An early warning system for building security

Abstract

This invention relates to the field of building security technology and discloses an early warning system for building security, comprising: constructing a normal behavior profile of each area of ​​the building through multi-source basic sensing data, extracting current behavior features in real time and comparing them with the profile, and generating an early warning signal when multi-dimensional features deviate collaboratively and continuously exceed a threshold. This invention realizes a paradigm shift from single threshold alarm to abnormal behavior pattern early warning, significantly improves early risk identification capability through a multi-dimensional collaborative deviation detection mechanism, and maintains high accuracy and environmental adaptability in complex scenarios with the dual guarantee of adaptive profile update and power grid harmonic noise verification.

Description

Technical Field

[0001] This invention relates to an early warning system for building security, belonging to the field of building security technology. Background Technology

[0002] Building safety early warning systems currently rely mainly on single physical quantity threshold triggering mechanisms, such as excessive smoke concentration or abnormal magnetic switches of doors and windows. These systems follow event-driven logic and can only respond after obvious damage occurs. They generally lack the ability to monitor the dynamic continuity of the overall building behavior pattern. In practical applications, risks such as the concealed development of initial fires and unauthorized intrusion attempts often lead to spatial rhythm abnormalities before traditional alarms are triggered. These include surges in activity in non-time-period areas, disordered equipment operation sequences, subtle behavioral changes such as lighting and air conditioning failures or energy consumption baseline drift.

[0003] However, existing technologies are limited by three fundamental bottlenecks: 1. They cannot construct multi-dimensional historical normal profiles, resulting in the inability to identify early collaborative deviations; 2. The confidence level of profiles during the cold start period is insufficient, which easily leads to false alarms, and there is a lack of flexible adjustment mechanisms when authorized activities change drastically; 3. In ambiguous situations, there is a lack of physical evidence independent of the behavioral layer, which limits the confidence level of decisions.

[0004] Some improvements have attempted to introduce complex learning models, but due to high computing power requirements and high implementation costs, they have failed to solve the core contradiction of early risk generalization warning under multi-source sparse data. Therefore, how to achieve early and accurate warning of sudden changes in building behavior patterns, and at the same time improve the adaptability and decision confidence in complex scenarios, has become the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides an early warning system for building security, the main purpose of which is to solve the problems of difficulty in timely detection of hidden building security risks, insufficient accuracy of early warning in complex scenarios, and lack of decision confidence.

[0006] To achieve the above objectives, the present invention provides an early warning system for building security, the system comprising:

[0007] The information sensing module is configured inside the building to acquire multi-source basic sensing data that reflects the building's space utilization patterns, equipment operating time series, and energy consumption baseline status.

[0008] The normal behavior profile construction module is electrically connected to the information perception module. It is used to adaptively construct and dynamically update the normal behavior profile of a specific geographical area and time period combination within a building based on the historical sequence of multi-source basic perception data and for specific geographical areas and time period combinations within the building using statistical learning. The normal behavior profile represents the multidimensional statistical distribution characteristics of multi-source basic perception data under historical normal conditions and the interrelationship between behavioral features.

[0009] The real-time behavior feature extraction module is electrically connected to the information perception module and is used to acquire multi-source behavior features under the current geographical area and time period combination in real time.

[0010] The behavior mutation detection module is electrically connected to the normal profile construction module and the real-time behavior feature extraction module. It is used to compare the current multi-source behavior features with the normal behavior profile in real time. When the combined deviation of the current multi-source behavior features exceeds a preset first threshold and the duration exceeds a preset first time length, an early warning signal representing the building safety risk is generated. The judgment criteria for the combined deviation include anomalies in a single dimension, coordinated deviations between multiple dimensions, and loss of correlation of historical related behavior features.

[0011] Preferably, the multi-source basic sensing data acquired by the information sensing module includes one or more combinations of the following: output signals from human activity sensors; opening and closing status signals from door and window magnetic switches; real-time energy consumption readings from energy meters, including measurements from electricity meters, water meters, and gas meters; ambient light signals from light intensity sensors; and ambient sound signals from sound intensity sensors. This multi-source basic sensing data comprehensively reflects personnel activities, entry and exit status, equipment operating load, ambient light levels, and ambient sound conditions within the building, thereby providing comprehensive behavioral information input for constructing a profile of normal behavior.

[0012] Preferably, the statistical learning method adopted by the normal profile construction module includes one or more of the following: calculating the statistical parameters of each behavioral feature under the corresponding geographical region and time period combination in the same historical period, the statistical parameters including but not limited to mean, median, standard deviation, and quantile interval; calculating the stable variation range of Pearson correlation coefficient between different behavioral features; or identifying specific behavioral feature combination patterns that frequently occur in the same historical period.

[0013] Preferably, the behavior mutation detection module determines that the combined deviation of the current multi-source behavioral features exceeds a preset first threshold, including: at least two behavioral features in the current multi-source behavioral features simultaneously deviate from their respective normal ranges defined in the normal behavioral profile, or a combination of behavioral features that has historically been strongly correlated shows that the absolute value of its Pearson correlation coefficient has decreased beyond a preset second threshold, and the duration of the deviation or loss of correlation exceeds a preset first time length, when a warning signal is generated.

[0014] Preferably, the normal profile construction module performs periodic rolling updates based on newly collected multi-source basic perception data and real-time behavioral characteristics at a frequency of once per hour to once per day, in order to adapt to the long-term evolution of building behavior patterns.

[0015] Preferably, it also includes: a profile adaptive adjustment module, electrically connected to the normal profile construction module, configured to set an initial high-density observation period when the early warning system is initially deployed or when a new geographical area is configured, with the observation period lasting from seven to thirty days; during this observation period, the normal profile construction module initially constructs a normal behavior profile at a frequency of once every ten minutes to once every hour, and the profile adaptive adjustment module dynamically adjusts the sensitivity threshold for the behavior change detection module to judge behavior changes according to the accumulation of profile data, with the sensitivity threshold decreasing linearly or non-linearly according to the amount of data accumulation.

[0016] Preferably, it also includes: an activity context receiving module, configured to receive context annotation information input by authorized users for a specific geographic area and future time period combination, representing expected atypical activity types; and a profile adaptation adjustment module, electrically connected to the normal profile construction module, the real-time behavior feature extraction module, and the activity context receiving module, configured so that when the activity context receiving module receives context annotation information for the current geographic area and current time period combination, the behavior mutation detection module, when comparing the current multi-source behavior features with the normal behavior profile, adjusts the weight of historical statistical parameters in the comparison in the normal behavior profile according to the type and intensity of the context annotation information, or increases the tolerance of the current behavior dynamic baseline, and when updating the normal behavior profile, based on the context annotation information, excludes the multi-source behavior feature data collected in the corresponding time period from the statistical samples of long-term profile updates.

[0017] Preferably, it further includes: a power grid noise sensing module, configured to passively acquire voltage and / or current signals at at least one location of the building's AC power supply network, and extract time-series data of background harmonic noise characteristics in a preset frequency band of 10 Hz to 2 kHz from the signals; a harmonic mode baseline construction module, electrically connected to the power grid noise sensing module, used to construct and dynamically update the harmonic noise mode baseline representing the building's power supply network under normal operating conditions based on historical time-series data of background harmonic noise characteristics using statistical learning; and an early warning collaborative verification module, electrically connected to the behavior mutation detection module and the harmonic mode baseline construction module, used to collaboratively analyze the deviation of multi-source behavioral characteristics and the deviation of background harmonic noise characteristics when the behavior mutation detection module generates an early warning signal, or when the degree of deviation of the real-time monitored background harmonic noise characteristics from the harmonic noise mode baseline exceeds a preset third threshold, and based on the results of the collaborative analysis, to increase or decrease the confidence level of the early warning signal, or to trigger targeted risk warnings or enhance the monitoring of specific behavioral patterns when only the background harmonic noise characteristics deviate from the harmonic mode baseline.

[0018] Preferably, the early warning collaborative verification module calculates the collaborative confidence factor between the background harmonic noise characteristics and the behavioral change early warning based on the degree of deviation of the characteristics. Cooperative confidence factor The following relationship must be satisfied: ,in, This index represents the degree of deviation in the behavioral pattern as determined by the behavioral mutation detection module, and its value ranges from zero to one. The index represents the degree of deviation of the background harmonic noise as determined by the harmonic mode baseline construction module, and its value ranges from zero to one. and The weighting coefficients are preset and satisfy the following conditions: .

[0019] Preferably, in the information sensing module, the human activity sensor is a passive infrared sensor, the door and window magnetic switch is a wireless magnetic sensor, the energy meter includes at least one of a smart electricity meter, a smart water meter, and a smart gas meter, the light intensity sensor is a photoresistor sensor, and the sound intensity sensor is a sound pressure sensor.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. The system is based on multi-source sparse data streams generated by low-cost sensors in the building, such as human activity, door and window status, and energy consumption readings. Through statistical learning, it constructs normal behavior profiles under different spatiotemporal scenarios. When real-time behavioral characteristics deviate from the normal fluctuation range defined by historical profiles in multiple dimensions at the same time, or when historically strongly correlated feature combinations lose correlation, the system automatically identifies it as a sudden change in behavior pattern. This multi-dimensional collaborative deviation judgment mechanism transforms the judgment logic of traditional alarms, which relies on a single physical quantity threshold, into the protection of the continuity and consistency of the overall behavior of the building. This allows for early warnings to be triggered based solely on subtle abnormalities in behavior patterns before physical damage or dangerous indicators appear, avoiding the problem of existing technologies missing early hidden risks.

[0022] 2. The routine profile building module automatically adjusts profile parameters according to the long-term gradual changes in building behavior patterns through periodic rolling updates, such as daily / weekly. To address the cold start issue in the initial deployment of the system, the warning sensitivity threshold is dynamically adjusted during the initial high-density observation period: the judgment conditions are relaxed when the confidence of the initial profile is low, and gradually tightened as data accumulates to avoid false alarms; for sudden authorized activities, such as temporary overtime, the weight of historical profiles is temporarily reduced and the tolerance for the current dynamic baseline is increased through simplified annotation information input by the user, while isolating the data of this period from contaminating the long-term profile; this flexible adjustment mechanism enables the system to maintain the accuracy of warnings and user trust in scenarios with changes in profile maturity and drastic changes in known behaviors.

[0023] 3. The system passively collects background harmonic noise from the building's power grid, such as the amplitude of the 3rd, 5th, and 7th harmonics, to construct a baseline for its normal operation. When the behavioral change detection module generates an early warning, it simultaneously analyzes whether the power grid noise deviates from the baseline. If the two are highly coordinated in time, such as abnormal human activity in a certain area accompanied by poor power grid contact, the warning level is raised to a high-risk state. If only the power grid noise is abnormal, the monitoring of related behaviors is enhanced in a targeted manner. This cross-validation of physical layer stress signals and behavioral layer pattern anomalies achieves zero-cost efficiency improvement using the building's existing power facilities, providing a fundamentally differentiated supporting dimension for early warning decisions in fuzzy scenarios. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the structure of the behavioral pattern analysis and adaptive early warning system of the present invention;

[0025] Figure 2 This is a timing diagram of the behavior mutation joint early warning processing flow of the present invention;

[0026] Figure 3 This is a graph showing the changes in the collaborative early warning indicators of behavioral deviation and power grid noise at different time points in this invention.

[0027] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0028] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0029] This application provides an early warning system for building security, the system comprising:

[0030] The information sensing module is configured inside the building to acquire multi-source basic sensing data that reflects the building's space utilization patterns, equipment operating time series, and energy consumption baseline status.

[0031] The normal behavior profile construction module is electrically connected to the information perception module. It is used to adaptively construct and dynamically update the normal behavior profile of a specific geographical area and time period combination within a building based on the historical sequence of multi-source basic perception data and for specific geographical areas and time period combinations within the building using statistical learning. The normal behavior profile represents the multidimensional statistical distribution characteristics of multi-source basic perception data under historical normal conditions and the interrelationship between behavioral features.

[0032] The real-time behavior feature extraction module is electrically connected to the information perception module and is used to acquire multi-source behavior features under the current geographical area and time period combination in real time.

[0033] The behavior mutation detection module is electrically connected to the normal profile construction module and the real-time behavior feature extraction module. It is used to compare the current multi-source behavior features with the normal behavior profile in real time. When the combined deviation of the current multi-source behavior features exceeds a preset first threshold and the duration exceeds a preset first time length, an early warning signal representing the building safety risk is generated. The judgment criteria for the combined deviation include anomalies in a single dimension, coordinated deviations between multiple dimensions, and loss of correlation of historical related behavior features.

[0034] Preferably, the multi-source basic sensing data acquired by the information sensing module includes one or more combinations of the following: output signals from human activity sensors; opening and closing status signals from door and window magnetic switches; real-time energy consumption readings from energy meters, including measurements from electricity meters, water meters, and gas meters; ambient light signals from light intensity sensors; and ambient sound signals from sound intensity sensors. This multi-source basic sensing data comprehensively reflects personnel activities, entry and exit status, equipment operating load, ambient light levels, and ambient sound conditions within the building, thereby providing comprehensive behavioral information input for constructing a profile of normal behavior.

[0035] Preferably, the statistical learning method adopted by the normal profile construction module includes one or more of the following: calculating the statistical parameters of each behavioral feature under the corresponding geographical region and time period combination in the same historical period, the statistical parameters including but not limited to mean, median, standard deviation, and quantile interval; calculating the stable variation range of Pearson correlation coefficient between different behavioral features; or identifying specific behavioral feature combination patterns that frequently occur in the same historical period.

[0036] Preferably, the behavior mutation detection module determines that the combined deviation of the current multi-source behavioral features exceeds a preset first threshold, including: at least two behavioral features in the current multi-source behavioral features simultaneously deviate from their respective normal ranges defined in the normal behavioral profile, or a combination of behavioral features that has historically been strongly correlated shows that the absolute value of its Pearson correlation coefficient has decreased beyond a preset second threshold, and the duration of the deviation or loss of correlation exceeds a preset first time length, when a warning signal is generated.

[0037] Preferably, the normal profile construction module performs periodic rolling updates based on newly collected multi-source basic perception data and real-time behavioral characteristics at a frequency of once per hour to once per day, in order to adapt to the long-term evolution of building behavior patterns.

[0038] Preferably, it also includes: a profile adaptive adjustment module, electrically connected to the normal profile construction module, configured to set an initial high-density observation period when the early warning system is initially deployed or when a new geographical area is configured, with the observation period lasting from seven to thirty days; during this observation period, the normal profile construction module initially constructs a normal behavior profile at a frequency of once every ten minutes to once every hour, and the profile adaptive adjustment module dynamically adjusts the sensitivity threshold for the behavior change detection module to judge behavior changes according to the accumulation of profile data, with the sensitivity threshold decreasing linearly or non-linearly according to the amount of data accumulation.

[0039] Preferably, it also includes: an activity context receiving module, configured to receive context annotation information input by authorized users for a specific geographic area and future time period combination, representing expected atypical activity types; and a profile adaptation adjustment module, electrically connected to the normal profile construction module, the real-time behavior feature extraction module, and the activity context receiving module, configured so that when the activity context receiving module receives context annotation information for the current geographic area and current time period combination, the behavior mutation detection module, when comparing the current multi-source behavior features with the normal behavior profile, adjusts the weight of historical statistical parameters in the comparison in the normal behavior profile according to the type and intensity of the context annotation information, or increases the tolerance of the current behavior dynamic baseline, and when updating the normal behavior profile, based on the context annotation information, excludes the multi-source behavior feature data collected in the corresponding time period from the statistical samples of long-term profile updates.

[0040] Preferably, it further includes: a power grid noise sensing module, configured to passively acquire voltage and / or current signals at at least one location of the building's AC power supply network, and extract time-series data of background harmonic noise characteristics in a preset frequency band of 10 Hz to 2 kHz from the signals; a harmonic mode baseline construction module, electrically connected to the power grid noise sensing module, used to construct and dynamically update the harmonic noise mode baseline representing the building's power supply network under normal operating conditions based on historical time-series data of background harmonic noise characteristics using statistical learning; and an early warning collaborative verification module, electrically connected to the behavior mutation detection module and the harmonic mode baseline construction module, used to collaboratively analyze the deviation of multi-source behavioral characteristics and the deviation of background harmonic noise characteristics when the behavior mutation detection module generates an early warning signal, or when the degree of deviation of the real-time monitored background harmonic noise characteristics from the harmonic noise mode baseline exceeds a preset third threshold, and based on the results of the collaborative analysis, to increase or decrease the confidence level of the early warning signal, or to trigger targeted risk warnings or enhance the monitoring of specific behavioral patterns when only the background harmonic noise characteristics deviate from the harmonic mode baseline.

[0041] Preferably, the early warning collaborative verification module calculates the collaborative confidence factor between the background harmonic noise characteristics and the behavioral change early warning based on the degree of deviation of the characteristics. Cooperative confidence factor The following relationship must be satisfied: ,in, This index represents the degree of deviation in the behavioral pattern as determined by the behavioral mutation detection module, and its value ranges from zero to one. The index represents the degree of deviation of the background harmonic noise as determined by the harmonic mode baseline construction module, and its value ranges from zero to one. and The weighting coefficients are preset and satisfy the following conditions: In the information sensing module, the human activity sensor is a passive infrared sensor, the door and window magnetic switches are wireless magnetic sensors, the energy meter includes at least one of smart meters, smart water meters, and smart gas meters, the light intensity sensor is a photoresistor sensor, and the sound intensity sensor is a sound pressure sensor. Meanwhile, to ensure user privacy and system data security, this invention fully considers data compliance and user information protection principles during design and implementation. For example, the multi-source basic sensing data collected by the system does not involve user identity information, and the recorded behavioral characteristic parameters, such as the frequency of human activity, door and window status, energy consumption fluctuations, ambient light, and sound intensity, are all desensitized, non-identifiable data used only for behavioral pattern modeling and deviation analysis. The detection, data collection, and transmission processes employ encryption mechanisms to ensure security during communication links and storage. In extended implementations, the system supports localized deployment and edge computing processing. Routine behavior profiling and early warning analysis can be completed on terminal devices or local servers without uploading to external networks, reducing the risk of data leakage at the source. Furthermore, contextual annotation information is proactively provided by authorized users and stored in isolation within the system, not directly bound to behavioral characteristics to avoid creating personal profiles. Additionally, a data access control mechanism can be equipped to ensure that only authorized personnel can view and manage relevant data and early warning information, thereby comprehensively enhancing user privacy protection. All of these are extended implementation methods known to those skilled in the art.

[0042] Example 1: This invention relates to an early warning system for building security. Based on multi-source basic sensing data, it constructs a profile of normal behavior in different areas of a building during a specific time period. By extracting current behavioral features in real time and comparing them with historical profiles, it identifies abnormal behavior patterns and achieves early warning. This invention aims to avoid the limitations of existing technologies that rely on thresholds of single physical quantities. Through a multi-dimensional collaborative deviation judgment mechanism, a dynamic profile update mechanism, and verification methods based on power grid harmonic noise, it significantly improves risk identification capabilities and early warning accuracy in complex application scenarios. The overall early warning system includes the following main functional modules: an information sensing module, a normal profile construction module, a real-time behavioral feature extraction module, a behavioral change detection module, a profile adaptive adjustment module, an activity context receiving module, a power grid noise sensing module, a harmonic mode baseline construction module, and an early warning collaborative verification module. These modules are logically connected through data interfaces, forming a complete data sensing and risk early warning closed loop. To ensure the timing coordination of these modules during actual operation, the system has a unified time scheduling mechanism. This mechanism sets the sampling and update frequency in layers, specifying the time intervals by second. The system organically combines minute-level or minute-level data acquisition, real-time feature extraction every ten minutes to hour, and daily routine profile updates. Utilizing timestamp alignment and task scheduling control, it ensures temporal continuity and logical consistency between data flow and model updates, effectively avoiding misjudgments or delayed warnings due to inconsistent processing rhythms. This scheduling strategy is also applicable to the high-frequency modeling phase during the cold start observation period, ensuring the system's stability and consistency across different operational stages. The information sensing module, deployed within the building, is the system's fundamental data source. It acquires various types of sensing data reflecting building space usage patterns, equipment operating status, and environmental parameters. The collected data includes, but is not limited to: activity signals output by human activity sensors (e.g., passive infrared sensors); opening and closing status signals of door and window magnetic switches (e.g., wireless magnetic sensors); real-time energy consumption readings from energy meters; ambient light intensity signals (e.g., values ​​collected by photoresistor sensors); and ambient sound intensity signals (e.g., output from sound pressure sensors). This module supports multiple data input combinations and performs unified sampling, preprocessing, and time alignment to generate a multi-source sensing data stream for analysis.The routine behavior profile construction module constructs routine behavior profiles for different geographical regions and time periods through statistical analysis of historical multi-source sensing data. Specifically, this module uses statistical learning methods to calculate the statistical parameters of each behavioral feature under similar historical spatiotemporal conditions, including mean, median, standard deviation, and quantile intervals. It also establishes Pearson correlation coefficient intervals between behavioral features to characterize the stable range of correlations between features. Furthermore, the module identifies high-frequency combinations of specific behavioral features in historical data to supplement the expressive power of traditional parametric models for complex behaviors. These routine behavior profiles encompass both univariate behavioral patterns and reflect historically stable patterns of multi-dimensional collaboration between features. The real-time behavior feature extraction module extracts and constructs current multi-source behavioral feature combinations consistent with the profile structure based on the sensing data collected at the current moment. The extracted feature types include... The current behavioral feature set generated by this module, including the number of personnel activities per unit time, energy usage rate, entry and exit frequency, changes in ambient light, and fluctuations in ambient sound intensity, is sent in real time to the behavioral mutation detection module for comparison and judgment. The behavioral mutation detection module compares the real-time features with the normal behavioral profile under corresponding spatiotemporal conditions to determine whether there is a structural deviation. The criteria for deviation judgment include, but are not limited to: at least two behavioral features simultaneously exceeding the normal fluctuation range defined in the normal profile, or historically highly correlated behavioral features showing significant loss of correlation in the current data, with the absolute value of their Pearson correlation coefficient decreasing beyond a preset second threshold. Furthermore, the system only determines a sudden change in behavioral pattern and generates a corresponding warning signal after the above deviation state persists for more than a first time period (e.g., 30 minutes), thereby effectively suppressing the risk of false alarms caused by short-term anomalies.

[0043] To adapt to insufficient data during the initial system deployment and changes in scenarios during subsequent operation, the system introduces a profile adaptive adjustment module. This module sets an initial high-density observation period of seven to thirty days when the system is first deployed or a new monitoring area is added. During this period, the normal profile construction module constructs preliminary profiles at a frequency of ten minutes to one hour. As the amount of data accumulates, the sensitivity threshold will decrease according to a linear or non-linear function, thereby appropriately relaxing the deviation judgment conditions in the early stage when the profile confidence is low to avoid false alarms, and gradually increasing the recognition standard after the profile stabilizes. During system operation, if there are atypical but predictable changes in authorized behavior, such as working overtime at night, working overtime on holidays, or temporary performances, users can submit context annotation information through the activity context receiving module. The system considers this context information together with the current behavior characteristics and historical profiles, dynamically adjusting the weight of the normal profile during the behavior comparison process, or appropriately increasing the dynamic tolerance range of the current behavior characteristics, and updating the profile accordingly. Data within a given segment is removed from the long-term profile sample to avoid affecting the long-term stability of the profile, thus achieving robust profile maintenance under drastic behavioral changes. Simultaneously, to further enhance the confidence of the early warning signal, the system also includes a power grid noise perception module and a harmonic mode baseline construction module. These modules are used to construct and monitor the background harmonic noise patterns of the building's power supply system in the 10 Hz to 2 kHz frequency band. By passively collecting voltage and / or current signals from one or more nodes in the building's power grid and extracting their spectral characteristics, the system can form a statistical baseline characterizing power supply stability. When the current behavioral change detection module generates an early warning signal, the system will simultaneously determine whether the current background harmonic noise also deviates from the normal pattern. If the behavioral pattern deviation and the harmonic pattern deviation are highly consistent in time, the early warning collaborative verification module will automatically increase the confidence level of the current early warning. If only power grid noise anomalies exist, the system will enhance the dynamic monitoring of the behavior in the relevant area, indicating possible equipment anomalies or potential risks. The system includes a confidence factor calculation method for collaborative analysis to quantify the combined impact of behavioral deviation and power grid noise deviation. This confidence factor is synthesized by weighted summation of the Behavioral Pattern Deviation Index (BPI) and the Noise Deviation Index (NPI). The specific weighting method satisfies the condition that the total weight is one, i.e., the sum of the behavioral weight and the noise weight is one. By reasonably setting the two weight coefficients, the influence of the deviation index on the final early warning decision can be flexibly adjusted according to the focus of different building scenarios.

[0044] Example 2: In this example, a security and early warning system for building scenarios is proposed. It achieves dynamic mutation identification based on multi-source behavioral characteristics and utilizes power grid harmonic noise as an auxiliary physical signal for verification, thereby enhancing the stability and confidence of the overall early warning system. The design and implementation of this system revolve around various types of sensor data from daily building operations. A complete early warning process is constructed by combining statistical learning and feature correlation analysis. The information sensing module is deployed in key spatial areas of the building to collect various types of sensor data. The acquired data includes: human activity signals output by human activity sensors, opening and closing status signals of door and window magnetic switches, energy consumption data such as electricity, water, and gas, and light intensity change data. The system collects data on environmental sound intensity fluctuations, human activity signals (sampled every minute to count activity frequency per unit time), energy usage data (sampled every five minutes, with the rate of change calculated from the difference between adjacent time periods), and light and sound signals (sampled every two seconds, preprocessed using a moving average). All perceived data undergoes unified timestamp calibration and format standardization to construct a multi-source behavioral data unit with a unified temporal structure. The routine profile construction module uses a sliding window mechanism for modeling, dividing the building space into multiple areas based on physical location, and further dividing the daily time of each area into at least twelve time periods. Within each time period, this module calculates the historical data of all behavioral characteristics. Statistical parameters, including mean, median, standard deviation, and upper and lower quantiles, are used. The Pearson correlation coefficient distribution range between each behavioral feature is constructed simultaneously to represent its normal correlation fluctuation range. For feature pairs that show a stable synchronous relationship in historical data, such as a long-term high consistency between personnel activity and energy consumption readings in a certain area, the system marks them as highly coupled feature groups. This profile model is updated daily, discarding the earliest data and integrating the latest data to maintain the timeliness and stability of the profile structure. The real-time behavioral feature extraction module receives and processes the latest perceived data every minute, extracting a set of behavioral features consistent with the above profile. These features include: the number of personnel activities in the current minute, and the percentage of activity per unit time. Energy volatility, frequency of changes in door and window status, instantaneous changes in light levels, and fluctuations in ambient sound intensity are all standardized features and then sent to the behavior mutation detection module for further comparative analysis. The behavior mutation detection module operates on a combined deviation judgment mechanism. Its specific process is as follows: First, it checks whether the real-time feature value exceeds its normal fluctuation range defined in the normal profile (e.g., exceeding twice the standard deviation). Then, it determines whether two or more features deviate from their respective normal ranges simultaneously. In addition, for combinations that have been marked as highly coupled feature pairs, the module compares the correlation coefficient in the real-time data with the reference interval in the historical profile. If the absolute value decreases by more than a set threshold (e.g., a decrease of more than 0), the module will detect the deviation.If any of the above deviations fails to return to the normal range within six consecutive monitoring periods, it is considered a behavioral mutation event, and a corresponding early warning signal is generated to avoid false alarms caused by short-term anomalies.

[0045] To effectively address the risk of misjudgment due to insufficient data samples during the initial deployment phase of the system, an initial high-density observation period was set, lasting from seven to thirty days. During this period, the normal profile building module rapidly models at a frequency of ten minutes to one hour, while appropriately relaxing the behavioral deviation judgment criteria, setting the sensitivity threshold to twice the normal value, and gradually tightening it to the standard level in a linear or non-linear manner as the sample volume accumulates. If foreseeable abnormal behaviors occur during system operation, such as temporary overtime work at night, holiday activities, or emergencies, users can submit annotation information through the activity context receiving module, and the system will receive the annotation information. This information is then considered in conjunction with current behavioral characteristics and historical profiles. During the behavioral deviation determination process, the historical statistical weights are dynamically adjusted, or the tolerance range of the real-time behavioral baseline is appropriately expanded. Simultaneously, during subsequent profile updates, data from the corresponding time period is removed from the long-term profile sample to maintain the long-term stability and representativeness of the profile model. Furthermore, to further enhance the reliability of early warning signal determination, the system adds a power grid noise sensing module and a harmonic mode baseline construction module. The former collects voltage or current signals through several nodes deployed in the building's power system and performs frequency domain feature extraction, focusing on frequencies between 10 Hz and 2000 Hz. Background harmonic noise between Hertz, with particular attention to odd harmonic components such as the third, fifth, and seventh harmonics, is collected. The collected data is used to construct a harmonic mode baseline for the power supply system under normal operating conditions. This baseline is updated every four hours, forming a dynamically adjustable power environment stability model. When the behavioral mutation detection module generates an early warning signal, the early warning collaborative verification module is activated simultaneously. It determines whether there is an anomaly in grid noise by comparing the deviation between the harmonic characteristics at the current moment and the harmonic mode baseline. If the behavioral mutation and the harmonic noise deviation highly overlap in the time dimension, the system will comprehensively calculate the confidence factor and accordingly increase the risk of the early warning event. If the risk level is low and only harmonic deviation exists without abnormal behavior characteristics, the system will initiate enhanced monitoring of the behavior characteristics of the relevant area and generate alerts to help identify potential equipment malfunctions or hidden dangers. Regarding confidence quantification, the system uses a weighted calculation of the co-confidence factor based on the behavioral pattern deviation index and the noise deviation index. The sum of the weighting coefficients of the two indicators is one, and this can be flexibly adjusted according to the specific needs of different building operating environments. Through these methods, the system maintains high-confidence early warning capabilities while also considering false alarm suppression and flexible adaptation in complex scenarios, ensuring long-term stable operation.

[0046] Example 3: This example was conducted on a typical multi-functional office floor, which includes offices, meeting rooms, a tea room, and public corridors. The floor exhibited significant temporal and spatial variations in personnel activity, equipment operation, and energy consumption patterns. The experiment spanned three consecutive months, covering various typical time combinations, including weekdays and non-weekdays, and daytime and nighttime. The experiment included an information sensing module: Multiple sensors were strategically deployed within the selected office floor to comprehensively acquire multi-source basic sensing data reflecting building space utilization patterns, equipment operation time sequences, and baseline energy consumption. Passive infrared sensors were used for human activity, sampling once per minute to statistically analyze activity frequency and evenly cover all offices and public areas. Wireless magnetic sensors were used for door and window magnetic switches, monitoring their opening and closing status in real time and installed at all entrances and important windows. Energy meters, including at least one of smart electricity meters, smart water meters, and smart gas meters, were read every five minutes to calculate energy consumption rates and integrated into the floor's electrical distribution box and main water and gas valves. The light intensity sensor uses a photoresistor sensor, sampling every two seconds in areas with ample natural light and artificial lighting. The sound intensity sensor uses a sound pressure sensor, sampling every two seconds in areas susceptible to ambient noise. All sensor data is aggregated to the local data processing unit via the LoRaWAN IoT protocol, where unified timestamp calibration, data type conversion, and preliminary outlier filtering are performed to generate a standardized multi-source sensing data stream for analysis. The power grid noise sensing module deploys high-precision AC voltage and / or current sensors at the core distribution cabinet on this floor, passively acquiring real-time voltage and / or current signals from the building's AC power supply network. The data sampling frequency is set to 5kHz to ensure accurate capture of background harmonic noise characteristics within the preset 10Hz to 2kHz frequency band. Using a fast Fourier transform algorithm, the amplitude and phase information of key odd harmonic components such as the 3rd, 5th, and 7th harmonics are extracted and tracked in real time. These harmonic characteristics are considered sensitive indicators reflecting the state of the building's internal power system and potential load anomalies.

[0047] Routine Profile Construction Module: This module divides the floor space into 12 logical zones and further subdivides each 24-hour day into 48 half-hour time slots. Under each zone-time slot combination, the system adaptively constructs and dynamically updates routine behavior profiles for geographical areas and time slot combinations based on the historical sequence of multi-source basic sensing data from the past 30 days using statistical learning. Specifically, for each behavioral feature, such as the frequency of human activity, energy consumption rate, number of times doors and windows are opened and closed, mean light intensity, and standard deviation of sound intensity, its statistical parameters in the historical data of the same period are calculated, including the mean, median, standard deviation, and quantile interval. Simultaneously, by calculating different behaviors... The stable range of Pearson correlation coefficients between features is established to determine their stable correlation under normal conditions. Furthermore, it identifies high-frequency combinations of specific behavioral features in historical data, such as no activity in the office area between midnight and 6 AM and energy consumption rates below a baseline. The normal behavior profile is updated daily at 2 AM, incorporating the latest 24-hour data and removing the oldest data to adapt to the long-term evolution of building behavior patterns. The behavior mutation detection module acquires multi-source behavioral features under the current geographical area and time period in real time and compares them with the normal behavior profile. If the deviation of the current multi-source behavioral feature combination exceeds a preset first threshold (…), the detection module detects the mutation. =0.5, indicating a deviation index) and the duration exceeds the preset first time length ( =30 minutes), the system generates an early warning signal characterizing building safety risks. The judgment criteria for the degree of combined deviation include: single-dimensional anomaly, that is, at least two behavioral features in the current multi-source behavioral features deviate from their respective normal ranges defined in the normal behavioral profile, such as exceeding 2 standard deviation intervals; coordinated deviation between multiple dimensions and loss of correlation of historically related behavioral features, that is, the combination of behavioral features that were historically strongly correlated shows that the absolute value of its Pearson correlation coefficient has decreased by more than the preset second threshold. =0.6), for example, the frequency of personnel activity in an office area during the daytime on weekdays is usually strongly positively correlated with the rate of energy consumption. If personnel activity is frequent but energy consumption is abnormally low during a certain period, it is considered to be uncorrelated.

[0048] Harmonic Mode Baseline Construction Module: This module constructs and dynamically updates the harmonic noise mode baseline representing the building's power supply network under normal operating conditions, based on historical time-series data of background harmonic noise characteristics collected over the past 7 days, at a frequency of once every 4 hours. The baseline model establishes its normal fluctuation range by statistically analyzing the mean, standard deviation, and variation trend of the amplitude of each harmonic (3rd, 5th, and 7th). Early Warning Collaborative Verification Module: When the behavior change detection module generates an early warning signal, or when the degree to which the real-time monitored background harmonic noise characteristics deviate from the harmonic noise mode baseline exceeds a preset third threshold (… When the value is 0.4 (indicating noise deviation index), this module collaboratively analyzes the deviation of multi-source behavioral characteristics from background harmonic noise characteristics. Based on the results of the collaborative analysis, it raises or lowers the confidence level of the warning signal. The warning collaborative verification module calculates the collaborative confidence factor between the background harmonic noise characteristics and the behavioral change warning based on the degree of deviation of the background harmonic noise characteristics. ,in, This index represents the degree of deviation in the behavioral pattern as determined by the behavioral mutation detection module, and its value ranges from zero to one. The index represents the degree of deviation of the background harmonic noise as determined by the harmonic mode baseline construction module, and its value ranges from zero to one. and The weighting coefficients are preset and satisfy the following conditions: In this experiment, the following settings were made: =0.7, =0.3, when When the value exceeds the preset 0.6, the confidence level of the warning signal is raised to high risk; otherwise, it is lowered or remains at a moderate level. If only the background harmonic noise characteristics deviate from the harmonic mode baseline, a targeted risk warning is triggered or the monitoring of specific behavioral patterns is enhanced. The profile adaptation adjustment module: The cold start period is an initial high-density observation period set when the warning system is initially deployed or when a new geographical area is configured. The observation period lasts from seven to thirty days; in this experiment, it is set to seven days. During this observation period, the normal profile construction module initially constructs a normal behavioral profile at a frequency of once every ten minutes to once per hour. The sensitivity threshold judged by the behavioral change detection module is dynamically adjusted, initially set to... =0.8. As the amount of accumulated data increases (in terms of cumulative data points), the sensitivity threshold decreases linearly or non-linearly depending on the amount of accumulated data. In this experiment, it is calculated as a linear function. Gradually tighten to 0.5.

[0049] Activity Context Adjustment: When the activity context receiving module receives context annotation information for the combination of the current geographic area and the current time period, the behavior mutation detection module, when comparing the current multi-source behavioral features with the normal behavioral profile, adjusts the weight of historical statistical parameters in the comparison based on the type and intensity of the context annotation information, or increases the tolerance of the current dynamic baseline of behavior. In this experiment, the normal fluctuation range of relevant features can be temporarily expanded by 1.5 times. Furthermore, when updating the normal behavioral profile, based on the context annotation information, the multi-source behavioral feature data collected within the corresponding time period is excluded from the statistical samples of long-term profile updates. During the experiment, approximately 1.5TB of multi-source sensing data and power grid harmonic noise data were recorded. The experimental data of three typical scenarios are selected below for detailed analysis. Scenario 1: Identification of behavioral mutations in the early stage of a concealed fire. On a weekday afternoon, the process of a concealed fire from the initial smoke generation to the spread of the fire was simulated. Traditional smoke detectors only trigger after the smoke concentration reaches a preset threshold. This experiment focuses on the slight deviation of the initial behavioral pattern. See Table 1:

[0050] Table 1. Evolution Record of Behavioral Characteristic Deviation and Collaborative Early Warning Indicators (Office A)

[0051]

[0052] Between 15:51 and 16:00, although there was no activity in office A and the frequency of human activity was zero, which was on the edge of the normal range, the energy consumption rate (180 watts / minute) began to deviate significantly from the normal range defined by the historical profile (200-300 watts / minute). Furthermore, the light and sound intensities also showed an abnormal, slow but continuous deviation trend. The light intensity should be between 280-350 lux, and the sound intensity should be between 50-60 dB. At this time, the BPI calculated by the behavioral mutation detection module reached 0.7, exceeding the first threshold of 0.5. Between 16:01 and 16:10, the energy consumption rate further abnormally increased to 400 watts / minute, causing the BPI to reach 0.85. Simultaneously, the NPI of grid harmonic noise also increased from 0.12 to 0.25 during this period. Although it had not yet exceeded the independent warning threshold, its co-deviation with the behavioral pattern increased the co-confidence factor (…). The value reached 0.67, triggering a high-risk warning. This warning occurred approximately 15 minutes before the simulated smoke alarm was triggered, fully demonstrating the superior ability of this invention in early, concealed risk identification. It can provide early warnings based solely on the coordinated deviation of multi-dimensional behavioral characteristics, effectively compensating for the lag of traditional single-physical-quantity threshold alarms. Scenario 2: Cold Start Adaptability and Nighttime Authorized Activity Processing. This scenario verifies the system's adaptability during the initial deployment phase (cold start period) and its flexible adjustment capability to known authorized activities. See Table 2:

[0053] Table 2. Relationship between sensitivity threshold adjustment and early warning response during cold start (public corridor).

[0054]

[0055] During the cold start phase, the initial data volume of the system is insufficient, and the confidence level of normal profiles is low. To avoid false alarms, the profile adaptive adjustment module dynamically adjusts the sensitivity threshold of the behavior change detection module. As shown in the table above, on the first day of cold start, although the public corridor area showed... A deviation of 0.75, for example, indicates unusual personnel activity during normal non-working hours, but because the sensitivity threshold is set to 0.8, the system does not trigger an alert. As data accumulates, by the seventh day, the sensitivity threshold linearly tightens to 0.5. At this point, the same level of deviation ( This allows for accurate triggering of alerts, effectively addressing false alarms during the cold start period by dynamically adjusting sensitivity thresholds, and gradually improving recognition standards as the profile matures. Furthermore, on a certain night between 8:00 PM and 10:00 PM, a user input contextual annotation information about working overtime at night through the activity context receiving module. During this period, the frequency of human activity and energy consumption rate in office area B significantly exceeded its normal nighttime profile range. The accuracy reached 0.9. However, due to the system receiving contextual annotation information, the profile adaptive adjustment module increased the tolerance of the current dynamic baseline of behavior and excluded the data during this period from the long-term profile update samples. Ultimately, despite significant deviations in behavioral characteristics, the system did not trigger an alert, avoiding false alarms caused by authorized activities and ensuring the stability of the long-term profile. This demonstrated the system's flexible adaptability to drastic changes in known user behavior in complex scenarios. Scenario 3: Power grid harmonic noise-assisted verification and risk warning. This scenario verifies the role of power grid harmonic noise in improving alert confidence and independent risk warnings. See Table 3:

[0056] Table 3, Record of Results of Coordinated Early Warning of Behavior and Grid Noise Deviation.

[0057]

[0058] In the tea room area, between 10:00 and 10:15, a temporary equipment malfunction caused [a situation / incident]. Reaching 0.6, the system issued a moderate warning. Subsequently, between 10:16 and 10:30, the impact of the fault intensified, increasing the power grid harmonic noise. The value increased from 0.1 to 0.42, exceeding the preset third threshold of 0.4. At this point, the early warning collaborative verification module calculated... Reaching 0.589, close to the high-risk threshold, prompted the system to upgrade the warning level to high-risk. This indicates that grid harmonic noise, as independent physical evidence, can significantly improve the confidence of abnormal behavior warnings, especially when behavioral pattern deviations are not extreme but are accompanied by power system anomalies. Collaborative verification can effectively avoid misjudgments. Meanwhile, during the 11:00-11:15 period, the behavioral pattern in the conference room area was normal. However, power grid harmonic noise When the value reached 0.45, the system did not trigger a behavioral warning, but the warning collaboration verification module... The independent deviation triggered enhanced monitoring and risk alerts. After manual investigation, it was found that the aging of the power supply circuit inside an old projector in the conference room caused local circuit harmonics to exceed the standard. If this continued, it could lead to safety hazards. This case proves that the power grid noise perception module can not only help verify behavioral warnings, but also independently identify potential equipment failures or power system risks when the behavioral pattern does not show any abnormalities, providing an additional warning dimension for building maintenance at zero cost.

[0059] Example 4: This example combines Figures 1 to 3 This paper describes the implementation of an early warning system for building security. Figure 1 As shown, the behavior pattern analysis and adaptive early warning section consists of the following functional modules: First, the information perception module is responsible for collecting historical and real-time perception data, providing basic data input for subsequent analysis; the real-time behavior feature extraction module extracts current behavior features based on real-time perception data and sends them to the behavior mutation detection module; the normal profile construction module dynamically generates normal behavior profiles based on historical perception data and supports the dynamic update mechanism of the profile adaptive adjustment module, which can adaptively adjust the profile update rhythm and content based on the context annotations provided by the activity context receiving module, so as to flexibly adjust the image update rhythm during the cold start phase or sudden authorized activities; the interaction of the above modules converges to the behavior mutation detection module, which... Based on the matching relationship between current behavioral characteristics and normal behavioral profiles, the module identifies abrupt behaviors and generates behavioral abrupt behavior warning signals. These signals are then transmitted to the early warning collaborative verification module, which then enters the power grid harmonic collaborative verification section. The power grid harmonic collaborative verification section includes a power grid noise sensing module, which collects historical and real-time harmonic features. The harmonic mode baseline construction module then generates a harmonic mode baseline. This baseline is compared with the real-time harmonic features and, together with the behavioral abrupt behavior signal from the behavior analysis section, is sent to the early warning collaborative verification module for comprehensive judgment. Finally, the early warning collaborative verification module outputs a collaborative verification decision and transmits it to the final early warning decision module, generating a system-level safety early warning output to ensure a high-confidence early warning response in complex scenarios.

[0060] like Figure 2As shown, firstly, the real-time behavior feature extraction module requests the corresponding normal behavior profile from the normal profile construction module and receives the returned normal behavior profile. Then, it sends the current multi-source behavior features and profile to the behavior mutation detection module. The behavior mutation detection module compares the behavior features with the profile and determines whether the condition of "deviation exceeding the threshold and continuous exceeding the limit" is met based on the combined deviation degree and duration. If the condition is met, a preliminary behavior warning signal is generated. If the "continuous condition is not met", no alarm is output. After the preliminary behavior warning signal is generated, the behavior mutation detection module requests the power grid noise perception module to return the real-time power grid harmonic noise features. The power grid noise perception module transmits the noise features to the harmonic mode baseline construction module, which returns the harmonic noise mode baseline. Subsequently, the warning collaborative verification module performs collaborative analysis and calculates the confidence factor for the behavior mutation and harmonic deviation features. Finally, the warning collaborative verification module outputs the final joint warning signal and adjusts the confidence level, sending it to the user / system to complete the final output warning signal (confidence level adjustment).

[0061] like Figure 3 As shown in the figure, the indicator types include three categories: behavioral pattern deviation indicators ( ), background harmonic noise deviation index ( ) and co-confidence factor ( ),in, Represented by a solid curve, Indicated by dashed lines, Represented by dotted lines, all three indicators show an increasing trend in the vertical direction, indicating that the building gradually entered an abnormal behavior state during this period. The graph also marks the warning signal status, corresponding to the system judgment result at each time point, including normal, warning, and high-risk warning, which are identified by circles, squares, and triangles in the legend, respectively. The graph shows that at 15:30, , and All three indicators were at low levels, and the system status was normal; at 15:41, the three indicators began to rise, and the system status remained normal; at 15:51, A significant increase The system is approaching its set risk threshold, triggering an alert; as of 16:01, BPI, and All exceeded key thresholds, among which Close to 0.9 Approximately 0.3 The value has exceeded 0.6, triggering a high-risk warning.

[0062] Example 5: In this example, a building is under closed management from 23:00 to 06:00 the next day, allowing only on-duty security guards and registered authorized night construction workers to enter and exit. During this period, no other personnel are allowed to enter except for the aforementioned authorized personnel, and the building as a whole should be in a static operating state, with the power load maintained at the basic operating level. If unauthorized personnel enter, abnormal energy consumption fluctuations, or abnormally frequent electrical noise occur, it may constitute security risks such as illegal intrusion, unauthorized operations, or equipment failure. In this application scenario, the normal behavior profile construction module in the system is based on the key feature data such as entry and exit flow, lighting and elevator energy consumption load data, and voiceprint change rate collected in 10-minute sampling periods over 60 consecutive nighttime periods. It adopts a multi-dimensional joint probabilistic modeling method (optimal). A nighttime behavior profile is constructed using an empirical Bayesian network. The criteria for constructing the profile include, but are not limited to: an average entry and exit flow rate of less than 1.2 people per 10 minutes, energy consumption power stable within ±5% of the building's basic load, and a standard deviation of voiceprint variation not exceeding 1.1 times its historical average. This serves as the image basis for the nighttime static state. When any behavioral feature in the real-time data collected by the information sensing unit exceeds the 95% upper limit of the historical confidence interval for three consecutive sampling periods (i.e., within 30 minutes) and is not included in the registered and filed nighttime construction plan, the behavior mutation identification module will activate a multi-level judgment mechanism. First, it will calculate the Mahalanobis distance of the current data relative to the behavior profile. If the distance exceeds a set threshold (e.g., 2.5), the behavior will be marked as an atypical event and enter the temporary risk state observation interval.

[0063] The Mahalanobis distance threshold was initially set to 3.0 during model building and then dynamically adjusted based on the increase in sample size. As the sample size increases (each time it doubles), the model automatically reduces this threshold, eventually settling between 2.5 and 2.7. This threshold adjustment strategy is based on the convergence pattern of behavior observed in actual building operations; that is, behavioral characteristics tend to stabilize after data accumulation, enhancing the model's ability to distinguish outliers. Therefore, a smaller threshold is suitable to improve recognition sensitivity. The power grid noise identification module, running synchronously with the behavioral profiling, performs harmonic waveform analysis on the main distribution lines (such as the lighting circuit in Area A) connected to the system. This module captures current signals in real time at a sampling frequency of 3000 times per second, evaluating the amplitude of the third and fifth harmonics. If any harmonic band shows an abnormal increase in amplitude within two consecutive minutes, and there is no record of authorized lighting load changes, it is marked as potentially illegal power injection. Ingress behavior is commonly caused by unauthorized external equipment or short circuits in the wiring, resulting in abnormal feedback. When the behavior mutation identification module and the power grid noise identification module independently issue warning signals, the system will activate the two-way back-check mechanism in the collaborative warning module. This mechanism pushes structured information such as the timestamp of the abnormal event, event type, associated access control number, and personnel identification confidence level from the behavior side to the power grid side. At the same time, the power grid side provides feedback on load source distribution information, abnormal frequency spectrum, and similarity analysis results. If the time overlap between the two exceeds 80%, the spatial coordinate distance of the abnormal point is less than 15 meters, and the confidence level of each abnormal classification is higher than 0.7, the collaborative module calculates a collaborative confidence factor (with a value range of 0 to 1) based on the fusion algorithm. When the factor exceeds 0.65, it is considered a valid event, and the system will trigger a level 3 alarm, linking the audible and visual alarm devices deployed in the warning area, and pushing a structured event spectrum through the management platform for security personnel to make decisions. In the above application scenarios, after actual deployment and verification, the system identified 5 unauthorized device access events and 2 unauthorized nighttime construction trespass events during the 14-day trial operation. The overall false alarm rate was controlled within 0.9%, effectively verifying the practical application effect of the multimodal feature fusion mechanism and collaborative triggering strategy. In addition, when the administrator temporarily changes the nighttime permission policy, such as enabling the temporary construction mode, the system can update the behavior profile judgment logic in real time. That is, the construction status is embedded into the behavior profile modeling parameters through the permission change interface. The system then adjusts the upper and lower limits of the feature confidence interval to expand by 20%, and relaxes the Mahalanobis distance threshold to 3.2, thereby effectively avoiding misjudgment caused by policy changes.

[0064] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An early warning system for building security, characterized in that, The system includes: The information sensing module is configured inside the building to acquire multi-source basic sensing data that reflects the building's space utilization patterns, equipment operating time series, and energy consumption baseline status. The normal behavior profile construction module is electrically connected to the information perception module. It is used to adaptively construct and dynamically update the normal behavior profile of a specific geographical area and time period combination within a building based on the historical sequence of multi-source basic perception data and for specific geographical areas and time period combinations within the building using statistical learning. The normal behavior profile represents the multidimensional statistical distribution characteristics of multi-source basic perception data under historical normal conditions and the interrelationship between behavioral features. The real-time behavior feature extraction module is electrically connected to the information perception module and is used to acquire multi-source behavior features under the current geographical area and time period combination in real time. The behavior mutation detection module is electrically connected to the normal profile construction module and the real-time behavior feature extraction module. It is used to compare the current multi-source behavior features with the normal behavior profile in real time. When the combined deviation of the current multi-source behavior features exceeds the preset first threshold and the duration exceeds the preset first time length, an early warning signal representing the building safety risk is generated. The judgment criteria for the combined deviation include anomalies in a single dimension, coordinated deviations between multiple dimensions, and loss of correlation of historical related behavior features. The system includes a power grid noise sensing module, configured to passively acquire voltage and / or current signals at at least one location in the building's AC power supply network, and extract time-series data of background harmonic noise characteristics in a preset frequency band from 10 Hz to 2 kHz from the signals; a harmonic mode baseline construction module, electrically connected to the power grid noise sensing module, used to construct and dynamically update the harmonic noise mode baseline representing the building's power supply network under normal operating conditions based on historical time-series data of background harmonic noise characteristics using statistical learning; and an early warning collaborative verification module, electrically connected to the behavior mutation detection module and the harmonic mode baseline construction module, used to collaboratively analyze the deviation of multi-source behavioral characteristics and the deviation of background harmonic noise characteristics when the behavior mutation detection module generates an early warning signal, or when the deviation of the real-time monitored background harmonic noise characteristics from the harmonic noise mode baseline exceeds a preset third threshold, and based on the results of the collaborative analysis, to increase or decrease the confidence level of the early warning signal, or to trigger targeted risk warnings or enhance the monitoring of specific behavioral patterns when only the background harmonic noise characteristics deviate from the harmonic mode baseline; The early warning collaborative verification module calculates the collaborative confidence factor between the background harmonic noise characteristics and the behavioral change early warning based on the degree of deviation of the characteristics. Cooperative confidence factor The following relationship must be satisfied: ,in, This index represents the degree of deviation in the behavioral pattern as determined by the behavioral mutation detection module, and its value ranges from zero to one. The index represents the degree of deviation of the background harmonic noise as determined by the harmonic mode baseline construction module, and its value ranges from zero to one. and The weighting coefficients are preset and satisfy the following conditions: .

2. The early warning system for building security according to claim 1, characterized in that, The information sensing module acquires multi-source basic sensing data including one or more of the following: output signals from human activity sensors; opening and closing status signals of door and window magnetic switches; real-time energy consumption readings from energy meters, including the measured values ​​of electricity meters, water meters, and gas meters; ambient light signals from light intensity sensors; and ambient sound signals from sound intensity sensors.

3. The early warning system for building security according to claim 1, characterized in that, The statistical learning methods adopted by the normal profile construction module include one or more of the following: calculating the statistical parameters of each behavioral feature under the corresponding geographical region and time period combination in the same historical period, including but not limited to the mean, median, standard deviation, and quantile interval; calculating the stable variation range of the Pearson correlation coefficient between different behavioral features; or identifying specific behavioral feature combination patterns that frequently occur in the same historical period.

4. The early warning system for building security according to claim 1, characterized in that, The behavioral mutation detection module determines that the combined deviation of the current multi-source behavioral features exceeds a preset first threshold, including: at least two behavioral features in the current multi-source behavioral features simultaneously deviate from their respective normal ranges defined in the normal behavioral profile, or a combination of behavioral features that have historically been strongly correlated shows that the absolute value of its Pearson correlation coefficient has decreased beyond a preset second threshold, and the duration of the deviation or loss of correlation exceeds a preset first time length, and then generates a warning signal.

5. The early warning system for building security according to claim 1, characterized in that, The normal profile construction module updates periodically, from once per hour to once per day, based on newly collected multi-source basic perception data and real-time behavioral characteristics, in order to adapt to the long-term evolution of building behavior patterns.

6. The early warning system for building security according to claim 1, characterized in that, Also includes: The profile adaptive adjustment module is electrically connected to the normal profile construction module. It is configured to set an initial high-density observation period when the early warning system is initially deployed or when a new geographical area is configured. The observation period lasts from seven to thirty days. During this observation period, the normal profile construction module initially constructs a normal behavior profile at a frequency of once every ten minutes to once every hour. The profile adaptive adjustment module dynamically adjusts the sensitivity threshold for the behavior change detection module to judge behavior changes based on the accumulation of profile data. The sensitivity threshold decreases linearly or non-linearly based on the amount of data accumulated.

7. The early warning system for building security according to claim 1, characterized in that, Also includes: The activity context receiving module is configured to receive context annotation information input by authorized users for a specific geographical area and future time period, which represents the expected atypical activity type; The system also includes a profile adaptation adjustment module, which is electrically connected to the normal profile construction module, the real-time behavior feature extraction module, and the activity context receiving module. The system is configured such that when the activity context receiving module receives context annotation information for the combination of the current geographic region and the current time period, the behavior mutation detection module, when comparing the current multi-source behavior features with the normal behavior profile, adjusts the weight of historical statistical parameters in the comparison based on the type and intensity of the context annotation information, or increases the tolerance of the current behavior dynamic baseline. Furthermore, when updating the normal behavior profile, the system excludes multi-source behavior feature data collected within the corresponding time period from the statistical samples of long-term profile updates based on the context annotation information.

8. The early warning system for building security according to claim 1, characterized in that, In the information sensing module, the human activity sensor is a passive infrared sensor, the door and window magnetic switch is a wireless magnetic sensor, the energy meter includes at least one of smart electricity meter, smart water meter and smart gas meter, the light intensity sensor is a photoresistor sensor, and the sound intensity sensor is a sound pressure sensor.

Images