T2; if all processed data between time i and time i-T2 are greater than the normal threshold of processed data (this value is determined in step 305), that is, all processed data between time i and time i-T2 are greater than the normal threshold of processed data, and are in an abnormal state, and all processed data between time i and time i-T2 are greater than the target data in processed data, then the duration between time i and time i-T2 is the target processing duration (including time i and time i-T2), and the maximum value among all processed data between time i and time i-T2 (i.e. within the target processing duration) is the jump point, that is, processed data that will switch instantaneously from "normal state (0)" to "warning state (1)" or "high risk state (2)", so it is the jump value of processed data.

[0273] For the target data (e.g., processed data j in all processed data between time i and time i-T2, where the acquisition time corresponding to processed data j is time j) in all processed data prior to processed data j (i.e., acquired at a time earlier than time j), all processed data prior to processed data j (i.e., all processed data acquired at a time earlier than time j in all processed data between time i and time i-T2, including time j) exhibit an increasing trend or a constant trend over time (i.e., all processed data acquired at a time earlier than time j in all processed data between time i and time i-T2 are sorted by acquisition time to obtain the first sequence; if the processed data value of the next element in the first sequence is not less than the processed data value of the previous element), then it indicates that the processed data corresponding to the acquisition time before processed data j between time i and time i-T2 is not less than processed data j; and, processed data j prior to processed data j (i.e., acquired at a time later than time j) exhibits an increasing trend or a constant trend over time. The processed data of time j (i.e., all processed data collected at a time later than time j among all processed data between time i and time i-T2, including time j) shows a decreasing trend or a constant trend over time (i.e., all processed data collected at a time later than time j among all processed data between time i and time i-T2 are sorted by collection time to obtain a second sequence. If the processed data value of the next element in the second sequence is not greater than the processed data value of the previous element), it means that the processed data corresponding to the collection time after processed data j between time i and time i-T2 is not less than processed data j; it means that processed data j is the most normal among all abnormal data between time i and time i-T2, and is also the turning point from the most normal to abnormal. Then this point is the jump point, which is the processed data that will switch instantaneously from "normal state (0)" to "warning state (1)" or "high risk state (2)", and is the jump value of processed data.

[0274] Taking the processed data including humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2) as an example, the jump values ​​of humidity, temperature, first red light, second red light, first blue light, and second blue light can be determined.

[0275] Among them, the jump value of the humidity value is the maximum value of the humidity value within the target processing time; the humidity values ​​within the target processing time are all greater than the normal threshold of the humidity value (H_AlarmValue), and there is at least one target data of humidity value within the target processing time; within the target processing time, the humidity value of the target data before the target data shows an upward trend or a constant trend over time, and the humidity value of the target data after the target data shows a downward trend or a constant trend over time.

[0276] The jump value of the temperature value is the maximum temperature value within the target processing time; all temperature values ​​within the target processing time are greater than the normal threshold of the temperature value (T_AlarmValue), and there is at least one target data for the temperature value within the target processing time; within the target processing time, the temperature value of the target data before the target data shows an upward trend or a constant trend over time, and the temperature value of the target data after the target data shows a downward trend or a constant trend over time.

[0277] The jump value of the first red light value is the maximum value of the first red light value within the target processing time; all first red light values ​​within the target processing time are greater than the normal threshold (IR1_AlarmValue) of the first red light value, and there is at least one target data in the first red light value within the target processing time; within the target processing time, the first red light value of the target data forward shows an upward trend or a constant trend over time, and the first red light value of the target data backward shows a downward trend or a constant trend over time.

[0278] The jump value of the second red light value is the maximum value of the second red light value within the target processing time; all second red light values ​​within the target processing time are greater than the normal threshold (IR2_AlarmValue) of the second red light value, and there is at least one target data in the second red light value within the target processing time; within the target processing time, the second red light value of the target data before the target data shows an upward trend or a constant trend over time, and the second red light value of the target data after the target data shows a downward trend or a constant trend over time.

[0279] The jump value of the first blue light value is the maximum value of the first blue light value within the target processing time; all first blue light values ​​within the target processing time are greater than the normal threshold (BL1_AlarmValue) of the first blue light value, and there is at least one target data for the first blue light value within the target processing time; within the target processing time, the first blue light value of the target data before the target data shows an upward trend or a constant trend over time, and the first blue light value of the target data after the target data shows a downward trend or a constant trend over time.

[0280] The jump value of the second blue light value is the maximum value of the second blue light value within the target processing time; all second blue light values ​​within the target processing time are greater than the normal threshold (BL2_AlarmValue) of the second blue light value, and there is at least one target data for the second blue light value within the target processing time; within the target processing time, the second blue light value before the target data shows an upward trend or a constant trend over time, and the second blue light value after the target data shows a downward trend or a constant trend over time.

[0281] The target processing time is a special processing time, which is the processing time during which all processed data are greater than the normal threshold of processed data, and at least one target data exists in the processed data.

[0282] Processing time is determined through the following steps:

[0283] 401, obtain the first duration during which the processed data in each time window is continuously greater than the standard value of each time window.

[0284] The standard value for any time window is the sum of the mean of the processed data for that time window and the standard deviation of the processed data for that time window.

[0285] The time window here is the same as the time window in "2) Time Window Feature Operator", so it will not be explained again here. Please refer to "2) Time Window Feature Operator" for details.

[0286] The mean and standard deviation of the processed data for any time window are determined in “2) Time Window Feature Operator”.

[0287] If the time window sequence is {time window 4, time window 5}, then the standard value of time window 4 is the mean of the processed data of time window 4 plus the standard deviation of the processed data of time window 4, and the standard value of time window 5 is the mean of the processed data of time window 5 plus the standard deviation of the processed data of time window 5.

[0288] Determine whether each processed data included in time window 4 is greater than the standard value of time window 4. For example, time window 4 includes processed data 41, processed data 42, processed data 43, processed data 44, processed data 45, and processed data 46. Processed data 41 is greater than the standard value of time window 4, processed data 42 is greater than the standard value of time window 4, processed data 43 is not greater than the standard value of time window 4, processed data 44 is greater than the standard value of time window 4, processed data 45 is greater than the standard value of time window 4, and processed data 46 is greater than the standard value of time window 4.

[0289] Determine whether each processed data included in time window 5 is greater than the standard value of time window 5.

[0290] Within time window 4, the maximum duration covered by processed data that is consecutively greater than the standard value of time window 4 is determined. This maximum duration is the first duration of time window 4. For example, if time window 4 contains two consecutive processed data sets that are consecutively greater than the standard value of time window 4, one set consists of processed data 41 and processed data 42, whose duration is the time between the acquisition time of processed data 42 and the acquisition time of processed data 41 (inclusive); the other set consists of processed data 44, processed data 45, and processed data 46, whose duration is the time between the acquisition time of processed data 46 and the acquisition time of processed data 44 (inclusive). The longest duration between the acquisition times of processed data 46 and processed data 44 (inclusive) is determined as the first duration of time window 4.

[0291] Within time window 5, the maximum duration covered by processed data that is continuously greater than the standard value of time window 5 is determined. This maximum duration is the first duration of time window 5.

[0292] Taking any given time window as an example, if the processed data includes humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2), then in step 401, the first duration for which the humidity value within any given time window is continuously greater than the standard humidity value of any given time window will be obtained. The standard humidity value for any given time window is the mean humidity (H) of any given time window plus the standard deviation of humidity (std(H)) of any given time window.

[0293] Obtain the first duration for which the temperature value within any given time window is consistently greater than the standard temperature value for that time window. The standard temperature value for any given time window is the sum of the mean temperature (mean(T)) and the standard deviation temperature (std(T)) for that time window.

[0294] Obtain the first duration for which the first red light value is continuously greater than the standard value of the first red light value in any time window. The standard value of the first red light value in any time window is the sum of the mean (IR1) and the standard deviation (std) of the first red light value in any time window.

[0295] Obtain the first duration for which the second red light value is continuously greater than the standard value of the second red light value within any time window. The standard value of the second red light value within any time window is the sum of the mean (IR2) and the standard deviation (std(IR2)) of the second red light value within that time window.

[0296] Obtain the first duration for which the first blue light value within any time window is continuously greater than the standard value of the first blue light value within any time window. The standard value of the first blue light value within any time window is the sum of the mean (BL1) and the standard deviation (std) of the first blue light value within any time window.

[0297] Obtain the first duration for which the second blue light value is continuously greater than the standard value of the second blue light value within any time window. The standard value of the second blue light value within any time window is the sum of the mean (BL2) and the standard deviation (std) of the second blue light value within that time window.

[0298] 402, determine the minimum and standard deviation of all first durations.

[0299] If the processed data includes humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2), then in step 402, the minimum and standard deviation of the first duration for all humidity values, the minimum and standard deviation of the first duration for all temperature values, the minimum and standard deviation of the first duration for all first red light values, the minimum and standard deviation of the first duration for all second red light values, the minimum and standard deviation of the first duration for all first blue light values, and the minimum and standard deviation of the first duration for all second blue light values ​​will be determined.

[0300] 403. The processing time is determined to be the minimum of all first-time durations plus twice the standard deviation of all first-time durations. That is, the processing time is the minimum of all first-time durations plus 2 × the standard deviation of all first-time durations.

[0301] If the processed data includes humidity (H), temperature (T), first red light value (IR1), second red light value (IR2), first blue light value (BL1), and second blue light value (BL2), then in step 403, the processing time for humidity value will be determined as the minimum of the first processing time for all humidity values ​​+ 2 × the standard deviation of the first processing time for all humidity values; the processing time for temperature value will be determined as the minimum of the first processing time for all temperature values ​​+ 2 × the standard deviation of the first processing time for all temperature values; the processing time for first red light value will be determined as the minimum of the first processing time for all first red light values ​​+ 2 × the standard deviation of the first processing time for all first red light values; the processing time for second red light value will be determined as the minimum of the first processing time for all second red light values ​​+ 2 × the standard deviation of the first processing time for all second red light values; the processing time for first blue light value will be determined as the minimum of the first processing time for all first blue light values ​​+ 2 × the standard deviation of the first processing time for all first blue light values; and the processing time for second blue light value will be determined as the minimum of the first processing time for all second blue light values ​​+ 2 × the standard deviation of the first processing time for all second blue light values.

[0302] The first duration of any time window reflects the maximum duration of persistently high processed data within that time window. The minimum of all first durations is the minimum of the maximum duration of persistently high processed data across all time windows, and this is used as the base duration. The standard deviation of all first durations is the fluctuation of the maximum duration of persistently high processed data across all time windows. The processing duration is calculated by doubling the fluctuation of the base duration upwards. This ensures that the processing duration comprehensively considers the true state of the processed data and better reflects its actual characteristics, thus enabling precise determination of pattern change feature operators.

[0303] In practical implementation, the multidimensional feature construction module can encapsulate the extracted feature operators into a one-dimensional high-dimensional vector.

[0304] 4. Classification Decision Module

[0305] The classification decision module uses a Lightweight Gradient Boosting Machine (LightGBM, a distributed gradient boosting framework based on decision tree algorithm) as a classifier to classify the feature operators extracted by the multi-dimensional feature construction module, thus obtaining smoke identification results. It outputs warning reasons (including: smoking, dust, cooking fumes, burnt food, steam, or actual fire).

[0306] For example, the classification decision module uses Lightweight Gradient Boosting Machine (LightGBM) as the fitting model. Lightweight Gradient Boosting Machine (LightGBM) greatly reduces the computational overhead while ensuring high accuracy through histogram-based decision trees, making it suitable for real-time analysis of embedded or edge sensors.

[0307] For example, a one-dimensional high-dimensional vector, encapsulated from a multi-dimensional feature construction module, is input into a pre-trained Lightweight Gradient Boosting Machine (LightGBM) model. The Lightweight Gradient Boosting Machine (LightGBM) model calculates non-linear weights on the one-dimensional high-dimensional vector based on the branching logic of its internal decision tree. The Lightweight Gradient Boosting Machine (LightGBM) ultimately outputs the classification result with the highest probability.

[0308] If the smoke is determined to be a "real fire," the smoke identification system based on multi-feature fusion and a lightweight gradient booster provided in this embodiment can immediately trigger a high-level alarm. If it is determined to be interference such as "smoking" or "cooking fumes," the smoke identification system based on multi-feature fusion and a lightweight gradient booster provided in this embodiment can also execute early warning filtering logic. This allows the classification decision module to use a lightweight gradient booster (LightGBM) as a classifier to classify the feature operators extracted by the multi-dimensional feature construction module. After obtaining the smoke identification result, the smoke identification result will also be used as input for hierarchical scheduling and closed-loop optimization of operation and maintenance decision-making.

[0309] The closed loop refers to the entire process from alarm discovery, verification, handling, feedback to archiving, which is completed and traceable.

[0310] This hierarchical scheduling and closed-loop optimization can achieve risk classification, differentiated handling, dynamic escalation, and merging of repeated alarms for alarm events while ensuring safety as the priority. It can also achieve continuous iterative optimization of thresholds, strategies, and models through the feedback of handling results, thereby reducing the cost of relocation caused by false alarms and improving the scalability of large-scale operation and maintenance.

[0311] Risk grading is the process of classifying alarm events into different risk levels based on the type of hazard, confidence level, and auxiliary information.

[0312] "On-site inspection" refers to the actions taken by maintenance or patrol personnel to conduct on-site verification after receiving an alarm.

[0313] The hierarchical scheduling and closed-loop optimization data involved in the hierarchical scheduling and closed-loop optimization process includes, but is not limited to:

[0314] a) Terminal ID, installation location, and alarm timestamp;

[0315] b) Resident and contact person information (at least one contact method);

[0316] c) The classification decision module obtains the smoke identification results and the corresponding probability / confidence vector;

[0317] The confidence level is the degree of credibility or probability of the Lightweight Gradient Boosting Machine (LightGBM) model's judgment result for a certain hazard category.

[0318] d) Optional enhanced information: saturation flag, trend characteristics (decline / continuous rise), device health status (battery level, offline status, fault codes, etc.), historical alarm statistics, etc.

[0319] The results of hierarchical scheduling and closed-loop optimization include:

[0320] a) Risk level L (e.g., L1 low risk, L2 medium risk, L3 high risk, L4 critical);

[0321] b) The policy package should include at least the following: the contact verification policy, the observation window and reassessment frequency, the dispatch resource type, the on-site SLA, the linkage level, the escalation / degradation rules, the evidence return fields and the closed-loop conditions.

[0322] c) Work Ticket, which includes the work ticket number, executing organization / personnel, status, handling SOP, and feedback requirements.

[0323] The hierarchical scheduling and closed-loop optimization process is as follows:

[0324] 501, Receive the smoke identification results and classification output (category and confidence level) from the classification decision module.

[0325] While receiving the smoke identification results output by the classification decision module, it also receives the probability / confidence vectors for each category from the classification decision module for subsequent mapping and dynamic re-judgment.

[0326] 502. The risk level is determined according to the mapping rules;

[0327] Risk level mapping can be achieved using threshold rules, expected cost minimization rules, or a combination of both. The preferred implementation is threshold rule mapping.

[0328] 1. When the classification is "real fire" and its confidence level reaches the first highest threshold, the risk level is mapped to L3 or L4.

[0329] 2. When the classification is "real fire" but the confidence level is in the medium range, the risk level is mapped to L2 or L3;

[0330] 3. When the judgment category is "burnt pot" and the confidence level reaches the set threshold, the risk level is mapped to L2 (and can be configured to L3 according to the regional strategy).

[0331] 4. When the category is determined to be low-risk, such as "cooking fumes," "water vapor," "smoke," or "dust," and the confidence level reaches the set threshold, the risk level is mapped to L1.

[0332] 5. When the confidence level of the classification output is insufficient, the category is unstable, or cannot be confirmed, the risk level is mapped to L2 as a fallback level.

[0333] Among the optional enhancement methods, when any of the following conditions are met, such as sustained high concentration saturation, significant temperature rise, or a continuous upward trend, the mapping level can be increased by one level or directly mapped to L3 / L4 to improve safety and response speed in high-risk scenarios.

[0334] 503, generate a handling strategy package and execute it to reach out for verification / dispatch / linkage;

[0335] For example, a policy package is generated based on the risk level L and the judgment category. This policy package should contain at least the following:

[0336] 1. Outbound verification strategy: including outbound calling method, number of redials, redial interval, rules for reaching backup contacts, SMS / APP push rules, and conditions for human agent intervention;

[0337] 2. Observation window and re-judgment strategy: including observation window duration, re-judgment frequency, re-judgment triggering conditions, and re-judgment termination conditions;

[0338] 3. Dispatch strategy: including whether to generate a disposal work order, dispatch resource type (station / inspection, property management, grid force, fire station / mini fire station, etc.), dispatch parallel / serial relationship, on-site SLA (service level time limit) and disposal SOP template;

[0339] 4. Levels of linkage: including notification, standby, parallel deployment, and automatic linkage with emergency channels;

[0340] 5. Upgrade / downgrade rules: These include triggering conditions such as no answer, increased risk, continued saturation, insufficient evidence, and changes in the retrial result;

[0341] 6. Evidence feedback fields and closed-loop conditions: including photo / video / text records, on-site readings, final conclusion labels, handling action records, equipment restoration and re-inspection requirements, etc.

[0342] 504, dynamically review the judgment within the observation window and adjust the strategy according to the upgrade / downgrade rules;

[0343] The observation window can be a window with a preset time duration, which is oriented towards future time. For example, the preset time duration starting from the current acquisition time constitutes the observation window.

[0344] To avoid mis-assignment due to short-term interference, and to ensure rapid response in high-risk scenarios, dynamic review and adjustment of strategies according to escalation / degradation rules include, but are not limited to:

[0345] 1. Continuously receive new sensor data within the observation window and repeatedly perform classification reasoning to obtain updated categories and confidence levels;

[0346] 2. When any upgrade condition is met, the risk level will be upgraded immediately and the strategy package will be adjusted, including but not limited to: the confidence level of the real fire situation continues to rise and exceeds the threshold, the category shifts from low risk to high risk, the saturation flag is continuously triggered, and the trend continues to rise.

[0347] 3. When any of the downgrade or closure conditions are met, the risk level will be reduced or the system will be closed directly, including but not limited to the risk falling back and residents confirming no abnormalities, and multiple consecutive reassessments showing a stable low-risk category with high confidence.

[0348] 505, merge duplicate alarms and update the work order within the merge window;

[0349] To reduce duplicate order dispatches caused by repeated alarms from the same terminal within a short period of time, a merge window will be set in step 505:

[0350] 1. Within the merge window, multiple alarm events from the same terminal are merged into a single main event;

[0351] 2. When a new alarm arrives, update the category, confidence level, and risk level of the main event, and refresh the policy package and work order status;

[0352] 3. If a work order has already been generated, do not create a new work order. Instead, update the work order content and processing level to reduce the cost of repeated visits and processing.

[0353] 506. Collect the on-site verification results and archive them in a closed loop, then return them for threshold, strategy and model optimization.

[0354] After the disposal is completed, the smoke identification system based on multi-feature fusion and lightweight gradient booster provided in this embodiment will structurally archive the final verification conclusion label, evidence materials, disposal actions, work order timeliness, upgrade and merging information, equipment health status and maintenance records, and bind the results with the classification output when the alarm is triggered for feedback, which is used for threshold calibration, strategy parameter optimization and model iterative training, thereby forming a continuous optimization of "identification-scheduling-disposal-feedback".

[0355] The above-mentioned hierarchical scheduling and closed-loop optimization process can achieve hierarchical handling.

[0356] For example, (1) L1 low-risk treatment:

[0357] When the risk level is L1, the smoke identification system based on multi-feature fusion and lightweight gradient booster provided in this embodiment prioritizes the remote verification process, including automatic outbound calls and SMS / APP push, and performs re-judgment at a set frequency in the observation window; if the resident confirms that there is no abnormality or the risk of re-judgment decreases, the event is closed-loop archived; if no one answers and the re-judgment shows that the risk has increased, the upgrade to L2 or L3 is triggered.

[0358] (2) Risk management in L2:

[0359] When the risk level is L2, the smoke identification system based on multi-feature fusion and lightweight gradient lift provided in this embodiment generates a disposal work order and dispatches resources such as dispatching personnel / patrols to conduct on-site verification, and can simultaneously notify property management or grid forces; upon arrival, the system verifies the information according to the SOP and returns evidence and conclusion labels; if signs of fire are found or the risk is reassessed and continues to rise, the system is upgraded to L3 / L4 and triggers corresponding linkage.

[0360] (3) L3 high-risk management:

[0361] When the risk level is L3, the smoke identification system based on multi-feature fusion and lightweight gradient lift provided in this embodiment adopts a parallel handling strategy, which includes at least parallel dispatching of dispatching resources for patrol and inspection and fire stations / mini fire stations, and enters a high-frequency review and real-time monitoring state; when necessary, it triggers manual intervention and additional linkage; the entire handling process is recorded and archived afterward.

[0362] (4) L4 emergency response:

[0363] When the risk level is L4, the smoke identification system based on multi-feature fusion and lightweight gradient lift provided in this embodiment triggers the emergency response process, including automatically linking emergency exits and simultaneously notifying fire stations / property management / grid forces, pushing evacuation tips to residents, and recording and mandating closed-loop archiving of the entire process.

[0364] This smoke detection system based on multi-feature fusion and a lightweight gradient booster can achieve a closed loop of alarm handling and maintenance services, and reduce the manpower costs caused by false alarms, thereby supporting large-scale, low-cost deployment. For example, the smoke detection system based on multi-feature fusion and a lightweight gradient booster provided in this embodiment establishes an alarm reach and confirmation mechanism, improving response accessibility and solving the problem of "no one responding when the alarm sounds." By automatically triggering outbound calls, SMS / message push notifications, and other reach methods when an alarm occurs, combined with timeout escalation and multi-channel confirmation mechanisms, it ensures that alarms can be verified and responded to in a timely manner, improving resident accessibility and the timeliness of handling.

[0365] This smoke identification system, based on multi-feature fusion and a lightweight gradient lift, establishes a closed loop for equipment health monitoring and maintenance, solving the problems of "no one repairing it when it breaks down / it becomes ineffective over time." Through platform-based access and status monitoring, it can identify and alarm on health statuses such as offline, low battery, abnormal drift, and frequent false alarms. Through dispatching maintenance orders and closed-loop verification mechanisms, it ensures the long-term effective operation of the equipment.

[0366] This smoke identification system based on multi-feature fusion and a lightweight gradient booster can improve the ability to identify interference scenarios and reduce false alarms and disturbances. As provided in this embodiment, the smoke identification system based on multi-feature fusion and a lightweight gradient booster can distinguish between non-fire scenarios such as cooking fumes, water vapor, smoking, dust, and burnt food, and real fires by performing more refined data collection and modeling of smoke and environmental signals, thereby reducing invalid alarms and minimizing disturbances and operational burden.

[0367] This smoke identification system based on multi-feature fusion and a lightweight gradient booster can achieve risk classification and differentiated handling, reducing unnecessary on-site operations and improving handling efficiency. As provided in this embodiment, the smoke identification system based on multi-feature fusion and a lightweight gradient booster formulates different handling strategies and personnel scheduling plans according to the alarm risk level, confidence level, signal persistence, and scene category: low-risk events are mainly verified remotely, medium-risk events trigger targeted on-site verification, and high-risk events involve multiple parties for rapid handling, thereby significantly reducing the proportion of unnecessary on-site verification and supporting controllable manpower costs when expanding coverage.

[0368] This smoke identification system, based on multi-feature fusion and a lightweight gradient booster, enhances the system's adaptability to resident differences and environmental changes, improving its generalization ability and stability. For example, by characterizing alarm evolution patterns through multi-sensor composite features and temporal features (such as time window statistics, rate of change, pattern changes, and state memory), the identification and grading strategies can adapt to differences in different households, seasons, and ventilation conditions, improving long-term stability and reducing rule maintenance costs.

[0369] This smoke recognition system, based on multi-feature fusion and a lightweight gradient booster, provides lightweight, edge-deployable intelligent classification capabilities, ensuring real-time performance and low cost. The use of lightweight models and real-time computational feature construction methods allows for flexible deployment of inference on the edge, cloud, and device sides, reducing computational and communication overhead while maintaining recognition accuracy and achieving real-time response.

[0370] This smoke identification system, based on multi-feature fusion and a lightweight gradient booster, forms a continuous optimization loop of "response feedback," supporting model and strategy iteration. For example, information such as resident feedback, on-site verification results, and final qualitative assessments are structured, stored, and fed back for threshold calibration, model retraining, and strategy optimization, achieving adaptive improvement over time and providing a data foundation for operational quality assessment and regulatory record-keeping.

[0371] This embodiment relates to an IoT-based linkage response method for residential smoke detectors, which acquires alarm information reported by smoke detector terminals. Multiple smoke detector terminals are located at the residential end, and all terminals establish a communication connection via the Internet of Things (IoT). The method determines whether a fire has occurred based on the alarm information. If a fire has occurred, a processing work order is generated and issued based on the smoke detector terminals. This method improves the responsiveness of alarm information by determining whether a fire has occurred based on alarm information, and then generating and issuing a processing work order after confirming a fire, thus solving the problem of "no one responding when the alarm sounds."

[0372] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0373] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0374] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0375] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0376] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0377] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0378] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A method for IoT-based coordinated response to smoke detectors in residential settings, characterized in that: The method includes: Obtain alarm information reported by smoke detection terminals; wherein there are multiple smoke detection terminals, the smoke detection terminals are located at the residential end, and all smoke detection terminals establish a communication connection based on the Internet of Things; Determine whether a fire has occurred based on the alarm information; In the event of a fire, a processing work order is generated based on the smoke detector terminal, and the processing work order is issued.

2. The method according to claim 1, characterized in that, Before determining whether a fire has occurred based on the alarm information, the process also includes: The raw electrical signals within the monitoring space corresponding to each smoke detector terminal are periodically collected; The original electrical signal is subjected to analog-to-digital conversion and standardization to obtain the processed data; Extract feature operators from the processed data; The step of determining whether a fire has occurred based on the alarm information includes: Based on the alarm information, determine the feature operator corresponding to the target monitoring space; A lightweight gradient booster is used as a classifier to classify the feature operators corresponding to the target monitoring space to determine whether a fire has occurred.

3. The method according to claim 2, characterized in that, The original electrical signals include: a humidity signal, a temperature signal, a first red light signal, a second red light signal, a first blue light signal, and a second blue light signal. The process of performing analog-to-digital conversion and standardization on the original electrical signal to obtain processed data includes: The humidity electrical signal is converted from analog to digital and standardized to obtain the humidity value; The electrical signal of temperature is converted from analog to digital and standardized to obtain the temperature value; The electrical signal of the first red light is subjected to analog-to-digital conversion and standardization to obtain the value of the first red light; The electrical signal of the second red light is subjected to analog-to-digital conversion and standardization to obtain the value of the second red light; The electrical signal of the first blue light is subjected to analog-to-digital conversion and standardization to obtain the first blue light value; The electrical signal of the second blue light is subjected to analog-to-digital conversion and standardization to obtain the second blue light value; Among them, the humidity value, temperature value, first red light value, second red light value, first blue light value, and second blue light value are all identified by the collection timestamp.

4. The method according to claim 2, characterized in that, The feature extraction operator from the processed data includes: After acquiring a raw electrical signal, instantaneous feature operators, time window feature operators, mode change feature operators, and state memory feature operators are extracted from the processed data acquired in that acquisition. Among them, the instantaneous feature operator, the time window feature operator, the mode change feature operator, and the state memory feature operator all use the collection timestamp as an identifier; Instantaneous feature operators are those that reflect the characteristics of a single data acquisition. The time window feature operator is an operator that reflects continuous changes; The pattern change characteristic operator is an operator that reflects jump changes; The state memory feature operator is an operator that reflects the normal threshold.

5. The method according to claim 4, characterized in that, The processed data includes humidity value, temperature value, first red light value, second red light value, first blue light value, and second blue light value; The instantaneous feature operator includes: humidity value, temperature value, first red light value, second red light value, first blue light value, second blue light value, blue-red light difference value, blue-red light ratio value, and saturation flag bit; Wherein, the difference between blue and red light = first blue light value - first red light value; Blue-to-red light ratio = first blue light value / (first red light value + epsilon); where epsilon is a preset minimum positive number; The saturation flag includes a red light flag and a blue light flag. When the second red light value is greater than the preset full-scale threshold, the red light flag is 1, and when the second red light value is not greater than the preset full-scale threshold, the red light flag is 0. When the second blue light value is greater than the preset full-scale threshold, the blue light flag is 1, and when the second blue light value is not greater than the preset full-scale threshold, the blue light flag is 0.

6. The method according to claim 4, characterized in that, The time window feature operator includes: the mean of the processed data for each time window, the maximum value of the processed data for each time window, the minimum value of the processed data for each time window, the standard deviation of the processed data for each time window, the rate of change of the processed data, and the duration for which the processed data meets the preset rules. The time window is a sliding window that starts from the current acquisition time and moves forward according to a preset time step. The rate of change of the processed data is the difference between the mean of the processed data in the second time window and the mean of the processed data in the first time window, divided by 2 times the preset time step. If the preset rule is that the saturation flag is 1, then the duration for which the processed data satisfies the preset rule is the duration between the first moment and the second moment; where the first moment is the first acquisition moment when both the red and blue flags are 1 forward from the current acquisition moment, and the second moment is the last acquisition moment when both the red and blue flags are 1 forward from the first moment.

7. The method according to claim 1, characterized in that, The method further includes: Obtain device operation data reported by the smoke detector terminal; When an anomaly is identified based on the equipment's operating data, an early warning is issued.

8. The method according to claim 1, characterized in that, The location of the smoke detector terminal at the resident's end is determined through the following steps: The electricity consumption of each meter at the residential end is collected once during each electricity collection period. Whenever a new energy consumption is collected from any meter, the following steps 201-203 are executed:

201. If the newly collected electricity consumption is greater than the first consumption threshold, then the household corresponding to any of the electricity meters is determined as the location of the smoke detector terminal at the resident end.

202. If the newly collected energy consumption is less than the second consumption threshold, then when the energy consumption collected by any meter in a consecutive first number of energy collection periods is less than the second consumption threshold, the location of the smoke detector terminal is determined based on the neighboring meters of any meter; when the energy consumption collected by any meter in a consecutive first number of energy collection periods is not all less than the second consumption threshold, the location of the smoke detector terminal is determined based on the historical energy consumption of any meter; wherein, the second consumption threshold is less than the first consumption threshold.

203. If the newly collected energy consumption is not greater than the first consumption threshold and not less than the second consumption threshold, then determine whether it is the location of the residential end where the smoke detector terminal is located based on the historical energy consumption of any of the meters.

9. The method according to claim 8, characterized in that, The step of determining whether the location of the smoke detector terminal is the residential end based on the neighboring meters of any of the aforementioned meters includes: Based on the location of each meter, determine the neighboring meters of any given meter; If a first target meter exists in the neighboring meters, or if the proportion of second target meters in the neighboring meters is greater than a preset proportion, then the household corresponding to any of the meters is determined as the location of the smoke detector terminal at the resident end. Among them, the latest energy consumption collected by the first target meter is greater than the first consumption threshold. The latest energy consumption collected by the second target meter is not greater than the first consumption threshold and not less than the second consumption threshold. However, the energy consumption collected in the second consecutive number of energy collection periods from the current collection period is not less than the first consumption threshold - (first consumption threshold - second consumption threshold) / 3.

10. The method according to claim 8, characterized in that, The step of determining whether the location of the smoke detector terminal is the residential end based on the historical energy consumption of any of the electricity meters includes: Determine the mean and standard deviation of the energy consumption collected by any meter during each energy collection time period; If the energy consumption collected in the third consecutive energy consumption collection period preceding the current collection period is greater than the sum of the average energy consumption and the standard deviation of the energy consumption, and the energy consumption collected in the third consecutive energy consumption collection period preceding the current collection period shows an upward trend over time, then the household corresponding to any of the electricity meters is determined as the location of the smoke detector terminal at the resident end.

Images