A fire-fighting equipment fault alarm method and system based on multi-sensor fusion

Abstract

The application relates to the technical field of fire safety monitoring, and relates to a fire-fighting equipment fault alarm method and system based on multi-sensor fusion, which comprises the following steps: obtaining an equipment operation file based on a fire-fighting equipment basic file, regional topological data and collection configuration, controlling preset fire-fighting equipment to perform data collection based on the equipment operation file, adding an identifier according to an equipment number, obtaining an original sampling frame, performing a queue write operation on the standardized state vector to obtain an actual state queue, obtaining a single fault feature index according to the standardized state vector and an equipment type, obtaining a sensor correlation state set based on the regional topological relationship, and determining a preliminary alarm judgment according to the fault confidence and an alarm threshold. The application can realize multi-sensor fusion monitoring and accurate alarm of the running state of fire-fighting equipment.

Description

Technical Field

[0001] This invention relates to the field of fire safety monitoring technology, and in particular to a method and system for alarming fire equipment faults based on multi-sensor fusion. Background Technology

[0002] Fire equipment fault monitoring is a crucial component of building fire safety assurance systems. Its accuracy and timeliness of response directly impact fire early warning capabilities and the safety of personnel and property. With the continuous growth in building size and complexity, the number of fire equipment deployed in individual buildings has expanded from dozens to thousands, encompassing various types such as smoke detectors, heat detectors, and fire water pressure sensors. This places higher demands on equipment status monitoring and fault identification capabilities.

[0003] Currently, fire equipment fault detection mainly relies on a combination of regular inspections and fixed threshold alarms. It typically focuses only on whether the instantaneous value of a single sensor exceeds its limit, making it difficult to detect early signs of faults such as slow performance degradation or intermittent anomalies. Because data from different types of sensors are independent, it's impossible to distinguish between equipment malfunctions and data fluctuations caused by environmental changes through cross-reference of multi-source information, leading to a high false alarm rate and the potential for overlooking genuine faults.

[0004] Existing solutions lack cumulative analysis over time. Occasional anomalies from a single sampling can trigger alarms, while persistent slight deviations are ignored because they do not reach fixed thresholds. Furthermore, it is difficult to differentiate between single-point faults and regional environmental anomalies by considering the spatial topology of the area where the equipment is located. Therefore, how to achieve fusion analysis of multi-type sensor data, trend prediction of equipment status, and accurate fault-based alarms has become a pressing technical problem to be solved in the operation and maintenance management of fire protection equipment. Summary of the Invention

[0005] This invention provides a method for alarming faults in fire-fighting equipment based on multi-sensor fusion and a computer-readable storage medium. Its main purpose is to achieve multi-sensor fusion monitoring and accurate alarm of the operating status of fire-fighting equipment.

[0006] To achieve the above objectives, the present invention provides a method for fire equipment fault alarm based on multi-sensor fusion, comprising: Acquire basic fire equipment files, regional topology data, and acquisition configuration. Based on the basic fire equipment files, regional topology data, and acquisition configuration, acquire equipment operation files, which include equipment number, equipment type, regional topology relationship, and alarm threshold. Data is collected from preset fire-fighting equipment based on equipment operation records, and an identifier is added according to the equipment number to obtain the original sampling frame. The original sampling frame is then subjected to multi-dimensional verification to obtain valid sampling data. The effective sampled data is time-aligned to obtain a device state vector, and the device state vector is then standardized to obtain a standardized state vector. A queue writing operation is performed on the standardized state vector to obtain an actual state queue. When the actual state queue meets the preset queue length condition, a predicted state value is generated based on the actual state queue, and a predicted residual index is calculated based on the standardized state vector and the predicted state value. A single fault characteristic index is obtained based on the standardized state vector and the device type, and a sensor associated state set is obtained based on the regional topology relationship. A sensor consistency index is generated based on the sensor associated state set. Time window writing is performed on the individual fault feature index, prediction residual index and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused and calculated to obtain the fault confidence. The initial alarm judgment is determined based on the fault confidence and alarm threshold. The initial alarm judgment is normal, warning or fault. When the initial alarm judgment is warning or fault, the fault type and alarm level are confirmed based on window-level fusion features, and the alarm event is obtained and pushed to the preset operation and maintenance terminal.

[0007] Optionally, the step of performing multi-dimensional verification on the original sampled frame to obtain valid sampled data includes: Perform field integrity verification on the original sampled frame to obtain missing field markers, where the missing field marker indicates whether the field is missing or not. When the missing field is marked as not missing, the original sampled frame is checked for sequence number to obtain a duplicate mark. The duplicate mark is either duplicate or not duplicate. If the duplicate mark is duplicate, the original sampled frame is discarded. Otherwise, the original sampled frame is checked for time validity to obtain a delay mark. The delay mark is either timed out or not timed out. When the time delay mark is not exceeded, the original sampling frame is range checked to obtain the boundary mark, which is either out of bounds or not out of bounds; When the boundary crossing mark is not crossed, the original sampled frame is subjected to outlier smoothing to obtain suspicious value marks and smoothed values, where suspicious value marks are either suspicious or normal. The missing field marker, delay marker, out-of-bounds marker, suspicious value marker, and smoothing value are combined with the original sampling frame to obtain valid sampling data.

[0008] Optionally, the step of time-aligning the valid sampled data to obtain a device state vector, and then standardizing the device state vector to obtain a standardized state vector, includes: The current fusion time slice is determined based on the acquisition configuration, and the valid sampled data falling into the current fusion time slice is queried from the valid sampled data to obtain the valid sampled data of the current slice and the query status, wherein the query status is either present or absent; If the query status is "existing", then the current slice status data is obtained based on the current slice's valid sampled data, missing field marker, delay marker, out-of-bounds marker, suspicious value marker, and smoothing value. Otherwise, a missing test flag is obtained, and missing test status data is acquired based on the device number, device type, current fusion time slice, and missing test flag; The current chip status data or missing test status data is determined as the device status vector; The device state vector is standardized to obtain standardized state data; The standardized state data, missing field markers, delay markers, out-of-bounds markers, suspicious value markers, or missing test markers are combined to obtain a standardized state vector.

[0009] Optionally, the step of performing a queue writing operation on the standardized state vector to obtain an actual state queue, and when the actual state queue meets a preset queue length condition, generating a predicted state value based on the actual state queue, and calculating the predicted residual exponent based on the standardized state vector and the predicted state value, includes: Obtain the historical state queue, write the standardized state vector into the historical state queue, and obtain the actual state queue; The actual state of the queue is determined based on the queue length condition to obtain the queue state, wherein the queue state is either satisfied or not satisfied; When the queue state is not satisfied, a predicted not started flag is generated, and the predicted not started flag is associated with the standardized state vector to obtain a predicted unavailable state node. Otherwise, the predicted state value is obtained based on the actual state queue; By comparing the difference between the standardized state vector and the predicted state value, the predicted residual data is obtained; The prediction residual index is calculated based on the prediction residual data.

[0010] Optionally, the step of calculating the prediction residual exponent based on the prediction residual data includes: The predicted residual data is extracted item by item to obtain multiple predicted residual values; Multiple residual tolerance thresholds are obtained, wherein the multiple residual tolerance thresholds correspond one-to-one with the multiple predicted residual values; Multiple prediction residual indices are calculated based on the multiple prediction residual values ​​and multiple residual tolerance thresholds, wherein each prediction residual index corresponds one-to-one with a prediction residual value, and the calculation formula is as follows:

[0011] in, This represents the j-th prediction residual index among multiple prediction residual indices. This represents the j-th prediction residual value among multiple prediction residual values. This represents the residual tolerance threshold corresponding to the j-th predicted residual value among multiple predicted residual values, and Greater than 0.

[0012] Optionally, the step of obtaining a single fault characteristic index based on the standardized state vector and device type, obtaining a sensor associated state set based on the regional topology, and generating a sensor consistency index based on the sensor associated state set includes: Based on the device type, the standardized state vector is extracted into multiple state component data. Anomaly feature calculations are performed on the multiple state item data to obtain multiple individual fault feature values; Multiple individual fault characteristic values ​​are combined to obtain an individual fault characteristic index; Based on the regional topological relationships, the standardized state vectors are correlated and extracted to obtain the sensor associated state set; Statistical processing is performed on the sensor-associated state set to obtain sensor-associated statistical data; The sensor consistency index is obtained by comparing the difference between the standardized state vector and the sensor-related statistical data.

[0013] Optionally, the step of writing the individual fault feature index, prediction residual index, and sensor consistency index into a time window to obtain window-level fusion features, and then performing fusion calculation on the window-level fusion features to obtain the fault confidence, includes: The individual fault characteristic index, the predicted residual index, the predicted unavailable state node, and the sensor consistency index are combined to obtain the current abnormal characteristic data. The current abnormal feature data is written into a preset sliding time window to obtain a time window feature node, wherein the time window feature node includes the time window abnormal feature data and the time window writing order; Statistical processing is performed on the individual fault feature indices in the time window anomaly feature data to obtain individual fault window features; Statistical processing is performed on the prediction residual index in the time window anomaly feature data to obtain the prediction residual window feature; Statistical processing is performed on the sensor consistency index in the time window anomaly feature data to obtain the sensor consistency window feature; The single fault window feature, the prediction residual window feature, and the sensor consistency window feature are combined to obtain the window-level fusion feature; The fault confidence is obtained by performing fusion calculation based on the window-level fusion features.

[0014] Optionally, the step of statistically processing the prediction residual index in the time window anomaly feature data to obtain the prediction residual window feature includes: The predicted residual index in the time window anomaly feature data is extracted by sub-item extraction within the time window to obtain a residual index sequence, wherein the residual index sequence consists of multiple residual index values ​​arranged in the writing order of the time window; The largest prediction residual index is extracted from multiple prediction residual indices, and the largest prediction residual index is used as the residual peak feature. The mean value of the multiple predicted residual indices is calculated to obtain the residual mean characteristic. Obtain the residual statistical threshold and the sudden increase judgment threshold, and perform threshold exceedance statistics on the multiple predicted residual indices based on the residual statistical threshold to obtain the residual threshold exceedance frequency characteristics; Extract the residual exponent values ​​sequentially from the residual exponent sequence, and perform the following operations on the extracted residual exponent values: Adjacent residual index values ​​are identified based on the residual index values, wherein adjacent residual index values ​​are adjacent and lag behind the residual index values. Calculate the difference between adjacent residual index values ​​and residual index values ​​to obtain the residual index difference value, where adjacent residual index values ​​are minuends and residual index values ​​are subtrahends; The residual index difference is judged and summarized based on the sudden increase judgment threshold to obtain the set of difference values ​​exceeding the threshold; By counting the number of out-of-threshold differences in the set of out-of-threshold differences, the characteristic of the number of residual bursts can be obtained; The residual peak value feature, residual mean value feature, residual over-threshold frequency feature, and residual surge frequency feature are combined to obtain the prediction residual window feature.

[0015] Optionally, the step of performing fusion calculation based on the window-level fusion features to obtain the fault confidence includes: The single fault window feature, the prediction residual window feature, and the sensor consistency window feature in the window-level fusion feature are normalized respectively to obtain the single fault comprehensive value, the prediction residual comprehensive value, and the sensor consistency comprehensive value. The fault confidence level is obtained by calculating the mean of the individual fault comprehensive value, the prediction residual comprehensive value, and the sensor consistency comprehensive value.

[0016] To achieve the above objectives, the present invention also provides a fire equipment fault alarm system based on multi-sensor fusion, comprising: The data acquisition and verification module is used to acquire basic files of fire-fighting equipment, regional topology data and acquisition configuration, and to acquire equipment operation files based on the basic files of fire-fighting equipment, regional topology data and acquisition configuration. The equipment operation files include equipment number, equipment type, regional topology relationship and alarm threshold. Data is collected from preset fire-fighting equipment based on equipment operation records, and an identifier is added according to the equipment number to obtain the original sampling frame. The original sampling frame is then subjected to multi-dimensional verification to obtain valid sampling data. The state standardization prediction module is used to time-align the effective sampled data to obtain a device state vector, and to standardize the device state vector to obtain a standardized state vector. A queue writing operation is performed on the standardized state vector to obtain an actual state queue. When the actual state queue meets the preset queue length condition, a predicted state value is generated based on the actual state queue, and a predicted residual index is calculated based on the standardized state vector and the predicted state value. The feature fusion calculation module is used to obtain a single fault feature index based on the standardized state vector and the device type, obtain a sensor associated state set based on the regional topology, and generate a sensor consistency index based on the sensor associated state set. Time window writing is performed on the individual fault feature index, prediction residual index and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused and calculated to obtain the fault confidence. The alarm judgment and push module is used to determine the preliminary alarm judgment based on the fault confidence and alarm threshold. The preliminary alarm judgment is normal, warning or fault. When the preliminary alarm judgment is warning or fault, the fault type and alarm level are confirmed based on window-level fusion features, the alarm event is obtained and pushed to the preset operation and maintenance terminal.

[0017] To address the above problems, the present invention also provides an electronic device, the electronic device comprising: Memory, storing at least one instruction; The processor executes the instructions stored in the memory to implement the above-described method for fire equipment fault alarm based on multi-sensor fusion.

[0018] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the aforementioned method for alarming faults in fire-fighting equipment based on multi-sensor fusion.

[0019] To address the problems described in the background art, this invention acquires basic fire equipment files, regional topology data, and acquisition configurations. Based on these files, it obtains equipment operation files, including equipment number, equipment type, regional topology relationships, and alarm thresholds. Based on these files, it controls preset fire equipment to collect data, adding identifiers based on equipment numbers to obtain raw sampling frames. Multi-dimensional verification is performed on these frames to obtain valid sampled data. The valid sampled data is time-aligned to obtain equipment state vectors. These vectors are then standardized to obtain standardized state vectors. A queue writing operation is performed on these standardized state vectors to obtain actual state queues. When the actual state queue meets a preset queue length condition, predicted state values ​​are generated based on the actual state queues. A predicted residual index is calculated based on the standardized state vectors and the predicted state values. This invention, by comparing actual and predicted states and normalizing them to an index of 0 to 1, can detect early anomalies deviating from normal trends. Furthermore, it obtains individual fault characteristic indices based on the standardized state vectors and equipment types. The invention obtains a sensor-associated state set based on the regional topology and generates a sensor consistency index based on the sensor-associated state set. It is evident that by calculating the individual fault feature index and the sensor consistency index separately, the invention can distinguish between single-device faults and regional environmental changes, reducing the probability of single-point false alarms. Time-window writing is performed on the individual fault feature index, prediction residual index, and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused to obtain the fault confidence level. This invention improves the ability to identify intermittent faults and equipment drift faults by fusing statistical features of multiple types of abnormal evidence within a sliding time window. A preliminary alarm judgment is determined based on the fault confidence level and alarm threshold, where the preliminary alarm judgment is normal, warning, or fault. When the preliminary alarm judgment is warning or fault, the fault type and alarm level are confirmed based on the window-level fusion features, resulting in an alarm event that is pushed to a preset maintenance terminal. This invention achieves graded alarms by comparing the fault confidence level with the graded threshold and determines the fault type by combining the feature item with the highest value in the window-level fusion features, providing maintenance personnel with clear fault location and handling suggestions. Therefore, the present invention can realize multi-sensor fusion monitoring and accurate alarm of the operating status of fire-fighting equipment. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a fire equipment fault alarm method based on multi-sensor fusion, provided in an embodiment of the present invention. Figure 2 A functional block diagram of a fire equipment fault alarm system based on multi-sensor fusion is provided in an embodiment of the present invention; Figure 3This is a schematic diagram of the structure of an electronic device that implements the fire equipment fault alarm method based on multi-sensor fusion, according to an embodiment of the present invention.

[0021] Explanation of reference numerals in the attached figures: 10. Electronic device; 11. Processor; 12. Memory; 13. Bus.

[0022] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0024] This application provides a method for fire equipment fault alarm based on multi-sensor fusion. The executing entity of this method includes, but is not limited to, at least one electronic device that can be configured to execute the method provided in this application, such as a server or a terminal. In other words, the method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0025] Reference Figure 1 The diagram shown is a flowchart illustrating a fire equipment fault alarm method based on multi-sensor fusion according to an embodiment of the present invention. In this embodiment, the fire equipment fault alarm method based on multi-sensor fusion includes: S1. Obtain basic fire equipment files, regional topology data and acquisition configuration. Based on the basic fire equipment files, regional topology data and acquisition configuration, obtain equipment operation files, wherein the equipment operation files include equipment number, equipment type, regional topology relationship and alarm threshold.

[0026] It should be explained that the basic fire equipment file is a record of basic information about fire equipment stored on a server. The regional topology data is structured data describing the spatial and logical relationships between fire equipment, including spatial adjacency, grouping of similar equipment, upstream and downstream relationships in the fire water system, power supply relationships, and communication affiliation. The acquisition configuration is a set of parameters used to control the data acquisition and processing of fire equipment. The equipment operation file is a comprehensive record of operational information obtained by integrating the basic fire equipment file, regional topology data, and acquisition configuration, used to guide subsequent data acquisition, verification, and alarm judgment. The equipment number is a unique identifier for fire equipment, used to distinguish different fire equipment nodes. The equipment type is a functional category identifier for fire equipment, such as smoke detectors, heat detectors, and manual alarm buttons. The regional topology relationship is spatial and logical association information related to the current fire equipment extracted from the regional topology data. The alarm threshold is a numerical limit extracted from the acquisition configuration, used to determine whether the equipment status is abnormal.

[0027] Understandably, the process of obtaining the equipment operation file based on the fire equipment basic file, regional topology data, and acquisition configuration is as follows: the server reads the fire equipment basic file to determine the equipment number, equipment type, and installation location attributes of each fire equipment node; then it reads the regional topology data to determine the set of neighboring fire equipment, the set of similar fire equipment, the set of upstream and downstream fire equipment, and the gateway to which the fire equipment node belongs; finally, it reads the acquisition configuration to determine the sampling period, upload period, prediction queue length, and alarm threshold of the fire equipment node, and combines the above information to form the equipment operation file.

[0028] For example, a fire monitoring system in a commercial complex deploys various fire protection devices. The server reads the basic files of the fire protection devices and obtains a smoke detector with the serial number SM-B2-001. Its installation location is on the east corridor of the 2nd floor of Building B, its rated operating range is smoke concentration from 0 to 20% per meter, and its inspection cycle is 30 days. The server reads the area topology data and determines that the set of neighboring fire protection devices for this smoke detector includes fire protection devices with serial numbers SM-B2-002 and TM-B2-001. The set of similar fire protection devices includes all smoke detectors on the 2nd floor of Building B, and their respective gateway is GW-B2-01. The server reads the acquisition configuration and determines that the sampling period is 10 seconds, the upload period is 60 seconds, the prediction queue length is 30, the warning threshold is 0.6, and the fault threshold is 0.8. After combining the above information, the device operation file for this smoke detector is obtained.

[0029] S2. Based on the equipment operation file, control the preset fire equipment to collect data, add an identifier according to the equipment number to obtain the original sampling frame, and perform multi-dimensional verification on the original sampling frame to obtain valid sampling data.

[0030] It should be explained that the "data collection based on equipment operation records" refers to the following: the server sends a fire equipment node configuration package to the edge gateway according to the sampling and upload cycles in the fire equipment operation records; the edge gateway then sends a sampling configuration to the fire equipment nodes; and the fire equipment nodes collect data according to the sampling cycles specified in the sampling configuration, generating real-time sampling data. The "adding identifiers based on equipment numbers" refers to the fire equipment nodes adding an equipment number, sampling time, configuration version number, and data sequence number to each piece of real-time sampling data. The original sampling frame is the real-time sampling data after adding the identifiers. The multi-dimensional verification is a process of checking the data quality of the original sampling frame from multiple dimensions. The valid sampling data is the sampling data retained after multi-dimensional verification, with accompanying verification marker information.

[0031] Understandably, to ensure the reliability and integrity of the data entering subsequent state calculations, it is necessary to verify the original sampling frame from multiple dimensions, including field integrity, data repeatability, time validity, range reasonableness, and outlier smoothing. Therefore, the multi-dimensional verification of the original sampling frame to obtain valid sampling data includes: Perform field integrity verification on the original sampled frame to obtain missing field markers, where the missing field marker indicates whether the field is missing or not. When the missing field is marked as not missing, the original sampled frame is checked for sequence number to obtain a duplicate mark. The duplicate mark is either duplicate or not duplicate. If the duplicate mark is duplicate, the original sampled frame is discarded. Otherwise, the original sampled frame is checked for time validity to obtain a delay mark. The delay mark is either timed out or not timed out. When the time delay mark is not exceeded, the original sampling frame is range checked to obtain the boundary mark, which is either out of bounds or not out of bounds; When the boundary crossing mark is not crossed, the original sampled frame is subjected to outlier smoothing to obtain suspicious value marks and smoothed values, where suspicious value marks are either suspicious or normal. The missing field marker, delay marker, out-of-bounds marker, suspicious value marker, and smoothing value are combined with the original sampling frame to obtain valid sampling data.

[0032] It should be understood that the field integrity check is the process of checking whether the original sampling frame is missing a device number, sampling time, or core sampling field. The sequence number check is the process of checking whether the same fire-fighting equipment has received an original sampling frame with the same data sequence number. The time validity check is the process of checking whether the sampling time of the original sampling frame exceeds the allowable delay range. The range check is the process of checking whether the numerical fields in the original sampling frame exceed the physical reach range. The outlier smoothing process is a process of statistical analysis of continuous fields in the original sampling frames that have passed the basic checks. Specifically, it involves calculating the local mean and standard deviation using the most recent valid value queue. When the current value deviates from the local mean by more than the configured threshold but does not exceed the physical range, the field is marked as a suspicious value, the original value is retained, and a smoothed value is calculated. The smoothed value is the local mean of the most recent valid value queue. The missing field marker, delay marker, out-of-bounds marker, and suspicious value marker are collectively referred to as verification markers, which are used as evidence of data quality anomalies in subsequent steps.

[0033] For example, taking a smoke detector with the serial number SM-B2-001 as an example, this smoke detector collects data at a sampling period of 10 seconds, generating a real-time sampling data point, including a smoke concentration of 2.5% per meter, a temperature of 26 degrees Celsius, a battery level of 85%, a signal strength of -65dBm, and a packet loss rate of 0.02%. The fire equipment node adds the device number SM-B2-001, sampling time 2025-03-15-10:30:00, configuration version number V3.2, and data sequence number 10086 to this data, forming an original sampling frame. After receiving this original sampling frame, the edge gateway performs multi-dimensional verification: first, it performs a field integrity verification to confirm that the device number, sampling time, and core sampling fields are all present, and marks missing fields as not missing. Then, it performs a sequence number verification to confirm that data sequence number 10086 has not been received repeatedly, and marks duplicates as not duplicates. Next, a time validity check is performed. The difference between the sampling time and the current time is 3 seconds, which is within the allowable delay range of 60 seconds, and the delay is marked as not exceeding the timeout. Then, a range check is performed. The smoke concentration of 2.5% per meter is within the rated operating range of 0 to 20% per meter, and the temperature of 26 degrees Celsius is within the range of -10 to 80 degrees Celsius, and out-of-range values ​​are marked as not exceeding the range. Finally, outlier smoothing is performed. Using the most recent 10 valid values, the local mean is calculated to be 2.3% per meter, and the standard deviation is 0.4% per meter. The current value of 2.5% per meter deviates from the local mean by 0.2% per meter, which is within the configured threshold (3 times the standard deviation, i.e., 1.2% per meter), and the suspicious value is marked as normal. The above check marks are combined with the original sampling frame to obtain the valid sampling data.

[0034] S3. Time-align the effective sampled data to obtain a device state vector, and standardize the device state vector to obtain a standardized state vector.

[0035] It should be explained that the time alignment is the process of uniformly grouping valid sampled data collected at different times into the same fusion time slice, where the fusion time slice is a time interval divided according to the fusion cycle in the acquisition configuration. The device state vector is a vector composed of all fields of the same fire-fighting equipment within the same fusion time slice. The standardization process is the process of converting continuous fields in the device state vector into a uniform dimension, making different types of numerical fields comparable. The standardized state vector is the device state vector after standardization.

[0036] Understandably, since the sampling times of different fire-fighting equipment may vary, it is necessary to unify the valid sampling data to the same time reference for processing. Therefore, the process of aligning the valid sampling data by time to obtain the equipment state vector, and then standardizing the equipment state vector to obtain the standardized state vector, includes: The current fusion time slice is determined based on the acquisition configuration, and the valid sampled data falling into the current fusion time slice is queried from the valid sampled data to obtain the valid sampled data of the current slice and the query status, wherein the query status is either present or absent; If the query status is "existing", then the current slice status data is obtained based on the current slice's valid sampled data, missing field marker, delay marker, out-of-bounds marker, suspicious value marker, and smoothing value. Otherwise, a missing test flag is obtained, and missing test status data is acquired based on the device number, device type, current fusion time slice, and missing test flag; The current chip status data or missing test status data is determined as the device status vector; The device state vector is standardized to obtain standardized state data; The standardized state data, missing field markers, delay markers, out-of-bounds markers, suspicious value markers, or missing test markers are combined to obtain a standardized state vector.

[0037] It should be understood that the current fusion time slice is a time interval established by the edge gateway according to the fusion period in the acquisition configuration. For example, if the fusion period is 60 seconds, then every 60 seconds is a fusion time slice. The current slice's valid sampled data is the latest valid sampled data belonging to the same fire-fighting equipment received within the current fusion time slice. The query status is used to indicate whether valid sampled data for the fire-fighting equipment exists within the current fusion time slice. The current slice status data is status data obtained by combining the values ​​of each field in the current slice's valid sampled data with the corresponding verification flags. The missing test flag is a flag generated when a fire-fighting equipment has not received new valid sampled data within the current fusion time slice. The missing test status data is status data composed of the equipment number, equipment type, current fusion time slice number, and missing test flag.

[0038] Furthermore, the specific method of the standardization process is as follows: The edge gateway performs normalization calculations on continuous fields in the device state vector based on the device type and rated operating range in the device operation file. Values ​​such as water pressure, voltage, current, battery charge, temperature, and smoke concentration are converted into relative state values ​​between 0 and 1. The conversion method is to subtract the rated lower limit from the current value and then divide by the difference between the rated upper and lower limits. For discrete fields, the edge gateway performs state encoding, converting valve opening / closing status, pump start / stop status, relay status, self-test results, and online status into corresponding numerical state codes.

[0039] For example, taking a smoke detector with the serial number SM-B2-001 as an example, the fusion period in the acquisition configuration is 60 seconds, and the current fusion time slice is the 100th time slice, corresponding to the time interval from 2025-03-15-10:30:00 to 2025-03-15-10:30:59. The edge gateway queries the valid sampled data for data falling within this time slice, and finds valid sampled data with a sampling time of 2025-03-15-10:30:00, with a query status of "existing". Based on the valid sampled data of the current slice and its verification flags, the current slice status data is obtained, including smoke concentration of 2.5% per meter, temperature of 26 degrees Celsius, battery power of 85%, signal strength of -65dBm, packet loss rate of 0.02%, and all verification flags being normal. The current slice status data is determined as the device status vector. The equipment state vector is standardized as follows: the rated operating range of smoke concentration is 0 to 20 percent per meter, and the standardized value is 2.5 divided by 20, which equals 0.125; the rated operating range of temperature is -10 to 80 degrees Celsius, and the standardized value is (26 minus (-10)) divided by (80 minus (-10)) equals 36 divided by 90, which equals 0.4; the standardized value of battery charge is 85 divided by 100, which equals 0.85. The standardized state data is then combined with each calibration marker to obtain the standardized state vector.

[0040] S4. Perform a queue writing operation on the standardized state vector to obtain the actual state queue. When the actual state queue meets the preset queue length condition, generate the predicted state value based on the actual state queue, and calculate the predicted residual index based on the standardized state vector and the predicted state value.

[0041] It should be explained that the queue write operation is the process of writing the standardized state vector of the current fusion time slice into the historical state queue of the fire-fighting equipment. The actual state queue is a first-in, first-out queue established for each fire-fighting equipment to store the standardized state data of the most recent fusion time slices. The queue length condition is the preset prediction queue length in the acquisition configuration; prediction calculation is only initiated when the number of data entries stored in the actual state queue reaches this prediction queue length. The predicted state value is the expected state value of the next fusion time slice generated using a short-term prediction model based on the historical standardized state data in the actual state queue. The prediction residual index is a quantitative indicator of the degree of difference between the standardized state vector of the current fusion time slice and the predicted state value generated in the previous fusion time slice, used to measure the degree to which the actual state deviates from the predicted state.

[0042] Understandably, by establishing a historical state queue and performing short-term predictions, early faults that have deviated from the normal trend but have not reached a fixed threshold can be identified. Therefore, the process of performing a queue writing operation on the standardized state vector to obtain an actual state queue, and generating a predicted state value based on the actual state queue when the actual state queue meets a preset queue length condition, and calculating the prediction residual exponent based on the standardized state vector and the predicted state value, including: Obtain the historical state queue, write the standardized state vector into the historical state queue, and obtain the actual state queue; The actual state of the queue is determined based on the queue length condition to obtain the queue state, wherein the queue state is either satisfied or not satisfied; When the queue state is not satisfied, a predicted not started flag is generated, and the predicted not started flag is associated with the standardized state vector to obtain a predicted unavailable state node. Otherwise, the predicted state value is obtained based on the actual state queue; By comparing the difference between the standardized state vector and the predicted state value, the predicted residual data is obtained; The prediction residual index is calculated based on the prediction residual data.

[0043] It should be understood that the historical state queue is a first-in, first-out queue pre-established by the edge gateway for each fire-fighting device, and the queue length is determined by the prediction queue length in the data acquisition configuration. The process of writing the standardized state vector into the historical state queue is as follows: the continuous standardized field in the standardized state vector is written into the historical state queue of the corresponding fire-fighting device. If the historical state queue is full, the data from the earliest fusion time slice is deleted, and then the data from the current fusion time slice is written, forming an updated actual state queue. The queue status is used to identify whether the number of data entries in the actual state queue has reached the prediction queue length. The prediction not started flag is a flag generated when the queue status is not satisfied, indicating that prediction calculation cannot be performed at present, and the prediction residual index is not used in subsequent fusion steps. The prediction unavailable state node is a state record obtained by associating the prediction not started flag with the current standardized state vector, used to indicate that the prediction residual index is unavailable in subsequent fusion calculations.

[0044] Furthermore, the process of obtaining the predicted state value based on the actual state queue is as follows: the edge gateway establishes a short-term prediction model for each continuous field in the actual state queue, generating the predicted state value for the next fusion time slice. The short-term prediction model can employ an exponentially weighted moving average method, specifically: for a historical standardized value sequence of a certain continuous field in the actual state queue, increasing exponential weights are assigned according to time from oldest to newest, and the weighted average is calculated as the predicted state value for the next fusion time slice. The predicted residual data is the difference between the standardized value of each continuous field in the standardized state vector of the current fusion time slice and the corresponding predicted state value generated in the previous fusion time slice.

[0045] Specifically, the calculation of the prediction residual index based on the prediction residual data includes: The predicted residual data is extracted item by item to obtain multiple predicted residual values; Multiple residual tolerance thresholds are obtained, wherein the multiple residual tolerance thresholds correspond one-to-one with the multiple predicted residual values; Multiple prediction residual indices are calculated based on the multiple prediction residual values ​​and multiple residual tolerance thresholds, wherein each prediction residual index corresponds one-to-one with a prediction residual value, and the calculation formula is as follows:

[0046] in, Represents the first of multiple prediction residual indices One predicted residual index, Represents the first of multiple predicted residual values Each predicted residual value, Represents the first of multiple predicted residual values The residual tolerance threshold corresponding to each predicted residual value, and Greater than 0.

[0047] Understandably, the extraction of prediction residual data involves splitting the prediction residual data item by item according to continuous fields. Each continuous field corresponds to a prediction residual value; for example, the smoke concentration field corresponds to one prediction residual value, and the temperature field corresponds to one prediction residual value. The multiple prediction residual values ​​are the set of differences between the current standardized value and the corresponding predicted state value of each continuous field. The multiple residual tolerance thresholds are the upper limits of the allowed prediction deviation for each continuous field. The residual tolerance thresholds are set as follows: based on the historical fluctuation characteristics of each field, the upper bound of the fluctuation range of the standardized value of that field under normal operating conditions is taken as the residual tolerance threshold. For example, the fluctuation range of the standardized value of the smoke concentration field under normal conditions is 0 to 0.1, so the residual tolerance threshold for that field is set to 0.1. The prediction residual index ranges from 0 to 1. When the absolute value of the prediction residual value divided by the residual tolerance threshold is greater than or equal to 1, the prediction residual index is set to 1, indicating that the actual state of the field has completely deviated from the predicted state.

[0048] For example, taking a smoke detector with the serial number SM-B2-001 as an example, the prediction queue length in the acquisition configuration is 30. Assuming the current time slice is the 100th fusion time slice, the historical state queue of this smoke detector has stored 30 standardized state data entries, and the queue state is satisfied. The edge gateway calculates the predicted state value for the 100th fusion time slice as 0.12 based on the 30 historical standardized values ​​of the smoke concentration field in the actual state queue using an exponentially weighted moving average method. The standardized smoke concentration value in the current 100th fusion time slice's standardized state vector is 0.125, and the difference between this and the predicted state value of 0.12 is 0.005, i.e., the prediction residual value is 0.005. The residual tolerance threshold for the smoke concentration field is 0.1. According to the formula, the prediction residual exponent is min(1, 0.005 divided by 0.1) equal to min(1, 0.05), which equals 0.05. This indicates that the deviation between the actual state and the predicted state of the smoke concentration field is small and within the normal range. Similarly, prediction residual indices are calculated for the temperature field and the battery charge field respectively, resulting in multiple prediction residual indices. This embodiment of the invention, by comparing the actual state with the predicted state and normalizing them to an index ranging from 0 to 1, can detect early anomalies deviating from the normal trend.

[0049] S5. Obtain a single fault characteristic index based on the standardized state vector and device type, and obtain a sensor associated state set based on the regional topology relationship. Generate a sensor consistency index based on the sensor associated state set.

[0050] It should be explained that the single-item fault characteristic index is a set of anomaly indicators calculated based on the standardized values ​​of each field in the standardized state vector, combined with the fault judgment rules corresponding to the equipment type. This set is used to characterize the anomaly degree of each state component. The sensor-associated state set is a set of standardized state vectors of other sensor devices associated with the target fire-fighting equipment, extracted from the current fusion time slice based on regional topology. The sensor consistency index is an indicator obtained by comparing the statistical characteristics between the standardized state vector of the target fire-fighting equipment and the sensor-associated state set. This index is used to determine whether the target fire-fighting equipment exhibits isolated anomalies relative to neighboring fire-fighting equipment.

[0051] It is understandable that different types of fire-fighting equipment exhibit different fault manifestations. Therefore, it is necessary to calculate the degree of abnormality for each state component based on the equipment type, and to distinguish between single-equipment faults and regional environmental changes by comparing the states of neighboring fire-fighting equipment. Thus, the process of obtaining a single fault characteristic index based on the standardized state vector and equipment type, obtaining a sensor-associated state set based on the regional topology, and generating a sensor consistency index based on the sensor-associated state set includes: Based on the device type, the standardized state vector is extracted into multiple state component data. Anomaly feature calculations are performed on the multiple state item data to obtain multiple individual fault feature values; Multiple individual fault characteristic values ​​are combined to obtain an individual fault characteristic index; Based on the regional topological relationships, the standardized state vectors are correlated and extracted to obtain the sensor associated state set; Statistical processing is performed on the sensor-associated state set to obtain sensor-associated statistical data; The sensor consistency index is obtained by comparing the difference between the standardized state vector and the sensor-related statistical data.

[0052] It should be understood that the extraction of standardized state vectors based on equipment type refers to: determining the state fields that need to be monitored for this fire protection equipment based on the equipment type, and extracting the standardized values ​​of the corresponding fields one by one from the standardized state vector. The multiple state sub-items data are the collection of the extracted standardized values ​​of each field. The anomaly feature calculation is the process of calculating the degree of anomaly for each state sub-item data according to the fault judgment rules corresponding to the equipment type. For example, for a smoke detector, the anomaly feature calculation includes: the smoke drift index is calculated from the deviation of the standardized smoke concentration value from the historical smoothing trend; the self-test anomaly index is calculated from the self-test status code; and the communication anomaly index is calculated from the missing test marker, delay marker, and packet loss rate. See the example for specific calculations. The multiple individual fault feature values ​​are the set of anomaly degree values ​​obtained after the anomaly feature calculation for each state sub-item. The individual fault feature index is the set of indices obtained by combining multiple individual fault feature values.

[0053] Further, the step of extracting standardized state vectors based on regional topology relationships refers to the edge gateway reading the sets of neighboring fire-fighting equipment, similar fire-fighting equipment, and upstream and downstream fire-fighting equipment based on the regional topology relationships in the device operation file. It then extracts the standardized state vectors of these associated fire-fighting equipment from the standardized state vectors generated in the current fusion time slice, forming a sensor-associated state set. If the sensor-associated state set is empty (i.e., no standardized state vectors have been generated for any associated fire-fighting equipment in the current fusion time slice), a neighborhood unavailable marker is generated, and the sensor consistency index is not calculated. The sensor association statistics are reference values ​​obtained by statistically calculating the standardized values ​​of the corresponding fields of each fire-fighting equipment in the sensor-associated state set. Specifically, it calculates the mean of the standardized values ​​of the corresponding fields of similar fire-fighting equipment in the same region, which is used as the neighborhood reference state. The step of comparing the difference between the standardized state vectors and the sensor association statistics refers to calculating the absolute difference between the standardized values ​​of each field in the standardized state vector of the target fire-fighting equipment and the mean of the corresponding fields in the neighborhood reference state, then dividing by the standard deviation of the neighborhood reference state for normalization, to obtain the sensor consistency index. The larger the sensor consistency index, the greater the state difference between the target fire-fighting equipment and the neighboring fire-fighting equipment, and the more likely it is a single-device failure.

[0054] For example, taking a smoke detector with the serial number SM-B2-001 as an example, the device type is a smoke detector. The edge gateway extracts the standardized state vector according to the device type, obtaining a standardized value of smoke concentration of 0.125, a self-test status code of normal (coded as 0), a missing test marker of no missing test, a delay marker of no timeout, and a standardized value of packet loss rate of 0.02. Anomaly feature calculations are performed on each state item data: the smoke drift index is calculated by dividing the absolute value of the difference between the current standardized smoke concentration value of 0.125 and the historical smoothing trend value of 0.12 (0.005) by the drift tolerance threshold of 0.15, resulting in a smoke drift index of 0.033. The self-test anomaly index is directly obtained as 0 from the self-test status code of normal. The communication anomaly index is calculated as 0.02 by combining the missing test marker of no missing test, the delay marker of no timeout, and the packet loss rate of 0.02. The smoke drift index (0.033), self-test anomaly index (0), and communication anomaly index (0.02) are combined to obtain a single fault characteristic index.

[0055] For example, the edge gateway reads the set of neighboring fire-fighting devices for SM-B2-001 based on the regional topology, including SM-B2-002 (smoke detector in the same area) and SM-B2-003 (smoke detector in the same area). It extracts the standardized state vectors of these two fire-fighting devices from the current fusion time slice to form a sensor-associated state set. Statistical processing is performed on the smoke concentration field in the sensor-associated state set. The standardized smoke concentration value for SM-B2-002 is 0.11, and for SM-B2-003 it is 0.13. The calculated mean is 0.12, and the standard deviation is 0.01. The standardized smoke concentration value for the target fire-fighting device SM-B2-001 is 0.125. The absolute difference between this and the mean of the neighboring reference state (0.12) is 0.005. Dividing this by the standard deviation (0.01) yields a sensor consistency index of 0.5, indicating that the smoke concentration state difference between the target fire-fighting device and neighboring fire-fighting devices is small, and no isolated anomalies have occurred. By calculating the individual fault characteristic index and the sensor consistency index separately, this invention can distinguish between single device faults and regional environmental changes, thereby reducing the probability of false alarms at single points.

[0056] S6. Perform time window writing on the individual fault feature index, prediction residual index and sensor consistency index to obtain window-level fusion features, and perform fusion calculation on the window-level fusion features to obtain fault confidence.

[0057] It should be explained that the time window writing is the process of writing various abnormal feature data of the current fusion time slice into a preset sliding time window. The sliding time window is a first-in, first-out window established for each fire-fighting device to store abnormal feature data from the most recent fusion time slices; the window length is determined by the sliding time window length in the acquisition configuration. The window-level fusion feature is a comprehensive feature set obtained by statistically processing the abnormal feature data of all fusion time slices within the sliding time window. The fusion calculation is a process of comprehensively evaluating the window-level fusion features. The fault confidence is a comprehensive anomaly assessment value obtained after fusion calculation; a higher value indicates a higher probability of equipment failure.

[0058] Understandably, the anomalous features of a single fusion time slice may be affected by transient interference. Therefore, it is necessary to statistically fuse the anomalous features of multiple fusion time slices within a sliding time window to improve the reliability of fault diagnosis. Thus, the step of writing the individual fault feature index, prediction residual index, and sensor consistency index into a time window to obtain window-level fusion features, and then performing fusion calculations on these window-level fusion features to obtain the fault confidence score, including: The individual fault characteristic index, the predicted residual index, the predicted unavailable state node, and the sensor consistency index are combined to obtain the current abnormal characteristic data. The current abnormal feature data is written into a preset sliding time window to obtain a time window feature node, wherein the time window feature node includes the time window abnormal feature data and the time window writing order; Statistical processing is performed on the individual fault feature indices in the time window anomaly feature data to obtain individual fault window features; Statistical processing is performed on the prediction residual index in the time window anomaly feature data to obtain the prediction residual window feature; Statistical processing is performed on the sensor consistency index in the time window anomaly feature data to obtain the sensor consistency window feature; The single fault window feature, the prediction residual window feature, and the sensor consistency window feature are combined to obtain the window-level fusion feature; The fault confidence is obtained by performing fusion calculation based on the window-level fusion features.

[0059] It should be understood that the current abnormal feature data is a comprehensive abnormal data obtained by combining the single fault feature index, the prediction residual index (if available), the prediction unavailable state node (if prediction not started), and the sensor consistency index (if available) of the current fusion time slice. The process of writing the current abnormal feature data into the sliding time window is as follows: if the sliding time window is full, the abnormal feature data of the earliest fusion time slice is deleted, and then the current abnormal feature data is written to form an updated sliding time window. The time window feature node is the basic storage unit in the sliding time window, containing the abnormal feature data of the fusion time slice and the writing order of the fusion time slice in the time window. The time window abnormal feature data is the collection of abnormal feature data of all time window feature nodes in the sliding time window. The time window writing order is the sequential numbering of each time window feature node according to its writing time from earliest to latest.

[0060] Furthermore, the process of statistically processing the individual fault feature indices in the time window anomaly feature data to obtain individual fault window features is as follows: The maximum value, mean value, and number of consecutive threshold exceedances of the individual fault feature indices for all fused time slices within the sliding time window are calculated, and the above statistical results are combined into individual fault window features. Here, the number of consecutive threshold exceedances refers to the number of fused time slices where the individual fault feature index consecutively exceeds the individual fault threshold.

[0061] Specifically, the statistical processing of the prediction residual index in the time window anomaly feature data to obtain the prediction residual window feature includes: The predicted residual index in the time window anomaly feature data is extracted by sub-item extraction within the time window to obtain a residual index sequence, wherein the residual index sequence consists of multiple residual index values ​​arranged in the writing order of the time window; The largest prediction residual index is extracted from multiple prediction residual indices, and the largest prediction residual index is used as the residual peak feature. The mean value of the multiple predicted residual indices is calculated to obtain the residual mean characteristic. Obtain the residual statistical threshold and the sudden increase judgment threshold, and perform threshold exceedance statistics on the multiple predicted residual indices based on the residual statistical threshold to obtain the residual threshold exceedance frequency characteristics; Extract the residual exponent values ​​sequentially from the residual exponent sequence, and perform the following operations on the extracted residual exponent values: Adjacent residual index values ​​are identified based on the residual index values, wherein adjacent residual index values ​​are adjacent and lag behind the residual index values. Calculate the difference between adjacent residual index values ​​and residual index values ​​to obtain the residual index difference value, where adjacent residual index values ​​are minuends and residual index values ​​are subtrahends; The residual index difference is judged and summarized based on the sudden increase judgment threshold to obtain the set of difference values ​​exceeding the threshold; By counting the number of out-of-threshold differences in the set of out-of-threshold differences, the characteristic of the number of residual bursts can be obtained; The residual peak value feature, residual mean value feature, residual over-threshold frequency feature, and residual surge frequency feature are combined to obtain the prediction residual window feature.

[0062] Understandably, the sub-item extraction within the time window involves extracting the prediction residual index of each fusion time slice sequentially from the sliding time window according to the time window's writing order, forming a residual index sequence arranged chronologically. This residual index sequence contains multiple residual index values, each corresponding to the prediction residual index of a fusion time slice. The residual peak feature is the prediction residual index with the largest value in the residual index sequence, used to reflect the peak level of the prediction residual within the sliding time window. The residual mean feature is the arithmetic mean of all prediction residual indices in the residual index sequence, used to reflect the overall level of the prediction residual within the sliding time window. The residual statistical threshold is a numerical limit used to count the number of times the prediction residual index exceeds a threshold. The residual statistical threshold is set by taking the upper bound of the higher set of values ​​from the historical data of the predicted residual index under normal operating conditions for this device type. The residual threshold exceedance frequency feature is the number of times the predicted residual index in the residual index sequence is greater than the residual statistical threshold. The sudden increase judgment threshold is a pre-set numerical limit used to determine whether the predicted residual index suddenly increases between adjacent fusion time slices. The sudden increase judgment threshold is set as follows: based on historical data of the difference between the predicted residual indexes of adjacent fusion time slices under normal operating conditions for this equipment type, the upper bound of the higher set of differences in the historical data is taken as the sudden increase judgment threshold. The adjacent residual index value is the next residual index value in the residual index sequence that is adjacent to the current residual index value and lags behind it in time. The residual index difference is the result of subtracting the current residual index value from the adjacent residual index value. The set of excess threshold differences is the set of all residual index differences greater than the sudden increase judgment threshold. The residual sudden increase frequency feature is the number of excess threshold differences in the set of excess threshold differences, used to reflect the number of times the predicted residual index suddenly increases within the sliding time window.

[0063] It should be explained that the process of statistically processing the sensor consistency index in the time window anomaly feature data to obtain the sensor consistency window feature is as follows: The maximum value and duration of the sensor consistency index for all available fusion time slices within the sliding time window are calculated, and the above statistical results are combined into the sensor consistency window feature. Here, the duration refers to the number of fusion time slices where the sensor consistency index continuously exceeds the consistency anomaly threshold. If the sensor consistency index for some fusion time slices is not generated due to unavailable neighborhood markers, these fusion time slices are skipped during the statistical analysis.

[0064] Further, the step of performing fusion calculation based on the window-level fusion features to obtain the fault confidence includes: The single fault window feature, the prediction residual window feature, and the sensor consistency window feature in the window-level fusion feature are normalized respectively to obtain the single fault comprehensive value, the prediction residual comprehensive value, and the sensor consistency comprehensive value. The fault confidence level is obtained by calculating the mean of the individual fault comprehensive value, the prediction residual comprehensive value, and the sensor consistency comprehensive value.

[0065] It should be understood that the normalization process is the process of combining multiple statistical indicators from each window feature into a comprehensive value between 0 and 1. The normalization method for the comprehensive value of a single fault is as follows: the maximum value, mean value, and number of consecutive threshold exceedances in the single fault window feature are normalized separately, and then a weighted average is taken. The maximum value and mean value are already within the range of 0 to 1, and the number of consecutive threshold exceedances is normalized by dividing by the sliding time window length. The normalization method for the comprehensive value of the prediction residual is as follows: the residual peak value, residual mean value, number of residual threshold exceedances, and number of residual spikes in the prediction residual window feature are normalized separately, and then a weighted average is taken. The residual peak value and residual mean value are already within the range of 0 to 1, and the number of residual threshold exceedances and number of residual spikes are normalized by dividing by the sliding time window length. The normalization method for the comprehensive value of sensor consistency is as follows: the maximum value in the sensor consistency window feature is normalized (treated as 1 if it exceeds 1), the duration is normalized by dividing by the sliding time window length, and then a weighted average is taken. If the predicted residual composite value or sensor consistency composite value is not generated due to unavailable markers, the corresponding item is deleted when calculating the mean, and the mean is only calculated for the generated composite values.

[0066] For example, taking a smoke detector with the serial number SM-B2-001 as an example, the sliding time window length is 10 fusion time slices. Assume that abnormal feature data for 10 fusion time slices has been written into the current sliding time window. Statistical processing of the individual fault feature indices is performed: the smoke drift indices for the 10 fusion time slices are 0.033, 0.035, 0.04, 0.038, 0.042, 0.045, 0.05, 0.055, 0.06, and 0.065, respectively, with a maximum value of 0.065, an average value of 0.046, an individual fault threshold of 0.3, and 0 consecutive times exceeding the threshold. Statistical processing was performed on the predicted residual indices: the predicted residual indices for the 10 fusion time slices were 0.05, 0.06, 0.055, 0.07, 0.065, 0.08, 0.09, 0.12, 0.15, and 0.18, respectively, with a residual peak value of 0.18 and a residual mean value of 0.092. The residual statistical threshold was set to 0.1 (determined based on a higher set of values ​​from the historical predicted residual index data under normal operating conditions of the smoke detector), and the residual exceeding the threshold was defined as 3 times (the predicted residual indices for the 8th, 9th, and 10th fusion time slices exceeded 0.1). The threshold for judging sudden increases is set to 0.05 (determined based on the higher set of differences in the historical data of the differences in the predicted residual index between adjacent fusion time slices under normal operating conditions). The differences in the residual index between adjacent fusion time slices are calculated as follows: the difference between the 7th and 8th fusion time slices is 0.12 minus 0.09 equals 0.03; the difference between the 8th and 9th fusion time slices is 0.15 minus 0.12 equals 0.03; and the difference between the 9th and 10th fusion time slices is 0.18 minus 0.15 equals 0.03. None of these exceed the threshold for judging sudden increases of 0.05, and the feature of the number of residual sudden increases is 0. The feature of the residual peak value (0.18), the feature of the residual mean value (0.092), the feature of the number of times the residual exceeds the threshold (3), and the feature of the number of times the residual increases (0) are combined to obtain the feature of the predicted residual window.

[0067] For example, the sensor consistency index is statistically processed: the sensor consistency indices for 10 fusion time slices are 0.5, 0.4, 0.6, 0.5, 0.55, 0.7, 0.8, 0.9, 1.0, and 1.1, respectively, with a maximum value of 1.1. The consistency anomaly threshold is 0.8, and the duration is 4 fusion time slices (the sensor consistency indices for the 7th to 10th fusion time slices are 0.8, 0.9, 1.0, and 1.1, respectively, all greater than or equal to the consistency anomaly threshold of 0.8). The single-fault window feature, the prediction residual window feature, and the sensor consistency window feature are combined to obtain the window-level fusion feature. Fusion calculation is performed: the normalized calculation of the single-fault comprehensive value is a weighted average of (maximum value 0.065 plus mean 0.046 plus the normalized value of the number of consecutive exceedances of the threshold, 0 divided by 10, equal to 0), and the equally weighted average is (0.065 plus 0.046 plus 0) divided by 3, equal to 0.037. The normalized value of the predicted residual composite is calculated as a weighted average of (residual peak feature 0.18 + residual mean feature 0.092 + normalized value of residual exceedance times 3 divided by 10 equals 0.3 + normalized value of residual spike times 0 divided by 10 equals 0), and the equal weighted average is (0.18 + 0.092 + 0.3 + 0) divided by 4 equals 0.143. The normalized value of the sensor consistency composite is calculated as a weighted average of (maximum value normalized to min(1, 1.1), i.e., treated as 1 when exceeding 1 equals 1 + duration normalized to 4 divided by 10 equals 0.4), and the equal weighted average is (1 + 0.4) divided by 2 equals 0.7. The fault confidence is the mean of the individual fault composite value, the predicted residual composite value, and the sensor consistency composite value, i.e., (0.037 + 0.143 + 0.7) divided by 3 equals 0.293. The embodiments of the present invention improve the ability to identify intermittent faults and equipment drift faults by fusing statistical features of multiple types of abnormal evidence within a sliding time window.

[0068] S7. Determine the preliminary alarm judgment based on the fault confidence and alarm threshold, wherein the preliminary alarm judgment is normal, warning or fault. When the preliminary alarm judgment is warning or fault, the fault type and alarm level are confirmed based on window-level fusion features, the alarm event is obtained and pushed to the preset operation and maintenance terminal.

[0069] It should be explained that the alarm thresholds include warning thresholds and fault thresholds, both of which are derived from the fused alarm thresholds in the data acquisition configuration. The preliminary alarm judgment is a device status judgment determined based on the comparison between the fault confidence level and the alarm threshold. The fault type is the fault cause category determined based on the feature item with the highest value in the window-level fused features. The alarm level is an alarm severity level determined comprehensively based on the fault confidence level, preliminary alarm judgment, fault type, device location, and device type. The alarm event is a comprehensive alarm record containing device number, device type, time slice number, installation location, fault confidence level, preliminary alarm judgment, fault type, alarm level, main evidence features, and related neighboring fire equipment numbers. The maintenance terminal is a terminal device used to receive alarm information and confirm the handling results, including the monitoring terminal in the fire control room and the mobile terminal of maintenance personnel.

[0070] For example, taking the fire water pressure sensor with serial number WP-A3-005 as an example, this sensor is installed at the end of the fire hydrant network on the 3rd floor of Building A. Assume that after sliding time window fusion calculation, the fault confidence level is 0.72. The warning threshold in the data acquisition configuration is 0.6, and the fault threshold is 0.8. Since the fault confidence level of 0.72 is greater than or equal to the warning threshold of 0.6 but lower than the fault threshold of 0.8, the initial alarm judgment is a warning. The edge gateway reads the window-level fusion features and finds that the maximum value of the insufficient water pressure index in the single fault window feature is 0.85, the average value is 0.7, and the number of consecutive exceedances of the threshold is 8. The feature with the highest value is the insufficient water pressure index in the single fault window feature, and the candidate fault cause is insufficient water pressure. Combined with the fact that the device type is a fire water pressure sensor, the fault type is identified as insufficient water pressure. According to the alarm level rules, the fault confidence level of 0.72 is within the warning range, the fault type is insufficient water pressure, and the device installation location is at the end of the fire hydrant network, which is an important monitoring point. Therefore, the alarm level is determined to be a level two warning. The edge gateway generates an alarm event, including device number WP-A3-005, device type fire water pressure sensor, installation location at the end of the fire hydrant network on the 3rd floor of Building A, fault confidence level 0.72, preliminary alarm judgment warning, fault type insufficient water pressure fault, alarm level level two warning, the main evidence being a persistently high water pressure deficiency index, and related neighboring fire equipment numbers WP-A3-004 and WP-A3-006. After the alarm event is uploaded to the server, the server reads recent alarm events in the same area for verification and finds that only WP-A3-005 has insufficient water pressure, while the water pressure of neighboring fire equipment WP-A3-004 and WP-A3-006 is normal, marking it as a device-level anomaly. The server generates an alarm push message, including alarm level two warning, fault type insufficient water pressure fault, device location at the end of the fire hydrant network on the 3rd floor of Building A, and suggested action to check the sensor's pipeline connection and valve status, and pushes it to the operation and maintenance terminal. This invention implements graded alarms by comparing fault confidence with grading thresholds, and determines the fault type by combining the feature with the highest value in the window-level fusion features, providing clear fault location and handling suggestions for operation and maintenance personnel.

[0071] To address the problems described in the background art, this invention acquires basic fire equipment files, regional topology data, and acquisition configurations. Based on these files, it obtains equipment operation files, including equipment number, equipment type, regional topology relationships, and alarm thresholds. Based on these files, it controls preset fire equipment to collect data, adding identifiers based on equipment numbers to obtain raw sampling frames. Multi-dimensional verification is performed on these frames to obtain valid sampled data. The valid sampled data is time-aligned to obtain equipment state vectors. These vectors are then standardized to obtain standardized state vectors. A queue writing operation is performed on these standardized state vectors to obtain actual state queues. When the actual state queue meets a preset queue length condition, predicted state values ​​are generated based on the actual state queues. A predicted residual index is calculated based on the standardized state vectors and the predicted state values. This invention, by comparing actual and predicted states and normalizing them to an index of 0 to 1, can detect early anomalies deviating from normal trends. Furthermore, it obtains individual fault characteristic indices based on the standardized state vectors and equipment types. The invention obtains a sensor-associated state set based on the regional topology and generates a sensor consistency index based on the sensor-associated state set. It is evident that by calculating the individual fault feature index and the sensor consistency index separately, the invention can distinguish between single-device faults and regional environmental changes, reducing the probability of single-point false alarms. Time-window writing is performed on the individual fault feature index, prediction residual index, and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused to obtain the fault confidence level. This invention improves the ability to identify intermittent faults and equipment drift faults by fusing statistical features of multiple types of abnormal evidence within a sliding time window. A preliminary alarm judgment is determined based on the fault confidence level and alarm threshold, where the preliminary alarm judgment is normal, warning, or fault. When the preliminary alarm judgment is warning or fault, the fault type and alarm level are confirmed based on the window-level fusion features, resulting in an alarm event that is pushed to a preset maintenance terminal. This invention achieves graded alarms by comparing the fault confidence level with the graded threshold and determines the fault type by combining the feature item with the highest value in the window-level fusion features, providing maintenance personnel with clear fault location and handling suggestions. Therefore, the present invention can realize multi-sensor fusion monitoring and accurate alarm of the operating status of fire-fighting equipment.

[0072] like Figure 2 The diagram shown is a functional block diagram of a fire equipment fault alarm system based on multi-sensor fusion provided in an embodiment of the present invention.

[0073] The fire equipment fault alarm system 100 based on multi-sensor fusion described in this invention can be installed in an electronic device. Depending on the functions implemented, the fire equipment fault alarm system 100 based on multi-sensor fusion may include a data acquisition and verification module 101, a state standardization prediction module 102, a feature fusion calculation module 103, and an alarm judgment and push module 104. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0074] The data acquisition and verification module 101 is used to acquire basic files of fire-fighting equipment, regional topology data and acquisition configuration, and acquire equipment operation files based on the basic files of fire-fighting equipment, regional topology data and acquisition configuration. The equipment operation files include equipment number, equipment type, regional topology relationship and alarm threshold. Data is collected from preset fire-fighting equipment based on equipment operation records, and an identifier is added according to the equipment number to obtain the original sampling frame. The original sampling frame is then subjected to multi-dimensional verification to obtain valid sampling data. The state standardization prediction module 102 is used to time-align the effective sampled data to obtain a device state vector, and to standardize the device state vector to obtain a standardized state vector. A queue writing operation is performed on the standardized state vector to obtain an actual state queue. When the actual state queue meets the preset queue length condition, a predicted state value is generated based on the actual state queue, and a predicted residual index is calculated based on the standardized state vector and the predicted state value. The feature fusion calculation module 103 is used to obtain a single fault feature index based on the standardized state vector and the device type, obtain a sensor associated state set based on the regional topology, and generate a sensor consistency index based on the sensor associated state set. Time window writing is performed on the individual fault feature index, prediction residual index and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused and calculated to obtain the fault confidence. The alarm judgment and push module 104 is used to determine the preliminary alarm judgment based on the fault confidence and alarm threshold. The preliminary alarm judgment is normal, warning or fault. When the preliminary alarm judgment is warning or fault, the fault type and alarm level are confirmed based on window-level fusion features, and the alarm event is obtained and pushed to the preset operation and maintenance terminal.

[0075] In detail, the modules in the fire equipment fault alarm system 100 based on multi-sensor fusion described in this embodiment of the invention employ the same methods as described above during use. Figure 1The method used is the same as the multi-sensor fusion-based fire equipment fault alarm method described in the article, and can produce the same technical effect, so it will not be repeated here.

[0076] like Figure 3 The diagram shown is a schematic representation of an electronic device that implements a fire equipment fault alarm method based on multi-sensor fusion, according to an embodiment of the present invention.

[0077] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a fire equipment fault alarm method program based on multi-sensor fusion.

[0078] The memory 11 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as the portable hard drive of the electronic device 1. In other embodiments, the memory 11 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 1. Furthermore, the memory 11 includes both internal storage units and external storage devices of the electronic device 1. The memory 11 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code of a fire equipment fault alarm method program based on multi-sensor fusion, but also to temporarily store data that has been output or will be output.

[0079] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., a fire equipment fault alarm method program based on multi-sensor fusion) and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0080] The bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0081] Figure 3 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0082] For example, although not shown, the electronic device 1 may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0083] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device 1 and other electronic devices.

[0084] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0085] The program for a fire equipment fault alarm method based on multi-sensor fusion, stored in the memory 11 of the electronic device 1, is a combination of multiple instructions. When run in the processor 10, it can achieve the following: Acquire basic fire equipment files, regional topology data, and acquisition configuration. Based on the basic fire equipment files, regional topology data, and acquisition configuration, acquire equipment operation files, which include equipment number, equipment type, regional topology relationship, and alarm threshold. Data is collected from preset fire-fighting equipment based on equipment operation records, and an identifier is added according to the equipment number to obtain the original sampling frame. The original sampling frame is then subjected to multi-dimensional verification to obtain valid sampling data. The effective sampled data is time-aligned to obtain a device state vector, and the device state vector is then standardized to obtain a standardized state vector. A queue writing operation is performed on the standardized state vector to obtain an actual state queue. When the actual state queue meets the preset queue length condition, a predicted state value is generated based on the actual state queue, and a predicted residual index is calculated based on the standardized state vector and the predicted state value. A single fault characteristic index is obtained based on the standardized state vector and the device type, and a sensor associated state set is obtained based on the regional topology relationship. A sensor consistency index is generated based on the sensor associated state set. Time window writing is performed on the individual fault feature index, prediction residual index and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused and calculated to obtain the fault confidence. The initial alarm judgment is determined based on the fault confidence and alarm threshold. The initial alarm judgment is normal, warning or fault. When the initial alarm judgment is warning or fault, the fault type and alarm level are confirmed based on window-level fusion features, and the alarm event is obtained and pushed to the preset operation and maintenance terminal.

[0086] Specifically, the processor 10's implementation method for the above instructions can be found in [reference needed]. Figures 1 to 3 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0087] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0088] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following: Acquire basic fire equipment files, regional topology data, and acquisition configuration. Based on the basic fire equipment files, regional topology data, and acquisition configuration, acquire equipment operation files, which include equipment number, equipment type, regional topology relationship, and alarm threshold. Data is collected from preset fire-fighting equipment based on equipment operation records, and an identifier is added according to the equipment number to obtain the original sampling frame. The original sampling frame is then subjected to multi-dimensional verification to obtain valid sampling data. The effective sampled data is time-aligned to obtain a device state vector, and the device state vector is then standardized to obtain a standardized state vector. A queue writing operation is performed on the standardized state vector to obtain an actual state queue. When the actual state queue meets the preset queue length condition, a predicted state value is generated based on the actual state queue, and a predicted residual index is calculated based on the standardized state vector and the predicted state value. A single fault characteristic index is obtained based on the standardized state vector and the device type, and a sensor associated state set is obtained based on the regional topology relationship. A sensor consistency index is generated based on the sensor associated state set. Time window writing is performed on the individual fault feature index, prediction residual index and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused and calculated to obtain the fault confidence. The initial alarm judgment is determined based on the fault confidence and alarm threshold. The initial alarm judgment is normal, warning or fault. When the initial alarm judgment is warning or fault, the fault type and alarm level are confirmed based on window-level fusion features, and the alarm event is obtained and pushed to the preset operation and maintenance terminal.

[0089] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and actual implementations may have other classification methods.

[0090] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0091] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0092] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for alarming faults in fire-fighting equipment based on multi-sensor fusion, characterized in that, The method includes: Acquire basic fire equipment files, regional topology data, and acquisition configuration. Based on the basic fire equipment files, regional topology data, and acquisition configuration, acquire equipment operation files, which include equipment number, equipment type, regional topology relationship, and alarm threshold. Data is collected from preset fire-fighting equipment based on equipment operation records, and an identifier is added according to the equipment number to obtain the original sampling frame. The original sampling frame is then subjected to multi-dimensional verification to obtain valid sampling data. The effective sampled data is time-aligned to obtain a device state vector, and the device state vector is then standardized to obtain a standardized state vector. A queue writing operation is performed on the standardized state vector to obtain an actual state queue. When the actual state queue meets the preset queue length condition, a predicted state value is generated based on the actual state queue, and a predicted residual index is calculated based on the standardized state vector and the predicted state value. The process involves performing a queue writing operation on the standardized state vector to obtain an actual state queue. When the actual state queue meets a preset queue length condition, a predicted state value is generated based on the actual state queue. The predicted residual exponent is then calculated based on the standardized state vector and the predicted state value, including: Obtain the historical state queue, write the standardized state vector into the historical state queue, and obtain the actual state queue; The actual state of the queue is determined based on the queue length condition to obtain the queue state, wherein the queue state is either satisfied or not satisfied; When the queue state is not satisfied, a predicted not started flag is generated, and the predicted not started flag is associated with the standardized state vector to obtain a predicted unavailable state node. Otherwise, the predicted state value is obtained based on the actual state queue; By comparing the difference between the standardized state vector and the predicted state value, the predicted residual data is obtained; Calculate the prediction residual index based on the prediction residual data; A single fault characteristic index is obtained based on the standardized state vector and the device type, and a sensor associated state set is obtained based on the regional topology relationship. A sensor consistency index is generated based on the sensor associated state set. The process of obtaining a single fault characteristic index based on the standardized state vector and device type, obtaining a sensor-associated state set based on the regional topology, and generating a sensor consistency index based on the sensor-associated state set includes: Based on the device type, the standardized state vector is extracted into multiple state component data. Anomaly feature calculations are performed on the multiple state item data to obtain multiple individual fault feature values; Multiple individual fault characteristic values ​​are combined to obtain an individual fault characteristic index; Based on the regional topological relationships, the standardized state vectors are correlated and extracted to obtain the sensor associated state set; Statistical processing is performed on the sensor-associated state set to obtain sensor-associated statistical data; By comparing the differences between the standardized state vector and the sensor correlation statistics, the sensor consistency index is obtained; Time window writing is performed on the individual fault feature index, prediction residual index and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused and calculated to obtain the fault confidence. The process involves writing the individual fault characteristic index, prediction residual index, and sensor consistency index into a time window to obtain window-level fused features. These window-level fused features are then fused and calculated to obtain the fault confidence level, including: The individual fault characteristic index, the predicted residual index, the predicted unavailable state node, and the sensor consistency index are combined to obtain the current abnormal characteristic data. The current abnormal feature data is written into a preset sliding time window to obtain a time window feature node, wherein the time window feature node includes the time window abnormal feature data and the time window writing order; Statistical processing is performed on the individual fault feature indices in the time window anomaly feature data to obtain individual fault window features; Statistical processing is performed on the prediction residual index in the time window anomaly feature data to obtain the prediction residual window feature; Statistical processing is performed on the sensor consistency index in the time window anomaly feature data to obtain the sensor consistency window feature; The single fault window feature, the prediction residual window feature, and the sensor consistency window feature are combined to obtain the window-level fusion feature; The fault confidence is obtained by performing fusion calculation based on the window-level fusion features. The initial alarm judgment is determined based on the fault confidence and alarm threshold. The initial alarm judgment is normal, warning or fault. When the initial alarm judgment is warning or fault, the fault type and alarm level are confirmed based on window-level fusion features, and the alarm event is obtained and pushed to the preset operation and maintenance terminal.

2. The fire equipment fault alarm method based on multi-sensor fusion as described in claim 1, characterized in that, The process of performing multi-dimensional verification on the original sampled frame to obtain valid sampled data includes: Perform field integrity verification on the original sampled frame to obtain missing field markers, where the missing field marker indicates whether the field is missing or not. When the missing field is marked as not missing, the original sampled frame is checked for sequence number to obtain a duplicate mark. The duplicate mark is either duplicate or not duplicate. If the duplicate mark is duplicate, the original sampled frame is discarded. Otherwise, the original sampled frame is checked for time validity to obtain a delay mark. The delay mark is either timed out or not timed out. When the time delay mark is not exceeded, the original sampling frame is range checked to obtain the boundary mark, which is either out of bounds or not out of bounds; When the boundary crossing mark is not crossed, the original sampled frame is subjected to outlier smoothing to obtain suspicious value marks and smoothed values, where suspicious value marks are either suspicious or normal. The missing field marker, delay marker, out-of-bounds marker, suspicious value marker, and smoothing value are combined with the original sampling frame to obtain valid sampling data.

3. The fire equipment fault alarm method based on multi-sensor fusion as described in claim 2, characterized in that, The step of aligning the valid sampled data by time to obtain a device state vector, and then standardizing the device state vector to obtain a standardized state vector, includes: The current fusion time slice is determined based on the acquisition configuration, and the valid sampled data falling into the current fusion time slice is queried from the valid sampled data to obtain the valid sampled data of the current slice and the query status, wherein the query status is either present or absent; If the query status is "existing", then the current slice status data is obtained based on the current slice's valid sampled data, missing field marker, delay marker, out-of-bounds marker, suspicious value marker, and smoothing value. Otherwise, a missing test flag is obtained, and missing test status data is acquired based on the device number, device type, current fusion time slice, and missing test flag; The current chip status data or missing test status data is determined as the device status vector; The device state vector is standardized to obtain standardized state data; The standardized state data, missing field markers, delay markers, out-of-bounds markers, suspicious value markers, or missing test markers are combined to obtain a standardized state vector.

4. The fire equipment fault alarm method based on multi-sensor fusion as described in claim 3, characterized in that, The step of calculating the prediction residual index based on the prediction residual data includes: The predicted residual data is extracted item by item to obtain multiple predicted residual values; Multiple residual tolerance thresholds are obtained, wherein the multiple residual tolerance thresholds correspond one-to-one with the multiple predicted residual values; Multiple prediction residual indices are calculated based on the multiple prediction residual values ​​and multiple residual tolerance thresholds, wherein each prediction residual index corresponds one-to-one with a prediction residual value, and the calculation formula is as follows: in, This represents the j-th prediction residual index among multiple prediction residual indices. This represents the j-th prediction residual value among multiple prediction residual values. This represents the residual tolerance threshold corresponding to the j-th predicted residual value among multiple predicted residual values, and Greater than 0.

5. The fire equipment fault alarm method based on multi-sensor fusion as described in claim 4, characterized in that, The statistical processing of the prediction residual index in the time window anomaly feature data to obtain the prediction residual window features includes: The predicted residual index in the time window anomaly feature data is extracted by sub-item extraction within the time window to obtain a residual index sequence, wherein the residual index sequence consists of multiple residual index values ​​arranged in the writing order of the time window; The largest prediction residual index is extracted from multiple prediction residual indices, and the largest prediction residual index is used as the residual peak feature. The mean value of the multiple predicted residual indices is calculated to obtain the residual mean characteristic. Obtain the residual statistical threshold and the sudden increase judgment threshold, and perform threshold exceedance statistics on the multiple predicted residual indices based on the residual statistical threshold to obtain the residual threshold exceedance frequency characteristics; Extract the residual exponent values ​​sequentially from the residual exponent sequence, and perform the following operations on the extracted residual exponent values: Adjacent residual index values ​​are identified based on the residual index values, wherein adjacent residual index values ​​are adjacent and lag behind the residual index values. Calculate the difference between adjacent residual index values ​​and residual index values ​​to obtain the residual index difference value, where adjacent residual index values ​​are minuends and residual index values ​​are subtrahends; The residual index difference is judged and summarized based on the sudden increase judgment threshold to obtain the set of difference values ​​exceeding the threshold; By counting the number of out-of-threshold differences in the set of out-of-threshold differences, the characteristic of the number of residual bursts can be obtained; The residual peak value feature, residual mean value feature, residual over-threshold frequency feature, and residual surge frequency feature are combined to obtain the prediction residual window feature.

6. The fire equipment fault alarm method based on multi-sensor fusion as described in claim 5, characterized in that, The step of performing fusion calculation based on the window-level fusion features to obtain the fault confidence includes: The single fault window feature, the prediction residual window feature, and the sensor consistency window feature in the window-level fusion feature are normalized respectively to obtain the single fault comprehensive value, the prediction residual comprehensive value, and the sensor consistency comprehensive value. The fault confidence level is obtained by calculating the mean of the individual fault comprehensive value, the prediction residual comprehensive value, and the sensor consistency comprehensive value.

7. A fire equipment fault alarm system based on multi-sensor fusion, characterized in that, The system includes: The data acquisition and verification module is used to acquire basic files of fire-fighting equipment, regional topology data and acquisition configuration, and to acquire equipment operation files based on the basic files of fire-fighting equipment, regional topology data and acquisition configuration. The equipment operation files include equipment number, equipment type, regional topology relationship and alarm threshold. Data is collected from preset fire-fighting equipment based on equipment operation records, and an identifier is added according to the equipment number to obtain the original sampling frame. The original sampling frame is then subjected to multi-dimensional verification to obtain valid sampling data. The state standardization prediction module is used to time-align the effective sampled data to obtain a device state vector, and to standardize the device state vector to obtain a standardized state vector. A queue writing operation is performed on the standardized state vector to obtain an actual state queue. When the actual state queue meets the preset queue length condition, a predicted state value is generated based on the actual state queue, and a predicted residual index is calculated based on the standardized state vector and the predicted state value. The process involves performing a queue writing operation on the standardized state vector to obtain an actual state queue. When the actual state queue meets a preset queue length condition, a predicted state value is generated based on the actual state queue. The predicted residual exponent is then calculated based on the standardized state vector and the predicted state value, including: Obtain the historical state queue, write the standardized state vector into the historical state queue, and obtain the actual state queue; The actual state of the queue is determined based on the queue length condition to obtain the queue state, wherein the queue state is either satisfied or not satisfied; When the queue state is not satisfied, a predicted not started flag is generated, and the predicted not started flag is associated with the standardized state vector to obtain a predicted unavailable state node. Otherwise, the predicted state value is obtained based on the actual state queue; By comparing the difference between the standardized state vector and the predicted state value, the predicted residual data is obtained; Calculate the prediction residual index based on the prediction residual data; The feature fusion calculation module is used to obtain a single fault feature index based on the standardized state vector and the device type, obtain a sensor associated state set based on the regional topology, and generate a sensor consistency index based on the sensor associated state set. The process of obtaining a single fault characteristic index based on the standardized state vector and device type, obtaining a sensor-associated state set based on the regional topology, and generating a sensor consistency index based on the sensor-associated state set includes: Based on the device type, the standardized state vector is extracted into multiple state component data. Anomaly feature calculations are performed on the multiple state item data to obtain multiple individual fault feature values; Multiple individual fault characteristic values ​​are combined to obtain an individual fault characteristic index; Based on the regional topological relationships, the standardized state vectors are correlated and extracted to obtain the sensor associated state set; Statistical processing is performed on the sensor-associated state set to obtain sensor-associated statistical data; By comparing the differences between the standardized state vector and the sensor correlation statistics, the sensor consistency index is obtained; Time window writing is performed on the individual fault feature index, prediction residual index and sensor consistency index to obtain window-level fusion features. The window-level fusion features are then fused and calculated to obtain the fault confidence. The process involves writing the individual fault characteristic index, prediction residual index, and sensor consistency index into a time window to obtain window-level fused features. These window-level fused features are then fused and calculated to obtain the fault confidence level, including: The individual fault characteristic index, the predicted residual index, the predicted unavailable state node, and the sensor consistency index are combined to obtain the current abnormal characteristic data. The current abnormal feature data is written into a preset sliding time window to obtain a time window feature node, wherein the time window feature node includes the time window abnormal feature data and the time window writing order; Statistical processing is performed on the individual fault feature indices in the time window anomaly feature data to obtain individual fault window features; Statistical processing is performed on the prediction residual index in the time window anomaly feature data to obtain the prediction residual window feature; Statistical processing is performed on the sensor consistency index in the time window anomaly feature data to obtain the sensor consistency window feature; The single fault window feature, the prediction residual window feature, and the sensor consistency window feature are combined to obtain the window-level fusion feature; The fault confidence is obtained by performing fusion calculation based on the window-level fusion features. The alarm judgment and push module is used to determine the preliminary alarm judgment based on the fault confidence and alarm threshold. The preliminary alarm judgment is normal, warning or fault. When the preliminary alarm judgment is warning or fault, the fault type and alarm level are confirmed based on window-level fusion features, the alarm event is obtained and pushed to the preset operation and maintenance terminal.

Images