Household intelligent alarm system

Abstract

The invention relates to the technical field of household intelligent alarm systems, and discloses a household intelligent alarm system, which comprises an environment sensing module, a state recognition module, a strategy generation module and an abnormity handling module. The environment sensing module integrates multi-source parameters such as temperature, smoke concentration, human body infrared signals and door and window opening and closing states through multi-mode detection, and an associated data space containing a normal state layer, an early warning state layer and an emergency state layer is constructed. The state recognition module extracts feature sets such as a parameter mean value, a change rate and a sudden change amplitude by adopting a multi-scene sampling and time sequence feature analysis model; the strategy generation module extracts a common rule from different scene and time period strategies, and constructs a rule matching engine to generate an identification strategy adapted to a new scene; the exception handling module monitors parameters in real time and triggers preset strategies such as parameter adjustment, early warning notification and emergency linkage. According to the system, comprehensive monitoring of the family environment, scene self-adaption and intelligent abnormity handling are achieved, and the intelligent level of safety protection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of home intelligent alarm systems, and in particular to a home intelligent alarm system. Background Art

[0002] As people's living standards improve and their awareness of home security grows, traditional home alarm systems are no longer able to meet the modern family's demand for safe, intelligent, and personalized protection. Traditional alarm systems often monitor only a single parameter, such as smoke or temperature, and are unable to comprehensively assess the home environment by integrating multiple sensory parameters. This makes them prone to false alarms and missed alarms in complex home scenarios. For example, a traditional smoke alarm may falsely trigger when a small amount of smoke is produced by cooking in the kitchen. Furthermore, when multiple abnormal conditions occur simultaneously, it may miss alarms due to its inability to comprehensively analyze the correlations between the various parameters.

[0003] Traditional systems lack adaptability to different living scenarios and time periods. The normal ranges of various environmental parameters vary across different home environments, such as during daily activities, nighttime rest, and when away from home. Traditional systems often employ fixed safety thresholds and identification strategies, failing to automatically adjust to changing living environments. For example, the pattern of changes in a person's infrared signal during nighttime rest differs from that during daily activities. If traditional systems fail to distinguish these differences, they may misjudge normal nighttime activity. Furthermore, traditional alarm systems offer limited handling options, typically limited to audible and visual alarms, lacking coordinated response capabilities and adaptive adjustment mechanisms. When an abnormality occurs, they are unable to tailor response measures based on its severity and frequency, nor can they optimize response thresholds based on actual conditions during system operation. This results in low system practicality and reliability. For example, if an abnormality occurs multiple times within a short period of time without causing serious consequences, traditional systems are unable to adjust their handling strategies, potentially resulting in continued unnecessary alarms and user inconvenience.

[0004] In terms of data processing, traditional systems lack effective feature extraction and analysis methods for multi-source sensory data. This makes it difficult to deeply explore the linkage relationships and temporal characteristics between parameters and build accurate environmental state recognition models. This results in low environmental state recognition accuracy and an inability to promptly and accurately identify potential safety hazards. Summary of the Invention

[0005] The purpose of the present invention is to provide a home intelligent alarm system to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a home intelligent alarm system, the system comprising: An environmental perception module, which uses multimodal detection to construct a correlation data space between environmental parameters and alarm trigger conditions, and integrates multi-source perception parameters, including temperature, smoke concentration, human infrared signals, and door and window switch status. The correlation data space includes a normal state layer, a warning state layer, and an emergency state layer. The state recognition module is used to execute the steady-state recognition strategy and the transient recognition strategy respectively. The steady-state recognition strategy uses a multi-scenario sampling method to extract a first feature set; the transient recognition strategy extracts a second feature set by constructing a time series feature analysis model. The first feature set includes parameter mean, rate of change, and duration; the second feature set includes mutation amplitude, rising edge slope, and correlation parameter coupling degree. The strategy generation module is used to extract common rules from the steady-state recognition strategies and transient recognition strategies trained in different scenarios and time periods, build a rule matching engine, and generate recognition strategies adapted to new residential scenarios based on the rule matching engine; The abnormality handling module is used to monitor multi-source perception parameters in real time based on the associated data space, and trigger the preset handling strategy when abnormal parameters are detected.

[0007] Preferably, a data space of association between environmental parameters and alarm triggering conditions is constructed through multimodal detection, and multi-source perception parameters are integrated. The specific steps include: Use synchronous acquisition devices to obtain time series data of environmental parameters and alarm trigger conditions, and perform data alignment and calibration; By analyzing the linkage relationship between parameters, normal state modeling, warning state simulation and emergency state analysis are performed on the associated data space respectively; The real-time multi-source perception parameters of the actual living environment are obtained and input into the associated data space.

[0008] Preferably, the specific steps of the steady-state identification strategy include: The monitoring data is divided using a fixed interval method, where the data segment includes M samples, and each sample is represented by a first feature vector composed of a first feature set; Construct a multi-scenario sampling framework, which includes daily activity mode, nighttime rest mode, and away-from-home mode; The samples were processed using a stratified processing approach; The specific steps of the layered processing method include: S1, calculates the daily characteristics, rest characteristics and leaving home characteristics of the sample based on the multi-scene sampling framework; S2, divides the samples into three groups according to feature stability: high-stable samples, medium-stable samples, and low-stable samples; S3, retains highly stable samples, performs interpolation and reconstruction operations on moderately stable samples, and performs high-probability elimination operations on low-stability samples; S4, based on the daily characteristics, rest characteristics, and leaving home characteristics of the samples, scene clustering is used to screen and select standard samples, among which the samples with the most balanced daily characteristics, rest characteristics, and leaving home characteristics are used as standard samples; S5, when the processing quantity reaches the preset maximum processing quantity, output the current standard sample, and select the sample that best meets the current living scenario according to actual needs, otherwise return to S2.

[0009] Preferably, the interpolation operation uses a linear interpolation method to generate new samples. The operation process of the linear interpolation method is: connecting the parameter curves of two adjacent samples through linear interpolation to generate a smooth transition sample data segment; The reconstruction operation uses the feature dimensionality reduction method to update the sample. The feature dimensionality reduction operation process is: mapping the feature vector of the transition sample to a low-dimensional space, extracting the key components through principal component analysis, and generating a simplified feature vector.

[0010] Preferably, the steady-state identification strategy and the transient identification strategy are converted into a steady-state rule vector and a transient rule vector, respectively, and cascaded to obtain a control rule vector; for the control rule vectors of different scenarios and different time periods, the repetition frequency of each rule in the control rule vector is calculated in sequence; when the repetition frequency is greater than a second preset threshold, the rule is determined to have high universality across different scenarios and time periods, and is regarded as a common law. The extracted common laws are combined to obtain a rule matching engine, which provides a universal recognition framework for new residential scenarios; The rule matching engine provides a general recognition framework for new living scenarios. The specific application methods include: inputting the basic parameters of the new living scenario into the rule matching engine, matching the closest historical scenarios and time periods, extracting the corresponding common rules, and combining the specific parameters of the new scenario to generate an adaptive recognition strategy. The specific parameters include: family member structure, daily routine habits and key protection areas.

[0011] Preferably, the exception handling module monitors the multi-source perception parameters in real time according to the associated data space. When the multi-source perception parameters exceed the preset safety threshold, the alarm system is judged to be operating abnormally and the preset handling strategy is triggered; a statistical period is set to record the number of times the same exception is triggered within the period. If the same exception occurs multiple times within the statistical period and the alarm system is still operating stably, the handling threshold of the exception is adjusted.

[0012] Preferably, the specific steps of the transient identification strategy include: According to the operating state of the alarm system, a state matrix containing the second feature set is constructed, wherein the expression of the state matrix is ​​a multidimensional array containing time step, temperature, smoke concentration, human infrared signal and door and window switch status; A fast computation model is used to construct a response network, which takes a state matrix as input and outputs a response index of a second feature set, wherein the response index represents a dynamic change rate of the feature; Constructing a feature space, where the feature space represents a value range of a response indicator of the second feature set; Construct an evaluation function for the time series feature analysis model; The response network is updated using a gradient optimization method.

[0013] Preferably, a synchronous acquisition device is used to obtain time series data of environmental parameters and alarm triggering conditions, and data alignment and calibration are performed. The specific steps include: Configure independent time series recording units for environmental parameters and alarm trigger conditions respectively; All recording units are started simultaneously through hardware synchronization signals to ensure the consistency of the initial time series; The collected time series data is subjected to offset detection, and the time deviation is corrected by polynomial fitting method to achieve data alignment between parameters.

[0014] Preferably, the preset disposal strategies include: parameter adjustment strategy, early warning notification strategy and emergency linkage strategy, wherein the parameter adjustment strategy is used to automatically adjust the sensor sensitivity and alarm threshold, the early warning notification strategy is used to push mobile phone APP messages and trigger sound and light alarms when an abnormality occurs, and the emergency linkage strategy is used to automatically call the emergency number and close the gas valve in the event of extreme abnormalities.

[0015] Preferably, the evaluation function of the time series feature analysis model adopts a mean square error function.

[0016] Compared with the prior art, the present invention has the following beneficial effects: At the environmental perception level, multimodal detection constructs a linked data space encompassing normal, warning, and emergency state layers. This integrates multi-source sensing parameters such as temperature, smoke concentration, human infrared signals, and door and window status to comprehensively and accurately reflect the home environment. Synchronous data collection devices acquire and align time series data to ensure accuracy and timeliness, laying a solid foundation for subsequent status identification and anomaly handling.

[0017] The state recognition module employs both steady-state and transient recognition strategies, significantly enhancing the system's recognition capabilities. The steady-state recognition strategy uses a multi-scenario sampling approach to extract a primary feature set, including parameter mean, rate of change, and duration, for different modes, such as daily activities, nighttime rest, and being away from home. It then applies a hierarchical processing approach to screen and process the samples, accurately capturing stable features across different scenarios and effectively distinguishing between normal and abnormal states. The transient recognition strategy constructs a time series feature analysis model to extract a secondary feature set, including mutation amplitude, rising edge slope, and associated parameter coupling. This allows for timely detection of parameter mutations, improving the system's response speed.

[0018] The strategy generation module extracts common patterns from steady-state and transient recognition strategies trained across different scenarios and time periods, building a rule-matching engine capable of generating recognition strategies adapted to new residential scenarios. The steady-state and transient recognition strategies are converted into rule vectors and concatenated. By calculating the frequency of rule repetition, the module extracts highly universal common patterns. Combined with scenario-specific parameters such as family structure, daily routines, and key protection areas, the system can quickly adapt to diverse home environments and meet personalized needs.

[0019] The anomaly handling module monitors multi-source sensing parameters in real time based on the associated data space. Upon detecting an anomaly, it triggers pre-defined handling strategies, including parameter adjustment, early warning notification, and emergency linkage. The parameter adjustment strategy automatically adjusts sensor sensitivity and alarm thresholds, improving system adaptability. The early warning notification strategy promptly alerts users through mobile app messages and audible and visual alarms. The emergency linkage strategy automatically dials emergency services and closes the gas valve in the event of an extreme anomaly, minimizing losses. Furthermore, a statistical cycle is established to record the number of anomaly triggers and adjust the handling thresholds, enabling continuous optimization of the system based on actual operating conditions, improving system reliability and accuracy.

[0020] The time series feature analysis model uses a mean square error function as its evaluation function, combined with a gradient optimization method to update the response network, improving the model's accuracy and stability. A feature dimensionality reduction method updates the samples, reducing the data dimension and improving the system's processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a working principle diagram of the home intelligent alarm system of the present invention; Figure 2 Flowcharts constructed for Linked Data Spaces; Figure 3 Flowchart built for the rule matching engine; Figure 4 A flowchart for adjusting the threshold for abnormal handling; Figure 5 A schematic diagram of the disposal strategy types. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0023] See also Figure 1-Figure 5 The present invention relates to a home intelligent alarm system, which includes: an environment perception module, a state recognition module, a strategy generation module and an abnormality handling module. The specific implementation is as follows: The environmental perception module uses multimodal detection to construct a linked data space between environmental parameters and alarm triggering conditions, integrating multi-source perception parameters. These multi-source perception parameters include temperature, smoke concentration, human infrared signals, and door and window switch status. The linked data space is divided into normal state, warning state, and emergency state layers. Specifically, the module uses synchronous acquisition devices to acquire time series data of environmental parameters and alarm triggering conditions. After data alignment and calibration, it analyzes the linkage between parameters and performs normal state modeling, warning state simulation, and emergency state analysis on the linked data space. Finally, the real-time multi-source perception parameters of the actual living environment are input into the linked data space.

[0024] The state recognition module implements both steady-state and transient recognition strategies. The steady-state recognition strategy uses a multi-scenario sampling method to extract a first feature set, which includes parameter mean, rate of change, and duration. The transient recognition strategy extracts a second feature set, including mutation amplitude, rising edge slope, and associated parameter coupling, by constructing a time series feature analysis model. Among them, the steady-state identification strategy divides the monitoring data through a fixed interval method. Each data segment contains M samples, and each sample is represented by a first feature vector composed of a first feature set. A multi-scenario sampling framework including daily activity patterns, nighttime rest patterns, and away-from-home patterns is constructed, and a hierarchical processing method is used to process the samples. The specific steps are: according to the multi-scenario sampling framework, the daily characteristics, rest characteristics, and away-from-home characteristics of the samples are calculated. According to the characteristic stability, the samples are divided into three groups: high stability, medium stability, and low stability. The highly stable samples are retained, interpolation and reconstruction operations are performed on the medium stable samples, and high-probability elimination operations are performed on the low-stable samples. Then, scene clustering is used to screen standard samples according to the sample characteristics, and the samples with the most balanced characteristics are used as standard samples. When the processing quantity reaches the preset maximum processing quantity, the current standard sample is output and the sample that best matches the current living scene is selected. Otherwise, it returns to continue processing. The transient recognition strategy constructs a state matrix containing the second feature set according to the operating status of the alarm system. The matrix is ​​a multidimensional array containing time steps, temperature, smoke concentration, human infrared signals, and door and window switch states. A fast calculation model is used to construct a response network. The state matrix is ​​used as input and output to generate response indicators of the second feature set. A feature space is constructed to represent the value range of the response indicator. An evaluation function of the time series feature analysis model is constructed (using the mean square error function), and a gradient optimization method is used to update the response network.

[0025] The strategy generation module extracts common rules from the steady-state recognition strategies and transient recognition strategies trained in different scenarios and time periods, builds a rule matching engine, and generates recognition strategies adapted to new residential scenarios based on the engine. Specifically, the steady-state and transient recognition strategies are converted into steady-state rule vectors and transient rule vectors and cascaded to obtain a control rule vector. For the control rule vectors of different scenarios and time periods, the repetition frequency of each rule is calculated. When the repetition frequency is greater than the second preset threshold, the rule is judged to have high universality and is regarded as a common rule. The extracted common rules are combined to obtain a rule matching engine, which provides a general recognition framework for new residential scenarios. When applied, the basic parameters of the new scenario are input into the engine, matching the closest historical scenarios and time periods, extracting the corresponding common rules, and combining the new scenario-specific parameters (family member structure, daily routines, and key protection areas) to generate an adapted recognition strategy.

[0026] The anomaly handling module monitors multi-source sensing parameters in real time based on the associated data space. Detecting an anomaly triggers a preset handling strategy. These include parameter adjustment, early warning notification, and emergency linkage strategies. The parameter adjustment strategy automatically adjusts sensor sensitivity and alarm thresholds. The early warning notification strategy pushes a mobile app message and triggers an audible and visual alarm when an anomaly occurs. The emergency linkage strategy automatically dials emergency services and closes the gas valve in the event of an extreme anomaly. A statistical period is also set to record the number of times the same anomaly is triggered. If the same anomaly occurs multiple times within a period and the alarm system remains stable, the handling threshold for that anomaly is adjusted.

[0027] Example 1: This example details the process by which the environmental perception module constructs a correlated data space and integrates multi-source perception parameters. The core of this module is to use multimodal detection to correlate various environmental parameters with alarm triggering conditions, forming a hierarchical data space that provides the foundation for subsequent state recognition and anomaly handling.

[0028] When acquiring time series data, it is necessary to configure independent time series recording units for environmental parameters and alarm trigger conditions. These recording units can be timing modules integrated into the sensor or independent hardware timing devices. Their function is to ensure that each parameter and trigger condition has a corresponding timestamp. For example, the temperature sensor is equipped with a recording unit to record the time of each temperature acquisition in real time; the smoke concentration sensor is also equipped with a recording unit, which is triggered at the hardware level by a synchronization signal with the temperature sensor's recording unit. Specifically, all recording units are started simultaneously through a hardware synchronization signal. The hardware synchronization signal here can be a clock pulse signal sent by the central controller to ensure that the initial time series of all recording units are consistent, avoiding time deviations caused by different startup times.

[0029] After collecting time series data, data alignment and calibration are required. Due to differences in sampling frequency, signal transmission delay, and other factors between different sensors, the collected data may have time offsets. Therefore, offset detection is required for the collected time series data. Correlation algorithms can be used to calculate the time offset between time series with different parameters. For example, by calculating the cross-correlation function of the temperature series and the smoke concentration series, the peak position is found to determine the time offset. After finding the offset, a polynomial fitting method is used to correct the time deviation and achieve data alignment between the parameters. The polynomial fitting method can select the appropriate polynomial order based on different offset conditions to fit and correct the time series, so that the time series with different parameters are aligned on the time axis, ensuring the accuracy of subsequent analysis.

[0030] After completing data alignment and calibration, it's necessary to analyze the linkages between parameters and perform normal state modeling, warning state simulation, and emergency state analysis on the associated data space. Modeling the normal state requires collecting a large amount of environmental parameter data under normal conditions. For example, in daily activity scenarios, the variation ranges and patterns of parameters such as temperature, smoke concentration, human infrared signals, and door and window switch status should be recorded over different time periods. By analyzing this data, the mean, variance, and variation range of each parameter under normal conditions can be determined to construct a normal state model. For example, for temperature parameters, the normal temperature range under daily activity patterns may be between 18 and 26 degrees Celsius, with a relatively slow rate of change.

[0031] Early warning simulation simulates the process by which parameters gradually deviate from their normal state and approach alarm thresholds. By setting up different simulation scenarios, we analyze the interconnected changes between parameters. For example, we simulate the slow rise in temperature caused by prolonged heating of food in a kitchen while simultaneously monitoring changes in smoke concentration. We analyze the interconnected relationship between temperature and smoke concentration to determine the parameter change patterns and thresholds during the early warning state.

[0032] Emergency analysis targets scenarios where sudden and dramatic changes in parameters could trigger dangerous situations. For example, during a fire, when temperatures and smoke concentrations rise sharply, the analysis analyzes the sudden change amplitude, rising slope, and coupling of related parameters to determine the triggering conditions and response strategies for the emergency.

[0033] Finally, real-time multi-source perception parameters of the actual living environment are acquired and input into the linked data space. Sensors in the actual living environment collect real-time parameters such as temperature, smoke concentration, human infrared signals, and door and window status. These parameters are sent to the environmental perception module via the data transmission module. The module compares and analyzes these real-time parameters with the normal state layer, warning state layer, and emergency state layer in the linked data space to determine the current state of the environment and provide input data for the subsequent state recognition module.

[0034] Throughout this process, the accuracy and stability of the synchronized data collection device are crucial, directly impacting the accuracy of time series data. Data alignment and calibration methods need to be selected and optimized based on actual conditions to ensure temporal consistency across different parameters. The construction of the linked data space must fully consider the changing patterns and interconnected relationships of parameters in various scenarios. Through extensive data collection and analysis, the accuracy and reliability of the model can be improved. Furthermore, to adapt to diverse living environments, the linked data space must possess a degree of self-learning and adaptive capabilities, enabling continuous updating and optimization of model parameters based on actual operational data. For example, if living habits change, the normal state model can automatically adjust to the new situation.

[0035] Furthermore, the integration of multi-source sensing parameters requires consideration of complementarity and redundancy. For example, infrared signals from human bodies and the status of doors and windows can be used to verify whether a resident is home, improving system reliability. During data processing, the impact of noise must also be considered, employing appropriate filtering algorithms to denoise the collected data and improve data quality.

[0036] Example 2: This embodiment describes in detail the specific execution steps of the steady-state identification strategy. The strategy achieves stable feature extraction of environmental parameters through multi-scene sampling and layered processing, providing a key basis for system state identification.

[0037] The steady-state identification strategy uses a fixed-interval method to divide monitoring data. Specifically, the system segments the real-time data collected by the sensor according to preset time intervals (such as every minute or every five minutes), with each data segment containing M samples. M here can be set based on actual monitoring needs. For example, in high-frequency monitoring scenarios, M might be set to 100 to ensure that the data segment contains sufficient information. Each sample is composed of a first feature vector composed of a first feature set. The first feature set covers the parameter mean, rate of change, and duration. Taking temperature parameters as an example, for each sample, the mean temperature for a period of time before and after the time point is calculated, the rate of change of temperature per unit time is analyzed, and the duration of a specific temperature state (such as above 28°C) is analyzed. These features together constitute the feature vector describing the sample.

[0038] A multi-scenario sampling framework was constructed, encompassing daily activity patterns, nighttime rest patterns, and away-from-home patterns. The daily activity pattern covers periods when family members engage in routine activities at home, such as breakfast, work, or study. During these periods, human infrared signals fluctuate frequently, and doors and windows are frequently opened and closed. The nighttime rest pattern corresponds to the period after family members fall asleep, when environmental parameters typically stabilize, human infrared signals weaken, and doors and windows are often closed. The away-from-home pattern targets periods when all family members have left their homes. During these periods, human infrared signals disappear, and doors and windows are closed and locked. When constructing the framework, historical data was first used to compile statistics on the time intervals and typical parameter characteristics of each pattern. For example, by analyzing door and window opening and closing records from the past month, it was determined that the most common time period for the away-from-home pattern is 8:00 AM to 6:00 PM on weekdays.

[0039] The samples are processed using a hierarchical processing method. The first step is to calculate the daily features, rest features, and away-from-home features of the samples based on the multi-scenario sampling framework. Taking the smoke concentration sample as an example, in the daily activity mode, the concentration fluctuation range of the sample during the kitchen cooking period needs to be calculated as the daily feature; in the night rest mode, the background concentration in the bedroom area is calculated as the rest feature; in the away-from-home mode, the concentration baseline in the closed state of the whole house is calculated as the away-from-home feature. The calculation of these features needs to be combined with time tags and scene rules. For example, by judging whether the sample collection time is between 23:00 and 6:00 at night and the human body infrared signal is continuously zero, it is determined whether it belongs to the night rest mode.

[0040] The second step is to divide the samples into three groups based on feature stability. Feature stability is determined by calculating the variance and coefficient of variation of each dimension of the feature vector. For example, if the mean variance of a temperature sample is less than a preset threshold and the rate of change fluctuates within a small range, it is considered a highly stable sample. If the mean variance is within the threshold but the rate of change fluctuates significantly, it is classified as a moderately stable sample. If the mean exceeds the normal range and the variance increases significantly, it is classified as a lowly stable sample. The classification criteria need to be set according to the characteristics of different parameters. For example, the stability threshold of the temperature parameter is different from that of the smoke concentration parameter.

[0041] The third step is to perform differentiated processing on samples of varying stability. Highly stable samples are retained directly because they accurately reflect the characteristics of the normal state. Moderately stable samples require interpolation and reconstruction. The interpolation process uses linear interpolation, connecting the parameter curves of two adjacent samples with a straight line to generate a smooth transition sample data segment. For example, if a temperature sample is 25°C at 10:00 and 26°C at 10:01, linear interpolation can generate a transition sample of 25.5°C at 10:00:30. The reconstruction process uses feature dimensionality reduction, mapping the feature vectors of the transition samples to a low-dimensional space. Principal component analysis is then used to extract key components. For example, a three-dimensional feature vector containing the mean, rate of change, and duration is reduced to two dimensions, retaining the most critical features for state recognition.

[0042] The fourth step is clustering and screening based on the scene characteristics of the samples. A clustering algorithm (such as K-means) is used to group the processed samples, with each group corresponding to a typical scene state. The sample with the most balanced daily, rest, and away-from-home characteristics is selected as the standard sample. Balance is determined by comprehensively considering the deviation of each characteristic from the typical value of the corresponding mode. For example, if a sample's daily characteristics deviate by 10% from the mean of the daily activity mode, its rest characteristics deviate by 8% from the mean of the nighttime mode, and its away-from-home characteristics deviate by 5% from the mean of the outing mode, the sample with the smallest deviation among similar samples is selected as the standard sample.

[0043] Step 5: Loop processing until the preset number of samples is reached. When the number of processed samples reaches the preset maximum number (e.g., 1000), the current standard sample set is output and the most suitable sample is selected based on the actual living scenario (such as the apartment type and family members' daily routines entered by the user). If the maximum number has not been reached, the system returns to step 2 to re-divide the sample stability and continue processing. For example, if the system initially sets the maximum number of processed samples to 500, when the 300th sample is processed, it will return to step 2 to stratify the subsequent samples.

[0044] During the execution of the entire steady-state identification strategy, attention should be paid to the dynamic update of the multi-scenario framework. After the system has been running for a period of time, the characteristic parameters of each mode can be re-counted based on the newly collected data. For example, the normal temperature range of the night rest mode in summer may be different from that in winter, and regular adjustments are required. In addition, the computational efficiency of sample processing needs to be optimized. For large-scale data segments, parallel computing technology can be used to accelerate the generation and clustering of feature vectors. At the same time, the judgment threshold of feature stability needs to be adaptive and can be dynamically adjusted according to the distribution of historical data to adapt to the parameter fluctuation characteristics of different living environments. For example, the formaldehyde concentration in a newly renovated house may fluctuate greatly in the initial stage. The system needs to automatically increase the stability judgment threshold of this parameter to avoid misjudgment.

[0045] Through the above steps, the steady-state identification strategy realizes the hierarchical screening and feature optimization of environmental parameters, ensuring that the extracted first feature set can accurately represent the stable state in different scenarios, providing reliable basic data for the subsequent strategy generation module, enabling the system to accurately identify normal and abnormal states in complex home environments, and improving the reliability and adaptability of the alarm system.

[0046] Example 3: This example details the process by which the strategy generation module constructs a rule matching engine and generates identification strategies adapted to new residential scenarios. The core of this module is to extract common patterns from identification strategies across different scenarios and time periods, forming a reusable rule framework that can adapt to diverse home environments.

[0047] The strategy generation module converts the steady-state identification strategy and the transient identification strategy into a steady-state rule vector and a transient rule vector, respectively. The steady-state rule vector is constructed based on the first set of features extracted from the steady-state identification strategy (parameter mean, rate of change, duration, etc.). For example, the normal mean range of temperature parameters and the allowable rate of change threshold for daily activity patterns are converted into elements of the rule vector. The transient rule vector is based on the second set of features from the transient identification strategy (mutation amplitude, rising edge slope, and coupling degree of associated parameters). For example, the mutation amplitude threshold during a sudden temperature rise and the coupling relationship with smoke concentration are converted into vector elements. These two rule vectors are combined into a control rule vector through a cascade operation, allowing the steady-state and transient features to be uniformly represented in the same vector space.

[0048] For control rule vectors for different scenarios (such as daily activities, nighttime rest, and being away from home) and different time periods (such as weekday daytime and weekend nighttime), the system calculates the repetition frequency of each rule. The repetition frequency here refers to the ratio of the number of times a rule appears in the control rule vectors for different scenarios or time periods to the total number of vectors. For example, when analyzing control rule vectors for 100 different time periods, if a rule stating "triggering an alert when the temperature exceeds 35°C and the smoke concentration exceeds a threshold" appears in 80 vectors, its repetition frequency is 80%. When the repetition frequency exceeds a second preset threshold (such as 70%), the rule is determined to be highly universal across different scenarios and time periods and is considered a common rule. The preset threshold should be determined based on historical data statistics, for example, by analyzing a large number of existing control rule vectors to determine a frequency threshold that effectively filters out common rules.

[0049] The extracted common patterns are combined into a rule-matching engine. This engine is essentially a database or algorithmic model containing multiple general rules and their applicable conditions, providing a universal recognition framework for new residential scenarios. For example, the rule-matching engine may include universal rules such as "If the human infrared signal disappears for more than 30 minutes and the doors and windows are closed, it is determined to be away from home mode." These rules are not dependent on specific residential environments and can be used as basic rules.

[0050] When applying the rule-matching engine to a new residential scene, the basic parameters of the new scene must first be entered into the engine. These basic parameters include the initial environmental parameter ranges collected by each sensor in the new scene, such as the initial normal temperature range and the baseline value of smoke concentration. The engine uses algorithms (such as similarity calculations) to match the closest historical scene and time period. For example, if the temperature distribution of the new scene is 90% similar to a scene of daily activities during summer daytime in historical data, the engine will extract the common patterns corresponding to this historical scene.

[0051] The engine needs to generate an adaptive recognition strategy based on the specific parameters of the new scene. Specific parameters include family structure, daily routines, and key protection areas. For example, if the family members in the new scene include elderly people, the system will set "the temperature threshold of the elderly activity area to a stricter range" as a specific rule; if the user sets the kitchen as a key protection area, the engine will strengthen the smoke concentration monitoring rules in the area and lower the warning threshold. Specifically, when the family member structure of the new scene is "with elderly people and children", the system will retrieve specific rules for this type of family from the rule library, such as "the monitoring frequency of the door and window switch status in the children's activity area is increased to once a minute", and combine it with the common rules obtained through historical scene matching to form the final recognition strategy.

[0052] During policy generation, it's important to consider the weighting of common rules and specific parameters. Generally, common rules serve as the foundational framework, ensuring the system's basic functionality; specific parameters are adjusted based on the specifics of new scenarios to make the policy more targeted. For example, while the common rule of "temperatures exceeding 40°C trigger an emergency alarm" remains unchanged, the threshold for key protection areas (such as kitchens) can be adjusted to 38°C to provide early warning.

[0053] The rule-matching engine needs to be self-updating. When new scene recognition strategies generate new valid rules during actual operation, the system automatically feeds these rules back to the engine to update the common rule library. For example, after a new scene has been running for a while, it may be discovered that "occasional brief fluctuations in human infrared signals in the living room between 11:00 PM and 6:00 AM are normal." This rule will be added to the engine's nighttime rest mode rules to reduce false positives.

[0054] Through these steps, the strategy generation module extracts common rules from historical data and adapts them to the characteristics of new scenarios. This eliminates the need for the smart home alarm system to redevelop identification strategies for each new living environment. Instead, it rapidly generates adaptive solutions through rule matching and parameter adjustments, improving system flexibility and deployment efficiency while ensuring the accuracy and reliability of identification strategies.

[0055] Example 4: This embodiment details the workflow and pre-set handling strategies of the exception handling module. This module is the core execution unit of the home intelligent alarm system, which responds to and handles home security risks promptly by monitoring multi-source sensing parameters in real time and triggering corresponding handling strategies.

[0056] The primary task of the anomaly handling module is to monitor multi-source sensor parameters in real time based on the linked data space. For example, in a home kitchen scenario, a temperature sensor collects real-time ambient temperature data, a smoke concentration sensor simultaneously monitors smoke levels, a human infrared sensor detects human activity, and door and window sensors record the open and closed status of windows and doors. This data is continuously fed into the linked data space and compared with pre-built normal state layers, warning state layers, and emergency state layers. For example, if the kitchen temperature rises continuously from 25°C to 35°C within 10 minutes, and the smoke concentration rises from 5%LEL (Lower Explosive Limit) to 15%LEL, the system will determine whether the current parameters exceed the preset safety thresholds based on the hierarchical structure of the linked data space.

[0057] When multi-source sensing parameters exceed safety thresholds, the system determines that the alarm system is operating abnormally and triggers pre-set response strategies. These strategies include parameter adjustment, early warning notification, and emergency linkage, with each triggered according to the severity of the anomaly. For example, if the temperature slowly rises from the normal range to the early warning level (e.g., exceeding 30°C but not reaching the emergency threshold), the system first initiates the parameter adjustment strategy: it automatically adjusts the temperature sensor's sensitivity, increasing its sampling frequency from 1 to 5 times per minute, while also slightly lowering the alarm threshold (e.g., from 35°C to 33°C) to more accurately capture temperature trends. During this process, sensor sensitivity adjustments are based on historical data. For example, if a kitchen has experienced repeated instances of slowly rising temperatures over the past week, the system will default to increasing the sampling frequency of sensors in that area.

[0058] If the temperature continues to rise to the emergency level (e.g., exceeding 40°C), or is accompanied by a sharp increase in smoke concentration (e.g., exceeding 25% LEL), the system triggers an early warning notification strategy. At this point, the mobile app will immediately push a message containing the specific abnormal location (e.g., "kitchen") and parameter details ("Temperature 42°C, Smoke concentration 28% LEL"). Simultaneously, the home's sound and light alarms will emit a high-frequency beep and flash red. For example, in a user's living room, if the system detects an abnormal bedroom temperature and a persistent human infrared signal, the app push message will clearly indicate "Bedroom temperature abnormal, please check." The frequency and brightness of the sound and light alarms will also adjust according to the abnormality level. In an emergency, the beep frequency can reach 1000Hz, and the red light flashes every 0.5 seconds.

[0059] When an extreme abnormality occurs, such as a temperature exceeding 60°C and a smoke concentration exceeding 50% LEL, the system immediately activates an emergency linkage strategy. Taking a typical fire scenario as an example, the system automatically dials a preset emergency number (such as 119), including the home address (via GPS or user pre-set) and the type of abnormality ("kitchen fire warning"). At the same time, it sends a shutdown command to the gas valve controller, cutting off the gas supply to prevent the danger from escalating. During the emergency linkage process, the system prioritizes ensuring the reliability of the command. For example, the command to close the gas valve will be sent three times in a row, with a one-second interval between each. If a shutdown signal is received from the valve, the command will stop. If not, the backup power supply will be triggered to continue sending the command.

[0060] Furthermore, the anomaly handling module must set a statistical period to optimize its handling strategy. For example, a statistical period of 24 hours is set to record the number of times the same anomaly is triggered within that period. For example, if a smoke sensor in a home kitchen triggers an alarm multiple times (e.g., five times) due to cooking fumes within a 24-hour period, and the system determines that it is during normal cooking hours (e.g., dinner time, 6:00 PM to 8:00 PM) based on human infrared signals and door and window status, and the alarm system continues to operate stably without false alarms, then the current smoke concentration threshold is considered overly sensitive. The system will automatically adjust the threshold from 15% LEL to 20% LEL. When adjusting the threshold, the system will reference historical smoke concentration data from normal cooking hours to ensure that the new threshold avoids false alarms while triggering an alarm promptly at the initial stage of a real fire.

[0061] In practice, the response logic of the exception handling module requires integrated multi-parameter linkage analysis. For example, if a single abnormal door or window status is detected (such as a door suddenly opening at 3 a.m.), but the human infrared signal is not triggered, the system will first initiate a warning notification strategy, push a "door opened abnormally" message to the user, and simultaneously activate the hallway camera to take photos for evidence. If the human infrared signal is also triggered at this time, it is determined that there may be an intrusion risk and immediately escalates to an emergency linkage strategy, calling the police and activating the whole house lights to deter people. This multi-parameter linkage approach effectively reduces the interference caused by false alarms from a single sensor.

[0062] The parameter adjustment strategy of the abnormality handling module must also be adaptable to specific scenarios. For example, in nighttime rest mode, to avoid disturbing family members with sound and light alarms, the system will automatically reduce the intensity of the early warning notification strategy's sound and light alarms to low-frequency beeps and dim flashing lights, while also increasing the vibration reminder of the mobile app. In away-from-home mode, if an abnormality is detected, the system will increase the response speed of the handling strategy, for example, by lowering the trigger threshold of the emergency linkage strategy by 10%, to more quickly respond to potential dangers.

[0063] The module's operation relies on the accuracy of the associated data space and the reliability of real-time data. Therefore, the system regularly performs self-calibration on its sensors. For example, the temperature sensor undergoes zero-point calibration at 4:00 AM daily, adjusting its deviation by comparing it to a built-in standard temperature source. The smoke sensor also undergoes a baseline calibration weekly to eliminate zero-point drift caused by dust accumulation. These calibrations ensure the accuracy of sensing parameters and prevent misjudgments due to sensor errors.

[0064] Through the above process, the exception handling module realizes a complete closed loop from real-time monitoring, hierarchical handling to strategy optimization, enabling the home smart alarm system to make accurate responses according to different abnormal situations, while ensuring family safety, minimizing false alarms and missed alarms, and improving user experience.

[0065] Example 5: This embodiment describes in detail the specific steps of the transient identification strategy, which achieves rapid detection and analysis of sudden changes in environmental parameters by constructing a state matrix, a response network, and a time series feature analysis model, providing key support for system early warning.

[0066] In the transient recognition strategy, a state matrix containing the second feature set is constructed based on the alarm system's operating status. For example, in a home living room scenario, when the system detects a sudden temperature rise, the state matrix records the event's time step, temperature, smoke concentration, human infrared signature, and door and window status in real time. Assuming the temperature suddenly rises from 26°C to 32°C at 3:10 PM, the elements corresponding to that time step in the state matrix are: timestamp "15:10:00," temperature "32°C," smoke concentration "8%LEL" (Lower Explosive Limit), human infrared signature "activated" (indicating someone is present in the living room), and door and window status "closed." This state matrix is ​​stored as a multidimensional array, with the parameters for each time step forming a dimension of the array, thus comprehensively recording the environmental conditions at the time of the transient event.

[0067] A fast computational model is used to construct a response network. This network takes the state matrix as input and outputs a response indicator for the second feature set. The response indicator represents the dynamic rate of change of a feature, such as the amplitude of a temperature jump or the slope of its rising edge. Using a temperature spike as an example, the response network analyzes the temperature data for adjacent time steps in the state matrix and calculates the slope of the temperature rise per unit time. Assuming the temperature is 32°C at 15:10:00 and 34°C at 15:10:05, the slope is (34-32) / 5 = 0.4°C / second. The network also calculates the amplitude of the temperature jump, which is the difference from the average temperature under normal conditions. If the normal average is 26°C, the amplitude of the jump is 6°C. The response network must be trained based on historical transient data. For example, state matrix data for all temperature spike events over the past six months can be collected. Machine learning algorithms can then be used to optimize network parameters to accurately extract the characteristics of the jump.

[0068] Constructing a feature space is a crucial step in transient recognition strategies. This space represents the range of values ​​for the response indicators of the second feature set, providing a reference for determining whether parameter changes are abnormal. For example, analyzing historical data reveals that the temperature slope is typically less than 0.1°C / second under normal circumstances, but can exceed 0.3°C / second in abnormal situations, such as during the initial stages of a fire. Therefore, the temperature slope in the feature space can be set to a range of 0 to 1°C / second, with 0.3°C / second serving as the warning threshold. The boundary values ​​of the feature space need to be determined based on the characteristics of different parameters and the distribution of historical data. For example, the feature space for the sudden change amplitude of smoke concentration might be set from 0 to 100% LEL, while the feature space for the coupling degree of the associated parameters of the human infrared signal is determined based on its linkage with the status of doors and windows.

[0069] The evaluation function of the time series feature analysis model uses a mean squared error (MSE) function to measure the deviation between the model's predicted value and the actual value, thereby optimizing the parameters of the response network. For example, when the actual temperature rise slope is 0.5°C / second, the model predicts a value of 0.48°C / second, resulting in a MSE of (0.5-0.48)² = 0.0004. By continuously calculating the MSE, a gradient optimization method is used to update the weight parameters of the response network, bringing the model's output closer to the actual value. The gradient optimization method adjusts parameters in the opposite direction of the error gradient. For example, if the model predicts a temperature jump that is too low, the optimization algorithm will increase the weight of the corresponding parameter, increasing the model's sensitivity to the sudden change.

[0070] For example, when cooking oil in a pot overheats and causes smoke, the transient recognition strategy executes as follows: First, the state matrix records the parameters at time step 18:30:15: temperature 28°C, smoke concentration 12% LEL, human infrared signal activation (user in the kitchen), and doors and windows closed. At 18:30:30, the temperature suddenly rises to 35°C, and the smoke concentration rises to 20% LEL. After receiving this state matrix, the response network calculates the temperature rise slope (35-28) / 15 = 0.47°C / second), the sudden change amplitude 9°C, and the sudden change amplitude of the smoke concentration 8% LEL. It also analyzes the coupling between the temperature and smoke concentration parameters (determined by calculating the correlation coefficient between the two; for example, a correlation coefficient of 0.9 indicates strong coupling). These response indicators are input into the feature space. Because the temperature rise slope of 0.47°C / second exceeds the preset threshold of 0.3°C / second, and the smoke concentration sudden change amplitude and coupling exceed the threshold, the system identifies this as an abnormal transient event and triggers the early warning notification strategy.

[0071] In practical applications, transient recognition strategies require adjusting feature space thresholds based on different scenarios. For example, during winter heating, the living room temperature may rise slowly due to the heater being turned on. In this case, the threshold for the temperature rise slope can be temporarily adjusted to 0.2°C / second to avoid false alarms. During summer air conditioning operation, the threshold can be restored to 0.1°C / second. This scenario-adaptive adjustment is achieved through scene tags in the associated data space. When the system detects that it is currently in "winter heating mode," it automatically loads the corresponding feature space threshold.

[0072] Transient recognition strategies also need to handle transient events involving multiple parameters. For example, when doors and windows are suddenly opened while the human infrared signal is inactive (possibly indicating a burglary), the state matrix records the time steps when the door and window state changes from "closed" to "open," while the human infrared signal is "inactive." The response network calculates the sudden change amplitude (the jump from 0 to 1) and the rising edge slope (the instantaneous change) of the door and window open and close states, and analyzes their coupling with the human infrared signal (in this case, the coupling is 0, indicating an anomaly). The threshold for abnormal door and window opening in the feature space is set as: a sudden change amplitude of 1 and a coupling less than 0.3. When these conditions are met, the system triggers an intrusion warning.

[0073] To improve the real-time performance of transient identification, response network calculations must be performed on edge devices (such as home gateways) to avoid data upload delays to the cloud. For example, real-time processing of the state matrix is ​​performed in the local gateway's CPU, keeping the latency from data acquisition to response indicator output to less than 50 milliseconds, ensuring timely detection of abnormal transient events. Simultaneously, the system regularly performs incremental training on the response network, updating model parameters with newly collected transient data. For example, the system automatically downloads the latest training data weekly to optimize network weights to adapt to parameter feature shifts caused by environmental changes.

[0074] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0075] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A home intelligent alarm system, characterized in that: include: An environmental perception module, which uses multimodal detection to construct a correlation data space between environmental parameters and alarm trigger conditions, and integrates multi-source perception parameters, including temperature, smoke concentration, human infrared signals, and door and window switch status. The correlation data space includes a normal state layer, a warning state layer, and an emergency state layer. The state recognition module is used to execute the steady-state recognition strategy and the transient recognition strategy respectively. The steady-state recognition strategy uses a multi-scenario sampling method to extract a first feature set; the transient recognition strategy extracts a second feature set by constructing a time series feature analysis model. The first feature set includes parameter mean, rate of change, and duration; the second feature set includes mutation amplitude, rising edge slope, and correlation parameter coupling degree. The strategy generation module is used to extract common rules from the steady-state recognition strategies and transient recognition strategies trained in different scenarios and time periods, build a rule matching engine, and generate recognition strategies adapted to new residential scenarios based on the rule matching engine; The abnormality handling module is used to monitor multi-source perception parameters in real time based on the associated data space, and trigger the preset handling strategy when abnormal parameters are detected.

2. A home intelligent alarm system according to claim 1, characterized in that: Through multimodal detection, a data space is constructed to associate environmental parameters with alarm triggering conditions, and multi-source perception parameters are integrated. The specific steps include: Use synchronous acquisition devices to obtain time series data of environmental parameters and alarm trigger conditions, and perform data alignment and calibration; By analyzing the linkage relationship between parameters, normal state modeling, warning state simulation and emergency state analysis are performed on the associated data space respectively; The real-time multi-source perception parameters of the actual living environment are obtained and input into the associated data space.

3. A home intelligent alarm system according to claim 1, characterized in that: The specific steps of the steady-state identification strategy include: The monitoring data is divided using a fixed interval method, where the data segment includes M samples, and each sample is represented by a first feature vector composed of a first feature set; Construct a multi-scenario sampling framework, which includes daily activity mode, nighttime rest mode, and away-from-home mode; The samples were processed using a stratified processing approach; The specific steps of the layered processing method include: S1, calculates the daily characteristics, rest characteristics and leaving home characteristics of the sample based on the multi-scene sampling framework; S2, divides the samples into three groups according to feature stability: high-stable samples, medium-stable samples, and low-stable samples; S3, retains highly stable samples, performs interpolation and reconstruction operations on moderately stable samples, and performs high-probability elimination operations on low-stability samples; S4, based on the daily characteristics, rest characteristics, and leaving home characteristics of the samples, scene clustering is used to screen and select standard samples, among which the samples with the most balanced daily characteristics, rest characteristics, and leaving home characteristics are used as standard samples; S5, when the processing quantity reaches the preset maximum processing quantity, output the current standard sample, and select the sample that best meets the current living scenario according to actual needs, otherwise return to S2.

4. A home intelligent alarm system according to claim 3, characterized in that: The interpolation operation uses the linear interpolation method to generate new samples. The operation process of the linear interpolation method is as follows: the parameter curves of two adjacent samples are connected by linear interpolation to generate a smooth transition sample data segment; The reconstruction operation uses the feature dimensionality reduction method to update the sample. The feature dimensionality reduction operation process is: mapping the feature vector of the transition sample to a low-dimensional space, extracting the key components through principal component analysis, and generating a simplified feature vector.

5. A home intelligent alarm system according to claim 1, characterized in that: The steady-state identification strategy and the transient identification strategy are respectively converted into a steady-state rule vector and a transient rule vector, and cascaded to obtain a control rule vector. For the control rule vectors of different scenarios and different time periods, the repetition frequency of each rule in the control rule vector is calculated in sequence. When the repetition frequency is greater than a second preset threshold, the rule is determined to have high universality across different scenarios and time periods and is regarded as a common rule. The extracted common rules are combined to obtain a rule matching engine, which provides a universal recognition framework for new residential scenarios. The rule matching engine provides a general recognition framework for new living scenarios. The specific application methods include: inputting the basic parameters of the new living scenario into the rule matching engine, matching the closest historical scenarios and time periods, extracting the corresponding common rules, and combining the specific parameters of the new scenario to generate an adaptive recognition strategy. The specific parameters include: family member structure, daily routine habits and key protection areas.

6. A home intelligent alarm system according to claim 1, characterized in that: The exception handling module monitors multi-source perception parameters in real time based on the associated data space. When the multi-source perception parameters exceed the preset safety threshold, the alarm system is judged to be operating abnormally and the preset handling strategy is triggered. The statistical period is set to record the number of times the same exception is triggered within the period. If the same exception occurs multiple times within the statistical period and the alarm system is still operating stably, the handling threshold of the exception will be adjusted.

7. A home intelligent alarm system according to claim 1, characterized in that: The specific steps of the transient identification strategy include: According to the operating state of the alarm system, a state matrix containing the second feature set is constructed, wherein the expression of the state matrix is ​​a multidimensional array containing time step, temperature, smoke concentration, human infrared signal and door and window switch status; A fast computation model is used to construct a response network, which takes a state matrix as input and outputs a response index of a second feature set, wherein the response index represents a dynamic change rate of the feature; Constructing a feature space, where the feature space represents a value range of a response indicator of the second feature set; Construct an evaluation function for the time series feature analysis model; The response network is updated using a gradient optimization method.

8. A home intelligent alarm system according to claim 2, characterized in that: Use a synchronous acquisition device to obtain time series data of environmental parameters and alarm trigger conditions, and perform data alignment and calibration. The specific steps include: Configure independent time series recording units for environmental parameters and alarm trigger conditions respectively; All recording units are started simultaneously through hardware synchronization signals to ensure the consistency of the initial time series; The collected time series data is subjected to offset detection, and the time deviation is corrected by polynomial fitting method to achieve data alignment between parameters.

9. A home intelligent alarm system according to claim 6, characterized in that: The preset disposal strategies include: parameter adjustment strategy, early warning notification strategy and emergency linkage strategy. Among them, the parameter adjustment strategy is used to automatically adjust the sensor sensitivity and alarm threshold. The early warning notification strategy is used to push mobile phone APP messages and trigger sound and light alarms when an abnormality occurs. The emergency linkage strategy is used to automatically call the emergency number and close the gas valve in extreme abnormalities.

10. A home intelligent alarm system according to claim 7, characterized in that: The evaluation function of the time series feature analysis model adopts the mean square error function.

Images