A four-level intelligent diagnosis method and system for the operation state of a fire water supply system

Abstract

The present application relates to the technical field of fire engineering and Internet of Things data analysis, and provides a four-level intelligent diagnosis method for the operation state of a fire water supply system, comprising: pre-processing: automatically identifying the type of fire system and the installation position of each sensor; first level: verifying the compliance threshold of single-point for each pressure record; second level: detecting discrete pressure drop events by using a double-threshold method combining the absolute threshold of amplitude and the relative threshold of rate; third level: establishing a rolling reference distribution of decay rate, triggering a leakage warning for abnormal decay rate continuously exceeding the upper limit of the reference and the positive trend of the decay rate time sequence; fourth level: obtaining the results of the water supply isolation determination within the system and the cross-system independence verification, and incorporating them into the diagnosis result aggregation; diagnosis result aggregation: forming a structured abnormal event list from the outputs of the four levels. The method can adapt to the system type and sensor position, and organically integrate the four diagnosis levels.

Description

Technical Field

[0001] This invention relates to the technical field of fire protection engineering and Internet of Things (IoT) data analysis, and particularly to a four-level intelligent diagnostic method and system for the operating status of fire water supply systems. Specifically, it integrates single-point pressure compliance detection, event-level anomaly detection, trend-level degradation detection, and system-level synchronization analysis into a comprehensive intelligent diagnostic system for the operating status of fire water supply systems at four levels. Background Technology

[0002] Fire water supply systems (including fire hydrant systems and automatic sprinkler systems) are among the most important active fire suppression facilities in building fire protection engineering. Their operational status directly determines the success rate of initial fire control and extinguishing. With the rapid development of IoT technology, more and more buildings and industrial sites have deployed pressure and flow sensors at key nodes of fire water supply networks, and are using IoT platforms to collect, display, and perform simple analysis of real-time monitoring data. Theoretically, this digital approach can significantly improve the operational efficiency and fault response speed of fire protection systems. However, in practical engineering applications, existing methods for diagnosing the operational status of fire water supply systems based on IoT data still face the following systemic challenges, resulting in diagnostic capabilities falling far short of expectations.

[0003] (1) Single diagnostic dimension Current IoT platforms' assessment of the operational status of fire-fighting water supply systems is largely limited to simple comparisons of single-point pressure values ​​with fixed thresholds—the so-called "single-point detection." For example, an alarm is triggered only when the pressure value at a given moment falls below a preset lower limit or rises above a preset upper limit. While this single-dimensional threshold detection method is simple to implement, it has a fundamental flaw: it cannot detect slow, gradual deterioration processes (such as decreased flow capacity due to corrosion of the pipe wall or long-term pressure decay caused by minor leaks), nor can it identify system-level structural problems involving multiple sensors (such as accidental closure of zone valves or mismatch between pump and pipe network characteristics). As a result, many early and hidden faults are completely missed, often only being discovered when the system's functionality is severely compromised or even completely collapses.

[0004] (2) The inherent differences between the fire hydrant system and the sprinkler system were not addressed. Fire hydrant systems and sprinkler systems differ significantly by orders of magnitude in engineering design, operating mechanisms, and monitoring data characteristics. Specifically: ① Regarding sampling frequency, fire hydrant systems typically employ high-frequency sampling at the second or minute level (intervals usually ≤2 minutes) to capture frequent start-ups and shutdowns of pressure-stabilizing pumps and pressure fluctuations; while sprinkler systems, due to their large pipe volumes and slow pressure changes, often require sampling intervals as long as 30 minutes or even several hours. ② Regarding operating pressure range, the normal pressure of fire hydrant main pipelines can reach over 1.00 MPa, while the upper limit for sprinkler systems is generally 0.80 MPa. ③ Regarding normal attenuation rate, the natural pressure attenuation rate of sprinkler systems is much lower than that of fire hydrant systems. However, most existing diagnostic methods ignore these differences, using uniform analysis parameters (such as the same pressure drop threshold and the same attenuation rate benchmark) for both types of systems. This "one-size-fits-all" approach leads to numerous false alarms (e.g., misjudging normal slow pressure drops in sprinkler systems as abnormal leaks) and missed alarms (e.g., failing to detect high-frequency, minute pressure drop anomalies in fire hydrant systems in a timely manner), severely reducing the reliability of diagnostic results and their practical guiding value.

[0005] (3) Sensor installation location information is missing or incorrect In fire-fighting water supply networks, there is a fundamental difference in the compliant pressure thresholds between main pipeline sensors and end-point test sensors (installed at the most unfavorable test point): main pipelines require maintaining relatively high static and dynamic pressures (e.g., not less than 0.15 MPa), while the allowable lower limit for end-point test points is significantly lower (not less than 0.07 MPa for fire hydrant ends and not less than 0.05 MPa for sprinkler ends). Correct sensor location labeling is a prerequisite for using correct thresholds for judgment. However, in actual engineering projects, due to incomplete construction records, errors in metadata entry during the commissioning phase, and failure to update the database synchronously during later system upgrades, a large amount of sensor location information (main pipeline / end-point) is missing or incorrectly labeled in the metadata of the IoT platform. When the fixed threshold method is applied to these incorrect or incomplete location labels, systemic diagnostic errors will occur: for example, a sensor actually installed at the end but incorrectly labeled as a main pipeline sensor will be judged as "collapsed" due to its normal low pressure value; conversely, mislabeling a main pipeline sensor as an end-point sensor may result in the failure to report serious low-pressure faults. These types of errors cannot be corrected by simply adjusting the global threshold; the sensor position must be adaptively inferred by the algorithm.

[0006] (4) Insufficient diagnostic levels and lack of multi-level organic integration The complete failure modes of fire-fighting water supply systems cover a broad timescale and spatial range, from millisecond-level transient events to weekly structural failures. Specifically: Single-point threshold detection is used to detect instantaneous over-limits (such as overpressure, collapse); event-level detection is used to identify discrete abnormal events (such as sudden pressure drop, abnormal pump start-up); trend analysis is used to capture slow degradation (such as the increase in decay rate caused by long-term leakage); cross-sensor synchronization analysis is used to diagnose system-level structural anomalies (such as water supply interruption, misconfiguration of shared pumps).

[0007] Each of the methods described above has its own applicable boundaries for different types of diagnostic problems. However, in the existing technology, these methods are often developed and applied in isolation: either only one or two are used, or multiple methods are integrated but lack an organic hierarchical architecture and parameter coordination mechanism. A single method cannot cover all failure modes, while simply piecing together methods can lead to redundant, conflicting, or even contradictory diagnostic results. Crucially, none of the existing methods introduce an adaptive parameter selection module that can automatically identify the system type (fire hydrant / sprinkler) and infer the installation location (main pipeline / terminal) based on sensor data. This prevents the analysis at each level from using the correct, targeted parameter set, fundamentally limiting the comprehensive diagnostic capability.

[0008] In summary, existing technologies lack a comprehensive diagnostic method for the operational status of fire-fighting water supply systems that can adapt to system type and sensor location and organically integrate four levels: single-point compliance detection, event-level anomaly detection, trend-level degradation detection, and system-level synchronization analysis. This invention addresses this technological gap. Summary of the Invention

[0009] To address the aforementioned problems, the present invention aims to provide a four-level intelligent diagnostic method and system for the operational status of fire-fighting water supply systems. This system can adaptively adapt to system type and sensor location, organically integrating the four diagnostic levels to cover all diagnostic dimensions, from single-point violations to system structural anomalies.

[0010] The above-mentioned objective of this invention is achieved through the following technical solutions: A four-level intelligent diagnostic method for the operating status of a fire water supply system includes the following preprocessing, four diagnostic levels, and aggregation of diagnostic results: Pre-processing: Automatically identify the fire protection system type and sensor installation location for each sensor, and select the corresponding set of diagnostic parameters accordingly; Level 1: Verify the single-point compliance threshold for each pressure record and determine network crashes, end-point water supply failures, and overpressure events; The second level: a dual-threshold method combining absolute amplitude threshold and relative rate threshold is used to detect discrete pressure drop events, and sampling hysteresis correction is performed on the pressure stabilizing pump start-up event; The third level: Establish a rolling baseline distribution of decay rate, and trigger leakage early warning for abnormal decay rates that continuously exceed the upper limit of the baseline and for positive trends in the decay rate time series. Fourth level: Obtain the results of water supply interruption determination within the same system and the results of cross-system independence verification, and incorporate them into the diagnostic results aggregation; Diagnostic result aggregation: The outputs from the four levels are combined into a structured list of abnormal events, sorted by severity, and then maintenance work order suggestions are output and pushed to the corresponding role's alarm notification.

[0011] Furthermore, in the pre-processing, the fire protection system type and sensor installation location are automatically identified for each sensor, and the corresponding set of diagnostic parameters is selected accordingly, specifically: A: System type identification: Automatic classification based on the pressure time-series statistical characteristics of each sensor within a preset time window, without relying on metadata; the statistical characteristics include at least the median of the pump cycle and the statistical distribution of the sampling interval; When the median pump cycle exceeds the first preset duration and the sampling interval is greater than or equal to the first sampling interval threshold, it is automatically identified as a sprinkler system. When the median pump cycle is within the second preset time range and the sampling interval is less than or equal to the second sampling interval threshold, it is automatically identified as a fire hydrant system. B: Sensor location inference: Automatically infers the installation location based on the continuous pressure range of each sensor during the continuous monitoring period; If the pressure value of a certain sensor is continuously lower than the first pressure threshold during the continuous monitoring period, the sensor will be automatically reclassified as an end test point and the corresponding end pressure compliance standard will be selected; among the end pressure compliance standards, the end pressure of fire hydrants is not lower than the second pressure threshold and the end pressure of sprinklers is not lower than the third pressure threshold. Otherwise, treat the sensor as a main network sensor and select the corresponding pressure diagnostic parameter set for the main network.

[0012] Furthermore, in the first level, each pressure record is verified against a single-point compliance threshold to determine network crashes, end-point water supply failures, and overpressure events, specifically: Each pressure record is independently verified in real time, without any accumulation over a time window, including the following types of judgments: A: Network crash determination: When the pressure record value is lower than the main pipeline crash threshold and the low pressure state continues for a preset short duration, it is determined to be a network crash; the main pipeline crash threshold corresponds to the minimum allowable pressure to maintain the basic integrity of the pipeline network; B: End-point water supply failure determination: When the sensor has been identified as an end-point test point by the preprocessing, if its pressure record value is lower than the specific lower limit threshold set for the end position, it is determined that the end-point water supply has failed. The specific lower limit threshold is preset according to different system types. C: Overpressure determination: When the pressure recorded value exceeds the pressure upper limit threshold of the corresponding system type, it is determined to be overpressure; among them, the pressure upper limit threshold of the sprinkler system is the first upper limit value, and the pressure upper limit threshold of the fire hydrant system is the second upper limit value; All of the above judgments are executed in real time, meaning that the compliance verification of each stress record is completed immediately upon receipt, without waiting for subsequent data.

[0013] Furthermore, in the second level, a dual-threshold method combining absolute amplitude thresholds and relative rate thresholds is used to detect discrete pressure drop events, and sampling lag correction is applied to the pressure stabilizing pump start-up event, specifically as follows: S21: Detect sudden pressure drop events A dual-threshold method is used, meaning that a pressure drop event is triggered only when both of the following conditions are met simultaneously; it is not triggered if only one condition is met or neither condition is met: Amplitude condition: Within a preset sliding time window, the magnitude of the pressure drop exceeds an absolute threshold set for the sensor location; wherein, for the main pipeline sensor, the absolute threshold is a first pressure drop amplitude; for the end-point test sensor, the absolute threshold is a second pressure drop amplitude. Rate condition: Within the sliding time window, the instantaneous pressure drop rate exceeds a preset multiple of the current normal attenuation rate of the system; the preset multiple is the first rate multiple; wherein, the instantaneous drop rate is calculated by dividing the pressure difference between adjacent sampling points by the sampling time interval; the normal attenuation rate of the system is dynamically determined based on historical data of the sensor during its recent normal operation period; The dual-threshold method, by simultaneously constraining the pressure drop amplitude and the rate of drop, can effectively distinguish between a real abnormal pressure drop and the pressure decay caused by normal pump circulation, thereby avoiding misjudging the decay of normal pump circulation as a sudden drop event. Compared with the single-rate detection method that only uses the rate threshold, this dual-threshold method can reduce false positive events by a preset proportion. S22: Sampling hysteresis correction for pump start-up events For pressure stabilizing pump start-up events, sampling lag correction is performed: by analyzing the time offset characteristics between the pump start-up marker and the actual pressure rise edge in the pressure timing sequence, the apparent early start phenomenon caused by sampling delay is identified and restored to a normal start caused by sampling delay; this correction can ensure that a preset proportion of apparent early starts are correctly classified as normal starts, avoiding false alarms. S23: Timeliness and Processing Method This layer employs a near real-time processing method, using a sliding time window to scan and detect the continuous pressure data stream window by window.

[0014] Furthermore, in the third level, a rolling baseline distribution of attenuation rate is established. Leakage warnings are triggered for abnormal attenuation rates that continuously exceed the upper limit of the baseline and for positive trends in the attenuation rate time series. Specifically: A: Establishment of rolling baseline distribution for attenuation rate Based on the pressure data collected by each sensor during its historical normal operation, a series of continuous attenuation rate sample intervals are calculated; the attenuation rate of each sample interval is determined by the rate of change of pressure over time within that interval. Once the accumulated number of valid sample intervals reaches the preset minimum sample size, a rolling baseline distribution of the decay rate is established, which includes the mean. and standard deviation ; The rolling baseline distribution is dynamically updated over time: each time a new sample interval is added, the value within the current time window is recalculated. and This is to reflect the normal attenuation characteristics of the system in the near future; B: Leakage warning triggering conditions Monitor the current decay rate in real time and compare it with the rolling baseline distribution; When the current decay rate continues to exceed the benchmark average Add the standard deviation of the preset multiple If the state continues for a preset duration or number of samples, a leakage warning will be triggered. The leakage early warning is used to indicate the presence of early, minor leaks in the pipeline network. This warning is issued before traditional fixed threshold alarms, providing an early warning time. C: Warning trigger condition for worsening leakage trend A rolling linear regression analysis was performed on the time series of decay rate, with the rolling window length set to a preset duration. Calculate the regression slope and its significance level p-value; When the regression slope is positive and its significance level p-value is less than the preset threshold, it is determined that the slope is positively significant, triggering an alarm for worsening leakage trend. D: Time Limit Explanation This level uses monthly calculations as the basic cycle and employs a preset rolling time window for data sampling and statistical analysis, balancing the sensitivity of trend capture with computational efficiency.

[0015] Furthermore, in the fourth level, the results of the water supply interruption determination within the same system and the cross-system independence verification are obtained, specifically as follows: A: The results of water supply interruption determination within the same system include the connection status of the corresponding pipe section for each sensor pair, including at least the normal connection status, the slightly damaged status, and the water supply interruption event. B: Cross-system independence verification results, including the independence determination results between sensor pairs belonging to different fire protection systems, specifically whether they are in normal independent operation or suspected shared pump configuration. The results obtained above are used as the fourth-level diagnostic output, and the diagnostic results are aggregated with those of the first, second, and third levels of abnormal events.

[0016] Furthermore, in the aggregation of diagnostic results, the outputs of the four levels are used to form a structured list of abnormal events. After being sorted by severity, maintenance work order suggestions are output and pushed to the corresponding role's alarm notification, specifically: A: Construction of a structured list of exception events Collect and integrate the raw abnormal event data output from the first, second, third, and fourth levels; Deduplication and merging of abnormal events from different levels: If the same physical fault is detected by multiple levels at the same time, it is merged into a single comprehensive anomaly record, and the source of all involved levels is marked. Generate a structured record for each exception event, which must contain at least the following fields: Types: Classified according to the diagnostic level to which the abnormality belongs, including first-level class, second-level class, third-level class, and fourth-level class; Time of occurrence: The timestamp when the anomaly was first detected; Duration: The duration from the first detection of the anomaly to the recovery or the current moment; Severity: Divided into two levels: Warning and Critical. The Critical level corresponds to anomalies that may cause immediate system failure or serious security risks; the Warning level corresponds to anomalies that require attention but do not affect the basic functions of the system at present. Recommended handling: Pre-set or dynamically generated maintenance suggestion text for this type of anomaly, including inspection parts, operation steps, priority prompts, etc.; B: Generation and sorting of maintenance work order suggestions Based on all structured anomaly records, they are sorted in descending order of severity: all critical anomalies are listed before warning anomalies. Within the same severity level, further sorting can be performed according to duration, time of occurrence, or preset weighting factors; The sorted set of abnormal records is converted into a maintenance work order suggestion list. Each work order suggestion includes an abnormality summary, recommended handling, and suggested response time limit. C: Alarm push notification Based on the anomaly type and severity suggested in the maintenance work order, a preset role distribution strategy is matched; Alarm notifications are sent to the terminal devices of the corresponding roles through at least one communication channel; For critical level anomalies, a repeat push mechanism can be added.

[0017] A four-level intelligent diagnostic system for fire water supply system operation status, used to perform the above-described four-level intelligent diagnostic method for fire water supply system operation status, includes: The adaptive parameter selection module is used to automatically identify the fire protection system type and sensor installation location for each sensor, and select the corresponding set of diagnostic parameters accordingly. The first-level single-point detection module is used to verify the single-point compliance threshold for each pressure record and to determine network crashes, end-point water supply failures, and overpressure events. The second-level event detection module is used to detect discrete pressure drop events using a dual-threshold method that combines absolute amplitude thresholds and relative rate thresholds, and to perform sampling hysteresis correction on pressure stabilizing pump start-up events. The third-level trend analysis module is used to establish a rolling baseline distribution of decay rate and to trigger leakage warnings for abnormal decay rates that continuously exceed the upper limit of the baseline and for positive trends in the decay rate time series. The fourth-level synchronization analysis module is used to obtain the water supply interruption judgment results within the same system and the cross-system independence verification results, and incorporate them into the diagnostic results aggregation; The diagnostic results aggregation module is used to construct a structured list of abnormal events from the four levels of output, sort them by severity, output maintenance work order suggestions, and push alarm notifications to the corresponding roles.

[0018] A computer device, characterized in that it includes a memory and one or more processors, wherein the memory stores computer code, and when the computer code is executed by the one or more processors, causes the one or more processors to perform the method as described above.

[0019] A computer-readable storage medium, characterized in that the computer-readable storage medium stores computer code, which, when executed, is performed as described above.

[0020] Compared with the prior art, the present invention has at least one of the following beneficial effects: (1) Full failure mode coverage, with no diagnostic blind spots This invention organically integrates four diagnostic levels, each targeting failure modes across different time scales and spatial ranges: the first level (single-point pressure compliance detection) can capture single-point over-limit events at the millisecond level; the second level (event-level anomaly detection) can identify discrete pressure drops ranging from seconds to minutes; the third level (trend-level degradation detection) can detect progressive degradation trends from days to months; and the fourth level (system-level synchronicity analysis) can diagnose structural isolation problems at the weekly level. These four levels complement each other, forming a complete diagnostic chain from instantaneous overpressure to long-term leakage and then to system-level water supply isolation, eliminating diagnostic blind spots commonly found in existing technologies.

[0021] (2) Adaptive parameter selection, fundamentally eliminating false alarms and missed alarms. This invention utilizes a pre-installed adaptive parameter selection module to automatically identify the system type (fire hydrant system / sprinkler system) based on pressure time-series statistical characteristics, and automatically infers the sensor installation location (main pipeline / end-point test point) based on the continuous pressure range. This method eliminates the need for metadata, ensuring that each sensor uses the correct diagnostic parameter set that matches its actual type and location. This eliminates systematic false alarms and false negatives caused by traditional uniform parameter schemes that ignore system differences and location errors, significantly improving the reliability of diagnostic results.

[0022] (3) The dual-threshold method significantly reduces the false positive rate and effectively alleviates alarm fatigue. In the second-level pressure drop event detection, this invention innovatively employs a dual-threshold method combining absolute amplitude thresholds and relative rate thresholds. This design requires that a pressure drop event is triggered only when the pressure drop amplitude exceeds a specific absolute threshold at a given location and the instantaneous drop rate exceeds a preset multiple (e.g., 5 times) of the system's normal attenuation rate; otherwise, no event is triggered. Practical verification shows that compared to a single-rate detection method using only rate thresholds, this invention's dual-threshold method reduces false positives by 99.3%, significantly avoiding invalid alarms caused by normal pump cycle attenuation being misjudged as abnormal. This effectively solves the problem of "alarm fatigue" for maintenance personnel, allowing limited human resources to focus on truly critical abnormal events.

[0023] (4) Large-scale engineering verification to prove the ability to detect significant anomalies The four-level intelligent diagnostic framework proposed in this invention has been successfully validated in large-scale engineering projects across 37 buildings of different types, connecting 85 pressure sensors and accumulating 208,002 effective pressure records. Validation results show that over 35% of the monitored systems exhibited at least one significant anomaly previously undetected by existing monitoring methods (including but not limited to pipe network leakage, water supply interruptions, incorrect sensor location labeling, and abnormal shared pump configuration). These results fully demonstrate the invention's powerful detection capability for hidden faults and progressive degradation in real-world engineering environments, showcasing significant technological advancements and broad prospects for industrial application. Attached Figure Description

[0024] Figure 1 This is the overall architecture diagram of the four-level intelligent diagnostic method for the operating status of fire water supply systems according to the present invention; Figure 2 This is an overall structural diagram of the four-level intelligent diagnostic structure for the operation status of the fire water supply system of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0027] First Embodiment like Figure 1 As shown, this embodiment provides a four-level intelligent diagnostic method for the operating status of a fire water supply system, including the following preprocessing, four diagnostic levels, and diagnostic result aggregation: (a) Pre-processing: Automatically identify the fire protection system type and sensor installation location for each sensor, and select the corresponding set of diagnostic parameters accordingly.

[0028] A: System type identification: Automatic classification based on the pressure time-series statistical characteristics of each sensor within a preset time window, without relying on metadata; the statistical characteristics include at least the median of the pump cycle and the statistical distribution of the sampling interval; When the median pump cycle exceeds the first preset duration (e.g., more than 2 hours) and the sampling interval is greater than or equal to the first sampling interval threshold (e.g., ≥30 minutes), it is automatically identified as a sprinkler system; When the median pump cycle time is within the second preset duration range (e.g., 520 minutes) and the sampling interval is less than or equal to the second sampling interval threshold (e.g., ≤2 minutes), it is automatically identified as a fire hydrant system. B: Sensor location inference: Automatically infers the installation location based on the continuous pressure range of each sensor during the continuous monitoring period; If a sensor's pressure value remains below the first pressure threshold (e.g., 0.15 MPa) during the continuous monitoring period, the sensor will be automatically reclassified as an end-point test point, and the corresponding end-point pressure compliance standard will be selected. Among the end-point pressure compliance standards, the fire hydrant end pressure is not lower than the second pressure threshold (e.g., 0.07 MPa), and the sprinkler end pressure is not lower than the third pressure threshold (e.g., 0.05 MPa). Otherwise, treat the sensor as a main network sensor and select the corresponding pressure diagnostic parameter set for the main network.

[0029] The core of the aforementioned pre-processing module lies in its ability to automatically identify system type and infer sensor location solely through the statistical characteristics of the pressure time-series data, without relying on any manual annotation or metadata. Specifically: In terms of system type identification, fire hydrant systems and sprinkler systems exhibit naturally separable statistical characteristics in terms of pump cycle time and sampling frequency due to fundamental differences in design specifications and operating mechanisms. Fire hydrant systems rely on frequent start-stop cycles of pressure-stabilizing pumps to maintain network pressure; therefore, the median pump cycle time is typically on the order of several minutes, and the sampling interval is correspondingly short. In contrast, sprinkler systems have large network volumes and smooth pressure fluctuations, resulting in pump cycle times that often last for several hours, and significantly longer sampling intervals. This invention utilizes this objective difference by calculating the statistical distribution of the median pump cycle time and sampling interval, accurately classifying sensors without any prior labels, thereby avoiding parameter mismatch problems caused by missing construction records or data entry errors.

[0030] Regarding sensor location inference, the essential difference between main pipeline sensors and end-point test points lies in the fact that end-point test points are located at the most unfavorable point in the pipeline network, and their continuous pressure under normal operating conditions is significantly lower than that of the main pipeline network. This invention sets a reasonable first pressure threshold (e.g., 0.15 MPa) to judge the pressure range of sensors during continuous monitoring: if the pressure of a sensor remains below this threshold, it indicates that its location cannot maintain the static pressure at the main pipeline network level, and it should be classified as an end-point test point, automatically switching to a lower end-point pressure compliance standard (fire hydrant end ≥ 0.07 MPa, sprinkler end ≥ 0.05 MPa); otherwise, it is treated as a main pipeline network. This inference method does not rely on location labels and is entirely data-driven, fundamentally eliminating systematic diagnostic biases caused by incorrect or missing location labels.

[0031] Through the aforementioned adaptive parameter selection mechanism, this invention provides accurate and personalized parameter inputs for the subsequent four diagnostic levels, ensuring that the thresholds, attenuation rate benchmarks, correlation coefficient criteria, etc. used in each level (single-point compliance detection, event detection, trend analysis, and synchronization analysis) are matched with the actual system type and installation location of the sensor, thus laying the foundation for the high precision and low false alarm of the entire four-level intelligent diagnostic method.

[0032] (ii) First level: Verify the single-point compliance threshold for each pressure record and determine network crash, end-point water supply failure and overpressure event.

[0033] Each pressure record is independently verified in real time, without any accumulation over a time window, including the following types of judgments: A: Network crash determination: When the pressure record value is lower than the main pipeline crash threshold (e.g., <0.15MPa) and the low pressure state continues for a preset short duration (e.g., more than 3 consecutive records or for more than a certain number of seconds), it is determined to be a network crash; the main pipeline crash threshold corresponds to the minimum allowable pressure to maintain the basic integrity of the pipeline network; B: End-point water supply failure determination: When the sensor has been identified as an end-point test point by the preprocessing, if its pressure record value is lower than the specific lower limit threshold set for that end position, it is determined that the end-point water supply has failed. The specific lower limit threshold is preset according to different system types (the specific lower limit threshold is preset according to different system types, for example, ≥0.07MPa for fire hydrant end and ≥0.05MPa for sprinkler end). C: Overpressure determination: When the pressure recorded value exceeds the pressure upper limit threshold of the corresponding system type, it is determined to be overpressure; among them, the pressure upper limit threshold of the sprinkler system is the first upper limit value (e.g., <0.80MPa, that is, the normal upper limit is 0.80MPa, exceeding it is overpressure), and the pressure upper limit threshold of the fire hydrant system is the second upper limit value (e.g., <1.00MPa, that is, the normal upper limit is 1.00MPa, exceeding it is overpressure). All of the above judgments are executed in real time, meaning that the compliance verification of each stress record is completed immediately upon receipt, without waiting for subsequent data.

[0034] The design goal of the first level (single-point pressure compliance detection) is to capture the most direct and urgent single-point pressure anomalies in the fire water supply system with the lowest latency and the most stringent real-time performance. The core feature of this level is "real-time verification of each record without window accumulation," meaning that for each pressure record received, the system immediately and independently completes a compliance judgment without relying on historical or future data, thereby ensuring millisecond-level anomaly detection capability and providing basic event input for subsequent high-level analysis.

[0035] Specifically, this level sets up three mutually exclusive and comprehensive exception handling logics: A. Network Collapse Detection: This detection applies to the main pipeline sensors. A network collapse is determined when the recorded pressure value falls below a preset main pipeline collapse threshold (e.g., 0.15 MPa), and this low-pressure state persists for a very short, preset duration (e.g., more than three consecutive records or lasting for several seconds). The "brief duration" condition is introduced here to avoid misjudgments caused by instantaneous pressure fluctuations (e.g., sensor noise or water flow impact). Once triggered, it indicates that the pipeline has lost its basic pressure-maintaining capacity, possibly due to main pump failure, severe pipeline leakage, or water supply interruption, representing the highest level of emergency.

[0036] B. End-point water supply failure determination: This determination only applies to sensors identified as "end-point test points" by the pre-processing module. Since the end-point test point is located at the most unfavorable location in the pipe network, its normal operating pressure is much lower than that of the main pipe. Therefore, an independent, lower specific lower threshold is used (e.g., ≥0.07 MPa for fire hydrant ends, ≥0.05 MPa for sprinkler ends). When the pressure of the end sensor is lower than this threshold, it means that the water supply to the most unfavorable point is insufficient, which will directly affect the fire extinguishing effect, but the rest of the system may still be basically normal. This determination does not require a continuous condition, because once the end pressure falls below the lower limit, even if it occurs momentarily, it indicates insufficient water supply capacity.

[0037] C. Overpressure Detection: This detection applies to all sensors, selecting the corresponding upper pressure threshold based on the system type determined by preprocessing (0.80 MPa for sprinkler systems and 1.00 MPa for fire hydrant systems). An overpressure alarm is immediately triggered when the recorded pressure exceeds this limit. Overpressure can lead to pipe rupture, loose connections, or accidental sprinkler discharge, all of which are considered critical anomalies. This detection method uses a single limit exceedance trigger mechanism because even a momentary overpressure event can cause severe damage.

[0038] Through the aforementioned three-tiered judgment logic, the first level achieves real-time "first line of defense" against the most urgent faults in the fire water supply system. All judgment results directly enter the diagnostic result aggregation module and are prioritized and pushed according to the urgency level. This level complements the subsequent levels (event-level, trend-level, and system-level) in terms of time scale: the first level focuses on instantaneous limit exceedances, while the subsequent levels focus on the change process, slow degradation, and structural anomalies, together forming a blind-spot-free, full-dimensional diagnostic chain.

[0039] (III) Second level: The dual threshold method combining absolute amplitude threshold and relative rate threshold is used to detect discrete pressure drop events, and sampling lag correction is performed on the pressure stabilizing pump start-up event.

[0040] S21: Detect sudden pressure drop events A dual-threshold method is used, meaning that a pressure drop event is triggered only when both of the following conditions are met simultaneously; it is not triggered if only one condition is met or neither condition is met: Amplitude condition: Within a preset sliding time window, the magnitude of the pressure drop exceeds an absolute threshold set for the sensor location; wherein, for the main pipeline sensor, the absolute threshold is a first pressure drop amplitude (e.g., ≥30kPa); for the end-point test sensor, the absolute threshold is a second pressure drop amplitude (e.g., ≥20kPa). Rate condition: Within the sliding time window, the instantaneous pressure drop rate exceeds a preset multiple (e.g., 5 times) of the current normal attenuation rate of the system; the preset multiple is the first rate multiple; wherein, the instantaneous drop rate is calculated by dividing the pressure difference between adjacent sampling points by the sampling time interval; the normal attenuation rate of the system is dynamically determined based on historical data of the sensor during recent normal operation periods (e.g., taking the median or mean of the historical attenuation rate distribution). The dual-threshold method, by simultaneously constraining the pressure drop amplitude and the rate of drop, can effectively distinguish between a real abnormal pressure drop and the pressure decay caused by normal pump circulation, thereby avoiding misjudging the decay of normal pump circulation as a sudden drop event. Compared with the single-rate detection method that only uses the rate threshold, this dual-threshold method can reduce false positive events by a preset percentage (e.g., 99.3%). S22: Sampling hysteresis correction for pump start-up events For pressure-stabilizing pump start-up events, sampling lag correction is performed: by analyzing the time offset characteristics between the pump start-up marker and the actual pressure rise edge in the pressure timing sequence, the apparent early start phenomenon caused by sampling delay is identified and restored to a normal start caused by sampling delay; this correction can correctly classify a preset proportion (e.g., about 37.9%) of apparent early starts as normal starts, avoiding false alarms; S23: Timeliness and Processing Method This layer employs a near real-time processing method, using a sliding time window to scan and detect the continuous pressure data stream window by window.

[0041] The second level (event-level anomaly detection) focuses on identifying discrete pressure anomaly events in fire-fighting water supply systems that are short-lived, significant in magnitude, but potentially submerged by normal pump circulation. The core design of this level lies in significantly improving the accuracy of event detection through dual constraints and physical correction, specifically in the following three aspects: S21: Dual-threshold mechanism for detecting sudden pressure drop events Traditional methods, which rely solely on instantaneous rate to determine sudden drops, are prone to misinterpreting pressure decay caused by the normal start-up and shutdown of a pressure-stabilizing pump as a leakage or rupture event, leading to a massive number of false positive alarms. The dual-threshold method proposed in this invention requires both amplitude and rate conditions to be met simultaneously; neither can be omitted. Amplitude conditions ensure that the detected pressure drop is significant in an engineering sense (e.g., ≥30 kPa for main lines and ≥20 kPa for terminal lines), filtering out minor pressure fluctuations; The rate condition requires that the instantaneous rate of pressure drop exceeds several times (e.g., 5 times) the recent normal decay rate of the system, thereby distinguishing between normal decay (such as slow leakage) and sudden drop (such as pipe rupture or valve accidental opening).

[0042] Due to the AND logic of the dual-threshold method, only events exhibiting both the characteristics of "significant pressure drop" and "rapid pressure drop" are considered abnormal. This design stems from the analysis of real-world failure physical processes: pressure drops caused by normal pump circulation are typically slow and limited in magnitude, while genuine pipeline ruptures or large-flow releases inevitably produce both significant and rapid pressure drops. Practical verification shows that compared to the single-rate method, this dual-threshold method reduces false positives by 99.3%, fundamentally solving the problem of alarm overload.

[0043] S22: Sampling hysteresis correction for pump start-up events In IoT systems, an unavoidable time lag exists between the sensor sampling time and the actual start-up time of the pressure-stabilized pump (due to communication delays, data buffering, etc.). This lag leads to an "apparent early start" phenomenon: the pump's start-up record appears abnormally early in the timeline, triggering unnecessary alarms. This invention analyzes the time lag characteristics between the pump start-up marker and the actual pressure rise edge to automatically identify false early starts caused by sampling delays and correct them as normal start-up events. Verification shows that this method can correctly classify approximately 37.9% of apparent early starts, avoiding false alarms caused by sampling system defects.

[0044] S23: Near Real-Time Sliding Window Processing This level does not require independent judgment for each record (as in the first level), but instead uses a sliding time window to scan the continuous data stream. The window length and step size can be set according to system characteristics (e.g., the window covers several pump cycle periods). This near real-time processing ensures timely detection of abnormal events while leveraging local data features within the window (such as rate calculation) to improve the robustness of the criteria, achieving a good balance between response latency and detection accuracy.

[0045] In summary, the second level, through a dual-threshold mechanism and sampling lag correction, achieves accurate identification of discrete pressure drop events and pump start-up events, significantly reducing the false alarm rate and providing maintenance personnel with highly reliable event-level alarms. Simultaneously, the results from this level (such as the actual time of the pressure drop and the corrected pump start-up record) are also included as part of the structured abnormal events, providing comprehensive evaluation for the diagnostic result aggregation module.

[0046] (iv) Third level: Establish a rolling baseline distribution of decay rate, and trigger leakage warning for abnormal decay rates that continuously exceed the upper limit of the baseline and for positive trends in the decay rate time series.

[0047] A: Establishment of rolling baseline distribution for attenuation rate Based on the pressure data collected by each sensor during its historical normal operation, a series of continuous attenuation rate sample intervals are calculated; the attenuation rate of each sample interval is determined by the rate of change of pressure over time within that interval. Once the accumulated number of valid sample intervals reaches the preset minimum sample size (e.g., N ≥ 30 sample intervals), a rolling baseline distribution of the decay rate is established, which includes the mean. and standard deviation ; The rolling baseline distribution is dynamically updated over time: each time a new sample interval is added, the value within the current time window is recalculated. and This is to reflect the normal attenuation characteristics of the system in the near future; B: Leakage warning triggering conditions Monitor the current decay rate in real time and compare it with the rolling baseline distribution; When the current decay rate continues to exceed the benchmark average Add the standard deviation of the preset multiple (e.g., 2x) When the current attenuation rate is greater than 1, the condition is met. +2 If this state continues for a preset duration or number of samples, a leakage warning will be triggered. The leakage early warning is used to indicate the presence of early, minor leaks in the pipeline network. This warning is issued before traditional fixed threshold alarms, providing an early warning time. C: Warning trigger condition for worsening leakage trend Perform rolling linear regression analysis on the time series of decay rate, with the rolling window length set to a preset duration (e.g., 30 days). Calculate the regression slope and its significance level p-value; When the regression slope is positive (i.e., the decay rate increases over time) and its significance level p-value is less than a preset threshold (e.g., p < 0.05), it is determined that the slope is positively significant, triggering an alarm for worsening leakage trend. D: Time Limit Explanation This level uses monthly calculations as the basic cycle and employs a preset rolling time window for data sampling and statistical analysis, balancing the sensitivity of trend capture with computational efficiency.

[0048] The aforementioned third level (trend-level degradation detection) focuses on the slow-developing, gradual degradation process in fire-fighting water supply systems that is difficult to detect using instantaneous or event-level methods, especially minor leaks in the pipeline network. The core design of this level is to utilize rolling statistical benchmarks and trend significance tests to achieve early warning of leaks and dynamic tracking of deterioration trends, issuing alarms before the fault develops to a serious level.

[0049] A. Adaptive establishment of rolling baseline distribution Because the normal attenuation rates vary among different fire protection systems and pipe sections (affected by pipe diameter, material, temperature, water usage habits, etc.), using a fixed threshold to determine leakage will inevitably lead to a large number of false alarms or missed alarms. This invention employs a data-driven rolling benchmark establishment method: based on historical pressure data from each sensor during normal operation, a series of continuous attenuation rate sample intervals are calculated (the attenuation rate of each interval is determined by the rate of pressure change over time within that interval). When the accumulated number of valid sample intervals reaches a preset minimum sample size (e.g., N≥30), a benchmark including the mean is established. and standard deviation The decay rate is a rolling baseline distribution. This baseline is not static but dynamically updated over time: each time a new sample interval is added, the decay rate within the current time window is recalculated. and This allows for continuous tracking of the system's recent normal degradation characteristics. This design automatically adapts to the slow drift in degradation rate caused by system aging, seasonal changes, or alterations in operating conditions.

[0050] B. Statistical Criteria for Leakage Early Warning When the current attenuation rate monitored in real time continues to exceed the benchmark average value Add the standard deviation of the preset multiple (e.g., 2x) When the current attenuation rate is greater than 1, the condition is met. +2 If this state persists for a preset duration or number of consecutive samples, a leakage warning will be triggered. Select " +2 "As an upper bound, it statistically corresponds to approximately 97.5% of the normal data quantiles, maintaining sensitivity to minor drifts while controlling the false positive rate. The 'continuous exceedance' condition is introduced to avoid triggering warnings due to short-term random fluctuations. The core advantage of this leakage warning lies in its earlier warning than traditional threshold alarms: when the decay rate has not yet reached any absolute safety lower limit, if it has significantly deviated from its historical benchmark, the system can issue a warning, thus giving maintenance personnel valuable advance response time."

[0051] C. Warning of worsening leakage trend Simple benchmark out-of-bounds analysis can only detect existing abnormal attenuation, but cannot determine whether the abnormality is stabilizing or continuing to worsen. Therefore, this invention further introduces a trend significance test: a rolling linear regression analysis is performed on the time series of the attenuation rate (the rolling window length is preset to 30 days), and the regression slope and its significance level p-value are calculated. When the slope is positive (i.e., the attenuation rate shows an upward trend over time) and the p-value is less than a preset threshold (e.g., p < 0.05), it is determined to be "positively significant," triggering a leakage trend deterioration alarm. This alarm indicates that leakage is not only occurring but also continuously worsening, requiring immediate intervention. By separating the judgment of "whether an anomaly exists" and "whether the anomaly is worsening," this level can provide more refined guidance for maintenance decisions.

[0052] D. Timeliness and Calculation Strategy Considering that the timescale for decay rate changes is typically measured in days or weeks, this level uses monthly calculations as the basic cycle, employing a 30-day rolling window for data sampling and statistical analysis. Compared to the real-time line-by-line detection of the first level and the near-real-time window scanning of the second level, this long-cycle design significantly reduces the computational burden while ensuring sufficient trend capture sensitivity, making it suitable for periodic execution in the cloud or on a server.

[0053] In summary, the third level, through a triple mechanism of rolling baseline distribution, statistical out-of-bounds early warning, and trend significance testing, achieves early detection and tracing of pipeline leaks, filling the gap in the diagnosis of progressive faults using traditional threshold methods. This level complements the real-time anomaly detection of the first and second levels on a time scale, together forming a complete diagnostic system ranging from milliseconds to monthly measurements.

[0054] (v) Fourth level: Obtain the results of water supply interruption determination within the same system and the results of cross-system independence verification, and incorporate them into the diagnostic results aggregation.

[0055] This level directly obtains the results of water supply interruption determination within the same system and the results of cross-system independence verification. Among them: The water supply interruption determination results within the same system include the connectivity status of the corresponding pipe section for each sensor, divided into three levels: normal connectivity, minor damage, and water supply interruption. Water supply interruption indicates that the corresponding pipe section has experienced a hydraulic disconnection from the main pipeline, which may be caused by factors such as the closure of control valves.

[0056] The results of cross-system independence verification include the determination of the independence between sensor pairs belonging to different fire protection systems (fire hydrant system and sprinkler system), which are divided into two types: normal independent operation (each system uses an independent pump set, and the pressure dynamics are independent of each other) or suspected shared pump configuration (the two systems may have undue shared water supply pipe sections or pump sets).

[0057] The results obtained above are used as the fourth-level diagnostic output and incorporated into the diagnostic result aggregation module. This level, together with the first three levels (single point, event, trend), forms a complete diagnostic matrix, with the fourth level specifically responsible for the identification and integrated reporting of system-level structural anomalies.

[0058] (vi) Diagnostic result aggregation: The output of the four levels is used to form a structured list of abnormal events. After sorting by severity, maintenance work order suggestions are output and pushed to the corresponding role's alarm notification.

[0059] A: Construction of a structured list of exception events Collect and integrate the raw abnormal event data output from the first, second, third, and fourth levels; Deduplication and merging of abnormal events from different levels: If the same physical fault is detected by multiple levels at the same time, it is merged into a single comprehensive anomaly record, and the source of all involved levels is marked. Generate a structured record for each exception event, which must contain at least the following fields: Types: Classified according to the diagnostic level to which the abnormality belongs, including first-level class, second-level class, third-level class, and fourth-level class; Time of occurrence: The timestamp when the anomaly was first detected; Duration: The duration from the first detection of the anomaly to the recovery or the current moment; Severity: Divided into two levels: Warning and Critical. The Critical level corresponds to anomalies that may cause immediate system failure or serious security risks; the Warning level corresponds to anomalies that require attention but do not affect the basic functions of the system at present. Recommended handling: Pre-set or dynamically generated maintenance suggestion text for this type of anomaly, including inspection parts, operation steps, priority prompts, etc.; B: Generation and sorting of maintenance work order suggestions Based on all structured anomaly records, they are sorted in descending order of severity: all critical anomalies are listed before warning anomalies. Within the same severity level, further sorting can be performed according to duration, time of occurrence, or preset weighting factors; The sorted set of abnormal records is converted into a maintenance work order suggestion list. Each work order suggestion includes an abnormality summary, recommended handling, and suggested response time limit. C: Alarm push notification Based on the anomaly type and severity suggested in the maintenance work order, a preset role distribution strategy is matched; Alarm notifications are sent to the terminal devices of the corresponding roles through at least one communication channel; For critical level anomalies, a repeat push mechanism can be added.

[0060] The aforementioned diagnostic result aggregation module serves as the terminal output and decision support link in the entire four-level intelligent diagnostic method. Its core task is to unify, merge, and sort the heterogeneous, scattered, and potentially overlapping abnormal events generated in the first four levels, ultimately transforming them into executable work orders and tiered alarms that can directly guide maintenance actions. The design of this module reflects the engineering principle that "diagnostic results must serve actual operation and maintenance."

[0061] A. Construction of a structured list of exception events Because the four levels output abnormal events from different time scales, spatial ranges, and analytical dimensions, their original data formats vary (e.g., the first level outputs single-point limit exceedance records, the second level outputs sudden drop event windows, the third level outputs attenuation rate drift warnings, and the fourth level outputs correlation coefficient anomalies). The aggregation module first collects all these original abnormal events and then performs deduplication and merging processing: when the same physical fault (such as leakage in a pipe section) is detected simultaneously by both the attenuation rate exceeding the limit in the third level and the correlation decrease in the fourth level, the system merges them into a single comprehensive abnormal record and marks the sources of all involved levels. This merging avoids generating multiple duplicate alarms for the same problem, reducing the processing burden on maintenance personnel.

[0062] For each merged anomaly event, a structured record containing five core fields is generated: Type (indicating the level and specific anomaly subclass), Occurrence Time (first detection timestamp), Duration (used to assess urgency), Severity (divided into "Critical" and "Warning" levels), and Recommended Action (preset or dynamically generated maintenance suggestions for this anomaly type, including inspection locations, operating procedures, priority prompts, etc.). The "Critical" level corresponds to anomalies that may cause immediate system failure or serious security risks (such as network crashes, water supply interruptions, overpressure, and shared pump configuration anomalies); the "Warning" level corresponds to anomalies that require attention but do not currently affect basic system functions (such as minor damage and leak warnings). This hierarchical approach allows the operations and maintenance team to quickly distinguish handling priorities.

[0063] B. Generation and sorting of maintenance work order suggestions After the structured list of abnormal events is generated, the aggregation module sorts them in descending order of severity: all "critical" level events are placed before "warning" level events, ensuring that the most critical issues are addressed first. Within the same severity level, further sorting can be performed based on duration (longer-lasting anomalies have higher priority), occurrence time (earlier events have higher priority), or preset weighting factors (e.g., certain equipment areas are more important). The sorted set of anomaly records is converted into a maintenance work order suggestion list. Each work order suggestion includes an anomaly summary and recommended handling, as well as a suggested response time limit (e.g., "critical" level requires a response within 2 hours, and "warning" level can be arranged within 24 hours). This sorting and time constraint directly aligns with actual fire protection maintenance management processes, significantly improving the usability of the work order system.

[0064] C. Role-based distribution of alert push notifications After a work order suggestion is generated, the system matches a preset role-based distribution strategy according to the anomaly type and severity. For example, critical anomalies such as "network crash" and "water supply interruption" need to be pushed to the fire supervisor, maintenance team leader, and on-duty engineer simultaneously; while warning-level anomalies such as "leakage warning" may only be pushed to the maintenance engineer and routine inspection personnel. Push channels support at least one communication method, such as SMS, mobile application push, email, or in-system messages, to ensure timely information delivery. For "critical" level anomalies, a repeat push mechanism can be added (e.g., resending every 15 minutes until the anomaly is confirmed and processed) to avoid delays due to missed messages. Through a role-based, multi-channel, and repeatable push design, the aggregation module seamlessly integrates diagnostic results into the user's existing operation and maintenance responsibility system.

[0065] In summary, the diagnostic results aggregation module transforms multi-level heterogeneous diagnostic data into standardized, executable, role-oriented maintenance instructions. It solves the critical engineering loop problem of "what to do after a problem is diagnosed," enabling the technological innovations of the first four levels to ultimately translate into improved operational efficiency and fire safety levels.

[0066] Second Embodiment This embodiment uses the comprehensive diagnosis of a 7-sensor system in a chemical plant as a specific example to further explain the technical content of the present invention, as detailed below: A chemical plant's fire protection system has a total of 7 pressure sensors (5 for fire hydrant systems and 2 for sprinkler systems). The following are the diagnostic results for each level: Adaptive parameter selection: Among the 5 fire hydrant sensors, 1 (#4) had a continuous pressure of only 0.062MPa, and was automatically reclassified as an end-point test point; the median cycle time of the 2 sprinkler system sensors was 30.8 hours, which confirmed it as a sprinkler type.

[0067] Level 1: Sprinkler system sensor #6 repeatedly recorded 0.000 MPa (far below the 0.15 MPa crash threshold), resulting in 66 crash events, which was determined to be a network crash. Another sensor, #7, labeled "sprinkler end," had an actual pressure between 0.875 and 1.492 MPa, exceeding the sprinkler system's upper limit of 0.80 MPa (78% of records exceeded the limit), which was determined to be an incorrect sensor location label (it was actually a main pipeline sensor), triggering an overpressure alarm.

[0068] Level 2: The fire hydrant system experienced frequent pressure drop events after December 14, with amplitudes of 38-52 kPa (exceeding the absolute threshold of 30 kPa) and a rate exceeding 8 times the baseline attenuation rate, which were identified as 10 real pressure drop events (not sampling noise).

[0069] Level 3: If the current decay rate of the fire hydrant system is continuously higher than the upper limit of its rolling baseline distribution (i.e., exceeding the sum of the baseline mean and the preset multiple standard deviation), a leakage warning is triggered; if a rolling linear regression is performed on the decay rate time series, and the regression slope is positive and the significance level p value is less than the preset threshold (p=0.03 in this example, significant), a leakage trend deterioration alarm is triggered.

[0070] Level 4: The correlation coefficients of the 3 sensors in the fire hydrant system with the other 4 sensors are all between 0.04 and 0.09 (below 0.30), indicating a water supply interruption - the pipe section to which the 3 sensors belong is suspected to be hydraulically interrupted due to the closure of the control valve.

[0071] Overall output: Four types of anomalies are identified (network crash, overpressure / location error, pipeline leakage, and water supply interruption). Maintenance work orders are generated according to the severity, with priority given to handling the two most severe anomalies, water supply interruption and network crash.

[0072] Third Embodiment like Figure 2 As shown, this embodiment provides a four-level intelligent diagnostic system for fire water supply system operation status, used to execute the four-level intelligent diagnostic method for fire water supply system operation status as described in the first embodiment, comprising: The adaptive parameter selection module is used to automatically identify the fire protection system type and sensor installation location for each sensor, and select the corresponding set of diagnostic parameters accordingly. The first-level single-point detection module is used to verify the single-point compliance threshold for each pressure record and to determine network crashes, end-point water supply failures, and overpressure events. The second-level event detection module is used to detect discrete pressure drop events using a dual-threshold method that combines absolute amplitude thresholds and relative rate thresholds, and to perform sampling hysteresis correction on pressure stabilizing pump start-up events. The third-level trend analysis module is used to establish a rolling baseline distribution of decay rate and to trigger leakage warnings for abnormal decay rates that continuously exceed the upper limit of the baseline and for positive trends in the decay rate time series. The fourth-level synchronization analysis module obtains the water supply interruption judgment results within the same system and the cross-system independence verification results, and incorporates them into the diagnostic results aggregation; The diagnostic results aggregation module is used to construct a structured list of abnormal events from the four levels of output, sort them by severity, output maintenance work order suggestions, and push alarm notifications to the corresponding roles.

[0073] A computer-readable storage medium stores computer code that, when executed, performs the methods described above. Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0074] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

[0075] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] It should be noted that the above embodiments can be freely combined as needed. The above description is only a preferred embodiment of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A four-level intelligent diagnostic method for the operating status of a fire-fighting water supply system, characterized in that, This includes the following pre-processing, four diagnostic levels, and diagnostic result aggregation: Pre-processing: Automatically identify the fire protection system type and sensor installation location for each sensor, and select the corresponding set of diagnostic parameters accordingly; Level 1: Verify the single-point compliance threshold for each pressure record and determine network crashes, end-point water supply failures, and overpressure events; The second level: a dual-threshold method combining absolute amplitude threshold and relative rate threshold is used to detect discrete pressure drop events, and sampling hysteresis correction is performed on the pressure stabilizing pump start-up event; The third level: Establish a rolling baseline distribution of decay rate, and trigger leakage early warning for abnormal decay rates that continuously exceed the upper limit of the baseline and for positive trends in the decay rate time series. Fourth level: Obtain the results of water supply interruption determination within the same system and the results of cross-system independence verification, and incorporate them into the diagnostic results aggregation; Diagnostic result aggregation: The outputs from the four levels are combined into a structured list of abnormal events, sorted by severity, and then maintenance work order suggestions are output and pushed to the corresponding role's alarm notification.

2. The four-level intelligent diagnostic method for the operating status of a fire-fighting water supply system according to claim 1, characterized in that, In the pre-processing, the fire protection system type and sensor installation location are automatically identified for each sensor, and the corresponding set of diagnostic parameters is selected accordingly, specifically: A: System type identification: Automatic classification based on the pressure time-series statistical characteristics of each sensor within a preset time window, without relying on metadata; the statistical characteristics include at least the median of the pump cycle and the statistical distribution of the sampling interval; When the median pump cycle exceeds the first preset duration and the sampling interval is greater than or equal to the first sampling interval threshold, it is automatically identified as a sprinkler system. When the median pump cycle is within the second preset time range and the sampling interval is less than or equal to the second sampling interval threshold, it is automatically identified as a fire hydrant system. B: Sensor location inference: Automatically infers the installation location based on the continuous pressure range of each sensor during the continuous monitoring period; If the pressure value of a certain sensor is continuously lower than the first pressure threshold during the continuous monitoring period, the sensor will be automatically reclassified as an end test point and the corresponding end pressure compliance standard will be selected; among the end pressure compliance standards, the end pressure of fire hydrants is not lower than the second pressure threshold and the end pressure of sprinklers is not lower than the third pressure threshold. Otherwise, treat the sensor as a main network sensor and select the corresponding pressure diagnostic parameter set for the main network.

3. The four-level intelligent diagnostic method for the operating status of a fire-fighting water supply system according to claim 1, characterized in that, In the first level, each pressure record is verified against a single-point compliance threshold to determine network crashes, end-point water supply failures, and overpressure events, specifically: Each pressure record is independently verified in real time, without any accumulation over a time window, including the following types of judgments: A: Network crash determination: When the pressure record value is lower than the main pipeline crash threshold and the low pressure state continues for a preset short duration, it is determined to be a network crash; the main pipeline crash threshold corresponds to the minimum allowable pressure to maintain the basic integrity of the pipeline network; B: End-point water supply failure determination: When the sensor has been identified as an end-point test point by the preprocessing, if its pressure record value is lower than the specific lower limit threshold set for the end position, it is determined that the end-point water supply has failed. The specific lower limit threshold is preset according to different system types. C: Overpressure determination: When the pressure recorded value exceeds the pressure upper limit threshold of the corresponding system type, it is determined to be overpressure; among them, the pressure upper limit threshold of the sprinkler system is the first upper limit value, and the pressure upper limit threshold of the fire hydrant system is the second upper limit value; All of the above judgments are executed in real time, meaning that the compliance verification of each stress record is completed immediately upon receipt, without waiting for subsequent data.

4. The four-level intelligent diagnostic method for the operating status of a fire-fighting water supply system according to claim 1, characterized in that, In the second level, a dual-threshold method combining absolute amplitude threshold and relative rate threshold is used to detect discrete pressure drop events, and sampling hysteresis correction is applied to the pressure stabilizing pump start-up event, specifically as follows: S21: Detect sudden pressure drop events A dual-threshold method is used, meaning that a pressure drop event is triggered only when both of the following conditions are met simultaneously; it is not triggered if only one condition is met or neither condition is met: Amplitude condition: Within a preset sliding time window, the magnitude of the pressure drop exceeds an absolute threshold set for the sensor location; wherein, for the main pipeline sensor, the absolute threshold is a first pressure drop amplitude; for the end-point test sensor, the absolute threshold is a second pressure drop amplitude. Rate condition: Within the sliding time window, the instantaneous pressure drop rate exceeds a preset multiple of the current normal attenuation rate of the system; the preset multiple is the first rate multiple; wherein, the instantaneous drop rate is calculated by dividing the pressure difference between adjacent sampling points by the sampling time interval; the normal attenuation rate of the system is dynamically determined based on historical data of the sensor during its recent normal operation period; The dual-threshold method, by simultaneously constraining the pressure drop amplitude and the rate of drop, can effectively distinguish between a real abnormal pressure drop and the pressure decay caused by normal pump circulation, thereby avoiding misjudging the decay of normal pump circulation as a sudden drop event. Compared with the single-rate detection method that only uses the rate threshold, this dual-threshold method can reduce false positive events by a preset proportion. S22: Sampling hysteresis correction for pump start-up events For pressure stabilizing pump start-up events, sampling lag correction is performed: by analyzing the time offset characteristics between the pump start-up marker and the actual pressure rise edge in the pressure timing sequence, the apparent early start phenomenon caused by sampling delay is identified and restored to a normal start caused by sampling delay; this correction can ensure that a preset proportion of apparent early starts are correctly classified as normal starts, avoiding false alarms. S23: Timeliness and Processing Method This layer employs a near real-time processing method, using a sliding time window to scan and detect the continuous pressure data stream window by window.

5. The four-level intelligent diagnostic method for the operating status of a fire-fighting water supply system according to claim 1, characterized in that, In the third level, a rolling baseline distribution of attenuation rate is established. Leakage warnings are triggered for abnormal attenuation rates that continuously exceed the upper limit of the baseline, as well as for positive trends in the attenuation rate time series. Specifically: A: Establishment of rolling baseline distribution for attenuation rate Based on the pressure data collected by each sensor during its historical normal operation, a series of continuous attenuation rate sample intervals are calculated; the attenuation rate of each sample interval is determined by the rate of change of pressure over time within that interval. Once the accumulated number of valid sample intervals reaches the preset minimum sample size, a rolling baseline distribution of the decay rate is established, which includes the mean. and standard deviation ; The rolling baseline distribution is dynamically updated over time: each time a new sample interval is added, the value within the current time window is recalculated. and This is to reflect the normal attenuation characteristics of the system in the near future; B: Leakage warning triggering conditions Monitor the current decay rate in real time and compare it with the rolling baseline distribution; When the current decay rate continues to exceed the benchmark average Add the standard deviation of the preset multiple If the state continues for a preset duration or number of samples, a leakage warning will be triggered. The leakage early warning is used to indicate the presence of early, minor leaks in the pipeline network. This warning is issued before traditional fixed threshold alarms, providing an early warning time. C: Warning trigger condition for worsening leakage trend A rolling linear regression analysis was performed on the time series of decay rate, with the rolling window length set to a preset duration. Calculate the regression slope and its significance level p-value; When the regression slope is positive and its significance level p-value is less than the preset threshold, it is determined that the slope is positively significant, triggering an alarm for worsening leakage trend. D: Time Limit Explanation This level uses monthly calculations as the basic cycle and employs a preset rolling time window for data sampling and statistical analysis, balancing the sensitivity of trend capture with computational efficiency.

6. The four-level intelligent diagnostic method for the operating status of a fire-fighting water supply system according to claim 1, characterized in that, In the fourth level, the results of the water supply interruption determination within the same system and the cross-system independence verification are obtained, specifically as follows: A: The results of water supply interruption determination within the same system include the connection status of the corresponding pipe section for each sensor, including at least the normal connection status, the slightly damaged status, and the water supply interruption event. B: Cross-system independence verification results, including the independence determination results between sensor pairs belonging to different fire protection systems, specifically whether they are in normal independent operation or suspected shared pump configuration. The results obtained above are used as the fourth-level diagnostic output, and the diagnostic results are aggregated with those of the first, second, and third levels of abnormal events.

7. The four-level intelligent diagnostic method for the operating status of a fire-fighting water supply system according to claim 1, characterized in that, In the aggregation of diagnostic results, the outputs from the four levels are combined into a structured list of abnormal events. After being sorted by severity, maintenance work order suggestions are output and pushed to the corresponding role's alarm notification. Specifically: A: Construction of a structured list of exception events Collect and integrate the raw abnormal event data output from the first, second, third, and fourth levels; Deduplication and merging of abnormal events from different levels: If the same physical fault is detected by multiple levels at the same time, it is merged into a single comprehensive anomaly record, and the source of all involved levels is marked. Generate a structured record for each exception event, which must contain at least the following fields: Types: Classified according to the diagnostic level to which the abnormality belongs, including first-level class, second-level class, third-level class, and fourth-level class; Time of occurrence: The timestamp when the anomaly was first detected; Duration: The duration from the first detection of the anomaly to the recovery or the current moment; Severity: Divided into two levels: Warning and Critical. The Critical level corresponds to anomalies that may cause immediate system failure or serious security risks; the Warning level corresponds to anomalies that require attention but do not affect the basic functions of the system at present. Recommended handling: Pre-set or dynamically generated maintenance suggestion text for this type of anomaly, including inspection parts, operation steps, priority prompts, etc.; B: Generation and sorting of maintenance work order suggestions Based on all structured anomaly records, they are sorted in descending order of severity: all critical anomalies are listed before warning anomalies. Within the same severity level, further sorting can be performed according to duration, time of occurrence, or preset weighting factors; The sorted set of abnormal records is converted into a maintenance work order suggestion list. Each work order suggestion includes an abnormality summary, recommended handling, and suggested response time limit. C: Alarm push notification Based on the anomaly type and severity suggested in the maintenance work order, a preset role distribution strategy is matched; Alarm notifications are sent to the terminal devices of the corresponding roles through at least one communication channel; For critical level anomalies, a repeat push mechanism can be added.

8. A four-level intelligent diagnostic system for the operating status of a fire-fighting water supply system, used for executing the four-level intelligent diagnostic method for the operating status of a fire-fighting water supply system as described in any one of claims 1-7, characterized in that, include: The adaptive parameter selection module is used to automatically identify the fire protection system type and sensor installation location for each sensor, and select the corresponding set of diagnostic parameters accordingly. The first-level single-point detection module is used to verify the single-point compliance threshold for each pressure record and to determine network crashes, end-point water supply failures, and overpressure events. The second-level event detection module is used to detect discrete pressure drop events using a dual-threshold method that combines absolute amplitude thresholds and relative rate thresholds, and to perform sampling hysteresis correction on pressure stabilizing pump start-up events. The third-level trend analysis module is used to establish a rolling baseline distribution of decay rate and to trigger leakage warnings for abnormal decay rates that continuously exceed the upper limit of the baseline and for positive trends in the decay rate time series. The fourth-level synchronization analysis module is used to obtain the water supply interruption judgment results within the same system and the cross-system independence verification results, and incorporate them into the diagnostic results aggregation; The diagnostic results aggregation module is used to construct a structured list of abnormal events from the four levels of output, sort them by severity, output maintenance work order suggestions, and push alarm notifications to the corresponding roles.

9. A computer device, characterized in that, The device includes a memory and one or more processors, wherein the memory stores computer code that, when executed by the one or more processors, causes the one or more processors to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer code, and when the computer code is executed, the method as described in any one of claims 1 to 7 is performed.

Images