Mesh Network Alarm System and Control Method Based on Fire Risk
By hierarchically fusing fire sensing and network status features, a cross-layer control vector is generated, and the communication parameters of the MESH network are dynamically adjusted. This solves the problem of inappropriate reliability and energy consumption in fire alarms and achieves balanced transmission under different fire urgency levels and network conditions.
Patent Information
- Application Number
- CN202611418801.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-14
- Publication Date
- 2026-11-10
AI Technical Summary
Existing fire alarm MESH networks struggle to balance low power consumption and high reliability. Fixed communication parameters cannot adapt to different fire urgency levels and network conditions, resulting in insufficient information transmission reliability and inappropriate energy consumption.
By collecting fire sensor features and network status features, layered fusion is performed to form node, regional and network risks, generate cross-layer control vectors, and dynamically adjust routing selection, retransmission, polling and power consumption parameters to meet the balance of reliability, timeliness and energy consumption.
To improve transmission reliability and timeliness when the risk of fire increases, maintain low power consumption during normal operation, avoid network-wide interference caused by false alarms, achieve dynamic balance of protocol stack parameters, and adapt to different levels of fire urgency and network conditions.
Smart Images

Figure CN123072136A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of fire alarm technology, specifically relating to a MESH network alarm system based on fire risk and its control method. Background Technology
[0002] Battery-powered smoke, temperature, and carbon monoxide fire alarms typically operate as low-power terminals connected to a mesh network. Under normal conditions, these devices need to maximize sleep time and minimize polling and message transmission to maintain battery life for several years. However, in the event of a fire, alarm information must be reliably transmitted to the gateway, other alarms, and linked devices within strict time constraints. This forces the fire alarm network's protocol stack to constantly operate under conflicting requirements of low power consumption, high reliability, and low latency, with the degree of this conflict continuously changing depending on the situation on-site.
[0003] Currently, most wireless protocol stacks used in alarm systems employ fixed communication parameters, such as fixed polling cycles, fixed maximum retransmission counts, fixed transmit power, fixed routing metric weights, and fixed heartbeat cycles. These fixed parameters are set as a compromise based on general operating conditions and are difficult to adapt to operating states with vastly different levels of urgency, such as normal monitoring, suspected fire observation, and confirmed fire alarms. They are also difficult to handle scenarios where network topology is damaged, such as gateway power outages or failure of some routing nodes: if the parameters are too conservative, alarm information may not be delivered within the specified time; if the parameters are too aggressive, the device will waste power during long periods of normal operation.
[0004] Some solutions increase the transmission frequency, transmission power, or number of retries after an alarm is detected in an attempt to improve the transmission performance of alarm information. However, this approach, which relies solely on local sensor triggering and unidirectionally enhances communication parameters, has significant drawbacks: Firstly, in a fire, multiple nodes often trigger alarms almost simultaneously. If each node synchronously increases its transmission strength and number of retries, it will cause a sharp increase in channel contention, collisions, and duplicate messages, thereby reducing the actual arrival rate of alarm information. Secondly, relying solely on whether local sensors exceed their limits to determine the communication strategy cannot distinguish between a real fire, consistent alarms from multiple devices, and localized false alarms from a single device. This can easily lead to unnecessary upgrades to the entire network communication due to individual false alarms, or insufficient response to real fire situations. Furthermore, selecting a forwarding path solely based on the received signal strength can easily lead to the selection of nodes with poor link quality, queue congestion, or depleted battery power, resulting in transmission failures or frequent route switching. Fixedly shortening the polling cycle significantly increases receiving power consumption, while a fixed number of retries cannot accommodate link conditions with vastly different quality levels, wasting energy on good links and failing to guarantee delivery on poor links. Moreover, when sensor readings fluctuate around a threshold, communication parameters frequently switch between high and low power states, causing protocol stack oscillations and accelerating battery wear.
[0005] Therefore, it is necessary to enable the protocol stack parameters of alarm-related communications in MESH networks to be adjusted in an orderly manner according to the actual level of fire risk and the actual state of the network. This ensures reliable and timely transmission of alarm information when the fire risk increases, maintains low-power operation when the risk is normal or low, and avoids indiscriminate synchronous enhancement of transmission by all network nodes when network congestion or topology damage occurs. This achieves a dynamic balance between reliability, timeliness, energy consumption and network stability. Summary of the Invention
[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of insufficient reliability of MESH networks used for fire alarm in the prior art, thereby providing a MESH network alarm system and its control method based on fire risk.
[0007] A control method for a fire risk-based mesh network alarm system, the method comprising: The fire sensing characteristics and network status characteristics of the alarm node are collected. The network status characteristics include link quality characteristics, channel congestion characteristics, topology impairment characteristics, and remaining energy characteristics. The fire sensing features and the network status features are fused in a hierarchical manner to obtain node risk, regional risk and network risk, and the risk status of the alarm node is determined accordingly. Based on the risk status, a candidate combination of cross-layer control vectors is generated from a limited set of candidate parameters allowed by the protocol. The cross-layer control vectors include at least routing parameters, retransmission parameters, polling parameters, and power consumption parameters. Candidate combinations that do not meet the constraints are filtered out, based on the constraints of target reliability, message deadline, upper limit of protocol parameters, battery safety reserve energy and maximum duty cycle. The multi-objective cost is calculated for each of the selected candidate combinations, and the candidate combination with the lowest cost is selected as the target control vector. The weights of each item in the multi-objective cost change with the risk state. The protocol stack parameters are adjusted according to the target control vector.
[0008] Furthermore, the layered fusion includes: The node risk is determined by sensing risk, trend risk, cross-sensor consistency risk, duration risk, and false alarm suppression term, wherein the false alarm suppression term is only used to reduce the escalation of communication parameters without masking local alarms; The regional risk is determined by fusion of neighbor risks with time and spatial weights, and the regional risk is formed based at least on the maximum value of neighbor risk, the number of neighbor nodes that reach the warning level or above, and the weighted average of neighbor risks. The network risk is obtained by fusing link risk, congestion risk, and topology damage risk, and a segmented rule is used for fusion: when any one of the link risk, the congestion risk, or the topology damage risk reaches the severity level, the network risk is not lower than the level corresponding to that severity level.
[0009] Furthermore, the process of generating candidate combinations of the cross-layer control vectors includes: Based on the basic parameter template corresponding to the risk state, the candidate combination is generated by combining discrete parameter tables. The discrete parameter tables are preset with several discrete levels for each parameter in the cross-layer control vector. When generating the candidate combinations, only one or two level changes are made around the currently executed parameter level.
[0010] Furthermore, the process of determining the retransmission parameters includes: The success rate of a single transmission on the current link is estimated by statistically analyzing the transmission results of each physical transmission attempt. Based on the message type and the target success rate corresponding to the risk status, the base number of attempts is inferred from the probability relationship that at least one attempt will succeed in multiple attempts. The initial retransmission count is obtained by subtracting one from the base number of attempts. The initial retransmission count is then limited by the maximum number of retransmissions allowed by the protocol, the number of attempts that can be accommodated within the remaining deadline, the energy budget for this event, and the current congestion level to obtain the final retransmission parameters.
[0011] Furthermore, the method also includes: distinguishing the cause of transmission failure based on the link quality characteristics and the channel congestion characteristics: when the link quality is stable but the channel busy rate, queue occupancy, duplicate packet ratio, and backoff count increase, it is determined to be channel contention congestion; when the channel busy rate and queue occupancy are low but the link quality continues to decline, it is determined to be link fading. When channel contention and congestion are identified, the following actions are performed in order of priority: restricting low-priority services, merging periodic data, sending data in staggered shifts according to node-related time slots, selecting a parent node or channel with lower congestion, adjusting backoff parameters, and increasing the number of retransmissions to a limited extent if the remaining reliability is still insufficient after the aforementioned actions are performed. When a link fading is detected, adjust the transmit power or switch the transmission path.
[0012] Furthermore, the process of determining the polling parameters and the power consumption parameters includes: Based on the risk status, the amount of data to be sent cached by the parent node for this node, the historical downlink arrival rate, and the battery available energy after deducting the safety reserve, the polling cycle is adjusted step by step in discrete polling cycle levels. The duration of the receiving window is determined, and the duration of the receiving window shall at least cover the clock drift between the current node and the parent node, the parent node response delay, the downlink packet air interface time, and the security processing time. When the risk status increases, the duration of the receiving window is extended. When no data is received in multiple consecutive receiving windows and the risk decreases, the duration of the receiving window is shortened step by step. The duration of the receiving window shall not be less than the base duration that covers the clock drift, the parent node response delay, the air interface time, and the security processing time. The polling period, the receiving window duration, the transmit power, and the retransmission parameters are jointly evaluated, and the combination with the lower expected total energy consumption is selected from the combinations that meet the target reliability and message deadline.
[0013] Furthermore, the power consumption parameter includes the transmission power, and the process of adjusting the transmission power includes: The actual link margin is estimated based on the received signal strength of acknowledgment frames, neighbor reports, or probe frames, and the target link margin is determined based on the risk status. When the difference between the target link margin and the actual link margin exceeds the positive dead zone for multiple consecutive observation windows, the transmit power is increased by one power level; when the actual link margin is higher than the target link margin and exceeds the negative dead zone and remains stable, the transmit power is decreased by one power level; the magnitude of a single adjustment does not exceed the preset change limit. When the transmission failure is determined to be caused by channel congestion, the transmit power is not increased; when the transmit power reaches the upper limit and the actual link margin is still lower than the target link margin, a parent node switch or path switch is performed.
[0014] Furthermore, the process of determining the routing parameters includes: Calculate the routing cost of candidate parent nodes or candidate paths. The routing cost includes at least the expected number of link transmissions, the expected end-to-end delay, the energy consumption per packet, the remaining energy of the parent node, the congestion of the parent node queue, the spatial risk exposure, and the handover cost. The spatial risk exposure is determined based on the regional risk of the area traversed by the path, and the weight of each cost changes with the risk status. When the routing cost of the new path is better than that of the current path for a preset number of consecutive periods and the improvement exceeds the switching hysteresis threshold, a path switch is performed. The higher the risk level, the smaller the preset number and the switching hysteresis threshold, and the path switch is limited by the maximum switching frequency and routing loop constraints.
[0015] Furthermore, the risk states include at least normal state, observation state, early warning state, alarm state, and emergency state, and the entry threshold for each risk state is higher than the corresponding exit threshold; when the risk increases, the higher risk state is entered when the entry threshold is met, and when the risk decreases, it is necessary to remain below the exit threshold for a preset number of consecutive periods and the network indicators must be stable before gradually recovering. When no candidate combination satisfies the constraints, a security degradation process is initiated: low-priority periodic reports are suspended, diagnostic data is merged, alarm summaries are sent with priority, an alternate parent node or redundant path is selected, and controlled reliable flooding is performed when the risk state is an emergency state.
[0016] A fire risk-based MESH network alarm system includes a gateway and multiple alarm nodes, wherein the alarm nodes include: The risk fusion module is used to collect fire sensor features and network status features, and fuse them in layers to obtain node risk, regional risk and network risk, and determine the risk status of the node. The strategy optimization module is used to generate candidate combinations of cross-layer control vectors from a limited set of candidate parameters allowed by the protocol based on the risk state. The candidate combinations are screened under constraints such as target reliability, message deadline, protocol parameter upper limit, battery safety reserve energy, and maximum duty cycle. The target control vector is selected from the screened candidate combinations according to the multi-objective cost that changes with the risk state. The cross-layer control vector includes at least routing parameters, retransmission parameters, polling parameters, and power consumption parameters. The parameter execution module is used to perform protocol stack parameter adjustments according to the target control vector.
[0017] Beneficial effects: This invention incorporates fire sensing characteristics along with network status characteristics such as link quality, channel congestion, topology impairment, and remaining energy into the basis for adjusting protocol stack parameters. Through layered fusion, node risk, regional risk, and network risk are identified, and risk states are determined accordingly. This ensures that the adjustment range of protocol stack parameters such as routing, retransmission, polling, and power consumption matches the actual urgency of the fire and the actual condition of the on-site network. When the actual fire risk increases, alarm information achieves higher transmission reliability and timeliness. During normal monitoring, the equipment maintains low power consumption, avoiding long-term unnecessary energy consumption. In the event of local false alarms or single-point anomalies, communication upgrades are limited to a level commensurate with the actual risk, reducing interference to the entire network.
[0018] Because candidate combinations are screened under constraints such as target reliability, message deadline, protocol parameter upper limit, battery safety reserve energy, and maximum duty cycle before cost comparison, the parameter adjustments always fall within the range allowed by the protocol and acceptable to the node. This ensures the basic reachability of alarm communication while avoiding overdrafting the battery or violating protocol specifications due to overly aggressive parameters.
[0019] Meanwhile, the multi-objective cost incorporates channel congestion increments, and the weights of each cost change with the risk state. This ensures that parameter selection prioritizes reliability and timeliness during fire emergencies, energy efficiency during normal times, and suppresses additional air interface occupation when the channel is congested. This avoids channel contention deterioration caused by indiscriminate synchronous enhanced transmissions from all nodes during alarms, reducing collisions and duplicate messages, and improving the actual arrival performance of alarm information. Furthermore, protocol stack adjustments are made from a limited set of candidate parameters allowed by the protocol, resulting in limited computation. This allows for real-time operation on resource-constrained embedded platforms used by the alarm system, without requiring modifications to the original protocol's message format and security mechanisms, demonstrating good engineering feasibility and protocol compatibility. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a schematic flowchart of the main method steps of the present invention; Figure 2 This is a schematic diagram of the overall control method architecture of the present invention; Figure 3 This is a schematic diagram illustrating the relationship between the protocol stack cross-layer controller and the parameter interface of the present invention; Figure 4 This is a schematic diagram of the layered fusion method of the present invention; Figure 5 This is a schematic diagram of the mapping from risk state to protocol stack control vector and the finite candidate search method of the present invention; Figure 6 This is a schematic diagram of the fault transfer method of the present invention; Figure 7 This is a schematic diagram of the joint power consumption evaluation method of the present invention; Figure 8 This is a schematic diagram of the parent node switching method of the present invention; Figure 9 This is a schematic diagram of the congestion identification and distributed coordination method of the present invention; Figure 10 This is a schematic diagram of the state machine for the risk state of the present invention; Figure 11 This is a schematic diagram illustrating the adaptive adjustment of protocol stack parameters in a multi-story real fire scenario according to the present invention. Detailed Implementation
[0022] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.
[0023] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0024] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise expressly limited. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.
[0025] Example 1: Reference Figures 1-11 As shown, this embodiment provides a control method for a MESH network alarm system based on fire risk. This method is applied to a MESH network including a gateway and multiple alarm nodes. These alarm nodes are, for example, battery-powered smoke detectors, temperature detectors, or carbon monoxide detectors, typically operating as low-power terminals. The network may also include mains-powered or battery-powered routing nodes. Alarm information is transmitted via single-hop or multi-hop transmission to the gateway, other alarm nodes, and linked devices. The method in this embodiment is executed by a protocol stack cross-layer controller within the alarm nodes. This controller modifies the protocol stack's operational behavior through parameter interfaces, policy hooks, queue callbacks, or vendor-extended control modules, while maintaining the original protocol stack's address, security, networking, and standard message formats, thus ensuring compatibility with Zigbee, Thread / RPL, Bluetooth Mesh, and private Sub-G networks. Figure 1 As shown, the system simultaneously collects sensor and alarm characteristics, link and network status, and node energy and role information. A risk fusion engine generates various risks, and a cross-layer policy optimizer produces control decisions that are applied to the protocol stack. Figure 2As shown, the alarm service layer provides the controller with event types, alarm levels, reliability targets, and deadlines, while the network layer, MAC layer, and physical / power management layer are responsible for executing the parameters.
[0026] Specifically, the method includes the following steps.
[0027] Step 1: Collect fire sensing characteristics and network status characteristics of the alarm node. The network status characteristics include link quality characteristics, channel congestion characteristics, topology impairment characteristics, and remaining energy characteristics.
[0028] Among them, fire sensing characteristics include smoke response value, smoke growth rate, temperature, temperature rise rate, carbon monoxide concentration, carbon monoxide growth rate, humidity change, alarm duration, and equipment self-test status; link quality characteristics include Received Signal Strength Indicator (RSSI), Link Quality Indicator (LQI), Expected Transmissions (ETX), and Acknowledgment timeout rate; channel congestion characteristics include channel busy rate, queue occupancy, duplicate packet ratio, and average backoff times; topology impairment characteristics include the number of parent node disconnections, the power failure ratio of surrounding routing nodes, and network partitioning indicators; and remaining energy characteristics include battery voltage and the percentage of available energy after deducting safety reserve energy.
[0029] The input features mentioned above are divided into directly acquired parameters and derived parameters. Smoke response, temperature, carbon monoxide concentration, humidity, RSSI, LQI, battery voltage, queue length, and the busy / idle result of the radio frequency idle channel assessment (CCA) can be directly read from sensors, radio frequency chips, or protocol stack counters. Smoke growth rate, temperature rise rate, carbon monoxide growth rate, ETX, acknowledgment timeout rate, channel busy rate, neighbor power failure ratio, and network partitioning indications are derived parameters, calculated from valid samples within a fixed observation window. Before entering the calculation, the acquired values undergo range checks, invalid code checks, and timestamp continuity checks; missing or out-of-bounds samples are not included in risk accumulation, and the most recent reliable value and its validity period are retained to avoid abnormal samples contaminating the risk calculation results.
[0030] The specific methods for obtaining each derived quantity are as follows: ETX is obtained by dividing the total number of actual physical transmission attempts within the observation window by the number of successfully delivered data packets, with a minimum sample size and an upper limit set; the acknowledgment timeout rate is the number of acknowledgment timeouts divided by the number of transmissions requiring acknowledgment; the channel busy rate is the number of samples or busy durations used for CCA-determined busy status divided by the total number of samples or observation duration; queue occupancy is the number of queue items or bytes used divided by the queue capacity; the number of parent node disconnections is the number of consecutive heartbeats, Poll responses, or missing data acknowledgments; the power failure ratio of surrounding routing nodes is the number of disconnected routing nodes that have been cross-confirmed by neighbors divided by the number of originally valid routing nodes within the observation range; network partition indicators are jointly confirmed by gateway unreachability, simultaneous failure of multiple originally reachable destinations, a sudden decrease in the candidate parent node set, and inconsistencies in the reachable sets given by different neighbors.
[0031] The environmental baseline is updated only when nodes are in normal condition, without self-test faults, without alarm messages, and without neighboring nodes reporting anomalies. The baseline can be the median of the most recent N valid samples, or it can be smoothed exponentially. The baseline is frozen after entering a warning, alarm, or emergency state to prevent the actual fire escalation process from being incorrectly absorbed into the baseline. N can be set according to the sampling period and the rate of environmental change, for example, covering several minutes to several hours, with the specific value determined through actual measurements in normal environments. The rate of change is obtained by subtracting the smoothed value of the previous moment or the previous window from the current smoothed value, and then dividing by the time interval between the two. The time interval is the difference between two valid timestamps. When smoke response, temperature, or carbon monoxide concentration are variables, the units of the calculated results are response time per second, degrees Celsius per minute, or ppm per minute, respectively. To reduce single-point noise, a straight line can be fitted to the most recent 3 to 10 valid points, and the fitted slope can be used as the rate of change. To reduce the impact of single-point spikes, median filtering, truncated averaging, or exponential smoothing can also be used for each feature.
[0032] Different characteristics are mapped to a uniform dimension, such as a range of 0 to 1 or 0 to 100, based on factory calibration values, environmental baselines, or product safety thresholds. For characteristics with higher numerical values indicating higher risk, normalization is performed using the following formula: x norm = min(1, max(0, (x L) / (H L))); In the formula, x is the current effective feature value, L is the normal upper bound or the point where risk begins to rise, H is the upper bound that needs to be considered high risk, and min and max are used to limit the normalization result to between 0 and 1. norm These are the normalized feature values. L and H are primarily derived from product safety thresholds, factory calibrations, certification tests, or statistical analysis of normal, false alarm, and fire samples, and are not arbitrarily set based on single field data, to ensure the stability and repeatability of the normalization benchmark. For features with lower values indicating higher risk, such as remaining energy or link margin, a reverse mapping is used. For resource-constrained microcontrollers, the above normalization and weight mapping can be implemented as integer lookup tables, fixed-point arithmetic, piecewise linear mapping, or finite candidate search.
[0033] Step 2: Perform hierarchical fusion of the fire sensing features and the network status features to obtain node risk, regional risk and network risk, and determine the risk status of the alarm node accordingly.
[0034] Node risk R node At least by sensing risk R s Trend risk R t Cross-sensor consistency risk R c Duration risk R dura And the false alarm suppression term Ffalse The determination can be achieved through weighted summation, segmentation rules, fuzzy reasoning, logistic regression, decision trees, or small neural networks.
[0035] In this embodiment, the node risk R node Represented as: R node = clamp(w s ·R s + w t ·R t + w c ·R c + w d ·R dura w f ·F false ); In the formula, R s R t R c R dura and F false These are sensor risk, trend risk, cross-sensor consistency risk, duration risk, and false alarm suppression term, w s w t w c w d and w f These represent the weights for each component, with `clamp` indicating that the final risk is limited to a specified range, such as 0 to 1 or 0 to 100. Each weight can be determined using engineering calibration or training data: During rule implementation, the maximum risk increment each component should contribute when transitioning from normal to hazardous is first determined based on single-factor experiments, and then the positive weights are normalized to sum to 1. A separate upper limit is set for the false alarm suppression weight to ensure that the false alarm suppression item cannot reduce a confirmed fire to a normal state. When using logistic regression or decision trees, the weights or splitting thresholds are obtained through training with labeled normal, false alarm, smoldering, and open flame samples and are fixed before firmware release. Weights can also be configured based on equipment type and installation scenario.
[0036] Among them, sensing risk R s The risk is formed by normalization of smoke, temperature, carbon monoxide, and other direct hazards; single-sensor products can take the risk of the primary sensed quantity, while multi-sensor products can be weighted according to sensor reliability, and R is limited when only one channel is abnormal and the other channels are stable. s The magnitude of the single-cycle increase. Trend risk R t R is formed by the smoke growth rate, temperature rise rate, carbon monoxide growth rate, and the number of consecutive unidirectional trends. A sustained trend with an increasing slope improves R. t Reduce R when the trend reverses or only a single peak occurs. t Cross-sensor consistency risk R cThe risk level (R) is determined by whether different sensors within the same node change in the same direction towards danger within the permissible time window. In implementation, one anomaly, two consistent indicators, and three or more consistent indicators can be mapped to low, medium, and high consistency, respectively. Known physical relationships are then verified; for example, water vapor is usually accompanied by a rapid increase in humidity, while actual combustion is more likely to simultaneously cause a continuous rise in smoke and changes in temperature or carbon monoxide. dura The result is obtained by comparing the cumulative time when the risk exceeds the entry threshold with the confirmation time specified by the product. Short-term threshold exceedances only produce lower values, while continuous threshold exceedances gradually increase.
[0037] False alarm suppression term F false Formed from known non-fire modes and equipment malfunction modes, including sudden humidity rise accompanied by short pulses of smoke, cooking particles appearing but temperature and carbon monoxide levels remaining inconsistent over a long period, sensor self-test failure, labyrinth contamination causing long-term offset, single-point burrs, and maintenance test modes; F false It is only used to reduce the magnitude of communication upgrades and does not block audible and visual alarms that already meet the conditions for mandatory local alarms, thereby suppressing communication side escalation while ensuring that local alarms related to personal safety are not affected.
[0038] Regional Risk R zone To express the spatial consistency of multiple nodes in the same room, adjacent rooms, on the same floor, or across floors, a neighbor risk fusion method with time and spatial weights is used: nodes located in the same room, the same fire compartment, or directly adjacent areas have higher weights, while nodes across floors or at greater distances have lower weights; only neighbors whose time falls within the consistency window and whose data freshness meets the requirements are included in the calculation. To avoid a large number of low-risk nodes diluting the high risk near the fire source, the maximum neighbor risk, the number of nodes reaching the warning level or above, and the weighted average are retained, and R is formed based on these three factors. zone If multiple adjacent devices change in the same direction within a similar timeframe, then R is increased. zone If only a single device briefly exceeds the threshold while surrounding devices remain normal, then the upgrade of communication parameters is limited to that node and its parent node to prevent the entire network from entering a high-power state, ensuring that the overhead of communication resources is commensurate with the spatial range of risk.
[0039] Network Risk R net Link risk R l Congestion risk R g and topological damage risk R damage The results are fused and used to determine whether to change routing weights, add redundant paths, or initiate emergency communication. Link risk R l The congestion risk is determined by ETX, confirmation timeout rate, RSSI / LQI degradation, parent node disconnection, and the degree of reduction in candidate parent nodes; R gThe risk of topology impairment is formed by channel busy rate, queue occupancy, backoff count, duplicate packet ratio, and neighbor congestion summary; R damage It is formed by the proportion of disconnected routing nodes, failure of critical cut points, unreachable gateways, decrease in the number of backup paths, and evidence of partitioning. net A segmented rule can be prioritized: if any item reaches a severe level, it should not be offset by other low-risk items on average. In other cases, weighted fusion should be performed to ensure that severe network damage is not masked by seemingly normal average indicators.
[0040] Based on node risk, regional risk, and network risk, nodes are classified into five risk states: Normal (S0), Observation (S1), Early Warning (S2), Alarm (S3), and Emergency (S4). State thresholds are determined through product testing and employ a hysteresis mechanism where the entry threshold is higher than the exit threshold: when risk increases, meeting the entry threshold allows for a relatively quick transition to a higher-level state; when risk decreases, K consecutive periods below the fallback threshold and stable network indicators are required for gradual recovery. An implementable example is: below the first entry threshold is Normal, progressing sequentially to Observation, Early Warning, Alarm, and Emergency; exit uses a threshold 5% to 15% lower than the corresponding entry threshold, requiring K consecutive periods to meet it; K is calculated from the sampling period and allowable recovery delay, with higher-risk states using longer exit confirmation times. This hysteresis structure prevents fluctuations in sensor values near the threshold from causing frequent switching of polling cycles, power, and routing parameters, avoiding protocol stack oscillations and additional battery consumption.
[0041] Step 3: Based on the risk status, generate candidate combinations of cross-layer control vectors from the limited set of candidate parameters allowed by the protocol. The cross-layer control vectors include at least routing parameters, retransmission parameters, polling parameters, and power consumption parameters.
[0042] The cross-layer control vector P includes at least: the Poll period T poll Receive window T rx Maximum number of retransmissions N retry Confirmation of timeout T ack CCA / retreat parameter B mac Transmit power P tx Heart rate cycle T heart Routing repair threshold H repair Parent node switching lag H switch Time to Live (TTL) and Number of Redundant Paths (K) path Business queue weight W q Merging window T with periodic data agg Parameters that are not supported by different protocols can be removed from the control vector.
[0043] Considering the limited microcontroller resources of the alarm device, this embodiment does not require real-time solving of complex continuous optimization problems. Instead, it pre-establishes several basic parameter templates for each risk state, and then generates a finite candidate set by combining discrete parameter tables based on the current link, congestion, and battery status. The discrete parameter tables pre-set several discrete levels for each parameter in the cross-layer control vector. When generating candidate combinations, only one or two levels are changed around the currently executed parameter level, without traversing all continuous values, thus limiting the computational load to no more than ten candidates. For example, in one embodiment, T... poll Take 1 second, 5 seconds, 15 seconds, and 60 seconds, T rx Take 40 milliseconds, 80 milliseconds, and 160 milliseconds, N retry Take 0 to 3 times, P tx Select several power levels allowed by the chip, K path Choose 1 or 2; the above values are only for illustrating the candidate generation method, and will be ultimately tailored to specific protocols, RF regulations and chip capabilities.
[0044] For example, under alarm conditions, low-latency schemes, high-reliability schemes, low-energy backup schemes, and congestion suppression schemes can be generated: In the low-latency scheme, T... poll Take the shorter gear, T rx Select the mid-to-high-end values that can cover the parent node's response, N retry Based on the allowable average value according to the deadline, the alarm queue is set to the highest priority, the normal data merging window is extended, and P... tx Only increase K when link margin is insufficient. path Typically, 1 is maintained to reduce additional duty cycle; in high-reliability solutions, N is calculated by working backward from the target success rate. retry Activation confirmation, T rx Prioritize high-end components, increase the target link margin, and, if necessary, K path Choosing option 2 and using the primary and backup paths lowers the route repair threshold, but if the channel is already congested, it restricts repeated transmissions and prioritizes the use of two paths with low spatial or channel correlation; in the low-energy backup scheme, T poll Maintain mid-to-long range, T rx Select the lowest safe level that satisfies clock drift and response latency, P tx Maintain the minimum target margin, N retry Limiting to a lower value, merging periodic diagnostic data, and extending the normal heartbeat, this scheme is used for nodes where the power is close to the safe reserve value but still needs to maintain basic alarm accessibility; The congestion suppression scheme does not increase or appropriately reduces retry times for ordinary packets, extends the backoff period for ordinary services and increases node-related random jitter, suspends low-priority reporting, maintains controlled retry times for alarm packets, and prioritizes parent nodes with lower channel busy rates and queue occupancy. txThe vector remains unchanged or is adjusted only when it is confirmed to be a link fading issue. After candidate vectors are generated, checks are performed on protocol validity, loops, maximum duty cycle, maximum power, energy budget, and deadline before proceeding to cost comparison.
[0045] Step 4: Using target reliability, message deadline, protocol parameter upper limit, battery safety reserve energy, and maximum duty cycle as constraints, filter out candidate combinations that do not meet the constraints.
[0046] Reliability, deadline, protocol limits, battery safety reserve, and maximum duty cycle are hard constraints. Unqualified candidates are screened out before cost calculation, and these cannot be offset by other lower costs, thus ensuring that any executed parameter combination first meets the bottom-line requirements of alarm communication. Specifically, the energy budget for this event is determined by the available remaining battery capacity, operating voltage, DC / DC efficiency, and the safety reserve energy. The safety reserve energy must at least cover the specified duration of local audible and visual alarms, continuous sensor sampling, a small number of emergency reports, and critical information storage. The expected energy of candidate schemes includes the energy consumption for each wake-up, receive window, transmission, acknowledgment waiting, retries, encryption, and routing control messages. Only schemes whose expected event energy consumption does not exceed the available remaining energy minus the safety reserve energy can be executed.
[0047] Step 5: Calculate the multi-objective cost for each of the selected candidate combinations and select the candidate combination with the lowest cost as the target control vector. The weights of each item in the multi-objective cost change with the risk state.
[0048] For each candidate parameter vector, calculate the comprehensive cost J, which includes at least the expected end-to-end delay, the expected packet loss probability, the single-cycle communication energy consumption, the channel congestion increment, and the routing oscillation cost. Each cost term is first mapped to 0 to 1.
[0049] In this embodiment, the overall cost J is expressed as: J = w D ·D cost + w L ·L cost + w E ·E cost + w C ·C cost + w S ·S cost ; In the formula, D cost L cost E cost C cost and S cost These are the normalized latency cost, packet loss cost, energy cost, congestion cost, and stability cost, w. D w L wE w C and w S Let w be the non-negative weight corresponding to the current risk state, and the sum of the weights is 1. The normal state increases w. E and w S Alarm status increased w D and w L Congestion increases w C When the parent node is unstable, the routing stability weight is increased; the weight table is fixed by simulation or real network testing, and the microcontroller only looks up the table when running.
[0050] The expected end-to-end latency is calculated by adding the waiting time for each hop queue, MAC backoff, transmission duration, acknowledgment waiting time, expected retries, and worst-case polling waiting time for low-power nodes, and then comparing it with the message deadline. The expected packet loss cost is calculated by estimating the final delivery probability based on the single-hop success rate, number of retries, and backup path relationships, and then taking the insufficient portion of the estimate. The single-cycle communication energy consumption is estimated by the current and duration of wake-up, receive window, transmission, acknowledgment waiting, retries, and encryption processing. The congestion increment is estimated by the candidate new air interface duration, the current channel busy rate, and the probability of simultaneous transmission by neighbors. The routing oscillation cost is determined by the number of parent node switches per unit time, the difference between the candidate path and the current path, and the overhead of routing control messages. The system selects the candidate with the lowest cost that meets the alarm reliability and deadline constraints. If the J values of multiple candidates are close, the candidate with a larger reliability margin, less change in relative current parameters, no triggering of routing switches, and lower expected energy consumption is given priority, so that parameter adjustments are as smooth and energy-efficient as possible while meeting alarm requirements.
[0051] Step 6: Perform protocol stack parameter adjustments according to the target control vector.
[0052] After execution, in a preferred embodiment, the actual delay, confirmation result, energy consumption estimation error and congestion change are also recorded for the next observation window to correct the prediction; if the prediction error continuously exceeds the set range, it will fall back to the verified conservative parameter template and prohibit further expansion of parameter changes to ensure the robustness of the system in long-term operation.
[0053] The determination process for each parameter is explained in detail below.
[0054] Regarding the determination of retransmission parameters, the node estimates the single transmission success rate p based on the acknowledgment results of the most recent unicast messages, the number of retransmissions, RSSI / LQI, channel busy rate, and parent node load. linkThe single transmission success rate is calculated based on each physical transmission attempt, not just the success of the final data packet. In the most recent W valid attempts, S represents the number of successful attempts with valid acknowledgments or verifiable forwarding results, F represents the number of timeouts or explicit failures, and W can be set according to the number of packets or a time window, with the weight of older samples gradually decreasing. To avoid drastic fluctuations in estimation when the sample size is small, Bayesian smoothing, exponential smoothing, or upper and lower bound constraints can be used. In this embodiment, small-sample smoothing is expressed as: p link = (S+1) / (S+F+2); In the formula, S represents the number of successes and F represents the number of failures. Adding 1 and 2 in the formula provides a neutral prior when there are no samples, avoiding extreme estimates such as 100% for the first success or 0% for the first failure. When RSSI, LQI, channel busy rate, parent node queue, or transmit power change significantly, the old link estimate is not used indefinitely. Instead, the weight of the old samples is reduced after changes in path, channel, or power level, and the counting results are merged with the chip's measured lookup table values. When RSSI / LQI remains consistently good and the channel busy rate is low, p is allowed to be slowly increased. link When the channel busy rate is high, and acknowledgment timeouts and backoffs increase simultaneously, reduce p. link However, it is marked as a congestion cause; when RSSI / LQI is low and the channel busy rate is not high, it is marked as a link fading cause.
[0055] For broadcast or unacknowledged messages, the probability of effective delivery is estimated by the ratio of the number of neighbors within the observation range that have declared forwarding, bitmap acknowledgment, or have been repeatedly detected and forwarded by the source node to the number of expected reachable neighbors. The set of expected reachable neighbors uses the neighbor table that is still valid before sending. Nodes that are disconnected or whose dormant state is unknown are not directly counted as failures. If there are too few verifiable neighbors, a conservative lower limit is adopted, and it is prohibited to determine that the entire network has been delivered based on a single duplicate packet.
[0056] The alarm processing layer provides the target success rate P based on the message type and risk status. target and deadlines D, P target Solidify during the product requirements phase. For example, a target of 90% to 95% can be used for normal status reporting, 95% to 98% for early warning summaries, and 99% or higher for confirmed fire alarms. Local audible and visual alarms, regulatory requirements, and network protocol capabilities take precedence over this example. Under the same risk, alarm confirmation, alarm disarming, and critical linkage commands use a higher P than the normal diagnostic data. targetThe deadline D is the maximum allowed time from when the event is acknowledged by the business layer to when the destination node completes reception. The controller uses the remaining deadline, which is the total deadline minus the sensor acknowledgment, business queuing, and already consumed transmission time. If the message traverses multiple hops, a budget is allocated to each hop based on the estimated number of hops along the current path, and a safety margin is reserved for the final failover. The sources of D include product alarm linkage requirements, certification test requirements, and real-world latency verification results. If we approximate that multiple attempts are independent of each other, then the probability of at least one success out of N attempts is: P(N) = 1 (1 p link )N; In the formula, N represents the number of attempts, including the first attempt, and p link Let P(N) be the success rate of a single transmission, and P(N) be the probability of at least one success in N attempts. Conversely, we can deduce that P satisfies... target When determining the base number of attempts, logarithmic operations are unnecessary. Instead, start with N=1 and increment the number of attempts, substituting the values into the above formula until P is reached. target Protocol parameter N retry Subtract 1 from N. The final N retry It is also subject to limitations imposed by the protocol's maximum allowed value, the number of attempts that can be accommodated within the remaining deadline, the event energy budget, and the congestion level. If the loss is a series of bursts and does not satisfy the independent approximation, the length of the most recent burst is used to approximate p. link Instead of endlessly adding retries to the same path, implement conservative reductions or add an independent backup path when resources allow.
[0057] When applied, the controller first calculates the base N. retry Then determine the cause of failure: if the link is fading and the channel is idle, increase the number of retries or increase the target link margin within the upper limit; if the channel is congested, limit N. retry Prioritize peak-shifting, compressing regular services, or switching paths; if the remaining deadline is insufficient, directly access the backup parent node, redundant path, or controlled flooding; if P cannot be satisfied... target When performing D, instead of returning a seemingly precise but actually unworkable retry value, the system records the unmet reliability target and performs a safety degradation. This prevents good links from wasting energy by performing excessive retries, while poor links achieve sufficient reliability within acceptable limits.
[0058] Regarding congestion and energy constraints, when acknowledgment timeouts are caused by channel congestion, blindly increasing retries can lead to a positive feedback avalanche. Therefore, a congestion factor C is formed based on channel busy rate, queue occupancy, duplicate packet ratio, and neighbor congestion summary. The congestion factor C uses a hierarchical rather than a complex continuous formula: for each observation window, channel busy rate, queue occupancy, acknowledgment timeout rate, duplicate packet ratio, average CCA backoff count, and neighbor congestion level are calculated separately, and each is mapped to four levels: low, medium, high, and severe. Channel busy rate and queue occupancy are primary indicators. When any primary indicator reaches the severe level, or two or more indicators reach the high level and persist for two windows, C enters a high congestion state. Only when a single indicator increases briefly is observation maintained to avoid instantaneous interference triggering network-wide speed reduction. When C is high, priority is given to suspending ordinary services, increasing the priority of alarm queues, staggering transmission times, or switching paths, rather than directly increasing N for all nodes. retry When C reaches a medium to high level, actions are executed according to priority: first, low-priority periodic data is paused or merged; then, the alarm queue weight is increased and node hash jitter is added; subsequently, a relatively idle parent node or backup channel is selected; and finally, N is increased to a limited extent when the remaining reliability is insufficient and there is still deadline time, energy, and air interface margin. retry After C is reduced, multiple windows need to be confirmed before normal services are restored step by step.
[0059] The energy to be sent and the energy to be received must be less than the energy budget available to the node for this event; otherwise, a lower-energy path will be selected or a nearby routing node will be requested to handle the forwarding.
[0060] Regarding failover after a retry failure, when N is reached... retry If unsuccessful, different actions are taken based on the risk and remaining deadline: low-risk messages can be retried; warning messages can switch parent nodes; alarm messages can immediately initiate local route repair or alternative paths; emergency messages can initiate controlled reliable flooding; the result of each action is used to update p. link ETX and routing costs.
[0061] Regarding the determination of polling parameters, for low-power terminals, the poll period determines the worst-case wait time for downlink commands and cascading alarms, T. poll From the maximum period T max Minimum period T min The risk (R), parent node cache usage (B), historical downlink arrival rate (λ), and battery status (E) collectively determine the T value: the higher the risk or the more data the parent node has to send, the higher the T value. poll The shorter the time; after several consecutive Polls have failed to acquire data and the risk has decreased, T poll The increments are gradual. When the device battery approaches the safe retention threshold, the controller limits the minimum number of poll cycles and can request the parent node to reduce invalid reception through protocol-permitted fast polling hints, indirect transmission flags, or application-layer wake-up strategies.
[0062] Among them, T min The minimum stable sleep / wake interval of the chip, the allowed range of the protocol, the maximum receive duty cycle, and the peak battery current are all determined together; T max The following parameters are used: The worst-case downlink latency allowed under normal conditions, parent node child device timeout, and product heartbeat requirements. Parent node buffer usage (B) is the number of message items or bytes to be sent to the child node divided by the corresponding buffer capacity. Historical downlink arrival rate (λ) is the number of valid downlink messages arriving at the child node within the most recent observation period divided by the observation duration, using a slow smoothing method. Battery status (E) uses the proportion of available energy after deducting safety reserves, rather than just instantaneous voltage. The controller maintains discrete Poll levels, such as 0.5 seconds, 1 second, 2 seconds, 5 seconds, 15 seconds, 60 seconds, and 300 seconds, and adjusts them incrementally from the current level: When entering an alarm or emergency state, when the parent node explicitly has pending data, when B exceeds the high water level, or when the probability of the next downlink arrival is expected to be high, the level is lowered by one or two levels; for consecutive N... empty If no data is available for the next Poll, the risk has decreased, and B is at a low level, the level will be raised by one level; N empty High and low water levels are determined through caching experiments at different network scales. High water levels can be verified starting from 70% to 80% of the capacity.
[0063] For messages with a clearly defined downlink deadline, T poll The value should not exceed the remaining downlink cutoff time minus the parent node's response and reception processing time; if the current minimum available level is still insufficient, the node enters a temporary fast polling or continuous reception mode, but its duration is limited by the maximum continuous reception time and energy budget; when the battery level is close to the safety reserve, T cannot be shortened indefinitely. poll The duration of a single T cycle can be extended first. rx Use the parent node's pending notification or request nearby normally powered nodes to take over forwarding. Record each adjustment as follows: empty Poll, successful data retrieval, parent node no response, receipt of pending notification, and actual downlink latency; if T is shortened... poll If downlink latency does not improve while channel busy rate or empty poll ratio increases, then backtrack by one level; if the risk continues to rise and there is real data to be sent, then maintain short poll until the message is completed or fast polling times out.
[0064] Regarding the determination of the receiving window, the receiving window T rx It should at least cover clock drift, parent node response delay, message length, and security processing time; the window should be appropriately extended in alarm states to improve the probability of downlink command and acknowledgment reception, and then gradually shortened after consecutive empty windows. This method is more precise than simply shortening the Poll cycle and can avoid missing downlink data due to an excessively short window after frequent wake-ups. The basic receive window is determined by the following formula: T rx_base = 2·T drift + Tresp + T air + T sec + Tguard; In the formula, T drift This represents the maximum unidirectional offset that the clock error between this node and its parent node may cause within this sleep duration. Offsets before and after this offset need to be covered, therefore multiplied by 2; T resp T is the high quantile delay from when the parent node receives the Poll to when it begins to respond. air T is the longest allowed air interface time for downlink frames at the current physical rate; sec Reserved for decryption, verification, and protocol processing; T guard This is for interrupt, scheduling, and measurement error protection time; all time quantities use the same unit and are rounded up to the hardware timer step size. Where T... drift Estimated by multiplying the sum of the local crystal oscillator error upper limit and the parent node clock error upper limit by the sleep duration; T resp Use the 90th to 99th percentile of the most recent effective response delay and set a protocol upper limit; T air The value is obtained by dividing the total number of bits in the preamble, header, payload, and parity field by the physical rate; T sec Determined by measuring the worst-case processing time of the target microcontroller; re-estimated after changes in path, rate, security suite, or crystal mode.
[0065] In alarm mode, if the parent node confirms that the message has been sent but the terminal has not received it, the message is received only at the end of the receiving window, or the clock synchronization quality deteriorates, T will be activated. rx Increase by one discrete level or in T rx_base Increase the protection level by 10% to 30% on top of the existing level; when multiple consecutive windows are empty and the parent node has no pending data, the risk decreases, and clock synchronization is stable, shorten the protection level step by step, but not less than T. rx_base If the window reaches its limit and still fails to collect data, check the parent node, clock, or path instead of continuing to extend indefinitely.
[0066] Regarding transmit power adjustment, the device estimates the actual link margin M based on acknowledgment frames, neighbor reports, or historical RSSI. actual The risk status determines the target link margin M. target The controller adjusts P in small steps according to the difference between the two. tx It also sets minimum and maximum power levels and upper limits for single-cycle changes; increases M for high-risk situations. target Normal state reduces M target The actual link margin is calculated using the following formula: M actual = RSSI filtered – RSSI required; In the formula, RSSI filteredRSSI is the median or truncated average of the most recent valid acknowledgment frames, neighbor reports, or bidirectional probe frames. required The minimum reliable receive level corresponds to the current modulation scheme, data rate, and target packet error rate. This level can be determined from the initial value in the chip datasheet, combined with adjustments based on the overall antenna, casing, and certification tests. Both values are in dBm. Subtracting the latter yields the link margin in dB. The higher the receive level is from the minimum reliable level, the greater the margin. actual The larger the margin, the better. When link asymmetry exists, the smaller value of the uplink and downlink margins should be used first; when the parent node's measurement value cannot be obtained, the RSSI of the parent node's received messages should be used and a conservative correction should be added; when the RSSI sample size is less than the minimum, or when the channel has just been switched or the power has just changed, low confidence should be marked and discontinuous adjustment should be temporarily suspended. target Determined by risk status, message level, and link fluctuation: Under normal conditions, a basic margin is adopted to ensure stable communication; under alarm conditions, a protection margin against obstruction and personnel movement is added. The specific dB value is solidified through anechoic chamber, whole machine, and field tests.
[0067] Setting dead zone and continuous confirmation during adjustment: If M target Reduce M actual If two consecutive observation windows are larger than the positive dead zone, for example, 2 dB, then P tx Increase the chip power level by one level; if M actual Higher than M target And if it exceeds the negative dead zone and remains stable for more consecutive windows, then P tx Downgrade by one level; single changes are typically limited to one level or 1 to 3 dB, and are subject to maximum transmit power, regulations, peak current, and duty cycle limitations. Even when the channel busy rate is high, M... actual If there are shortcomings, first determine whether the problem is caused by a collision rather than a weak signal; under congested conditions, all nodes are not allowed to increase power simultaneously; to achieve P tx Still below M after the upper limit target If the success rate does not improve after adjustment, switch to parent node switching, alternative path or rate, or channel adjustment, and feed the result back to p. link And routing costs. This closed-loop method avoids all nodes using maximum power when an alarm is triggered, reducing mutual interference and energy consumption.
[0068] Regarding joint energy consumption assessment, shortening the Poll cycle mainly increases the energy consumption for receiving and waking up, thus improving P. tx The main increase is in transmission energy consumption, and the increase in retries also increases the energy consumption of both transmission and reception waiting. Therefore, the policy optimizer adjusts the T... poll T rx P tx and N retry The joint evaluation employs four steps: discrete candidate generation, hard constraint screening, energy consumption comparison, and execution feedback. First, it focuses on the current T...poll T rx P tx and N retry A finite number of combinations are generated, including hold, increase by one level, and decrease by one level. Combinations violating protocol limits, reliability targets, remaining cutoff time, maximum duty cycle, peak current, and safety reserve energy are then removed. To prevent oscillation, only two parameters are typically allowed to change levels within a single control cycle. The receive power consumption of each candidate is calculated based on the number of polls per unit time, the power consumption per wake-up, and T. rx Internal receive current estimation; transmit power consumption based on P tx The corresponding transmission current, frame air interface time, and estimated number of attempts are estimated; confirmation waiting, encryption processing, routing control, and backup path transmission are included separately; for downlink services, the worst-case Poll waiting time is used instead of the average for the deadline assessment; for uplink alarms, queuing, backoff, per-hop transmission, and failover time are mainly included.
[0069] Candidates must simultaneously meet the condition that the expected delivery probability is not less than P. target The worst-case latency is expected to not exceed the remaining D, and the expected event energy consumption is expected to not exceed the budget. If these conditions are met, the option with the lowest expected total energy consumption is selected. When energy consumption is similar, priority is given to options with larger reliability margins, fewer parameter changes, and lower congestion increments. After executing one or more observation windows, the estimate is corrected using actual success rate, latency, empty poll ratio, and energy consumption count. If the effect deteriorates, it is rolled back to the previous stable vector. For example, if the link margin is sufficient but the downlink wait time is too long, T is shortened first. poll Without increasing P tx When the link margin is insufficient but the parent node has no data to send, prioritize increasing P in small steps. tx Or change routes without shortening T poll When congestion is high, limit N. retry and P tx The parameters are adjusted upwards, and off-peak and low-congestion paths are used instead. This causal constraint ensures that the four parameters do not simply increase simultaneously, but are adjusted in a coordinated manner according to their actual effects on latency, reliability, and energy consumption.
[0070] Regarding the determination of routing selection parameters, the cost of a candidate parent node or path includes at least ETX, estimated end-to-end latency, single packet energy consumption, remaining energy of the parent node, parent node queue congestion, spatial risk exposure, path redundancy, and handover cost. Higher alarm levels result in greater weights for ETX and latency, while under normal conditions, energy and handover stability weights are increased. Spatial risk exposure refers to whether the path passes through areas prone to fire, power outages, or multiple node anomalies. Specifically, the ETX of a candidate parent node is estimated from the number of transmission attempts and successful deliveries within the most recent window, and the path ETX is the sum of the ETX of each hop. Estimated end-to-end latency is the sum of queue waiting, backoff, air interface transmission, acknowledgment, and estimated retry times for each hop. Single packet energy consumption includes the energy consumption of the node itself and an estimate of the forwarding energy consumption of the parent node when available. Remaining energy of the parent node is the available percentage after deducting security reserves; a lower percentage results in a higher cost. Parent node queue congestion is determined by queue occupancy, high-priority queue length, and drop rate.
[0071] Spatial risk exposure is calculated by mapping each relay node on the path to a room, fire compartment, or floor risk area. The cost increases when the path passes through areas with high risk, a high proportion of routing nodes experiencing power outages, or areas expected to be blocked by fire. The maximum regional risk of the path, the number of nodes passing through high-risk areas, and the expected number of hops spent in those areas can be used simultaneously to avoid simply averaging and diluting localized high risks. In alarm states, high-risk relay nodes exceeding safety thresholds can be directly excluded. Path redundancy is determined by the number of available non-intersecting or non-intersecting backup paths, the proportion of shared failure areas for primary and backup paths, and the quality of backup paths. Higher redundancy results in lower reliability costs, but additional energy consumption is still separately accounted for. Switching costs are formed by routing update messages, neighbor table and security state migrations, cached message redirection, the number of recent switches, and potential short-term packet loss.
[0072] Each item is mapped to a range of 0 to 1 using normal and unacceptable values, and then weighted according to risk status using a lookup table. Before entering the candidate list, parent nodes that may form loops, have invalid rank or hop counts, have less remaining energy than the safe retention level, are in a full queue, or have excessive spatial risk are excluded. A new parent node is not switched immediately simply because its instantaneous score is better: under normal and warning conditions, the new path must be better than the current path for K consecutive cycles, and the improvement must exceed the switching hysteresis threshold H. switch Only then does the switching occur; in alarm and emergency states, K can be shortened and the difference threshold lowered, but the maximum switching frequency, routing loop checks, and Rank and hop count constraints are still retained. This allows for rapid bypassing of failed nodes during disasters while avoiding frequent oscillations between multiple similar paths. The routing repair triggering conditions also change with the risk: in normal states, observation is allowed after short-term packet loss to save control overhead; in alarm states, a small number of consecutive failures or loss of parent node heartbeat can initiate partial repair; in emergency states, primary and backup paths or a limited number of redundant parent nodes can be maintained simultaneously.
[0073] Regarding congestion assessment and distributed coordination, nodes measure channel busy rate (CBR), queue occupancy (Q), acknowledgment timeout rate (A), duplicate packet proportion (Dup), average backoff count, and neighbor congestion digest to form congestion levels from C0 to C4. Specifically, CBR is the ratio of the radio frequency busy time or CCA busy count within the observation window to the total observation time or total CCA count; Q is the ratio of the current queue occupancy bytes or queue items to the current queue capacity; A is the ratio of acknowledgment timeouts in unicast attempts requiring acknowledgment; Dup is the proportion of duplicate packets identified in received packets based on source address, sequence number, and time window; average backoff count is the total number of CCA backoffs performed before successful or failed transmission divided by the number of transmission attempts; and the neighbor congestion digest includes at least the neighbor's own level, queue high-water mark, and digest expiration date, excluding expired digests. C0 to C4 are determined using segmentation rules: C0 indicates that all indicators are within the normal range; C1 indicates that a single indicator is slightly elevated but there is no packet dropping; C2 indicates that CBR or Q reaches a moderate level, or two or more auxiliary indicators continue to rise; C3 indicates that CBR, Q, or A reaches a high level and delay exceeds the standard, and duplicate packets increase significantly; C4 indicates queue overflow, consecutive deadline failures, multiple neighbors simultaneously reporting severe congestion, or available channels approaching saturation; the thresholds for each level are determined by the measured distribution under different network sizes and interference intensities, and a hysteresis of high entry and low exit is adopted.
[0074] The fault cause is determined using combined evidence: low RSSI / LQI, decreased neighbor reachability, and low CBR and backoff indicate link fading; stable RSSI / LQI, still reachable neighbors, and increased CBR, Q, Dup, and backoff indicate channel contention; high Q of a specific parent node but low local CBR indicate local congestion of the parent node; simultaneous loss of connection of a gateway or multiple routing nodes accompanied by a decrease in the number of paths indicates topology damage; conflicting evidence is treated as a mixed fault, and the problem of a single action continuously amplifying the fault is avoided.
[0075] The specific adjustment methods are as follows: C0 maintains normal parameters; C1 adds random jitter to ordinary services and merges adjacent period data; C2 extends the ordinary heartbeat and reporting period, restricts ordinary message retry, and increases the alarm queue weight; C3 suspends unnecessary services, staggers peak times according to node hash time slots, prioritizes switching to low CBR, low Q parent nodes or backup channels, and limits the number of alarm repetitions; C4 retains only alarm summaries, acknowledgments, and key linkages, and initiates partial repair, limited redundant paths, or controlled reliable flooding. If an alarm packet is very close to the deadline, a short-term emergency preemption can be allowed: increase its queue priority, shorten the initial backoff, or select a redundant path, but limit the preemption duration, number of neighbors, and maximum number of repetitions to prevent a single event from occupying the channel for a long time. After congestion is reduced, ordinary services are not immediately fully restored, but require continuous K... rIf an observation window falls below the current level's exit threshold, it will decrease by only one level at a time, and will be restored in batches according to queue priority and node hash order to avoid a second congestion peak caused by all nodes in the network simultaneously restoring heartbeats and reporting. If the CBR decreases after a certain adjustment action but the alarm arrival rate deteriorates, the action will be canceled or the alarm queue resources will be increased; if the CBR does not decrease and duplicate packets continue to increase, further retries will be prohibited, and the process will be transferred to path or channel processing.
[0076] Regarding state machines and fail-safety, each risk state corresponds to a set of parameter ranges rather than a single fixed value. Entering a high-risk state depends on continuously escalating risk, multi-sensor consistency, alarm confirmation, or network impairment; exiting requires a decrease in risk, stable sensor data, and recovery of network indicators. All states are subject to hard boundary constraints, including battery safety reserve energy, maximum RF duty cycle, maximum retransmission count, maximum transmit power, maximum routing switching frequency, and maximum continuous reception time. If a boundary is about to be breached, the system prioritizes retaining alarm summaries, alarm confirmations, and critical linkage commands, while suspending routine diagnostics and periodic reports. When power, link, or congestion conditions prevent all candidates from simultaneously meeting the objective, the controller does not increase parameters indefinitely but instead enters a safety degradation phase: suspending low-priority periodic reports, merging diagnostic data, prioritizing alarm summaries, selecting backup parent nodes or redundant paths, and triggering controlled reliable flooding when necessary to ensure that the most important alarm communications are prioritized even when resources are insufficient.
[0077] The following describes the actual operation of the method in this embodiment using typical application scenarios. In the kitchen smoke false alarm observation scenario, a short-term increase in smoke is detected by the kitchen smoke detector, but the temperature, carbon monoxide, and smoke detectors in adjacent corridors do not change synchronously. The node risk enters the observation state. The device only shortens the Poll cycle from 60 seconds to 15 seconds, the heartbeat from 30 minutes to 5 minutes, and sends one warning summary. The number of retries and transmission power remain moderate, without triggering a network-wide routing upgrade. If the smoke subsequently decreases, the device returns to normal parameters after meeting the fallback conditions for several consecutive cycles. In a multi-story real fire scenario, smoke and temperature detectors on the first floor rise rapidly at the same time. Adjacent nodes show consistent changes in a short period of time, and the regional risk increases rapidly. Nodes near the fire source enter alarm or emergency states. The Poll cycle is shortened, the target link margin is increased, a higher number of retries is calculated, and low ETX and low latency paths are prioritized. Nodes on adjacent floors increase the receiving frequency and routing priority, while nodes on distant floors only moderately increase monitoring and heartbeat to avoid the entire network simultaneously entering the highest power consumption state.
[0078] In scenarios involving power outages and localized congestion in mains-powered routers, a fire causes two mains-powered routers to lose power. The remaining nodes simultaneously reselect routes, leading to congestion. The controller identifies parent node disconnection and high channel busy rates, reduces the normal reporting rate, distributes routing control messages by hash time slots, and increases the remaining energy and congestion weight of parent nodes in candidate paths. Alarm packets are given high priority, while normal diagnostic packets are merged to prevent network avalanche caused by all nodes simultaneously increasing retries. In scenarios where the battery is nearing its safe retention threshold, a smoke detector is at the edge of the alarm zone, but its remaining power is insufficient to maintain a 1-second poll for an extended period. The controller limits the minimum poll cycle based on the available duration, prioritizing improving single-link quality and selecting parent nodes with lower energy consumption. Simultaneously, it requests nearby routing nodes with higher power to handle redundant forwarding. This node still retains local audible and visual alarms and key summary reporting, preventing battery depletion due to communication optimization.
[0079] The direction of change of key algorithm parameters under each risk state can be seen in Table 1. The values in the table are only examples to illustrate the direction of parameter change. The actual values are determined based on product agreements, chip resources, certification requirements, battery capacity, network scale, and alarm deadlines. Table 1: Direction of Change of Key Algorithm Parameters under Different Risk States
[0080] To verify the effectiveness of this embodiment, hardware-in-the-loop or real-network testing can be conducted: Different smoke levels, temperatures, carbon monoxide variations, network sizes, routing node power-down ratios, interference intensity, and battery status are set. The differences between the fixed-parameter scheme and this scheme in end-to-end latency, alarm arrival rate, energy consumption per successful alarm, channel busy rate, number of duplicate packets, number of route switching times, and static battery life loss are compared. During verification, the input window, risk components, selected control vectors, reasons for rejection, prediction success rate, latency, energy consumption, and actual results are saved simultaneously, allowing for replay of each parameter change. The test should cover at least normal long-term operation, single-point false alarms, smoldering, open flames, simultaneous alarms from multiple nodes, continuous interference, parent node power-down, gateway power-down, low battery, and recovery phases. The comparison group uses the same hardware, packet load, and RF environment.
[0081] Example 2: This embodiment provides a fire risk-based MESH network alarm system, including a gateway and multiple alarm nodes. Each alarm node includes a risk fusion module, a strategy optimization module, and a parameter execution module. The risk fusion module collects fire sensing characteristics and network status characteristics, performs hierarchical fusion to obtain node risk, area risk, and network risk, and determines the risk status of the node. Its specific fusion method is the same as described in steps 1 and 2 of Embodiment 1. The strategy optimization module generates candidate combinations of cross-layer control vectors from a limited set of protocol-allowed parameter candidates based on the risk status. It filters the candidate combinations under constraints such as target reliability, message deadline, protocol parameter upper limit, battery safety reserve energy, and maximum duty cycle. It then selects a target control vector from the filtered candidate combinations based on a multi-objective cost that varies with the risk status. The cross-layer control vector includes at least routing parameters, retransmission parameters, polling parameters, and power consumption parameters. Its specific optimization process is the same as described in steps 3 to 5 of Embodiment 1. The parameter execution module adjusts protocol stack parameters according to the target control vector, and its execution method is the same as described in step 6 of Embodiment 1.
[0082] The system's alarm service layer provides the strategy optimization module with event types, alarm levels, reliability targets, and deadlines; each module can be implemented by the processor in the alarm node executing computer programs stored in memory, or it can be implemented by a combination of software and dedicated hardware circuits.
[0083] This embodiment also provides an alarm device, including a fire sensor, a wireless communication module, a memory, and a processor. The memory stores a computer program, which, when executed by the processor, implements the steps performed by the alarm device node in the method described in Embodiment 1. The fire sensor is, for example, one or more of a smoke sensor, a temperature sensor, or a carbon monoxide sensor; the wireless communication module supports Zigbee, Thread, Bluetooth Mesh, or proprietary Sub-G radio frequency.
[0084] The implementation methods in different protocols are as follows. In a Zigbee network, this can be achieved by dynamically adjusting terminal device Poll, application retry, APS acknowledgment policy, MAC retry, route discovery triggering conditions, broadcast radius, transmission power, packet priority queue, and parent node selection policy. For internal parameters that the standard does not allow to be directly modified, equivalent control can be achieved at the application layer through transmission interval, message merging, repeated transmission, route status monitoring, and NCP / Host policy interfaces.
[0085] In Thread or RPL-based networks, the polling cycle of sub-devices, parent node candidate scores, RPL objective function weights, Trickle-related control message rhythm, link modes, IPv6 queue priorities, and route repair triggering conditions can be adjusted. Risk weights can be added as vendor-internal routing cost supplements, but do not change standard secure sessions and IPv6 packet formats. In Bluetooth Mesh, the behavior of Relay, Friend, and Low-Power Nodes, message TTL, network retransmission, Relay retransmission, Friend Poll, and buffer priority can be adjusted according to risk. To avoid broadcast storms, congestion levels, duplicate packet ratios, and selected forwarding node mechanisms must be combined, rather than all Relays simultaneously increasing retransmission rates. In private Sub-G networks, complete control vectors can be directly implemented, including TDMA / CSMA time slots, receiver wake-up, transmit power, acknowledgments and retransmissions, primary / backup routing, heartbeats, and channel switching. Due to the high protocol autonomy, a hybrid centralized and distributed optimization approach can be adopted.
[0086] It should be noted that the mathematical expressions in the above embodiments are used to describe the relationships between variables and can be implemented by weighted summation, table lookup, fuzzy rules, decision trees, reinforcement learning strategies, or other equivalent algorithms. In resource-constrained microcontrollers, fixed-point arithmetic, table lookup, piecewise linear mapping, or finite candidate search can be used. This solution does not depend on a specific chip model, nor does it require modification of the security mechanisms of publicly available standards. The specific parameter names of each protocol may differ, but as long as routing, reliability, monitoring, and power consumption parameters are adjusted collaboratively using fire risk and network status, they fall within the technical concept of this invention. Equivalent substitutions or improvements made by those skilled in the art without departing from the inventive concept should be included within the protection scope of this invention.
Claims
1. A control method for a MESH network alarm system based on fire risk, characterized in that, The method includes: The fire sensing characteristics and network status characteristics of the alarm node are collected. The network status characteristics include link quality characteristics, channel congestion characteristics, topology impairment characteristics, and remaining energy characteristics. The fire sensing features and the network status features are fused in a hierarchical manner to obtain node risk, regional risk and network risk, and the risk status of the alarm node is determined accordingly. Based on the risk status, a candidate combination of cross-layer control vectors is generated from a limited set of candidate parameters allowed by the protocol. The cross-layer control vectors include at least routing parameters, retransmission parameters, polling parameters, and power consumption parameters. Candidate combinations that do not meet the constraints are filtered out, based on the constraints of target reliability, message deadline, upper limit of protocol parameters, battery safety reserve energy and maximum duty cycle. The multi-objective cost is calculated for each of the selected candidate combinations, and the candidate combination with the lowest cost is selected as the target control vector. The weights of each item in the multi-objective cost change with the risk state. The protocol stack parameters are adjusted according to the target control vector.
2. The control method for a fire risk-based MESH network alarm system according to claim 1, characterized in that, The layered fusion includes: The node risk is determined by sensing risk, trend risk, cross-sensor consistency risk, duration risk, and false alarm suppression term, wherein the false alarm suppression term is used to reduce the escalation magnitude of communication parameters; The regional risk is determined by fusion of neighbor risks with time and spatial weights, and the regional risk is formed based at least on the maximum value of neighbor risk, the number of neighbor nodes that reach the warning level or above, and the weighted average of neighbor risks. The network risk is obtained by fusing link risk, congestion risk, and topology damage risk, and a segmented rule is used for fusion: when any one of the link risk, the congestion risk, or the topology damage risk reaches the severity level, the network risk is not lower than the level corresponding to that severity level.
3. The control method for a fire risk-based MESH network alarm system according to claim 1, characterized in that, The process of generating the candidate combinations of the cross-layer control vectors includes: Based on the basic parameter template corresponding to the risk state, the candidate combination is generated by combining discrete parameter tables. The discrete parameter tables are preset with several discrete levels for each parameter in the cross-layer control vector. When generating the candidate combination, one or two level changes are made around the currently executed parameter level.
4. The control method for a fire risk-based MESH network alarm system according to claim 1, characterized in that, The process of determining the retransmission parameters includes: The success rate of a single transmission on the current link is estimated by statistically analyzing the transmission results of each physical transmission attempt. Based on the message type and the target success rate corresponding to the risk status, the base number of attempts is inferred from the probability relationship that at least one attempt will succeed in multiple attempts. The initial retransmission count is obtained by subtracting one from the base number of attempts. The initial retransmission count is then limited by the maximum number of retransmissions allowed by the protocol, the number of attempts that can be accommodated within the remaining deadline, the energy budget for this event, and the current congestion level to obtain the final retransmission parameters.
5. The control method for a fire risk-based MESH network alarm system according to claim 4, characterized in that, The method further includes: The causes of transmission failure are distinguished based on the link quality characteristics and the channel congestion characteristics: when the link quality is stable but the channel busy rate, queue occupancy, duplicate packet ratio, and backoff count increase, it is determined to be channel contention congestion; when the channel busy rate and queue occupancy are low but the link quality continues to decline, it is determined to be link fading. When channel contention and congestion are identified, the following actions are performed in order of priority: restricting low-priority services, merging periodic data, sending data in staggered shifts according to node-related time slots, selecting a parent node or channel with lower congestion, adjusting backoff parameters, and increasing the number of retransmissions to a limited extent if the remaining reliability is still insufficient after the aforementioned actions are performed. When a link fading is detected, adjust the transmit power or switch the transmission path.
6. The control method for a fire risk-based MESH network alarm system according to claim 1, characterized in that, The process of determining the polling parameters and the power consumption parameters includes: Based on the risk status, the amount of data to be sent cached by the parent node for this node, the historical downlink arrival rate, and the battery available energy after deducting the safety reserve, the polling cycle is adjusted step by step in discrete polling cycle levels. The duration of the receiving window is determined, and the duration of the receiving window shall at least cover the clock drift between the current node and the parent node, the parent node response delay, the downlink packet air interface time, and the security processing time. When the risk status increases, the duration of the receiving window is extended. When no data is received in multiple consecutive receiving windows and the risk decreases, the duration of the receiving window is shortened step by step. The duration of the receiving window shall not be less than the base duration that covers the clock drift, the parent node response delay, the air interface time, and the security processing time. The polling period, the receiving window duration, the transmit power, and the retransmission parameters are jointly evaluated, and the combination with the lower expected total energy consumption is selected from the combinations that meet the target reliability and message deadline.
7. The control method for a fire risk-based MESH network alarm system according to claim 1 or 6, characterized in that, The power consumption parameter includes the transmission power, and the process of adjusting the transmission power includes: The actual link margin is estimated based on the received signal strength of acknowledgment frames, neighbor reports, or probe frames, and the target link margin is determined based on the risk status. When the difference between the target link margin and the actual link margin exceeds the positive dead zone for multiple consecutive observation windows, the transmit power is increased by one power level; when the actual link margin is higher than the target link margin and exceeds the negative dead zone and remains stable, the transmit power is decreased by one power level; the magnitude of a single adjustment does not exceed the preset change limit. When the transmission failure is determined to be caused by channel congestion, the transmit power is not increased; when the transmit power reaches the upper limit and the actual link margin is still lower than the target link margin, a parent node switch or path switch is performed.
8. The control method for a fire risk-based MESH network alarm system according to claim 1, characterized in that, The process of determining the routing parameters includes: Calculate the routing cost of candidate parent nodes or candidate paths. The routing cost includes at least the expected number of link transmissions, the expected end-to-end delay, the energy consumption per packet, the remaining energy of the parent node, the congestion of the parent node queue, the spatial risk exposure, and the handover cost. The spatial risk exposure is determined based on the regional risk of the area traversed by the path, and the weight of each cost changes with the risk status. When the routing cost of the new path is better than that of the current path for a preset number of consecutive periods and the improvement exceeds the switching hysteresis threshold, a path switch is performed. The higher the risk level, the smaller the preset number and the switching hysteresis threshold, and the path switch is limited by the maximum switching frequency and routing loop constraints.
9. The control method for a fire risk-based MESH network alarm system according to claim 1, characterized in that, The risk states include at least normal state, observation state, early warning state, alarm state, and emergency state. The entry threshold for each risk state is higher than the corresponding exit threshold. When the risk increases, the higher risk state is entered when the entry threshold is met. When the risk decreases, it needs to be below the exit threshold for a preset number of consecutive cycles and the network indicators need to be stable before gradually recovering. When no candidate combination satisfies the constraints, a security degradation process is initiated: low-priority periodic reports are suspended, diagnostic data is merged, alarm summaries are sent with priority, an alternate parent node or redundant path is selected, and controlled reliable flooding is performed when the risk state is an emergency state.
10. A MESH network alarm system based on fire risk, characterized in that, It includes a gateway and multiple alarm nodes, wherein the alarm nodes include: The risk fusion module is used to collect fire sensor features and network status features, and fuse them in layers to obtain node risk, regional risk and network risk, and determine the risk status of the node. The strategy optimization module is used to generate candidate combinations of cross-layer control vectors from a limited set of candidate parameters allowed by the protocol based on the risk state. The candidate combinations are screened under constraints such as target reliability, message deadline, protocol parameter upper limit, battery safety reserve energy, and maximum duty cycle. The target control vector is selected from the screened candidate combinations according to the multi-objective cost that changes with the risk state. The cross-layer control vector includes at least routing parameters, retransmission parameters, polling parameters, and power consumption parameters. The parameter execution module is used to perform protocol stack parameter adjustments according to the target control vector.
