Wireless intelligent smoke sensing and power utilization safety cooperative monitoring network system

Abstract

The invention relates to the technical field of intelligent safety monitoring, and discloses a wireless intelligent smoke sensing and power utilization safety cooperative monitoring network system, which comprises a plurality of multi-mode sensing units configured by intelligent sensing nodes in a distributed manner and used for synchronously acquiring chemical fingerprints, electromagnetic acoustics and environmental data; an edge processor performs edge analysis through a preset AI model, generates and broadcasts a local event vector after identifying a persistent or mutable risk, performs multi-source information cross validation by calculating a regional evidence factor, and evaluates a quantized regional threat level in combination with local confidence. The gateway device may also perform hierarchical intervention according to threat levels, and predict a risk propagation path based on an architectural space environment map model. Through combination of edge intelligence and center cooperation, high-precision identification of early-stage hidden fire disasters is realized, the false alarm rate is significantly reduced through multi-source information cross validation, and the reliability of early warning and the accuracy of intervention are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent security monitoring technology, specifically to a wireless intelligent smoke detector and electrical safety collaborative monitoring network system. Background Technology

[0002] Fire safety is a critical issue in building, industrial, and residential environments. Currently, widely deployed automatic fire alarm systems primarily rely on smoke detectors and heat detectors installed within monitored areas. These traditional detectors play a fundamental role in detecting fires caused by open flames or significant smoke, providing valuable early warning time for evacuation and firefighting. Their technological maturity and widespread application are widely recognized.

[0003] However, with the rapid increase in electrical equipment and the growing complexity of electrical circuits, the proportion of fires caused by electrical faults has been rising year by year, becoming one of the main sources of modern fires. The limitations of traditional detection systems are becoming increasingly apparent when facing such fires. Traditional smoke and heat detectors are consequence-based detection devices; their alarm mechanisms rely on the passive sensing of physical phenomena generated after combustion (i.e., smoke particles or a sharp increase in ambient temperature). For electrical fires, there is often a long, hidden early incubation stage before visible smoke is produced or the ambient temperature rises significantly. This stage may manifest as the overheating and decomposition (pyrolysis) of cable insulation releasing specific chemical gases, or as weak but continuous arcing discharge in the circuit. These early, non-visible warning signs of risk are difficult for traditional detectors to effectively capture.

[0004] Furthermore, existing detection systems, relying on relatively simple physical quantities for judgment, lack sufficient anti-interference capabilities in complex environments and are easily affected by non-fire factors such as cooking fumes, water vapor, and dust, resulting in a persistently high false alarm rate. Frequent false alarms not only cause unnecessary resource allocation and manpower waste, but more seriously, they gradually erode users' trust in the alarm system, potentially leading to alarms being ignored when a real fire occurs, resulting in irreparable losses. At the same time, most detectors in existing systems are single-function sensing terminals, lacking local intelligent analysis and decision-making capabilities. They cannot conduct multi-dimensional comprehensive assessments of the on-site environment and struggle to improve the overall accuracy and reliability of the monitoring network through inter-node collaboration. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a wireless intelligent smoke detector and electrical safety collaborative monitoring network system, which solves the problems of insufficient detection capability and high false alarm rate caused by the inability of traditional fire alarm systems to detect early, hidden signs of electrical fires and reliance on single information for judgment.

[0006] To achieve the above objectives, the present invention provides a wireless intelligent smoke detector and electrical safety collaborative monitoring network system. The first aspect of the present invention provides a wireless intelligent smoke detector and electrical safety collaborative monitoring network system, the system comprising multiple intelligent sensing nodes and a gateway device.

[0007] Each of the multiple intelligent sensing nodes is configured with:

[0008] The multimodal sensing unit is used to simultaneously collect chemical fingerprint data, electromagnetic and acoustic data, and conventional environmental data within the monitoring area.

[0009] An edge processor, connected to the multimodal sensing unit, receives and analyzes data collected by the multimodal sensing unit using a pre-set AI model. Upon identifying potential risks, it generates a structured local event vector. This local event vector is a multidimensional data packet containing node identifier, location, timestamp, risk type probability vector, local confidence level, and local strength.

[0010] A communication module, connected to the edge processor, is used to broadcast the local event vector.

[0011] The gateway device is configured with:

[0012] The collaborative reasoning module is used to receive local event vectors broadcast by one or more of the intelligent sensing nodes. Based on preset collaborative reasoning rules, the collaborative reasoning module performs comprehensive analysis on the received one or more local event vectors, calculates regional corroboration factors, and then combines the local confidence levels in the local event vectors to assess a quantified regional threat level.

[0013] A collaborative intervention module, which is connected to the collaborative reasoning module, is used to trigger a preset intervention action when the regional threat level exceeds a preset intervention threshold.

[0014] This technical solution generates structured local event vectors through edge computing at intelligent sensing nodes. Then, the gateway device aggregates the local event vectors from multiple nodes for collaborative reasoning, realizing the transformation from single-point perception to networked collaborative decision-making and improving the accuracy and timeliness of risk identification.

[0015] Preferably, the multimodal sensing unit specifically includes:

[0016] The chemical fingerprint sensing unit uses broadband absorption spectroscopy to identify a specific combination of characteristic gas molecules released by an object in the early stages of pyrolysis, thereby acquiring the chemical fingerprint data. This method enables early detection of risks before visible smoke forms.

[0017] An electromagnetic acoustic sensing unit is used to simultaneously monitor high-frequency electromagnetic field signals and acoustic signals in a specific frequency band generated when weak electric arc discharge occurs in electrical equipment or when insulation materials break down, and to acquire the electromagnetic and acoustic data. This method is used to detect hidden electrical fire hazards.

[0018] Preferably, the preset collaborative reasoning rule specifically defines the calculation steps for the regional corroboration factor. In a specific embodiment, for any node... Generated local event vectors Its regional supporting factors The calculation method is as follows:

[0019]

[0020] in, Within a preset time window, nodes The set of neighboring nodes; For nodes Reported event risk types and nodes Correlation weighting coefficients between event risk types; For nodes Local confidence in the reported local event vector; For nodes The local intensity in the reported local event vector; It is a decay function whose value depends on the nodes. With nodes Spatial distance between The time difference between the event and the occurrence of the event .

[0021] Preferably, the gateway device is further configured with a risk propagation prediction module. The risk propagation prediction module is activated after the collaborative reasoning module confirms the regional threat level. It uses the identified risk source locations as initial nodes and, based on a pre-set building space environment map model, it iteratively calculates the future distribution and propagation path of the risk in different spatial units through a state transition model.

[0022] In one specific embodiment, the state transition model is used to calculate any spatial unit. In the next moment risk value The calculation content is as follows:

[0023]

[0024] in, spatial unit At the present moment The risk value; spatial unit The risk naturally dissipates at a certain coefficient. For all spatial units and adjacent spatial units A set; Adjacent spatial units At the present moment The risk value; Risk from space units Towards spatial units The probability of propagation and transfer.

[0025] Preferably, the intervention actions triggered by the collaborative intervention module include performing tiered intervention. Specifically, the tiered intervention is configured as follows: when the regional threat level exceeds the first-level threshold, a source-blocking intervention is performed, sending instructions to the intelligent control device associated with the risk source to cut off the risk source; when the regional threat level exceeds the second-level threshold, a regional emergency response intervention is performed, coordinating with fire-fighting or security equipment within the region.

[0026] In one specific embodiment, the edge processor is further configured with an event generation and decision module. This module receives multiple risk type probability vectors with temporal information continuously output by the AI ​​model and stores them in a temporary vector sequence buffer. The module uses preset temporal analysis rules to determine whether the vector sequence in the temporary buffer meets a preset risk growth pattern or mutation pattern. Only when the vector sequence meets the risk growth pattern or mutation pattern does the edge processor generate a local event vector and trigger the communication module to broadcast it. This scheme is used to suppress invalid event reporting caused by instantaneous signal fluctuations.

[0027] Preferably, the edge processor is further configured with a collaborative sampling control program. The collaborative sampling control program has at least two preset working states that are linked to the AI ​​model analysis results, specifically including: a low-power inspection state, in which, when the probability of all risk types output by the AI ​​model is lower than a first preset threshold, the collaborative sampling control program sets the chemical fingerprint sensing unit and the electromagnetic acoustic sensing unit in the multimodal sensing unit to a first preset sampling frequency; and a high-frequency verification state, in which, when the probability of any risk type exceeds the first preset threshold, the collaborative sampling control program switches the chemical fingerprint sensing unit and the electromagnetic acoustic sensing unit to a second preset sampling frequency higher than the first preset sampling frequency, for high-density data collection and verification of potential risks. This scheme is used to balance the monitoring sensitivity and power consumption of the node.

[0028] Preferably, the gateway device is further configured with a model and rule base update module. This module receives update instructions from an external cloud platform via the communication interface. Based on the update instructions, the module performs online updates and parameter tuning of the collaborative inference rules used to calculate regional corroboration factors in the collaborative inference module, as well as the AI ​​models in the multiple intelligent sensing nodes. This solution enables the system to have remote iteration and optimization capabilities.

[0029] This invention provides a wireless intelligent smoke detector and electrical safety collaborative monitoring network system. It has the following beneficial effects:

[0030] 1. This invention acquires multi-dimensional data using a multimodal sensing unit at the intelligent sensing node, and generates structured local event vectors through preliminary analysis by an AI model on the edge processor. Furthermore, the collaborative reasoning module in the gateway device can aggregate local event vectors from multiple nodes and perform cross-validation by calculating regional corroboration factors. This overcomes the shortcomings of false alarms caused by environmental interference or malfunctions of a single sensor, ensuring that system-level decisions are based on reliable information confirmed by multi-point collaboration, rather than judgments from isolated nodes.

[0031] 2. The gateway device in this invention is equipped with a risk propagation prediction module, which can predict the future propagation path of the risk after the risk is confirmed by using a state transition model, thereby issuing predictive warnings to potentially affected areas. At the same time, the collaborative intervention module performs graded intervention based on the quantified regional threat level, and can accurately trigger response actions of different intensities, such as cutting off the source or linking regional emergency equipment, according to the severity of the risk, making the intervention measures more targeted and effective.

[0032] 3. This invention, by configuring a collaborative sampling control program on the intelligent sensing node, enables the node to adaptively switch between low-power inspection state and high-frequency verification state based on the analysis results of the local AI model. This significantly optimizes the balance between power consumption and monitoring performance of the wireless node. In addition, the model and rule base update module configured in the gateway device allows for online updates and optimizations of the AI ​​model deployed in all nodes and the gateway's own collaborative inference rules. This enables the entire system to have the ability to continuously learn and evolve remotely, and to adapt to constantly changing environments and risk characteristics. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0034] Figure 2 This is a schematic diagram of the hardware structure of the intelligent sensing node of the present invention;

[0035] Figure 3 This is a schematic diagram of the internal structure of the multimodal sensing unit of the present invention;

[0036] Figure 4 This is a schematic diagram of the edge processor software module structure of the present invention;

[0037] Figure 5 This is a schematic diagram of the gateway device software module structure of the present invention;

[0038] Figure 6 This is a schematic diagram of the system collaborative working method of the present invention. Detailed Implementation

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] See attached document Figure 1 This invention provides a wireless intelligent smoke detector and electrical safety collaborative monitoring network system, comprising multiple intelligent sensing nodes and a gateway device. The multiple intelligent sensing nodes are distributed and deployed in one or more monitoring areas for on-site data acquisition and edge analysis. The gateway device is centrally deployed to receive and process data reported by all intelligent sensing nodes. In one embodiment, the multiple intelligent sensing nodes communicate via a wireless ad hoc network and ultimately aggregate the data to the gateway device.

[0041] In one embodiment, each intelligent sensing node is equipped with a multimodal sensing unit, an edge processor, and a communication module. The multimodal sensing unit synchronously collects chemical fingerprint data, electromagnetic and acoustic data, and general environmental data within the monitored area. The edge processor connects to the multimodal sensing unit and receives and processes the data. The edge processor analyzes the data using a pre-built AI model and generates a local event vector when a potential risk is identified. The local event vector is a structured data packet containing fields such as: node identifier, node location, event timestamp, risk type probability vector, local confidence level, and local strength. The communication module connects to the edge processor and broadcasts the local event vector to the wireless network.

[0042] The gateway device is configured with a collaborative reasoning module and a collaborative intervention module. The collaborative reasoning module receives local event vectors broadcast by one or more intelligent sensing nodes. Based on preset collaborative reasoning rules, the module analyzes the received local event vectors, calculates regional corroboration factors, and, combined with the local confidence levels in the local event vectors, assesses a quantified regional threat level. The collaborative intervention module, connected to the collaborative reasoning module, triggers one or more preset intervention actions when the regional threat level exceeds a preset intervention threshold.

[0043] The information processing flow of the system of this invention is as follows: First, multiple distributed intelligent sensing nodes perform front-end sensing and edge analysis. When any node identifies a potential risk, it generates and broadcasts a local event vector. Subsequently, a centralized gateway device receives one or more such vectors, performs collaborative reasoning, and confirms the risk and assesses its threat level through cross-validation of multi-source information. Finally, the gateway device makes a decision and executes corresponding intervention actions based on the assessed threat level, forming a complete closed loop from distributed sensing to centralized collaborative decision-making and then to precise intervention.

[0044] In one specific implementation, the step of the collaborative reasoning module calculating the regional corroboration factors is a concrete manifestation of the preset collaborative reasoning rules. For any node... Generated local event vectors Its regional supporting factors The calculation method is as follows:

[0045]

[0046] in, Within a preset time window, nodes The set of neighboring nodes; For nodes Reported event risk types and nodes Correlation weighting coefficients between event risk types; For nodes Local confidence in the reported local event vector; For nodes The local intensity in the reported local event vector; It is a decay function whose value depends on the nodes. With nodes Spatial distance between The time difference between the event and the occurrence of the event .

[0047] See attached document Figure 2In one specific embodiment, the intelligent sensing node is designed as an integrated, low-power, standalone device. All its electronic components are distributed on a printed circuit board (PCB) and encapsulated in a durable housing. The hardware of the intelligent sensing node includes an edge processor, multimodal sensing units, a communication module, a power module, and memory.

[0048] The edge processor is the core control and computing unit of the intelligent sensing node, and it is specifically implemented as a low-power microcontroller unit (MCU) or an embedded microprocessor unit (MPU). The selection of the edge processor depends on the computational complexity of the pre-built AI model. It is responsible for executing firmware programs, processing raw data from multimodal sensing units, running AI models, and executing control logic for event generation decisions and collaborative sampling.

[0049] The multimodal sensing unit is physically integrated on a printed circuit board or connected to an edge processor via a dedicated data interface. Its constituent chemical fingerprint sensing unit, electromagnetic acoustic sensing unit, and sensor array for acquiring general environmental data are integrated into a single layout to ensure spatial consistency of multi-source data acquisition.

[0050] The communication module is a sub-GHz band wireless transceiver that supports long-range, low-power communication protocols such as LoRaWAN or Zigbee. It connects to the edge processor via a Serial Peripheral Interface (SPI) or a Universal Asynchronous Receiver / Transmitter (UART) interface, and is responsible for encoding, modulating, and transmitting local event vectors generated by the edge processor, as well as receiving instructions from the gateway device.

[0051] The memory includes non-volatile flash memory and volatile random access memory (RAM). Non-volatile flash memory is used to store the operating system of the storage node, pre-configured AI model parameters, and the node's unique identifier. Random access memory is used for temporary data storage during system operation, including the space required to provide a temporary vector sequence buffer for the event generation and decision-making module.

[0052] The power module is responsible for providing a stable operating voltage to all components of the intelligent sensing node. In one embodiment, the power module has a built-in power management unit (PMU) that supports dual power inputs, including an internal lithium battery for long-term operation and an external DC power interface, and has the function of automatically switching between the two inputs.

[0053] The power module is responsible for providing electrical power for the stable operation of the equipment. In one specific implementation, to ensure continuous operation in various environments, the power module is designed as a highly reliable power supply unit with redundant inputs and seamless switching capabilities. It includes an interface for connecting a built-in backup battery (e.g., a lithium battery) and an external DC power interface for connecting to an AC adapter or a centralized power supply bus.

[0054] The gateway device is designed as a high-performance central processing unit, with its hardware configured within an industrial-grade chassis suitable for inter-device use. The gateway device's hardware components include a central processor, large-capacity memory, multiple communication interfaces, and intervention control interfaces.

[0055] The central processor is based on a high-performance embedded motherboard, which is equipped with a multi-core processor that provides parallel processing capabilities to run the software programs of the collaborative reasoning module, the risk propagation prediction module, and the model and rule base update module simultaneously, and to process concurrent data streams from all intelligent sensing nodes in the network.

[0056] Massive storage includes random access memory (RAM) and solid-state drives (SSDs). RAM provides the central processor with operating space and is used to cache local event vectors received from various intelligent sensing nodes. SSDs are used to install the operating system, store the building space environment graph model, the collaborative reasoning rule base, the historical event database, and AI model update packages downloaded from the cloud platform.

[0057] Multiple communication interfaces provide the gateway device with data exchange capabilities. These include: a wireless communication interface for receiving data from smart sensing nodes, such as a LoRaWAN concentrator; a wired Ethernet interface for connecting to an external cloud platform; and a 4G / 5G cellular network communication module for redundancy or use when the wired network is unavailable.

[0058] The intervention control interface is the physical output portion of the collaborative intervention module. In one embodiment, this interface includes a programmable array of reed relays for outputting switching signals to directly drive external devices such as circuit breakers or audible and visual alarms. Furthermore, the interface includes an RS-485 bus interface and an Ethernet interface supporting the Modbus-TCP protocol for sending standardized control commands or data messages to the building automation system (BAS) or fire alarm controller (FACP).

[0059] See attached document Figure 3The multimodal sensing unit is an integrated sensor module that includes a chemical fingerprint sensing unit, an electromagnetic acoustic sensing unit, and sensors for collecting general environmental data. All sensors in this module are centrally configured in a physical layout to ensure synchronous acquisition of multi-dimensional data from the same monitoring point.

[0060] A chemical fingerprint sensing unit is used to acquire information about chemical substances released by an object during the initial stages of overheating or pyrolysis. In one specific embodiment, this unit is based on non-dispersive infrared (NDIR) absorption spectroscopy. Its internal structure includes a broadband infrared light source that emits a continuous mid-infrared spectrum, an optical chamber communicating with the ambient air, an optical filter wheel or filter array configured with multiple narrowband filters, and a corresponding high-sensitivity photodetector array. When ambient air enters the optical chamber, specific gas molecules (e.g., carbon monoxide (CO) and total hydrocarbons (THCs) absorb infrared light at their characteristic wavelengths. By measuring the attenuation of infrared light intensity at each characteristic wavelength after passing through the filters, the concentration of the corresponding gas can be calculated according to the Beer-Lambert law. This unit simultaneously acquires the concentration data of multiple characteristic gases and combines them into a multi-dimensional vector, which is the chemical fingerprint data used to characterize the combustion composition of substances in an early fire.

[0061] The electromagnetic acoustic sensing unit is used to acquire high-frequency physical signals caused by electrical faults. This unit includes a high-frequency current transformer and a microelectromechanical system (MEMS) microphone. The high-frequency current transformer is wrapped non-contactly around the monitored power cable. Its internal circuitry is designed as a high-pass filter, insensitive to 50 / 60 Hz power frequency currents, but capable of accurately capturing and outputting time-series data of transient high-frequency electromagnetic field signals ranging from several kilohertz (kHz) to several megahertz (MHz) generated by weak arc discharges or insulation breakdowns.

[0062] In one specific embodiment of the present invention, a high-pass filter is integrated into the internal front-end analog circuit of the electromagnetic acoustic sensing unit. The core technical objective of this filter is to accurately filter out low-frequency background noise generated by power frequency (50 / 60Hz) power supply and its harmonics from the acquired wide-spectrum electromagnetic or acoustic signals, while selectively allowing signals with distinct high-frequency characteristics generated by early electrical faults such as arc discharge and electric sparks to pass through. This design significantly improves the signal-to-noise ratio (SNR), providing a clean, fault-feature-rich data input for subsequent edge AI model analysis, thereby significantly enhancing the sensitivity and accuracy of detection.

[0063] Meanwhile, MEMS microphones are used to acquire acoustic signals within the monitored area. The acquired broadband acoustic signals, after being digitized by an analog-to-digital converter (ADC), are processed by a digital bandpass filter. This filter has a specific passband, for example, from 20kHz to 100kHz. The purpose of this processing step is to filter out low-frequency ambient noise audible to the human ear and specifically extract acoustic characteristic signals in the ultrasonic frequency band generated by electric arc discharge or thermal cracking of insulating materials.

[0064] A key technical feature of the electromagnetic acoustic sensing unit is that its sampling process for high-frequency electromagnetic field signals and acoustic signals is synchronously triggered by the same clock signal emitted by the edge processor. This synchronization mechanism ensures that the two acquired time-series data are precisely aligned in timestamps, providing a data foundation for subsequent AI models to perform correlation analysis of cross-modal signal features.

[0065] In addition, the multimodal sensing unit integrates temperature, humidity, and barometric pressure sensors. These sensors collect routine environmental data, providing background parameters of the environmental state for risk analysis by the AI ​​model on the edge processor. All raw or preprocessed data from the sensors within the multimodal sensing unit is transmitted to the edge processor for unified analysis and processing via an internal data bus, such as a Serial Peripheral Interface (SPI) or an Integrated Circuit Interconnect Bus (I2C).

[0066] See attached document Figure 4 The edge processor is the computing and control core of the intelligent sensing node. Multiple collaborative software modules run on it, including a pre-built AI model, an event generation and decision-making module, and a collaborative sampling control program.

[0067] In one specific implementation, the pre-built AI model is a lightweight multi-stream fusion neural network. The input to this model is a unified time-series feature tensor, which is composed of three parts of data concatenated within each sampling period:

[0068] Multidimensional gas concentration vector from the chemical fingerprint sensing unit;

[0069] Frequency domain characteristics of high-frequency electromagnetic signals and frequency domain characteristics of acoustic signals from electromagnetic acoustic sensing units after processing with Fast Fourier Transform (FFT);

[0070] The model uses scalar data such as temperature and humidity from conventional environmental sensors. Its main structure includes parallel multi-path one-dimensional convolutional layers (1D-CNN) to extract local temporal features from time-series data of different modalities, such as chemical, electromagnetic, and acoustic data.

[0071] The features extracted by each convolutional layer are then flattened and concatenated, and fed into a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU) layer to model the temporal dependencies of cross-modal features. The final layer of the model is a Softmax activation function layer, whose output is a... A probability vector of risk types. Each dimension of this vector corresponds to a predefined risk type (e.g., electropyrolysis, arc fault, slow smoldering of solids), and its value represents the probability that the current state belongs to that risk type. The sum of the values ​​of all dimensions in the vector is 1.

[0072] The event generation decision module performs a second, time-series-level screening of the risk type probability vectors continuously output by the AI ​​model to suppress single-point misjudgments caused by transient interference. Internally, this module maintains a fixed-length... The first-in-first-out (FIFO) queue serves as a temporary vector sequence buffer, used to store the most recently accessed data. The probability vector output by the AI ​​model. This module continuously analyzes the vector sequence in the cache using preset time-series analysis rules. In one embodiment, the time-series analysis rules include:

[0073] Risk growth pattern judgment: For any risk type If its probability value Continuous within the cache Each sampling period It satisfies the condition of monotonically increasing, and the linear regression slope of the sequence exceeds a preset growth slope threshold. If so, it is determined that the risk growth pattern is met.

[0074] Mutation pattern determination: for any risk type If its probability value Probability value at the previous time step The difference, i.e. Exceeding a preset mutation threshold If the vector sequence in the buffer satisfies at least one of the above modes, the event generation decision module determines that the potential risk is persistent or mutable, and triggers the edge processor to generate a local event vector containing the current node state.

[0075] The collaborative sampling control program manages the operating state of the multimodal sensing unit based on the real-time analysis results of the AI ​​model, balancing monitoring accuracy and node power consumption. This program is implemented as a finite state machine (FSM) with at least two preset operating states.

[0076] One is a low-power inspection state. In this state, the values ​​of all dimensions in the risk type probability vector output by the AI ​​model are below a first preset threshold. At that time, the program sends control commands to the multimodal sensing unit, setting the chemical fingerprint sensing unit and the electromagnetic acoustic sensing unit to a first preset sampling frequency, for example, sampling once every 60 seconds.

[0077] Another type is the high-frequency verification state. This occurs when the value of any dimension in the risk type probability vector output by the AI ​​model exceeds a first preset threshold. Upon this, the program immediately switches to high-frequency verification mode. In this mode, the program sets the two sensing units to a second preset sampling frequency higher than the first preset sampling frequency, for example, sampling 100 times per second, to obtain high-density data for rapid confirmation and detailed analysis of potential risks. When all dimensions of the risk type probability vector continuously remain at the first preset threshold... The following exceeds a preset cooldown time After that, the state machine will switch back to the low-power inspection state.

[0078] See attached document Figure 5 The software system running on the gateway device includes a collaborative reasoning module, a risk propagation prediction module, a collaborative intervention module, and a model and rule base update module. These modules together constitute the system's central decision-making and management functions.

[0079] The collaborative reasoning module is responsible for performing correlation analysis and comprehensive evaluation on discrete local event vectors received from various intelligent sensing nodes to generate a global, high-confidence regional threat assessment. In one specific implementation, when this module receives a threat vector from a node... Local event vectors At that time, it first calculates the regional corroboration factor of the event. The calculation process has been explained in the aforementioned system architecture formula; its physical meaning is elaborated here: Correlation weight coefficient Stored as a preset × Matrix, where This represents the total number of predefined risk types. This matrix defines the corroboration strength between different risk types. For example, if an electrical pyrolysis event reported by a node has a high correlation with an arc fault event reported by a neighboring node, the corresponding weight coefficient will be close to 1; while the correlation with an unrelated event (such as electromagnetic interference caused by routine equipment start-up and shutdown) will be close to 0. Attenuation function. Specifically, it can be implemented using a Gaussian decay function or an exponential decay function to ensure that events with greater distance and longer time intervals provide lower corroborating weight.

[0080] Calculate the regional corroborating factors Then, the collaborative reasoning module further calculates the final regional threat level. This level of confidence is calculated by combining the local confidence of the event itself with the strength of corroboration from neighboring nodes. The formula is as follows:

[0081]

[0082] in, In response to the incident The assessed regional threat level; For the event The local confidence contained in the local event vector; These are the regional supporting factors calculated above; This is a weighting coefficient, ranging from 0 to 1, used to adjust the proportion of local confidence and regional corroborating factors in the final assessment result. This calculation process transforms an isolated local event judgment into a regional threat assessment that has undergone networked cross-validation.

[0083] The risk propagation prediction module is activated after the collaborative reasoning module confirms the regional threat level and identifies the location of the risk source. The core function of this module is to predict the future spatial propagation path and distribution of risk based on a pre-set built space environment map model and through iterative calculations using a state transition model.

[0084] In one implementation, the architectural space environment graphical model is a directed weighted graph. Among them, the vertex set Each vertex in Each vertex represents a single physical spatial unit, such as a room, a corridor, or a shaft. It includes a set of attributes, such as space volume, combustible material load, and ventilation conditions. (Edge set) Each edge in Representative spatial unit and There is a physical connection between them, such as a door, a vent, or a cable run-through. Each edge The weighting represents the risk from spread to The difficulty or probability of transition. This graphical model is initially constructed based on the building's BIM (Building Information Modeling) data or CAD drawings during the initial system deployment.

[0085] The state transition model is based on this graphical model and performs iterative calculations. For any spatial unit... In the next moment risk value The calculation formula is:

[0086]

[0087] wherein, is the risk value of the spatial unit at the current moment ; is the natural dissipation coefficient of the risk of the spatial unit ; is the set of all adjacent spatial units to the spatial unit ; is the risk value of the adjacent spatial unit at the current moment ; is the propagation transfer probability of the risk from the spatial unit to the spatial unit .

[0088] By repeatedly executing this iterative calculation, the risk propagation prediction module can obtain the distribution map of the risk values at multiple future time steps (for example, in the future 1 minute, 5 minutes, 10 minutes) in all spatial units, thereby forming a visual propagation path prediction.

[0089] The collaborative intervention module is the execution unit of the system decision-making. This module continuously monitors the regional threat level output by the collaborative reasoning module, and compares it with the primary intervention threshold and the secondary intervention threshold stored in the local configuration to perform hierarchical intervention. In one implementation, when the value of the regional threat level exceeds the primary intervention threshold but is lower than the secondary intervention threshold, this module performs source blocking intervention. The specific operation is as follows: The module obtains the specific location information of the risk source from the collaborative reasoning module, queries the preset device location mapping table, determines the intelligent control device directly associated with the risk source (such as an intelligent circuit breaker or solenoid valve), and then sends an exact cut-off command, such as a write register command conforming to the Modbus protocol, to the address of this device through the intervention control interface to cut off the energy supply of the risk source.

[0090] When the value of the regional threat level exceeds the secondary intervention threshold, the collaborative intervention module performs regional emergency response intervention. This intervention action is a system-level linkage response. The module sends a predefined fire alarm signal or linkage protocol message to the building automation system or the fire alarm controller through the intervention control interface. This signal will trigger a more extensive emergency plan, such as starting the fire exhaust system, audible and visual alarm in the area, or unlocking the access control of the fire escape.

[0091] Furthermore, the collaborative intervention module also works in conjunction with the risk propagation prediction module. After the risk propagation prediction module outputs the future risk propagation path, the collaborative intervention module analyzes the path data and identifies all spatial units that are on the predicted path but currently have a risk value of zero. Subsequently, the module sends predictive alarm commands to the alarm devices (such as smart speakers or dedicated alarms) deployed in these spatial units. The commands include warning information and the direction of the risk source, used to guide personnel evacuation in advance.

[0092] The model and rule base update module is responsible for maintaining and iterating the software algorithms and decision logic of the entire system. In one implementation, this module establishes an encrypted communication connection based on Transport Layer Security (TLS) with an externally deployed cloud platform through the wired Ethernet interface or cellular network communication module of the gateway device. The module periodically queries the cloud platform for update instructions, or the cloud platform proactively pushes update instructions.

[0093] Upon receiving an update instruction, the module downloads a digitally signed and encrypted update package from the cloud platform. The update package contains one or more new AI model files, an updated collaborative inference rule base file, and a manifest file containing version information and file hash checksums (e.g., SHA-256). After downloading, the module first verifies the digital signature of the update package using a pre-configured public key to confirm its authenticity and integrity. After successful verification, it calculates the hash value of each file within the package and compares it with the records in the manifest file to ensure that the files have not been tampered with during transmission.

[0094] After completing all verification steps, the module performs an update operation. For the collaborative inference rule base file, the module writes it to the local storage of the gateway device, replacing the old version file, and sends an internal reload configuration signal to the collaborative inference module. For the new AI model file, the module uses the Firmware Over-the-Air (FOTA) protocol via the wireless communication interface to distribute it to one or more smart sensing nodes specified in the instruction. Upon receiving the new model file, the smart sensing node also verifies it and then writes it to the node's non-volatile flash memory, loading the new AI model upon the next startup. After the update operation is complete, the module reports the execution result of this update (success or failure, and the specific status of each node) to the cloud platform, forming a complete closed-loop update process.

[0095] See attached document Figure 6 To clarify the internal working mechanism of the system of the present invention and the collaborative process between its modules, the following will be explained step by step through a specific application scenario. The scenario is set as follows: In a power distribution cabinet, a power cable undergoes initial pyrolysis due to insulation aging, which eventually develops into an arc fault.

[0096] Phase 1: Initial System State and Emerging Risks

[0097] In its initial state, the collaborative sampling control program inside the intelligent sensing node (hereinafter referred to as Node A) deployed near the power distribution cabinet is in a low-power inspection state. In this state, its multimodal sensing unit periodically monitors the environment at a first preset sampling frequency (e.g., once every 60 seconds). When the cable insulation begins to undergo slow pyrolysis, Node A's chemical fingerprint sensing unit detects a slight increase in the concentration of trace amounts of carbon monoxide and total hydrocarbons in the environment in its next sampling. This chemical fingerprint data is sent to Node A's edge processor.

[0098] The pre-built AI model on the edge processor analyzes the input data. In its output risk type probability vector, the probability value of electropyrolysis increases slightly and exceeds the preset first threshold. This change immediately triggers the collaborative sampling control procedure, causing node A to switch from a low-power inspection state to a high-frequency verification state, and the sampling frequency of the multimodal sensing unit is increased to a second preset sampling frequency (e.g., 100 times per second).

[0099] Phase Two: Local Event Confirmation and Broadcast

[0100] After entering the high-frequency verification state, node A begins high-density data collection on the risks. As cable pyrolysis continues, the electropyrolysis probability values ​​continuously output by the AI ​​model form a steadily increasing sequence within the temporary vector sequence buffer of the event generation and decision module at node A. The event generation and decision module, through its time series analysis rules, determines that this sequence meets a preset risk growth pattern (i.e., the sequence slope exceeds a growth slope threshold).

[0101] After confirming the persistence of the risk, the edge processor generates the first local event vector. This vector's data structure includes: a unique identifier and physical location coordinates of node A, the current timestamp, a risk type probability vector with the highest probability of electropyrolysis, and a moderate local confidence level calculated based on the stability and magnitude of the probability growth sequence. Subsequently, this local event vector is broadcast to the wireless network via the communication module.

[0102] Phase Three: Central Collaborative Reasoning and Escalation of the Situation

[0103] The gateway device receives a local event vector from node A. Its collaborative inference module initiates analysis. Since no other local event vectors from neighboring nodes of node A were received within the set time window, the calculated regional corroboration factor... The value is zero. At this point, the gateway device calculates the area threat level. The system determines the event as a low-confidence event from a single source based solely on the local confidence level of node A, which is below the first-level intervention threshold. Monitoring will continue, but no intervention will be implemented for the time being.

[0104] Subsequently, the cable insulation broke down, generating a continuous, weak electric arc. The electromagnetic acoustic sensing unit at node A detected previously absent high-frequency electromagnetic signals and ultrasonic acoustic signals. Upon receiving the new multimodal input, the AI ​​model's output risk type probability vector showed an instantaneous jump in the probability value of the arc fault type to an extremely high value. The event generation decision module detected this change as satisfying a mutation pattern (i.e., the probability difference exceeded the mutation threshold). Immediately generate and broadcast an updated local event vector, where the local confidence is set to a high value.

[0105] At almost the same time, another intelligent sensing node (hereinafter referred to as node B) deployed in the same room as the power distribution cabinet also detected characteristic gases such as ozone generated by the electric arc through its chemical fingerprint sensing unit, and similarly generated and broadcast a local event vector of type electrical fault byproduct.

[0106] The gateway device's collaborative inference module receives two local event vectors from node A and node B, respectively. When processing the update vector from node A, the module uses the vector from node B as supporting information. Based on a preset association weight matrix, (Arc faults, byproducts of electrical faults) have high weight values. Therefore, the calculated regional corroborating factors... This is a significantly positive value. Finally, combining node A's high local confidence and high regional corroboration factor, the regional threat level is calculated. The value significantly exceeded the first-level intervention threshold.

[0107] Phase Four: Tiered Intervention, Proactive Prediction, and System Iteration

[0108] Once the regional threat level exceeds the Level 1 intervention threshold, the collaborative intervention module is triggered. The module first performs source blocking intervention. By querying the device location mapping table, it determines the address of the smart circuit breaker controlling the power supply of the distribution cabinet and sends a trip command to it through the intervention control interface, cutting off the power supply to the fault source.

[0109] Simultaneously, the risk propagation prediction module was activated, using node A as the risk initiation point. Based on a pre-set building space environment diagram model, the state transition model began iterative calculations. The calculation results showed that the high-temperature fumes generated by the electric arc had a very high probability of spreading along the cable tray, through the wall penetrations, and into the adjacent storage room within the next three minutes.

[0110] Upon receiving the propagation path prediction result, the collaborative intervention module immediately executes a predictive alarm intervention. The module sends a warning command to an audible and visual alarm deployed in the storage room, causing it to emit a warning signal that differs from a conventional fire alarm, indicating the direction of the potential hazard's source.

[0111] After the event concluded, all data from the entire event, including the raw sensor data from nodes A and B, the generated local event vectors, and the decision logs from the gateway device, was packaged and uploaded to the cloud platform. This validated real-world data was used as new training samples to optimize the AI ​​model. Several days later, a new, higher-performing AI model was securely pushed and updated to all intelligent sensing nodes in the network via the model and rule base update module, completing a self-evolutionary closed loop for the system.

Claims

1. A wireless intelligent smoke detector and electrical safety collaborative monitoring network system, characterized in that, include: Multiple intelligent sensing nodes, each of which is configured with: The multimodal sensing unit is used to simultaneously collect chemical fingerprint data, electromagnetic and acoustic data, and conventional environmental data within the monitoring area; An edge processor is used to analyze the data collected by the multimodal perception unit through a pre-set AI model, and when potential risks are identified, generate a local event vector containing node identifier, location, timestamp, risk type probability vector, local confidence and local strength. The communication module is used to broadcast the local event vector to external systems. Gateway device, the gateway device being configured with: The collaborative reasoning module is used to receive local event vectors broadcast by one or more of the intelligent sensing nodes, calculate regional corroboration factors based on preset collaborative reasoning rules, and then combine the local confidence in the local event vectors to assess the regional threat level. The collaborative intervention module is used to trigger a preset intervention action when the threat level in the area exceeds a preset intervention threshold.

2. The wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 1, characterized in that, The multimodal sensing unit specifically includes: The chemical fingerprint sensing unit is used to identify characteristic gas molecule combinations produced by the early pyrolysis of an object using broadband absorption spectroscopy technology, and to acquire the chemical fingerprint data. An electromagnetic acoustic sensing unit is used to synchronously monitor specific high-frequency electromagnetic field signals and acoustic signals generated by weak arc discharge or insulation breakdown, and to acquire the electromagnetic and acoustic data.

3. The wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 1, characterized in that, The preset collaborative reasoning rules specifically define the calculation steps for the regional corroboration factors, which include: For any local event vector, aggregate other local event vectors broadcast by neighboring nodes of the local event vector within a preset time window; Taking into account the correlation weights between the risk types of other local event vectors and the current event type, the local confidence and local strength of other local event vectors, as well as the spatial distance and time difference between events, a weighted summation operation is performed to obtain the regional corroborating factor.

4. The wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 1, characterized in that, The gateway device is also configured with: The risk propagation prediction module is used to calculate the future distribution and propagation path of the risk in different spatial units by taking the risk source location as the starting point and using a state transition model based on a pre-set building space environment map model after the collaborative reasoning module has confirmed the regional threat level.

5. The wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 4, characterized in that, The specific calculation content of the state transition model for calculating the risk value of any spatial unit at the next moment is as follows: The risk value of the spatial unit at the current moment, the risk natural dissipation coefficient of the spatial unit, and the risk values ​​and corresponding propagation and transfer probabilities of all spatial units adjacent to the spatial unit at the current moment.

6. The wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 1, characterized in that, The intervention action triggered by the collaborative intervention module is as follows: Based on the propagation path output by the risk propagation prediction module, a predictive alarm command is sent to the alarm device in the spatial unit that is on the path and has not yet experienced a risk.

7. The wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 1, characterized in that, The intervention actions triggered by the collaborative intervention module include performing tiered interventions, which include: When the threat level of the area exceeds the first-level threshold, source blocking intervention is implemented, and instructions are sent to the intelligent control device associated with the risk source to cut off the risk source; When the threat level of the area exceeds the level 2 threshold, an area emergency response intervention will be implemented, and fire protection or security equipment within the area will be activated.

8. The wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 1, characterized in that, The edge processor is further configured with an event generation decision module; The event generation decision module is used to receive multiple risk type probability vectors with time-series information continuously output by the AI ​​model and store them in a temporary vector sequence buffer. The event generation decision module determines whether the vector sequence in the temporary vector sequence buffer meets the risk growth pattern or mutation pattern by using preset time series analysis rules. The edge processor generates and triggers the communication module to broadcast the local event vector only when the vector sequence satisfies the risk growth pattern or mutation pattern.

9. The wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 1, characterized in that, The edge processor is also configured with a collaborative sampling control program, which has at least two preset working states that are linked to the analysis results of the AI ​​model, including: In a low-power inspection state, when the probability of risk type output by the AI ​​model is lower than a first preset threshold, the collaborative sampling control program sets the chemical fingerprint sensing unit and the electromagnetic acoustic sensing unit in the multimodal sensing unit to a first preset sampling frequency. In a high-frequency verification state, when the probability of any risk type exceeds the first preset threshold, the collaborative sampling control program switches the chemical fingerprint sensing unit and the electromagnetic acoustic sensing unit to a second preset sampling frequency higher than the first preset sampling frequency, for high-density data collection and verification of potential risks.

10. A wireless intelligent smoke detector and electrical safety collaborative monitoring network system according to claim 1, characterized in that, The gateway device is also equipped with a model and rule base update module; The model and rule base update module is used to receive update instructions from an external cloud platform through the communication interface, and according to the update instructions, to perform online updates and parameter tuning of the collaborative reasoning rules used to calculate regional corroborating factors in the collaborative reasoning module and the AI ​​models in the multiple intelligent sensing nodes.

Images