Household Internet of Things integrated fire alarm system and linkage control method thereof

Abstract

The invention discloses a household Internet of Things integrated fire alarm system and a linkage control method thereof, and relates to the technical field of smart home security. A first sensing recognition module, a second deep recognition module, a linkage control module, a cloud management platform module and a system self-checking module are integrated; the problems that in an existing fire alarm system, the false alarm rate is high, linkage equipment is prone to false triggering, linkage control logic is complex, and errors are prone to occurring are effectively solved. According to the invention, multi-modal data fusion and a deep learning algorithm are adopted to carry out fire identification, a dynamic threshold adjustment algorithm is combined to optimize the sensitivity of a sensor, and intelligent linkage configuration and an anomaly detection function are provided through a cloud management platform; the accuracy of fire identification and the stability of the system are greatly improved, the credibility of a user to the system and the comprehensiveness of family safety protection are remarkably improved, and resource waste and unnecessary panic caused by false alarm are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of smart home security technology, and more specifically to a home Internet of Things integrated fire alarm system and a linkage control method thereof. Background Art

[0002] With the rapid development of Internet of Things technology, smart homes have gradually become a part of modern family life, greatly improving the convenience and comfort of residents' lives. Especially in the field of fire safety, home fire alarm systems based on the Internet of Things have become an important means of ensuring family safety. In recent years, with the continuous advancement of sensor technology, wireless communication technology and cloud computing, home fire alarm systems can not only achieve early warning of fires, but also realize automatic fire extinguishing and emergency evacuation through linkage control, significantly improving the ability of families to deal with fires;

[0003] The existing home IoT integrated fire alarm systems mainly rely on a variety of sensors (such as smoke, temperature, gas, etc.) for fire monitoring, and use IoT technology to achieve remote monitoring and alarm. These systems usually also have linkage control functions, which can automatically trigger other smart home devices (such as closing gas valves, turning on exhaust fans, etc.) when a fire occurs to reduce the harm of the fire.

[0004] Patent CN118658260B discloses a passive smoke fire detector, a smoke alarm system and a detection method, wherein the optical signal emitted by the controller is transmitted to the receiving part through the transmission part, and the filter unit and the reflection unit of the receiving part filter, separate and reflect the optical signal respectively, so that the optical signal meeting the characteristic wavelength of the smoke sensor enters the detection part, and then is transmitted through the receiving part at the other end, and smoke detection is performed by the controller, and then an early warning is issued; however, in actual use, since the sensors in the system (such as smoke sensors and temperature sensors) are easily interfered by non-fire factors in daily life, such as cooking fumes, steam, cigarette smoke, etc. These interference factors may cause the sensor to misjudge, thereby triggering false alarms, and thus often encounter the problem of a high false alarm rate; false alarms will not only frequently disturb the user's life, but also cause the user's trust in the system to decrease, and even ignore the alarm when a real fire occurs, thereby missing the best escape opportunity. In addition, the fire alarm system integrated with the home Internet of Things usually needs to work in conjunction with other smart devices in the home (such as smart door locks, window sensors, smart sockets, etc.) to achieve more comprehensive security protection; however, false alarms will not only lead to false alarms, but may also cause false triggering of linkage devices; such as the home fire alarm system based on the Internet of Things disclosed in patent CN203941592U, although it can prevent home fires, it may trigger unnecessary linkage responses due to false alarms when integrating linkage devices, resulting in waste of resources and unnecessary panic. Secondly, the linkage control logic of the fire alarm system integrated with the home Internet of Things usually requires complex condition settings, including state monitoring, condition judgment, and action execution; however, if these conditions are not set properly, the system may not respond correctly at critical moments. For example, the linkage control logic may need to make a comprehensive judgment based on multiple parameters such as smoke concentration, ambient temperature, and flame conditions. If the threshold setting of a certain parameter is unreasonable, the system may fail to initiate the corresponding safety measures in time at the early stage of the fire. In addition, the complexity of linkage control also increases the probability of system errors, which may cause some key equipment to fail to work properly in an emergency. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention discloses a home Internet of Things integrated fire alarm system and a linkage control method thereof, aiming to solve the problems in the background technology.

[0006] In order to achieve the above technical effects, the present invention adopts the following technical solutions:

[0007] A fire alarm system integrated with the home Internet of Things, comprising: a first perception and recognition module, a second depth recognition module, a linkage control module, a cloud management platform module and a system self-check module;

[0008] The first perception and recognition module is used to monitor environmental parameters in real time, and to make preliminary judgments and early warnings on fire situations through a multimodal data fusion method, and to make preliminary misjudgment corrections through an environmental correction algorithm;

[0009] The second deep recognition module is used to comprehensively analyze the real-time monitoring environmental parameters output by the first perception recognition module through a fire recognition algorithm based on deep learning, identify fire characteristics, and judge fire and non-fire events; when the second deep recognition module identifies a fire event, it transmits the fire characteristic data to the linkage control module and the cloud management platform module through a wireless communication method; when the second deep recognition module does not identify a fire event, it dynamically adjusts the recognition sensitivity of each sensor of the first perception recognition module according to environmental changes through a dynamic threshold adjustment algorithm;

[0010] The linkage control module is used to control the home security device through a linkage control strategy based on the fire feature data output by the second deep recognition module;

[0011] The cloud management platform module is used to generate linkage device configuration solution suggestions through a linkage logic analysis algorithm based on machine learning, and provide users with linkage device configuration, alarm notification and device status display functions;

[0012] The system self-check module is used to regularly check the operating status of linkage devices and sensors, and generate an abnormality detection report and transmit it to the cloud management platform module.

[0013] As a further technical solution of the present invention, the first perception and recognition module detects smoke concentration through a smoke sensor, measures ambient temperature through a temperature sensor, detects infrared radiation of the flame through a flame sensor, monitors ambient humidity through a humidity sensor, detects carbon monoxide and carbon dioxide concentrations through an air quality sensor, and captures video images through a camera; when the smoke concentration, ambient temperature or flame intensity exceeds a preset threshold, the first perception and recognition module judges the ambient humidity and air quality; if the ambient humidity and air quality data exceed the preset threshold range, a multimodal data fusion method is used to comprehensively judge it as a preliminary fire warning, and a control signal is transmitted to the cloud management platform module for early warning notification, and the smoke concentration, ambient temperature, flame intensity, ambient humidity, carbon monoxide and carbon dioxide concentrations and video image data are transmitted to the second deep recognition module; when the smoke concentration exceeds the preset threshold, if the ambient humidity and air quality data do not exceed the preset threshold range, the smoke sensor data is corrected through an environmental correction algorithm to reduce false alarms, and monitoring continues.

[0014] As a further technical solution of the present invention, the working principle of the multimodal data fusion method is as follows: receiving monitoring parameters of smoke sensors, temperature sensors, flame sensors, humidity sensors and air quality sensors; if any sensor data of the smoke sensors, temperature sensors and flame sensors exceeds a preset threshold, a preliminary fire judgment is made by combining the monitoring parameters of other sensors through a weighted average method, and if yes, further analysis is made through a conditional probability fusion mechanism, and if not, monitoring is continued; if multiple sensor data of the smoke sensors, temperature sensors and flame sensors exceed a preset threshold, the joint probability distribution of smoke concentration, ambient temperature and flame intensity parameters under specific conditions is calculated through a conditional probability fusion mechanism to evaluate the probability of fire occurrence; the conditional probability fusion mechanism combines sensor information according to the probability principle through a multi-Bayesian estimation method, and synthesizes the associated probability distribution of each sensor parameter into a joint conditional probability, which is used to represent the probability of fire occurrence; if the conditional probability exceeds the preset threshold, it is judged that there is a fire risk and an alarm mechanism is triggered; if the conditional probability is lower than the preset threshold, environmental changes are continued to be monitored without triggering an alarm.

[0015] As a further technical solution of the present invention, the working method of the environmental correction algorithm is:

[0016] Step 401: pre-process the smoke density data, remove noise and outliers through Kalman filter, and the pre-processed smoke density data is recorded as C' smoke ;

[0017] Step 402: Extract smoke concentration C from historical data smoke , ambient humidity H and the characteristic vector F of air quality index AQI = [C smoke ,H,AQI];

[0018] Step 403: Calculate the posterior probability P(C smoke |H, AQI); the posterior probability calculation formula of the Bayesian network is:

[0019]

[0020] In formula (1), P(C smoke |H,AQI) indicates the smoke concentration C smoke Under the condition of ambient humidity H and air quality index AQI, the posterior probability P(H,AQI|C smoke ) is the conditional probability, indicating that given the smoke concentration C smoke The joint probability of ambient humidity H and air quality index AQI under smoke ) represents the smoke density C smokeThe prior probability of; P(H,AQI) represents the joint probability of ambient humidity H and air quality index AQI;

[0021] Step 404: Corrected smoke density data C" smoke Calculated by the following formula:

[0022]

[0023] In formula (2), C” smoke Indicates the corrected smoke density data; a indicates the correction factor, which is used to adjust the correction intensity; Indicates the smoke concentration C under current environmental conditions. smoke Credibility;

[0024] Step 405: dynamically adjust the smoke concentration threshold T according to the real-time changes of the ambient humidity H and the air quality index AQI smoke , the adjustment formula is:

[0025] T' smoke =T smoke +k 1 (HT H )+k 2 (AQI-T AQI )-1 (3)

[0026] In formula (3), T' smoke Indicates the smoke density threshold after dynamic adjustment; k 1 Represents the humidity adjustment coefficient, which is used to adjust the effect of humidity on the threshold; k 2 Represents the air quality adjustment factor, which is used to adjust the impact of air quality on the threshold; T H Indicates the preset ambient humidity threshold; T AQI Indicates the preset air quality threshold;

[0027] Step 406: Corrected smoke density data C" smoke and the dynamically adjusted threshold T' smoke Compare; if C" smoke ≤T' smoke , it is judged as a false alarm and the system continues to monitor; otherwise, it is judged as a real fire and the alarm mechanism is triggered.

[0028] As a further technical solution of the present invention, the working method of the fire identification algorithm based on deep learning is:

[0029] Step 501: Based on the video monitoring data output by the first perception recognition module, feature extraction is performed on the real-time image through a convolutional neural network; the convolutional neural network extracts feature maps layer by layer from the original data through multi-layer convolution kernels and pooling operations, and the feature maps include flame color, shape, texture, and dynamic changes;

[0030] Step 502: Based on the feature graph output in step 501 and the environmental monitoring data output by the first perception and recognition module, a multi-layer stacked LSTM is used to capture the changing trend of environmental parameters in the time series when a fire occurs. The formula expression is:

[0031]

[0032] In formula (4), represents the hidden state of the fire feature at time step t, which contains the fire-related temporal information extracted from the environmental parameters; represents the environmental parameter input at time step t, including smoke density, ambient temperature, flame intensity, ambient humidity, carbon monoxide and carbon dioxide concentrations; L fire represents the number of LSTM layers, which is used to capture the complex changes of fire characteristics in time series; and Respectively represent the weight matrices from hidden state to hidden state, input to hidden state, and hidden state to forget gate of the i-th layer LSTM; and They represent the bias term of the i-th layer LSTM and the bias term of the final hidden state, respectively, which are used to adjust the representation of fire characteristics; r represents the activation function, which is used to control the flow and scaling of information;

[0033] Step 503: Map the output of step 502 to a multi-dimensional classification space through a fully connected layer, where each dimension corresponds to a fire level or a category of non-fire events, and calculate the probability distribution of each category through a multi-classification strategy function, and select the category with the highest probability as the final recognition result; the formula expression of the multi-classification strategy function is:

[0034]

[0035] In formula (5), Y fire represents the fire feature map, which contains the discriminative features related to fire extracted from the image data; X img represents the image input, including smoke and flame features; M fire Represents the number of convolutional layers, which is used to increase the depth and feature extraction capability of the model; N fire Represents the number of convolution kernels in the mth convolution layer; represents the weight matrix corresponding to the nth convolution kernel of the mth layer; CON represents a two-dimensional convolution operation, and the input is the image data X img , convolution kernel Step Length and fill Represents the nth convolution kernel of the mth layer; Indicates the step size of the nth convolution operation in the mth layer; Represents the padding of the nth convolution operation in the mth layer; Represents the bias term corresponding to the nth convolution kernel of the mth layer; ReLU represents the linear rectification function, which is used to increase the nonlinearity of the model;

[0036] Step 504: If the identification result is a fire event, the fire feature map is mapped to the final fire level through a level decision function according to the fire feature intensity and change trend; the fire feature intensity includes flame area and smoke density; the change trend includes rising rate and duration; the formula expression of the level decision function is:

[0037]

[0038] In formula (6), C fire Indicates the final fire grade classification results, including grade one, grade two, and grade three; J fire Indicates the number of classifiers; n jfire represents the weight coefficient of the jth classifier, which is used to adjust the contribution of different classifiers to the final fire grade classification result; softmax is the softmax function, which is used to convert the output of the classifier into a probability distribution; W jfire represents the weight matrix of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire Represents the bias term of the j-th classifier; Represents the weight coefficient of the kth fire level, which is used to adjust the contribution of different fire levels to the final result; () k Represents the k-th element of a vector or matrix.

[0039] As a further technical solution of the present invention, the cloud management platform module includes a user linkage device configuration submodule, an alarm notification submodule and a device status display submodule; the user linkage device configuration submodule includes a device management unit and a scene linkage unit; the device management unit implements device registration and authentication through the OAuth 2.0 protocol, and implements device status display through the RESTful The API interface realizes the management and update of device configuration; if the user modifies the device configuration, the user linkage device configuration submodule pushes the updated configuration information to the corresponding device through the MQTT protocol; the scene linkage unit provides users with the function of creating and editing linkage scenes through a data interaction method; the scene linkage unit also generates a linkage device configuration plan through a linkage logic analysis algorithm based on machine learning; when a fire event is detected, the alarm notification submodule notifies the user of the alarm information through SMS, phone, email and push notification, and stores the alarm record through a relational database, including the alarm time, location, type and processing result information; the device status display submodule collects the status information of each device in real time through the Zigbee protocol, including online status, working mode, battery power, and displays it; the device status display submodule also uses a time series database to store the historical status records of the device, including the timestamp and specific value of the status change; based on the historical status data, the historical status records of the device are also analyzed through time series analysis to generate a status change trend report and display it.

[0040] As a further technical solution of the present invention, the working method of the dynamic threshold adjustment algorithm for dynamically adjusting the recognition sensitivity of each sensor according to environmental changes is:

[0041] Step 701: When the second depth recognition module does not recognize a fire event, the dynamic threshold adjustment algorithm obtains the current environmental parameters, including the smoke density C, from the first perception recognition module through the Zigbee protocol. smoke 、Ambient temperature E evn 、Flame intensity I flame , ambient humidity H and air quality index AQI;

[0042] Step 702: Obtain historical environmental data through the cloud platform, including historical records of smoke concentration, ambient temperature, flame intensity, ambient humidity, and air quality index in the past 24 hours;

[0043] Step 703: Based on the historical environmental data obtained in step 702, an environmental change model is constructed by a time series analysis method to predict future environmental change trends; the time series analysis method models the time series data through three parts: autoregression, difference, and moving average, and captures the change pattern of environmental parameters through data smoothing and trend extraction;

[0044] Step 704: Calculate the threshold adjustment factor of each sensor according to the future environmental parameters predicted by the environmental change model. The calculation formula is:

[0045]

[0046] In formula (7), represents the threshold adjustment factor of the εth sensor; h ε Indicates the adjustment coefficient; E ε Represents the predicted future environmental parameters; E ε,avg Represents the average value of historical environmental parameters;

[0047] Step 705: Based on the calculated threshold adjustment factor, the recognition sensitivity of each sensor is dynamically updated by weighted average. During the adjustment process, the stability of the threshold is ensured by threshold stability test;

[0048] Step 706: divide the historical data into a training set and a test set, use the training set data to train the model, and use the test set data to verify the performance of the model. If the performance of the model on the test set meets expectations, it is considered that the threshold adjustment is effective; otherwise, readjust the threshold;

[0049] Step 707: Apply the verified effective threshold update to each sensor of the first perception and recognition module through the Zigbee protocol.

[0050] As a further technical solution of the present invention, the working method of the system self-check module is: regularly triggering self-check tasks through a timed task scheduling mechanism to collect real-time status data of each linkage device and sensor, including the online status, working mode, battery power, and communication status of the device; based on the collected real-time status data, analyzing through an abnormality detection algorithm to detect the abnormal state of the device; if no abnormal state is detected, ending this round of self-check; if an abnormal state is detected, classifying through a support vector machine to identify the type of abnormality, including communication failure, low battery, and sensor failure; the classified abnormal state generates an abnormality detection report through a report generation module, including the detected abnormal state, abnormality type, occurrence time, and device ID; the generated abnormality detection report is transmitted to the cloud management platform module through the MQTT protocol, and the cloud management platform module automatically handles the abnormality according to preset rules through a rule engine, including sending maintenance notifications, restarting the device, and adjusting device parameters; and notifying the user of the abnormal situation and processing results through the email communication protocol.

[0051] As a further technical solution of the present invention, the working method of the linkage logic analysis algorithm based on machine learning is:

[0052] Step 901: collect real-time data from various sensors and devices through the Zigbee protocol and wireless communication network, and pre-process the collected data through data cleaning and normalization;

[0053] Step 902: historical data is obtained through the cloud platform, including historical fire event records and equipment operation records; the historical data is annotated through semi-automatic annotation to mark the triggering conditions and corresponding linkage operations of each fire event; the semi-automatic annotation is assisted by a machine learning algorithm to generate historical data with labels;

[0054] Step 903: Based on the historical data with labels, feature extraction is performed on the historical data by recursive feature elimination to generate a feature vector, where the feature vector includes environmental parameters and device status; the environmental parameters include smoke concentration, ambient temperature, flame intensity, ambient humidity, and air quality index; the device status includes a sprinkler, automatic door lock, and smart socket;

[0055] Step 904: Train the annotated historical data through a support vector machine to generate a fire identification and linkage control model;

[0056] Step 905: Generate optimal linkage logic under different fire scenarios through a generative adversarial network; the generative adversarial network is composed of a generator and a discriminator, the generator is used to generate linkage logic, and the discriminator is used to evaluate the rationality of the generated linkage logic; the generator generates linkage logic through a deep neural network, including the operation sequence and conditions of the equipment; the discriminator evaluates the rationality of the generated linkage logic through a deep neural network;

[0057] Step 906: Based on the linkage logic generated in step 905, define the state, action and reward function through a reinforcement learning algorithm to optimize the linkage logic;

[0058] Step 907: Based on the optimized linkage logic, a series of rules are generated through a decision tree algorithm to guide the specific operation of the equipment under different fire levels; the working method of the decision tree algorithm to generate a specific linkage configuration scheme is:

[0059] r1. Convert linkage logic into rule instructions through rule generation;

[0060] r2. Optimize the decision tree through pruning method to reduce redundant rules;

[0061] r3. Convert the generated linkage configuration plan into executable rules through the rule engine.

[0062] As a further technical solution of the present invention, a linkage control method of a fire alarm system integrated with a home Internet of Things comprises the following steps:

[0063] S1. After the system is powered on, all linkage devices are initialized through the system self-check module, and the working status of each device is checked. If the device is working normally, it will proceed to the next step; if an abnormality is detected in the device, the fault information will be transmitted to the cloud platform through the wireless communication network to notify the user;

[0064] S2, receiving the threshold judgment result and sensor data output by the first perception recognition module through the wireless communication protocol Zigbee; if the threshold exceeds the set range, proceed to the next step; if not, continue monitoring and remain in standby state;

[0065] S3, transmit the fire signal to the cloud platform through the wireless communication network to give early warning notification to the user; and use the multimodal data fusion method to comprehensively judge the smoke concentration, temperature, flame, humidity and air quality parameters. If the smoke concentration exceeds the preset threshold and the temperature rise range exceeds the preset threshold, start the water sprinkler fire extinguishing device; if not, proceed to the next step;

[0066] S4. Call video analysis to identify fire features through a deep learning-based fire recognition algorithm to determine whether it is a fire event or a non-fire event. If it is confirmed to be a fire event, proceed to the next step; if not, continue monitoring and remain on standby;

[0067] S5, control the water sprinkler fire extinguishing device to start through the relay, open the automatic door lock through the Zigbee module, start the escape channel indicator light, and if the air quality sensor detects that the carbon monoxide concentration exceeds the standard, start the ventilation equipment through the relay to reduce the indoor harmful gas concentration, and cut off the power supply of household appliances through the smart socket;

[0068] S6. Dynamically adjust the linkage strategy according to the development of the fire; if the fire continues to expand, increase the water spraying volume of the sprinkler fire extinguishing device, and search and notify nearby smart smoke alarms through the wireless communication protocol Zigbee to form a regional linkage; if the fire is under control, gradually reduce the water spraying volume;

[0069] S7. The alarm information is transmitted to the cloud platform through the wireless communication network, and the user is notified through the mobile phone APP or the web page; the user checks the status of the joint control device through the cloud management platform module, receives the alarm information, and remotely controls the joint control device; if the user confirms that it is a false alarm, the linkage control is cancelled through the application interaction; if not, the linkage device continues to be started;

[0070] S8. Regularly check the working status of all linked devices and record the operating data of the devices; if a device failure or communication interruption is found, try to restart the failed device or re-establish the communication connection through the system self-check module, and send the fault information to the cloud platform through the wireless communication network to notify the user; if the failed device cannot be repaired automatically, notify the user through the cloud platform to perform manual intervention.

[0071] Based on the above technical solutions, the positive and beneficial effects of the present invention are:

[0072] 1. Through the multimodal data fusion method and environmental correction algorithm of the first perception recognition module, the system can more accurately identify fire situations and make preliminary misjudgment corrections for non-fire factors (such as cooking fumes, steam, cigarette smoke, etc.). This effectively reduces sensor misjudgments caused by daily interference factors, thereby significantly reducing the false alarm rate. At the same time, the second deep recognition module uses a fire recognition algorithm based on deep learning to conduct a comprehensive analysis of real-time monitoring environmental parameters, further improving the accuracy and robustness of fire recognition. Even under complex and changing environmental conditions, the system can accurately distinguish between fire and non-fire events, further reducing the possibility of false alarms.

[0073] 2. Due to the significant reduction in false alarm rate, the system reduces unnecessary alarms, thus avoiding frequent interruptions to users' lives. This improves users' trust in the system, making them more willing to believe and respond to the system's alarms when a real fire occurs, thereby increasing the success rate of escape and rescue; it also avoids the misoperation of linkage equipment triggered by false alarms, reducing unnecessary waste of resources and panic; through the user linkage equipment configuration, alarm notification and equipment status display functions provided by the cloud management platform module, users can more intuitively understand the system's operating status and alarm conditions, enhancing the transparency and credibility of the system.

[0074] 3. The linkage control module controls the home security devices through the linkage control strategy based on the fire feature data output by the second deep recognition module. This linkage control logic based on fire features is more accurate and efficient, and can promptly initiate corresponding safety measures at the early stage of a fire. The cloud management platform module generates linkage device configuration solution suggestions through a linkage logic analysis algorithm based on machine learning to help users optimize the linkage control logic. This reduces the problem of the system failing to respond correctly due to improper condition settings and improves the system's responsiveness at critical moments.

[0075] 4. The system self-check module regularly checks the operating status of linkage devices and sensors and generates abnormal detection reports. This helps to timely discover and solve potential problems in the system and enhance the stability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0077] Figure 1 This is an architecture diagram of a fire alarm system integrated with a home Internet of Things according to the present invention;

[0078] Figure 2 It is a working principle diagram of the multimodal data fusion method of the present invention;

[0079] Figure 3 A working method step diagram of the environmental correction algorithm of the present invention;

[0080] Figure 4 A diagram showing the working method of the system self-check module of the present invention;

[0081] Figure 5 It is a flow chart of the working method of the dynamic threshold adjustment algorithm of the present invention;

[0082] Figure 6 This is a module framework diagram of the cloud management platform of the present invention. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0084] like Figure 1 As shown, a fire alarm system integrated with the Internet of Things for a home comprises: a first perception and recognition module, a second depth recognition module, a linkage control module, a cloud management platform module and a system self-check module; wherein the output end of the first perception and recognition module is connected to the input end of the second depth recognition module and the cloud management platform module, and outputs preliminary judgment and warning results; the output end of the second depth recognition module is connected to the input end of the first perception and recognition module, the linkage control module and the cloud management platform module; the judgment result is output for linkage control, alarm, or output sensor sensitivity adjustment to the first perception and recognition module; the output end of the cloud management platform module is connected to the input end of the linkage control module, and the control strategy is output to the linkage control module; the output end of the system self-check module is connected to the input end of the cloud management platform module and the first perception and recognition module, and outputs a self-check report and performs self-check on sensors and equipment.

[0085] The first perception and recognition module is used to monitor environmental parameters in real time, and to make preliminary judgments and warnings on fire situations through a multimodal data fusion method, and is also used to make preliminary misjudgment corrections through an environmental correction algorithm; the first perception and recognition module detects smoke concentration through a smoke sensor, measures ambient temperature through a temperature sensor, detects infrared radiation of flames through a flame sensor, monitors ambient humidity through a humidity sensor, detects carbon monoxide and carbon dioxide concentrations through an air quality sensor, and captures video images through a camera; the first perception and recognition module determines ambient humidity and air quality when smoke concentration, ambient temperature or flame intensity exceeds a preset threshold; if ambient humidity and air quality data exceed a preset threshold range, a comprehensive judgment is made as a preliminary fire warning through a multimodal data fusion method, a control signal is transmitted to the cloud management platform module for a warning notification, and smoke concentration, ambient temperature, flame intensity, ambient humidity, carbon monoxide and carbon dioxide concentrations and video image data are transmitted to the second deep recognition module; when smoke concentration exceeds a preset threshold, if ambient humidity and air quality data do not exceed a preset threshold range, the smoke sensor data is corrected through an environmental correction algorithm to reduce false alarms, and monitoring continues. For further information, please refer to Figure 2 The working principle of the multimodal data fusion method is as follows: receiving the monitoring parameters of the smoke sensor, temperature sensor, flame sensor, humidity sensor and air quality sensor; if any sensor data of the smoke sensor, temperature sensor and flame sensor exceeds the preset threshold, a preliminary fire judgment is made by combining the monitoring parameters of other sensors through the weighted average method, and if yes, further analysis is made through the conditional probability fusion mechanism, and if no, monitoring is continued; if multiple sensor data of the smoke sensor, temperature sensor and flame sensor exceed the preset threshold, the joint probability distribution of the smoke concentration, ambient temperature and flame intensity parameters under specific conditions is calculated through the conditional probability fusion mechanism to evaluate the probability of fire; the conditional probability fusion mechanism combines the sensor information according to the probability principle through the multi-Bayesian estimation method, and synthesizes the associated probability distribution of each sensor parameter into a joint conditional probability to represent the probability of fire; if the conditional probability exceeds the preset threshold, it is judged that there is a fire risk and the alarm mechanism is triggered; if the conditional probability is lower than the preset threshold, environmental changes are continued to be monitored without triggering an alarm.

[0086] In specific implementation, the preset thresholds are set based on historical data and expert experience to ensure that fire risks can be accurately judged under different environmental conditions;

[0087] The logical operation process of the multimodal data fusion method can be described as follows: First, the system receives the monitoring data from each sensor through the interface and stores it in the corresponding data structure. Then, the system determines whether the data of each sensor is abnormal based on the preset threshold. If the data of any sensor is abnormal, the weighted average function is called to combine the data of other sensors to calculate the preliminary fire risk assessment value. If the preliminary assessment value exceeds the preset threshold, the conditional probability fusion mechanism function is called to calculate the joint probability distribution of each sensor parameter and evaluate the probability of fire. The assessment result is compared with the preset alarm threshold through a comparison operation. If it exceeds the threshold, the alarm mechanism is triggered, otherwise it continues to monitor environmental changes. The entire logical operation process is implemented through loops and conditional judgment statements to ensure that the system can assess the fire risk in real time and accurately, and take corresponding alarm measures.

[0088] Compared with the existing technology, the multimodal data fusion method integrates the monitoring data of multiple sensors to achieve a comprehensive assessment of fire risks, thereby improving the accuracy and reliability of the system. Secondly, the weighted average method and conditional probability fusion mechanism are used to fully consider the correlation between sensor data and the possibility of fire under different conditions, further improving the assessment accuracy of the system. In addition, this method can also dynamically adjust the threshold according to environmental changes, avoiding false alarms and missed alarms caused by environmental interference.

[0089] For further information, see Figure 3 , the working method of the environmental correction algorithm is:

[0090] Step 401: pre-process the smoke density data, remove noise and outliers through Kalman filter, and the pre-processed smoke density data is recorded as C' smoke Step 402: Extract smoke concentration C from historical data smoke , ambient humidity H and the characteristic vector F of air quality index AQI = [C smoke ,H,AQI]; Step 403, calculate the posterior probability P(C smoke |H, AQI); the posterior probability calculation formula of the Bayesian network is:

[0091]

[0092] In formula (1), P(C smoke |H,AQI) indicates the smoke concentration C smoke Under the condition of ambient humidity H and air quality index AQI, the posterior probability P(H,AQI|C smoke ) is the conditional probability, indicating that given the smoke concentration C smoke The joint probability of ambient humidity H and air quality index AQI undersmoke ) represents the smoke density C smoke The prior probability of the ambient humidity H and the air quality index AQI is P(H, AQI); Step 404, the corrected smoke concentration data C" smoke Calculated by the following formula:

[0093]

[0094] In formula (2), C” smoke Indicates the corrected smoke density data; a indicates the correction factor, which is used to adjust the correction intensity; Indicates the smoke concentration C under current environmental conditions. smoke Step 405, according to the real-time changes of ambient humidity H and air quality index AQI, dynamically adjust the smoke concentration threshold T smoke , the adjustment formula is:

[0095] T' smoke =T smoke +k 1 (HT H )+k 2 (AQI-T AQI )-1 (3)

[0096] In formula (3), T' smoke Indicates the smoke density threshold after dynamic adjustment; k 1 Represents the humidity adjustment coefficient, which is used to adjust the effect of humidity on the threshold; k 2 Represents the air quality adjustment factor, which is used to adjust the impact of air quality on the threshold; T H Indicates the preset ambient humidity threshold; T AQI represents the preset air quality threshold; Step 406, the corrected smoke concentration data C" smoke and the dynamically adjusted threshold T' smoke Compare; if C" smoke ≤T' smoke , it is judged as a false alarm and the system continues to monitor; otherwise, it is judged as a real fire and the alarm mechanism is triggered.

[0097] In practical applications, the working mode and structural characteristics of the environmental correction algorithm are as follows: First, the system collects data such as smoke concentration, ambient humidity and air quality index in real time through the first perception recognition module. Then, the Kalman filter is used to preprocess the smoke concentration data to remove noise and outliers, and the preprocessed smoke concentration data is obtained. Next, the feature vectors of smoke concentration, ambient humidity and air quality index similar to the current environment are extracted from the historical database for subsequent probability calculation. Subsequently, the posterior probability of the smoke concentration data is calculated by combining the preprocessed smoke concentration data and the extracted feature vectors through the constructed Bayesian network. Based on the posterior probability and the current environmental conditions, the corrected smoke concentration data and the dynamically adjusted smoke concentration threshold are calculated by a specific formula. Finally, the corrected smoke concentration data is compared with the dynamically adjusted threshold, and whether it is a false alarm or a real fire is determined according to the comparison result, and corresponding measures are taken. Compared with the prior art, the environmental correction algorithm uses the Bayesian network to calculate the posterior probability, makes full use of the information of historical data and current environmental conditions, and improves the accuracy of fire warning. In addition, by dynamically adjusting the threshold of smoke concentration, it adapts to the changes of different environmental conditions and reduces false alarms caused by environmental interference. Finally, the entire algorithm has a clear structure, rigorous logic, and is easy to implement and maintain, providing a more intelligent and reliable early warning mechanism for the fire alarm system integrated with the home IoT.

[0098] In a specific embodiment, the hardware working environment of the first perception and recognition module mainly includes: sensors and microprocessors for data processing, algorithm operation and decision making; communication modules for interconnection between the system and the home Internet of Things; and alarm devices, such as sound and light alarms, for sounding an alarm when a fire risk is detected.

[0099] Based on the above hardware equipment, a comparative experiment was designed. The experimental environment simulated a home kitchen environment, including interference factors such as cooking fumes, steam, and cigarette smoke. Fire alarm systems using the first perception recognition module (Group A) and the traditional method (Group B) were installed for testing. The traditional method mainly relies on a single smoke sensor for fire warning. When the smoke concentration exceeds the preset threshold, the alarm is triggered. The experimental record table is shown in Table 1:

[0100] Table 1. Comparative experimental record of the first perception recognition module

[0101]

[0102] Through comparative experiments, it can be seen that Group A using the first perception recognition module did not have any false alarms in the cooking smoke scene, while Group B using the traditional method did. This shows that the first perception recognition module effectively improves the accuracy of fire warnings and reduces the false alarm rate through multimodal data fusion methods and environmental correction algorithms, providing more reliable technical support for the fire alarm system integrated with the home Internet of Things.

[0103] The second deep recognition module is used to comprehensively analyze the real-time monitoring environmental parameters output by the first perception recognition module through a fire recognition algorithm based on deep learning, identify fire characteristics, and judge fire and non-fire events; when the second deep recognition module recognizes a fire event, it transmits fire characteristic data to the linkage control module and the cloud management platform module through a wireless communication method; when the second deep recognition module does not recognize a fire event, it dynamically adjusts the recognition sensitivity of each sensor of the first perception recognition module according to environmental changes through a dynamic threshold adjustment algorithm; the working method of the fire recognition algorithm based on deep learning is as follows: step 501, based on the video monitoring data output by the first perception recognition module, the real-time image is extracted through a convolutional neural network; the convolutional neural network extracts feature maps from the original data layer by layer through multi-layer convolution kernels and pooling operations, and the feature maps include flame color, shape, texture and dynamic changes; step 502, based on the feature map output in step 501 and the environmental monitoring data output by the first perception recognition module, the change trend of environmental parameters in the time series when the fire occurs is captured through multi-layer stacked LSTM, and the formula expression is:

[0104]

[0105] In formula (4), represents the hidden state of the fire feature at time step t, which contains the fire-related temporal information extracted from the environmental parameters; represents the environmental parameter input at time step t, including smoke density, ambient temperature, flame intensity, ambient humidity, carbon monoxide and carbon dioxide concentrations; L fire represents the number of LSTM layers, which is used to capture the complex changes of fire characteristics in time series; and Respectively represent the weight matrices from hidden state to hidden state, input to hidden state, and hidden state to forget gate of the i-th layer LSTM; and Respectively represent the bias item of the i-th layer LSTM and the bias item of the final hidden state, which are used to adjust the representation of fire features; r represents the activation function, which is used to control the flow and scaling of information; Step 503, map the output of step 502 to a multi-dimensional classification space through the fully connected layer, each dimension corresponds to a fire level or a category of non-fire events, and calculate the probability distribution of each category through the multi-classification strategy function, and select the category with the highest probability as the final recognition result; the formula expression of the multi-classification strategy function is:

[0106]

[0107] In formula (5), Y fire represents the fire feature map, which contains the discriminative features related to fire extracted from the image data; X img represents the image input, including smoke and flame features; M fire Represents the number of convolutional layers, which is used to increase the depth and feature extraction capability of the model; N fire Represents the number of convolution kernels in the mth convolution layer; represents the weight matrix corresponding to the nth convolution kernel of the mth layer; CON represents a two-dimensional convolution operation, and the input is the image data X img , convolution kernel Step Length and fill Represents the nth convolution kernel of the mth layer; Indicates the step size of the nth convolution operation in the mth layer; Represents the padding of the nth convolution operation in the mth layer; ReLU represents the bias term corresponding to the nth convolution kernel of the mth layer; ReLU represents the linear rectification function, which is used to increase the nonlinearity of the model; Step 504, if the recognition result is a fire event, the fire feature map is mapped to the final fire level through the level decision function according to the fire feature intensity and change trend; the fire feature intensity includes the flame area and smoke density; the change trend includes the rising rate and duration; the formula expression of the level decision function is:

[0108]

[0109] In formula (6), C fire Indicates the final fire grade classification results, including grade one, grade two, and grade three; J fire Indicates the number of classifiers; n jfire represents the weight coefficient of the jth classifier, which is used to adjust the contribution of different classifiers to the final fire grade classification result; softmax is the softmax function, which is used to convert the output of the classifier into a probability distribution; W jfirerepresents the weight matrix of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire Represents the bias term of the j-th classifier; Represents the weight coefficient of the kth fire level, which is used to adjust the contribution of different fire levels to the final result; () k Represents the k-th element of a vector or matrix.

[0110] See also Figure 5 The working method of the dynamic threshold adjustment algorithm for dynamically adjusting the recognition sensitivity of each sensor according to environmental changes is as follows: Step 701, when the second depth recognition module does not recognize the fire event, the dynamic threshold adjustment algorithm obtains the current environmental parameters from the first perception recognition module through the Zigbee protocol, including the smoke concentration C smoke 、Ambient temperature E evn 、Flame intensity I flame , environmental humidity H and air quality index AQI; step 702, obtain historical environmental data through the cloud platform, including historical records of smoke concentration, environmental temperature, flame intensity, environmental humidity and air quality index in the past 24 hours; step 703, based on the historical environmental data obtained in step 702, build an environmental change model through a time series analysis method to predict future environmental change trends; the time series analysis method models the time series data through autoregression, difference and moving average, and captures the change law of environmental parameters through data smoothing and trend extraction; step 704, calculate the threshold adjustment factor of each sensor according to the future environmental parameters predicted by the environmental change model, and the calculation formula is:

[0111]

[0112] In formula (7), represents the threshold adjustment factor of the εth sensor; h ε Indicates the adjustment coefficient; E ε Represents the predicted future environmental parameters; E ε,avgRepresents the average value of historical environmental parameters; Step 705, based on the calculated threshold adjustment factor, dynamically update the recognition sensitivity of each sensor through weighted average. During the adjustment process, ensure the stability of the threshold through threshold stability test; Step 706, divide the historical data into training set and test set, use the training set data to train the model, and use the test set data to verify the performance of the model. If the performance of the model on the test set meets expectations, the threshold adjustment is considered to be effective; otherwise, readjust the threshold; Step 707, apply the verified effective threshold update to each sensor of the first perception and recognition module through the Zigbee protocol.

[0113] In actual applications, the second deep recognition module integrates a deep learning algorithm model, which can receive and process data input from the first perception and recognition module; secondly, the module is equipped with a wireless communication interface for data transmission with the linkage control module and the cloud management platform module; finally, the module also includes a dynamic threshold adjustment algorithm module for real-time monitoring of environmental changes and adjusting sensor sensitivity.

[0114] The specific implementation method includes: the second deep recognition module first receives the data input from the first perception recognition module, and cleans and formats the data through the preprocessing step. Then, the module inputs the processed data into the deep learning algorithm model for feature extraction and fire recognition. During the recognition process, the module will match and judge according to the preset fire feature library. If a fire event is recognized, the fire feature data will be sent to the linkage control module and the cloud management platform module through the wireless communication interface.

[0115] If no fire event is identified, the module will enter the dynamic threshold adjustment phase. In this phase, the module will monitor environmental changes (such as temperature, humidity, etc.) in real time and adjust the sensor sensitivity according to these changes. During the adjustment process, the module will calculate the new threshold and apply it to the sensor of the first perception recognition module to ensure the accuracy and stability of the recognition performance.

[0116] Among them, the deep learning algorithm is based on the construction and training of artificial neural networks (ANN). The artificial neural network consists of multiple layers, including input layer, hidden layer and output layer, each layer contains multiple neurons. These neurons perform linear transformation on the input data through connection weights and biases, and introduce nonlinear factors through activation functions, so that the network can learn and approximate arbitrarily complex functions. In the fire identification task, the deep learning algorithm automatically extracts fire features and builds a fire identification model by learning a large number of fire and non-fire images, smoke concentration, temperature and other data. When new real-time monitoring data is input, the model can process it efficiently and output the results of fire identification. In the fire alarm system, wireless communication technology usually adopts standards such as Wi-Fi, Bluetooth or Zigbee to ensure the reliability and efficiency of data transmission. When a fire event is identified, the second deep identification module will immediately send the fire feature data to the relevant system through wireless communication technology so that timely countermeasures can be taken.

[0117] In fire alarm systems, the dynamic threshold adjustment algorithm is mainly used in practical applications to dynamically adjust the threshold of the sensor according to environmental changes. The implementation of this algorithm usually relies on real-time monitoring data and changes in environmental parameters. For example, when the ambient temperature increases, the sensitivity of the smoke sensor may decrease, so the threshold needs to be increased to avoid false alarms. Conversely, when the ambient temperature decreases, the threshold may need to be lowered to increase sensitivity. The dynamic threshold adjustment algorithm monitors these changes in real time and automatically adjusts the threshold to adapt to environmental changes, thereby ensuring that the sensor maintains the best recognition performance in different environments. In practical applications, the algorithm parameters need to be regularly checked and updated to ensure their accuracy and reliability.

[0118] Compared with the existing technology, the second deep recognition module has the following advantages and characteristics: First, through the application of deep learning algorithms, the module can automatically mine fire features and achieve high-precision fire recognition, greatly improving the accuracy and reliability of fire warnings; second, the introduction of wireless communication technology enables fire information to be quickly transmitted to relevant systems, providing a strong guarantee for timely response measures; second, the application of dynamic threshold adjustment algorithms enables sensors to maintain optimal recognition performance in different environments, further improving the adaptability and stability of the fire warning system. Finally, the second deep recognition module realizes the intelligence and adaptability of the system through deep learning models and dynamic threshold adjustment algorithms, and can automatically adjust the recognition strategy according to environmental changes to ensure that the system is always in the best working state.

[0119] The hardware working environment of the fire recognition algorithm based on deep learning applied to the fire alarm system integrated with the home Internet of Things mainly includes: cameras, sensor arrays, embedded processors equipped with deep learning algorithms, wireless communication modules and power supply systems; based on these hardware, the fire recognition algorithm based on deep learning (Group A) and the traditional threshold-based fire detection algorithm (Group B) are compared. Among them, the traditional method uses a threshold-based fire detection algorithm in this experiment. The algorithm determines whether a fire occurs by preset smoke concentration and temperature thresholds. When the smoke concentration or temperature exceeds the preset threshold, the system triggers an alarm. In the experiment, a variety of fire scenes and non-fire scenes were simulated, including smoke generated by kitchen cooking, smoke and flames generated by real fires, etc. Each group of experiments was conducted five times, and the recognition accuracy, false alarm rate and missed alarm rate were recorded. The experimental record table is shown in Table 2:

[0120] Table 2 Algorithm comparison experiment record table

[0121]

[0122] Among them, the traditional method is based on thresholds and does not involve target detection, so there is no IoU indicator; as shown in Table 2, the fire recognition algorithm based on deep learning (Group A) is superior to the traditional threshold-based fire detection algorithm (Group B) in terms of accuracy, precision, recall, F1 score and IoU. The algorithm of Group A can more accurately identify fire scenes, reduce false alarms and missed alarms, and has high target detection accuracy. Although the model complexity of Group A is higher and the hardware requirements are higher, its performance advantages in practical applications make it the preferred solution in the fire alarm system integrated with the home Internet of Things.

[0123] A linkage control module is used to control the home security device through a linkage control strategy based on the fire feature data output by the second deep recognition module; in a specific embodiment, one of the exemplary linkage control strategies of the system is applied, specifically:

[0124] The fire level is judged based on the fire recognition algorithm based on deep learning; if the fire level is level one, the first-level linkage control strategy is triggered, including controlling the water sprinkler fire extinguishing device to start through the relay; opening the automatic door lock through the Zigbee protocol; starting the escape channel indicator light; if the air quality sensor detects that the carbon monoxide concentration exceeds the standard, the ventilation equipment is started through the relay; the smart lighting equipment is controlled through the Zigbee protocol to increase the brightness; the voice alarm information is played through the smart speaker; the home security camera is started through the Zigbee protocol to record the fire scene video; if the fire level is level two, the second-level linkage control strategy is executed, including all the first-level linkage control strategies, and the smart socket is controlled through the Zigbee protocol to cut off the power supply of household appliances; the fire alarm information is sent to the cloud platform through the wireless communication network, and the user is notified through the mobile phone APP or web page, and an emergency rescue phone is called at the same time to request external support; if the fire level is level three, the third-level linkage control strategy is triggered, including all the second-level linkage control strategies, and the smoke purification equipment is started through the relay to reduce the indoor smoke concentration; the automatic sprinkler system is controlled through the relay; the smart window is controlled to open through the Zigbee protocol to increase ventilation.

[0125] The cloud management platform module is used to generate linkage device configuration solution suggestions through linkage logic analysis algorithms based on machine learning, and provide users with linkage device configuration, alarm notification and device status display functions; such as Figure 6As shown: the cloud management platform module includes a user linkage device configuration submodule, an alarm notification submodule and a device status display submodule; the user linkage device configuration submodule includes a device management unit and a scene linkage unit; the device management unit implements device registration and authentication through the OAuth2.0 protocol and implements device status display through the RESTful The API interface realizes the management and update of device configuration; if the user modifies the device configuration, the user linkage device configuration submodule pushes the updated configuration information to the corresponding device through the MQTT protocol; the scene linkage unit provides users with the function of creating and editing linkage scenes through a data interaction method; the scene linkage unit also generates a linkage device configuration plan through a linkage logic analysis algorithm based on machine learning; when a fire event is detected, the alarm notification submodule notifies the user of the alarm information through SMS, phone, email and push notification, and stores the alarm record through a relational database, including the alarm time, location, type and processing result information; the device status display submodule collects the status information of each device in real time through the Zigbee protocol, including online status, working mode, battery power, and displays it; the device status display submodule also uses a time series database to store the historical status records of the device, including the timestamp and specific value of the status change; based on the historical status data, the historical status records of the device are also analyzed through time series analysis to generate a status change trend report and display it.The working method of the linkage logic analysis algorithm based on machine learning is as follows: step 901, collect real-time data from various sensors and devices through the Zigbee protocol and wireless communication network, and pre-process the collected data through data cleaning and normalization; step 902, obtain historical data through the cloud platform, including historical fire event records and equipment operation records; and annotate the historical data through semi-automatic annotation to mark the triggering conditions and corresponding linkage operations of each fire event; the semi-automatic annotation is assisted by machine learning algorithms to generate labeled historical data; step 903, based on the labeled historical data, extract features from the historical data through recursive feature elimination to generate feature vectors, which include environmental parameters and equipment status; the environmental parameters include smoke concentration, ambient temperature, flame intensity, ambient humidity, and air quality index; the equipment status includes sprinkler fire extinguishing devices, automatic door locks, and smart sockets; step 904, train the annotated historical data through a support vector machine to generate fire recognition and Linkage control model; step 905, generate the optimal linkage logic under different fire scenarios through a generative adversarial network; the generative adversarial network is composed of a generator and a discriminator, the generator is used to generate the linkage logic, and the discriminator is used to evaluate the rationality of the generated linkage logic; the generator generates the linkage logic through a deep neural network, including the operation sequence and conditions of the equipment; the discriminator evaluates the rationality of the generated linkage logic through a deep neural network; step 906, based on the linkage logic generated in step 905, define the state, action and reward function through a reinforcement learning algorithm, and optimize the linkage logic; step 907, according to the optimized linkage logic, generate a series of rules through a decision tree algorithm to guide the specific operation of the equipment under different fire levels; the working method of the decision tree algorithm to generate a specific linkage configuration scheme is: r1, convert the linkage logic into rule instructions through rule generation; r2, optimize the decision tree through a pruning method to reduce redundant rules; r3, convert the generated linkage configuration scheme into an executable rule through a rule engine.

[0126] In a specific embodiment, the linkage logic analysis algorithm based on machine learning trains the machine learning model through historical data and labels (such as fire events, user operation records, etc.), so that it can generate the optimal linkage device configuration plan according to the current environmental parameters and user needs. Through the simulation environment and reward mechanism, the training model automatically optimizes the linkage logic in different scenarios to improve the adaptability and robustness of the system. By extracting and selecting features from historical data, effective features that are helpful for model training are generated, such as smoke concentration, ambient temperature, flame intensity, ambient humidity, and air quality.

[0127] The user linkage device configuration submodule generates the optimal linkage device configuration scheme based on the machine learning model, including the selection of linkage devices, the setting of linkage conditions and the execution of linkage actions. It also provides a user-friendly interface, allowing users to manually configure linkage devices, and the system generates the corresponding linkage logic according to the user configuration. For example, if the user wants to automatically close the gas valve and turn on the smoke exhaust fan when a fire is detected, the system will generate the corresponding linkage logic and configure it in the linkage control module. The alarm notification submodule uses message queues (such as MQTT) to achieve real-time transmission of alarm information to ensure the timeliness and reliability of alarm notifications. Alarm notifications are sent to users through multiple channels (such as mobile APP, SMS, email, etc.) to ensure that users can receive alarm information in a timely manner. For example, when the system detects a fire, an alarm notification is sent to the user through the mobile APP, and detailed information is sent through SMS and email. The device status display submodule displays the real-time status of the device through charts and dashboards, such as smoke concentration, ambient temperature, flame intensity, etc. It also records the historical status data of the device, provides historical data query and analysis functions, and helps users understand the long-term operation of the device. For example, users can view the current smoke concentration and ambient temperature, as well as the historical data change trend over the past week through the mobile APP.

[0128] Compared with the existing technology, the cloud management platform module uses a linkage logic analysis algorithm based on machine learning. The system can intelligently generate the optimal linkage device configuration plan and automatically execute the linkage logic, which greatly improves the intelligence and automation level of the system. In addition, the cloud management platform supports users to configure and adjust equipment according to actual needs. At the same time, the system has good scalability and can be continuously upgraded and improved with the development of home Internet of Things. In addition, the system can collect and process sensor data in real time, respond to fire alarms quickly, and help users quickly understand the fire situation and take corresponding measures through accurate equipment status display and alarm notifications. At the same time, the cloud management platform provides intuitive equipment configuration and status display functions through a graphical user interface (GUI), allowing users to easily grasp the safety status of their homes, greatly improving the ease of use and user satisfaction of the system.

[0129] In specific implementation, the linkage logic analysis algorithm based on machine learning relies on a large amount of historical fire data and equipment linkage data. Through data preprocessing and feature extraction, the algorithm can learn the correlation and response mode between different devices when a fire occurs. The data includes fire-related parameters such as smoke concentration, temperature, flame image characteristics, as well as the response time and linkage sequence of various sensors (such as smoke detectors, temperature sensors, cameras, etc.).

[0130] Commonly used machine learning models include decision trees, support vector machines (SVM), random forests, convolutional neural networks (CNN), etc. These models use training data to learn the rules of equipment linkage when a fire occurs. CNN is particularly suitable for image-based fire recognition, which can extract deep features of smoke and flames and improve recognition accuracy and robustness.

[0131] By analyzing the learned data features and model parameters, the algorithm can intelligently infer the type and order of equipment that should be linked when a fire occurs. For example, when a smoke detector detects smoke, the algorithm can infer that it is necessary to link equipment such as opening the exhaust fan and closing the gas valve, and send an alarm notification to the user. In implementation, the algorithm needs to be able to process sensor data in real time and respond quickly when a fire occurs. At the same time, the algorithm also needs to have a certain degree of adaptability and be able to automatically adjust the linkage logic according to changes in the home environment and updates to the equipment.

[0132] In actual applications, the linkage logic analysis algorithm based on machine learning works as follows: First, data from various sensors, including smoke concentration, temperature, flame images, etc., are collected through the home Internet of Things. The data undergoes preprocessing steps such as cleaning, denoising, and formatting to improve the quality and consistency of the data. Then, the preprocessed data is used to train the machine learning model so that it can learn the rules of device linkage when a fire occurs. The model performance is optimized through cross-validation, parameter adjustment, and other methods to improve the accuracy and robustness of recognition. When a fire occurs, the algorithm quickly analyzes the type and order of devices that should be linked based on real-time sensor data. The system displays linkage device configuration suggestions to the user through the GUI and allows the user to make manual adjustments or confirmations. After the user confirms the linkage device configuration, the system sends linkage instructions to the relevant devices through the Internet of Things technology. At the same time, the system sends fire alarm notifications to users through SMS, APP push, etc., to ensure that users can quickly understand the fire situation and take corresponding measures.

[0133] Compared with the existing technology, through machine learning algorithms, the system can intelligently analyze the equipment linkage rules when a fire occurs, and automatically generate the optimal linkage equipment configuration plan. It reduces the possibility of manual intervention and configuration errors, and improves the intelligence and automation level of the system. In addition, the machine learning algorithm can extract the deep characteristics of smoke and flames, and improve the accuracy and robustness of fire identification. At the same time, the algorithm can also automatically adjust the linkage logic according to changes in the home environment and equipment updates to meet the needs of different scenarios. Secondly, the system can process sensor data in real time and respond quickly when a fire occurs. This improves the timeliness and accuracy of fire alarms and helps reduce losses caused by fires. In addition, the system displays linkage equipment configuration suggestions to users through the GUI, allowing users to easily grasp the safety status of their homes.

[0134] In the specific implementation, the hardware operating environment of the linkage logic analysis algorithm based on machine learning mainly includes: sensor network, Internet of Things gateway, cloud management platform server, linkage equipment: including smart exhaust fans, automatic gas closing valves, sound and light alarms, etc., which perform corresponding operations according to the linkage instructions generated by the cloud management platform; and user terminals: such as smart phones, tablets, etc., with remote monitoring and alarm notification functions.

[0135] Based on the above hardware environment, the linkage logic analysis algorithm based on machine learning (Group A) and the traditional method (Group B) were used for comparison. During the experiment, by simulating different types of fire scenes (such as smoldering fire, open fire, etc.), the fire feature recognition rate, linkage equipment response time, fire spread control effect and system false alarm rate of the two methods were recorded. Each group of experiments was conducted five times, and the average value was taken as the final result. During the experiment, all equipment was ensured to be in normal working condition, and the experimental conditions (such as fire scale, sensor layout, etc.) were kept consistent. Among them, the traditional method mainly relies on threshold judgment and simple logical rules, combined with the single or combined use of fire detectors (such as smoke detectors, temperature sensors). For example, when the smoke concentration of the smoke detector reaches or exceeds the preset threshold, the system will trigger an alarm and link the gas valve to close, the exhaust fan and other equipment according to the preset logical rules. The experimental record table is shown in Table 3:

[0136] Table 3. Comparison experiment table of linkage logic algorithms

[0137]

[0138] Through comparative experiments, it can be seen that the linkage logic analysis algorithm based on machine learning (Group A) is significantly superior to the traditional method (Group B) in terms of fire feature recognition rate, linkage equipment response time, fire spread control effect and system false alarm rate. The fire feature recognition rate of Group A is as high as over 98%, the linkage equipment response time is about 3 seconds on average, the fire spread control effect is over 95%, and there are no false alarms. The fire feature recognition rate of Group B is only about 86%, the linkage equipment response time is about 4.5 seconds on average, the fire spread control effect is only about 80%, and the false alarm rate is high. Therefore, the linkage logic analysis algorithm based on machine learning has higher accuracy and reliability in practical applications, can better identify fire features, respond quickly and control fire spread, thereby more effectively protecting family safety.

[0139] The system self-check module is used to regularly check the operating status of linkage devices and sensors, and generate an abnormality detection report to be transmitted to the cloud management platform module. Figure 4As shown, the working method of the system self-check module is: regularly triggering self-check tasks through a scheduled task scheduling mechanism to collect real-time status data of each linkage device and sensor, including the online status, working mode, battery power, and communication status of the device; based on the collected real-time status data, analyzing through anomaly detection algorithms to detect abnormal status of the device; if no abnormal status is detected, ending this round of self-check; if an abnormal status is detected, classifying through a support vector machine to identify the type of abnormality, including communication failure, low battery, and sensor failure; the classified abnormal status generates an abnormal detection report through a report generation module, including the detected abnormal status, abnormal type, occurrence time, and device ID; the generated abnormal detection report is transmitted to the cloud management platform module through the MQTT protocol, and the cloud management platform module automatically handles the abnormality according to preset rules through a rule engine, including sending maintenance notifications, restarting the device, and adjusting device parameters; and notifying the user of the abnormal situation and processing results through the email communication protocol.

[0140] In the specific implementation, the system self-check module works in the fire alarm system integrated with the home Internet of Things as follows: the system self-check module sets the scheduled task through the Cron expression and triggers the self-check task regularly. The task scheduler (such as Quartz, cron) manages the scheduled task to ensure that the self-check task can be executed on time. After the self-check task is triggered, the system collects real-time status data from each linkage device and sensor through the Zigbee protocol, including the online status, working mode, battery power, communication status, etc. of the device. The collected data is transmitted to the central processing unit through the Wi-Fi module. The central processing unit pre-processes the collected real-time status data, such as data cleaning and normalization, to ensure the consistency and comparability of the data. The pre-processed data is detected by the isolation forest algorithm to identify the abnormal state of the device. The isolation forest algorithm isolates data points in different leaf nodes by constructing multiple random decision trees, and abnormal data points are usually isolated faster. Then, the detected abnormal state is classified by the support vector machine to identify the abnormal type, such as communication failure, low battery, sensor failure, etc. The support vector machine classifies abnormal states into different categories by learning the characteristics of historical abnormal data. The classified abnormal states generate an abnormal detection report through the report generation module, including the detected abnormal state, abnormal type, occurrence time, device ID and other information. The generated anomaly detection report is transmitted to the cloud management platform module through the MQTT protocol to ensure efficient and reliable transmission of the report. The cloud management platform module automatically handles exceptions according to preset rules through a rule engine (such as Drools) and generates corresponding processing instructions, such as sending maintenance notifications, restarting devices, adjusting device parameters, etc. The user is informed of the abnormal situation and processing results through the email communication protocol (such as SMTP) to ensure that the user can understand and handle the exception in a timely manner.

[0141] In the specific implementation, through the timed task scheduling mechanism, the system self-check module can trigger the self-check task regularly to ensure that the system can automatically and periodically perform self-checks. This helps to timely discover and handle equipment failures and improve the reliability of the system. Compared with traditional manual inspections, regular self-checks can reduce human intervention and increase the frequency and efficiency of inspections. In addition, through the isolation forest algorithm and data preprocessing technology, the system self-check module can efficiently detect abnormal states of equipment and reduce false positives and false negatives. Through support vector machine classification, the system can accurately identify abnormal types, such as communication failures, low battery, sensor failure, etc., to provide a clear direction for subsequent processing. Compared with traditional threshold judgment methods, machine learning-based anomaly detection and classification methods can more accurately identify and classify anomalies and improve the intelligence level of the system. By analyzing historical data and real-time status data, the system self-check module can generate data-based anomaly detection reports to improve the scientificity and accuracy of decision-making. Compared with traditional experience-based inspection methods, data-driven decisions can better reflect the actual situation and improve the reliability and effectiveness of the system.

[0142] In implementation, a linkage control method for a fire alarm system integrated with a home Internet of Things includes the following steps:

[0143] S1. After the system is powered on, all linkage devices are initialized through the system self-check module, and the working status of each device is checked. If the device is working normally, it will proceed to the next step; if an abnormality is detected in the device, the fault information will be transmitted to the cloud platform through the wireless communication network to notify the user;

[0144] S2, receiving the threshold judgment result and sensor data output by the first perception recognition module through the wireless communication protocol Zigbee; if the threshold exceeds the set range, proceed to the next step; if not, continue monitoring and remain in standby state;

[0145] S3, transmit the fire signal to the cloud platform through the wireless communication network to give early warning notification to the user; and use the multimodal data fusion method to comprehensively judge the smoke concentration, temperature, flame, humidity and air quality parameters. If the smoke concentration exceeds the preset threshold and the temperature rise range exceeds the preset threshold, start the water sprinkler fire extinguishing device; if not, proceed to the next step;

[0146] S4. Call video analysis to identify fire features through a deep learning-based fire recognition algorithm to determine whether it is a fire event or a non-fire event. If it is confirmed to be a fire event, proceed to the next step; if not, continue monitoring and remain on standby;

[0147] S5, control the water sprinkler fire extinguishing device to start through the relay, open the automatic door lock through the Zigbee module, start the escape channel indicator light, and if the air quality sensor detects that the carbon monoxide concentration exceeds the standard, start the ventilation equipment through the relay to reduce the indoor harmful gas concentration, and cut off the power supply of household appliances through the smart socket;

[0148] S6. Dynamically adjust the linkage strategy according to the development of the fire; if the fire continues to expand, increase the water spraying volume of the sprinkler fire extinguishing device, and search and notify nearby smart smoke alarms through the wireless communication protocol Zigbee to form a regional linkage; if the fire is under control, gradually reduce the water spraying volume;

[0149] S7. The alarm information is transmitted to the cloud platform through the wireless communication network, and the user is notified through the mobile phone APP or the web page; the user checks the status of the joint control device through the cloud management platform module, receives the alarm information, and remotely controls the joint control device; if the user confirms that it is a false alarm, the linkage control is cancelled through the application interaction; if not, the linkage device continues to be started;

[0150] S8. Regularly check the working status of all linked devices and record the operating data of the devices; if a device failure or communication interruption is found, try to restart the failed device or re-establish the communication connection through the system self-check module, and send the fault information to the cloud platform through the wireless communication network to notify the user; if the failed device cannot be repaired automatically, notify the user through the cloud platform to perform manual intervention.

[0151] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that these specific embodiments are only illustrative, and those skilled in the art may omit, replace, and change the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, merging the above method steps so as to perform substantially the same functions in substantially the same manner to achieve substantially the same results is within the scope of the present invention. Therefore, the scope of the present invention is limited only by the appended claims.

Claims

1. A fire alarm system integrated with home Internet of Things; characterized by: The system comprises: a first perception and recognition module, a second depth recognition module, a linkage control module, a cloud management platform module and a system self-check module; The first perception and recognition module is used to monitor environmental parameters in real time, and to make preliminary judgments and early warnings on fire situations through a multimodal data fusion method, and to make preliminary misjudgment corrections through an environmental correction algorithm; The second deep recognition module is used to comprehensively analyze the real-time monitoring environmental parameters output by the first perception recognition module through a fire recognition algorithm based on deep learning, identify fire characteristics, and judge fire and non-fire events; when the second deep recognition module identifies a fire event, it transmits the fire characteristic data to the linkage control module and the cloud management platform module through a wireless communication method; when the second deep recognition module does not identify a fire event, it dynamically adjusts the recognition sensitivity of each sensor of the first perception recognition module according to environmental changes through a dynamic threshold adjustment algorithm; The linkage control module is used to control the home security device through a linkage control strategy based on the fire feature data output by the second deep recognition module; The cloud management platform module is used to generate linkage device configuration solution suggestions through a linkage logic analysis algorithm based on machine learning, and provide users with linkage device configuration, alarm notification and device status display functions; The system self-check module is used to regularly check the operating status of linkage devices and sensors, and generate an abnormality detection report and transmit it to the cloud management platform module.

2. A home Internet of Things integrated fire alarm system according to claim 1, characterized in that: The first perception and recognition module detects smoke concentration through a smoke sensor, measures ambient temperature through a temperature sensor, detects infrared radiation of the flame through a flame sensor, monitors ambient humidity through a humidity sensor, detects carbon monoxide and carbon dioxide concentrations through an air quality sensor, and captures video images through a camera; the first perception and recognition module determines the ambient humidity and air quality when the smoke concentration, ambient temperature or flame intensity exceeds a preset threshold; if the ambient humidity and air quality data exceed the preset threshold range, a multimodal data fusion method is used to comprehensively judge it as a preliminary fire warning, and a control signal is transmitted to the cloud management platform module for early warning notification, and smoke concentration, ambient temperature, flame intensity, ambient humidity, carbon monoxide and carbon dioxide concentrations and video image data are transmitted to the second deep recognition module; when the smoke concentration exceeds the preset threshold, if the ambient humidity and air quality data do not exceed the preset threshold range, the smoke sensor data is corrected through an environmental correction algorithm to reduce false alarms, and monitoring continues.

3. A home Internet of Things integrated fire alarm system according to claim 2, characterized in that: The working principle of the multimodal data fusion method is as follows: receiving monitoring parameters of smoke sensors, temperature sensors, flame sensors, humidity sensors and air quality sensors; if any sensor data among the smoke sensors, temperature sensors and flame sensors exceeds a preset threshold, a preliminary fire judgment is made by combining the monitoring parameters of other sensors through the weighted average method, and if yes, further analysis is made through the conditional probability fusion mechanism, and if no, monitoring is continued; if multiple sensor data among the smoke sensors, temperature sensors and flame sensors exceed the preset threshold, the joint probability distribution of smoke concentration, ambient temperature and flame intensity parameters under specific conditions is calculated through the conditional probability fusion mechanism to evaluate the probability of fire occurrence. The conditional probability fusion mechanism combines sensor information according to the probability principle through the multi-Bayesian estimation method, and synthesizes the associated probability distribution of each sensor parameter into a joint conditional probability, which is used to represent the probability of fire occurrence. If the conditional probability exceeds the preset threshold, it is judged that there is a fire risk and the alarm mechanism is triggered; if the conditional probability is lower than the preset threshold, environmental changes are continued to be monitored without triggering an alarm.

4. A fire alarm system integrated with home Internet of Things according to claim 1, characterized in that The working method of the environmental correction algorithm is: Step 401: pre-process the smoke density data, remove noise and outliers through Kalman filter, and the pre-processed smoke density data is recorded as C' smoke ; Step 402: Extract smoke concentration C from historical data smoke , ambient humidity H and the characteristic vector F of air quality index AQI = [C smoke ,H,AQI]; Step 403: Calculate the posterior probability P(C smoke |H, AQI); the posterior probability calculation formula of the Bayesian network is: In formula (1), P(C smoke |H, AQI) means the smoke concentration C smoke Under the condition of ambient humidity H and air quality index AQI, the posterior probability P(H,AQI|C smoke ) is the conditional probability, indicating that given the smoke concentration C smoke The joint probability of ambient humidity H and air quality index AQI under smoke ) represents the smoke density C smoke The prior probability of; P(H,AQI) represents the joint probability of ambient humidity H and air quality index AQI; Step 404: Corrected smoke density data C" smoke Calculated by the following formula: In formula (2), C” smoke Indicates the corrected smoke density data; a indicates the correction factor, which is used to adjust the correction intensity; Indicates the smoke concentration C under current environmental conditions. smoke Credibility; Step 405: dynamically adjust the smoke concentration threshold T according to the real-time changes of the ambient humidity H and the air quality index AQI smoke , the adjustment formula is: T’ smoke =T smoke +k1(H-T H )+k2(AQI-T AQI )-1 (3) In formula (3), T' smoke represents the smoke concentration threshold after dynamic adjustment; k1 represents the humidity adjustment coefficient, which is used to adjust the effect of humidity on the threshold; k2 represents the air quality adjustment coefficient, which is used to adjust the effect of air quality on the threshold; T H Indicates the preset ambient humidity threshold; T AQI Indicates the preset air quality threshold; Step 406: Corrected smoke density data C" smoke and the dynamically adjusted threshold T' smoke Compare; if C" smoke ≤T' smoke , it is judged as a false alarm and the system continues to monitor; Otherwise, it is judged as a real fire and the alarm mechanism is triggered.

5. The fire alarm system integrated with home Internet of Things according to claim 1, characterized in that: The working method of the fire identification algorithm based on deep learning is: Step 501: Based on the video monitoring data output by the first perception recognition module, feature extraction is performed on the real-time image through a convolutional neural network; the convolutional neural network extracts feature maps layer by layer from the original data through multi-layer convolution kernels and pooling operations, and the feature maps include flame color, shape, texture, and dynamic changes; Step 502: Based on the feature graph output in step 501 and the environmental monitoring data output by the first perception and recognition module, a multi-layer stacked LSTM is used to capture the changing trend of environmental parameters in the time series when a fire occurs. The formula expression is: In formula (4), represents the hidden state of the fire feature at time step t, which contains the fire-related temporal information extracted from the environmental parameters; represents the environmental parameter input at time step t, including smoke density, ambient temperature, flame intensity, ambient humidity, carbon monoxide and carbon dioxide concentrations; L fire represents the number of LSTM layers, which is used to capture the complex changes of fire characteristics in time series; and Respectively represent the weight matrices from hidden state to hidden state, input to hidden state, and hidden state to forget gate of the i-th layer LSTM; and They represent the bias term of the i-th layer LSTM and the bias term of the final hidden state, respectively, which are used to adjust the representation of fire characteristics; r represents the activation function, which is used to control the flow and scaling of information; Step 503: Map the output of step 502 to a multi-dimensional classification space through a fully connected layer, where each dimension corresponds to a fire level or a category of non-fire events, and calculate the probability distribution of each category through a multi-classification strategy function, and select the category with the highest probability as the final recognition result; the formula expression of the multi-classification strategy function is: In formula (5), Y fire represents the fire feature map, which contains the discriminative features related to fire extracted from the image data; X img Represents image input, including smoke and fire features; M fire Represents the number of convolutional layers, which is used to increase the depth and feature extraction capability of the model; N fire Represents the number of convolution kernels in the mth convolution layer; represents the weight matrix corresponding to the nth convolution kernel of the mth layer; CON represents a two-dimensional convolution operation, and the input is the image data X img , convolution kernel Step Length and fill Represents the nth convolution kernel of the mth layer; Indicates the step size of the nth convolution operation in the mth layer; Represents the padding of the nth convolution operation in the mth layer; Represents the bias term corresponding to the nth convolution kernel of the mth layer; ReLU represents the linear rectification function, which is used to increase the nonlinearity of the model; Step 504: If the identification result is a fire event, the fire feature map is mapped to the final fire level through a level decision function according to the fire feature intensity and change trend; the fire feature intensity includes flame area and smoke density; the change trend includes rising rate and duration; the formula expression of the level decision function is: In formula (6), C fire Indicates the final fire grade classification results, including grade one, grade two, and grade three; J fire Indicates the number of classifiers; n fire represents the weight coefficient of the jth classifier, which is used to adjust the contribution of different classifiers to the final fire grade classification result; softmax is the softmax function, which is used to convert the output of the classifier into a probability distribution; W jfire represents the weight matrix of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire represents the bias term of the jth classifier; b jfire Represents the bias term of the j-th classifier; Represents the weight coefficient of the kth fire level, which is used to adjust the contribution of different fire levels to the final result; () k Represents the k-th element of a vector or matrix.

6. A home Internet of Things integrated fire alarm system according to claim 1, characterized in that: The cloud management platform module includes a user linkage device configuration submodule, an alarm notification submodule and a device status display submodule; the user linkage device configuration submodule includes a device management unit and a scene linkage unit; the device management unit implements device registration and authentication through the OAuth 2.0 protocol, and implements device configuration management and update through the RESTful API interface; if the user modifies the device configuration, the user linkage device configuration submodule pushes the updated configuration information to the corresponding device through the MQTT protocol; The scene linkage unit provides users with the function of creating and editing linkage scenes through a data interaction method; the scene linkage unit also generates linkage equipment configuration schemes through a linkage logic analysis algorithm based on machine learning; when a fire event is detected, the alarm notification submodule notifies the user of the alarm information through SMS, phone, email and push notification, and stores the alarm record through a relational database, including the alarm time, location, type and processing result information; The device status display submodule collects the status information of each device in real time through the Zigbee protocol, including online status, working mode, and battery power, and displays it; the device status display submodule also uses a time series database to store the historical status records of the device, including the timestamp and specific value of the status change; based on the historical status data, the historical status records of the device are analyzed through time series analysis, and a status change trend report is generated and displayed.

7. A home Internet of Things integrated fire alarm system according to claim 1, characterized in that: The working method of the dynamic threshold adjustment algorithm to dynamically adjust the recognition sensitivity of each sensor according to environmental changes is: Step 701: When the second depth recognition module does not recognize a fire event, the dynamic threshold adjustment algorithm obtains current environmental parameters, including smoke density C, from the first perception recognition module through the Zigbee protocol. smoke 、Ambient temperature E evn 、Flame intensity I flame , ambient humidity H and air quality index AQI; Step 702: Obtain historical environmental data through the cloud platform, including historical records of smoke concentration, ambient temperature, flame intensity, ambient humidity, and air quality index in the past 24 hours; Step 703: Based on the historical environmental data obtained in step 702, an environmental change model is constructed by a time series analysis method to predict future environmental change trends; the time series analysis method models the time series data through three parts: autoregression, difference, and moving average, and captures the change pattern of environmental parameters through data smoothing and trend extraction; Step 704: Calculate the threshold adjustment factor of each sensor according to the future environmental parameters predicted by the environmental change model. The calculation formula is: In formula (7), represents the threshold adjustment factor of the εth sensor; h ε Indicates the adjustment coefficient; E ε Represents the predicted future environmental parameters; E ε,avg Represents the average value of historical environmental parameters; Step 705: Based on the calculated threshold adjustment factor, the recognition sensitivity of each sensor is dynamically updated by weighted average. During the adjustment process, the stability of the threshold is ensured by threshold stability test; Step 706: divide the historical data into a training set and a test set, use the training set data to train the model, and use the test set data to verify the performance of the model. If the performance of the model on the test set meets expectations, it is considered that the threshold adjustment is effective; otherwise, readjust the threshold; Step 707: Apply the verified effective threshold update to each sensor of the first perception and recognition module through the Zigbee protocol.

8. The home Internet of Things integrated fire alarm system according to claim 1, characterized in that: The working method of the system self-check module is: regularly triggering self-check tasks through a scheduled task scheduling mechanism to collect real-time status data of each linkage device and sensor, including the online status, working mode, battery power, and communication status of the device; based on the collected real-time status data, analyzing it through an anomaly detection algorithm to detect abnormal status of the device; If no abnormal state is detected, this round of self-checking ends; if an abnormal state is detected, it is classified through a support vector machine to identify the type of abnormality, including communication failure, low battery, and sensor failure; the classified abnormal state generates an abnormal detection report through a report generation module, including the detected abnormal state, abnormality type, occurrence time, and device ID; the generated abnormal detection report is transmitted to the cloud management platform module through the MQTT protocol, and the cloud management platform module automatically handles the exception according to preset rules through a rule engine, including sending maintenance notifications, restarting the device, and adjusting device parameters; and the user is informed of the abnormal situation and processing results through the email communication protocol.

9. A home Internet of Things integrated fire alarm system according to claim 1, characterized in that: The working method of the linkage logic analysis algorithm based on machine learning is: Step 901: collect real-time data from various sensors and devices through the Zigbee protocol and wireless communication network, and pre-process the collected data through data cleaning and normalization; Step 902: historical data is obtained through the cloud platform, including historical fire event records and equipment operation records; the historical data is annotated through semi-automatic annotation to mark the triggering conditions and corresponding linkage operations of each fire event; the semi-automatic annotation is assisted by a machine learning algorithm to generate historical data with labels; Step 903: Based on the historical data with labels, feature extraction is performed on the historical data by recursive feature elimination to generate a feature vector, where the feature vector includes environmental parameters and device status; the environmental parameters include smoke concentration, ambient temperature, flame intensity, ambient humidity, and air quality index; the device status includes a sprinkler, automatic door lock, and smart socket; Step 904: Train the annotated historical data through a support vector machine to generate a fire identification and linkage control model; Step 905: Generate optimal linkage logic under different fire scenarios through a generative adversarial network; the generative adversarial network is composed of a generator and a discriminator, the generator is used to generate linkage logic, and the discriminator is used to evaluate the rationality of the generated linkage logic; the generator generates linkage logic through a deep neural network, including the operation sequence and conditions of the equipment; the discriminator evaluates the rationality of the generated linkage logic through a deep neural network; Step 906: Based on the linkage logic generated in step 905, define the state, action and reward function through a reinforcement learning algorithm to optimize the linkage logic; Step 907: Based on the optimized linkage logic, a series of rules are generated through a decision tree algorithm to guide the specific operation of the equipment under different fire levels; the working method of the decision tree algorithm to generate a specific linkage configuration scheme is: r1. Convert linkage logic into rule instructions through rule generation; r2. Optimize the decision tree through pruning method to reduce redundant rules; r3. Convert the generated linkage configuration plan into executable rules through the rule engine.

10. A linkage control method for a fire alarm system integrated with a home Internet of Things, characterized in that: A home Internet of Things integrated fire alarm system applied to any one of claims 1 to 9 comprises the following steps: S1. After the system is powered on, all linkage devices are initialized through the system self-check module, and the working status of each device is checked. If the device is working normally, it will proceed to the next step; if an abnormality is detected in the device, the fault information will be transmitted to the cloud platform through the wireless communication network to notify the user; S2, receiving the threshold judgment result and sensor data output by the first perception recognition module through the wireless communication protocol Zigbee; if the threshold exceeds the set range, proceed to the next step; if not, continue monitoring and remain in standby state; S3, transmit the fire signal to the cloud platform through the wireless communication network to give early warning notification to the user; and use the multimodal data fusion method to comprehensively judge the smoke concentration, temperature, flame, humidity and air quality parameters. If the smoke concentration exceeds the preset threshold and the temperature rise range exceeds the preset threshold, start the water sprinkler fire extinguishing device; if not, proceed to the next step; S4. Call video analysis to identify fire features through a deep learning-based fire recognition algorithm to determine whether it is a fire event or a non-fire event. If it is confirmed to be a fire event, proceed to the next step; if not, continue monitoring and remain on standby; S5, control the water sprinkler fire extinguishing device to start through the relay, open the automatic door lock through the Zigbee module, start the escape channel indicator light, and if the air quality sensor detects that the carbon monoxide concentration exceeds the standard, start the ventilation equipment through the relay to reduce the indoor harmful gas concentration, and cut off the power supply of household appliances through the smart socket; S6. Dynamically adjust the linkage strategy according to the development of the fire; if the fire continues to expand, increase the water spraying volume of the sprinkler fire extinguishing device, and search and notify nearby smart smoke alarms through the wireless communication protocol Zigbee to form a regional linkage; if the fire is under control, gradually reduce the water spraying volume; S7. The alarm information is transmitted to the cloud platform through the wireless communication network, and the user is notified through the mobile phone APP or the web page; the user checks the status of the joint control device through the cloud management platform module, receives the alarm information, and remotely controls the joint control device; if the user confirms that it is a false alarm, the linkage control is cancelled through the application interaction; if not, the linkage device continues to be started; S8. Regularly check the working status of all linked devices and record the operating data of the devices; if a device failure or communication interruption is found, try to restart the failed device or re-establish the communication connection through the system self-check module, and send the fault information to the cloud platform through the wireless communication network to notify the user; if the failed device cannot be repaired automatically, notify the user through the cloud platform to perform manual intervention.

Images