Adaptive Reference Reconstruction Pipeline Pressure Co-stabilization and Leakage Diagnosis Method and System

Abstract

This invention provides an adaptive benchmark reconstruction method and system for coordinated pressure stabilization and leakage diagnosis in a pipe network, belonging to the field of agricultural water conservancy information technology. Addressing the challenges of monitoring hydraulic instability in drip irrigation systems caused by terrain undulations, pump station fluctuations, and minute leaks caused by bird pecking and insect bites, this invention constructs a multi-mode dynamic hydraulic benchmark library by collecting healthy operation data; calculates the dynamic residual matrix between real-time pressure and the benchmark using a sliding time window; decouples the residuals from their spatiotemporal characteristics; when the residuals exhibit synchronous drift in the same direction across the entire network, it is identified as a systemic fluctuation, and pump-valve coordinated stabilization is implemented; when the residuals exhibit localized negative outliers and decreased neighborhood correlation, it is identified as a micro-leak, and the faulty branch is identified. This invention eliminates the need for precision flow meters, achieving sensitive detection of minute damage and constant pressure control using only pressure data. Furthermore, by updating the benchmark through self-healing, it adapts to long-term network evolution, improving water and fertilizer utilization and operational reliability of the drip irrigation system.

Description

Technical Field

[0001] This invention belongs to the technical field of new-generation information technology and smart agriculture water conservancy engineering, and particularly relates to an adaptive benchmark reconstruction pipeline pressure collaborative stabilization and leakage diagnosis method and system. Background Technology

[0002] Drip irrigation technology, as an important method of efficient and water-saving irrigation in modern agriculture, has the core advantage of delivering water and fertilizer to the crop root zone at a precise flow rate, reducing deep seepage and evaporation losses, thereby improving water and fertilizer use efficiency. However, the realization of this advantage depends heavily on maintaining a constant inlet pressure of the drip emitter. Drip emitters typically exhibit pressure-flow sensitivity, meaning that even small changes in inlet pressure can directly cause deviations in outflow, thus affecting irrigation uniformity. In practical field applications, maintaining a constant pipeline pressure and effectively monitoring for faults pose significant challenges due to constantly changing factors such as water source conditions, equipment status, and the external environment.

[0003] Pressure fluctuations are caused by a complex set of factors. Seasonal changes in water source levels alter pump inlet pressure, gradual clogging of the head filter increases local resistance in the pipeline network and causes a drop in downstream pressure, and voltage fluctuations in the power grid caused by the start-up and shutdown of pumping stations in adjacent areas affect pump speed and outlet pressure. These factors combine to create nonlinear, time-varying, and systematic fluctuations in pipeline network pressure. These pressure fluctuations directly cause the flow rate of the irrigation emitters to deviate from the design value, reducing the uniformity of irrigation across the entire network, resulting in water shortages in some crops and waterlogging in others, thus severely impacting water and fertilizer use efficiency.

[0004] Minor leaks are difficult to detect. Drip irrigation tape has thin walls and is laid in the field, making it highly susceptible to pecking by birds, gnawing by rodents, or scratches from agricultural machinery. The leakage caused by such minor damage is usually far less than the total system flow rate, often on the order of magnitude below the flow meter's own measurement error range, and therefore easily masked by flow metering noise or normal flow fluctuations. Traditional alarm systems based on total flow thresholds are ineffective in identifying such minor leaks, and in most cases, they fail completely.

[0005] Traditional alarm mechanisms have significant limitations. Existing alarm systems mostly employ fixed threshold methods, such as triggering an alarm when the pressure drops below 0.2 MPa. However, pipeline pressure itself exhibits time-varying characteristics; pressure differs between the initial and final stages of irrigation, and switching between different irrigation groups or changes in terrain elevation can cause deviations in normal operating pressure. Fixed thresholds cannot adapt to these pressure variations under normal operating conditions. When the threshold is set too high, normal fluctuations are easily misinterpreted as faults, resulting in false alarms; when the threshold is set too low, pressure drops caused by minor leaks may not reach the threshold, leading to missed alarms. Therefore, fixed threshold methods are insufficient to meet the accuracy and reliability requirements of fault detection in practical engineering projects.

[0006] Therefore, there is an urgent need for an intelligent method and system that can learn and memorize the health status of the pipeline network and decouple the characteristics of minor faults from complex pressure fluctuations. This would solve the problems of difficulty in identifying minor leaks in existing drip irrigation systems under time-varying pressure conditions and the high false alarm and missed alarm rates of fixed threshold alarms, thereby ensuring drip irrigation uniformity, improving water and fertilizer utilization, and enhancing system operational reliability. Summary of the Invention

[0007] The summary section of this application is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0008] To address the problems and shortcomings of existing technologies, this invention aims to provide an adaptive benchmark reconstruction method and system for pressure stabilization and leakage diagnosis in drip irrigation networks. This method involves progressively constructing a multi-mode dynamic hydraulic benchmark library, calculating the dynamic residual matrix in real time, decoupling the dynamic residuals from their spatiotemporal characteristics, implementing pump-valve coordinated stabilization for systemic disturbances, performing topology localization and early warning for local anomalies, and updating the benchmark through self-healing based on long-term operational data. This achieves the ability to sensitively detect minute leaks in drip irrigation networks and maintain constant pressure control, thereby solving the problems mentioned in the background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] As a first aspect of this application, the present invention discloses an adaptive baseline reconstructed pipeline pressure co-stabilization and leakage diagnosis method, comprising the following steps:

[0011] Step S1: Collect node pressure data for different irrigation modes under multiple healthy operating cycles and construct mode labels. Use Gaussian process regression to train and generate pressure change prediction models and dynamic confidence envelopes for each node, and construct a multi-mode dynamic hydraulic benchmark library.

[0012] Step S2: Read the current mode label, match the nearest pressure change prediction model from the multi-mode dynamic hydraulic benchmark library, calculate the pressure deviation, rate of change deviation and spatial correlation deviation using a sliding time window and standardize to form a dynamic residual matrix;

[0013] Step S3: Decouple the dynamic residual matrix from its spatiotemporal features, introduce the spatial Pearson correlation coefficient as a discrimination index to distinguish between global pressure disturbances and local micro-leakage, and output the anomaly type and node set.

[0014] Step S4: For global pressure disturbance, the pressure compensation amount is generated by using a discrete PI algorithm based on the weighted average drift of the network residuals, and control commands are generated to the variable frequency pump and electric pressure regulating valve according to the pump and valve sensitivity.

[0015] Step S5: Based on the local micro-leakage, perform a breadth-first search on the negative outlier residual nodes based on the pipeline topology adjacency matrix, and select the smallest continuous topological unit as the leak candidate region by combining the topological inclusion relationship.

[0016] Step S6: Based on the anomaly judgment, control execution and maintenance registration records of steps S3 to S5, the benchmark reconstruction is triggered. Healthy samples are selected, the pressure change prediction model is retrained and verified, and the old model is replaced. Feedback is sent to step S1 to complete the self-healing closed loop.

[0017] Preferably, the construction of the multi-mode dynamic hydraulic reference database in step S1 includes:

[0018] Collect node pressure data for different irrigation modes under multiple healthy operating cycles, and use the set of open valves, irrigation branch number, target flow level and pump set operating level to form mode labels;

[0019] For healthy samples with the same pattern label, a pressure change prediction model for each node as a function of irrigation time is trained using Gaussian process regression, and the predicted mean is output simultaneously. With prediction variance ;

[0020] Optimizing the hyperparameters of the pressure change prediction model using maximum likelihood estimation and the L-BFGS-B algorithm with boundary conditions. The dynamic confidence envelope is determined based on the predicted mean and predicted variance.

[0021] The optimized pressure change prediction model and dynamic confidence envelope parameters are stored in the benchmark library to form a multi-mode dynamic hydraulic benchmark library.

[0022] Preferably, the calculation of the dynamic residual matrix in step S2 includes: reading the current valve feedback, pump frequency and irrigation task number to form the current mode label, and matching the nearest neighbor model from the multi-mode dynamic hydraulic benchmark library;

[0023] Calculate the pressure deviation at each node using a sliding time window. Deviation in rate of change Spatial correlation deviation ;

[0024] The three biases are standardized according to the standard deviation of the healthy samples to form a node residual vector, which is then arranged by node number to form a dynamic residual matrix. .

[0025] Preferably, the spatiotemporal feature decoupling in step S3 includes:

[0026] Calculate the spatial Pearson correlation coefficient between nodes and the average spatial correlation between the target node and its neighboring nodes, and compare it with the synchronous change threshold and the local anomaly threshold.

[0027] If the dynamic residual sequence of all nodes in the network shows synchronous drift in the same direction and the average spatial correlation is not lower than the synchronous change threshold, it is judged as a systemic pressure disturbance.

[0028] If the dynamic residual sequence of a local node shows a persistent negative abnormal deviation and the average spatial correlation is lower than the local anomaly threshold, it is judged as a local micro-leakage event.

[0029] For local anomalies, pressure drop rate fingerprints and high-frequency energy of pressure signals are further used to distinguish between normal irrigation switching transients, systemic slow pressure loss, sudden pipe ruptures, and micro-hole leaks.

[0030] Preferably, the high-frequency energy identification using pressure drop rate fingerprint and pressure signal includes:

[0031] Compute the pressure change rate of the node And establish pressure drop time scale discrimination parameters. in This represents the duration of the maximum pressure drop during the tube rupture. For the minimum duration of microleakage, , These are the pressure drop rate thresholds for pipe rupture and micro-leakage, respectively. These are the cumulative pressure drop thresholds for pipe rupture and micro-leakage, respectively.

[0032] If a local abnormal node has a known valve action event identifier, and the pressure change rate satisfies Meanwhile, cumulative pressure reduction Pressure within the time window If the internal state recovers to the dynamic confidence envelope, it is determined to be a normal rotational irrigation switching transient.

[0033] If the proportion of nodes with average spatial correlation below the synchronous change threshold exceeds a preset proportion threshold, and the pressure change rate satisfies... Meanwhile, cumulative pressure reduction satisfy Pressure within the time window If the pressure continues to decrease, it is determined to be a systemic slow decompression;

[0034] If the node pressure is within the time window A rapid decrease occurs within, and the rate of pressure change satisfies Meanwhile, cumulative pressure reduction If so, it is determined to be a sudden pipe rupture;

[0035] If the node pressure is within the time window The pressure decreases slowly and continuously, and the rate of pressure change satisfies... Meanwhile, cumulative pressure reduction High-frequency energy of pressure signals If the leakage rate is higher than the health status threshold, it is determined to be a micropore leakage.

[0036] Preferred,

[0037] The drift amount is obtained by weighted average of the residuals of all nodes in the network. The pressure compensation control quantity is generated using a discrete PI algorithm. ;

[0038] Based on the pump pressure sensitivity and valve opening sensitivity, the compensation amount is allocated as the frequency adjustment amount of the variable frequency water pump. and the adjustment amount of the electric pressure regulating valve opening ;

[0039] Current frequency of variable frequency water pump With frequency adjustment amount The sum of these values ​​forms the frequency setpoint of the variable frequency water pump, which in turn determines the current opening degree of the electric pressure regulating valve. Adjustment amount of electric pressure regulating valve opening The sum of these values ​​determines the opening degree of the electric pressure regulating valve.

[0040] After the rate of change is limited, upper and lower limits are checked, and dead zone is determined, control commands are generated to drive the variable frequency water pump and electric pressure regulating valve to restore the real-time pressure to the reference envelope.

[0041] Preferred,

[0042] Based on a pre-stored drip irrigation network topology diagram, pressure sensor nodes are mapped to nodes in the topology diagram, and a node adjacency matrix is ​​established according to the pipe segment connection relationships. Simultaneously establish the side length matrix ;

[0043] Calculate the absolute value of the comprehensive dynamic residual of each node, and form an anomaly set with negative residual nodes that exceed the 99th percentile threshold of healthy samples.

[0044] Using the node with the largest residual as the initial outlier node, based on the adjacency matrix A breadth-first search is performed on the initial abnormal nodes, and only connected nodes with negative residuals that exceed a threshold are accepted to obtain a set of connected abnormal nodes.

[0045] Based on the inclusion relationship between the set of abnormal nodes and the pipeline topology, using the side length matrix... Calculate the total length of pipe segments in the candidate topology units, and select the smallest continuous topology unit as the leakage candidate area and trigger an alarm in order of least number of pipe segments and shortest total length of pipe segments.

[0046] Preferred,

[0047] Determine whether a baseline reconstruction candidate is triggered based on maintenance registration events or preset same-direction mean drift conditions.

[0048] Healthy samples with normal irrigation cycles were screened, and those with alarm indicators, comprehensive residuals exceeding the 99th percentile threshold, or average spatial correlation below the local anomaly threshold were excluded. ;

[0049] Retrain the Gaussian process regression model and validate the health coverage and anomaly detection rate. Replace the old model only if the health coverage of the validation set is not less than 95%, the detection rate of known anomaly samples is not less than that of the old model, and there are no anomaly alarms for two consecutive trial runs; otherwise, revert to the old model.

[0050] As a second aspect of this application, the present invention also discloses an adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis system, characterized in that it comprises:

[0051] A wireless pressure sensor network is used to collect pressure data from various nodes in the drip irrigation network; an edge computing gateway is used to store the standard hydraulic baseline model and perform dynamic residual analysis, including:

[0052] The model dynamic hydraulic reference library module is used to build and store a multi-model dynamic hydraulic reference library;

[0053] The dynamic residual calculation module is used to calculate the dynamic residual matrix based on real-time pressure data and the baseline model;

[0054] The spatiotemporal feature decoupling module is used to decouple the dynamic residual matrix from spatiotemporal features, distinguishing between systemic disturbances and local micro-leakages;

[0055] The collaborative stabilization control module is used to generate collaborative control commands for pumps and valves in response to systemic disturbances.

[0056] The topology location and early warning module is used to locate the leakage branch for localized micro-leakage and trigger an alarm.

[0057] The self-healing update module is used to update the hydraulic reference model based on long-term operating data.

[0058] The actuator, including the first variable frequency water pump and the field electric pressure regulating valve, is used to receive control commands from the edge computing gateway and regulate the pressure.

[0059] The human-machine interface is used to display alarm information, potential leakage areas, and maintenance suggestions, and allows maintenance personnel to register maintenance events and confirm baseline reconstruction.

[0060] Preferably, the wireless pressure sensor network includes one sensor at the head outlet, one sensor at the inlet of each main pipe, and one sensor at the inlet of each irrigation branch pipe; for branch pipes with a length exceeding 100m or an elevation change exceeding 5m, one additional sensor is added every 50-100m or at each significant elevation inflection point, and at least one verification sensor is set at the end; the sensor arrangement is constrained by the observability of candidate pipe segments, so that there are at least two monitoring nodes on both sides of any candidate leaking pipe segment that can form a residual propagation direction.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] This invention provides an adaptive benchmark reconstruction method and system for coordinated stabilization and leakage diagnosis of pipeline network pressure. By collecting historical pressure data of the drip irrigation network under healthy operating conditions, and using valve set opening, irrigation branch number, target flow level, and pump operating level as pattern labels, an adaptive hydraulic benchmark model is established using Gaussian process regression. The benchmark reconstruction program is triggered based on equipment maintenance, water source changes, and changes in pipeline network operating conditions, achieving dynamic updates of the pressure benchmark with the pipeline network's operating state. This avoids the benchmark drift problem caused by traditional fixed theoretical models and improves the accuracy and adaptability of pressure state judgment under long-term operating conditions. Simultaneously, by constructing a dynamic residual sequence between real-time pressure data and the dynamic hydraulic benchmark, and combining the spatial Pearson correlation coefficient between nodes, pressure deviation, and rate of change deviation, different types of pressure anomalies are decoupled, enabling the differentiation between systemic pressure fluctuations and local micro-leakage events. This reduces false alarms caused by changes in normal operating conditions such as water source fluctuations, equipment adjustments, and irrigation switching, and improves the reliability of anomaly identification.

[0063] To address systemic pressure fluctuations, this application generates pressure compensation control quantities by calculating the weighted average drift of the pressure residuals across the entire network. It then utilizes an edge computing gateway to collaboratively regulate the head unit's variable frequency pumps and field electric pressure regulating valves, achieving proactive regulation and coordinated stabilization of the network pressure, thus improving the pressure stability and irrigation uniformity of the drip irrigation network. For localized abnormal pressure changes, the application combines network topology, node adjacency matrix, and breadth-first search to differentiate between sudden pipe bursts and micro-hole leaks using pressure drop rate characteristics. This enables rapid identification of leak locations, improving fault location efficiency and reducing manual inspection costs. Furthermore, the edge computing gateway integrates pressure data acquisition, model updates, anomaly identification, and control command output, reducing reliance on cloud computing and complex hydraulic models, and enhancing the system's real-time response and field application capabilities. Attached Figure Description

[0064] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application.

[0065] In the attached diagram:

[0066] Figure 1 This is a block diagram showing the steps of the adaptive benchmark reconstruction pipeline pressure co-stabilization and leakage diagnosis method in this embodiment of the invention.

[0067] Figure 2 This is a logical flowchart of the adaptive benchmark reconstruction pipeline pressure collaborative stabilization and leakage diagnosis method in this embodiment of the invention;

[0068] Figure 3 This is a waveform diagram comparing irrigation time under normal pipeline pressure, system disturbance, and local micro-leakage in an embodiment of the present invention.

[0069] Figure 4 This is a topology diagram of the adaptive benchmark reconstruction pipeline pressure co-stabilization and micro-leakage diagnosis system in an embodiment of the present invention.

[0070] Figure 5 This is a schematic diagram of the structure of the electronic device in an embodiment of the present invention. Detailed Implementation

[0071] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0072] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0073] Example 1

[0074] In this embodiment of the invention, an adaptive benchmark reconstruction method for coordinated stabilization and leakage diagnosis of pipeline network pressure is disclosed. The following will describe this disclosure in detail with reference to the accompanying drawings and embodiments. (Refer to...) Figures 1 to 4 As shown, the present invention includes the following steps:

[0075] Step S1: Collect node pressure data for different irrigation modes under multiple healthy operating cycles and construct mode labels. Use Gaussian process regression to train and generate pressure change prediction models and dynamic confidence envelopes for each node, and construct a multi-mode dynamic hydraulic benchmark library.

[0076] Step S2: Read the current mode label, match the nearest pressure change prediction model from the multi-mode dynamic hydraulic benchmark library, calculate the pressure deviation, rate of change deviation and spatial correlation deviation using a sliding time window and standardize to form a dynamic residual matrix;

[0077] Step S3: Decouple the dynamic residual matrix from its spatiotemporal features, introduce the spatial Pearson correlation coefficient as a discrimination index to distinguish between global pressure disturbances and local micro-leakage, and output the anomaly type and node set.

[0078] Step S4: For global pressure disturbance, the pressure compensation amount is generated by using a discrete PI algorithm based on the weighted average drift of the network residuals, and control commands are generated to the variable frequency pump and electric pressure regulating valve according to the pump and valve sensitivity.

[0079] Step S5: Based on the local micro-leakage, perform a breadth-first search on the negative outlier residual nodes based on the pipeline topology adjacency matrix, and select the smallest continuous topological unit as the leak candidate region by combining the topological inclusion relationship.

[0080] Step S6: Based on the alarm and maintenance records in steps S4 and S5, determine the trigger benchmark reconstruction, select healthy samples, retrain the pressure change prediction model, verify it, and replace the old model. Feedback is then sent to step S1 to complete the self-healing closed loop.

[0081] Regarding step S1, pressure data for different irrigation modes under multiple healthy operating cycles are collected. Mode labels are constructed using the set of open valves, irrigation branch number, target flow rate, and pump operating level. A multi-mode dynamic hydraulic benchmark library is then trained and built for healthy samples with the same mode label. Specifically, the following steps are also included:

[0082] Collect node pressure data for different irrigation modes under multiple healthy operating cycles, and use the set of open valves, irrigation branch number, target flow level and pump set operating level to form mode labels;

[0083] For healthy samples with the same pattern label, a Gaussian process regression model was used to train a pressure change prediction model for each node as a function of irrigation time, and the predicted mean was output. With prediction variance ;

[0084] Optimizing the hyperparameters of the pressure change prediction model using maximum likelihood estimation and the L-BFGS-B algorithm with boundary conditions. The dynamic confidence envelope is determined based on the predicted mean and predicted variance.

[0085] The optimized pressure change prediction model and dynamic confidence envelope parameters are stored in the benchmark library to form a multi-mode dynamic hydraulic benchmark library.

[0086] Specifically, step S1 serves as the data foundation and benchmark for the entire method. First, with the drip irrigation system in a healthy operating state, a wireless pressure sensor network deployed at the head outlet, main pipe inlet, irrigation branch inlet, and key pipe sections collects node pressure data for different irrigation modes during multiple healthy operating cycles. Node pressure data refers to the measured pressure value at each pressure monitoring node at each sampling moment, with a sampling period preferably between 1 and 5 seconds. Each set of collected data is accompanied by a mode tag, which consists of four elements: the set of open valves, the irrigation branch number, the target flow rate, and the pump operating level, uniquely identifying the current operating condition. The mode tag distinguishes the pressure characteristics under different irrigation modes, ensuring that the benchmark model obtained through subsequent training accurately corresponds to the actual operating conditions and avoiding mutual interference between pressure fluctuation characteristics of different modes.

[0087] Then, statistical modeling was performed on healthy samples with the same pattern label, specifically using the Gaussian process regression (GPR) algorithm. Using time-series variables within the irrigation cycle as input and pressure measurements at corresponding times as output, a pressure change prediction model was established for each node's pressure over irrigation time. The Gaussian process regression model can be expressed as follows: ,in It is a function of the pressure mean. The covariance function is used to describe the correlation between pressure values ​​at different times. For a moment The predicted value of node pressure. The covariance function is represented by the squared exponential kernel function. ,in This is a parameter representing the magnitude of pressure change amplitude, indicating the magnitude of pressure fluctuation. The time-dependent length scale parameter characterizes the temporal correlation range of pressure changes, i.e., how long apart pressure observations still exhibit a strong correlation. The squared exponential kernel function is chosen because pressure over time is typically continuous and smooth within a healthy drip irrigation cycle, offering better nonlinear smoothing interpolation and uncertainty quantification capabilities compared to fixed-order fitting, facilitating the formation of a time-varying pressure confidence envelope. After training with a Gaussian process regression (GPR) model, the predicted mean of pressure changes at each node over irrigation time is output. and prediction variance The calculation of the predicted mean and predicted variance is implemented using existing formulas, which will not be elaborated here. This indicates the node pressure at time t in a healthy state. Expected value; This indicates the degree of uncertainty in the prediction at that moment, and is used to quantify the baseline fluctuation range for different irrigation times and nodes. The larger the prediction variance, the greater the natural pressure fluctuation at that moment or the sparser the historical data, and the more relaxed the dynamic confidence envelope should be in the subsequent construction.

[0088] Next, the established pressure change prediction model is optimized for hyperparameters, and the dynamic pressure confidence envelope is determined. The hyperparameter set is as follows: ,in To measure noise parameters, the model hyperparameters are optimized using maximum likelihood estimation, specifically employing the bounded L-BFGS-B algorithm to minimize the negative logarithmic marginal likelihood. To reduce the influence of local optima, at least 10 initial values ​​are randomly generated from a log-uniform distribution and optimized separately. The set with the minimum negative logarithmic prediction likelihood and satisfactory coverage in the validation set is selected. The iteration stops when the relative change of the objective function is less than 1 / 2 consecutive times. Or the number of iterations reaches 500. Normal irrigation cycle number. The optimal number of samples is 20–60. When there are many rotational irrigation patterns or significant seasonal fluctuations, each pattern should ideally have at least 30 cycles. The training and validation sets should be divided in an 8:2 ratio, ensuring that each rotational irrigation pattern has validation samples. When the validation set envelope coverage falls below 95%, additional samples should be added or the training should be retrained. Coverage refers to the proportion of healthy samples in the validation set whose stress values ​​fall within the dynamic confidence interval.

[0089] After optimization, the dynamic confidence envelope coverage is validated using a validation set. The dynamic confidence interval for stress is determined by the predicted mean. and variance Build: , ,in The lower bound of the dynamic confidence envelope for pressure. The upper bound of the dynamic confidence envelope for pressure. Confidence coefficient. The preferred value is 1.96 to 3.00. Under normal operating conditions, 1.96 is used to form a confidence envelope of about 95%. When the noise is strong or the suppression of false alarms is emphasized, 2.58 to 3.00 is used. The validation set is calibrated to ensure that at least 95% of healthy samples fall within the validation envelope, while maximizing the detection rate of known leaked samples. If the validation set envelope coverage falls below 95%, additional samples or retraining are required. (The last sentence appears to be incomplete and possibly refers to real-time pressure.) or Furthermore, if the duration of consecutive crossings of the dynamic confidence envelope reaches a time threshold, the node is determined to have deviated from the normal operating range. The duration threshold is preferably set to 3–10 sampling periods, but not less than 10 seconds. Specifically, it is the larger of the 99th percentile of the healthy sample's out-of-range duration and the duration of crossing within the three sampling periods, and is calibrated using a validation set to ensure a false alarm rate of no more than 1%. This dynamic confidence envelope can adaptively adjust its width; a wider envelope is allowed when data is sparse or naturally fluctuating, while a narrower envelope is used when data is stable, thereby reducing false alarms and false negatives caused by a fixed threshold.

[0090] Finally, the optimized and validated pressure change prediction model and dynamic confidence envelope parameters are stored in the benchmark library, forming a multi-model dynamic hydraulic benchmark library. This multi-model dynamic hydraulic benchmark library is organized and stored according to model tags, and contains the predicted mean values ​​of each node under different irrigation modes. Prediction variance Dynamic confidence envelope upper and lower bounds and and the standard deviation required for residual standardization , , Parameters such as these are used. The role of the benchmark library is to provide a health status reference for subsequent steps. Specifically, during real-time operation, the corresponding benchmark model is called through pattern label matching to calculate the dynamic residual between the real-time data and the benchmark. Thus, the multi-mode dynamic hydraulic benchmark library essentially preserves the hydraulic fingerprint of the pipeline network in a healthy state in a statistical manner, replacing the traditional fixed threshold method and laying the foundation for the extraction of minor fault characteristics.

[0091] Regarding step S2, the current valve feedback, pump frequency, and irrigation task number are read to form the current mode label. The nearest neighbor model is matched from the multi-mode dynamic hydraulic benchmark library. Pressure deviation, rate of change deviation, and spatial correlation deviation are calculated using a sliding time window and standardized to form a dynamic residual matrix. Specifically, this step receives the multi-mode dynamic hydraulic benchmark library constructed in main step S1 and calculates the deviation between the current operating state and the healthy benchmark during real-time operation. First, the current valve feedback, pump frequency, and irrigation task number are read to form the current mode label. The composition of the mode label is the same as in main step one, consisting of the set of open valves, the irrigation branch number, the target flow rate level, and the pump set operating level. A benchmark model with completely identical labels is preferred; if no completely identical model is found, the nearest neighbor model is selected based on the weighted distance of the Hamming distance of the valve state, the relative deviation of the target flow rate, and the relative deviation of the pump frequency. The Hamming distance refers to the number of different elements in two valve state vectors. The relative deviation of the target flow rate is the difference between the current target flow rate and the corresponding flow rate of the benchmark model divided by the benchmark flow rate. The relative deviation of the pump frequency is calculated similarly. After matching the baseline model, a sliding time window is used to calculate the time of each node at the sampling time in real time. The three types of deviation characteristics. Let the first one be... Each pressure node at the sampling time The measured pressure, the baseline predicted mean, and the sampling interval are respectively , and The first characteristic is pressure deviation. ,in This represents the deviation of the current measured pressure from the healthy baseline; a positive value indicates that the pressure is too high, and a negative value indicates that the pressure is too low. The second type of feature is represented by the rate of change deviation.

[0092] ;

[0093] in It represents the difference between the current rate of pressure change and the reference rate of pressure change, and is used to detect anomalies in the dynamic trend of pressure. Even if the absolute deviation of pressure is not obvious, an anomaly in the rate of change may indicate an early failure.

[0094] The third type of feature is spatial correlation deviation, let's assume... For nodes The set of topologically adjacent nodes, Let be the number of nodes in this set, and let be the spatial association deviation.

[0095] ;

[0096] in Nodes within the sliding window With nodes Pearson correlation coefficient of stress series. Represents a node The degree of decrease in pressure that is consistent with the pressure changes of other nodes in its topological neighborhood. When a node's pressure change is highly synchronized with its neighborhood... Approaching 1 Approaching 0; when a node experiences a local anomaly that causes it to become disconnected from changes in pressure within its neighborhood, Decline Increase.

[0097] To eliminate the influence of different nodes and different feature dimensions, the above three types of features are standardized according to the standard deviation of healthy samples. Let... For nodes Standard deviation of pressure bias in healthy samples The standard deviation of the rate of change of healthy samples. Let be the standard deviation of healthy samples with spatial correlation bias. The standardized node residual vector is .

[0098] ;

[0099] From all Row vectors of nodes The dynamic residual matrix is ​​formed by arranging nodes according to their numbers. Each row of this matrix corresponds to a pressure node, and the three columns correspond to the standardized pressure deviation, rate of change deviation, and spatial correlation deviation, respectively. The dynamic residual matrix output in this step... This serves as the input for the next step of spatiotemporal feature decoupling.

[0100] Regarding step S3, the dynamic residual matrix undergoes spatiotemporal feature decoupling, and the spatial Pearson correlation coefficient is introduced as a discrimination index. Network-wide unidirectional drift is identified as global pressure disturbance, and local negative outlier mutations are identified as local micro-leakage. Specifically, this step receives the dynamic residual matrix output from step S2. Based on spatial synchronicity and local anomalies, global pressure disturbances and local micro-leakages are decoupled. Decoupling refers to decoupling the dynamic residual matrix. It is decomposed into a global common component that represents synchronous changes across the entire network and a local singular component that represents local outliers and anomalies. That is, it determines whether the anomaly is a systemic fluctuation caused by a common factor across the entire network or an isolated change caused by a fault at a certain local location.

[0101] First, the spatial Pearson correlation coefficient is introduced as a discrimination index. For any two pressure monitoring nodes... With nodes Pressure time series were collected within the same irrigation cycle. , The spatial Pearson correlation coefficient between the two nodes is calculated as follows:

[0102] ;

[0103] in, , These are the average values ​​of the pressure sequences at the corresponding nodes. This represents the number of pressure sampling points within a single irrigation cycle. For nodes With nodes Spatial Pearson correlation coefficient, , These are the two nodes respectively. The pressure values ​​at each sampling time. Based on the correlation coefficient matrix between multiple pressure nodes,

[0104] ;

[0105] The average spatial correlation between the target node and its neighboring nodes is calculated as follows: , This represents the total number of nodes participating in the calculation. Reflects the nodes The pressure changes are compared with the average synchronization level of other nodes in the entire network. A threshold for judging synchronization changes is set. The 10th quantile of the average correlation coefficient between adjacent nodes in a healthy state is selected, preferably between 0.75 and 0.90; the threshold for determining local anomalies is... Take the 90th quantile of the average correlation coefficient of known local leakage samples, preferably between 0.30 and 0.60, and it is required that... The threshold was determined through a validation set grid search, with the goal of minimizing the combined classification error rate of systemic disturbances and local leaks.

[0106] The judgment rule is: when When the pressure change at this node is considered to be highly synchronized with that of surrounding nodes, the source of the anomaly is attributed to overall network pressure fluctuations or common disturbances; when At that time, it was considered that the pressure change at this node had a low correlation with surrounding nodes, and its anomaly originated from a local node failure; when At this point, it is determined to be a transitional state. Further calculation of the comprehensive residual is then performed. and its duration ,like Exceeding the local anomaly threshold and If the time threshold is reached and the anomaly only extends along a single topology branch, it is classified as a local micro-leakage; if the proportion of anomalous nodes in the same direction is not less than 70% and continues for two sliding windows, it is classified as a systemic pressure disturbance; otherwise, observation is maintained and control commands are not immediately output to reduce malfunctions caused by transients during irrigation switching.

[0107] In particular, for anomalies involving localized micro-leakage, pressure drop rate fingerprinting is further used to identify the fault type, distinguishing between four scenarios: normal irrigation switching transients, systemic slow pressure loss, sudden pipe rupture, and micro-hole leakage. Among these, normal irrigation switching transients are characterized by known valve event indicators and short-term recovery; systemic pressure loss manifests as unidirectional changes at most nodes; pipe rupture is characterized by a short-term, large step pressure drop; and micro-hole leakage is characterized by a sustained, slow pressure drop at a localized node accompanied by an increase in high-frequency energy. The calculated node pressure change rate is...

[0108] ;

[0109] in, For the current moment's node pressure, The pressure sampling time interval is defined as follows. Based on historical healthy operating conditions, known pipe rupture events, and micro-leakage event samples, the pressure drop time scale discrimination parameter is established as follows: ,in This represents the duration of the maximum pressure drop during the tube rupture. For the minimum duration of microleakage, , These are the pressure drop rate thresholds for pipe rupture and micro-leakage, respectively. These are the cumulative pressure drop thresholds for pipe rupture and micro-leakage, respectively. The parameters are determined by the quantiles of the labeled samples. Take the 95th percentile value of the duration of the burst tube sample, preferably 5–30 seconds; Take the 5th percentile value of the duration of the microleakage sample, preferably 60–300 s; Take the 5th percentile of the absolute value of the pressure drop rate of the burst tube sample, preferably 0.005 to 0.020 MPa / s; Take the 99th percentile of the absolute value of the pressure drop rate of the healthy sample and below The preferred value is 0.0001 to 0.0020 MPa / s; The preferred pressure is 0.05–0.15 MPa. The preferred value is 0.005–0.030 MPa. The final value is determined by selecting the threshold that maximizes the Youden exponent using the ROC curve of the validation set.

[0110] If a local abnormal node has a known valve action event identifier, and the pressure change rate satisfies Meanwhile, cumulative pressure reduction Pressure within the time window If the internal state recovers to the dynamic confidence envelope, it is determined to be a normal rotational irrigation switching transient.

[0111] If the proportion of nodes with average spatial correlation below the synchronous change threshold exceeds a preset proportion threshold, and the pressure change rate satisfies... Meanwhile, cumulative pressure reduction satisfy Pressure within the time window If the pressure continues to decrease, it is determined to be a systemic slow decompression;

[0112] If the node pressure is within the time window A rapid decrease occurs within, and the rate of pressure change satisfies Meanwhile, cumulative pressure reduction If so, it is determined to be a sudden pipe rupture;

[0113] If the node pressure is within the time window The pressure decreases slowly and continuously, and the rate of pressure change satisfies... Meanwhile, cumulative pressure reduction High-frequency energy of pressure signals If the leakage rate is higher than the health status threshold, it is determined to be a micropore leakage.

[0114] Pressure signal high frequency energy Calculated using discrete wavelet decomposition of the detrended stress sequence within the sliding window, with a sampling frequency of [missing value]. Time selection frequency band [0.1] 0.5 ] Detail coefficients within, calculated .in, For wavelet detail coefficients, This is a detrended stress sequence. The health status threshold is taken from healthy samples of the same irrigation pattern. The 99th percentile. Cumulative pressure drop. The definition is as follows: From the first sustained period of a node's pressure below its dynamic baseline envelope lower bound. The moment Starting from this point, the node's mean relative to the baseline prediction within the anomaly persistence window... The cumulative value of the negative bias over the sampling time, i.e. ,in For the actual pressure measured at the current moment, This is the baseline predicted mean at the current time. The sampling interval is indicated by a negative sign, meaning sampling only occurs when... Accumulated over time to ensure that positive fluctuations are not included in the pressure drop. When the node pressure recovers to within the reference envelope or enters a positive deviation state and the duration exceeds the preset recovery confirmation window (preferably 3-5 sampling periods), Reset to zero, awaiting the next abnormal event to begin accumulating. Correspondingly, the cumulative pressure drop threshold for pipe rupture. The 5th percentile of the cumulative pressure drop from the initial moment to the moment of maximum pressure drop in historical pipe rupture samples is used as the cumulative pressure drop threshold for micro-leakage. The 95th percentile value of the cumulative pressure drop from the start time to the judgment time in the historical micro-leakage sample is taken and determined by statistical analysis of the labeled sample, and meets the following conditions: .

[0115] Regarding step S4, it is triggered only when step S3 determines it to be a systemic pressure disturbance, i.e., when the residuals of the entire network or most nodes show a unidirectional drift and the pressure changes between nodes are highly synchronized. Causes of this situation include changes in the water source level, gradual clogging of the head filter, or voltage fluctuations caused by the start-up and shutdown of pumping stations in adjacent areas. Such disturbances cause an overall shift in the network pressure, requiring coordinated regulation of pumps and valves to restore the network pressure to the reference envelope range. Specifically, when the dynamic residual sequence represents a systemic pressure fluctuation, the average drift is calculated based on the pressure residuals of all network nodes, and the average drift is converted into a pressure compensation control quantity. The edge computing gateway generates frequency adjustment commands for the head variable frequency pump and opening adjustment commands for the field electric pressure regulating valves based on the pressure compensation control quantity. Through coordinated control of global pressure compensation and local pressure correction, the real-time pressure is restored to the dynamic hydraulic reference envelope range.

[0116] First, calculate the weighted average drift of the pressure residuals of all nodes in the network. Let the nodes... The reliability weight is This weight can be set based on sensor accuracy, historical data quality, or the importance of the node in the pipeline network. Calculate the average drift across the entire network. ,in For nodes At sampling time The pressure deviation. The pressure compensation amount is generated using the discrete PI algorithm.

[0117] ;

[0118] in, , These are the proportional and integral coefficients, respectively. This represents the length of the integration window. This is the pressure compensation control value. This indicates a need for increased pressure. This indicates that pressure reduction is needed. The proportional term is used to quickly respond to the current drift, and the integral term is used to eliminate steady-state error.

[0119] After pressure compensation is obtained, the compensation amount is distributed to the actuator based on the pump pressure sensitivity and valve opening sensitivity. Pump pressure sensitivity and the The opening sensitivity of each pressure regulating valve The compensation allocation formula is as follows:

[0120] ,

[0121] ;

[0122] in, This refers to the frequency adjustment of the variable frequency water pump. For the first The opening adjustment amount of the electric pressure regulating valve. This is the pump-valve allocation coefficient, with a value range of 0 to 1, used to adjust the compensation weight between the pump and the valve. For the first The proportion of the absolute value of the residual of each partition to the sum of the absolute values ​​of the residuals of all partitions, and satisfying the following conditions: . For the first The residuals corresponding to the partitions, , These are the lower and upper limits for the frequency adjustment amount; , These are the lower and upper limits of the valve opening adjustment amount; This is a limiting function used to restrict the adjustment amount within the allowable range.

[0123] Pump valve distribution coefficient The value selection strategy is as follows: when the proportion of anomalous nodes in the same direction is not less than 70% or the head pressure deviates from the envelope at the same time, take... Prioritize adjusting the variable frequency water pump; when the anomaly is concentrated in a single zone and the head pressure is within the envelope, take... Prioritize adjusting the corresponding zone's electric pressure regulating valve; otherwise, take... Implement coordinated pump and valve regulation. The edge computing gateway will control the current pump frequency. With frequency adjustment The sums form the frequency setpoint, which is then used to determine the current valve opening. With opening adjustment amount The values ​​are added together to form the opening setpoint, and after passing through rate of change limiting, upper and lower limit checks, and dead zone determination, the final control command is generated. Rate of change limiting prevents sudden command changes from impacting the actuator; upper and lower limit checks ensure the setpoint does not exceed the equipment's allowable range; and dead zone determination avoids frequent actions caused by minor adjustments. The pump and valve control command generated in this step directly acts on the actuator, restoring the real-time pressure to the dynamic hydraulic reference envelope established in step S1.

[0124] Regarding step S5, it is triggered when a local micro-leakage event is determined in step S3, i.e., when a negative outlier occurs in the residual of a specific node and its correlation with the pressure of neighboring nodes decreases. Causes of this situation include branch pipe rupture, loose joints, or tiny holes caused by bird pecking or insect bites. This step utilizes network topology information and residual propagation direction to locate the leaking branch, without relying on a precision flow meter. First, based on a pre-stored drip irrigation network topology diagram, pressure sensor nodes are mapped to nodes in the topology diagram, and a node adjacency matrix is ​​established according to the pipe segment connection relationships. If pressure node and If adjacent monitoring locations are located on the same unbranched pipe section, or if the two are separated only by a single pipe fitting without a sensor and there are no other sensor nodes, then... Otherwise, it is 0, and Simultaneously establish the side length matrix. , Take the actual length of the pipe segment between the two nodes.

[0125] Calculate the absolute value of the dynamic residuals at each node. The comprehensive dynamic residual is defined as follows:

[0126] ;

[0127] in, For pressure deviation, For the rate of change deviation, Spatial correlation deviation; , , These represent the standard deviations of healthy samples for the three categories of characteristics mentioned above. , The weights are used to adjust the contributions of the rate of change bias and spatial correlation bias to the overall residual, and it is preferred that both be set to 1. This comprehensively reflects the degree of anomaly at the nodes across three dimensions: pressure amplitude, changing trend, and spatial correlation. Nodes exceeding the 99th percentile threshold of healthy samples constitute an anomaly set, which is then sorted in descending order of residual magnitude, with the node with the largest residual magnitude serving as the initial anomaly node. Subsequently, based on the adjacency matrix... A breadth-first search is performed on the initial abnormal nodes, accepting only connected nodes with negative residuals exceeding a threshold, thus obtaining the set of abnormal nodes connected to them. A negative residual refers to pressure deviation. This means that the pressure at this node is below the healthy baseline.

[0128] Based on the inclusion relationship between the set of anomalous nodes and the pipeline topology, the corresponding minimum continuous topological unit is determined. This inclusion relationship means that the node set of the candidate topological unit contains all accepted anomalous nodes, and the corresponding pipeline segments are topologically connected. In the candidate subgraphs that satisfy this condition, the edge length matrix is ​​used... The actual total length of pipe segments contained in each candidate topology unit is calculated. The smallest continuous topology unit is selected based on the principle of having the fewest pipe segments and the shortest total pipe length. The location corresponding to this topology unit is designated as the leak candidate area, and an alarm is triggered. When the smallest continuous topology unit contains only two adjacent monitoring nodes, the pipe segment between the two nodes is designated as the leak candidate area. When the smallest continuous topology unit contains a bifurcation point, the downstream branch after the bifurcation is selected based on the increasing direction of the residual along the water flow direction, further narrowing down the leak location range. The leak candidate area and alarm information output in this step serve as the basis for maintenance decisions and as one of the inputs for subsequent determination of whether baseline reconstruction is needed.

[0129] Regarding step S6, the alarm flags, residual statistics, and maintenance records generated during long-term operation in steps S4 and S5 are received to determine whether baseline reconstruction is necessary. Conditions triggering baseline reconstruction include: maintenance personnel registering equipment repair or replacement events on the human-machine interface; or, under the same irrigation rotation mode, at least 70% of the nodes in the entire network exhibit unidirectional mean drift for five consecutive irrigation cycles, and the average residual exceeds twice the original healthy standard deviation, while no local anomalies are detected (preset unidirectional mean drift conditions). These conditions indicate that the hydraulic characteristics of the pipeline network have undergone irreversible changes, the original baseline is no longer applicable, and manual confirmation is required before entering the baseline reconstruction mode. After entering the reconstruction mode, the edge computing gateway continuously collects pressure data for at least 20 normal irrigation cycles. To ensure the purity of the reconstructed baseline, samples need to be screened: samples with alarm flags, samples with comprehensive residuals exceeding the 99th percentile threshold, and samples with average spatial correlation below the local anomaly threshold are removed. Only healthy samples whose valve status, pump frequency, and target flow rate match the corresponding mode label are retained.

[0130] Using the selected healthy samples, the stress change prediction model was retrained according to the Gaussian process regression training method in step S1. Specifically, this included: using time as input and stress as output, performing Gaussian process regression with a squared exponential kernel function; and optimizing the hyperparameters using the L-BFGS-B multi-initial-value optimization method. , and Recalculate the predicted mean for each node. Prediction variance Dynamic confidence envelope upper and lower bounds and and the standard deviation required for residual standardization. , , The new model replacing the old model must meet the following validation conditions: the validation set health coverage rate is not less than 95%; the detection rate of known abnormal samples is not less than that of the old model; and there are no abnormal alarms for two consecutive complete rotation irrigation group operation cycles. If any condition is not met, the old pressure change prediction model will be rolled back, and the original baseline will continue to be used. After the new pressure change prediction model passes validation, it will officially replace the old model, and the updated baseline model will be fed back to step S1, thus completing the closed-loop update.

[0131] This self-healing mechanism enables the system to adapt to long-term evolution such as equipment aging, water source changes, and pipeline renovation, maintaining consistency between the baseline model and the actual pipeline state.

[0132] This embodiment also includes an adaptive reference reconstruction network pressure collaborative stabilization and leakage diagnosis system, which consists of four parts: a sensing layer, an edge computing layer, an execution layer, and a human-machine interaction layer. The sensing layer is a wireless pressure sensor network responsible for collecting pressure data from key nodes in the drip irrigation network; the edge computing layer is an edge computing gateway with built-in data storage units and processors, running core algorithms such as adaptive hydraulic reference models, dynamic residual analysis, feature decoupling, collaborative stabilization, topology positioning, and reference self-healing updates; the execution layer includes a head-mounted variable frequency pump and field electric pressure regulating valves, responsible for executing pressure regulation commands; the human-machine interaction layer is used for alarm display, maintenance registration, and reference reconstruction confirmation. The entire system connects sensors, gateways, and actuators via wireless communication to achieve real-time monitoring, anomaly diagnosis, and intelligent control of drip irrigation network pressure.

[0133] The sensing layer is a wireless pressure sensor network, including one sensor at the head outlet, one sensor at the inlet of each main pipe, and one sensor at the inlet of each irrigation branch pipe. For branch pipes longer than 100 m or with elevation changes exceeding 5 m, an additional sensor is added every 50–100 m or at each significant elevation inflection point, and at least one check sensor is installed at the end. Sensor placement optimization is constrained by the observability of candidate pipe segments, ensuring that there are at least two monitoring nodes on both sides of any candidate leaking pipe segment that can form a residual propagation direction; under this constraint, the location is selected with the goal of minimizing the number of sensors and the shortest maximum unmonitored pipe segment length. The pressure data acquisition cycle is preferably 1–5 s. The sensors upload real-time pressure data to the edge computing gateway via wireless communication, providing a data foundation for dynamic residual calculation.

[0134] The edge computing layer consists of edge computing gateways. These gateways have built-in data storage units and processors for storing standard hydraulic benchmark models and executing dynamic residual analysis algorithms. Their main functional modules are as follows:

[0135] The multi-mode dynamic hydraulic benchmark module collects node pressure data for different irrigation modes across multiple healthy operating cycles during the system's healthy operation period. Mode labels are defined by the set of open valves, irrigation branch number, target flow rate, and pump operating level. For healthy samples with the same mode label, Gaussian process regression is used to train and generate pressure change prediction models for each node, outputting the predicted mean and variance. Maximum likelihood estimation and the L-BFGS-B algorithm with boundaries are used to optimize the model hyperparameters, and the dynamic pressure confidence envelope is determined through validation set coverage verification. The optimized pressure change prediction model and dynamic confidence envelope parameters are stored in the benchmark library, forming the multi-mode dynamic hydraulic benchmark library, which serves as a comparison benchmark for subsequent residual calculations.

[0136] During real-time operation, the dynamic residual calculation module reads the current valve feedback, pump frequency, and irrigation task number to form the current mode label, and matches the nearest neighbor model from the benchmark library. It then uses a sliding time window to calculate the pressure deviation, rate of change deviation, and spatial correlation deviation for each node, and standardizes them according to the standard deviation of healthy samples to form a three-dimensional dynamic residual matrix. The pressure deviation is the difference between the measured pressure and the baseline predicted mean; the rate of change deviation is the difference between the pressure change rate and the baseline change rate; and the spatial correlation deviation is constructed based on the Pearson correlation coefficient between the pressure sequences of the node and its neighboring nodes.

[0137] The spatiotemporal feature decoupling module decouples the dynamic residual matrix spatiotemporally, introducing spatial Pearson correlation coefficient and pressure drop rate fingerprint. It calculates the average spatial correlation between nodes and compares it with synchronous change thresholds and local anomaly thresholds: if the residuals of all nodes in the network drift in the same direction and have a high correlation, it is determined to be a systemic pressure disturbance; if the residuals of local nodes are negatively outliers and their correlation with their neighbors decreases, it is determined to be a local micro-leakage event; if multiple spatially adjacent nodes form a continuous anomalous region and propagation is topologically constrained, it is determined to be a regional hydraulic anomaly; if no obvious spatial correlation or continuous anomaly is observed, it is determined to be a normal random disturbance. For local anomalies, pressure drop rate fingerprint and high-frequency energy of pressure signals are further used to distinguish between micro-hole leaks and sudden pipe bursts.

[0138] The collaborative stabilization control module, addressing systemic pressure disturbances, generates pressure compensation control quantities using a discrete PI algorithm based on the weighted average drift of pressure residuals across all network nodes. The compensation quantities are allocated to the frequency adjustment of the variable frequency pump and the opening adjustment of the electric pressure regulating valve, based on pump pressure sensitivity and valve opening sensitivity. After rate of change limiting, upper and lower limit verification, and dead zone determination, a control command is generated to drive the actuators and restore the real-time pressure to the reference envelope range.

[0139] The topology localization and early warning module targets localized micro-leakage events by establishing a node adjacency matrix and side length matrix based on a pre-stored drip irrigation network topology map. It calculates the absolute value of the comprehensive dynamic residual for each node, identifying anomaly sets composed of nodes with negative residuals exceeding the 99th percentile threshold of healthy samples. A breadth-first search is then performed starting from the node with the largest residual to obtain a set of connected anomaly nodes. Based on the inclusion relationship between the anomaly node set and the topology, the smallest continuous topological unit is selected as a leak candidate region, triggering an alarm.

[0140] The baseline self-healing update module determines a permanent change in operating conditions and triggers a baseline reconstruction candidate when maintenance personnel register a repair event, or when at least 50% of nodes across the entire network exhibit a mean drift in the same direction for five consecutive irrigation cycles under the same irrigation mode, and the average residual exceeds twice the original healthy standard deviation without detecting any local anomalies. After manual confirmation, healthy samples from normal irrigation cycles are collected and screened, eliminating alarm markers, samples with abnormal comprehensive residuals, and samples with excessively low spatial correlation. The Gaussian process regression model is retrained, and the healthy coverage and anomaly detection rate are verified. If the conditions are met, the old model is replaced and fed back to the baseline library; otherwise, the old model is reverted.

[0141] The processing performance of the edge computing gateway meets the following requirements: the processing time from the end of a computing window to the output of an anomaly determination does not exceed 2 seconds; the response time from the determination of a systemic disturbance to the issuance of a pump and valve control command does not exceed 3 seconds; and the status feedback cycle of the actuator does not exceed 5 seconds.

[0142] The execution layer, or actuators, includes the head variable frequency water pump and the field electric pressure regulating valve. The head variable frequency water pump changes the outlet pressure by adjusting the frequency, while the field electric pressure regulating valve corrects the local pressure by changing its opening degree. The edge computing gateway adds the current pump frequency to the frequency adjustment amount to form a frequency setpoint, and adds the current valve opening degree to the opening adjustment amount to form an opening setpoint. After passing the rate of change limit, upper and lower limit verification, and dead zone judgment, control commands are generated. Upon receiving the commands, the actuators adjust their operating status and feed the status back to the edge computing gateway.

[0143] The human-machine interface (HMI) layer displays pipeline pressure status, alarm information, potential leak areas, and maintenance suggestions. Maintenance personnel can register equipment repair or replacement events through the interface and confirm the initiation of the baseline reconfiguration mode. When the system triggers a baseline reconfiguration candidate, it enters baseline reconfiguration mode after manual confirmation. During the trial operation of the new baseline model, administrators can view coverage, detection rate, and trial alarm status to decide whether to replace the old model.

[0144] After the system powers on, the edge computing gateway loads a multi-mode dynamic hydraulic benchmark library, receives pressure sensor data in real time, and generates a current mode label. During each sampling cycle, the dynamic residual calculation module generates a dynamic residual matrix, and the spatiotemporal feature decoupling module determines the anomaly type. If it is a systemic disturbance, the collaborative stabilization control module generates pump and valve control commands and issues them for execution; if it is a localized micro-leak, the topology location and early warning module identifies the leak branch and issues an alarm; if it is a normal disturbance, no control response is triggered. During long-term operation, the benchmark self-healing update module determines whether to reconstruct the benchmark based on alarm records and maintenance events. After reconstruction, the benchmark library is updated to form a closed loop.

[0145] Example 2

[0146] Regarding the adaptive baseline reconstruction pipeline pressure collaborative stabilization and leakage diagnosis system and method in Example 1 during actual operation, this example diagnoses and stabilizes the following different types of abnormal situations, which are described below in conjunction with specific scenarios.

[0147] A localized micro-leakage event caused by bird pecking occurred. During the third round of irrigation operation on May 10th, bird pecking in the field caused a hole of approximately 2 mm in diameter in the drip irrigation tape near node P3. Due to the small leakage, the head flow meter did not show significant changes. At the 30-minute mark of irrigation operation, the edge computing gateway detected a pressure of 0.28 MPa at the head pressure node P0, which was basically consistent with the reference pressure. However, the pressure at node P3 dropped to 0.138 MPa, resulting in a dynamic residual of approximately -0.012 MPa compared to the normal reference pressure of 0.15 MPa. Further calculation of the spatial Pearson correlation coefficient between the pressure nodes revealed a decreased synchronicity of pressure changes between node P3 and other normal nodes, indicating that the anomaly was not caused by changes in water source pressure or disturbances in the head equipment, but rather was a localized anomaly. Furthermore, the absolute value of the residual at node P3 exceeded the range of the dynamic pressure envelope. Further analysis of the P3 pressure change process revealed that it exhibited a slow and prolonged decrease, accompanied by an increase in the high-frequency disturbance energy of the pressure signal, consistent with the characteristics of localized jet disturbance in a hole. Therefore, the system determined that the event was a localized micro-leak and located it in the drip irrigation tape area near P3.

[0148] A sudden rupture of drip irrigation tape occurred. In another test scenario, after artificially simulating a sudden rupture of drip irrigation tape in a branch pipe, the pressure at the abnormal node rapidly dropped from 0.25 MPa to 0.08 MPa. The duration of the pressure drop was less than the set burst time threshold Tburst, while the rate of pressure change exceeded the burst detection threshold. The edge computing gateway detected a rapid increase in the dynamic residual of the abnormal node and determined the largest abnormal node and its adjacent area through a network topology search, identifying the fault location in the corresponding branch pipe area. Due to the short-term step drop characteristic of this pressure change, the system identified it as a sudden rupture event, sent an alarm message to the management personnel, and simultaneously shut off the corresponding area's electric valve to reduce the impact of continuous leakage.

[0149] The system experienced a baseline drift event due to long-term changes in water source pressure. During long-term operation, the system inlet pressure decreased from 0.30 MPa to 0.24 MPa due to changes in the water source level, causing a synchronous decrease in pressure at all network nodes. At this time, the dynamic residuals of each pressure node all showed the same direction of change, and the spatial Pearson correlation coefficient between nodes remained at a high level (ρi>θsync). The system determined that this anomaly was a systemic pressure fluctuation rather than a local leakage. The edge computing gateway generated a pressure compensation control quantity based on the average pressure drift of the entire network. By coordinating the adjustment of the frequency of the variable frequency water pumps and the opening of the field electric pressure regulating valves, the network pressure was restored to the normal operating range.

[0150] Automatic hydraulic baseline reconstruction after equipment maintenance. When the system completes drip irrigation tape maintenance, equipment replacement, or a permanent change in water source conditions, management personnel confirm "repair complete" through the human-machine interface, and the system automatically enters baseline reconstruction mode. The edge computing gateway re-collects pressure data from multiple consecutive normal irrigation cycles and filters healthy operating samples based on pressure residual thresholds, spatial correlation indicators, and abnormal state records. The filtered healthy data is used to retrain the pressure baseline model, update the expected pressure curves and dynamic pressure envelope range of each node, and replace the original baseline model, thereby eliminating baseline drift caused by equipment aging, changes in water source, and changes in pipeline conditions.

[0151] The above-mentioned multiple operating scenarios demonstrate that this application does not target the detection of a single type of leakage. Instead, it uses an adaptive hydraulic benchmark model, dynamic residual analysis, spatial feature decoupling, and topology positioning methods to identify and respond to various abnormal states in drip irrigation networks, such as local micro-leakage, sudden rupture, system pressure fluctuations, and long-term operating status changes. Combined with pump and valve collaborative control and automatic benchmark reconstruction, it improves the operational stability, fault diagnosis reliability, and intelligent operation and maintenance capabilities of drip irrigation networks.

[0152] To implement the above embodiments, this application also discloses an electronic device. (Refer to...) Figure 5 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0153] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although... Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0154] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by the processing device 501, it performs the functions defined in the methods of some embodiments of this disclosure.

[0155] It should be noted that, in some embodiments of this disclosure, the computer storage medium described above can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0156] In some embodiments of this disclosure, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer storage medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0157] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and end-to-end networks (e.g., ad hoc end-to-end networks), as well as any currently known or future-developed networks.

[0158] The aforementioned computer storage medium may be included within the aforementioned electronic device, or it may exist independently without being assembled into the electronic device. The aforementioned computer storage medium carries one or more programs that, when executed by the electronic device, enable the electronic device to implement an adaptive reference reconfiguration pipeline pressure co-stabilization and leakage diagnosis method.

[0159] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0160] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings.

[0161] For example, two consecutively represented blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be located in a processor, and the names of these units do not necessarily constitute a limitation on the unit itself.

[0162] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0163] All technologies not described in detail in this invention are existing technologies. The above descriptions are merely some preferred embodiments of this disclosure and explanations of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalent features without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for adaptive baseline reconstruction of pipeline network pressure co-stabilization and leakage diagnosis, characterized in that, Includes the following steps: Step S1: Collect node pressure data of different irrigation modes under multiple healthy operating cycles to form mode labels, use Gaussian process regression to train and generate pressure change prediction models and dynamic confidence envelopes for each node, and construct a multi-mode dynamic hydraulic benchmark library. Step S2: Read the current mode label, match the nearest pressure change prediction model from the multi-mode dynamic hydraulic benchmark library, calculate the pressure deviation, change rate deviation and spatial correlation deviation using a sliding time window and standardize to form a dynamic residual matrix; Step S3: Decouple the dynamic residual matrix by spatiotemporal features, introduce the spatial Pearson correlation coefficient as a discrimination index to distinguish between global pressure disturbances and local micro-leakage, and output the anomaly type and node set; Step S4: For the global pressure disturbance, a pressure compensation amount is generated using a discrete PI algorithm based on the weighted average drift of the network residuals, and control commands are generated to the variable frequency pump and electric pressure regulating valve according to the pump and valve sensitivity. Step S5: Based on the local micro-leakage, perform a breadth-first search on the negative outlier residual nodes based on the pipeline topology adjacency matrix, and select the smallest continuous topological unit as the leak candidate region by combining the topological inclusion relationship. Step S6: Based on the anomaly determination, control execution, and maintenance registration records from steps S3 to S5, a benchmark reconstruction is triggered. Healthy samples are selected to retrain the pressure change prediction model, and after verification, the old model is replaced. Feedback is then sent to step S1 to complete the self-healing closed loop.

2. The adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis method according to claim 1, characterized in that, The construction of the multi-mode dynamic hydraulic reference pool in step S1 includes: Collect node pressure data for different irrigation modes under multiple healthy operating cycles, and use the set of open valves, irrigation branch number, target flow level and pump set operating level to form the mode label; For healthy samples with the same pattern label, a Gaussian process regression model is used to train the pressure change prediction model for each node as a function of irrigation time, and the predicted mean is output simultaneously. and prediction variance ; The hyperparameters of the pressure change prediction model are optimized using maximum likelihood estimation and the L-BFGS-B algorithm with boundary conditions. And based on the predicted mean and prediction variance Determine the dynamic confidence envelope; The optimized pressure change prediction model and the dynamic confidence envelope parameters are stored in the benchmark library to form a multi-mode dynamic hydraulic benchmark library.

3. The adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis method according to claim 1, characterized in that, The calculation of the dynamic residual matrix in step S2 includes: reading the current valve feedback, pump frequency and irrigation task number to form the current mode label, and matching the nearest neighbor model from the multi-mode dynamic hydraulic benchmark library; The pressure deviation at each node is calculated using a sliding time window. Deviation in rate of change Spatial correlation deviation ; The three biases are standardized according to the standard deviation of the healthy samples to form a node residual vector, which is then arranged by node number to form a dynamic residual matrix. .

4. The adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis method according to claim 1, characterized in that, The spatiotemporal feature decoupling in step S3 includes: calculating the spatial Pearson correlation coefficient between nodes and the average spatial correlation between the target node and its neighboring nodes, and comparing it with the synchronization change threshold and the local anomaly threshold; if the dynamic residual sequence of all nodes in the network shows synchronous drift in the same direction and the average spatial correlation is not lower than the synchronization change threshold, it is determined to be a systemic pressure disturbance; if the dynamic residual sequence of local nodes shows a continuous negative abnormal deviation and the average spatial correlation is lower than the local anomaly threshold, it is determined to be a local micro-leakage event; the local micro-leakage event is further identified by pressure drop rate fingerprint and high-frequency energy of pressure signal to distinguish between normal irrigation switching transients, systemic slow pressure loss, sudden pipe bursts and micro-hole leaks.

5. The adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis method according to claim 4, characterized in that, The identification method utilizing pressure drop rate fingerprinting and high-frequency energy of pressure signals includes: Compute the pressure change rate of the node And establish pressure drop time scale discrimination parameters. in This represents the duration of the maximum pressure drop during the tube rupture. For the minimum duration of microleakage, , These are the pressure drop rate thresholds for pipe rupture and micro-leakage, respectively. These are the cumulative pressure drop thresholds for pipe rupture and micro-leakage, respectively. If a local abnormal node has a known valve action event identifier, and the pressure change rate satisfies Meanwhile, cumulative pressure reduction Pressure within the time window If the internal state recovers to the dynamic confidence envelope, it is determined to be a normal rotational irrigation switching transient. If the proportion of nodes with average spatial correlation below the synchronous change threshold exceeds a preset proportion threshold, and the pressure change rate satisfies... Meanwhile, cumulative pressure reduction satisfy Pressure within the time window If the pressure continues to decrease, it is determined to be a systemic slow decompression; If the node pressure is within the time window A rapid decrease occurs within, and the rate of pressure change satisfies Meanwhile, cumulative pressure reduction If so, it is determined to be a sudden pipe rupture; If the node pressure is within the time window The pressure decreases slowly and continuously, and the rate of pressure change satisfies... Meanwhile, cumulative pressure reduction High-frequency energy of pressure signals If the leakage rate is higher than the health status threshold, it is determined to be a micropore leakage.

6. The adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis method according to claim 1, characterized in that, The definition The drift amount is obtained by weighted average of the residuals of all nodes in the network. The pressure compensation control quantity is generated using a discrete PI algorithm. ; Based on the pump pressure sensitivity and valve opening sensitivity, the pressure compensation control quantity is allocated as the frequency adjustment quantity of the variable frequency water pump. and the adjustment amount of the electric pressure regulating valve opening ; Current frequency of variable frequency water pump With the frequency adjustment amount The sums are the frequency setpoint of the variable frequency water pump. Set the current valve opening of the electric pressure regulating valve Adjustment amount of the opening of the electric pressure regulating valve The sum of these values ​​determines the opening degree of the electric pressure regulating valve. After the rate of change is limited, upper and lower limits are checked, and dead zone is determined, control commands are generated to drive the variable frequency water pump and electric pressure regulating valve to restore the real-time pressure to the reference envelope.

7. The adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis method according to claim 1, characterized in that, The Based on a pre-stored drip irrigation network topology diagram, pressure sensor nodes are mapped to nodes in the topology diagram, and a node adjacency matrix is ​​established according to the pipe segment connection relationships. Simultaneously establish the side length matrix ; Calculate the absolute value of the comprehensive dynamic residual of each node, and form an anomaly set with negative residual nodes that exceed the 99th percentile threshold of healthy samples. Using the node with the largest residual as the initial outlier node, based on the adjacency matrix A breadth-first search is performed on the initial abnormal nodes, and only connected nodes with negative residuals that exceed a threshold are accepted to obtain a set of connected abnormal nodes. Based on the inclusion relationship between the abnormal node set and the pipeline topology, the side length matrix is ​​used. Calculate the total length of pipe segments in the candidate topology units, and select the smallest continuous topology unit as the leakage candidate area and trigger an alarm in order of least number of pipe segments and shortest total length of pipe segments.

8. The adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis method according to claim 1, characterized in that, The Determine whether a baseline reconstruction candidate is triggered based on maintenance registration events or preset same-direction mean drift conditions. After manual verification, healthy samples from normal irrigation cycles were collected and screened, and samples with alarm indicators, comprehensive residuals exceeding the 99th percentile threshold, and average spatial correlations below the local anomaly threshold were excluded. ; Retrain the Gaussian process regression model and validate the health coverage and anomaly detection rate. Replace the old model only if the health coverage of the validation set is not less than 95%, the detection rate of known anomalies is not less than that of the old model, and there are no abnormal alarms for two consecutive complete irrigation group operation cycles; otherwise, revert to the old model.

9. An adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis system, characterized in that, include: A wireless pressure sensor network is used to collect pressure data from various nodes in the drip irrigation network. An edge computing gateway, used to store standard hydraulic benchmark models and perform dynamic residual analysis, includes: The model dynamic hydraulic reference library module is used to build and store a multi-model dynamic hydraulic reference library; The system includes: a dynamic residual calculation module for calculating the dynamic residual matrix based on real-time pressure data and a baseline model; a spatiotemporal feature decoupling module for decoupling the dynamic residual matrix to distinguish between systemic disturbances and local micro-leakages; a collaborative stabilization control module for generating pump-valve collaborative control commands for systemic disturbances; a topology positioning and early warning module for locating leak branches and triggering alarms for local micro-leakages; a baseline self-healing update module for updating the hydraulic baseline model based on long-term operating data; an actuator including a head variable frequency pump and a field electric pressure regulating valve for receiving control commands from the edge computing gateway and adjusting pressure; and a human-machine interface for displaying alarm information, leak candidate areas, maintenance suggestions, and allowing maintenance personnel to register maintenance events and confirm baseline reconstruction.

10. The adaptive reference reconstruction pipeline pressure co-stabilization and leakage diagnosis system according to claim 9, characterized in that: The wireless pressure sensor network includes one sensor at the head outlet, one sensor at the inlet of each main pipe, and one sensor at the inlet of each irrigation branch pipe. For branch pipes with a length exceeding 100 m or an elevation change exceeding 5 m, an additional sensor is added every 50–100 m or at each significant elevation inflection point, and at least one verification sensor is installed at the end. The sensor arrangement is constrained by the observability of candidate pipe segments, ensuring that there are at least two monitoring nodes on both sides of any candidate leaking pipe segment that can form a residual propagation direction.

Images