Warehouse intelligent fire alarm system based on multi-source perception fusion

Abstract

The application discloses a warehouse intelligent fire alarm system based on multi-source sensing fusion, and relates to the technical field of warehouse fire alarm systems.The warehouse intelligent fire alarm system comprises a three-level heterogeneous sensing array, a multi-source fusion research and judgment module, a partition linkage execution module and a bidirectional feedback closed loop module.The three-level heterogeneous sensing array and a dynamic self-adaptive sampling calibration mechanism are used to break through the framework limitation of traditional warehouse fire alarm systems which only rely on ceiling plane detection, effectively eliminate the detection blind area formed by the shelf shielding and internal smoldering of goods in a high and large stereoscopic warehouse, and significantly shorten the alarm response time of smoldering fire;in combination with a dynamic frequency increasing sampling and sensitivity self-compensation mechanism, the equipment can be maintained in low-power operation and the service life can be prolonged under normal conditions, abnormal situations can be quickly formed into high-density monitoring grids, the sensitivity attenuation caused by equipment dust accumulation and aging can be automatically offset, the workload of manual calibration and troubleshooting can be reduced, and the detection accuracy can be continuously ensured.

Description

Technical Field

[0001] This invention relates to the field of warehouse fire alarm system technology, specifically a warehouse intelligent fire alarm system based on multi-source sensing fusion. Background Technology

[0002] As the warehousing and logistics industry develops towards higher levels, automation, and greater density, the importance of fire alarm systems in warehouse scenarios is becoming increasingly prominent. Simply put, a fire alarm system is based on detectors to monitor the composition of smoke in the warehouse in real time. When the smoke composition exceeds the threshold, an alarm is issued in a timely manner and corresponding fire-fighting operations are performed to prevent the fire from spreading.

[0003] Existing warehouse fire alarm systems generally adopt a ceiling-mounted point detector architecture. In tall, automated warehouses, due to the vertical dilution of smoke and the obstruction of multiple layers of shelves, the alarm delay for smoldering inside goods or fires deep within shelves can be more than 10 minutes, resulting in a large number of blind spots. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a warehouse intelligent fire alarm system based on multi-source sensing fusion, which solves the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a warehouse intelligent fire alarm system based on multi-source sensing fusion, comprising a three-level heterogeneous sensing array, a multi-source fusion analysis module, a zone linkage execution module, and a two-way feedback closed-loop module; the three-level heterogeneous sensing array adopts a three-dimensional distributed deployment structure with top-level full-domain detection, middle-level shelf monitoring, and bottom-level cargo location sensing, used to collect fire characteristic parameters at different spatial scales and upload them to the multi-source fusion analysis module; The multi-source fusion analysis module has a built-in fire confidence fusion algorithm with multi-dimensional correction of space, cargo, and environment, which is used to perform weighted fusion and dynamic correction of the perceived data and output the fire confidence of the target area. The zone linkage execution module has a built-in "risk-loss" dual-objective linkage decision model, which is used to dynamically select the optimal linkage zone scheme based on the fire confidence level and trigger the fire-fighting equipment in the corresponding area. The bidirectional feedback closed-loop module is used to feed back the judgment results and the linkage effect to the three-level heterogeneous sensing array and the multi-source fusion judgment module, dynamically adjust the sensing sampling strategy, environmental baseline threshold and algorithm weight parameters, and realize the self-closed-loop operation of perception-judgment-linkage-optimization.

[0006] Furthermore, the three-level heterogeneous sensing array specifically includes: The top-level global detection layer, located on the warehouse roof, consists of an aspirating early smoke detection pipeline and a wide-angle infrared thermal imager, responsible for monitoring the aerosol concentration and temperature field of the entire space. The middle-level shelving distribution layer is deployed on the beams of each shelf layer and consists of multiple wireless composite detection nodes. Each node integrates photoelectric smoke detection, digital temperature detection, temperature and humidity detection and RFID reading and writing units, and is responsible for collecting fire characteristic parameters between shelf layers. The bottom layer of cargo location sensing consists of passive RFID temperature sensing tags embedded in cargo pallets. It is powered and activated by the radio frequency field of the middle layer composite node to directly monitor the internal temperature rise of the cargo.

[0007] Furthermore, the three-level heterogeneous sensing array is equipped with a dynamic adaptive sampling calibration mechanism: under normal conditions, the middle-layer nodes and the bottom-layer tags are in a low-power sampling state, while the top layer maintains regular sampling; when any node parameter triggers the warning threshold, all sensing nodes in the spatial neighborhood are immediately activated to upsample; at the same time, it receives instructions from the bidirectional feedback closed-loop module to dynamically update the environmental baseline threshold and compensate for the attenuation of device sensitivity.

[0008] Furthermore, the multi-source fusion judgment module first normalizes the detection values ​​of a single sensor and calculates the normalized risk level of that single sensor. The calculation formula is as follows:

[0009] in, For the first Normalized hazard level of each sensor; This refers to the real-time detection value of the sensor. This refers to the environmental baseline value for the corresponding area; The national standard legal fire alarm threshold.

[0010] Furthermore, the fire confidence fusion algorithm, based on weighted fusion, introduces three types of dynamic parameters: spatial location weight coefficient, cargo combustion characteristic coefficient, and environmental interference correction factor. The final fire confidence calculation formula is as follows:

[0011] in, The fire confidence level of the target fire compartment; This represents the total number of online sensors within the partition. The inherent weights are for the sensor type. Spatial location weighting coefficient; This refers to the combustion characteristic coefficient of the cargo. This is an environmental disturbance correction factor.

[0012] Furthermore, the aforementioned "risk-loss" dual-objective linkage decision-making model aims to minimize the overall disposal cost. It selects the optimal linkage zone from multiple levels of solutions, including location-level, shelf-level, fire-prevention zone-level, and full-warehouse-level solutions. The calculation formula is as follows:

[0013] in, The optimal linkage partitioning scheme; This is the set of all possible linkage schemes; This refers to the expected secondary losses corresponding to the proposed solution. This represents the spread risk value under the corresponding fire confidence level; This is the loss-risk balance coefficient.

[0014] Furthermore, the bidirectional feedback closed-loop module includes a sub-module for verifying the handling effect: after the linkage action is executed, the fire confidence level of the target area is continuously monitored through a three-level heterogeneous sensing array; if the built-in confidence level continues to decline within a specified time, the handling is deemed effective and the linkage is gradually downgraded; if the confidence level continues to rise, the handling is deemed ineffective and the linkage level is automatically upgraded and the zoning range is expanded.

[0015] Furthermore, the system connects to the Warehouse Management System (WMS) to obtain real-time information on the type, flammability, and value of goods in each storage location, and dynamically updates the flammability coefficient and secondary loss calculation parameters.

[0016] Furthermore, the bidirectional feedback closed-loop module also includes a device status assessment submodule: by long-term tracking of the environmental baseline offset, it determines the degree of sensor sensitivity attenuation, automatically compensates for sampling warning thresholds, and generates accurate maintenance work orders to be pushed to the management terminal.

[0017] This invention provides a warehouse intelligent fire alarm system based on multi-source sensing fusion, which has the following beneficial effects: 1. This intelligent warehouse fire alarm system based on multi-source sensing fusion utilizes a three-level heterogeneous sensing array and a dynamic adaptive sampling calibration mechanism to break through the limitations of traditional warehouse fire alarms that rely solely on ceiling-level detection. Through a three-dimensional layered deployment of the top-level full-area, middle-level shelves, and bottom-level storage locations, it effectively eliminates detection blind spots caused by shelf obstruction and smoldering inside goods in tall, automated warehouses, significantly shortening the alarm response time for smoldering fires. Combined with dynamic upsampling and sensitivity self-compensation mechanisms, it maintains low-power operation and extends battery life under normal conditions, while quickly forming a high-density monitoring grid when anomalies occur. It can also automatically offset the sensitivity decay caused by equipment aging and dust accumulation, reducing the workload of manual calibration and troubleshooting, and continuously ensuring detection accuracy.

[0018] 2. This intelligent fire alarm system for warehouses, based on multi-source sensing fusion, employs a multi-dimensional fire confidence fusion algorithm and a two-way feedback linkage mechanism. This overcomes the adaptability limitations of traditional fixed-weight threshold judgments. By introducing three types of dynamic correction factors—spatial location, cargo combustion characteristics, and environmental interference—it can accurately distinguish non-combustion interferences such as dust and condensation in the storage environment, significantly reducing the probability of false alarms and missed alarms. Coupled with a "risk-loss" dual-objective linkage decision-making model, it can dynamically select the optimal handling range based on the fire level and cargo attributes, avoiding secondary losses to cargo caused by excessive linkage. Furthermore, relying on closed-loop verification of handling effects and parameter self-iteration capabilities, it can continuously optimize system performance as the warehouse environment and cargo layout change, achieving long-term stable adaptive operation. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the steps of a warehouse intelligent fire alarm system based on multi-source sensing fusion according to the present invention. Detailed Implementation

[0020] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0021] like Figure 1 As shown, the present invention provides a technical solution: a warehouse intelligent fire alarm system based on multi-source sensing fusion. The application scenario of this embodiment is an e-commerce automated warehouse. The warehouse is 80m long, 40m wide, and 12m high. It is equipped with 5 layers of beam-type shelves, each layer is 2.2m high. It mainly stores cardboard-packaged daily necessities and small household appliances. Goods are frequently entering and leaving the warehouse. The warehouse area has typical interferences such as dust from forklift operations and condensation due to temperature differences at loading and unloading points.

[0022] The system is divided into five logical layers, from bottom to top: perception layer, transmission layer, judgment layer, execution layer and feedback layer, with each layer being bidirectionally connected. The perception layer, namely the three-level heterogeneous perception array, is responsible for collecting fire characteristic data throughout the entire space. The transmission layer adopts a hybrid networking of fire protection dual bus and LoRa wireless. The top-level equipment uses wired transmission, while the middle-level wireless nodes and the bottom-level tags upload data via LoRa, balancing reliability and deployment flexibility. The analysis layer is deployed on local fire alarm controllers and cloud management platforms, running fusion algorithms and decision models; The execution layer connects to the existing sprinkler system, audible and visual alarms, fireproof roller shutters, non-fire-fighting power supply, ventilation and smoke extraction, access control and other equipment in the warehouse; The feedback layer enables reverse control of the judgment and execution results to the perception and judgment layers, thus completing the self-closed-loop optimization.

[0023] The specific deployment and operation of the three-level heterogeneous sensing array are as follows: Top-level global detection layer: Four aspirating early smoke detection sampling pipelines are laid longitudinally at a distance of 0.5m below the warehouse ceiling, with each pipeline spaced 10m apart. A sampling hole is opened every 3m on the pipeline. The pipeline can actively extract air from the warehouse area for smoke detection, with a sensitivity of up to 0.005%obs / m, capturing very early and weak smoke signals.

[0024] Meanwhile, one wide-angle infrared thermal imager is deployed for every 200 square meters, with a total of 8 units deployed throughout the warehouse. The temperature measurement range is -20℃ to 150℃, and the temperature measurement accuracy is ±0.5℃. It can generate one frame of the global temperature field every second and identify local temperature rises and open flame hotspots.

[0025] Under normal conditions, the inhalation system samples once every 2 seconds, while the thermal imager samples one frame every 10 seconds, responsible for comprehensive monitoring of the entire area.

[0026] Middle shelf distribution levels: On the beams of each shelf, one wireless composite detection node is installed every 5 meters, for a total of 160 nodes across the five shelves of the warehouse. Each node integrates a photoelectric smoke sensor, a digital temperature sensor, a temperature and humidity sensor, a LoRa communication module, and an RFID reader / writer module, and is powered by a lithium-ion battery, eliminating the need for wiring.

[0027] This layer of nodes is deployed directly between the shelf layers, closer to the goods, and not obstructed by the upper shelves. It can capture the smoke and temperature rise spreading between the shelves in the first time, making up for the shortcomings of the roof detectors which have a long vertical distance and diluted signal.

[0028] Under normal conditions, the node samples smoke, temperature, and humidity data every 30 seconds, operating in a low-power state.

[0029] Bottom layer of cargo location perception: A passive RFID temperature sensing tag is embedded inside each cargo pallet. When goods enter the warehouse, the tag ID is bound to the location and cargo information by scanning the code. The tag itself does not have a battery; it is powered and activated by the radio frequency field of the middle-layer composite node. After activation, it can collect the internal temperature of the cargo and transmit the data back.

[0030] This layer is in direct contact with the cargo and can monitor the smoldering temperature rise inside the cargo, completely solving the traditional blind spot where "the smoke has already spread to the cargo before it even appears".

[0031] Under normal conditions, the middle-level node wakes up every 5 minutes to read the temperature of the RFID tags in the corresponding area, balancing low power consumption and basic monitoring.

[0032] During normal operation, the system maintains a low-power sampling strategy, with overall power consumption only 15% of that of full-frequency sampling, and wireless nodes can last for more than 3 years. The complete sampling calibration and interference self-learning process of the system is as follows: Early warning triggering and local frequency upsampling process The system pre-sets the smoke warning threshold for mid-level nodes to 30% of the national standard fire alarm threshold of 2.0%obs / m, i.e., 0.6%obs / m. When the mid-level composite node M2-A-07 in Zone A of the second floor detects that the smoke concentration instantaneously reaches 0.62%obs / m, triggering the warning threshold, the system immediately performs the following local frequency boosting operation: The sampling frequency of node M2-A-07 itself has been increased from 30 seconds / time to 1 second / time, continuously collecting smoke, temperature, and humidity data; Automatically activate 8 adjacent mid-level nodes within a 10m radius centered on this node, and simultaneously increase the sampling frequency to 1 second / time; The reading frequency of the 12 passive RFID temperature tags within the corresponding coverage area has been increased from 5 minutes / time to 10 seconds / time. The two infrared thermal imagers in the corresponding area of ​​the ceiling were upgraded from 10 seconds / frame to 1 second / frame to focus on monitoring the temperature field changes in that area.

[0033] This allows for the rapid formation of a high-density three-dimensional monitoring grid in abnormal areas, accurately capturing parameter change trends and avoiding missed fire detections caused by routine low-frequency sampling.

[0034] After 60 seconds of upsampling, the multi-source fusion analysis module initiates the interference identification and attribute determination process, employing a three-level determination logic for forklift dust interference: Multi-parameter synergy verification: Comparing the synchronicity of changes in three core fire characteristic parameters—smoke, temperature, and CO—within the same spatial range. In actual combustion, an increase in smoke concentration is inevitably accompanied by a simultaneous increase in temperature and CO concentration. In this test, only the photoelectric smoke detector showed a step increase, while the temperature rise rate of the digital temperature sensor remained within 0.1℃ / min, and the CO concentration remained stable at 2-3 ppm. The relative changes were all less than 5%, which was initially determined to be interference from non-combustible particulate matter.

[0035] Time curve feature matching: Extract the complete change curve of smoke concentration within 60 seconds for feature comparison. Forklift dust exhibits typical pulse characteristics of "rapid rise, short peak, and rapid fall": The data shows that the smoke concentration took 8 seconds to rise from the baseline to the peak, the peak remained stable for 18 seconds, and by 60 seconds it had fallen back to near the baseline level, which is consistent with the curve characteristics of dust interference; while the smoke concentration of a real smoldering fire shows a continuous and slow upward trend, without a rapid fall, and the curve characteristics of the two are significantly different.

[0036] Scene attribute cross-validation: Matching the spatial attributes of this location with the historical sample database. This location is next to the main forklift operating aisle, which is a high-dust-prone area. Furthermore, the historical database shows that this location has exhibited similar curve characteristics anomalies 6 times, all of which have been verified on-site as forklift dust interference, further increasing the confidence level of the interference determination.

[0037] When all three of the above conditions are met, the analysis module will ultimately determine that the anomaly is dust interference from forklift operations, and will not escalate it to a fire alarm level, but will only record it as an interference event.

[0038] Upon determining that dust interference has occurred, the two-way feedback closed-loop module immediately sends a baseline update command to all sensing nodes in the area, and updates the smoke environment baseline value using a damped adaptive algorithm. The specific calculation and adjustment rules are as follows: First, extract the average value of the stable segment during the last 10 seconds of the smoke concentration decline phase during this interference process as the interference baseline value. Peak spikes were removed. The results of this calculation are... The new environmental baseline value was calculated using the damping increment method. The calculation formula is:

[0039] in, The environmental baseline value before the update is 0.1%obs / m. The damping coefficient is updated to a fixed value of 0.25 to ensure that a single disturbance only slightly raises the baseline, avoiding sensitivity fluctuations caused by large baseline jumps; the new baseline value is calculated to be 0.115%obs / m. The updated baseline value must not exceed 20% of the national standard fire alarm threshold. After three consecutive baseline updates at the same node, a seven-day observation period is required. During the observation period, baseline adjustments are suspended to prevent the baseline from continuously rising and ultimately leading to missed fire reports. If there are no similar interference events for the next seven consecutive days, the baseline will gradually fall back to the initial calibration level at a rate of 2% per day to avoid sensitivity loss caused by permanent increases. Through this mechanism, the system can gradually adapt to the normal dust levels in the area, and subsequent forklift dust of the same intensity will no longer trigger an alarm, achieving interference adaptive learning. To address the sensitivity degradation issue during long-term equipment operation, the system's equipment status assessment submodule performs periodic self-diagnosis and compensation adjustments. The specific judgment and adjustment process is as follows: Sensitivity attenuation determination: Using a 7-day evaluation cycle, the daily average environmental baseline value of a single node is calculated and compared with the average daily baseline value of three adjacent normal nodes in the same area. If the baseline value of the node is more than 15% higher for two consecutive evaluation cycles, and external factors have been ruled out through temperature and humidity correction, cargo change investigation, and regional environmental comparison, then the sensor is determined to have sensitivity attenuation due to dust accumulation in the optical cavity and component aging.

[0040] Early warning compensation coefficient adjustment: The system sets an independent early warning compensation coefficient for each node. The initial value is 1.0, used to correct the sensor's detection output value. The formula for calculating the corrected effective detection value is:

[0041] After each sensitivity decay is detected, the compensation coefficient is increased in a fixed step of 0.05 to amplify the detected value and offset the numerical deviation caused by the decrease in sensitivity. The compensation coefficient is set with an upper limit of 1.5 to prevent over-compensation from causing false alarms. When the compensation coefficient reaches the upper limit, the system will generate a first-level maintenance work order, accurately marking the node location, attenuation degree and suggested handling method, and push it to the warehouse management terminal to force the management personnel to clean and calibrate the node or replace the hardware.

[0042] After the adjustment is completed, the warning threshold and fire alarm threshold of this node are calculated based on the corrected effective detection value, so as to offset the detection deviation caused by sensitivity decay and continuously ensure detection accuracy before manual operation and maintenance intervention.

[0043] When smoldering of cardboard boxes occurs at the storage location corresponding to node M2-A-07 in Zone A on the second floor, all multi-layer sensors in this area detect abnormal signals. The system calculates the fire confidence level in the following three steps: Step 1: Calculate the normalized hazard level for a single sensor First, the detection values ​​of sensors of different types and dimensions are normalized and mapped uniformly to the [0,1] interval to eliminate dimensional differences. The normalization formula is as follows: in These are the effective detection values ​​of the sensors after correction by the compensation coefficient. Taking five typical sensors in this area as examples: Mid-layer photoelectric smoke sensor: Corrected real-time value Environmental baseline National standard fire alarm threshold Substituting into the formula, we get ; Mid-layer digital temperature sensor: real-time value baseline Fire alarm threshold ,have to ; Mid-layer CO sensor: Real-time value baseline Fire alarm threshold ,have to ; Passive RFID Temperature Tag: Real-time Value baseline Alarm threshold ,have to ; Top-level infrared thermal imager (temperature rise rate): Real-time value baseline Alarm threshold ,have to ; Step 2: Determine the dynamic correction parameters Sensor type inherent weights Based on the preset detection reliability, the values ​​are: smoke sensor 0.25, temperature sensor 0.15, CO 0.2, RFID temperature tag 0.2, and thermal imager 0.2. Spatial location weighting coefficient The coefficient is calculated based on the sensor's vertical height and the degree of shelf obstruction; the lower the height and the less obstruction, the larger the coefficient. A coefficient of 0.9 is used for mid-level nodes, 0.5 for top-level thermal imagers due to smoke dilution and distance effects, and 1.0 for RFID tags directly contacting goods. Cargo combustion characteristic coefficient : Connect to the WMS system to obtain information about the goods in this location. The goods are daily necessities packaged in cardboard boxes. They are flammable and produce a lot of smoke. Take 1.2. Environmental disturbance correction factor Based on the current temperature and humidity, airflow speed, and historical interference frequency, the current regional environment is stable, so we take 0.05.

[0044] Step 3: Calculate the final fire confidence level Substitute the above parameters into the fire confidence fusion formula:

[0045] Calculated The fire alarm falls within the Level 2 fire alarm confidence range (0.4~0.7). The system determines this as a confirmed fire and triggers the linkage decision-making process.

[0046] The system has a pre-set four-level linkage scheme: storage location level, shelf row level, fire compartment level, and overall warehouse level. It has a built-in "risk-loss" dual-objective decision-making model, which selects the optimal solution with the goal of minimizing the overall disposal cost. The decision formula is as follows: In conjunction with this Based on the fire alarm level and the characteristics of the goods in the area (ordinary daily necessities, waterproof but with low unit value), the warehouse pre-sets a loss-risk balance coefficient. This approach balances fire containment with the control of secondary losses.

[0047] Calculate the combined cost of each of the four options: Location-level linkage: Only the local aerosol fire suppression system at that location is activated, simultaneously triggering a single-point audible and visual alert. Expected secondary losses. risk of spread , the comprehensive cost = 0.4×0.05 + 0.6×0.7 = 0.44.

[0048] Rack row-level linkage: Activate the audible and visual alarms of the entire corresponding rack row, cut off the non-fire power supply of this rack row, start the high-pressure water mist fire extinguishing device in the corresponding area, and close 3 groups of adjacent vents. Expected secondary loss , spread risk , the comprehensive cost = 0.4×0.2 + 0.6×0.25 = 0.23.

[0049] Fire compartment-level linkage: Activate the sprinkler system of the entire fire compartment, full-area audible and visual alarms and fire broadcast, cut off the non-fire power supply in the compartment, and start the smoke exhaust fan. Expected secondary loss , spread risk , the comprehensive cost = 0.4×0.5 + 0.6×0.05 = 0.23.

[0050] Whole warehouse-level linkage: Power off and shut down the entire warehouse, activate the audible and visual alarms of the whole warehouse and all fire-fighting facilities. Expected secondary loss , spread risk , the comprehensive cost = 0.4×0.95 + 0.6×0.01 = 0.386.

[0051] It can be seen from the comparison that the rack row level and the fire compartment level have the same comprehensive cost, and the system further selects the rack row level scheme with a smaller scope as the optimal solution, which not only effectively controls the spread of the fire, but also avoids the large-area wet damage of goods caused by the sprinkler in the entire compartment.

[0052] After the linkage action is executed, the disposal effect verification sub-module continuously monitors the change of the fire confidence level in the target area through the three-level sensing array, performs hierarchical verification according to time nodes and makes dynamic adjustments: First verification at 30 seconds: 30 seconds after the linkage is executed, the change trend of the fire confidence level is verified for the first time. If the confidence level continues to drop below 0.2 within 30 seconds, it is determined that the initial disposal is effective, the current linkage level is maintained, and continuous monitoring is carried out; if the confidence level still remains above 0.4 with no downward trend, it is determined that the initial disposal effect is not good, and the linkage scope is automatically expanded to the adjacent rack row.

[0053] Second verification at 60 seconds: The second verification is performed 60 seconds after the linkage is executed. If the fire confidence level continues to rise above 0.8, it is determined that the single-row linkage is insufficient to control the fire, the system automatically upgrades the linkage to the fire compartment level, activates the sprinkler and smoke exhaust systems, and simultaneously pushes fire alarm information to the warehouse person in charge and the local fire department, with key information such as the warehouse location, type of stored materials, fire area, and real-time fire data attached.

[0054] After the fire is extinguished, the system will be downgraded and reset: If the fire confidence level remains below 0.1 for 10 consecutive minutes, the fire is considered to be extinguished. The system will then gradually downgrade and reset: First, the fire extinguishing device will be shut down. After 5 minutes, the power supply will be restored. After 10 minutes, all devices will resume normal low-power sampling, and the system will return to normal monitoring status.

[0055] During long-term system operation, the bidirectional feedback closed-loop module continuously drives the system's self-optimization: Algorithm iteration and optimization: The system automatically records complete data and handling results for each interference, warning, and fire alarm. The environmental interference factor calculation model and sensor inherent weights are iteratively calibrated monthly to continuously improve adaptability to the warehouse scenario and continuously reduce the false alarm rate.

[0056] Risk warning and auxiliary management: The system counts the frequency of warnings and the number of abnormal temperature rises in each area on a weekly basis, generates a warehouse fire risk heat map, marks high-risk areas, and assists managers in optimizing cargo layout, clearing flammable accumulation hazards, and repairing aging electrical circuits, thus upgrading from "passive alarm" to "active fire prevention".

[0057] Equipment health management: Continuously track the baseline offset, signal strength, and battery status of all sensing nodes, predict equipment failures in advance, generate hierarchical maintenance work orders, and transform the traditional "annual inspection" into "precise on-demand maintenance", which greatly reduces the workload and cost of maintenance.

[0058] In summary, this intelligent warehouse fire alarm system based on multi-source sensing fusion utilizes a three-level heterogeneous sensing array and a dynamic adaptive sampling calibration mechanism. It breaks through the limitations of traditional warehouse fire alarms that rely solely on ceiling-level detection. Through a three-dimensional layered deployment of the top-level full-area, middle-level shelving, and bottom-level storage locations, it effectively eliminates detection blind spots caused by shelving obstruction and smoldering inside goods in tall, automated warehouses, significantly shortening the alarm response time for smoldering fires. Combined with dynamic upsampling and sensitivity self-compensation mechanisms, it maintains low-power operation and extends battery life under normal conditions, while rapidly forming a high-density monitoring grid when anomalies occur. Simultaneously, it automatically offsets sensitivity attenuation caused by dust accumulation and aging, reducing the workload of manual calibration and troubleshooting, and continuously ensuring detection accuracy.

[0059] The multi-dimensional fire confidence fusion algorithm and two-way feedback linkage mechanism overcome the adaptability defects of traditional fixed-weight threshold judgment. By introducing three types of dynamic correction factors—spatial location, cargo combustion characteristics, and environmental interference—it can accurately distinguish non-combustion interferences such as dust and condensation in the storage scenario, significantly reducing the probability of false alarms and false negatives. Coupled with the "risk-loss" dual-objective linkage decision model, it can dynamically select the optimal handling range based on the fire level and cargo attributes, avoiding secondary cargo losses caused by excessive linkage. At the same time, relying on the closed-loop verification of handling effects and parameter self-iteration capabilities, it can continuously optimize system performance as the warehouse environment and cargo layout change, achieving long-term stable adaptive operation.

[0060] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A warehouse intelligent fire alarm system based on multi-source sensing fusion, comprising a three-level heterogeneous sensing array, a multi-source fusion analysis module, a zone linkage execution module, and a bidirectional feedback closed-loop module; characterized in that: The three-level heterogeneous sensing array adopts a three-dimensional distributed deployment structure with top-level full-domain detection, middle-level shelf monitoring, and bottom-level cargo location sensing. It is used to collect fire characteristic parameters at different spatial scales and upload them to the multi-source fusion analysis module. The multi-source fusion analysis module has a built-in fire confidence fusion algorithm with multi-dimensional correction of space, cargo, and environment, which is used to perform weighted fusion and dynamic correction of the perceived data and output the fire confidence of the target area. The zone linkage execution module has a built-in "risk-loss" dual-objective linkage decision model, which is used to dynamically select the optimal linkage zone scheme based on the fire confidence level and trigger the fire-fighting equipment in the corresponding area. The bidirectional feedback closed-loop module is used to feed back the judgment results and the linkage effect to the three-level heterogeneous sensing array and the multi-source fusion judgment module, dynamically adjust the sensing sampling strategy, environmental baseline threshold and algorithm weight parameters, and realize the self-closed-loop operation of perception-judgment-linkage-optimization.

2. The intelligent fire alarm system for warehouses based on multi-source sensing fusion according to claim 1, characterized in that: The three-level heterogeneous sensing array specifically includes: The top-level global detection layer, located on the warehouse roof, consists of an aspirating early smoke detection pipeline and a wide-angle infrared thermal imager, responsible for monitoring the aerosol concentration and temperature field of the entire space. The middle-level shelving distribution layer is deployed on the beams of each shelf layer and consists of multiple wireless composite detection nodes. Each node integrates photoelectric smoke detection, digital temperature detection, temperature and humidity detection and RFID reading and writing units, and is responsible for collecting fire characteristic parameters between shelf layers. The bottom layer of cargo location sensing consists of passive RFID temperature sensing tags embedded in cargo pallets. It is powered and activated by the radio frequency field of the middle layer composite node to directly monitor the internal temperature rise of the cargo.

3. The intelligent fire alarm system for warehouses based on multi-source sensing fusion according to claim 1, characterized in that: The three-level heterogeneous sensing array is equipped with a dynamic adaptive sampling and calibration mechanism: under normal conditions, the middle-level nodes and the bottom-level tags are in a low-power sampling state, while the top-level nodes maintain regular sampling; when any node parameter triggers the warning threshold, all sensing nodes in the spatial neighborhood are immediately activated to upsample; at the same time, it receives instructions from the bidirectional feedback closed-loop module to dynamically update the environmental baseline threshold and compensate for the attenuation of device sensitivity.

4. The intelligent fire alarm system for warehouses based on multi-source sensing fusion according to claim 1, characterized in that: The multi-source fusion assessment module first normalizes the detection values ​​of a single sensor and calculates the normalized risk level of that single sensor using the following formula: ;in, For the first Normalized hazard level of each sensor; This refers to the real-time detection value of the sensor. This refers to the environmental baseline value for the corresponding area; The national standard legal fire alarm threshold.

5. A warehouse intelligent fire alarm system based on multi-source sensing fusion according to claim 4, characterized in that: The fire confidence fusion algorithm, based on weighted fusion, introduces three types of dynamic parameters: spatial location weight coefficient, cargo combustion characteristic coefficient, and environmental interference correction factor. The final fire confidence calculation formula is as follows: ;in, The fire confidence level of the target fire compartment; This represents the total number of online sensors within the partition. The inherent weights are for the sensor type. Spatial location weighting coefficient; This refers to the combustion characteristic coefficient of the cargo. This is an environmental disturbance correction factor.

6. A warehouse intelligent fire alarm system based on multi-source sensing fusion according to claim 1, characterized in that: The aforementioned "risk-loss" dual-objective linkage decision-making model aims to minimize the overall disposal cost. It selects the optimal linkage zone from multiple levels of solutions, including warehouse location, shelving row, fire prevention zone, and overall warehouse level. The calculation formula is as follows: ;in, The optimal linkage partitioning scheme; This is the set of all possible linkage schemes; This refers to the expected secondary losses corresponding to the proposed solution. This represents the spread risk value under the corresponding fire confidence level; This is the loss-risk balance coefficient.

7. A warehouse intelligent fire alarm system based on multi-source sensing fusion according to claim 1, characterized in that: The bidirectional feedback closed-loop module includes a sub-module for verifying the handling effect: after the linkage action is executed, the fire confidence level of the target area is continuously monitored through a three-level heterogeneous sensing array; if the built-in confidence level continues to decline within a specified time, the handling is deemed effective and the linkage is gradually downgraded; if the confidence level continues to rise, the handling is deemed ineffective and the linkage level is automatically upgraded and the zoning range is expanded.

8. A warehouse intelligent fire alarm system based on multi-source sensing fusion according to claim 1, characterized in that: The system connects to the Warehouse Management System (WMS) to obtain real-time information on the type, flammability, and value of goods in each storage location, and dynamically updates the flammability coefficient and secondary loss calculation parameters.

9. A warehouse intelligent fire alarm system based on multi-source sensing fusion according to claim 1, characterized in that: The bidirectional feedback closed-loop module also includes a device status assessment submodule: by long-term tracking of the environmental baseline offset, it determines the degree of sensor sensitivity attenuation, automatically compensates for sampling warning thresholds, and generates accurate maintenance work orders to be pushed to the management terminal.

Images