Intelligent building fire safety early warning system based on Internet platform

Abstract

The invention discloses an intelligent building fire safety early warning system based on an internet platform, relates to the technical field of building fire safety early warning, and aims to identify a potential fire scene in advance by performing deep modeling on personnel behaviors, elevator use, lighting states and load power operation data, so that the potential fire scene can be identified in advance even if smoke or high temperature is not generated in a fire. And systematic anomalies can also be perceived. The system establishes a reference behavior reference value in a sliding time window through a building behavior track modeling module, and dynamically adapts to operation rules of different buildings, different time periods and different floors. The method avoids false alarm and missing alarm caused by a static threshold value, and is suitable for various building types such as office buildings and shopping malls. The energy load anomaly inversion module and the behavior deviation recognition module work cooperatively, the anomaly that no person enters but the elevator operates frequently can be recognized, then coupling index judgment is conducted through the logic conflict detection module, and the capacity of recognizing complex atypical fire premonition is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building fire safety early warning, in particular to a smart building fire safety early warning system based on an Internet platform. BACKGROUND

[0002] Under the background of the continuous growth of urban high-density building clusters, the personnel flow, facility operation and energy consumption change in the building constitute rich and observable behavior data streams. Through the analysis of these behavior data streams, it is expected to achieve in-depth understanding and prediction of the building state. Especially in the fire safety scene, if a "daily behavior model" can be built based on the running behavior characteristics of the building, when a period or space that is significantly inconsistent with the normal operation characteristics appears, potential fire hazard can be identified in advance, thereby gaining time for subsequent response and reducing losses.

[0003] The current common smart building fire safety system mainly relies on independently deployed physical sensors such as smoke sensors, temperature sensors and open fire recognition to form a fire sensing network. Although this method can achieve a faster response in the open fire stage, it is difficult to detect the "hidden stage" of fire development, such as equipment overload heating in unoccupied office areas at night, abnormal load in pipe wells, etc.

[0004] The above-mentioned traditional fire safety system relying on physical sensors has significant time delay and identification error. When the fire is in the stage of not generating smoke or open fire, the building may have shown signs of imbalance in behavior structure: such as the lighting being on for a long time, the elevator being frequently called, and the air conditioner being continuously running in the office building at night without personnel entering; or the load of the power distribution circuit increases sharply for a short time, but there is no sign of personnel activity. These phenomena often do not occur in isolation, but are manifestations of prelude signals such as accumulation of potential high heat sources, air conditioner condenser short circuit, and cable overheating. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a smart building fire safety early warning system based on an Internet platform, which solves the problems mentioned in the background art.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: a smart building fire safety early warning system based on an Internet platform, comprising a building behavior trajectory modeling module, a nonlinear behavior deviation identification module, an energy load anomaly inversion module, an associated logic conflict detection module, a potential fire reasoning evaluation module and an early warning release and instruction feedback module.

[0007] The building behavior trajectory modeling module collects data parameters of the building based on the existing Internet connection system of the building, fits the building data set LW, and establishes a behavior reference benchmark model in a time window Tw to obtain a behavior benchmark value ABS.

[0008] The nonlinear behavior deviation identification module combines the behavior benchmark value ABS with the building data set LW to obtain a behavior deviation index Pab;

[0009] The energy load anomaly inversion module analyzes the building data set LW to obtain the average load usage per person, and evaluates the deviation degree with the average load level to obtain an energy load anomaly index Pene;

[0010] The correlation logic conflict detection module fuses the behavior deviation index Pab and the energy load anomaly index Pene to calculate a logic conflict value Cerr;

[0011] The potential fire reasoning evaluation module analyzes the logic conflict value Cerr and combines the introduced building silence state to calculate a fire potential index Rfth;

[0012] The early warning release and instruction feedback module compares the obtained fire potential index Rfth with a preset fire threshold Trf to determine the fire state.

[0013] Preferably, the building behavior trajectory modeling module includes a data structure normalization processing unit and a behavior modeling and benchmark fitting unit;

[0014] The data structure normalization processing unit collects data parameters from the existing internet connection system of the building, including the number of people entering An, the elevator call frequency EL, the lighting use cycle Lm, and the floor average load power Ap, and fits them into the original data set YW;

[0015] The number of people entering An is obtained through the records of access card swiping and face recognition in the network access control system;

[0016] The elevator call frequency EL is obtained through the elevator dispatch request events recorded in the elevator control system

[0017] The lighting use cycle Lm is obtained through the cumulative duration of the lighting control system loop opening;

[0018] The floor average load power Ap is obtained by weighting the real-time power of each floor power distribution branch through the power distribution monitoring system;

[0019] The original data set YW is noise-removed by filtering method and normalized to obtain the building data set LW;

[0020] The building data set LW is obtained by the following formula:

[0021]

[0022] In the formula, LWo represents the oth data in the building data set LW, YWo represents the oth data in the original data set YW, minYWo represents the valley value of the oth data in the original data set YW, and maxYWo represents the peak value of the oth data in the original data set YW.

[0023] Preferably, the behavior modeling and benchmark fitting unit models the statistical distribution of the building data set LW within the time window Tw, extracts the behavior benchmark value ABS of the normal behavior state of the building through the median value analysis method;

[0024] The behavior benchmark value ABS is obtained by the following formula:

[0025]

[0026] In the formula, Med(An) represents the median value of the number of personnel entering An within the time window Tw, Med(EL) represents the median value of the elevator call frequency EL within the time window Tw, Med(Lm) represents the median value of the lighting usage period Lm within the time window Tw, and Med(Ap) represents the median value of the average load power Ap of the floor within the time window Tw.

[0027] Preferably, the nonlinear behavior deviation identification module includes a multidimensional behavior projection calculation unit and a comprehensive behavior discrimination unit.

[0028] The multidimensional behavior projection calculation unit extracts the behavior parameters, including the number of personnel entering An, the elevator call frequency EL, and the lighting usage period Lm, from the building data set LW, and projects them into the difference space of the reference benchmark ABS to obtain the personnel projection deviation function DAn and the lighting projection deviation function DLm.

[0029] The personnel projection deviation function DAn is obtained in the following manner: the average level of the number of personnel entering An within a set historical time window, i.e., the average number of personnel entering, is extracted; then the number of personnel entering An(t) at time t is subtracted from the average number of personnel entering to obtain the original offset; and finally, the original offset is divided by "1 plus the elevator call frequency" to obtain the personnel projection deviation function DAn.

[0030] The lighting projection deviation function DLm is obtained in the following manner: the expected value of the lighting usage period Lm is extracted from the historical data; then the lighting usage period Lm(t) at time t is subtracted from the historical expected value to obtain the offset difference value; and finally, the offset difference value is divided by "1 plus the current number of entering personnel" to obtain the lighting projection deviation function DLm.

[0031] Preferably, the comprehensive behavior discrimination unit compares the obtained personnel projection deviation function DAn and the lighting projection deviation function DLm with the behavior benchmark value ABS, obtains the behavior deviation index Pab, and makes a judgment on the behavior state of the building.

[0032] The behavior deviation index Pab is obtained by the following formula:

[0033]

[0034] In the formula, Pab(t) represents the behavior deviation index at time t, ln represents the logarithmic function, DAn(t) represents the personnel projection deviation function at time t, DLm(t) represents the lighting projection deviation function at time t, and ABS(t) represents the behavior reference value at time t.

[0035] The behavior state of the building is matched by the following method:

[0036] When the behavior deviation index Pab is less than 0.8, it indicates that the behavior is normal.

[0037] When the behavior deviation index Pab is between 0.8 and 1.5, it indicates a slight abnormality.

[0038] When the behavior deviation index Pab is greater than or equal to 1.5, it indicates a strong deviation, triggering an alarm.

[0039] Preferably, the energy load anomaly inversion module includes a unit load intensity analysis unit and an average deviation fitting discrimination unit.

[0040] The unit load intensity analysis unit analyzes the building data set LW, extracts the floor average load power Ap and the number of personnel entering An, and calculates and obtains the unit personnel load intensity Qunit.

[0041] The unit personnel load intensity Qunit is obtained by first obtaining the average load power value of the building at the current time point, then extracting the number of personnel entering the building An at the current time from the access control system, and finally dividing the obtained average load power value by the value of "the number of personnel entering An plus one", which is the unit personnel load intensity Qunit.

[0042] The average deviation fitting discrimination unit compares the obtained unit personnel load intensity Qunit with the average load level μ(Ap) in the time window Tw to obtain the energy load anomaly index Pene.

[0043] The energy load anomaly index Pene is obtained by the following formula:

[0044]

[0045] In the formula, E represents a non-zero constant, which is used to avoid the risk of calculation with a denominator of 0.

[0046] Preferably, the correlation logic conflict detection module includes an anomaly index cross integration unit and a non-typical logic conflict identification unit.

[0047] The abnormal index cross-integration unit fuses the behavior deviation index Pab and the energy load abnormal index Pene obtained, and constructs a combined expression of the two through a mathematical structure to obtain a logic conflict value Cerr;

[0048] The logic conflict value Cerr is obtained by the following formula:

[0049] Cerr(t) = Pab(t) x ln(1 + Pene(t));

[0050] In the formula, Cerr(t) represents the logic conflict value at time t, Pab(t) represents the behavior deviation index at time t, Pene(t) represents the energy load abnormal index at time t, and ln represents a logarithmic function.

[0051] Preferably, the atypical logic conflict identification unit compares the fused logic conflict value Cerr with a dynamic threshold value Terr to determine whether the atypical operation state feature is met;

[0052] The dynamic threshold value Terr is obtained in the following manner:

[0053] The logic conflict values Cerr in the past 72 hours are extracted from the system operation history, and a historical sequence set is constructed;

[0054] Statistical feature extraction is performed on the historical sequence set, and the logic median MC and the logic standard deviation σC are calculated;

[0055] Based on the obtained median MC and logic standard deviation σC, the dynamic threshold value Terr is calculated;

[0056] The dynamic threshold value Terr is obtained by the following formula: Terr = MC + 1.5 x σC;

[0057] The operation state feature is matched and obtained in the following manner:

[0058] When the logic conflict value Cerr is less than the dynamic threshold value Terr, it indicates that the building is in an acceptable coupled fluctuation state;

[0059] When the logic conflict value Cerr is greater than or equal to the dynamic threshold value Terr, an atypical logic abnormality identification flag is triggered, indicating a potential high-risk node.

[0060] Preferably, the potential fire reasoning evaluation module collects and normalizes the vibration response event number Na, the motion response number Nb, and the air flow disturbance number Nc through acoustic sensors, human infrared devices, and air flow sensors, and calculates and obtains a silence degree index Qsi;

[0061] The silence degree index Qsi is obtained by the following formula:

[0062]

[0063] In the formula, Qsi(t) represents the silence degree index at time t, Na(t) represents the number of vibration response events at time t, Nb(t) represents the number of action response times at time t, Nc(t) represents the number of air flow disturbance times at time t, Nto represents the sum of the maximum triggering times of the acoustic sensor, the human infrared device and the air flow sensor, and Eo represents a non-zero constant;

[0064] The logic conflict value Cerr is fused with the silence degree index Qsi to obtain the fire potential index Rfth; the fire potential index Rfth is obtained through the following formula:

[0065]

[0066] Preferably, the early warning issuing and instruction feedback module analyzes the fire potential index Rfth and compares it with the preset fire threshold Trf to determine the fire state;

[0067] The fire state is matched and obtained through the following way:

[0068] When the fire potential index Rfth is less than the fire threshold Trf, it indicates that the fire state is normal, and the monitoring is maintained;

[0069] When the fire potential index Rfth is greater than or equal to the fire threshold Trf, it indicates that the fire state is abnormal, and enters the early warning response state;

[0070] When the early warning response state is performed, the building internal equipment, external platform and system log are linked to issue a response, and a full closed-loop operation chain for fire emergency is constructed, including the operations of “triggering→guiding→reporting→storing”;

[0071] Emergency prompt sound triggering: the warning sound channel is activated through the building broadcast system, and the standard fire alarm voice and the buzzer signal are emitted;

[0072] Imagery escape guidance: the visual display screen in the corridor and the hall area is switched to the evacuation passage guidance map;

[0073] Multi-terminal synchronous information pushing: the fire state information is pushed to the following terminals through the Internet interface / API: the fire control host; the property remote monitoring system; the designated management personnel mobile terminal App;

[0074] Historical data latching mechanism starting: all building data parameters in the current time are labeled with a time stamp, forming a “fire event log package”, and uploaded to the cloud and the local log area.

[0075] The application provides an intelligent building fire safety early warning system based on an Internet platform, which has the following beneficial effects:

[0076] (1) When the system is running, potential fire scene is identified in advance by deeply modeling multiple types of operation data such as personnel behavior, elevator use, lighting state and load power, and even when smoke or high temperature has not yet been generated in a fire, systematic abnormalities can be perceived. The system establishes reference behavior benchmark values in a sliding time window through a building behavior trajectory modeling module, and realizes dynamic adaptation of operation rules of different buildings, different time periods and different floors. This method avoids false positives and false negatives caused by static thresholds, and is suitable for various building types such as office buildings, shopping malls and dormitories.

[0077] The energy load anomaly inversion module and the behavior deviation identification module work cooperatively to identify combined abnormalities such as 'no personnel entering but frequent elevator operation' and 'night lighting on for a long time but no one', and then make coupling exponential judgments through a logic conflict detection module, thereby significantly improving the ability to identify complex and atypical fire precursors. The system introduces a silence index Qsi to comprehensively analyze weak response behaviors such as human perception, acoustics and airflow, and through reasoning of the 'active load in a silent background' scene, further supplements the judgment ability of potential fire in unoccupied areas, equipment rooms and pipe well areas that cannot be perceived by traditional systems.

[0078] (2) The data structure normalization processing unit collects, cleanses, denoises and normalizes heterogeneous raw data in multiple subsystems inside the building, successfully constructing a unified dimension building dataset LW. This dataset effectively solves the problem that traditional systems cannot be fused and modeled due to different data standards and large dimension differences in each subsystem, significantly improving the accuracy and expandability of behavior modeling. Through the behavior modeling and benchmark fitting unit, the concept of sliding time window Tw is introduced, and based on the median value analysis method, the behavior benchmark value ABS is constructed, so that the system can dynamically capture the 'typical operation trajectory' of the building in different time periods and different use modes.

[0079] (3) Through multi-dimensional behavior projection calculation, the system no longer relies on absolute numerical judgment of a single behavior index, but differentiates the interaction between the number of personnel entering and the frequency of elevator use and the length of lighting on time, enhances the identification sensitivity through a nonlinear function, especially for 'normal surface but abnormal structure' behavior combinations, has stronger discrimination ability, and effectively improves the accuracy of early fire warning.

[0080] (4) The logic conflict value Cerr is taken as a fusion index, not just as a subsidiary judgment value of behavior or energy consumption, but is independently defined as an important intermediary signal of potential abnormalities. A significant single-point increase in this index can directly trigger the marking of high-risk nodes and serve as the input basis for the subsequent "fire potential assessment module", providing a pilot warning mechanism for early intervention. In real-world scenarios, many fire precursors do not follow the linear rule of "severe abnormalities in a certain parameter", but rather exhibit mild or non-synchronous deviations in multiple indicators. BRIEF DESCRIPTION OF DRAWINGS

[0081] Figure 1 A block diagram flowchart of a smart building fire safety early warning system based on an Internet platform according to the present application;

[0082] Figure 2 A flowchart of a fire potential index acquisition according to the present application;

[0083] Figure 3 A behavior deviation index trend chart according to the present application. DETAILED DESCRIPTION

[0084] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0085] Embodiment 1

[0086] The present application provides a smart building fire safety early warning system based on an Internet platform, please refer to Figure 1 , which comprises a building behavior trajectory modeling module, a nonlinear behavior deviation identification module, an energy load anomaly inversion module, a correlation logic conflict detection module, a potential fire reasoning evaluation module, and an early warning release and instruction feedback module.

[0087] The building behavior trajectory modeling module collects data parameters of the building based on the existing Internet connection system of the building, fits the building data set LW, and establishes a behavior reference benchmark model within a time window Tw to obtain a behavior benchmark value ABS.

[0088] The nonlinear behavior deviation identification module combines the behavior benchmark value ABS with the building data set LW to obtain a behavior deviation index Pab.

[0089] The energy load anomaly inversion module analyzes the building data set LW to obtain the average load usage per unit of personnel, and evaluates the deviation degree with the average load level to obtain an energy load anomaly index Pene.

[0090] The association logic conflict detection module fuses the behavior deviation index Pab and the energy load anomaly index Pene to calculate a logic conflict value Cerr;

[0091] The potential fire reasoning evaluation module analyzes the logic conflict value Cerr and combines the introduced building silence state to calculate a fire potential index Rfth;

[0092] The early warning release and instruction feedback module compares the obtained fire potential index Rfth with a preset fire threshold Trf to determine the fire state.

[0093] In this embodiment, compared with the traditional fire-fighting system relying on physical triggers such as smoke and temperature, the system can recognize potential fire scenarios in advance by deeply modeling multiple types of operation data such as personnel behavior, elevator use, lighting state, and load power. Even when there is no smoke or high temperature, the system can also sense systematic abnormalities. The system establishes reference behavior benchmark values in a sliding time window through the building behavior trajectory modeling module, which dynamically adapts to the operation rules of different buildings, different time periods, and different floors. This approach avoids false positives and false negatives caused by static thresholds and is suitable for various building types such as office buildings, shopping malls, and dormitories.

[0094] The energy load anomaly inversion module and the behavior deviation identification module work together to identify combined abnormalities such as “no personnel entering but frequent elevator operation” and “night lighting is on for a long time but no one is present”. Then, the logic conflict detection module is used for coupling and index judgment, which significantly improves the ability to identify complex and atypical fire precursors. The system introduces the silence index Qsi to comprehensively analyze weak response behaviors such as human perception, acoustics, and airflow. By reasoning the “active load in a silent background” scenario, the system further supplements the judgment ability of potential fire in unoccupied areas, equipment rooms, and pipe shafts that cannot be sensed by traditional systems.

[0095] The finally generated fire potential index fuses multiple source indicators to support quantitative risk level judgment, which is convenient for platform managers and property guards to process response levels and can set multiple response mechanisms (prompt warning → notification → linkage device control → report fire) as needed.

[0096] Embodiment 2

[0097] This embodiment is an explanation and description in Embodiment 1. Please refer to Figure 1 , specifically: the building behavior trajectory modeling module includes a data structure normalization processing unit and a behavior modeling and benchmark fitting unit;

[0098] The data structure normalization processing unit collects data parameters from the existing Internet connection system of the building, including the number of personnel entering An, the elevator call frequency EL, the lighting use period Lm, and the average load power Ap of the floor, and fits them into the original data set YW;

[0099] Among them, the number of personnel entering An is obtained through the access control system of the network access control system and the face recognition record;

[0100] The elevator call frequency EL is obtained by collecting the elevator dispatch request events recorded in the elevator control system

[0101] The lighting use period Lm is obtained by the cumulative duration of the loop opening through the lighting control system;

[0102] The average load power Ap of the floor is obtained by weighting the real-time power of each floor distribution branch through the power distribution monitoring system;

[0103] The original data set YW is processed by filtering method to remove noise and normalized to obtain the building data set LW;

[0104] The building data set LW is obtained by the following formula:

[0105]

[0106] In the formula, LWo represents the oth data in the building data set LW, YWo represents the oth data in the original data set YW, minYWo represents the valley value of the oth data in the original data set YW, and maxYWo represents the peak value of the oth data in the original data set YW.

[0107] The behavior modeling and benchmark fitting unit statistically models the building data set LW within the time window Tw, and extracts the behavior benchmark value ABS of the normal behavior state of the building through the median value analysis method;

[0108] The behavior benchmark value ABS is obtained by the following formula:

[0109]

[0110] In the formula, Med(An) represents the median value of the number of personnel entering An within the time window Tw, Med(EL) represents the median value of the elevator call frequency EL within the time window Tw, Med(Lm) represents the median value of the lighting use period Lm within the time window Tw, and Med(Ap) represents the median value of the average load power Ap of the floor within the time window Tw.

[0111] In this embodiment, the heterogeneous raw data in the multiple subsystems inside the building is collected, cleaned, denoised and normalized by the data structure normalization processing unit, and a unified dimension building dataset LW is successfully constructed. The dataset effectively solves the problem of not being able to fuse modeling due to different data standards and large dimension differences in traditional systems, and significantly improves the accuracy and scalability of behavior modeling. Through the behavior modeling and benchmark fitting unit, the concept of sliding time window Tw is introduced, and the behavior benchmark value ABS is constructed based on the median value analysis method, so that the system can dynamically capture the "typical running trajectory" of the building in different time periods and different use modes. Compared with the traditional false alarm prone mechanism based on static threshold, this method is more suitable for the behavior characteristic changes of office buildings, dormitories, shopping malls and other scenes in the morning and evening, holidays, shift changes and other time periods.

[0112] The normalized building dataset LW and the behavior benchmark value ABS constitute the "standard behavior interval", and the subsequent deviation identification module and load inversion module are referenced, which has good discrimination ability. Especially when the behavior- energy consumption mismatch phenomenon occurs at night or during device dormancy, potential abnormalities can be identified earlier, effectively making up for the technical shortcoming of the existing fire safety system in the "unattended phase" of weak recognition ability.

[0113] Embodiment 3

[0114] This embodiment is an explanation and description in embodiment 2, please refer to Figure 1 , in particular: the nonlinear behavior deviation identification module includes a multi-dimensional behavior projection calculation unit and a comprehensive behavior judgment unit;

[0115] The multi-dimensional behavior projection calculation unit extracts behavior parameters from the building dataset LW, including the number of personnel entering An, the frequency of elevator calling EL and the lighting use cycle Lm, and projects them into the difference space of the reference benchmark ABS to obtain the personnel projection deviation function DAn and the lighting projection deviation function DLm;

[0116] The personnel projection deviation function DAn is obtained in the following way: the average level of the number of personnel entering An in the set historical time window, that is, the average number of personnel entering, is extracted; then the number of personnel entering An(t) at time t is subtracted from the average number of personnel entering to obtain the original offset; finally, the original offset is divided by "1 plus the frequency of elevator calling" to obtain the personnel projection deviation function DAn;

[0117] The lighting projection deviation function DLm is obtained in the following way: the expected value of the lighting use cycle Lm is extracted from the historical data; then the lighting use cycle Lm(t) at time t is subtracted from the historical expected value to obtain the offset difference value; finally, the offset difference value is divided by "1 plus the current number of entries" to obtain the lighting projection deviation function DLm.

[0118] The comprehensive behavior discrimination unit compares the personnel projection deviation function DAn and the lighting projection deviation function DLm with the behavior reference value ABS, obtains a behavior deviation index Pab, and judges the behavior state of the building;

[0119] The behavior deviation index Pab is obtained by the following formula:

[0120]

[0121] In the formula, Pab(t) represents the behavior deviation index at time t, ln represents a logarithmic function, DAn(t) represents the personnel projection deviation function at time t, DLm(t) represents the lighting projection deviation function at time t, and ABS(t) represents the behavior reference value at time t.

[0122] The behavior state of the building is obtained by matching in the following manner:

[0123] When the behavior deviation index Pab is less than 0.8, it indicates normal behavior.

[0124] When the behavior deviation index Pab is between 0.8 and 1.5, it indicates slight abnormality.

[0125] When the behavior deviation index Pab is greater than or equal to 1.5, it indicates strong deviation, triggering an alarm.

[0126] The energy load anomaly inversion module includes a unit load intensity analysis unit and an average deviation fitting discrimination unit.

[0127] The unit load intensity analysis unit analyzes the building data set LW, extracts the floor average load power Ap and the number of personnel entering An, and calculates and obtains the unit personnel load intensity Qunit.

[0128] The unit personnel load intensity Qunit is obtained in the following manner: first, obtain the average load power value of the building at the current time point; then extract the number of personnel entering An at the current time from the access control system; finally, divide the obtained average load power value by the value of "the number of personnel entering An plus one", which is the unit personnel load intensity Qunit.

[0129] The average deviation fitting discrimination unit compares the obtained unit personnel load intensity Qunit with the load average level μ(Ap) in the time window Tw to obtain an energy load anomaly index Pene.

[0130] The energy load anomaly index Pene is obtained by the following formula:

[0131]

[0132] In the formula, E represents a non-zero constant.

[0133] In this embodiment, through the multi-dimensional behavior projection calculation method, the judgment of the absolute value of a single behavior index is no longer relied on, but the interaction relationship between the number of personnel entering and the frequency of elevator use and the length of lighting on time is differentially processed, the recognition sensitivity is enhanced through a nonlinear function, especially for the behavior combination of "normal surface but abnormal structure", which has stronger discrimination ability, and the early warning accuracy of fire is effectively improved.

[0134] In this embodiment, the number of personnel entering and the lighting cycle are respectively constructed into independent deviation functions, and the transformation from "behavior intensity" to "behavior structure deviation" is realized by different projection with historical reference data. This method can identify typical early warning behavior combinations such as "few personnel entering at night, frequent elevator use" and "lighting on for a long time but no one enters", and provides the system with early judgment basis at the behavior chain level. The introduction of behavior deviation index Pab enables the system to make three-level judgment (normal, slight abnormal, strong deviation) on the current behavior state of the building, not only improves the recognition accuracy, but also provides clear trigger conditions for the graded response of the early warning module, realizes "risk early warning hierarchical management", and meets the needs of diversified control scenarios of large and medium-sized buildings.

[0135] The energy load anomaly inversion module effectively reveals the deep connection between "behavior intensity and energy consumption matching degree" through unit load intensity calculation. For example, in the case of zero personnel entering at night but significant increase in load power, it can be clearly determined as an abnormal load concentration phenomenon, which helps to identify potential risks such as illegal power use, unturned off appliances, and local heating of equipment, and breaks through the blind spot of traditional total load judgment method. The average deviation fitting discrimination unit uses the load mean value in the sliding time window as the reference benchmark, so that each determination can be dynamically adjusted based on the historical running state, and has self-adaptive judgment ability for different running scenarios such as daytime peak and nighttime valley, effectively reducing the false positive rate caused by stage load fluctuation.

[0136] Embodiment 4

[0137] This embodiment is an explanation and description in embodiment 3, please refer to Figure 1 , in particular: the correlation logic conflict detection module includes an abnormal index cross integration unit and a non-typical logic conflict identification unit;

[0138] The abnormal index cross integration unit fuses the behavior deviation index Pab and the energy load anomaly index Pene, and constructs the combination expression of the two through mathematical structure, to obtain the logic conflict value Cerr;

[0139] The logic conflict value Cerr is obtained by the following formula:

[0140] Cerr(t) = Pab(t) x ln(1 + Pene(t));

[0141] In the formula, Cerr(t) represents a logical conflict value at time t, Pab(t) represents a behavior deviation index at time t, Pene(t) represents an energy load anomaly index at time t, and ln represents a logarithmic function.

[0142] The atypical logical conflict identification unit compares the fused logical conflict value Cerr with the dynamic threshold Terr to determine whether the atypical operation state feature is met.

[0143] The dynamic threshold Terr is obtained in the following manner:

[0144] Logical conflict values Cerr in the past 72 hours are extracted from the system operation history, and a historical sequence set is constructed.

[0145] Statistical feature extraction is performed on the historical sequence set, and the logical median MC and the logical standard deviation σC are calculated.

[0146] Based on the obtained median MC and the logical standard deviation σC, the dynamic threshold Terr is calculated.

[0147] The dynamic threshold Terr is obtained by the following formula: Terr = MC + 1.5 x σC.

[0148] The operation state feature is matched and obtained in the following manner:

[0149] When the logical conflict value Cerr is less than the dynamic threshold Terr, it indicates that the building is in an acceptable coupling fluctuation state.

[0150] When the logical conflict value Cerr is greater than or equal to the dynamic threshold Terr, it indicates that the atypical logical anomaly identification flag is triggered, and it is marked as a potential high-risk node.

[0151] In this embodiment, the information silos are broken, and the personnel behavior and energy consumption features are deeply fused in the mathematical structure level. A combination expression of cross-system indicators is constructed, and behavior-energy joint anomaly analysis is realized. This fusion not only improves the coverage of identification, but also discovers potential safety hazards such as "personnel missing but load rising" or "frequent entry and exit but no load fluctuation" and other atypical phenomena. Compared with the defects of false positives and false negatives caused by traditional static preset thresholds, the introduction of the dynamic threshold Terr in this embodiment analyzes the median and standard deviation of the historical logical conflict values in the past 72 hours, and autonomously constructs the judgment threshold. The system can adaptively update the identification strategy according to the state changes of different buildings, different time periods, and different operation stages, greatly improving the stability and rationality of anomaly identification and reducing the environmental adaptation cost.

[0152] The logic conflict value Cerr is a fusion index, which is no longer a subsidiary judgment value of behavior or energy consumption, but is independently defined as an important intermediary signal of potential abnormality. A single point of the index significantly rising can directly trigger the marking of a high-risk node and serve as the input basis of the subsequent "fire potential assessment module", providing a pilot warning mechanism for early intervention. In real scenarios, many fire precursors do not follow the linear rule of "severe abnormality of a certain parameter", but show slight or non-synchronous deviation of multiple indicators. The embodiment has the ability to capture "weakly coupled abnormal combinations" or "crossed chronic abnormality" by identifying atypical logic conflict characteristics, which can effectively fill the recognition blind spot of existing systems for "preheating fire" and "abnormal night operation".

[0153] Embodiment 5

[0154] This embodiment is an explanation and description in embodiment 4, please refer to Figure 1 Specifically, the potential fire reasoning assessment module collects and normalizes the vibration response event number Na, the action response number Nb and the air flow disturbance number Nc through the acoustic sensor, the human infrared device and the air flow sensor, and calculates and obtains the silence degree index Qsi.

[0155] The silence degree index Qsi is obtained by the following formula:

[0156]

[0157] In the formula, Qsi(t) represents the silence degree index at time t, Na(t) represents the vibration response event number at time t, Nb(t) represents the action response number at time t, Nc(t) represents the air flow disturbance number at time t, Nto represents the total number of maximum trigger times of the acoustic sensor, the human infrared device and the air flow sensor, and Eo represents a non-zero constant.

[0158] The logic conflict value Cerr and the silence degree index Qsi are fused and operated to calculate and obtain the fire potential index Rfth; the fire potential index Rfth is obtained by the following formula:

[0159]

[0160] The warning release and instruction feedback module analyzes the fire potential index Rfth and compares it with the preset fire threshold Trf to judge the fire state.

[0161] The fire state is matched and obtained by the following way:

[0162] When the fire potential index Rfth is less than the fire threshold Trf, it indicates that the fire state is normal and the monitoring is maintained.

[0163] When the fire potential index Rfth is greater than or equal to the fire threshold Trf, it indicates that the fire state is abnormal, and the system enters the early warning response state;

[0164] When the early warning response state is entered, the system sends a response to the internal equipment of the building, the external platform, and the system log, and constructs a full closed-loop operation chain for fire emergency, including the operations of "triggering, guiding, reporting, and archiving";

[0165] Emergency prompt sound triggering: The system activates the warning sound channel through the building broadcast system and sends out standard fire alarm voice and buzzer signals.

[0166] Image-based escape guidance: The visual display screen in the corridor and hall area is switched to a evacuation passage guidance map.

[0167] Multi-terminal synchronous information pushing: The system uses the Internet interface / API to push the fire state information to the following terminals: fire control host, property remote monitoring system, and designated management personnel mobile terminal App.

[0168] Historical data latching mechanism activation: The system timestamps all building data parameters within the current time, forms a "fire event log package", and uploads it to the cloud and local log area.

[0169] In this embodiment, the traditional intelligent building fire alarm system mostly relies on physical triggers such as flame, smoke, and temperature rise. However, this module breaks through this passive triggering mechanism, fuses the building internal logic conflict value Cerr and the quietness index Qsi, identifies the composite state of "high abnormality + high quietness", accurately points to the fire risk caused by the "potential unattended + high energy consumption" scene, and enables the system to have an earlier and deeper "preliminary fire identification" capability. Through acoustic, human sensing, and airflow sensing data, the system dynamically obtains human activity traces in the space, proposes the concept of quietness index Qsi, and successfully quantifies the trend of "reduction of physical disturbance". When the quietness index Qsi is abnormally high, but the energy consumption behavior is still deviating from the norm, the system can effectively identify typical "static high-risk fire source points" such as left-over electrical appliances, long-time lighting, and independent spaces, and fill the perception blind spot of previous early warning systems for "vacant place implicit high-risk" areas.

[0170] The system fuses and calculates the behavior logic conflict value Cerr and the quietness index Qsi to construct a unified dimension fire potential index Rfth, which not only improves the judgment accuracy, but also provides a quantifiable index for automatic determination of the early warning response level. As a "state value" rather than an "event value", this index can achieve rolling tracking of periodic and non-sudden fire trends, and promote the transformation of fire alarm management to prediction and feedforward.

[0171] Once the system determines that the Rfth exceeds the fire threshold Trf, it immediately starts a full-linkage response mechanism, covering four levels of sound and light warning, image guidance, remote information synchronization and data sealing, effectively avoiding the signal lag problem of traditional one-way alarm in high noise and high dispersion scenes, ensuring that the fire disposal instructions cover multiple dimensions of "people, machines and terminals", and enhancing the response ability and decision execution efficiency of the system in emergency situations.

[0172] While embodiments of the present application have been shown and described with reference to particular embodiments thereof, it will be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1. A smart building fire safety early warning system based on an internet platform, characterized in that: It includes a building behavior trajectory modeling module, a nonlinear behavior deviation identification module, an energy load anomaly inversion module, a correlation logic conflict detection module, a potential fire reasoning and assessment module, and an early warning issuance and instruction feedback module; The building behavior trajectory modeling module is based on the building's existing Internet connection system, collects the building's data parameters, fits them into a building dataset, and establishes a behavior reference benchmark model within a time window to obtain behavior benchmark values; The nonlinear behavior deviation identification module combines the behavior baseline value with the building dataset to obtain the behavior deviation index; The energy load anomaly inversion module analyzes the building dataset to obtain the average load usage per person per unit, evaluates the deviation from the average load level, and obtains the energy load anomaly index. The correlation logic conflict detection module integrates the behavioral deviation index and the energy load anomaly index to calculate and obtain the logic conflict value. The potential fire reasoning and assessment module analyzes logical conflict values ​​and, in conjunction with the building inactivity status, calculates and obtains the fire potential index. The early warning and instruction feedback module compares the acquired fire potential index with the preset fire threshold to determine the fire status.

2. The intelligent building fire safety early warning system based on an internet platform according to claim 1, characterized in that: The building behavior trajectory modeling module includes a data structure normalization processing unit and a behavior modeling and benchmark fitting unit; The data structure normalization processing unit collects data parameters from the building's existing Internet connection system, including the number of people entering An, elevator call frequency EL, lighting usage cycle Lm, and average floor load power Ap, and fits them into the original dataset YW. The number of people entering, An, is obtained by collecting access control card swipe and facial recognition records through the networked access control system. Elevator call frequency (EL) is obtained by collecting elevator dispatch request events recorded in the elevator control system. The lighting usage cycle Lm is obtained through the lighting control system to determine the cumulative duration of the circuit being turned on. The average load power Ap of each floor is obtained by acquiring the real-time power of the power distribution branches on each floor through the power distribution monitoring system and then performing a weighted average. The original dataset YW was filtered to remove noise and then normalized to obtain the building dataset LW. The building dataset LW is obtained using the following formula: In the formula, LWo represents the o-th data in the building dataset LW, YWo represents the o-th data in the original dataset YW, minYWo represents the valley value of the o-th data in the original dataset YW, and maxYWo represents the peak value of the o-th data in the original dataset YW.

3. The intelligent building fire safety early warning system based on an internet platform according to claim 2, characterized in that: Within the time window Tw, the behavioral modeling and benchmark fitting unit performs statistical distribution modeling on the building dataset LW and extracts the behavioral benchmark value ABS of the normal behavioral state of the building through median analysis. The behavioral baseline (ABS) is obtained using the following formula: In the formula, Med(An) represents the median value of the number of people entering An within the time window Tw, Med(EL) represents the median value of the elevator call frequency EL within the time window Tw, Med(Lm) represents the median value of the lighting usage cycle Lm within the time window Tw, and Med(Ap) represents the median value of the average floor load power Ap within the time window Tw.

4. The intelligent building fire safety early warning system based on an internet platform according to claim 1, characterized in that: The nonlinear behavior deviation identification module includes a multi-dimensional behavior projection calculation unit and a comprehensive behavior discrimination unit; The multidimensional behavior projection calculation unit extracts behavior parameters from the building dataset LW, including the number of people entering An, the frequency of elevator calls EL, and the lighting usage cycle Lm, and projects them into the difference space of the reference ABS to obtain the personnel projection deviation function DAn and the lighting projection deviation function DLm. The personnel projection deviation function DAn is obtained as follows: extract the average level of the number of people entering An within a set historical time window, i.e., the average number of people entering; then, subtract the number of people entering An(t) at time t from the average number of people entering to obtain the original offset; finally, divide the original offset by 1 and add the elevator call frequency to obtain the personnel projection deviation function DAn. The lighting projection deviation function DLm is obtained as follows: extract the expected value of the lighting usage period Lm from the historical data; then, subtract the lighting usage period Lm(t) at time t from the historical expected value to obtain the offset difference; finally, divide the offset difference by "1 plus the current number of people entering" to obtain the lighting projection deviation function DLm.

5. The intelligent building fire safety early warning system based on an internet platform according to claim 1, characterized in that: The integrated behavior discrimination unit compares the acquired personnel projection deviation function DAn and lighting projection deviation function DLm with the behavior baseline value ABS to obtain the behavior deviation index Pab and make a judgment on the behavior status of the building. The behavioral deviation index Pab is obtained using the following formula: In the formula, Pab(t) represents the behavior deviation index at time t, ln represents the logarithmic function, DAn(t) represents the personnel projection deviation function at time t, DLm(t) represents the lighting projection deviation function at time t, and ABS(t) represents the behavior baseline value at time t. The building's behavioral status is obtained through matching in the following ways: When the behavioral deviation index Pab < 0.8, it indicates normal behavior; When 0.8 ≤ behavioral deviation index Pab < 1.5, it indicates an abnormality; When the behavior deviation index Pab ≥ 1.5, it indicates a deviation and triggers an alarm.

6. The intelligent building fire safety early warning system based on an internet platform according to claim 1, characterized in that: The energy load anomaly inversion module includes a unit load intensity analysis unit and an average deviation fitting discrimination unit; The unit load intensity analysis unit analyzes the building dataset LW, extracts the average floor load power Ap and the number of people entering An, and calculates the unit load intensity Qunit. The method for obtaining the unit personnel load intensity Qunit is as follows: First, obtain the average load power value of the building at the current time point; then, extract the number of people An actually entering the building at the current time from the access control system; finally, divide the obtained average load power value by the number of people An plus one, which is the unit personnel load intensity Qunit. The mean deviation fitting discriminant unit compares the obtained unit load intensity Qunit with the average load level μ(Ap) in the time window Tw to obtain the energy load anomaly index Pene. The Pene energy load anomaly index is obtained using the following formula: In the formula, E represents a non-zero constant.

7. A smart building fire safety early warning system based on an internet platform according to claim 6, characterized in that: The associated logical conflict detection module includes an anomaly index cross-integration unit and an atypical logical conflict identification unit; The anomaly index cross-integration unit merges the acquired behavioral deviation index Pab and energy load anomaly index Pene, and constructs a combined expression of the two through a mathematical structure to obtain the logical conflict value Cerr. The logical conflict value Cerr is obtained using the following formula: Cerr(t)=Pab(t)×ln(1+Pene(t)); In the formula, Cerr(t) represents the logical conflict value at time t, Pab(t) represents the behavioral deviation index at time t, Pene(t) represents the energy load anomaly index at time t, and ln represents the logarithmic function.

8. The intelligent building fire safety early warning system based on an Internet platform according to claim 7, characterized in that: The atypical logic conflict identification unit compares the fused logic conflict value Cerr with the dynamic threshold Terr to determine whether it meets the characteristics of atypical operating state. The dynamic threshold Terr is obtained as follows: Extract the logical conflict values ​​Cerr from the past 72 hours of system operation history and construct a historical sequence set; Statistical features are extracted from the historical sequence set, and the logical median (MC) and logical standard deviation (σC) are calculated. Based on the obtained median MC and logical standard deviation σC, the dynamic threshold Terr is calculated. The dynamic threshold Terr is obtained using the following formula: Terr = MC + 1.5 × σC; Running status characteristics are obtained through matching in the following ways: When the logical conflict value Cerr < the dynamic threshold Terr, it indicates that the building is in an acceptable state of coupling fluctuation. When the logical conflict value Cerr is greater than or equal to the dynamic threshold Terr, it indicates that the atypical logical anomaly identification flag is triggered and the node is marked as a potential high-risk node.

9. A smart building fire safety early warning system based on an internet platform according to claim 8, characterized in that: The potential fire reasoning and assessment module collects and normalizes the number of vibration response events Na, the number of action responses Nb, and the number of airflow disturbances Nc through acoustic sensors, human-sensing infrared devices, and airflow sensors, and calculates and obtains the silence index Qsi. The inactivity index Qsi is obtained using the following formula: In the formula, Qsi(t) represents the quietness index at time t, Na(t) represents the number of vibration response events at time t, Nb(t) represents the number of action responses at time t, Nc(t) represents the number of airflow disturbances at time t, Nto represents the sum of the maximum number of triggers of the acoustic sensor, human infrared sensor and airflow sensor, and Eo represents a non-zero constant. The logical conflict value Cerr and the inactivity index Qsi are fused together to calculate the fire potential index Rfth; The fire potential index Rfth is obtained using the following formula:

10. A smart building fire safety early warning system based on an internet platform according to claim 9, characterized in that: The early warning and instruction feedback module analyzes the fire potential index Rfth and compares it with the preset fire threshold TRF to determine the fire status. Fire status is obtained through matching in the following ways: When the fire potential index Rfth < the fire threshold Trf, it indicates that the fire situation is normal and monitoring should be maintained. When the fire potential index Rfth is greater than or equal to the fire threshold Trf, it indicates that the fire situation is abnormal and the warning response state is entered. When an early warning response is initiated, a coordinated response is sent to internal building equipment, external platforms, and system logs to construct a closed-loop operation chain for fire emergency response, including triggering, guiding, reporting, and sealing operations. Emergency alert tone trigger: Activate the warning tone channel through the building broadcast system to issue a standard fire alarm voice and buzzer signal; Visualized escape guidance: Switch to evacuation route guidance maps on the visual displays in the corridors and lobby areas; Multi-terminal synchronous information push: Utilize internet interfaces / APIs to push fire status information to the following terminals: fire control host; property remote monitoring system; designated management personnel's mobile app; Historical data latching mechanism activated: all building data parameters within the current time period are timestamped to form a "fire event log package", which is then uploaded to the cloud and local log area.

Images