A water taking metering monitoring and abnormal early warning system based on internet of things

Abstract

The present application belongs to the technical field of Internet of Things and industrial water intake monitoring, and is used to solve the problem that the prior art cannot dynamically identify the existence of one-way elements in the pipe network and their state by using sensor data flow itself, and correct the causal attribution logic between pressure and flow. Specifically, it is a water intake metering monitoring and abnormal warning system based on Internet of Things, which comprises an event detection module, a matching analysis module, a block identification module and a result output module. Through the cooperation of the adaptive fluctuation reference value and the dynamic survival window in the event detection module, and the distinction between the responsive matching rule and the non-responsive matching rule in the matching analysis module, the non-responsive event caused by the operation downstream of the check valve can be identified without clock synchronization and manual annotation of the topology. Through the daily statistics and periodic self-learning of the direction advantage ratio in the block identification module, the upstream and downstream topology relationship between sensors is passively inferred, and the existence of one-way block elements is automatically marked.

Description

Technical Field

[0001] This invention belongs to the field of Internet of Things and industrial water intake monitoring technology, specifically an Internet of Things-based water intake metering monitoring and anomaly early warning system. Background Technology

[0002] Water intake metering and monitoring systems are widely deployed in industrial water use scenarios. By installing flow meters and pressure sensors on water intake pipelines, they collect data in real time and upload it to a server to monitor whether water consumption exceeds limits, whether pipelines are leaking, and whether equipment is malfunctioning. When significant changes in pressure and flow are observed simultaneously in the system, existing technologies typically use the order of these changes to determine the cause of the anomaly: if the flow rate changes before the pressure, it is determined to be due to user-initiated operation; if the pressure changes before the flow rate, it is determined to be due to external network disturbance or pipeline rupture.

[0003] This time-series causal analysis method is generally effective in simple straight pipes; however, industrial water intake networks commonly contain unidirectional blocking elements such as check valves and water hammer eliminators, as well as multi-branch parallel structures, which block the propagation path of pressure waves and distort the waveform. For example, when a check valve closes downstream, the reverse pressure wave cannot pass through the check valve to reach the upstream pressure sensor, causing the system to observe a decrease in flow rate but no change in pressure, thus misjudging it as a flow sensor malfunction or pipe breakage. Water hammer eliminators absorb the pressure spikes generated by valve closure, making the downstream pressure waveform smooth, so the system cannot identify the safety hazards caused by the valve actuator closing too quickly. Further complicating matters, minor leaks in branch pipes often do not cause observable changes in the main pipe pressure, rendering conventional time-series judgments completely ineffective.

[0004] Therefore, how to dynamically identify the existence and state of unidirectional components in the pipeline network using sensor data streams and correct the causal attribution logic between pressure and flow rate has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide an Internet of Things-based water intake metering monitoring and anomaly early warning system to solve the problem that existing technologies cannot use sensor data streams to dynamically identify the existence and status of unidirectional components in the pipeline network, and thereby correct the causal attribution logic between pressure and flow.

[0006] The technical problem to be solved by this invention is: how to provide an IoT-based water intake metering monitoring and anomaly early warning system that can dynamically identify the existence and state of unidirectional elements in the pipeline network using sensor data streams and thereby correct the causal attribution logic between pressure and flow.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A water intake metering monitoring and anomaly early warning system based on the Internet of Things includes:

[0009] Event detection module: acquires real-time data streams reported by multiple sensors within the same hydraulic unit; for each sensor, dynamically detects candidate water intake events in the real-time data stream based on its historical data fluctuation amplitude; broadcasts the detected candidate water intake events to other sensors within the hydraulic unit, and maintains a queue of suspenseful events locally on each sensor.

[0010] Pairing Analysis Module: When the target sensor generates a new candidate water intake event, it attempts to pair the new candidate water intake event with the event in the suspense event queue according to the event sequence number. The pairing rules are based on the physical constraints of industrial water intake. If the pairing is successful, a causal link record is generated. The causal link record includes at least the upstream sensor identifier, the downstream sensor identifier, the event sequence number difference, and the causal type.

[0011] Obstruction identification module: Based on the accumulated causal link records, it counts the number of bidirectional event propagation between each pair of sensors, calculates the directional dominance ratio, and infers the upstream and downstream topological relationship between sensors and identifies the existence of unidirectional obstruction elements based on the directional dominance ratio.

[0012] Results output module: Based on the upstream and downstream topology, identify hidden pressure leaks, check valve blockage events, and external disturbances, and output corresponding abnormal warning information.

[0013] The present invention has the following beneficial effects:

[0014] 1. By combining the adaptive fluctuation benchmark value and dynamic survival window in the event detection module, and by distinguishing between response pairing rules and non-response pairing rules in the pairing analysis module, the system can accurately identify non-response events caused by downstream operations of check valves without clock synchronization or manual topology labeling, and classify them as check valve blocking events. This avoids the problem of misjudging normal valve closing operations as flow sensor failure or pipeline breakage in the existing technology, and significantly reduces the false alarm rate.

[0015] 2. By using the daily statistics and periodic self-learning of the directional dominance ratio in the barrier identification module, the system can passively infer the upstream and downstream topological relationships between sensors using only the internal operations and external disturbances that occur naturally during daily operation, and automatically mark the presence of unidirectional barrier elements when the reverse count is below the threshold; at the same time, by monitoring the historical trend of the directional dominance ratio, it can detect topological changes such as check valve failure, human removal or illegal bypass installation online, and realize non-sensory early warning of changes in pipeline structure.

[0016] 3. By combining the extraction of flow rate change rate, pressure response delay points, pre-pressure disturbance and waveform tail in the operation fingerprint analysis module with historical pattern comparison, and comparing the number of upstream and downstream pressure rise points and peak levels in the water hammer analysis module, the system can identify the existence of water hammer eliminators and monitor the wear trend of valve actuators without the need for additional sensors. Thus, even when the energy dissipation element masks the pressure peak, it can still warn of the risk of excessive water hammer caused by too fast valve closure, filling the gap in valve status monitoring under energy dissipation interference in the existing technology.

[0017] 4. Through the two-stage circuit breaker mechanism of cross-sensor aggregation analysis and single-sensor directional sequence alternation mode in the result output module, the system strictly distinguishes between global disturbance events, sensor noise drift and local real anomalies, avoiding false alarms caused by overall fluctuations in the external pipeline network or isolated events of multiple sensors in the same direction; at the same time, by combining the joint judgment of branch leakage candidate conditions and the lack of paired records of main pipe pressure, the early identification of slow leakage in the pressure concealment state is realized, overcoming the leakage underreporting defects caused by the pressure response being masked in the existing technology. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a system block diagram of Embodiment 1 of the present invention;

[0020] Figure 2 This is a flowchart of the system operation method according to Embodiment 1 of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the field of industrial water intake metering and monitoring, water supply networks typically consist of basic monitoring units composed of flow sensors on the water source side and pressure sensors on the user side. Abnormal events are identified by comparing the temporal sequence of changes in the data from these two sensors. Existing technologies generally assume that pressure waves can propagate freely in both directions within the pipeline, thus establishing a simplified causal attribution rule: a change in flow rate followed by a pressure response indicates active user operation; a change in pressure followed by a flow rate response indicates external disturbance or pipeline leakage. This linear time-series comparison method is generally effective in unobstructed straight pipelines.

[0023] However, industrial water intake networks widely contain unidirectional blocking elements such as check valves and water hammer eliminators, as well as asymmetrical topologies with multiple branches in parallel. From a fluid dynamics perspective, the physical essence of a check valve is to allow pressure waves to propagate only in the direction of water flow, while completely cutting off reverse pressure waves. Water hammer eliminators, on the other hand, absorb pressure spikes through air bladders or piston structures, causing significant distortion and broadening of the pressure waveform during propagation. When a network structure with these physical characteristics is introduced, the causal relationship between pressure and flow rate is systematically disrupted. Specifically, when a valve closure occurs downstream of the check valve, the reverse pressure wave cannot pass through the check valve to reach the upstream pressure sensor, resulting in a contradictory phenomenon in the data stream where flow rate decreases while pressure remains constant. According to the existing timing comparison logic, the system will misjudge this normal operation as a flow sensor malfunction or pipeline breakage. When a water hammer eliminator is present in the system, its energy dissipation effect flattens the pressure waveform, which should be steep. The system fails to detect pressure spikes and therefore misses the water hammer safety hazard caused by the valve actuator closing too quickly.

[0024] In scenarios where multiple branches share a main pipe, minor leaks in branch pipes often only cause a slow increase in flow rate in that branch, while the pressure in the main pipe is masked by the superposition of water usage fluctuations in other branches, and the pressure sensor hardly produces any observable changes. At this time, the physical correspondence between flow rate changes and pressure responses is completely lost, and existing causal attribution rules cannot distinguish between slow leaks and normal process fluctuations.

[0025] If the above problems are not solved, the industrial water intake monitoring system will continue to lose its ability to determine the true state of the pipeline network; when the check valve blocks the reverse pressure wave, normal valve closing operations will be repeatedly falsely reported as equipment failures, causing alarm fatigue among maintenance personnel and leading to the blocking of critical warnings; when the water hammer eliminator masks pressure spikes, the progressive wear of the valve actuator cannot be detected early, until the water hammer exceeds the standard and causes the pipeline to burst; when branch leaks are in a pressure-hidden state, small leaks will continue to expand, eventually leading to foundation erosion or water waste; it is evident that existing technologies cannot dynamically identify the location and energy dissipation characteristics of unidirectional blocking elements using the data flow itself without prior topological information, thus failing to establish a pressure-flow attribution logic consistent with physical causality.

[0026] Example 1: As Figure 1-2 As shown, an IoT-based water intake metering monitoring and anomaly early warning system includes:

[0027] Event Detection Module: Acquires real-time data streams reported by multiple sensors within the same hydraulic unit; for each sensor, dynamically detects candidate water intake events in the real-time data stream based on its historical data fluctuation amplitude; broadcasts the detected candidate water intake events to other sensors within the hydraulic unit, and maintains a queue of pending events locally on each sensor; a hydraulic unit refers to a pipeline system with hydraulic connectivity.

[0028] In this embodiment, the event detection module continuously receives real-time data streams reported by multiple sensors within the same hydraulic unit. Each sensor has a unique identifier, and the sensor types include pressure sensors and flow sensors. The sensors report data independently according to their respective sampling periods. Each data packet contains the sensor identifier, the value, and the device's internal timestamp. However, this embodiment does not require clock synchronization between the sensors, nor does it require that the sampling periods be strictly equal.

[0029] The event detection module maintains a circular buffer for each sensor to store all data points from that sensor within the most recent hour; each data point records two fields: global reception sequence number. The sum of values ​​is V; where the global reception sequence number is assigned by the server in the order in which the data packets arrive, monotonically increasing from the integer 1. This sequence number does not depend on the sampling time or timestamp of the sensor, but only reflects the order in which the data arrives at the server; the capacity of the circular buffer is preset to 3600, which is sufficient to hold the data points of a sensor with a sampling interval of 1 second within one hour; for sensors with a larger sampling interval, the actual number of storage points is less than the capacity value, but this does not affect subsequent calculations.

[0030] Inside the circular buffer, the event detection module calculates the maximum absolute value of the difference between adjacent data points of the sensor in real time, and records it as the fluctuation reference value of the sensor. Specifically, whenever a new data point is stored in the buffer, the module iterates through all adjacent data point pairs in the buffer. Assuming there are M data points, there are M-1 pairs of adjacent points. For the i-th pair of adjacent points, their values ​​are respectively... and Calculate the absolute value of the difference And take the maximum value among them. ;

[0031] If there are fewer than two data points in the buffer, the fluctuation baseline value is temporarily set to 0, waiting for subsequent data accumulation. This fluctuation baseline value reflects the maximum signal fluctuation amplitude of the sensor in the most recent hour and has an adaptive characteristic: when the sensor signal is stable, the baseline value is small and the sensitivity of subsequent detection is high; when the sensor signal fluctuates greatly due to drastic changes in operating conditions, the baseline value increases accordingly, and the detection threshold is automatically raised, thereby avoiding normal fluctuations from being misjudged as events.

[0032] To avoid false triggers due to insufficient data, when the number of adjacent data point pairs in the buffer is less than 2, the fluctuation baseline value is adjusted. It is set to a preset default positive number, which is one percent of the sensor's range; after the buffer accumulates at least 10 data points, it switches to the actual calculated maximum difference; if the sensor has not obtained any historical data during its initial operation, the event detection module will not generate candidate water intake events and will wait for at least one minute of data accumulation.

[0033] Based on the aforementioned fluctuation baseline value, the event detection module determines the candidate water intake event for each newly arrived data point; let the value of the current data point be... The value of the previous data point is The absolute value of the difference ;like If the result is positive, it is considered a candidate water intake event. It should be noted that the coefficient 0.5 is selected based on engineering experience: this value ensures that small but meaningful changes (such as flow rate rise caused by slow leakage) can be captured when the signal is stable, and provides sufficient noise suppression margin when the signal fluctuates violently.

[0034] Once a candidate water intake event is identified, the event detection module further determines the direction and magnitude of the event's change; the direction of change is determined by... and The comparison directly yields: if If it is "increase", then it is "decrease"; otherwise, the magnitude category is determined by the ratio. To determine the magnitude: if r > 0.8, it is marked as "significant"; if 0.5 < r ≤ 0.8, it is marked as "normal". The magnitude is divided into "significant" and "normal" because in the subsequent pairing rules, only significant and normal changes are considered meaningful water intake operation events, while changes below 0.5 times the fluctuation baseline value are considered background noise and do not generate events. Each candidate water intake event is recorded as a data structure containing sensor identifier, sensor type, global receiver sequence number, change direction, and magnitude category.

[0035] During system initialization, the measurement range (scale) of each sensor is pre-configured as a device attribute and stored in the server; for example, the flow sensor's range is 0–1000 m.3 / h, the pressure sensor's range is 0~1.0MPa; when insufficient buffer data causes fluctuations in the reference value If calculation is not possible, use one-hundredth of the sensor's range as a temporary fluctuation reference value; after accumulating 10 data points, switch to the actual calculated value.

[0036] After generating candidate water intake events for the local sensor, the event detection module broadcasts the event to all other sensors within the same hydraulic unit. The broadcast content only includes the key feature fields of the event, excluding the raw values. Simultaneously, each sensor maintains a local queue of suspenseful events to store events broadcast from other sensors that have not yet received a response from the local sensor. Each suspenseful event is assigned a survival window upon entering the queue, and the end number of this window is dynamically determined by the data arrival rate of the target sensor itself.

[0037] The survival window allocation process is as follows: The event detection module first obtains the global sequence number of the ten most recently reported data points from the target sensor. (Arranged in chronological order), calculate the difference in global sequence number between two consecutive reports. Where j = 1, 2, ..., 9; then the median of these nine differences is taken as the data arrival interval reference of the sensor. Specifically, the difference sequence is sorted from smallest to largest, and the 5th value is taken as the median (because there are 9 differences, and the median is in the 5th position). For example, if the global sequence numbers reported by a sensor in the last ten times are 100, 102, 105, 106, 109, 110, 113, 115, 118, 120, then the differences between adjacent sequence numbers are 2, 3, 1, 3, 1, 3, 2, 3, 2. After sorting, they become 1, 1, 2, 2, 2, 3, 3, 3, 3, with a median of 2. This means that the sensor's data has reached the interval reference. =2; This means that the sensor generates its own data point only for every two global sequence numbers it receives on average; For any suspenseful event, let its own global sequence number be 2. Then the end sequence number of its survival window. Defined as: ;

[0038] Taking the sensor in the example above as an example, if the sequence number of the suspenseful event is 1000, then =1000+3×2=1006;

[0039] To prevent prolonged invalid waiting due to excessively large windows caused by network interruptions, data arrival intervals are set as a baseline. The upper limit is limited to 100; that is, when the calculated median exceeds 100, a forced setting is implemented. =100; At the same time, if the maximum interval of the global sequence number reported by the target sensor in the last ten times exceeds 200, it is determined that the sensor may have a communication abnormality, and a survival window in the suspense event queue is not assigned to it for the time being, and a sensor communication interruption prompt is output.

[0040] The survival window also has a dynamic extension mechanism; within the survival window period (i.e., the current global sequence number) Whenever the target sensor generates an arbitrary new data point (regardless of whether the data point constitutes a candidate water intake event), the event detection module extends the survival window end number of the suspenseful event by one data arrival interval reference; the new end number after the extension operation is... ;

[0041] However, the total extension must not exceed five times the data reaching the interval reference, meaning the maximum end sequence number should not exceed... Continuing the example above, if the target sensor generates a new data point at time 1002, the end sequence number is updated to 1006+2=1008; if another new data point is generated at time 1005, the end sequence number is updated to 1008+2=1010. At this point, the upper limit of 1000+5×2=1010 has been reached, and no further extension will be made even if new data points are generated. The purpose of this extension mechanism is to deal with the data sparsity caused by network latency or low sampling frequency of the sensor, and to ensure that pairing events with a physical causal relationship are not prematurely timed out.

[0042] When the target sensor receives a new data point, and the global sequence number of that data point... The current survival window for the mystery event has expired. When the event is detected, the event detection module removes the suspicious event from the queue. The removed event is marked as an isolated event or enters the non-response pairing link depending on whether its subsequent events meet the non-response pairing conditions (handled by the pairing analysis module). Thus, the event detection module completes the entire processing flow from raw data stream to structured event broadcasting and suspicious queue management.

[0043] For example, suppose a water intake system has a flow sensor on the water source side. Fluctuation benchmark value =10m 3 / h, the current data point value starts from 100m 3 / h jumps to 70m 3 / h, difference Δ=30m 3 / h is greater than 0.5 × 10 = 5m 3The event is identified as a candidate water intake event because the value is / h; the direction of change is "decreasing", and the ratio r = 30 / 10 = 3 > 0.8, so the amplitude is marked as "significant" and the global sequence number is recorded as 500; the event detection module broadcasts this event to the plant pressure sensors. ; calculated in advance Data reaches the interval benchmark =3, therefore the survival window end number for the suspenseful event is 500 + 3 × 3 = 509; at subsequent times 502, 505, 508, etc., if If a pressure increase event occurs (meeting the pairing rules), pairing is successful; otherwise, if the server receives a sequence number exceeding 509 and... If there is still no response, the suspenseful event is removed from the queue and enters subsequent processing; this process provides input data with clear time constraints for the subsequent pairing analysis module.

[0044] Pairing Analysis Module: When the target sensor generates a new candidate water intake event, it attempts to pair the new candidate water intake event with the event in the suspense event queue according to the event sequence number. The pairing rules are based on the physical constraints of industrial water intake. If the pairing is successful, a causal link record is generated. The causal link record includes at least the upstream sensor identifier, the downstream sensor identifier, the event sequence number difference, and the causal type.

[0045] After the event detection module completes the broadcasting of candidate water intake events and the maintenance of the suspense event queue, the pairing analysis module takes over the subsequent processing. The pairing analysis module continuously listens for newly generated candidate water intake events of each sensor and attempts to pair the event with the broadcast event in the suspense event queue of the same sensor. The pairing process strictly follows the physical constraints of the industrial water intake scenario and distinguishes between responsive pairing and non-responsive pairing.

[0046] Specifically, for any target sensor, when it generates a new candidate water intake event... At this time, the pairing analysis module first obtains the sensor's suspense event queue; this queue stores suspense events broadcast from other sensors within the same hydraulic unit that have not yet timed out and have not yet been paired, each suspense event is assigned a global reception sequence number. Sort in ascending order; modules are arranged in ascending order of their serial numbers, and attempts are made sequentially to... Match the suspenseful event with the suspenseful event in the queue; once a match is successful, immediately stop subsequent attempts and remove the suspenseful event from the queue.

[0047] The pairing analysis module has three built-in responsive pairing rules, each of which is based on the physical causal relationship between pressure and flow rate during industrial water intake.

[0048] The first rule applies to scenarios where a decrease in upstream flow leads to an increase in downstream pressure; let the suspenseful event be... Its sensor type is a flow sensor, the direction of change is "decreasing", and the amplitude category is "large" or "normal"; assume the target sensor's own event. The sensor type is a pressure sensor, and the direction of change is "up". If all the above conditions are met, the pairing is successful, and the module marks the pairing as "internal valve closing operation". The physical basis of this rule is: when the user closes the water intake valve, the water flow velocity in the pipeline decreases, the flow rate decreases, and at the same time, the downstream pressure rises because the kinetic energy of the water flow is converted into pressure energy.

[0049] The second rule applies to scenarios where an increase in upstream flow leads to a decrease in downstream pressure; assuming... It is a flow sensor, the direction of change is "up", and the amplitude category is "large" or "normal"; It is a pressure sensor, and the direction of change is "down"; when the conditions are met, the pairing is successful and it is marked as "internal valve opening operation"; the physical basis is: when the user opens the valve to increase water intake, the flow rate increases, and the downstream pressure decreases due to the acceleration of the water flow.

[0050] The third rule applies to situations where changes in downstream pressure cause changes in upstream flow rate and there is no unidirectional blocking element; assuming It is a pressure sensor, and the direction of change is arbitrary; It is a flow sensor, and The direction of change and The changes are in opposite directions (i.e., when the pressure increases, the flow rate decreases, and when the pressure decreases, the flow rate increases). At this time, an additional condition needs to be checked: whether the system's barrier identification module has marked the sensor pair as having a unidirectional barrier element; if not marked, the pairing is successful and it is marked as "external disturbance". The physical basis of this rule is that external pipeline pressure fluctuations (such as upstream pump station regulation) will first cause pressure changes at this monitoring point, and then affect the water intake flow rate; however, when there is a check valve between the sensors, the reverse pressure wave cannot propagate, so this rule should be ignored.

[0051] For any of the above successful pairing scenarios, the pairing analysis module generates a causal link record; this record contains at least the following fields: upstream sensor identifier (i.e., ... (Source sensor), upstream event global sequence number Downstream sensor identification (i.e.) (Source sensor), downstream event global sequence number Response latency points And causal type labels; the formula for calculating response latency points is: It should be noted that the response latency points are defined here. This refers to the event sequence number difference mentioned earlier, which represents the difference in the global sequence numbers of upstream and downstream events.

[0052] This value is a non-negative integer, representing the global sequence number span from the occurrence of the upstream event to the occurrence of the downstream event, reflecting the "sequence distance" of the pressure wave propagation; since the global sequence number strictly corresponds to the arrival order of data packets, this delay point actually implies time information but does not require clock synchronization.

[0053] It is important to emphasize that the "flow change leads to reverse pressure change" described by the first and second rules in the pairing analysis module is a fixed physical causal relationship that does not change with the installation position of the sensor on the pipeline; the upstream and downstream topology relationship output by the blockage identification module is only used to determine whether there is a unidirectional blockage element and to assist in the determination of external disturbances (third rule), and must not be used to reverse the direction of the first and second rules.

[0054] In addition to the three types of responsive pairings mentioned above, the pairing analysis module also handles non-responsive pairings. The triggering time for non-responsive pairings is when a suspicious event is removed from the queue by the event detection module after its survival window ends. At this time, the pairing analysis module intervenes and checks whether the suspicious event meets the following two special conditions. If they are met, a causal link record is still generated, but it is not marked as a responsive pairing.

[0055] The first type of non-response pairing is designed for scenarios involving pressure-stealth branch leaks; assuming a suspenseful event that has been removed. The event originates from a branch flow sensor, with an amplitude category of "normal" and a direction of change of "increasing." The target sensor for this suspenseful event (i.e., the sensor awaiting a response) is the main pipe pressure sensor. If the above conditions are met, a causal link record is generated, with its causal type marked as "pressure-hidden branch leak." The physical phenomenon reflected by this rule is that when a small leak occurs in a branch pipe, the branch flow rate increases slightly, but due to fluctuations in other water-using equipment, the main pipe pressure remains almost unchanged, so the pressure sensor will not generate any response event. A branch flow sensor refers to a flow sensor installed on a branch pipe; a main pipe pressure sensor refers to a pressure sensor installed on the main pipe.

[0056] The second type of non-responsive pairing is for check valve blocking events; assuming a suspenseful event that has been removed. The flow sensor, which has been marked as "downstream of the check valve" by the blocking identification module, has an amplitude category of "large" or "normal" and an arbitrary direction of change. The target sensor of this suspicious event is the pressure sensor upstream of the check valve. If the conditions are met, a causal link record is generated, and the causal type is marked as "check valve blocking event". The physical basis of this rule is that the check valve only allows water to flow in the forward direction, and the reverse pressure wave is completely blocked. Therefore, the downstream flow change cannot be propagated to the upstream pressure sensor, and the upstream pressure will inevitably not respond.

[0057] For timeout-related unresolved events that do not meet any of the above conditions, the pairing analysis module marks them as isolated events and hands them over to the results output module for subsequent circuit breaker statistics and early warning analysis.

[0058] The following example illustrates the workflow of the pairing analysis module; suppose a water intake system has a flow sensor on the water source side. A candidate water intake event with a "decreasing" direction and a "significant" magnitude was generated at global index 500, and this event was broadcast to the plant pressure sensors. In the suspense queue, its survival window ends at index 509; subsequently, at global index 503, It itself generated a candidate water intake event in an "upward" direction and a "normal" direction; the pairing analysis module in The earliest mystery found in the mystery queue happened to be from And the sequence number is 500, check the first rule: For traffic, reduce, significantly For pressure and rise, all conditions are met; therefore, the module records the causal chain: upstream Downstream Response latency points =503-500=3, the cause-effect type is "internal valve closing operation"; this link is then sent to the barrier identification module for directional dominance ratio statistics, and at the same time to the operation fingerprint analysis module for extracting valve operation features.

[0059] If within the survival window If no event of its own is generated, the suspenseful event is removed from the queue after the timeout. The pairing analysis module checks and finds that the suspenseful event originates from the flow sensor and targets the pressure sensor, but the amplitude category is "large," which does not meet the "normal" amplitude category required by the branch leakage rule. Therefore, it does not constitute a pressure stealth branch leakage. At the same time, the flow sensor is not marked as downstream of the check valve, and the check valve blocking event condition is not met. Therefore, it is determined to be a regular isolated event and handed over to the result output module for processing. The entire pairing analysis process ensures the accurate capture of causal relationships and the reasonable classification of various abnormal scenarios.

[0060] The operation fingerprint analysis module performs the following steps for the causal chain of internal valve closing or opening: traversing backwards from the start point of the flow event, recording the number of data points where the flow value continuously changes in the same direction; comparing this number with the median of similar operations in the sensor's history; if less than or equal to half the median, it's considered fast; if greater than or equal to 1.5 times the median, it's considered slow; otherwise, it's considered medium. The difference between the pressure event sequence number and the flow event sequence number is calculated as the response delay point. It checks whether there is a pressure fluctuation within the two data points before the start point of the flow event that is opposite to the final pressure change direction and exceeds three-tenths of the fluctuation benchmark value. Three data points are taken backwards from the start point of the pressure event, and the tail type is marked according to the pressure change trend. These tags are combined into an operation fingerprint string.

[0061] The operation fingerprint analysis module runs independently after the pairing analysis module, and is specifically designed to handle causal links with the causal type of "internal valve closing operation" or "internal valve opening operation". The core objective of this module is to extract a set of structured operation fingerprint tags from the waveform data of each internal operation event, and to identify potential problems such as valve actuator wear, abnormal operation, or model change by comparing them with historical regular patterns. The entire analysis process does not rely on any external calibration or manual annotation, and is entirely based on self-learning from historical sensor data.

[0062] For each internal operational causal link, the operation fingerprint analysis module first extracts the corresponding original data sequence from the circular buffer maintained by the event detection module. Specifically, taking the upstream traffic event as the center, it takes several data points before and after the occurrence of the traffic event; taking the downstream pressure event as the center, it takes several data points before and after the occurrence of the pressure event. The extraction range needs to cover the entire process of traffic from initial change to stability and pressure from initial change to regression.

[0063] The first feature extracted by the operation fingerprint analysis module is the traffic change rate label; let the data point corresponding to the start point of the traffic event be... Its value is The direction of change is ( =+1 indicates an increase. =-1 indicates a decrease); the module starts from Begin iterating through the subsequent data points of this flow sensor; for the k-th data point reached (k=1, 2, 3, ...), its value is... Calculate the change ;when symbols and Conversely, or if the absolute values ​​of the changes in two consecutive data points are both less than (in When the data point reaches the fluctuation reference value of the flow sensor, the traversal stops; the number of data points traversed at this point (excluding the starting point itself) is denoted as... For example, if point 1 continues to change in the same direction after the starting point, point 2 continues to change in the same direction, and point 3 reverses direction, then... =2.

[0064] In order to To convert to rate tags, the module needs the reference median of similar historical operations for that sensor; specifically, the module maintains the historical data for each flow sensor under both "internal valve closing operation" and "internal valve opening operation" categories. A value queue is used, retaining the 30 most recent values ​​for each type of operation; once the historical data reaches 30 values, the median of the queue is calculated. For an even number of samples, the median is the arithmetic mean of the two middle values ​​(which may be non-integer); when comparing, and All threshold calculations in the code use real number operations and do not round down; if the current operation... If so, it will be marked as "fast"; if If the data is not sufficient, it will be marked as "slow"; otherwise, it will be marked as "medium". If there are fewer than 30 historical data points, the label will not be generated until sufficient data is accumulated.

[0065] The second feature extracted by the operation fingerprint analysis module is the number of pressure response delay points; this value is directly taken from the pairing analysis module and stored in the causal link record. This field (its definition can be found in the Paired Analysis module); therefore, this module does not recalculate this value.

[0066] The third feature extracted by the operation fingerprint analysis module is the pre-pressure disturbance flag; let the global sequence number of the upstream flow event be... The global sequence number of the downstream pressure event is The module's two global sequence number positions (i.e., sequence numbers) before the start point of the traffic event. and At position ), check if the pressure sensor's data point exists; if it exists, read the value of that data point and calculate its direction and magnitude of change relative to the previous data point; assume the found pressure data point is at sequence number . The value at is Its preceding data point (serial number) The value of ) is The change The direction of change is The final direction of pressure change in this internal operation The direction of change in the value of the pressure event is determined by: if the value of the pressure event increases, then... =+1, then a decrease =-1; If there exists at least one found data point that satisfies the following two conditions: its direction of change is... On the contrary (i.e.) ), and the absolute value of its change (in If the pressure sensor's fluctuation reference value is used, it is marked as "with pre-impact"; otherwise, it is marked as "without pre-impact". The physical meaning of this feature is that, for certain types of valves (such as pneumatic diaphragm valves), the pre-action of the valve diaphragm or valve plate before completely cutting off the water flow will produce a small reverse pressure fluctuation, which can serve as an indirect identifier of the valve type.

[0067] To further distinguish between genuine pre-impact and inherent sensor noise, the absolute value of the preceding pressure fluctuation must be no less than 0.02 MPa (for the typical range of industrial pressure sensors), and the fluctuation must occur before the start of the flow event and be no more than two data points. If the sensor has experienced more than three similar "pseudo-pre-impacts" in the past hour but no valve operation followed, the determination of this feature will not be enabled until the sensor signal stabilizes.

[0068] The fourth feature extracted by the operation fingerprint analysis module is the pressure waveform trail label; the module takes three consecutive data points after the pressure event start point (i.e., the pressure data point) (if less than three, it takes the actual number); let the pressure event value be... The values ​​of the first, second, and third data points are respectively The module analyzes the changing trend of pressure values, employing sign-based judgment logic; the specific judgment rules are as follows:

[0069] First, determine the direction of this pressure change: if If the pressure is higher than the steady-state pressure before the operation (i.e., pressure increases), then the extreme point direction is upward, and subsequent data points should gradually decline; if If the pressure is below the steady-state pressure (i.e., the pressure drops), the extreme point is decreasing, and subsequent data points should gradually rise; let the steady-state pressure before the operation be denoted as . This value is the average of the five data points preceding the start point of the traffic event.

[0070] Rule 1: If three consecutive data points , , All directions Monotonic change in direction (i.e., in events of rising pressure) ≥ ≥ Or during a stress reduction event ≤ ≤ ), and the third data point and The absolute value of the deviation is less than (in If the value is the fluctuation baseline of the pressure sensor, it is marked as "monotonic regression".

[0071] Rule 2: If Compared to Continue to change towards extreme values ​​(i.e., during a pressure rise event) > Or during a stress-reduction event < ),and and A reverse turn (i.e., towards) occurs If the direction changes, it is marked as "oscillation decay".

[0072] Rule 3: If , , All remain near extreme values, that is, each data point and The absolute values ​​of the deviations are all less than If so, it will be marked as "continuous non-return".

[0073] After extracting the above four features, the operation fingerprint analysis module combines the tags into a string in the format "{operation direction}_{rate tag}_{delay}". The operation direction is "valve off" or "valve on", corresponding to the internal valve off / on operation in the cause-effect type; for example, "valve off_fast_delay 2 points_no pre-impact_oscillation decay".

[0074] To identify abnormal operations, the operation fingerprint analysis module maintains an operation fingerprint frequency table for each sensor (the upstream flow sensor corresponding to each valve). This frequency table records the number of times each operation fingerprint string appears, using the operation fingerprint string as the key. Whenever a sensor generates a new operation fingerprint, the module queries that sensor's frequency table: if the fingerprint has appeared five times or more, it is marked as "normal mode." Then, the newly generated fingerprint is compared with all marked normal mode fingerprints of the current sensor. The comparison is done by comparing the new fingerprint with each normal mode fingerprint by label field, counting the number of different fields. For example, "valve closing_fast_delay 2 points_no pre-impact_oscillation decay" differs from "valve closing_medium speed_delay 3 points_no pre-impact_oscillation decay" in both the rate label and delay point number fields, so the number of different labels is 2. If the number of different labels between the new fingerprint and all normal mode fingerprints reaches two or more, it is determined as "abnormal valve operation behavior," and the corresponding warning information is output. If the difference between the new fingerprint and at least one normal mode fingerprint is less than two labels, it is considered normal operation, and only the appearance count of the fingerprint is incremented by one; no warning is output.

[0075] The following example illustrates the complete process of operating the fingerprint analysis module; assuming that the historical data of a certain electric butterfly valve contains the number of flow change points from the most recent 30 valve closing operations. The following numbers are given: 2, 2, 3, 2, 2, 3, 2, 2, 3, 2, 2, 3, 2, 2, 3, 2, 2, 3, 2, 2, 3, 2, 2, 3, 2, 2, 3. The median after sorting these numbers is: =2; In the current new operation, the flow rate drops from 100 to 0. After the starting point, the value at the first point is 30, the value at the second point is 5, the value at the third point is 0, and the value at the fourth point is 0 (and no longer changes). =3 (because point 3 is still changing in the same direction, point 4 is when the change stops); =3 and =2 Comparison: 3 ≥ 1.5 × 2 = 3, therefore marked as "slow"; pressure delay points =2; Pre-pressure detection: A pressure data point exists at sequence number -1 before the flow event, and its change direction is "decreasing". However, the final pressure change direction of this valve closure is "increasing", which is the opposite direction, and the change amplitude is 0.05MPa. The pressure sensor's fluctuation reference value... =0.12MPa, 0.3×0.12=0.036MPa, 0.05>0.036, so it is marked as "with pre-impact"; Pressure tail: The values ​​of the three data points from the pressure event point are 0.52, 0.50, and 0.48 respectively. The steady-state pressure before operation is 0.45MPa, which is obviously monotonically regressive, so it is marked as "monotonically regressive"; The combined fingerprint is "valve closed_slow_delay 2 points_with pre-impact_monotonically regressive"; The frequency table of this sensor already has a regular pattern: "valve closed_medium speed_delay 2 points_no pre-impact_oscillation decay" (occurred 12 times) and "valve closed_fast_delay". 1 point_no pre-impact_monotonic regression (occurred 8 times); Compared with the first regular pattern, the new fingerprint has different rate labels (slow vs. medium speed), different pre-impact (present vs. absent), and different tails (monotonic regression vs. oscillation decay), for a total of 3 differences; Compared with the second regular pattern, the rate is different, the pre-impact is different, and the number of delay points is different (2 vs. 1), for a total of 3 differences, all ≥2, thus triggering the "abnormal valve operation behavior" warning, indicating that the valve actuator may be worn, causing the valve closing speed to slow down and generate pre-impact; Based on this, the maintenance personnel found that the torque of the electric actuator of the butterfly valve had decreased, requiring maintenance.

[0076] Through the aforementioned sophisticated fingerprint extraction and comparison mechanism, the operation fingerprint analysis module can achieve non-destructive monitoring of the mechanical condition of valves and early warning of anomalies without the need for additional hardware.

[0077] The water hammer analysis module performs the following steps for water intake points equipped with both pump outlet pressure sensors and downstream pipeline pressure sensors: It collects at least ten internal valve closure operations, calculates the global sequence difference between the upstream pressure sensor's flow initiation point and pressure peak as the upstream rising edge count, and the downstream similar parameters as the downstream rising edge count; it calculates the ratio of the upstream peak pressure to the steady-state pressure before valve closure as the upstream peak level; when the downstream rising edge count exceeds twice the upstream rising edge count, and the upstream peak level is high while the downstream peak level is low, a water hammer eliminator is identified; it maintains a list of upstream peak ratios for the most recent thirty valve closure operations, taking the fourth largest value as the peak threshold; when the upstream peak ratio of a new valve closure operation exceeds 1.5 times the peak threshold, a water hammer exceeding warning is output.

[0078] The water hammer analysis module, as an optional enhancement module of this system, is only enabled when both the water pump outlet pressure sensor and the downstream pipeline pressure sensor are configured at the water intake point. The core task of this module includes two stages: first, identifying whether there is a water hammer eliminator in the pipeline system; and then, on the premise that the water hammer eliminator exists, monitoring the upstream pressure peak generated by each internal valve closing operation to determine whether the valve closing speed is too fast, thereby providing an early warning of the risk of excessive water hammer due to actuator wear.

[0079] The water hammer analysis module operates independently of the operation fingerprint analysis module, but its input also depends on the "internal valve closure operation" causal link output by the paired analysis module. For each internal valve closure operation link, the water hammer analysis module extracts the corresponding upstream pressure sensor (at the pump outlet, denoted as...) from the annular buffer of the event detection module. ) and downstream pressure sensor (a point downstream of the pipeline, denoted as The original data sequence; the extraction range needs to cover the five data points before the valve closing operation starts to the ten data points after the operation is completed, to ensure complete capture of the entire process of pressure wave rise and decay.

[0080] The water hammer eliminator identification phase is performed only after the initial run or system maintenance, and requires data from at least ten internal valve closure operations to make a determination; before collecting ten operations, the system assumes that there is no water hammer eliminator at that water intake point (i.e., =False), no second-stage monitoring will be performed; at the same time, the system will record the upstream peak ratio of each valve closing operation, and once three consecutive operations are detected, If all values ​​are greater than 1.6 (temporary threshold), the system will output "Suspected water hammer exceeding the standard, it is recommended to check the valve aging" in advance to avoid missing the early warning due to waiting for ten data points; for each collected valve closure operation, the water hammer analysis module calculates the following four indicators.

[0081] The first indicator is the number of upstream rising edge points. This metric is defined as: starting from the starting point of the upstream flow event (i.e., the data point where the flow begins to change), and traversing backwards through the upstream pressure sensors. The data points are recorded until the pressure value reaches the peak value of this valve closing operation (i.e., the highest point of the pressure rise phase), and the global sequence number difference experienced during this process is recorded; let the global sequence number of the flow event start point be . The global index of the pressure peak is ,but ;

[0082] This value reflects the number of data points the pressure wave passes through from the valve closing action to the appearance of the pressure peak, implying the steepness of the pressure rise.

[0083] The second indicator is the number of points where the downstream pressure rises. Using the exact same definition: starting from the same traffic event origin. Begin by iterating through the downstream pressure sensors. The data points are used until the pressure value reaches the downstream peak value. Let the global sequence number of the downstream pressure peak value be . ;Record ;

[0084] The third indicator is the upstream peak level; first, the steady-state pressure before valve closure is calculated. Take the average of the five data points before the start of the flow event (excluding abnormal jump points); then calculate the maximum peak value reached by the upstream pressure sensor during this valve closing operation. Upstream peak ratio Based on this ratio, the peak level is divided into three grades: if If the value is greater than 1.5, it is marked as "high"; if 1.2 ≤ If ≤1.5, it is marked as "Medium"; if <1.2 is marked as "low"; the physical basis for the threshold is: during normal valve closing operation, the pressure rise when shutting down a water pump or slowly closing the valve usually does not exceed 1.2 times the steady-state pressure; rapid valve closing may generate more than 1.5 times the water hammer pressure; those in between are considered medium level.

[0085] The fourth indicator is the downstream peak level; it uses the exact same calculation method: downstream steady-state pressure. Take the average of the five data points before the start point of the traffic event, and the downstream peak value. Calculate the ratio based on the maximum value of the downstream pressure sensor during this operation. Similarly, it is divided into three levels: "high", "medium" and "low" according to >1.5, 1.2~1.5 and <1.2.

[0086] After collecting ten valve-closing operations, the water hammer analysis module comprehensively judges the above four indicators for each operation; if the conditions are met... If the upstream peak level is "high" and the downstream peak level is "low", then it is determined that a water hammer eliminator exists in the pipeline system at the water intake point. The combined physical meaning of these three conditions is that the water hammer eliminator (such as a bladder-type or piston-type energy dissipation device) will significantly absorb the energy of the pressure wave and prolong the pressure rise time. Specifically, a high-amplitude pressure peak can still be detected upstream (upstream is "high"), while downstream, due to the effect of the energy dissipator, the pressure peak is greatly reduced (downstream is "low"), and the pressure rise process is broadened (the number of rise points downstream is significantly greater than upstream). The coefficient 2 and the threshold 1.2 / 1.5 are both determined based on the measured characteristics of typical industrial water hammer eliminators and have engineering universality.

[0087] Once the water hammer analysis module determines that a water hammer eliminator exists, it will set this flag. Set it to "true" and proceed to the second stage.

[0088] In the second phase, the water hammer analysis module continuously monitors each internal valve closure operation, focusing only on the upstream pressure sensor. peak ratio The module maintains a record of the last thirty valve closing operations. Value queue; when the queue length reaches thirty, each time a new operation arrives, the module retrieves the fourth largest value from the queue as the peak threshold. The definition of "4th largest" here is: sort the 30 values ​​in the queue from largest to smallest, and take the 4th one (i.e., the 4th ranked value); this threshold is equivalent to the upper quartile of the 30 samples. The value represents a relatively high level of the upstream pressure peak during normal valve closure operation. The maximum value is not used to exclude the excessive influence of a very small number of abnormal operations on the threshold. The mean value is not used because the distribution of the peak ratio may be skewed, and the median or quantile is more robust.

[0089] For a newly generated internal valve closure operation, calculate its upstream peak ratio. ;like ;

[0090] This triggers a "water hammer warning"; this warning indicates that the valve is closing too quickly, the valve actuator may be worn, resulting in a shortened valve closing time and the generation of dangerous water hammer. Even if the water hammer eliminator absorbs some energy, the abnormal increase in upstream pressure still indicates that the valve action has exceeded the safe range; the selection of coefficient 1.5 is based on the fact that when the pressure peak of a single valve closing operation deviates significantly from the higher level (upper quartile) of the past thirty normal operations, it is a statistical outlier and has a clear physical warning significance.

[0091] At water intake points where the water hammer eliminator is not identified (i.e.) =False), the water hammer analysis module does not perform the second stage monitoring, nor does it output any water hammer-related warnings; this is to avoid the normal pressure rise caused by valve closure in pipelines without energy dissipation devices being mistakenly judged as exceeding the standard.

[0092] The following example illustrates the complete process of the water hammer analysis module; a power plant's circulating water system has been equipped with a water pump outlet pressure sensor. and cooling tower inlet pressure sensor The system collected data from ten internal valve closing operations. The calculation result for one of these operations is as follows: number of upstream rising edges. =2, number of downstream rising edge points =6, therefore (6>4) Satisfied; Upstream peak level: steady-state pressure before valve closure 0.40MPa, peak 0.68MPa, ratio 1.70>1.5, which is "high"; downstream steady-state pressure 0.35MPa, peak 0.40MPa, ratio 1.14<1.2, which is "low"; if all three conditions are met simultaneously, it is determined that a water hammer eliminator exists and is installed. True; This is followed by the second phase of monitoring; the upstream peak ratio sequence for the most recent thirty valve closure operations at this water intake point is: 1.41, 1.38, 1.45, 1.42, 1.39, 1.43, 1.40, 1.44, 1.37, 1.42, ... (a total of thirty). The fourth largest value after sorting from largest to smallest is... =1.44; calculated during a new valve closing operation. =2.23, 1.5 × 1.44 = 2.16. Since 2.23 > 2.16, a "water hammer warning" is triggered, prompting the operator to check the valve closing time of the actuator. It was found that the closing time had shortened from the normal 2 seconds to 0.6 seconds, indicating a serious risk of wear. After repair and replacement of the actuator, subsequent valve closing operations... The index has recovered to around 1.40, and the warning has been lifted.

[0093] Through the above two-stage progressive judgment, the water hammer analysis module can accurately identify the existence of water hammer eliminators and effectively warn of water hammer hazards caused by valves closing too quickly without adding extra hardware, significantly improving the safe operation level of industrial water intake systems.

[0094] Obstruction identification module: Based on the accumulated causal link records, it counts the number of bidirectional event propagation between each pair of sensors, calculates the directional dominance ratio, and infers the upstream and downstream topological relationship between sensors and identifies the existence of unidirectional obstruction elements based on the directional dominance ratio.

[0095] The causal link records continuously generated by the pairing analysis module are sent to the obstruction identification module. The core task of this module is to count the number of bidirectional event propagation between each pair of sensors from a large number of causal links, thereby inferring the upstream and downstream topological relationships between sensors and identifying the existence of unidirectional obstruction elements. The obstruction identification module runs once a day, performing calculations based on all responsive pairing data accumulated in the past 24 hours (i.e. links with causal types of "internal valve closing operation", "internal valve opening operation" or "external disturbance"), and outputting the results for use by other modules and subsequent early warning.

[0096] For any pair of sensors A and B within the same hydraulic unit, the barrier identification module maintains two independent counters: the first counter... Second counter ;in Record the number of pairings where the upstream sensor is A and the downstream sensor is B in all causal links; Record the number of pairings where the upstream is B and the downstream is A; the initial values ​​of both counters are 0, and the pairing links are accumulated sequentially to the corresponding counters each day; it should be noted that the number of bidirectional transmissions between the same pair of sensors may be asymmetrical, and this asymmetry is the basis for inferring the upstream and downstream relationship.

[0097] At the end of each daily statistical cycle, the obstruction identification module calculates the directional dominance ratio for each pair of sensors. The calculation formula is as follows: ;

[0098] The purpose of adding a constant of 1 to the denominator is to avoid a division-by-zero error when both counters are zero. =0.

[0099] Based on the calculated orientation advantage ratio, the barrier recognition module executes the following upstream and downstream determination process: First, it calculates the total number of pairings. ;when >20 and When the value is greater than 0.85, sensor A is determined to be upstream of sensor B, meaning the water flow direction and the forward propagation direction of the pressure wave are both from A to B; conversely, when... >20 and When the value is less than 0.15, A is determined to be downstream of B. The threshold is selected based on the following criteria: when the propagation direction is completely unidirectional, the direction dominance ratio should be close to 1 or 0. 0.85 and 0.15 are the 15% deviation ranges from 1 and 0, respectively. This range is sufficient to eliminate misjudgments caused by random noise or a small number of abnormal pairings, while ensuring that the sample size is sufficient to converge to the correct direction. The empirical value of 20 total pairings in the threshold ensures that the judgment result is output only after the statistical significance is sufficient.

[0100] When the sum of the first counter and the second counter When the directional dominance ratio is <20, it is only used as an intermediate reference value and is not used for any upstream or downstream judgment; the system maintains a separate "undetermined state" flag, and the judgment result is output after the sample size reaches the target; when When R = 0, the R value is meaningless and no inference is made.

[0101] After determining that sensor A is upstream of sensor B, the blocking identification module further checks the number of back propagation cycles. ;like If the number of pairs is less than 5, it indicates the presence of a unidirectional blocking element (such as a check valve or pressure reducing valve) between the sensor pairs. The physical meaning of this condition is that if a check valve is installed in the pipeline, the pressure wave can only propagate from upstream to downstream, and reverse propagation is almost impossible. Therefore, the number of reverse pairings should be very low. The threshold of less than 5 is based on engineering experience: during long-term operation, a very small number of reverse pairing records may be generated due to data noise or occasional abnormal operation. However, as long as the number is much smaller than that of forward pairings, it can be determined that there is a unidirectional blockage. At the same time, the module marks sensor B as a "downstream of check valve" sensor for use by the pairing analysis module's no-response pairing rule.

[0102] For those not yet satisfied For sensor pairs with a ratio greater than 20, or sensor pairs with an orientation advantage ratio between 0.15 and 0.85, the blocking identification module will not make upstream or downstream determinations for the time being, and will remain in a pending state, waiting for a longer period of data accumulation.

[0103] The obstruction identification module also performs topology change anomaly detection; specifically, the module saves the historical orientation advantage ratio of each sensor pair daily; if on a certain day it originally met the requirements... Sensor pairs with a directional advantage ratio >0.85 and lasting for more than seven days suddenly experience a drop to a certain value. <0.5, and on the same day If the value is still greater than 20, a "topology change alarm" will be triggered. This alarm indicates that the check valve in the pipeline section may have failed, been removed by human intervention, or been illegally bypassed and installed, requiring maintenance personnel to check immediately.

[0104] The following example illustrates the calculation process of the barrier identification module; assuming a flow sensor on the water source side... Pressure sensors in the factory area A total of 150 causal links were accumulated between them in the past day, of which =142 entries, =8 lines; Calculate the directional advantage ratio ;

[0105] The total number of pairings T = 150 > 20, and 0.940 > 0.85, therefore, it is determined that... lie in Upstream; then check the reverse count. =8, which is greater than 5 and does not meet the condition that the reverse count is less than five. Therefore, it is not identified as having a one-way blocking element. This means that the pipe section allows the pressure wave to propagate in both directions, which is consistent with the expected situation without a one-way valve.

[0106] Consider another pair of sensors: the water source side flow sensor. upstream pressure sensor of water intake ; in the cumulative link =95, =2, total number of attempts T=97, direction advantage ratio R=95 / (95+2+1)=95 / 98≈0.969>0.85, judgment Upstream; at this time, the reverse count is 2, which is less than 5, therefore it is identified that there is a unidirectional blocking element between the sensor pairs, and The downstream sensor of the check valve is marked; thereafter, the pairing analysis module encounters a signal from... Suspenseful events and the target is When there is no response from the pair, it can be correctly attributed to a "check valve blocking event".

[0107] The upstream and downstream topology relationship table and the marking results of unidirectional blocking elements output daily by the barrier identification module are persistently stored and available for real-time querying by other modules. The third rule in the pairing analysis module relies on the condition that "the barrier identification module has not marked unidirectional blocking elements" to distinguish between external disturbances and internal operations. The operation fingerprint analysis module and the water hammer analysis module also indirectly use this topology information to optimize their respective judgment logic. The entire barrier identification process does not require manual input of pipeline drawings and is entirely based on data flow self-learning to obtain pipeline network structure characteristics.

[0108] Results output module: Based on the upstream and downstream topology, identify hidden pressure leaks, check valve blockage events, and external disturbances, and output corresponding abnormal warning information;

[0109] After the pairing analysis module, the blocking identification module, the operation fingerprint analysis module, and the water hammer analysis module complete their respective processing, all abnormal events, isolated events, and various early warning information are collected into the result output module. The core responsibility of the result output module is to comprehensively judge, filter, and finally output the above intermediate results, ensuring that the output early warning information has high accuracy and reliability, and avoiding false alarms caused by network jitter, instantaneous sensor noise, or temporary external disturbances.

[0110] The results output module includes a circuit breaker output submodule, which maintains an independent count of isolated events for each sensor and uses a sliding time window mechanism for statistics. Specifically, for each pressure sensor, the module maintains a ten-minute sliding window, recording the occurrence time of isolated events generated by that sensor within the window. The definition of an isolated event comes from the pairing analysis module: any suspenseful event that exceeds the end sequence number of the survival window and does not meet the no-response pairing condition is marked as an isolated event and sent to the results output module. When the accumulated isolated event count of a pressure sensor reaches three within the ten-minute sliding window, the circuit breaker output submodule outputs an "Upstream Pressure Continuous Disturbance Warning". For each flow sensor, the same process is used: maintaining an isolated event count within the ten-minute sliding window, and outputting a "Downstream Flow Continuous Abnormality Warning" when the count reaches three. The sliding window is implemented as follows: the module maintains a first-in-first-out queue for each sensor, storing the timestamp of each isolated event generated within the last ten minutes (represented by the system global sequence number or reception time). Whenever a new isolated event arrives, the module removes events whose timestamps exceed ten minutes from the current time in the queue, then checks the queue length; if the length reaches three, the corresponding warning is triggered.

[0111] However, simply counting isolated events can lead to false alarms. For example, when real water hammer or pressure fluctuations occur in the upstream pipeline, multiple sensors may generate isolated events almost simultaneously. If each sensor counts three times independently, a large number of duplicate and inaccurate alarms will be generated. Therefore, the results output module prioritizes cross-sensor aggregation analysis before counting isolated events. This analysis is performed across all sensors within the same hydraulic unit to distinguish between global disturbance events and local real anomalies.

[0112] The specific process of cross-sensor aggregation analysis is as follows: The module acquires isolated event records generated by all sensors within the same hydraulic unit within the current time window. Each isolated event record includes the source sensor identifier, direction of change, and generation time (global sequence number). The module checks whether there are at least two different sensors that have generated isolated events within the same preset time sub-window, and the direction of change of these isolated events is the same (i.e., all are "rising" or all are "falling"). The length of this preset time sub-window is the maximum value of the median of the data arrival interval reference of all sensors within the same hydraulic unit, rounded up. Specifically, let the data arrival interval reference of each sensor be... Then the window length If the data from any sensor has not yet reached the baseline interval (e.g., less than ten times), cross-sensor aggregation analysis will not be performed until all sensors meet the sample size requirements. The length of the preset time sub-window is the maximum value of the median of the data reaching the baseline interval for all sensors, which can cover a complete reporting cycle for most sensors. If the above conditions exist, it is determined as a "global disturbance event". At this time, the module outputs "upstream pipeline pressure fluctuation warning" and forces the isolated event count of all sensors involved in the global disturbance event to zero. Then, it directly exits the circuit breaker statistics process and no longer accumulates the circuit breaker count for this batch of isolated events. The physical basis of this design is that external disturbances (such as upstream pump station regulation and water hammer waves) will affect multiple sensors simultaneously, and the direction of the pressure waves is consistent. Therefore, the simultaneous generation of isolated events in the same direction by multiple sensors is a typical feature of global disturbances and should not be alarmed as independent faults of individual sensors.

[0113] If no global disturbance event occurs, the process proceeds to the single-sensor isolated event analysis workflow. This workflow performs in-depth deconstruction of the isolated event sequence for each sensor to distinguish between sensor noise drift and local real anomalies.

[0114] The single-sensor isolated event analysis process first obtains a sequence of continuous isolated events for a single sensor within the current time window, with each event carrying a direction of change; the module extracts the direction of change for each isolated event, forming a direction sequence; let the direction sequence be... Each of them The value is +1 (representing "up") or -1 (representing "down"); the module checks whether the direction of change of adjacent events in the sequence exhibits an alternating pattern; an alternating pattern is defined as: the direction of three consecutive events is "up-down-up" or "down-up-down", i.e. and If three consecutive events in the direction sequence satisfy the alternation pattern, it is determined to be "sensor noise drift". At this time, the module outputs the corresponding sensor calibration prompt, such as "sensor signal noise drift, it is recommended to perform zero-point calibration", and immediately stops counting the isolated events of the sensor in the fuse counter. The physical basis for this determination is that real pipeline leaks or valve abnormalities usually cause the flow or pressure to change continuously in one direction (unidirectional drift), while the sensor's own noise, electromagnetic interference or temperature drift often manifests as the value fluctuating around a certain mean, thus generating an alternation pattern in the direction sequence of isolated events. The alternation pattern is selected as the threshold for three consecutive occurrences because a single alternation may be caused by random noise, while three consecutive alternations have high statistical significance.

[0115] If the direction sequence does not satisfy the alternation pattern, the module further checks whether another pattern exists: the change direction of all isolated events in the direction sequence is completely consistent (all "ascending" or all "descending"); if the obstruction identification module has not yet completed the determination of the upstream and downstream directions of the sensor (i.e., the total number of pairings of the sensor pair). If the value is less than 20, information from other upstream and downstream sensors cannot be obtained. In this case, sensor noise drift is judged solely by the alternating pattern of the direction sequence. For cases with consistent directions, they are uniformly regarded as local real anomalies (i.e., included in the fuse count). More refined upstream and downstream consistency checks are then performed after the topology is established. Simultaneously, the module queries other sensors in the upstream and downstream directions of this sensor (this information is obtained from the topology table provided by the blockage identification module) to check for isolated events within the same time window. In particular, for pressure sensors, the module checks the upstream flow sensor and other downstream pressure sensors; for flow sensors, it checks the upstream pressure sensor and downstream flow sensor. If all other sensors in the upstream and downstream directions of this sensor have no isolated events within the same time window, it is determined to be a "local real anomaly". In this case, the module adds the isolated events of this sensor to the fuse counter for accumulation, i.e., retains the original fuse count logic. The physical basis for this determination is that local real anomalies (such as minor pipeline leaks or slow valve drift) only affect the measurement value of this sensor and will not quickly propagate to other upstream and downstream sensors. Therefore, isolated events only occur on a single sensor and are in the same direction.

[0116] For isolated event sequences that do not belong to any of the above modes (e.g., the directions are neither alternating nor completely consistent, or there are other isolated events upstream and downstream), the module adopts the default processing: the isolated events of the sensor are normally counted into the fuse counter, and the original fuse logic is retained.

[0117] In addition to the fuse output submodule, the result output module also directly outputs other types of warnings without going through fuse filtering. Among them, the "valve operation behavior abnormality warning" output by the operation fingerprint analysis module and the "water hammer exceeding the standard warning" output by the water hammer analysis module are output immediately each time they occur, because these events have clear physical meaning and occur infrequently, with a low risk of false alarms. For branch leakage warnings (from the "pressure stealth branch leakage" link in the pairing analysis module's unresponsive pairing), the result output module performs additional confirmation conditions: the branch leakage candidate conditions must be met for three consecutive half-hour periods (i.e., one and a half consecutive hours), and there must be no pairing record between the main pipe pressure sensor and the branch flow sensor during the period before a confirmed warning is output. At the same time, to prevent repeated alarms on the same branch, the same branch will only output a warning once within 24 hours.

[0118] The following three specific examples illustrate the workflow of the results output module.

[0119] Example 1: At a certain moment, the pressure sensor and Isolated events were generated within five consecutive global sequence numbers, all in the direction of "up"; cross-sensor aggregation analysis detected that at least two sensors generated isolated events in the same direction within the same time sub-window, which were identified as global disturbance events; the module output "upstream pipeline pressure fluctuation warning", and will and The isolated event count is forcibly reset to zero and does not enter the subsequent circuit breaker count; at this time, even if Even if three isolated events accumulate within ten minutes, the "upstream pressure continuous disturbance warning" will not be triggered, thus avoiding false alarms.

[0120] Example 2: Flow sensor Four isolated events were generated within ten minutes, with the direction sequence being "rise-fall-rise-fall". The module entered the single sensor isolated event analysis process, detected that "rise-fall-rise" met the alternation pattern three times in a row, determined it to be sensor noise drift, output a calibration prompt, and excluded these four isolated events from the fuse counter. Based on this, the maintenance personnel performed zero-point checks and adjusted the filter parameters of the flow sensor.

[0121] Example 3: Pressure sensor Three isolated events occurred within ten minutes, all in the direction of "decline"; the module checked its upstream and downstream sensors: upstream flow sensor and downstream pressure sensors There are no isolated events within the same time window; the alternation mode is not met, and it is determined to be a local real anomaly; the module adds these three isolated events to the circuit breaker counter. When the count reaches three, it triggers the "Upstream Pressure Continuous Disturbance Warning", prompting maintenance personnel to check whether there is a leak or valve abnormality in the pipeline section corresponding to the pressure measurement point.

[0122] Through the aforementioned hierarchical and progressive judgment logic, the result output module achieves accurate conversion from raw intermediate results to final early warning information, significantly reducing the false alarm rate and improving the practicality and reliability of the industrial water intake monitoring system.

[0123] The system simultaneously runs a global heartbeat monitoring task: if no new data is received from any sensor within twice the maximum reporting interval of all sensors (the maximum data arrival interval among all sensors multiplied by 2), a "communication link interruption" warning is output, and all analysis modules that rely on real-time data are suspended until data reception is restored.

[0124] Through the adaptive threshold and dynamic extension mechanism of the survival window based on historical fluctuation amplitude in the event detection module, the system can stably capture candidate water intake events under non-uniform sampling and network latency conditions without relying on clock synchronization. Through the three types of physical rules distinguishing between responsive and non-responsive components and the automatic generation of causal link records in the pairing analysis module, the temporal relationship between pressure and flow is transformed into quantifiable direction counts and response delays. Through the daily statistics and threshold determination of the directional dominance ratio in the barrier identification module, the system can passively identify the upstream and downstream topology of the pipeline and the presence of unidirectional barrier elements using only naturally occurring operational events during daily operation, without the need for manual input of pipeline diagrams or additional hardware. The operation fingerprint analysis module uses flow change rate and pressure response... By combining the extraction of delay, preceding disturbance, and waveform wake with historical pattern comparison, the mechanical state changes of the valve actuator are transformed into a self-learning operational fingerprint drift, enabling progressive early warning of visible wear. Through the comparison of the number of upstream and downstream pressure rise edges and peak levels in the water hammer analysis module, water hammer eliminators are independently identified without the assistance of flow sensors. The upper quartile threshold of the upstream peak ratio is used to dynamically monitor the valve's excessively fast closing behavior, overcoming the masking effect of energy dissipation elements on pressure signals. Through the two-stage fuse mechanism of cross-sensor aggregation analysis and single-sensor directional sequence alternation mode in the results output module, global disturbances, sensor noise drift, and local real anomalies are strictly distinguished, significantly reducing the false alarm rate and improving the reliability of the early warning.

[0125] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0126] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0127] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A water intake metering monitoring and anomaly early warning system based on the Internet of Things, characterized in that, include: Event detection module: Acquires real-time data streams reported by multiple sensors within the same hydraulic unit; For each sensor, candidate water intake events in the real-time data stream are dynamically detected based on the fluctuation amplitude of its historical data; the detected candidate water intake events are broadcast to other sensors in the hydraulic unit, and a queue of suspenseful events is maintained locally on each sensor. Pairing Analysis Module: When the target sensor generates a new candidate water intake event, it attempts to pair the new candidate water intake event with the event in the suspense event queue according to the event sequence number. The pairing rules are set based on the physical constraints of industrial water intake. If the pairing is successful, a causal link record is generated. The causal link record includes at least the upstream sensor identifier, the downstream sensor identifier, the event sequence number difference, and the causal type. Obstruction identification module: Based on the accumulated causal link records, it counts the number of bidirectional event propagation between each pair of sensors, calculates the directional dominance ratio, and infers the upstream and downstream topological relationship between sensors and identifies the existence of unidirectional obstruction elements based on the directional dominance ratio. Results output module: Based on the upstream and downstream topology, identify pressure leakage, check valve blockage events and external disturbances, and output corresponding abnormal warning information.

2. The water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 1, characterized in that, The process by which the event detection module dynamically detects candidate water intake events based on historical data fluctuation amplitude includes: maintaining a circular buffer for each sensor to store the data points of the most recent hour; calculating the maximum absolute value of the difference between all adjacent data points in the buffer as the fluctuation benchmark value; when the absolute value of the difference between the current data point and the previous data point exceeds half of the fluctuation benchmark value, it is determined as a candidate water intake event; determining the amplitude category according to the ratio of the absolute value of the difference to the fluctuation benchmark value: a ratio exceeding 80% indicates a significant change, otherwise it is a normal change; and recording the global reception sequence number, change direction, and amplitude category of the candidate water intake event.

3. The water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 1, characterized in that, The process by which the event detection module allocates a survival window for a suspenseful event includes: obtaining the global sequence number of the ten most recently reported data points from the target sensor, and calculating the median of the differences between adjacent global sequence numbers as the data arrival interval benchmark; using the sum of the global sequence number of the suspenseful event and three times the data arrival interval benchmark as the end sequence number of the survival window; during the survival window period, whenever the target sensor generates a new data point, extending the end sequence number by one data arrival interval benchmark, with a total extension not exceeding five times the data arrival interval benchmark; when the global sequence number of a new data point exceeds the end sequence number, removing the suspenseful event from the queue; and during the survival window period, this means that the global sequence number of the current data point is less than or equal to the current survival window end sequence number.

4. The water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 1, characterized in that, The pairing rules in the pairing analysis module include: First rule: when the suspenseful event originates from the flow sensor and the direction of change is downward, and the amplitude category is large or normal, while the event itself originates from the pressure sensor and the direction of change is upward, the pairing is successful and marked as an internal valve closing operation; Second rule: when the suspenseful event originates from the flow sensor and the direction of change is upward, and the amplitude category is large or normal, while the event itself originates from the pressure sensor and the direction of change is downward, the pairing is successful and marked as an internal valve opening operation; Third rule: when the suspenseful event originates from the pressure sensor, while the event itself originates from the flow sensor and the direction of change is opposite, and the obstruction identification module does not mark a unidirectional obstruction element, the pairing is successful and marked as an external disturbance.

5. The water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 4, characterized in that, The pairing analysis module handles unresponsive pairings: for suspenseful events that have not been paired after exceeding the end sequence number of the survival window, if the suspenseful event originates from a branch flow sensor and its amplitude category is normal with an upward change direction, while the target sensor is a main pipe pressure sensor, a pressure stealth branch leakage link record is generated; if the suspenseful event originates from a flow sensor that has been marked by the blockage identification module as being downstream of the check valve, while the target sensor is a pressure sensor upstream of the check valve, a check valve blocking event link record is generated; a branch flow sensor refers to a flow sensor installed on a branch pipe; a main pipe pressure sensor refers to a pressure sensor installed on the main pipe.

6. The water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 1, characterized in that, The process of calculating the directional dominance ratio by the obstruction identification module includes: maintaining a first counter and a second counter for each pair of sensors, recording the number of pairings as the event propagates from the first sensor to the second sensor and back; dividing the value of the first counter by the sum of the first counter, the second counter, and one on a daily cycle to obtain the directional dominance ratio; determining that the first sensor is upstream of the second sensor when the sum of the first counter and the second counter exceeds 20 and the directional dominance ratio exceeds 85%; determining that the first sensor is downstream of the second sensor when the directional dominance ratio is below 15%; and identifying the presence of a unidirectional obstruction element when the upstream determination is valid and the back-count is less than five times.

7. The water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 1, characterized in that, It also includes an operation fingerprint analysis module, which performs the following steps for the causal link of internal valve closing or opening: traversing backward from the starting point of the flow event, recording the number of data points where the flow value changes continuously in the same direction, comparing this number with the median of similar operations in the sensor's history; if less than or equal to half the median, it is considered fast; if greater than or equal to 1.5 times the median, it is considered slow; otherwise, it is considered medium. The difference between the pressure event sequence number and the flow event sequence number is calculated as the response delay point; it checks whether there is a pressure fluctuation within the two data points before the starting point of the flow event that is opposite to the final pressure change direction and exceeds three-tenths of the fluctuation benchmark value; and it takes three data points backward from the starting point of the pressure event and marks the tail type according to the pressure change trend. Combine the above tags into an operation fingerprint string.

8. A water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 7, characterized in that, The process by which the operation fingerprint analysis module compares the operation fingerprint with historical regular patterns includes: maintaining an operation fingerprint frequency table for each sensor, marking fingerprints that appear more than five times as regular patterns; comparing the newly generated operation fingerprint string with all regular pattern strings of the current sensor tag by tag, and counting the number of different tags; when the number of different tags reaches two or more, outputting an abnormal valve operation behavior warning.

9. A water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 1, characterized in that, It also includes a water hammer analysis module, which performs the following actions for water intake points that are equipped with both a pump outlet pressure sensor and a downstream pipeline pressure sensor: collecting at least ten internal valve closure operations; calculating the global sequence difference between the upstream pressure sensor's flow start point and pressure peak as the upstream rising edge number, and the downstream similar parameter as the downstream rising edge number; calculating the ratio of the upstream peak pressure to the steady-state pressure before valve closure as the upstream peak level; when the downstream rising edge number exceeds twice the upstream rising edge number, and the upstream peak level is high while the downstream peak level is low, it is determined that a water hammer eliminator exists; maintaining a list of upstream peak ratios for the most recent thirty valve closure operations, and taking the fourth largest value as the peak threshold; when the upstream peak ratio of a new valve closure operation exceeds 1.5 times the peak threshold, a water hammer exceeding warning is output.

10. A water intake metering monitoring and anomaly early warning system based on the Internet of Things according to claim 1, characterized in that, The result output module includes a fuse output submodule: it maintains an isolated event count within a ten-minute sliding window for each pressure sensor, and outputs an upstream pressure continuous disturbance warning when the count reaches three; it maintains the same count for each flow sensor, and outputs a downstream flow continuous abnormality warning when the count reaches three; for operation fingerprint abnormality and water hammer exceeding the standard warning, it outputs immediately each time it occurs; for branch leakage warning, it only outputs when three consecutive half-hour periods meet the candidate conditions and the main pipe pressure has no paired record within the period, and the same branch outputs only once within twenty-four hours.

Images