A gateway device-based fire-fighting facility alarm management method and system

Abstract

The application relates to the technical field of intelligent fire-fighting Internet of Things, and discloses a fire-fighting facility alarm management method and system based on a gateway device, which has the technical scheme as follows: joint punishment is performed in combination with system prior parameters and space-time multi-dimensional features, dynamic weight correction of edge perception data is realized, multi-modal situation weighting is generated in combination with visual recognition confidence probability to generate comprehensive fire disaster severity data, and then safety communication delay budget and control duty cycle are reversely and adaptively distributed, so that control of the fast safety fire-fighting linkage is realized from the bottom hardware level denoising decision.

Description

Technical Field

[0001] This invention relates to the field of smart fire protection Internet of Things technology, and more specifically, to a method and system for fire protection facility alarm management based on gateway devices. Background Technology

[0002] In existing technologies, the alarm and linkage of fire protection facilities typically rely on a single sensing device to set a fixed threshold for triggering. This problem is particularly prominent in complex fire protection scenarios such as industrial buildings and underground spaces, where such devices are highly susceptible to noise interference from airflow, dust, or local hardware failures, resulting in high false alarm and false alarm rates. Furthermore, existing IoT gateways lack the ability to dynamically reduce the weight of sensing data at the underlying level based on spatial discrete differences and temporal entropy characteristics when processing multi-source heterogeneous data, leading to low decision reliability. More critically, when issuing fire prevention and firefighting linkage control commands to end-point execution devices, existing systems typically introduce fixed secure encrypted handshake mechanisms to prevent malicious network tampering. This fails to effectively resolve the physical conflict between the time-consuming encrypted handshake and the rapid linkage response urgently needed in fire rescue, easily causing network communication congestion or execution delays for critical control commands in critical fire situations.

[0003] Therefore, the present invention provides a method and system for fire protection facility alarm management based on gateway devices, which improves the above-mentioned technical problems. Summary of the Invention

[0004] This invention aims to address the shortcomings of existing technologies by providing a fire alarm management method and system based on gateway devices. The invention combines prior system parameters with spatiotemporal multidimensional features to perform joint penalties, thereby achieving dynamic weighting correction of edge-aware data. It also combines visual recognition confidence probability to perform multimodal situational weighting to generate comprehensive fire severity data, and then adaptively allocates safety communication delay budget and control duty cycle in reverse, realizing a control closed loop from low-level hardware-level noise reduction decision to rapid safety prevention and firefighting linkage.

[0005] To achieve the above objectives, the present disclosure proposes the following technical solutions:

[0006] In a first aspect, this disclosure proposes a fire alarm management method based on a gateway device, applied to an edge computing gateway device, the method comprising:

[0007] The original environmental electrical signals of heterogeneous sensing devices in the target area are obtained through the communication controller of the gateway device and converted into basic probability allocation data for discrete physical environment states after preprocessing.

[0008] Based on the spatial discrete difference characteristics of data between sensing devices, the temporal information entropy characteristics, and the inherent prior trust weights of sensing devices, the basic probability allocation data is subjected to joint penalty calculation and dynamic weight reduction correction through an arithmetic logic unit, and the corrected basic probability allocation data is output.

[0009] The modified basic probability allocation data is processed by a digital signal processor to generate a global fire confidence score. The visual recognition confidence probability extracted by the edge AI accelerator is combined with multimodal situational weighting to generate comprehensive fire severity data.

[0010] Based on the comprehensive fire severity data, inverse proportional mapping control is performed to adaptively allocate a secure communication delay budget to the hardware timer within the secure encryption coprocessor.

[0011] Based on the comprehensive fire severity data and its differential slope, as well as the reciprocal of the safety communication delay budget, the actuator drive unit performs dimensional alignment and weighted summation operations to generate recommended operating intensity data for the corresponding target execution device. This data is then converted into control duty cycle and encrypted by hardware before being sent out.

[0012] As a preferred embodiment of the present invention, before acquiring the original environmental electrical signals of heterogeneous sensing devices within the target area, the method further includes:

[0013] The general computational parameters, including the physical sampling period, calibration parameter set, and dimension transformation and alignment physical matching coefficients, are loaded from non-volatile memory into the resident address area of ​​static random access memory.

[0014] The security encryption coprocessor extracts the message authentication code key and encryption algorithm configuration parameters from its built-in secure storage area to complete the initialization operation. The message authentication code key does not enter the system's shared memory addressing area.

[0015] As a preferred embodiment of the present invention, before performing joint penalty calculation and dynamic weight reduction correction on the basic probability allocation data, the following feature extraction steps are also included:

[0016] The discrete difference value of environmental data between any two sensing devices is quantified based on the Hellinger distance metric formula. The average deviation between the target sensing device and other devices is calculated by a hardware accumulator to generate a global conflict degree feature as the spatial discrete difference feature.

[0017] The temporal information entropy features of the basic probability allocation data are calculated based on the Shannon entropy mathematical model. When the single-state probability input value is detected to be zero, the zero-value bypass circuit built into the information entropy conversion logic gate is triggered to directly output the summation result as zero. When the calculated feature value is greater than the set maximum environmental noise information entropy tolerance constant, the hardware limiting circuit built into the information entropy conversion logic gate is triggered to perform a truncation operation.

[0018] As a preferred embodiment of the present invention, the step of performing joint penalty calculation and dynamic weight reduction correction on the basic probability allocation data, and outputting corrected basic probability allocation data, includes:

[0019] Prior trust weights are constructed based on the log-likelihood ratio of detection rate to false alarm rate and the log-likelihood ratio of false negative rate to correct negative rate of sensing devices.

[0020] For the target perception device, the basic probability allocation data, the linear penalty function term composed of the temporal information entropy features, the exponential penalty function term composed of the global conflict degree features, and the prior trust weight are multiplied together to calculate the denominator value of the sum of the penalty terms.

[0021] The zero-value detection circuit is used to verify the denominator of the sum of the penalty terms. If the denominator of the sum of the penalty terms is greater than or equal to the minimum weight determination threshold, normalization processing is performed to output the corrected basic probability allocation data. If the denominator of the sum of the penalty terms is less than the minimum weight determination threshold, the corrected basic probability allocation data of the target sensing device is reset to an unknown state.

[0022] As a preferred embodiment of the present invention, the step of performing evidence fusion processing on the corrected basic probability allocation data to generate a global fire confidence score includes:

[0023] Obtain the total number of sensing devices online simultaneously. In response to the total number of sensing devices being 1, directly output the corrected basic probability allocation data of that device as the fusion result.

[0024] When the total number of the sensing devices is greater than or equal to 2, fusion is performed using a pairwise iterative synthesis method.

[0025] Before each pairwise synthesis operation, the evidence synthesis conflict coefficient is calculated: if the evidence synthesis conflict coefficient is less than the conflict threshold, standard Dempster orthogonal synthesis is performed; if the evidence synthesis conflict coefficient is greater than or equal to the conflict threshold, downgraded fusion processing is triggered, and a weighted average is performed according to the prior trust weights of the participating synthesis devices to replace orthogonal synthesis, and the equivalent prior weights of the current fusion result are updated synchronously for use in subsequent iterative synthesis.

[0026] The numerical components pointing to the occurrence of open flame are extracted from the final fusion result after iteration and used as the global fire confidence level.

[0027] As a preferred embodiment of the present invention, the step of performing inverse proportional mapping control based on the comprehensive fire severity data to adaptively allocate a secure communication delay budget for the hardware timer within the secure encryption coprocessor includes:

[0028] Extract the pre-loaded baseline encryption tolerance latency parameters and network bandwidth adjustment constant;

[0029] Perform an inverse proportional mapping calculation, taking the benchmark encryption tolerance delay parameter as the numerator component and the sum of the products of constant 1, network bandwidth adjustment constant, and comprehensive fire severity data as the denominator component, and obtain the division operation result;

[0030] The result of the division operation is used as the budget for the secure communication delay and written into the hardware timer as the initial countdown value.

[0031] As a preferred embodiment of the present invention, the step of performing dimensional alignment and weighted summation operations based on the comprehensive fire severity data and its differential slope, and the reciprocal of the secure communication delay budget, to generate recommended operational intensity data for the corresponding target execution device includes:

[0032] Extract the pre-loaded first dimension transformation and alignment physical matching coefficients, second dimension transformation and alignment physical matching coefficients, and third dimension transformation and alignment physical matching coefficients;

[0033] The difference between the comprehensive fire severity data of the current period and the comprehensive fire severity data of the previous period is divided by the physical sampling period to obtain the differential slope.

[0034] The first operational component is calculated by multiplying the first dimension transformation and alignment physical matching coefficient with the comprehensive fire severity data of the current period; the second operational component is calculated by multiplying the second dimension transformation and alignment physical matching coefficient with the differential slope; and the third operational component is calculated by multiplying the third dimension transformation and alignment physical matching coefficient with the reciprocal of the secure communication delay budget.

[0035] A weighted summation operation is performed on the first operational component, the second operational component, and the third operational component to output dimensionless operational intensity recommendation data.

[0036] Secondly, embodiments of this disclosure propose a fire alarm management system based on a gateway device, wherein the system is configured on an edge computing gateway device and includes:

[0037] The initial data acquisition and probability allocation module, which runs on the communication controller and central processing unit, is used to acquire the original environmental electrical signals of heterogeneous sensing devices in the target area, and convert them into basic probability allocation data for discrete physical environment states after preprocessing.

[0038] The dynamic weight reduction correction module, which runs in the arithmetic logic unit, is used to perform joint penalty calculation and dynamic weight reduction correction on the basic probability allocation data based on the spatial discrete difference characteristics, temporal information entropy characteristics and the inherent prior trust weights of the data between sensing devices, and output the corrected basic probability allocation data.

[0039] The evidence fusion and multimodal assessment module, running on a digital signal processor and an edge AI accelerator, is used to perform evidence fusion processing on the corrected basic probability allocation data to generate a global fire confidence score, and to perform multimodal situation weighting by combining visual recognition confidence probability to generate comprehensive fire severity data.

[0040] The secure communication delay allocation module, which runs on the secure encryption coprocessor, is used to perform inverse proportional mapping control based on the comprehensive fire severity data and adaptively allocate the secure communication delay budget for the hardware timer.

[0041] The linkage recommendation and execution module, running in the actuator drive unit, is used to perform dimensional alignment and weighted summation operations based on the comprehensive fire severity data and its differential slope, as well as the reciprocal of the safety communication delay budget, to generate recommended operation intensity data for the corresponding target execution device, convert it into control duty cycle, and send it out after hardware encryption.

[0042] As a preferred embodiment of the present invention, the dynamic weight reduction correction module is specifically used for:

[0043] The global conflict degree feature is obtained by calculating the Hellinger distance metric formula and hardware accumulator, and is used as the spatial discrete difference feature.

[0044] The time-domain information entropy feature is calculated based on the Shannon entropy mathematical model. When the single-state probability input value is detected to be zero, the zero-value bypass circuit built into the information entropy conversion logic gate is triggered, and when the feature value exceeds the limit, the hardware limiting circuit is triggered to perform a truncation operation.

[0045] The prior trust weight is constructed based on the log-likelihood ratio of detection rate and false positive rate, and the log-likelihood ratio of false negative rate and correct rejection rate. This prior trust weight is then multiplied by the basic probability allocation data, the linear penalty function term composed of the temporal information entropy feature, and the exponential penalty function term composed of the global conflict degree feature. After verification by a hardware zero-value detection circuit, the corrected basic probability allocation data is output.

[0046] As a preferred embodiment of the present invention, the linkage recommendation and execution module is specifically used for:

[0047] Extract the three sets of dimensional transformation and alignment physical matching coefficients;

[0048] Calculate the differential slope of the comprehensive fire severity data on the first-order time axis;

[0049] The current fire severity, differential slope, and reciprocal delay are multiplied by the corresponding dimension transformation and alignment physical matching coefficients to cancel out the time reciprocal dimension of the differential slope and reciprocal delay terms. After summing, the dimensionless recommended operation intensity data is output.

[0050] In summary, the present invention has the following beneficial effects:

[0051] Firstly, this invention quantifies the spatial discreteness and temporal disorder jump characteristics of data between sensing devices by using global conflict degree features and temporal information entropy features respectively. It also performs joint penalty calculation of spatiotemporal multidimensional constraints on basic probability allocation data by combining the prior trust weights of sensing devices, thereby effectively suppressing extreme noise caused by environmental airflow disturbances or local device failures at the hardware level. Furthermore, it configures a hardware zero-value verification mechanism to isolate the effective evidence contribution of failed devices, reducing the risk of false alarms and false negatives caused by the drift of a single sensing device and improving the reliability of the underlying fused data.

[0052] Secondly, this invention generates comprehensive fire severity data by combining the visual recognition confidence probability output by the edge AI accelerator with the global fire confidence score to perform multimodal situation weighting. This effectively avoids false fire signals caused by complex physical interference on site. It also makes up for the physical limitations of a single IoT sensing mode, can more accurately characterize the fire spread, and improves the reliability of fire situation determination.

[0053] Third, this invention achieves a dynamic balance between the security encryption handshake time and the real-time fire response by using an inverse proportional mapping control based on comprehensive fire severity data to adaptively allocate a security communication delay budget as the initial countdown value for the hardware timer within the security encryption coprocessor. It can forcibly constrain the encryption packet and transmission window time of control commands at the hardware level during fire emergencies, effectively resolving the physical conflict between the encryption handshake time and rapid emergency response, and avoiding channel blockage of critical fire prevention and control linkage commands in emergency situations.

[0054] Fourth, this invention extracts the differential slope of the comprehensive fire severity data and the reciprocal of the safe communication delay budget, calls the dimension conversion physical matching coefficient to perform dimension alignment and weighted summation, and realizes the seamless connection of fire situation and network status to the control model of the end physical device; it outputs dimensionless operation intensity recommendation data and strictly maps it into the duty cycle control quantity of the target execution device, realizing hardware-level adaptive closed-loop linkage control from the bottom signal acquisition, multi-modal cleaning decision to the precise drive of the actuator. Attached Figure Description

[0055] Figure 1 A flowchart illustrating a fire protection facility alarm management method based on a gateway device, provided as an embodiment of the present invention;

[0056] Figure 2 This is a framework diagram of a fire alarm management system based on a gateway device, provided as an embodiment of the present invention. Detailed Implementation

[0057] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0060] Furthermore, the technical features involved in the various embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0061] The subject of this application is an edge computing gateway device. This gateway device physically includes a communication controller, a central processing unit (CPU), a digital signal processor (DSP), an edge AI accelerator (NPU), a security encryption coprocessor, non-volatile memory (EEPROM), a direct memory access (DMA) controller, and an actuator driver unit. The CPU integrates data cleaning control logic, and its internal arithmetic logic unit (ALU) is configured with spatial difference calculation hardware logic, information entropy conversion logic gates, logarithmic operation circuits, and exponential operation circuits. The DSP has a built-in evidence orthogonal synthesis hardware operation unit. The edge AI accelerator has built-in multiply-accumulate operation cores (MACs) and a critical situation control register group, which includes a current value register and a historical value register. The actuator driver unit has a built-in floating-point operation circuit. The security encryption coprocessor is configured with an independent secure storage area and a hardware timer.

[0062] like Figure 1 As shown, this application provides a fire protection facility alarm management method based on a gateway device. This method is executed collaboratively by the underlying hardware of the gateway device, and the specific implementation steps are as follows:

[0063] Step S0: System initialization and loading of physical prior parameters

[0064] After the gateway device is powered on and started, it loads the configuration according to the classification of execution parameters.

[0065] For general computing parameters, the central processing unit reads the system prior parameters and system calibration coefficients pre-programmed in non-volatile memory through the internal bus, and loads them into the resident address area of ​​static random access memory (SRAM) to establish the benchmark data required for end-to-end computing.

[0066] Specifically, this includes: the total number N of sensing devices simultaneously online within the target area; and the physical sampling period. ; Signal range calibration parameter set for each sensing device i; Detection rate inherent to each sensing device i at the factory. False alarm rate With underreporting rate And it is limited by the fact that the physical properties of the device always satisfy and The window length W of the sliding window mid-range filter and the threshold for determining amplitude spikes. Discrete probability mapping lookup table parameter set; Maximum environmental noise information entropy tolerance constant measured and calibrated in a fire-free, clean environment. Minimum weight threshold ; Conflict threshold for evidence synthesis Fire confidence wake-up threshold Flame visual recognition model weight parameters and inference configuration parameters; sensor modality matching coefficient α and visual modality matching coefficient β, satisfying α+β=1; baseline encryption tolerance delay parameters. With network bandwidth adjustment constant μ; dimensional transformation and alignment physical matching coefficient ,in The coefficient is dimensionless. and The dimension is time unit, used to offset the time reciprocal dimension of the rate of change and the reciprocal term, ensuring that the dimension is unified as dimensionless after the three operations; and the actuator duty cycle linear mapping coefficient and control output range parameter.

[0067] For secure communication parameters, the message authentication code key and encryption algorithm configuration parameters are pre-embedded in the secure storage area built into the secure encryption coprocessor and are initialized automatically by the secure encryption coprocessor after power-on.

[0068] Step S1: Obtain the initial behavioral data of the multi-source heterogeneous sensing devices and convert it into basic probability allocation data.

[0069] S1.1 Heterogeneous signal acquisition and preprocessing:

[0070] To obtain the data foundation characterizing the actual physical conditions on site, the gateway device's communication controller acquires raw environmental electrical signals from heterogeneous sensing devices (including infrared thermal imagers, combustible gas detectors, and photoelectric smoke detectors) within the target area in real time via the underlying RS485 or CAN bus, using a hardware timer-triggered interrupt. The central processing unit loads these raw environmental electrical signals into a volatile cache and calls the window length W and amplitude spike determination threshold loaded in SRAM. Perform sliding window mid-range filtering to remove abnormal spike pulses, and call the signal range calibration parameter set to perform dimensionless normalization processing to obtain standardized feature data.

[0071] S1.2 Discrete Environment State Confidence Mapping:

[0072] Based on the positive correlation between the calibrated feature amplitudes and physical state probabilities, the arithmetic logic unit retrieves the discrete probability mapping lookup table in SRAM, converts the standardized feature data into initial confidence levels for three types of discrete physical environment states, and generates basic probability assignment data. .

[0073] Where i represents the hardware number of the sensing device; t represents the current clock sampling time; k represents a discrete set of physical environment states, including three categories: open flame, no fire, and unknown state, respectively identified as [identification details missing]. ; The sum of all state components is equal to 1.

[0074] The above The vector is written to the first-level shared memory pool of the gateway device. The central processing unit issues a memory write completion flag through the direct memory access controller and simultaneously triggers the spatial difference calculation hardware logic and information entropy conversion logic gate in the arithmetic logic unit through the internal data bus of the system, executing steps S2 and S3 in parallel.

[0075] Step S2: Calculate the global conflict degree characteristics of each sensing device in the physical space dimension.

[0076] S2.1 Measurement of spatial differences between sensing devices:

[0077] The spatial difference calculation hardware logic extracts all basic probability allocation data within the same clock synchronization cycle from the shared memory pool. First, it reads the total number of sensing devices from the SRAM. ; in response to Directly output global conflict degree features ; in response to The Hellinger distance metric model is invoked to quantify the spatial discrete difference values ​​of environmental data between any sensing device i and j.

[0078] S2.2 Global Conflict Level Aggregation:

[0079] For target sensing device i, a hardware accumulator is used to traverse and calculate the spatial discrete difference values ​​between it and all other devices, and the average deviation is calculated to generate a global conflict degree feature. The larger this value, the more severe the deviation between the sensing device i and the spatial data of the global network.

[0080] Calculated The data is temporarily stored in the feature register of the central processing unit, and the spatial difference calculation hardware logic sends the first ready signal to the data cleaning control logic.

[0081] Step S3: Calculate the temporal information entropy characteristics of each sensing device in the time dimension.

[0082] S3.1 Time-domain information discreteness transformation:

[0083] The information entropy transformation logic gate reads the basic probability allocation data of target sensing device i from the shared memory pool, calls the Shannon entropy model, and calculates the temporal information entropy feature. The higher this eigenvalue, the more severe the local physical environment disturbance of the device. The information entropy conversion logic gate has a built-in zero-value bypass circuit, which detects a single-state probability input value. When the summation result is 0, the logarithmic operation will be output directly to avoid numerical overflow exceptions.

[0084] S3.2, Extreme value limiting cutoff:

[0085] In response to sudden interference The hardware limiting circuit within the logic gate performs a cutoff operation, taking... .

[0086] Calculated The status register of the central processing unit is written and a second ready signal is issued; when the central processing unit receives both the first ready signal and the second ready signal, it triggers the execution of step S4.

[0087] Step S4: Dynamically reduce the weight of the basic probability allocation data by combining the system prior parameters.

[0088] S4.1 Hardware Prior Trust Weight Extraction:

[0089] Prior trust weights are constructed based on the likelihood ratio of detection performance, and the data cleaning control logic reads data loaded from SRAM. , and Using the logarithmic operation circuit built into the arithmetic logic unit, the log-likelihood ratio of the detection rate to the false alarm rate is calculated, and the log-likelihood ratio of the false negative rate to the correct rejection rate is subtracted. The inherent prior trust weight of the sensing device i is then output. .

[0090] S4.2 Spatiotemporal Multidimensional Constraint Correction:

[0091] The joint penalty employs a product-based operational architecture, configuring an exponential penalty function in the spatial dimension and a linear penalty function in the temporal dimension to jointly suppress multidimensional noise. The overall data cleaning control logic utilizes a built-in exponential operation circuit and a hardware multiplier-adder to perform the joint penalty calculation, deriving the denominator value of the sum of the penalty terms. :

[0092] ;

[0093] The arithmetic logic unit's built-in zero-value detection circuit performs hard verification on the summation result of the denominator:

[0094] In response to Perform normalization processing to output corrected basic probability allocation data :

[0095] ;

[0096] In response to If it is determined that sensing device i is completely overwhelmed by noise in the current cycle, its correction value is directly reset to [value]. This avoids the division-to-zero exception at the hardware level.

[0097] generated The data are loaded one by one into the pipeline buffer of the digital signal processor; when the correction data of all sensing devices in the current cycle has been loaded, the buffer full flag is set, triggering the digital signal processor to execute step S5.

[0098] Step S5: Fuse the corrected basic probability allocation data to generate a global fire confidence score.

[0099] S5.1 Boundary Determination and Orthogonal Iterative Synthesis:

[0100] The digital signal processor reads the total number of sensing devices N; in response to The corrected basic probability allocation data of the device is directly output as the fusion result.

[0101] In response to Using the built-in evidence synthesis hardware computing unit, the correction data of the devices in the pipeline cache are synthesized sequentially by adopting a pairwise iterative synthesis method; before each pairwise synthesis operation, the evidence synthesis conflict coefficient K is first calculated.

[0102] In response to Invoke Dempster's orthogonal composition rules to perform evidence fusion; in response to Hardware-triggered degradation fusion logic is applied based on the prior trust weight of the current input device. We perform weighted averaging and normalization across all states to replace orthogonal synthesis and avoid iteration anomalies. After each synthesis, we synchronously update the equivalent prior weights of the current fusion result for use in subsequent iterations.

[0103] S5.2 Global Confidence Extraction:

[0104] The digital signal processor extracts the pointer from the joint probability result of the final iteration. The numerical components of the state are used as the global fire confidence level. .

[0105] Digital signal processors will Pushed to the global state bus, the hardware comparator within the edge AI accelerator compares the fire confidence wake-up threshold in SRAM in real time. In response to Then, the start level is issued to activate step S6.

[0106] Step S6: Combine edge visual features and the global fire confidence score to generate comprehensive fire severity data.

[0107] S6.1 Video Stream Takeover and Deep Inference:

[0108] Once activated, the edge AI accelerator takes over the real-time video stream from the edge camera via the MIPI-CSI physical interface and loads the model weight parameters from SRAM. The multiply-accumulate core performs frame-by-frame inference for a lightweight convolutional neural network, outputting the visual recognition confidence probability. .

[0109] S6.2, Multimodal Situation Weighting:

[0110] Edge AI accelerator extracts matching coefficients and Perform a multimodal weighted model to calculate the comprehensive fire severity data. :

[0111] ;

[0112] Generated comprehensive fire severity data The range of values ​​is strictly constrained to the interval [0,1].

[0113] Each time the current cycle is latched First, the edge AI accelerator moves the current value in the critical situation control register group into the historical value register; then, the newly generated... The value is locked into the current value register; the update of the current value register synchronously generates hardware control pulses, which are distributed in two channels to the security encryption coprocessor and the actuator driver unit, and concurrently trigger steps S7 and S8.

[0114] Step S7: Based on the comprehensive fire severity data, adaptively allocate a secure communication delay budget for the secure encryption chip.

[0115] S7.1, Inverse proportional delay mapping control:

[0116] An inverse proportional function is used to achieve a negative correlation between severity and encryption latency. The more severe the fire, the lower the allocated security encryption handshake latency budget, thereby increasing the network transmission priority of critical control commands. After receiving the trigger pulse, the security encryption coprocessor calls the built-in floating-point circuit to read the baseline encryption tolerance latency parameter from SRAM. Calculate using the network bandwidth adjustment constant μ:

[0117] ;

[0118] Calculated dynamic secure communication delay budget The value is directly written into the hardware timer of the secure encryption coprocessor as the initial countdown value; the timer immediately starts counting down, forcibly constraining the instruction encryption packet and sending window time of the downstream step S8 from the hardware level.

[0119] Step S8: Based on the comprehensive fire severity and security communication delay budget, generate recommended data on the operational intensity for the target execution equipment.

[0120] S8.1, Physically driven dimensional alignment and synthesis:

[0121] The actuator drive unit reads the current severity data from the critical situation control register group via the system's internal bus. Historical severity data Budget for secure communication delays During the first sampling period after system power-on, since there is no valid historical data, the differential slope term on the first-order time axis is set to 0 by default; from the second sampling period onwards, the floating-point arithmetic circuit built into the driver unit extracts the physical sampling period from the SRAM. Physical matching coefficient with dimension conversion Perform dimension alignment and weighted summation, and output dimensionless operation strength recommendation data. :

[0122] ;

[0123] Subsequently, the actuator drive unit calls the actuator duty cycle linear mapping coefficients and interval parameters loaded in SRAM to... The value of is strictly mapped to the actuator duty cycle control quantity of 0~100%.

[0124] Before the countdown set by the hardware timer in step S7 ends, the security encryption coprocessor extracts the message authentication code key from the secure storage area, performs hardware encryption and message authentication code encapsulation on the actuator duty cycle control quantity, and sends it to the target execution device (such as a smoke exhaust fan or fire extinguishing agent release valve) through the physical communication interface. Finally, the direct memory access controller clears the volatile cache and pipeline cache, resets the system-level interrupt suspension flag, and the system enters the next sampling cycle in step S1, completing the closed loop of edge fire prevention and suppression linkage control based on hardware scheduling.

[0125] This disclosure also proposes a fire alarm management system based on a gateway device, applied to the aforementioned edge computing gateway device, the structure of which is as follows: Figure 2 As shown, it includes: system initialization and parameter loading module, initial data acquisition and probability allocation module, spatial conflict degree calculation module, temporal information entropy evaluation module, dynamic weight reduction correction module, evidence fusion and confidence generation module, multimodal severity evaluation module, secure communication delay allocation module, and linkage recommendation and execution module.

[0126] The system initialization and parameter loading module is used to perform the classification and loading configuration of parameters after the gateway device is powered on and started. It loads general computing parameters into the resident address area of ​​static random access memory, and the security encryption coprocessor autonomously completes the initialization of secure communication parameters.

[0127] The initial data acquisition and probability allocation module is used to acquire the initial behavioral data of multi-source heterogeneous sensing devices in the target area. After sliding window midpoint filtering and normalization preprocessing, the standardized feature data is transformed into basic probability allocation data for discrete physical environment states.

[0128] The spatial conflict degree calculation module is used to calculate the global conflict degree characteristics of each sensing device in the physical space dimension, so as to quantify the spatial discrete differences of data between sensing devices and thus identify abnormal sensing devices in the physical space.

[0129] The temporal information entropy evaluation module is used to calculate the temporal information entropy characteristics of each sensing device in the time dimension to quantify the disordered jump characteristics of data. It also has built-in zero-value bypass and hardware amplitude limiting and truncation logic to avoid numerical overflow anomalies in logarithmic operations.

[0130] The dynamic weight reduction correction module is used to calculate the hardware prior trust weight by combining the system prior parameters, perform joint penalty calculation of spatiotemporal multidimensional constraints on the basic probability allocation data, configure a hardware zero value verification mechanism to avoid division by zero operation anomalies from the hardware level, and output the corrected basic probability allocation data.

[0131] The evidence fusion and confidence generation module is used to select standard Dempster orthogonal synthesis or downgraded fusion processing based on the evidence synthesis conflict coefficient using a pairwise iterative synthesis method, and supports boundary processing of the fusion results directly output by a single sensing device, and finally extracts and generates the global fire confidence of the target area.

[0132] The multimodal severity assessment module is used to combine the visual recognition confidence probability and the global fire confidence level to perform multimodal situation weighting and generate comprehensive fire severity data that characterizes the fire spread situation.

[0133] The secure communication delay allocation module is used to adaptively allocate a secure communication delay budget as an initial countdown value to the hardware timer in the secure encryption coprocessor based on the comprehensive fire severity data and through inverse proportional mapping control.

[0134] The linkage recommendation and execution module is used to perform physical drive dimension alignment and synthesis calculation based on the comprehensive fire severity data and the security communication delay budget, generate recommended operation intensity data for the target execution device and convert it into control duty cycle, and send it to the target execution device after hardware encryption and message authentication code encapsulation.

[0135] It should be noted that the specific underlying hardware scheduling mechanism, quantization calculation model and rules, parameter flow and hardware triggering mechanism, and full-boundary scene processing logic of each functional module included in the above system embodiments are all one-to-one corresponding to and completely consistent with the corresponding steps (steps S0 to S8) in the aforementioned fire protection facility alarm management method embodiment based on gateway devices. To avoid redundancy and ensure the consistency of logic before and after the application documents, for detailed implementation processes of each system module's operation, please refer to the corresponding descriptions in the aforementioned method embodiments, which will not be repeated here.

[0136] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for managing fire alarm systems based on gateway devices, characterized in that, Applied to edge computing gateway devices, the method includes: The original environmental electrical signals of heterogeneous sensing devices in the target area are obtained through the communication controller of the gateway device and converted into basic probability allocation data for discrete physical environment states after preprocessing. Based on the spatial discrete difference characteristics of data between sensing devices, the temporal information entropy characteristics, and the inherent prior trust weights of sensing devices, the basic probability allocation data is subjected to joint penalty calculation and dynamic weight reduction correction through an arithmetic logic unit, and the corrected basic probability allocation data is output. The modified basic probability allocation data is processed by a digital signal processor to generate a global fire confidence score. The visual recognition confidence probability extracted by the edge AI accelerator is combined with multimodal situational weighting to generate comprehensive fire severity data. Based on the comprehensive fire severity data, inverse proportional mapping control is performed to adaptively allocate a secure communication delay budget to the hardware timer within the secure encryption coprocessor. Based on the comprehensive fire severity data and its differential slope, as well as the reciprocal of the safety communication delay budget, the actuator drive unit performs dimensional alignment and weighted summation operations to generate recommended operating intensity data for the corresponding target execution device. This data is then converted into control duty cycle and encrypted by hardware before being sent out.

2. The fire alarm management method based on a gateway device according to claim 1, characterized in that, Before acquiring the raw environmental electrical signals from heterogeneous sensing devices within the target area, the method further includes: The general computational parameters, including the physical sampling period, calibration parameter set, and dimension transformation and alignment physical matching coefficients, are loaded from non-volatile memory into the resident address area of ​​static random access memory. The security encryption coprocessor extracts the message authentication code key and encryption algorithm configuration parameters from its built-in secure storage area to complete the initialization operation. The message authentication code key does not enter the system's shared memory addressing area.

3. The fire alarm management method based on a gateway device according to claim 1, characterized in that, Before performing joint penalty calculation and dynamic weight reduction correction on the basic probability allocation data, the following feature extraction steps are also included: The discrete difference value of environmental data between any two sensing devices is quantified based on the Hellinger distance metric formula. The average deviation between the target sensing device and other devices is calculated by a hardware accumulator to generate a global conflict degree feature as the spatial discrete difference feature. The temporal information entropy features of the basic probability allocation data are calculated based on the Shannon entropy mathematical model. When the single-state probability input value is detected to be zero, the zero-value bypass circuit built into the information entropy conversion logic gate is triggered to directly output the summation result as zero. When the calculated feature value is greater than the set maximum environmental noise information entropy tolerance constant, the hardware limiting circuit built into the information entropy conversion logic gate is triggered to perform a truncation operation.

4. The fire alarm management method based on a gateway device according to claim 3, characterized in that, The step of performing joint penalty calculation and dynamic weight reduction correction on the basic probability allocation data, and outputting corrected basic probability allocation data, includes: Prior trust weights are constructed based on the log-likelihood ratio of detection rate to false alarm rate and the log-likelihood ratio of false negative rate to correct negative rate of sensing devices. For the target perception device, the basic probability allocation data, the linear penalty function term composed of the temporal information entropy features, the exponential penalty function term composed of the global conflict degree features, and the prior trust weight are multiplied together to calculate the denominator value of the sum of the penalty terms. The zero-value detection circuit is used to verify the denominator of the sum of the penalty terms. If the denominator of the sum of the penalty terms is greater than or equal to the minimum weight determination threshold, normalization processing is performed to output the corrected basic probability allocation data. If the denominator of the sum of the penalty terms is less than the minimum weight determination threshold, the corrected basic probability allocation data of the target sensing device is reset to an unknown state.

5. The fire alarm management method based on a gateway device according to claim 1, characterized in that, The step of performing evidence fusion processing on the corrected basic probability allocation data to generate a global fire confidence score includes: Obtain the total number of sensing devices online simultaneously. In response to the total number of sensing devices being 1, directly output the corrected basic probability allocation data of that device as the fusion result. When the total number of the sensing devices is greater than or equal to 2, fusion is performed using a pairwise iterative synthesis method. Before each pairwise synthesis operation, the evidence synthesis conflict coefficient is calculated: if the evidence synthesis conflict coefficient is less than the conflict threshold, standard Dempster orthogonal synthesis is performed; if the evidence synthesis conflict coefficient is greater than or equal to the conflict threshold, downgraded fusion processing is triggered, and a weighted average is performed according to the prior trust weights of the participating synthesis devices to replace orthogonal synthesis, and the equivalent prior weights of the current fusion result are updated synchronously for use in subsequent iterative synthesis. The numerical components pointing to the occurrence of open flame are extracted from the final fusion result after iteration and used as the global fire confidence level.

6. The fire alarm management method based on a gateway device according to claim 1, characterized in that, The step of performing inverse proportional mapping control based on the comprehensive fire severity data, and adaptively allocating a secure communication delay budget for the hardware timer within the secure encryption coprocessor, includes: Extract the pre-loaded baseline encryption tolerance latency parameters and network bandwidth adjustment constant; Perform an inverse proportional mapping calculation, taking the benchmark encryption tolerance delay parameter as the numerator component and the sum of the products of constant 1, network bandwidth adjustment constant, and comprehensive fire severity data as the denominator component, and obtain the division operation result; The result of the division operation is used as the budget for the secure communication delay and written into the hardware timer as the initial countdown value.

7. The fire alarm management method based on a gateway device according to claim 1, characterized in that, Based on the comprehensive fire severity data and its differential slope, and the reciprocal of the secure communication delay budget, dimensional alignment and weighted summation operations are performed to generate recommended operational intensity data for the corresponding target execution equipment, including: Extract the pre-loaded first dimension transformation and alignment physical matching coefficients, second dimension transformation and alignment physical matching coefficients, and third dimension transformation and alignment physical matching coefficients; The difference between the comprehensive fire severity data of the current period and the comprehensive fire severity data of the previous period is divided by the physical sampling period to obtain the differential slope. The first operational component is calculated by multiplying the first dimension transformation and alignment physical matching coefficient with the comprehensive fire severity data of the current period; the second operational component is calculated by multiplying the second dimension transformation and alignment physical matching coefficient with the differential slope; and the third operational component is calculated by multiplying the third dimension transformation and alignment physical matching coefficient with the reciprocal of the secure communication delay budget. A weighted summation operation is performed on the first operational component, the second operational component, and the third operational component to output dimensionless operational intensity recommendation data.

8. A fire alarm management system based on a gateway device, characterized in that, The system is used to implement the fire protection facility alarm management method based on a gateway device as described in any one of claims 1 to 7, wherein the system is configured on an edge computing gateway device and includes: The initial data acquisition and probability allocation module, which runs on the communication controller and central processing unit, is used to acquire the original environmental electrical signals of heterogeneous sensing devices in the target area, and convert them into basic probability allocation data for discrete physical environment states after preprocessing. The dynamic weight reduction correction module, which runs in the arithmetic logic unit, is used to perform joint penalty calculation and dynamic weight reduction correction on the basic probability allocation data based on the spatial discrete difference characteristics, temporal information entropy characteristics and the inherent prior trust weights of the data between sensing devices, and output the corrected basic probability allocation data. The evidence fusion and multimodal assessment module, running on a digital signal processor and an edge AI accelerator, is used to perform evidence fusion processing on the corrected basic probability allocation data to generate a global fire confidence score, and to perform multimodal situation weighting by combining visual recognition confidence probability to generate comprehensive fire severity data. The secure communication delay allocation module, which runs on the secure encryption coprocessor, is used to perform inverse proportional mapping control based on the comprehensive fire severity data and adaptively allocate the secure communication delay budget for the hardware timer. The linkage recommendation and execution module, running in the actuator drive unit, is used to perform dimensional alignment and weighted summation operations based on the comprehensive fire severity data and its differential slope, as well as the reciprocal of the safety communication delay budget, to generate recommended operation intensity data for the corresponding target execution device, convert it into control duty cycle, and send it out after hardware encryption.

9. A method for managing fire alarm facilities based on a gateway device according to claim 8, characterized in that, The dynamic weight reduction correction module is specifically used for: The global conflict degree feature is obtained by calculating the Hellinger distance metric formula and hardware accumulator, and is used as the spatial discrete difference feature. The time-domain information entropy feature is calculated based on the Shannon entropy mathematical model. When the single-state probability input value is detected to be zero, the zero-value bypass circuit built into the information entropy conversion logic gate is triggered, and when the feature value exceeds the limit, the hardware limiting circuit is triggered to perform a truncation operation. The prior trust weight is constructed based on the log-likelihood ratio of detection rate and false positive rate, and the log-likelihood ratio of false negative rate and correct rejection rate. This prior trust weight is then multiplied by the basic probability allocation data, the linear penalty function term composed of the temporal information entropy feature, and the exponential penalty function term composed of the global conflict degree feature. After verification by a hardware zero-value detection circuit, the corrected basic probability allocation data is output.

10. A fire alarm management method based on a gateway device according to claim 8, characterized in that, The linked recommendation and execution module is specifically used for: Extract the three sets of dimensional transformation and alignment physical matching coefficients; Calculate the differential slope of the comprehensive fire severity data on the first-order time axis; The current fire severity, differential slope, and reciprocal delay are multiplied by the corresponding dimension transformation and alignment physical matching coefficients to cancel out the time reciprocal dimension of the differential slope and reciprocal delay terms. After summing, the dimensionless recommended operation intensity data is output.

Images