Integrated smart safety management system

The integrated smart safety management system addresses fragmented building safety systems by using AI for real-time risk prediction and control, enhancing disaster response and management in complex structures.

WO2026235252A1PCT designated stage Publication Date: 2026-11-12SM TECHNICAL GROUP LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/KR2025/095402
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-06-10
Filing Date
2025-06-12
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Existing building safety systems lack integrated data sharing, proactive risk prediction, and customized responses, leading to fragmented and passive fire and power management that can escalate into major disasters in complex structures.

Method used

An integrated smart safety management system that collects data from multiple sensors, uses AI for comprehensive analysis, and performs real-time control and prediction of risks, providing customized evacuation routes and efficient energy management.

Benefits of technology

Reduces response time, minimizes damage, and improves rescue success rates by predicting and actively responding to power anomalies, fires, and smoke spread, while ensuring high accuracy and interoperability with smart city infrastructure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure KR2025095402_12112026_PF_FP_ABST
    Figure KR2025095402_12112026_PF_FP_ABST
Patent Text Reader

Abstract

The present invention relates to an integrated smart safety management system capable of performing real-time detection, prediction, and automatic control for various safety threats such as power abnormality, arc occurrence, fire, smoke spread, and the like. The system comprises: a plurality of sensor units; a central integrated control server; an intelligence-based analysis engine unit; a smart device control unit; a user and firefighting interface; and an integrated learning management unit. Each risk factor is preemptively predicted through an artificial intelligence-based analysis engine, and is dealt with in association with a smart home controller, a fire shutter, a ventilation fan, a door, and the like. User location-based evacuation route guidance, rescue support heatmap provision, situation-specific electric load control, and switchover strategy implementation are possible, and optimal control specialized for each building environment is realized through continuous AI learning. The present invention relates to a highly integrated safety system that can be applied to smart cities and industrial facilities as well as residential, commercial, and public buildings.
Need to check novelty before this filing date? Find Prior Art

Description

Integrated Smart Safety Management System

[0001] The present invention relates to an integrated smart safety management system capable of predicting in advance and performing real-time responses to various safety threats that may occur within buildings or industrial facilities, such as power anomalies, electric arcs, fires, and smoke spread. In particular, the present invention relates to a technology that simultaneously realizes the protection of life and property by having an AI-based analysis engine evaluate risk levels in real time based on various environmental data collected from multiple sensors, such as power quality information, temperature, smoke concentration, gas concentration, human presence, and door opening / closing status, and by automatically generating control commands—such as smoke spread suppression, evacuation guidance, power switching, and arc interruption—and transmitting them to smart devices and control units within the building. This system is configured in a form that fuses IoT-based sensor networks, AI algorithms, digital twin technology, and smart control technology, and is applicable to various types of residential facilities, commercial facilities, public buildings, plants, etc. Furthermore, unlike existing individual safety systems, it belongs to the field of next-generation safety management technology that enables the complex recognition of various risks and optimal response through a central control architecture capable of integrated analysis and control.

[0002] Traditional building safety systems are installed as separate modules, such as fire detectors, gas detectors, evacuation announcements, and power backup devices, and most of them rely on passive detection and alarm methods. For example, it is common practice to simply emit an alarm sound when a smoke sensor detects a concentration above a certain level, or to switch to emergency power via a manual transfer switch in the event of a power outage. These systems have fundamental limitations: they lack data sharing between sensors, cannot provide customized responses based on user location or building structure, and cannot take proactive measures before risks such as fires or electrical arcs occur. Furthermore, issues such as harmonics, surges, voltage imbalances, overheating of connections, and electrical arcs caused by degraded power quality are difficult to effectively detect or preemptively prevent with existing systems. Particularly in complex structures such as high-rise buildings, smart homes, and industrial facilities, these problems can accumulate, leading to the simultaneous occurrence of multiple risk factors and posing a significant risk of escalating into major disasters. Therefore, there is a growing need for a next-generation integrated safety system capable of analyzing data from various sensors, making predictions and decisions based on AI, and responding in real-time in conjunction with smart devices.

[0003] The technical problem that this invention aims to solve is to overcome the structural limitations of existing fragmented and passive fire response and power management systems, and to implement an integrated safety management platform capable of predicting various threat factors in advance and actively responding to power quality anomalies, arcs, fires, and smoke spread. In particular, this invention aims for a structure that collects data from multiple heterogeneous sensors, comprehensively analyzes them based on time series to quantitatively determine risk levels, and performs warnings and control before an accident occurs through an AI-based prediction model. Furthermore, another major objective is to improve the efficiency and accuracy of rescue operations by providing customized evacuation routes to individual users through real-time analysis of user location information, door open / closed status, and evacuation route availability, and by providing information on rescue priority targets and smoke spread to fire terminals. In addition, this invention aims to enable efficient energy management through an advanced power switching algorithm that readjusts electrical loads according to various operating modes and includes energy saving and recovery modes.

[0004] To solve the above technical problem, the present invention provides an integrated smart safety management system composed of an integrated sensor network unit, a central integrated control server, an intelligent analysis engine unit, a smart device linkage control unit, a user and firefighting interface unit, and an integrated learning management unit. The integrated sensor network unit collects complex data such as power, temperature, smoke, gas, airflow, vibration, illuminance, and human body detection, and the central integrated control server performs situational awareness, decision-making, and predictive analysis functions after preprocessing and normalizing this data. The intelligent analysis engine unit is composed of four main engines, including smart power management, arc prediction, fire prediction, and smoke spread prediction; each engine uses deep learning, time series analysis, and digital twin-based simulation technologies to precisely analyze different risks and make control decisions. The smart device linkage control unit connects the analysis results to actual control to perform automatic switching, smart home device control, and fire shutter and door control, while the user / firefighting interface is responsible for functions such as real-time alarm transmission, evacuation route guidance, and providing rescue support information. In addition, the Integrated Learning Management Unit enables continuous performance improvement and field-specific optimization by performing data collection and cleaning, AI model training and transfer learning, and the creation of user-customized models.

[0005] Specific details of other embodiments are included in the detailed description and drawings.

[0006] According to the system of the present invention, comprehensive response to multiple risks becomes possible by implementing existing separate fire detection, power switching, arc prediction, and evacuation guidance systems into a single integrated platform. For example, if power quality degradation is detected, an arc prediction algorithm is activated in advance, and if a high-risk determination is made, circuit interruption and emergency power switching are performed. In the event of a fire, consistent automated response is carried out, ranging from smoke spread prediction and control by zone to the provision of user-customized evacuation routes, while real-time heatmaps and information on remaining personnel are also provided to support rescue operations. These integrated functions contribute to reducing response time, minimizing damage, and improving rescue success rates, while ensuring high performance in arc prediction, fire prediction, and smoke spread prediction accuracy. Furthermore, power switching and evacuation times are shortened, and the system is capable of interoperability with smart city infrastructure, allowing it to be expanded into a city-wide safety network.

[0007] FIG. 1 is a block diagram showing the overall configuration of an integrated smart safety management system according to one embodiment of the present invention.

[0008] Various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The advantages and technical features of the present invention, and the methods for achieving them, may become clear through the embodiments described together with the drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various forms. These embodiments are provided to ensure that the disclosure of the present invention is complete and to clearly convey the technical concept of the present invention to those skilled in the art to which the present invention pertains. Accordingly, the scope of the present invention is defined only by the claims.

[0009] The shapes, sizes, arrangements, quantities, etc., shown in the drawings illustrating embodiments of the present invention are merely examples to aid understanding and may differ in actual application. Throughout the specification, the same reference numerals refer to the same components. Furthermore, in describing the present invention, specific descriptions of known technologies may be omitted where necessary, as such descriptions could unnecessarily obscure the core content of the invention.

[0010] Terms such as "comprising," "having," and "consisting of" as used in this specification should be interpreted to mean that additional components or steps may be included unless specifically stated otherwise. Even if a component is expressed in the singular, it may be interpreted to include the plural unless explicitly stated otherwise.

[0011] Where numerical limits or criteria are not specified for the interpretation of each component, the configuration should be understood to include an interpretation within the usual tolerance range.

[0012] When describing the positional relationship of components, for example, when expressions such as 'on top of,' 'below,' 'adjacent to,' or 'inside' are used, one or more other components may be interposed between them unless phrases such as 'directly' or 'right therein' are present.

[0013] Furthermore, terms such as "first," "second," etc., are merely identifying terms used to distinguish different components and should not be interpreted as restricting a specific order or quantity. Accordingly, the first component may be replaced by the second component, and vice versa.

[0014] The technical features included in various embodiments of the present invention may be combined or combined with one another, in whole or in part, and those skilled in the art can easily derive various variations in which each feature can be implemented independently or in an interconnected form. The combination of such embodiments can be expanded in various ways depending on the technical interconnected structure or functional requirements.

[0015]

[0016] FIG. 1 is a block diagram showing the overall configuration of an integrated smart safety management system according to one embodiment of the present invention.

[0017] Referring to FIG. 1, the integrated smart safety management system (10000) of the present invention is composed of an integrated platform in which six main components are organically connected. The first component, the integrated sensor network unit (1000), is responsible for detecting the safety status of the entire building in real time. The power quality sensor group (1100) performs the functions of monitoring the status of commercial power and emergency power through voltage / current sensors, measuring voltage fluctuation rates, harmonic distortion rates, and frequency deviations through power quality monitors, and measuring contact resistance between power wires and junction boxes in real time through contact resistance sensors.

[0018] The sensor group (1200) for fire and arc detection collects more precise fire-related information. The smoke sensor performs precise measurements in the range of 0.110 mg / m³ using a light scattering method, and the temperature sensor is capable of precise measurements in the range of -20 degrees to 100 degrees in increments of 0.1 degrees. The IR thermal imaging sensor visualizes the surface temperature distribution in a non-contact manner, and the CO / VOC sensor measures the concentration of carbon monoxide and volatile organic compounds. The acoustic sensor detects arc-related acoustic signals in the range of 20 Hz to 20 kHz, thereby enabling the early detection of signs of arc occurrence. The environmental monitoring sensor group (1300) comprehensively monitors environmental changes such as humidity, illuminance, airflow, human detection, and vibration to maximize the system's situational awareness capabilities.

[0019] The second core component, the central integrated control server (2000), acts as the brain of the system and consists of a processor (2100), a storage unit (2200), and a communication unit (2300). The processor (2100) performs real-time collection, preprocessing, and normalization of multi-sensor data through the integrated data processing unit, and evaluates risk levels by comprehensively analyzing the overall system status through the situation awareness engine. The decision engine adjusts priorities among multiple modules and establishes integrated control strategies, while the predictive analysis engine is responsible for the risk prediction function using AI models. The storage unit (2200) consists of a real-time data storage, a history database, an AI model storage, and a configuration information storage, systematically managing all information required for system operation.

[0020] The third component, the intelligent analysis engine unit (3000), consists of AI analysis engines specialized for each safety function. The smart power management engine (3100) performs power quality analysis, load importance evaluation, management of five operating modes, and pattern analysis through learning algorithms. The arc prediction analysis engine (3200) predicts and controls arc occurrence in advance through a two-stage hybrid AI structure based on Bi-LSTM and XGBoost and digital twin simulation. The fire prediction analysis engine (3300) predicts fire risk in advance through the calculation of five fusion indicators and cause classification based on Random Forest and CNN, and the smoke spread prediction engine (3400) predicts smoke spread and generates personalized evacuation routes using graph neural networks and ConvLSTM.

[0021]

[0022] The smart power management engine (3100) is a core module responsible for intelligent switching between commercial power and emergency power, and differential power supply for each electrical load. The commercial power input unit receives power from a general commercial power grid, supplies it stably to the system, and provides commercial power status information to the power quality monitor unit. It is designed with a structure that includes surge protection circuits, voltage stabilization circuits, current limiting circuits, etc., to monitor the status of the commercial power in real time and immediately detect problems when they occur.

[0023] The emergency power input section is responsible for receiving backup power in the event of commercial power interruption or instability. It includes functions for battery charging and discharging management, battery status monitoring, stable power supply during emergencies, and charge / discharge control to optimize battery life. The emergency power source can be configured in various forms, such as lithium-ion batteries, lead-acid batteries, and supercapacitors. It includes a Battery Management System (BMS) to monitor the battery status in real time and control operation under optimal conditions.

[0024] The automatic transfer switch is a core device that switches between commercial and emergency power, performing the transition according to processor commands. It minimizes temporary power interruptions during switching to provide millisecond-level response, offers monitoring and feedback on the switching status, and allows for the simultaneous use of both power sources in hybrid mode under specific situations. Composed of a combination of mechanical relays and electronic relays (SSRs), it applies the optimal switching mechanism based on load characteristics.

[0025] The processor is a core component responsible for central control and decision-making across the entire system. It comprehensively analyzes all sensor data to determine operating modes and execute switching logic. It performs functions such as calculating electrical load importance scores and assigning groups, generating load control signals, issuing non-essential load cutoff commands, executing algorithms and analysis models, determining power switching points, and managing and optimizing the overall system state. It is implemented using high-performance microcontrollers or embedded processors and is equipped with a Real-Time Operating System (RTOS) to ensure time-deterministic operation.

[0026] The system has multiple operating modes to provide optimized responses to various environmental conditions and power states. Standard mode is the system's default operating state, balancing energy efficiency and user convenience; it utilizes commercial power as the primary power source, sets commercial power monitoring as the first cycle, and maintains the battery charge status within the SoC 1 threshold range. Energy saving mode focuses on minimizing power consumption and maximizing cost efficiency; the system switches to this mode when entering peak electricity pricing hours or when the battery charge status exceeds the 2nd SoC threshold.

[0027] Emergency Readiness Mode is designed for preemptive response in situations with a high probability of power outages. It calculates a risk score by comprehensively analyzing power quality factors, weather hazards, surrounding power outage conditions, and training data. The system switches to this mode when the overall risk score exceeds the first risk threshold and a danger signal is detected in at least one of the categories; this activates rapid battery charging, shortens the power quality monitoring cycle, and reduces switching response latency. Emergency Operation Mode focuses on maintaining essential functions by relying on the battery in situations where commercial power is cut off or severely unstable, while Recovery Mode safely restores the system to a normal state after commercial power is restored.

[0028] The importance score of an electrical load is calculated based on various factors such as occupancy status, environmental conditions, spatial importance, and time information. It is calculated using the mathematical formula: Importance Score = w1 X Occupancy + w2 X Environmental Factor + w3 X Time Weight + w4 X Spatial Importance. The value of each factor is normalized to a range between 0 and 1, and the weights are dynamically adjusted according to the current operating mode. Based on the calculated importance score, the electrical load is assigned to one of three groups: E-line (essential load), SE-line (semi-essential load), or NE-line (non-essential load). A control signal is generated by applying differentiated control parameters to each group.

[0029]

[0030] The arc prediction analysis engine (3200) is an intelligent system that monitors and predicts arcs that may occur in power facilities, including solar power generation facilities, in real time. Multiple sensor units are installed on solar arrays, junction boxes, power lines, etc., to collect various electrical and physical data in real time. Current sensors measure current values, abnormal currents, current imbalances, and current fluctuation trends for each circuit, while voltage sensors detect voltage values, voltage drops, and voltage imbalances at each point. Temperature sensors identify local heat generation and abnormal temperature rises in areas such as the inside of the junction box, wire connection points, and array surfaces, and contact resistance sensors measure contact resistance values ​​and increases or decreases in resistance relative to temperature changes at the connection terminals between the power lines and the junction box.

[0031] The central control server is the core control unit of the system that receives and analyzes data from multiple sensor units to predict the likelihood of arc occurrence and performs control decisions. The communication unit receives data from the sensor units and transmits analysis results and warning signals to user terminals, while the processor is an computational control unit that executes all information processing and decision logic within the system, and includes sub-modules of the data processing unit, arc prediction unit, digital twin simulation unit, and control decision unit.

[0032] The data processing unit performs preprocessing tasks such as normalization of sensor data, removal of outliers, construction of time-series waveforms, and feature extraction. Key derived features include contact resistance change-related features (rate of change of contact resistance per unit of time), temperature-resistance correlation coefficient (indicator of the correlation between temperature and resistance), current-voltage mismatch (degree of deviation from Ohm's law), high-frequency noise components (energy ratio of a specific frequency band in the Fast Fourier Transform result of the current waveform), and current ripple index (degree of minute fluctuation in DC current).

[0033] The arc prediction unit is implemented through a two-stage hybrid AI structure. The first stage is time-series-based risk prediction, utilizing a Bi-directional Long Short-Term Memory (Bi-LSTM) neural network model. The Bi-LSTM model receives multivariable time-series data, such as current, voltage, temperature, and contact resistance, within a fixed time window and outputs an 'arc risk score' between 0 and 1. The second stage is a risk pattern classification algorithm that utilizes XGBoost (eXtreme Gradient Boosting) to classify risk states into three classes (Class 0: Normal, Class 1: Abnormal Warning, Class 2: High Risk).

[0034] The Digital Twin Simulation Unit is activated when the prediction results from the Arc Prediction Unit satisfy preset conditions, performing physics-based verification of the probability of arc occurrence. It utilizes a digital twin model that implements the physical structure and electrical characteristics of actual power generation facilities, such as solar arrays, solar junction boxes, and power lines, in a virtual space. Built on CAD data, the physical properties of each component—including material, thermal conductivity, electrical resistance, and coefficient of thermal expansion—are precisely set. Real-time data collected from multiple sensor units is mapped to the digital twin model, allowing the virtual model to accurately reflect the current state of the actual physical world.

[0035] The control decision unit performs a final control decision by synthesizing the analysis results of the arc prediction unit and the digital twin simulation unit. It prevents power generation interruptions or equipment damage caused by unnecessary control measures through a double verification mechanism in which a control command is generated only when the risk score of the arc prediction unit is above a preset threshold and the simulation result of the digital twin simulation unit satisfies the conditions for arc occurrence. Depending on the severity of the situation, it is classified into four stages—Warning, Caution Required, Danger, and Urgent—and performs appropriate control measures for each.

[0036]

[0037] The fire prediction analysis engine (3300) is an intelligent system that goes beyond conventional simple fire detection systems and can predict signs of danger in advance and respond preemptively before a fire occurs. The multi-sensor fusion detection unit collectively includes various types of sensors such as temperature, smoke, carbon monoxide (CO), carbon dioxide (CO2), infrared (IR) thermal imaging, sound, humidity, illuminance, VOC (volatile organic compounds), and fine dust.

[0038] The temperature sensor precisely measures temperature changes in the range of -20 to 100 degrees in increments of 0.1 degrees, and the smoke sensor detects minute changes in smoke concentration ranging from 0.1 to 10 mg / m³ to capture the initial combustion stage. The CO sensor monitors carbon monoxide concentration in real-time in the range of 0 to 1000 ppm to detect risks in incomplete combustion or confined spaces, and the IR thermal imaging sensor visualizes surface temperature distribution non-contactually to precisely detect localized temperature rises and hot spots. The acoustic sensor detects abnormal acoustic signals such as explosions, bursts, and burning sounds in the range of 20 Hz to 20 kHz to capture auditory signs of a fire.

[0039] The intelligent fire analysis and response unit consists of three sub-components: a risk feature extraction unit, a fire cause identification unit, and an intelligent response judgment unit. The risk feature extraction unit fuses data collected from multiple sensors to calculate composite indicators such as the heat-smoke composite index (TSI), gas-acoustic variation rate index (GSI), IR spectrum variation rate (IRI), humidity-temperature relative index (HTI), and illuminance variation pattern (LPI), and classifies fire risk into levels 0 to 4 based on these indicators.

[0040] TSI distinguishes between simple cooking smoke and fire precursors by analyzing the ratio between rapid temperature rise and increase in smoke concentration, while GSI detects risks of gas leaks or electrical short circuits by analyzing the ratio between increased combustible gas concentration and the occurrence of loud bangs or bursts. IRI precisely detects localized overheating points by analyzing the trend of surface temperature distribution changes over time using IR thermal imaging sensors, and HTI comprehensively considers seasonal factors and the indoor environment by reflecting the increased probability of fire under low humidity and high temperature conditions. LPI captures power system anomalies or the moment of ignition by detecting sudden power outages or rapid changes in brightness caused by flames.

[0041] The fire cause identification unit is an intelligent module that accurately identifies the cause of a fire by integrating and analyzing sensor time-series data, IR thermal images, and user environment information. It consists of four sub-modules: an environmental context correction module, a multimodal feature extractor, a Random Forest-based cause classifier, and a CNN-based auxiliary image analyzer. The Random Forest-based multimodal classifier learns extracted multidimensional features to classify the cause as one of electrical leakage, cooking appliance overheating, smoking, heating device malfunction, or other abnormal heat sources, while the CNN-based auxiliary image analyzer analyzes the location, size, shape, and diffusion pattern of hot spots in IR images to make an auxiliary determination as to whether the cause is cooking activity or an electrical leakage phenomenon.

[0042] The Intelligent Response Decision Unit is a core module that determines the optimal response scenario by synthesizing the analysis results of the Risk Feature Extraction Unit and the Fire Cause Identification Unit. It evaluates the severity of the situation by comprehensively analyzing risk levels, cause classification results, the current status of smart home devices, and user environment information. Response scenarios are divided into four stages, from Stage A (Concern) to Stage D (Danger), and differentiated response measures are automatically executed for each stage. In Stage A, smartphone warning notifications are sent and air purifiers and ventilation systems are automatically activated; in Stage B, induction cooktops are automatically shut off and CCTV recording begins; in Stage C, smart sprinklers are prepared and the front door is automatically unlocked; and in Stage D, sprinklers are automatically sprayed and an automatic API report is made to the fire department.

[0043]

[0044] The smoke diffusion prediction engine (3400) is an integrated fire response system that predicts the spread of smoke in advance when a fire occurs, effectively suppresses it, and simultaneously provides an optimized evacuation route to occupants in the building in real time. Multiple sensor units detect and collect various environmental information in real time, such as smoke concentration, temperature, air flow, CO or VOC concentration, presence of human bodies, and door opening / closing status.

[0045] Smoke sensors detect early signs of fire by precisely detecting changes in the concentration of smoke particles in the air through light scattering or ionization methods, while heat and temperature sensors identify the extent of fire spread and the location of heat sources by detecting local temperature increases within the space. CO sensors measure carbon monoxide concentration to determine the state of incomplete combustion and monitor the diffusion of toxic gases, and VOC sensors contribute to early fire cause analysis by detecting volatile organic compounds. Human body detection sensors determine the presence of users within the space using PIR (Passive Infrared) or ultrasonic methods to determine rescue priorities, and door opening / closing detection sensors determine the availability of evacuation routes by recognizing the open and closed status of doors and windows.

[0046] The processor of the central control server executes advanced algorithms that analyze and process multidimensional time-series data received from multiple sensor units in real time to establish comprehensive judgments and control strategies regarding fire situations. During the real-time sensor data collection and preprocessing process, various sensor data such as smoke, heat, CO, VOCs, and human body detection are received in real time; noise filtering and valid value extraction are performed according to the characteristics of each sensor type; and abnormal signs are accurately detected through correlation analysis between multiple sensors within the same space.

[0047] The spatial risk assessment algorithm comprehensively analyzes factors such as the rate of increase or decrease in smoke concentration, the rate of temperature rise, hazardous gas concentration, and the presence of humans for each section to calculate a risk level of 0 to 4 in real time, and improves prediction accuracy by applying AI-based weights that reflect the time-series change rate and the correlation between spaces. The smoke diffusion path prediction algorithm utilizes a Graph Neural Network (GNN) to model each section as a node and adjacent connections as edges to learn the possibility of smoke transmission over time, and performs spatiotemporal smoke concentration time-series prediction using ConvLSTM to output a smoke distribution heatmap after 30 seconds to 3 minutes.

[0048] The smoke diffusion prediction algorithm is an advanced AI-based simulation system that operates after a fire occurs. It precisely predicts the diffusion path and time of smoke by utilizing a combination of deep learning models, such as Graph Neural Networks (GNN), Convolutional Recurrent Neural Networks (ConvLSTM), and Long-Term Short-Term Memory Networks (LSTM). As input variables, it comprehensively utilizes real-time sensor data such as smoke concentration, temperature, airflow direction, wind speed, and CO and VOC concentrations, building metadata such as floor plans, inter-floor structure, compartment division information, and window and door location information, as well as time information such as the time of smoke detection and the start time of sensor response.

[0049] Through GNN, each section is modeled as a node and adjacent connections such as doors, windows, and stairs are modeled as edges to learn the possibility of smoke transmission over time, and the connection strength between nodes is set as weights reflecting physical characteristics such as opening size, pressure difference, and air flow velocity. By utilizing ConvLSTM to perform spatiotemporal-based smoke concentration time series prediction, changes in smoke distribution in 3D space are continuously predicted along the time axis, and smoke distribution heatmaps after 30 seconds, 1 minute, and 3 minutes are accurately output.

[0050] The compartment-specific control decision logic dynamically determines the closure of fire shutters, automatic opening or closing of doors, deployment of smoke barriers, and the operation direction and intensity of ventilation fans for specific compartments by integrating smoke spread prediction results with current risk levels, and establishes an integrated control strategy that considers the relationships and connectivity between compartments. The central control server stores building compartment information in the form of a digital map and performs selective control by dynamically classifying the status of each compartment based on smoke spread prediction results.

[0051] In automatic compartment-level recognition, the server subdivides each zone based on a Building Information Modeling (BIM) or building floor plan database, precisely calculates the actual detection range, distance between spaces, and connectivity based on sensor locations, and quantitatively defines the influence range of each sensor using Voronoi diagrams. In the compartment status classification logic, the Multi-Criteria Decision Analysis (MCDA) technique is applied to establish complex risk judgment criteria, including cases where the smoke prediction arrival time is less than 60 seconds, human detection is confirmed, or there is no alternative evacuation route.

[0052] The evacuation route optimization system is a user location-based inter-floor evacuation system in which a central control server estimates each user's location with centimeter-level precision through multi-sensor fusion, such as PIR, BLE, app location signals, Wi-Fi fingerprinting, and Ultra-Wideband (UWB) signals from user terminals, and calculates a customized evacuation route for each individual user in real time by considering smoke spread prediction results. By applying an evacuation optimization technique based on Multi-Agent Reinforcement Learning, each user is modeled as an independent agent, and distributed route planning is performed using the principles of Swarm Intelligence.

[0053] By applying dynamic weights to the D* Lite algorithm, an extension of the A* algorithm, path search that adapts to real-time environmental changes is performed, and the shortest path and the safest path are calculated simultaneously by applying the Dijkstra algorithm and the Floyd-Warshall algorithm in parallel. In evacuation path optimization, a cost function is optimized by integrating multidimensional elements such as a mixed distance metric that simultaneously considers the Euclidean distance and Manhattan distance from the current location to the target point, an accessibility matrix that models the open state of doors and shutters as 0-1 binary variables, a risk function that models the smoke concentration and diffusion speed of the path as a Gaussian distribution, and a congestion index that calculates the density of adjacent users and traffic volume using a fluid dynamic crowd flow model.

[0054]

[0055] The smart device linkage control unit (4000) is an execution layer that implements the analysis results into actual physical control. The power control module (4100) performs automatic switching between commercial power and emergency power in millisecond intervals and differential load control classified into E-line (essential), SE-line (semi-essential), and NE-line (non-essential). During the switching process, temporary power interruption during switching is minimized, switching status monitoring and feedback are provided, and in specific situations, the two power sources are used simultaneously in hybrid mode.

[0056] The fire response control module (4200) is responsible for the automatic control of smart home devices and fire extinguishing equipment. It transmits and executes control commands to various smart home devices (IoT devices), such as smart plugs, lights, induction cooktops, boilers, air purifiers, gas valves, electric curtains, smart windows, smart door locks, and sprinklers. It performs situation-adaptive control that reflects user safety and lifestyle patterns through a situation-based control policy, and during nighttime hours, it prioritizes external warning sounds instead of opening doors, or automatically executes more aggressive blocking commands when the user is absent.

[0057] The smoke blocking control module (4300) suppresses smoke spread and secures evacuation routes through fire shutters, ventilation systems, and door control. For a section designated as an evacuation route, if the estimated time to reach smoke spread is 120 seconds or more and the section is included in the path that the current user must pass through for evacuation, it performs multimodal guidance enhancement through automatic door opening, maintaining passable entrances, turning on LED-based evacuation route lighting, and activating a voice guidance system. If a section is classified as a blocking target, it immediately implements active blocking measures such as closing the fire shutter, blocking the HVAC system to the section, stopping the operation of the positive pressure ventilation system, and deploying additional water curtain barriers.

[0058] The user / firefighting interface section (5000) provides interfaces specialized for general users and firefighters, respectively. The general user interface (5100) provides real-time safety status monitoring, alarm reception, and evacuation route guidance functions via a mobile app, and performs safety status display and simple control within the household via a wall-mounted panel, as well as multimodal alarms and guidance via voice / haptic feedback. In the event of a dangerous situation, it provides immediate warnings to the user through vibration, voice, and visual notifications, and automatically switches to emergency response mode if the user is absent or does not respond.

[0059] The firefighting dedicated interface (5200) provides information on the overall danger situation of the building and priority rescue targets through the firefighting terminal, and provides real-time smoke spread heatmaps and location information of remaining personnel through the rescue support system. Through the on-site control function, control of specific areas through manual intervention is possible, and the on-site situation can be accurately assessed by receiving real-time fire and smoke spread status information from the central control server. By receiving location information of remaining personnel within the building and a list of priority rescue targets, the efficiency of rescue operations is maximized, and the safety of rescue personnel is ensured by checking information on closed sections and inaccessible areas in real time.

[0060]

[0061] The integrated learning management unit (6000) is a core component responsible for the intelligence and personalization of the system and consists of data collection and refinement (6100), AI model training and optimization (6200), and customized model creation (6300). In the data collection and refinement unit, all data related to power, arc, fire, and smoke is collected in an integrated manner, and data quality management is performed through outlier removal, missing data interpolation, and noise filtering. Risk characteristics, fire cause classification results, response history, etc. are collected from the intelligent fire analysis response unit and processed into a training dataset.

[0062] In AI model training and optimization, knowledge is shared among AI models in each functional module through federated learning, and model performance is enhanced by utilizing external datasets through transfer learning. Continuous model updates are performed using real-time data through online training, and training performance is secured even in data-scarce environments by transferring knowledge from historical accident cases or external generalization models. In customized model generation, individual models are created based on building types and usage patterns, and user profile-based optimization is performed that reflects individual lifestyle patterns.

[0063] The power management system determines system operating modes and executes control logic based on algorithms trained on electrical load occupancy patterns, power usage history, power outage history, and weather history data. Occupancy pattern learning predicts future usage probabilities by analyzing the frequency and duration of usage by day of the week and time of day for each space, while power usage history learning constructs an energy consumption prediction model by analyzing usage and patterns by electrical load. Power outage history learning builds a power outage prediction model by analyzing the duration, duration, and pre-outage power quality indicators of past power outages, and weather history data learning analyzes the relationships between weather conditions, power quality, and power outages.

[0064] The arc prediction system employs a semi-supervised learning approach, enabling continuous learning even in environments where actual arc occurrence data is scarce. Even after initial training, the model is continuously improved by combining a large volume of unlabeled data collected during operation with a small amount of labeled data. The fire prediction system supports individual profile-based learning that reflects the user's residential structure, cooking habits, presence of pets, and ventilation patterns, and adjusts alarm sensitivity by incorporating user-specified false alarm cases.

[0065]

[0066] It goes without saying that various embodiments of the present invention may be combined with one or more others to form a new embodiment.

[0067]

[0068] An integrated smart safety management system according to one embodiment of the present invention may include: an integrated sensor network unit composed of a plurality of sensors that detects power quality, arc occurrence, fire risk, and smoke spread status in real time; an intelligent analysis engine unit that analyzes data collected from the integrated sensor network unit to calculate an integrated risk level and establish a response strategy for each situation; a smart device linkage control unit that controls power control, fire response, and smoke blocking devices in an integrated manner according to the analysis results of the intelligent analysis engine unit; a user / firefighting interface unit that provides situation information and response guidance to users and firefighters; and an integrated learning management unit that learns collected data to continuously improve an AI model.

[0069] According to another feature of the present invention, the intelligent analysis engine unit may include a smart power management engine through power quality analysis and load-specific importance evaluation, an arc prediction analysis engine that performs Bi-LSTM and XGBoost-based arc prediction and digital twin verification, a fire prediction analysis engine that performs fire risk prediction and cause identification through multi-sensor fusion, and a smoke spread prediction engine that performs graph neural network-based smoke spread prediction and evacuation path optimization.

[0070] According to another feature of the present invention, the smart device interlocking control unit may include a power control module that performs automatic switching between commercial power and emergency power and differential power supply for each load, a fire response control module that performs automatic control of smart home devices and fire extinguishing equipment, and a smoke blocking control module that performs smoke blocking control of fire shutters, ventilation systems, and doors.

[0071]

[0072] Although the present invention has been described above with specific details such as specific components, limited embodiments, and drawings, this is provided only to aid in a more comprehensive understanding of the invention and the invention is not limited to the embodiments, and a person skilled in the art to which the invention belongs can make various modifications and variations from this description.

[0073] Accordingly, the scope of the present invention is not limited to the described embodiments, and all variations equivalent to or equivalent to the claims set forth below, as well as the claims themselves, shall be considered to fall within the scope of the concept of the present invention.

Claims

1. An integrated sensor network unit composed of multiple sensors that detects power quality, arc generation, fire risk, and smoke spread status in real time; An intelligent analysis engine unit that analyzes data collected from the integrated sensor network unit to calculate the integrated risk level and establishes situation-specific response strategies; A smart device interlocking control unit that integrates and controls power control, fire response, and smoke blocking devices according to the analysis results of the intelligent analysis engine unit mentioned above; A user / firefighting interface section that provides situation information and response guidance to users and firefighters; and An integrated smart safety management system comprising an integrated learning management unit that continuously improves an AI model by learning from collected data.

2. The intelligent analysis engine unit comprises: a smart power management engine through power quality analysis and load-specific importance evaluation; an arc prediction analysis engine that performs Bi-LSTM and XGBoost-based arc prediction and digital twin verification; a fire prediction analysis engine that performs fire risk prediction and cause identification through multi-sensor fusion; and a smoke spread prediction engine that performs graph neural network-based smoke spread prediction and evacuation path optimization; comprising the integrated smart safety management system described in claim 1.

3. The smart device interlocking control unit described above comprises: a power control module that performs automatic switching between commercial power and emergency power and differential power supply for each load; a fire response control module that performs automatic control of smart home devices and fire extinguishing equipment; and a smoke blocking control module that performs smoke blocking control of fire shutters, ventilation systems, and doors; an integrated smart safety management system as described in claim 1.