An intelligent irrigation system and control method based on the Internet of Things
By combining the Internet of Things and digital twin models with a regularized linear inversion algorithm, the problems of accuracy, real-time performance, and economy in fault location of irrigation pipeline networks have been solved, achieving real-time location and control of minute faults with meter-level accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG LIANGCHENG ENVIRONMENTAL PROTECTION TECH CO LTD
- Filing Date
- 2026-03-02
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies struggle to balance accuracy, real-time performance, and cost-effectiveness in large-scale irrigation network monitoring. High-precision methods are expensive, while low-cost methods cannot accurately locate minute leaks or blockages.
An IoT-based intelligent irrigation control method is adopted. By acquiring real-time pressure data of the irrigation network, and using a digital twin model and a regularized linear inversion algorithm, combined with sparsely deployed pressure sensors, the system can achieve real-time fault location and control with meter-level accuracy.
With a reduced number of sensors deployed and limited computing resources, real-time location and control of minor faults were achieved, improving monitoring accuracy and cost-effectiveness.
Smart Images
Figure CN122172668A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of Internet of Things (IoT) and fluid pipeline monitoring technology, and in particular to an IoT-based intelligent irrigation system and control method. Background Technology
[0002] In modern large-scale precision agriculture, especially in the management of commercial drip irrigation systems, maintaining the health of the irrigation network and the uniformity of water supply is a core prerequisite for ensuring crop yield and quality. However, existing technologies for achieving refined health monitoring of large-scale networks are generally hampered by a fundamental technical contradiction: an irreconcilable conflict exists between monitoring accuracy and deployment economy. On the one hand, high-precision fault inversion methods based on complete hydraulic models rely on dense sensor deployments and massive real-time computations, resulting in excessively high hardware costs and computational loads, making them unsuitable for cost-sensitive agricultural scenarios. On the other hand, low-cost sparse monitoring schemes cannot accurately locate minute leaks or blockages due to insufficient information. This technical bias of either being expensive and precise or cheap and crude makes real-time, high-precision, and low-cost health diagnosis of irrigation networks an unresolved technical challenge. Summary of the Invention
[0003] Firstly, this application provides an intelligent irrigation control method based on the Internet of Things (IoT), aiming to solve the technical problem in existing technologies that makes it difficult to simultaneously achieve accuracy, real-time performance, and economy in locating faults in irrigation networks. The method includes: The system acquires real-time pressure vectors from multiple preset pressure sensor nodes within the irrigation network, as well as the real-time source status of the source nodes of the irrigation network. Based on the real-time source status and a pre-stored digital twin model of the irrigation network, a theoretical pressure baseline vector aligned with the timestamp of the real-time pressure vector is generated. Based on the real-time pressure vector and the theoretical pressure baseline vector, a pressure deviation vector is determined. Based on the pressure deviation vector and a pre-stored transfer admittance matrix, a regularized linear inversion algorithm is used to determine the equivalent pressure source vector characterizing the fault location and intensity. Based on the equivalent pressure source vector, the fault in the irrigation network is located and corresponding control is executed.
[0004] Optionally, obtaining the real-time source status of the source node of the irrigation network includes: synchronously collecting the instantaneous flow rate and instantaneous pressure of the source node to generate the real-time source status.
[0005] Optionally, after obtaining the real-time pressure vectors of multiple preset pressure sensor nodes within the irrigation network, the method further includes: performing data validity verification on the real-time pressure vectors to generate valid sensor masks; the determination of the pressure deviation vector specifically involves: performing the difference calculation between the real-time pressure vector and the theoretical pressure baseline vector only for valid sensors identified by the valid sensor masks, in order to generate the pressure deviation vector.
[0006] Optionally, before determining the equivalent pressure source vector, the method further includes: determining whether the number of effective sensors identified by the effective sensor mask is less than a preset minimum sensor number threshold; if the determination result is yes, then the execution of the regularized linear inversion algorithm is stopped, and a global sensor fault alarm is generated.
[0007] Optionally, generating the theoretical pressure baseline vector based on the real-time source state and the pre-stored digital twin model of the irrigation network specifically involves: dynamically updating the water head boundary conditions of the water source nodes in the digital twin model using the real-time source pressure in the real-time source state; and performing a steady-state hydraulic simulation in the dynamically updated digital twin model to generate the theoretical pressure baseline vector.
[0008] Optionally, the step of using a regularized linear inversion algorithm to determine the equivalent pressure source vector specifically involves: constructing a Tikhonov regularization objective function that includes the transfer admittance matrix, the pressure deviation vector, and regularization parameters; and solving the closed-form solution of the Tikhonov regularization objective function to obtain the equivalent pressure source vector.
[0009] Optionally, the step of locating the fault in the irrigation network based on the equivalent pressure source vector specifically involves: finding the element with the largest absolute value in the equivalent pressure source vector, determining the node index corresponding to the element as a fault candidate location, and determining the value of the element as the fault intensity; comparing the absolute value of the fault intensity with a preset diagnostic threshold, and confirming that a fault has occurred if the absolute value is greater than the diagnostic threshold.
[0010] Optionally, before obtaining the real-time pressure vector, an initialization step is further included: constructing an uncalibrated digital twin model based on the topology of the irrigation network; using a nonlinear optimization algorithm, with the hydraulic parameters of the pipe segments in the uncalibrated digital twin model as optimization variables and minimizing the error between the simulated pressure and the measured pressure as the objective function, performing iterative optimization to generate a calibrated digital twin model; and based on the calibrated digital twin model, constructing and storing the transfer admittance matrix by sequentially applying unit virtual pressure disturbances to each possible fault node and performing steady-state simulation.
[0011] Optionally, the step of locating the fault in the irrigation network and executing corresponding control based on the equivalent pressure source vector includes: querying the nearest controllable valve node upstream of the fault location in a pre-stored network topology database according to the fault location; generating a closing command for the controllable valve node and sending it to the corresponding field controller via an industrial control protocol to isolate the fault location.
[0012] Secondly, this application provides an IoT-based intelligent irrigation system, comprising: a data acquisition and preprocessing module configured to acquire real-time pressure vectors of multiple preset pressure sensor nodes within an irrigation network, and to acquire real-time source status of the source nodes of the irrigation network; a hydraulic digital twin and baseline generation module configured to generate a theoretical pressure baseline vector aligned with the timestamp of the real-time pressure vector based on the real-time source status and a pre-stored digital twin model of the irrigation network; an equivalent pressure source inversion and localization module configured to determine a pressure deviation vector based on the real-time pressure vector and the theoretical pressure baseline vector, and to determine an equivalent pressure source vector characterizing the fault location and intensity based on the pressure deviation vector and a pre-stored transfer admittance matrix using a regularized linear inversion algorithm; and a fault diagnosis and control decision module configured to locate faults in the irrigation network and execute corresponding controls based on the equivalent pressure source vector.
[0013] The beneficial effects of this application are as follows: The technical solution provided in this application transforms a computationally intensive nonlinear hydraulic dynamic simulation inversion problem into a sparse linear equation system problem that can be solved at the edge in milliseconds by constructing an equivalent pressure source inversion and positioning model. This makes it possible to achieve meter-level accuracy real-time positioning of minute faults under the stringent constraints of a significantly reduced number of sensor deployments and extremely limited computing resources. Attached Figure Description
[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0015] Figure 1 This is a schematic diagram of an IoT-based smart irrigation control method provided in an embodiment of this application.
[0016] Figure 2 This is a schematic diagram of the functional module structure of an intelligent irrigation system provided in an embodiment of this application.
[0017] Figure 3This is a schematic diagram illustrating the cross-domain principle mapping relationship between the equivalent pressure source model and the equivalent current dipole model in the embodiments of this application.
[0018] Figure 4 This is a schematic diagram of the system initialization and calibration process in an embodiment of this application. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0020] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0021] This application discloses an intelligent irrigation control method configured to execute in a distributed system comprised of resource-constrained, field-deployed edge computing nodes. The core of this method lies in deploying an equivalent pressure source inversion model. This model mathematically reconstructs a computationally intensive nonlinear hydraulic dynamic simulation problem into a system of linear equations, allowing the location and intensity of pipeline faults to be directly solved through a single matrix pseudo-inverse operation. Furthermore, the method couples a monitoring subsystem that performs high-frequency sampling of the pipeline source state to dynamically correct the boundary conditions of the linear equations. This enables millisecond-level response and meter-level accuracy in locating minute leaks or blockages using only the limited observations provided by sparsely deployed pressure sensors.
[0022] Reference Figure 1 This is a flowchart illustrating an IoT-based smart irrigation control method provided in an embodiment of this application. The method may include the following steps: S100: Obtain the real-time pressure vectors of multiple preset pressure sensor nodes within the irrigation network, and obtain the real-time source status of the source nodes of the irrigation network.
[0023] In a specific implementation scenario, the irrigation network is defined as a drip irrigation system with a complex topology used in a large-scale commercial farm (e.g., a vineyard covering hundreds of hectares). To monitor the system's condition, several pressure sensors are sparsely deployed at key nodes of the network, such as branches of the main pipe or pressure-sensitive end areas. Simultaneously, source sensors characterizing the overall operating status of the system are installed at the water source of the irrigation system, i.e., the main outlet pipe of the pumping station. The execution entity of this step S100 is an industrial-grade edge computing gateway deployed in the field. This gateway is configured to periodically and synchronously collect raw data streams reflecting the current state of the network from the physical sensors and convert these raw data streams into structured digital signals as data input for subsequent fault diagnosis processes. The execution of this step establishes the data foundation for the entire diagnostic process, and the synchronicity and fidelity of the collected data directly constrain the reliability of subsequent analysis.
[0024] In a specific implementation, S100 is deconstructed into a series of coupled data processing sub-steps, such as S105, S110, S120, and S130.
[0025] S105: Synchronously collect the instantaneous flow and instantaneous pressure of the source node to generate the real-time source status.
[0026] This sub-step is performed by a communication unit in the edge computing gateway configured with an industrial communication interface (e.g., an RS485 interface). The communication unit is programmed to operate in Modbus master mode and uses a preset, for example... The system polls at a frequency that broadcasts a read-hold register instruction with function code 0x03 to the source digital electromagnetic flowmeter and pressure transmitter, which are Modbus slaves. In response to this instruction, the digital electromagnetic flowmeter returns a register value representing the instantaneous volumetric flow rate through the mains, parsed to cubic meters per hour (m³ / h). The pressure transmitter then returns a register value representing the instantaneous gauge pressure in the main pipe, which, after parsing, is in bar (). These two synchronously collected parameters constitute a data pair that can accurately characterize the real-time operating status of the irrigation system's pumping stations. This data stream is then processed and formed into a time-series data stream. This data stream is cached in a fixed-size circular buffer in the volatile memory of the edge computing gateway for subsequent use by the dynamic baseline correction module.
[0027] For example, within a specific sampling period, the communication unit of the edge computing gateway sends the Modbus read command to a flow meter with address configuration 0x01 and a pressure gauge with address configuration 0x02 via its RS485 port. The 16-bit integer data returned by the flow meter is converted into a floating-point instantaneous flow value through a preset scaling factor. Meanwhile, the data returned by the pressure gauge is converted to obtain the instantaneous pressure value. These two values constitute a moment. A real-time snapshot of the source status is captured and pushed into the circular buffer. This precise, high-frequency capture of the source status enables the entire diagnostic system to adapt to the dynamics of pump stations in the real world caused by power grid voltage fluctuations or multi-region water supply switching, providing a crucial reference benchmark for subsequent accurate identification of anomalies in the pipeline network itself.
[0028] S110: Obtain the real-time pressure vector of multiple preset pressure sensor nodes within the irrigation network.
[0029] This sub-step is performed by an acquisition unit within the edge computing gateway, which integrates a multi-channel 24-bit analog-to-digital converter (ADC) module. Pressure sensors deployed throughout the pipeline network (e.g., pressure transmitters of the MPM489 model) are configured to linearly convert the water pressure sensed by their sensing elements into a range within... The standard industrial current signal is transmitted via shielded twisted-pair cable to the acquisition board of the edge computing gateway. The ADC module is configured to, for example, use a standard industrial current signal. The sampling rate is used to synchronously sample the current signal of each input channel. The sampled current value is first quantized into a dimensionless numerical value, and then, based on the pre-stored calibration parameters (i.e., range, for example) of each sensor... correspond ), via a linear transformation function: Mapped to have physical units ( The actual pressure value. At any given sampling time, all The pressure readings from each pressure sensor node are organized into a single... A column vector of dimension, that is, a sample in the original discrete pressure sequence. .
[0030] For example, assume the system is deployed with 5 ranges, all of which are The pressure sensor. At any given moment. The 5-channel ADC module of the edge computing gateway triggers a synchronous sampling, and the vector formed by the collected current values of each channel is... The processor in the acquisition unit applies the linear transformation function to each element of the current vector to generate the original pressure vector at that moment. For example, for the first channel, the mapped pressure value is... After performing this operation on all channels, the resulting original pressure vector is: This vector constitutes a pressure snapshot of the pipeline network over this tiny time slice. The sampling rate ensures that the system has the ability to capture potentially rapid pressure transients caused by minor leaks or blockages, providing a data foundation with sufficient time resolution for subsequent signal processing and feature extraction.
[0031] S120: Perform digital filtering on the original pressure sequence and source state data.
[0032] This sub-step, as a data preprocessing flow, is implemented in software by the edge computing gateway's processor. Its purpose is to suppress the high-frequency noise prevalent in the raw data stream. This noise may originate from electromagnetic interference in sensor circuitry, electronic noise within the sensors themselves, or minor water hammer effects caused by rapid valve movements in the pipeline network. To extract the pressure signal, typically exhibiting low-frequency variations and caused by physical events such as leaks or blockages, from the noise, this step deploys a digital low-pass filter. In a preferred embodiment, the filter is a fourth-order Butterworth infinite impulse response (IIR) filter. The filter's design parameters are configured as follows: The cutoff frequency, this configuration makes the frequency higher than The signal components are attenuated, while those with frequencies lower than [a certain value] are [attenuated]. The signal components characterizing real pressure changes are then allowed to pass through. The filters are applied respectively to… The sequence includes historical data sequences from each sensor and source state data. and The time series. After the filtering process, the generated denoised real-time pressure vector is labeled as... The smoothed source state is marked as and .
[0033] For example, the original pressure sequence of sensor 1 For example, this contains minute random fluctuations. When this sequence is fed into the cutoff frequency... After processing by the fourth-order Butterworth filter kernel, the output sequence will appear smooth, for example... If a tiny leak occurs at this point, causing a slow drop in pressure, the original sequence may evolve into... After the filtering process, the output sequence will retain this downward trend, for example... However, it filters out high-frequency jitter between sampling points. This filtering operation improves the signal-to-noise ratio, ensuring the accuracy of subsequent calculations of pressure deviation. Vectors can accurately reflect changes in the macroscopic hydraulic state of a pipe network, rather than irrelevant noise interference, which is crucial for detecting weak signals caused by minute leaks (which typically only cause pressure changes at the millibar level).
[0034] S130: Perform data validity verification on the denoised real-time pressure vector.
[0035] This sub-step is configured as a key component of the data quality assurance process and is also implemented in the processor of the edge computing gateway. Its function is to identify and flag invalid or abnormal data points caused by sensor hardware failures, signal transmission line interruptions, or transient strong interference. The verification process includes at least two orthogonal dimensions of checking logic. The first dimension is a range check logic, configured to determine whether each pressure reading falls within the sensor's predefined physical range (e.g., ...). Any reading outside this range (e.g., negative values or ADC saturation readings due to short circuits) is considered invalid. The second dimension is a statistical outlier detection logic. This logic maintains a statistical model for each sensor based on a sliding time window (e.g., the past 300 sampling points) and calculates the historical mean in real time. and standard deviation If the current reading exceeds a preset confidence interval, for example... If the reading is not found to be statistically outlier, it is marked as invalid. The output of this step is a validated pressure vector. And a corresponding Boolean effective sensor mask vector. .exist In the vector, data points deemed invalid are replaced with a non-numeric (NaN) identifier, while... In the vector, the Boolean value at the corresponding position is set to False, while valid data points are set to True.
[0036] For example, suppose the real-time stress vector has been denoised. for The verification logic first checks the fourth element in the vector. Perform a scope check, because it is smaller than The data point is deemed invalid because it exceeds the lower limit of the measurement range. Meanwhile, assuming statistical analysis of historical data from sensor 2 yields a mean value... The standard deviation is Thus defining a effective range Current reading Because it exceeds the upper bound of this range, it is deemed invalid by the outlier detection logic. Based on this, the final validated pressure vector generated in this step... for Its corresponding effective sensor mask for This mask vector will serve as a critical control input, guiding subsequent calculations to use only verified, reliable sensor data. This avoids the risk of contaminating the entire diagnostic result due to the failure of a single sensor, thus enhancing the robustness of the system.
[0037] S200: Determine the pressure deviation vector based on the real-time pressure vector and the theoretical pressure baseline vector.
[0038] After acquiring verified, high-quality field-measured pressure data, the next core task is to extract differential signals from this data that indicate pipeline anomalies. Since the pressure distribution in a healthy pipeline network dynamically changes with pump station operating conditions, valve scheduling, and normal fluctuations in source pressure, the purpose of step S200 is to precisely subtract the measured pressure from a theoretical healthy pressure predicted by a high-fidelity digital twin model under the same boundary conditions, thereby separating the pressure disturbance component caused solely by potential faults (such as leaks or blockages). This operation purifies the raw pressure information, which is mixed with multiple influencing factors, into a pure, clearly directional fault characteristic signal, namely, the pressure deviation vector.
[0039] In a specific implementation, S200 includes two sub-steps, S210 and S220, which are completed collaboratively by the hydraulic digital twin and baseline generation module (M200) and the equivalent pressure source inversion and positioning module (M300).
[0040] S210: Based on the real-time source status and the pre-stored digital twin model of the irrigation network, generate a theoretical pressure baseline vector aligned with the real-time pressure vector timestamp.
[0041] This sub-step is configured to use a pre-built and calibrated hydraulic digital twin model to calculate the theoretical pressure baseline in real time and dynamically, effectively suppressing interference caused by source pressure fluctuations. The digital twin model is a computational model that accurately reproduces the topology, hydraulic parameters, and sensor node locations of the physical pipe network in software. In each diagnostic cycle, the module receives the real-time source pressure collected by S105 and smoothed by S120. As a dynamic input, it first queries the current valve control commands to determine the pipeline connectivity status in the model; then, it inputs the measured data... The value (converted to head height) is set as the head boundary condition for the water source nodes in the model, replacing the preset static rated pressure value in the model. Based on this dynamically updated boundary condition, the module calls a lightweight hydraulic solver (e.g., solving the hydraulic balance equations based on the Hassen-Williams formula or the Darcy-Weisbach formula using the Newton-Raphson method) to perform a steady-state hydraulic simulation. The simulation output is the theoretical pressure distribution of all nodes in the pipe network. Finally, all... The theoretical pressure values at each sensor node location are collected and organized into a theoretical pressure baseline vector. .
[0042] For example, suppose at time... The smoothed source pressure output by S120 is Simultaneously, the control system status register shows that the solenoid valves of irrigation zones 1 and 3 are in the open state. In response to these inputs, the M200 module dynamically sets the water head value of the water source node within its internal digital twin model. (correspond The model then sets the valve components corresponding to partitions 1 and 3 in the model to be fully open. Subsequently, its hydraulic solver kernel is triggered, converges after iterative calculations, and solves for the steady-state pressure distribution of the pipeline network. Assuming the calculation results indicate that the theoretical pressures at the five sensor nodes are respectively... This vector is the vector relative to time. The measured data are perfectly aligned with the theoretical pressure baseline vector. Because this baseline vector is generated based on real, dynamic source pressure, it can accurately follow the normal fluctuations of source pressure. If the overall source pressure decreases, the calculated... The overall pressure will also decrease, thus offsetting this systemic pressure change in subsequent vector subtraction operations and avoiding false alarms.
[0043] S220: Determine the pressure deviation vector based on the real-time pressure vector and the theoretical pressure baseline vector.
[0044] This sub-step is logically executed by the M300 module, which is configured to perform the core's differential signal extraction operation. It receives the verified pressure vector from S130. and effective sensor mask and the theoretical pressure baseline vector from S210. The core operation of this step is a mask-based vector subtraction. Specifically, it first applies the... Vector pairs and Filter and keep only those in The pressure component corresponding to the valid sensors marked as True. Then, for the two filtered dimensions... ( The subvector of is subjected to element-wise subtraction. The output of the operation is a vector with dimension . Pressure deviation vector This vector Each element in They all precisely quantified the first The difference between the measured pressure of each effective sensor and its theoretical value under healthy conditions. Under ideal, fault-free conditions... It should be a noise vector close to zero. Conversely, if there is a leak or blockage in the pipeline network, this local disturbance will change the pressure field distribution, leading to... The system exhibits significant non-zero components with specific patterns, which together constitute a pressure fingerprint pointing to the source of the fault.
[0045] For example, continuing from the previous example, the output of S130 is... and S210 output This step first is based on from Extract the sub-vectors formed by the valid measured values. and from Extract the sub-vector formed by the baseline values at the corresponding positions. Then, perform vector subtraction: The generated A vector whose dimension is 3 (equal to) It encapsulates all the purified and valid information that can be used for subsequent inversion calculations.
[0046] S300: Based on the pressure deviation vector and the pre-stored transfer admittance matrix, a regularized linear inversion algorithm is used to determine the equivalent pressure source vector characterizing the fault location and intensity.
[0047] This step is configured to perform a reverse reasoning process from the phenomenon (pressure deviation vector) to the physical root cause (fault source). This reasoning is based on an Equivalent Pressure Source (EPS) model, which mathematically equates any minor fault (leak or blockage) in the piping network to a virtual pressure source applied to the fault point. Leaks are modeled as a pressure source with negative intensity (i.e., pressure sink), while blockages are modeled as a pressure source with positive intensity (i.e., pressure gain). Under the small perturbation assumption of this model, the pressure deviation observed on any sensor... Linearly represented as all in the pipeline network Equivalent pressure sources at each potential failure node The weighted sum of the equations. This linear mapping is fully defined by a pre-computed transfer admittance matrix (TAM) characterizing the static hydraulic transfer properties of the pipe network. Therefore, the fault location problem is transformed into solving a system of linear equations: The inversion problem. Given the number of sensors... Typically much smaller than the number of potential failure points. This problem is a typical underdetermined problem, and the TAM matrix is often ill-conditioned. To obtain a stable and physically meaningful unique solution, this step employs a regularized linear inversion algorithm.
[0048] In a specific implementation, S300 is decomposed into three sequentially executed sub-steps: S305, S310, and S320, all of which are completed in the M300 module.
[0049] S305: Determine whether the number of valid sensors is less than the preset minimum sensor number threshold.
[0050] This sub-step is configured to perform a pre-solvability check before invoking the computationally intensive inversion solver. It receives the valid sensor mask from S130. The number of valid sensors is determined by counting the elements that have a Boolean value of True. Subsequently, A minimum number of sensors is required, based on a pre-defined threshold specific to a particular network topology. Compare them. The value of is set based on the principle of ensuring the minimum observability of the inversion problem; for example, it can be set to ensure that there is at least one effective sensor on each main branch. If Less than This means that the currently available measurement information is insufficient to effectively constrain the inversion equations, and solving it will produce meaningless or misleading results. Therefore, if the comparison result is true, the system will suspend the inversion calculation for this cycle and route a specific global sensor fault alarm code to the fault diagnosis module M400.
[0051] For example, assuming a pipeline topology contains 5 main branches, what is the minimum sensor count threshold to ensure basic observability of the entire pipeline network? It was set to 5. At a certain moment, due to a circuit fault, the data from the three sensors were deemed invalid by S130. There are only 2 True values in the result. This step calculates... ,because The verification failed. Therefore, the inversion process was aborted, and an alarm code such as 0xF01A: SENSORSUBSYSTEMDEGRADED was generated. This verification logic serves as a crucial fault-tolerance mechanism, preventing the system from outputting unreliable location results when data is severely incomplete, and providing clear system-level maintenance alerts to operations and maintenance personnel.
[0052] S310: The equivalent pressure source vector is solved using a regularized linear inversion algorithm.
[0053] This step is the entity that performs the core mathematical operations. It first depends on... From the preloaded memory of the edge gateway, the dimension is From the complete TAM matrix, extract the rows corresponding to the effective sensors to form a dimension of submatrix Subsequently, it constructs an inversion objective function with Tikhonov regularization, which takes the form: This objective function consists of two terms: a data fitting term and a data fitting term. The required solution is After forward propagation through the model, the deviation from the measured values can be minimized. The L2 norm (i.e., Euclidean distance); regularization term Then for the solution vector Penalize its own L2 norm, where The regularization parameter controls the intensity of the penalty. This regularization term introduces a physical prior assumption that the fault source should be as simple as possible (minimize energy), thus selecting the smoothest and most stable solution when the underdetermined problem has infinitely many solutions. This least squares problem has a closed-form analytical solution: in The matrix is an identity matrix. The edge computing gateway performs this matrix operation by invoking a high-efficiency linear algebra library optimized for embedded platforms (such as Eigen or Armadillo).
[0054] For example, continuing from the previous example, The complete TAM matrix has the following dimensions: .according to The system extracts rows 1, 3, and 5 from the TAM matrix to form... submatrix Set regularization parameters (Dimensionless). The calculation process begins with matrix multiplication: (The result is one) (a square matrix), plus a regularization term: Next, for the matrix Perform the inverse operation Then calculate: Finally, through matrix-vector multiplication: Obtain a dimension as column vector solution Each element of this vector That is, in the first The equivalent pressure source intensity (in units) inverted at each potential failure node. ).
[0055] S320: Find the peak value in the equivalent pressure source vector to determine the fault candidate location and intensity.
[0056] Solve the complete equivalent pressure source distribution vector in S310. Then, this sub-step parses the data to extract the most direct fault indication information. The vector is viewed as a spatial distribution map of the fault source intensity across the entire pipeline network. This step is configured to locate the point of highest energy concentration on this distribution map. This step is based on the engineering assumption that in minor fault scenarios, there is usually only a single or dominant fault source; therefore, the most likely fault candidate location is determined by finding the global maximum value. Specifically, the equivalent pressure source vector is traversed. All of them Given a set of elements, we find and record the element with the largest absolute value by comparing the absolute values of each element. This element is denoted as . and its index position in the vector. The index position of the maximum value in the vector. They are directly mapped as candidate locations for faults. The algebraic value of this maximum value is then determined as the fault intensity at that location. The sign of the fault intensity is used to distinguish the fault type: if A negative value indicates a pressure sink, corresponding to a pipeline leak; if... A positive value indicates a pressure source, corresponding to a blockage in the pipe.
[0057] For example, suppose S310 calculates Vector (dimension 1) In the given information, the absolute values of most elements are less than 1. However, a significant value appears at index 257. After traversing the entire vector, confirm... This is the global maximum value. Therefore, the output of this step is a structured data object: {location_index: 257, magnitude: -0.12}. The physical interpretation of this result is: the system inverses the existence of a magnitude of [value missing] at node 257 of the pipeline network. The equivalent pressure source. The negative sign indicates the fault type is leakage. This concise {location, intensity} tuple encapsulates the core information ultimately produced by the entire inversion calculation process for downstream decision-making.
[0058] S400: Based on the equivalent pressure source vector, locate the fault in the irrigation network and execute corresponding control.
[0059] This step is the final decision-making stage of the methodology. It is configured to interpret and confirm the raw calculation results output by the S300, and ultimately generate user-operable diagnostic conclusions or trigger automated control responses. Because measurement noise and model errors can cause even healthy pipe networks to show small non-zero values... Peak value: This step uses a defined, statistically based threshold to distinguish real fault signals from background noise. Furthermore, if a fault is confirmed, its severity must be graded and mapped to specific alarm messages or control commands.
[0060] In a specific implementation, S400 is broken down into two sub-steps, S410 and S420, both of which are completed in the M400 module.
[0061] S410: Compare the absolute value of the fault intensity with a preset diagnostic threshold.
[0062] This sub-step is configured to execute the final fault confirmation logic. It receives the fault strength from S320. And compare it with a preset, floating-point diagnostic threshold. The comparison is performed. Its core logic is a Boolean judgment: If this condition is true, the system sets an internal status flag to FAULT_CONFIRMED; otherwise, the peak value is attributed to system noise, and the status flag remains HEALTHY. The diagnostic threshold... The value of is based on a large amount of historical data collected during the initial deployment of the system to confirm the health status of the pipeline network. The background noise peak value is obtained by running an inversion process on this data. It is determined by the statistical distribution. Specifically, This can be set to the 99.5th percentile of the distribution to ensure a false alarm rate of less than 0.5% under healthy conditions. Furthermore, this step is configured to process system-level alarm codes from the S305. If a global sensor fault alarm code is received, it will bypass the threshold comparison logic and directly generate a system-level maintenance alarm.
[0063] For example, suppose that a diagnostic threshold is determined through offline statistical analysis. S320 output fault strength This step calculates its absolute value. .because If the comparison result is true, the system's internal fault detection flag is set to True. If, in another calculation cycle, the S320 outputs... Its absolute value Not exceeding the threshold If the condition is not met, the fault determination flag remains False. This decision-making mechanism based on statistical thresholds provides an objective and quantitative standard for fault confirmation.
[0064] S420: Generates structured alarm information or control commands based on fault diagnosis indicators, location, and intensity.
[0065] After fault confirmation in S410, this sub-step is configured to translate the abstract diagnostic results into concrete information services or control actions. If the fault determination flag is True, it will integrate the fault location from S320. and intensity Information. It first queries a node index-geographic information mapping table pre-stored in non-volatile memory, and then retrieves the node index... (e.g., 257) is converted into actual geographic coordinates, pipe segment numbers, or human-readable location descriptions (e.g., the western branch pipe of the 5th row in the A section vineyard). Then, it is based on... The sign indicates the fault type (negative for leakage, positive for blockage). Then, based on... The absolute value of the fault severity level maps the severity to multiple levels (e.g., by comparing it to a set of grading thresholds to determine whether it is minor, moderate, or severe). Ultimately, this information is encapsulated into a structured alarm message, such as JSON: {"status": "FAULT", "type": "LEAK", "location": {"node_id": 257, "desc":"A-5-W"}, "severity": "MEDIUM", "magnitude": -0.12}. This JSON object can be published to a cloud-based IoT platform via the MQTT protocol for user apps or web management interfaces to subscribe to and visualize. In a more advanced automation implementation, the module can also automatically query the pipeline topology database based on the fault location, find the nearest solenoid valve upstream of the fault point, and send a write single coil instruction with function code 0x05 to the PLC controlling that valve via the Modbus / TCP protocol to achieve automatic fault isolation.
[0066] For example, for the confirmed fault mentioned above, the M400 module generates an alarm message: Medium-level leak detected! Location: West branch pipe of row 5 in area A (node 257). Equivalent pressure source intensity: -0.12 bar. Simultaneously, the system detects that node 257 is controlled by solenoid valve 3. Therefore, it sends a shut-off command to the corresponding address of the PLC, thereby cutting off the water supply to that area and awaiting maintenance personnel to arrive for repairs.
[0067] To enable those skilled in the art to better understand the technical solutions of this application, some initialization and mechanisms in the embodiments of this application will be further explained below.
[0068] In one embodiment, the method further includes an offline system initialization and calibration step S000 executed on a cloud server before performing S100.
[0069] This step forms the foundation for the entire high-precision positioning system. Its purpose is to generate a high-fidelity digital twin model that can reproduce the hydraulic characteristics of the real physical pipe network, and to pre-calculate the core data structure for real-time inversion based on this model—the transfer admittance matrix (TAM).
[0070] S010: Construct an uncalibrated digital twin model.
[0071] The input to this process is an engineering file (e.g., EPANET .inp format) describing the physical layout of the pipeline network. This file includes topological information such as node coordinates and elevations, pipe start and end points, lengths, and diameters, as well as sensor deployment locations. Based on this file, the system constructs a preliminary computational model in the software, represented by a graph data structure. At this stage, key hydraulic parameters in the model, especially the pipe roughness coefficient, are empirical values set based on material type, which deviate from actual conditions; therefore, this model is uncalibrated.
[0072] S015: Use a nonlinear optimization algorithm to calibrate the digital twin model.
[0073] To ensure that the simulation behavior of the digital twin model matches the actual response in the physical world, this step employs a global nonlinear optimization algorithm, such as a genetic algorithm, to automatically optimize the model parameters. The algorithm's input consists of the uncalibrated model and a set of pressure benchmark datasets manually measured at multiple non-sensor nodes using high-precision portable pressure gauges during the initial system deployment. In the implementation of the genetic algorithm, the roughness coefficients of all pipe segments... Encoded as a real-number vector, representing an individual's chromosome. The algorithm iteratively performs selection, crossover, and mutation operations, using the root mean square error (RMSE) between the simulated and measured stresses as the fitness function to evolve a chromosome population. After a predetermined number of iterations, the set of roughness coefficients represented by the individual with the highest fitness in the population is determined as the optimal solution and used to update the digital twin model, thereby generating a calibrated, high-fidelity digital twin base.
[0074] In a specific, reproducible embodiment, the genetic algorithm is implemented by the following steps: Encoding and Population Initialization: Assuming the pipeline network includes The pipe section to be calibrated will then... Roughness coefficient of each pipe section Encode into a length of A real-valued vector, which constitutes an individual (chromosome). The system generates this vector randomly. Such individuals (e.g., ), forming the initial population. Each coefficient The initial value is within its physically reasonable range (for example, for PVC pipes, the range can be set to...). (Randomly generated.)
[0075] Fitness function definition: The fitness function is explicitly defined as the reciprocal of the root mean square error (RMSE) between the model-simulated pressure and the measured pressure benchmark dataset, i.e.: in: This represents the total number of pressure measurement points in the baseline dataset. A higher fitness value indicates a better individual (i.e., a set of roughness coefficients).
[0076] Genetic operator operations: The algorithm iteratively performs the following operations: Selection: A roulette wheel selection algorithm is used, where the probability of each individual being selected is proportional to its fitness value, thus selecting... Individuals enter the next generation.
[0077] Crossover: Randomly pair individuals from the selected population with a preset crossover probability (e.g., ...). Perform a single-point crossover operation. That is, cut at a random position on the two parent chromosomes and exchange the latter half to generate two new daughter chromosomes.
[0078] Mutation: For each gene (roughness coefficient) of every individual in the population, a mutation is performed with a small probability (e.g., Perform Gaussian mutation on the original gene values. Add a parameter that has a mean of 0 and a standard deviation of 0. (For example Random numbers from a normal distribution are used to introduce new genetic diversity.
[0079] Termination conditions: When the number of iterations of the algorithm reaches the preset maximum number of generations (e.g., ), or the improvement in the optimal fitness value of the population over 20 consecutive generations is less than a preset small threshold (e.g., When the condition is met, the algorithm terminates. At this point, the roughness coefficient values represented by the individual with the highest fitness in the population are considered the optimal solution.
[0080] S020: Construct and store the transitive admittance matrix (TAM).
[0081] Based on the high-fidelity model, this step constructs the TAM matrix through a systematic virtual simulation experiment. This applies to all pipelines... The system sequentially processes each of the discretized potential fault nodes. Apply a virtual pressure perturbation of one unit size (e.g., (A negative pressure source is used to simulate leakage). For each applied disturbance, the system runs a steady-state hydraulic simulation and records the single disturbance across all [various parameters / conditions]. Sensor locations The resulting pressure response value This response value, after normalization, is assigned to the first element of the TAM matrix. line, number The elements of the column, namely: Traverse all By identifying potential faulty nodes and repeating this process, a complete set of faulty nodes can be constructed. A dimensional TAM matrix. Each column of this matrix physically represents the unique pressure fingerprint generated on all sensors when a unit fault occurs at a specific location in the pipeline network. After calculation, this matrix is serialized and sent to the non-volatile memory of the edge computing gateway for use in the real-time inversion process.
[0082] Reference Figure 2 This is a schematic diagram of the functional module structure of an intelligent irrigation system provided in an embodiment of this application. The system may include: Data Acquisition and Preprocessing Module M100: Configured as the interface between the system and the physical world, its hardware entity is an industrial-grade edge computing gateway. This module is configured to execute all sub-steps in S100, including polling sensors, performing signal sampling, quantization, digital filtering, and validity verification, and publishing the processed data to the internal message bus.
[0083] The hydraulic digital twin and baseline generation module M200 is configured as the system's theoretical reference frame and adopts a cloud-edge collaborative architecture. Its cloud submodule (M200-Cloud) is responsible for executing computationally intensive offline tasks such as S015 and S020. Its edge submodule (M200-Edge) loads the cloud-calibrated model parameters and is responsible for executing the lightweight real-time baseline generation task of S210.
[0084] The equivalent pressure source inversion and localization module M300 is configured as the core computing engine of the system, running on the edge gateway. It is responsible for performing the inverse inference calculations for S220 and the entire S300. This module is optimized for efficient matrix operations to calculate the core result—the equivalent pressure source vector. .
[0085] Fault Diagnosis and Control Decision Module M400: Configured as the system's decision-making brain, responsible for executing S400. This module interprets the calculation results of M300, confirms faults through threshold comparisons, and generates structured alarm information or sends control commands to field actuators (such as PLCs) via industrial protocols.
[0086] Reference Figure 3 This diagram illustrates the cross-domain principle mapping between the equivalent pressure source model in this embodiment and the equivalent current dipole (ECD) model in neuroscience. The core idea of this invention is to formally map the mathematical framework of the ECD model, used in the source domain (neuroscience) to invert neural activity sources within the brain from scalp electroencephalography (EEG) signals, to the target domain (hydraulic network). Tiny leaks in the network are analogized to neural activity sources; the network, as a flow-conducting medium, is analogized to the brain's conductive medium; distributed pressure sensors are analogized to scalp electrodes; and the hydraulic balance equation describing pressure propagation is mathematically similar to the Poisson equation describing potential field propagation. Therefore, the Lead Field Matrix connecting the source and measurement in neuroscience is mapped to a Transfer Admittance Matrix (TAM) in this invention, thereby applying an efficient linear inversion mathematical tool to solve the problem of fault location and control in irrigation networks.
[0087] This application also provides a computing device, including a processor and a memory, wherein a computer program is stored in the memory, and when the processor executes the computer program, it implements the method described in any of the above embodiments. The computing device may be an edge computing gateway or an industrial PC deployed in the field, or a server in the cloud.
[0088] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the methods described in any of the above embodiments. The computer-readable storage medium may be ROM, RAM, flash memory, hard disk, etc.
[0089] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.
[0090] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0091] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.
Claims
1. A smart irrigation control method based on the Internet of Things, characterized in that, include: The real-time pressure vectors of multiple preset pressure sensor nodes within the irrigation network are obtained, as well as the real-time source status of the source nodes of the irrigation network are obtained. Based on the real-time source status and the pre-stored digital twin model of the irrigation network, a theoretical pressure baseline vector aligned with the real-time pressure vector timestamp is generated. Based on the real-time pressure vector and the theoretical pressure baseline vector, the pressure deviation vector is determined; Based on the pressure deviation vector and the pre-stored transfer admittance matrix, a regularized linear inversion algorithm is used to determine the equivalent pressure source vector characterizing the fault location and intensity. Based on the equivalent pressure source vector, the fault in the irrigation network is located and corresponding control is executed.
2. The method according to claim 1, characterized in that, The step of obtaining the real-time source status of the source nodes of the irrigation network includes: The instantaneous flow and instantaneous pressure of the source node are collected synchronously to generate the real-time source status.
3. The method according to claim 1, characterized in that, After obtaining the real-time pressure vectors of multiple preset pressure sensor nodes within the irrigation network, the method further includes: The real-time pressure vector is validated for data validity, and a valid sensor mask is generated. The determination of the pressure deviation vector specifically involves: The pressure deviation vector is generated by performing the difference calculation between the real-time pressure vector and the theoretical pressure baseline vector only for the valid sensors identified by the valid sensor mask.
4. The method according to claim 3, characterized in that, Before determining the equivalent pressure source vector, the method further includes: Determine whether the number of valid sensors identified by the valid sensor mask is less than a preset minimum sensor number threshold; If the judgment result is yes, then the execution of the regularized linear inversion algorithm is stopped, and a global sensor fault alarm is generated.
5. The method according to claim 1, characterized in that, The theoretical pressure baseline vector is generated based on the real-time source status and the pre-stored digital twin model of the irrigation network, specifically as follows: The water head boundary conditions of the water source node in the digital twin model are dynamically updated using the real-time source pressure in the real-time source status. In the dynamically updated digital twin model, a steady-state hydraulic simulation is performed to generate the theoretical pressure baseline vector.
6. The method according to claim 1, characterized in that, The method of using a regularized linear inversion algorithm to determine the equivalent pressure source vector is as follows: Construct a Tikhonov regularization objective function that includes the transfer admittance matrix, the pressure deviation vector, and regularization parameters; Solve for the closed-form solution of the Tikhonov regularized objective function to obtain the equivalent pressure source vector.
7. The method according to claim 1, characterized in that, The method of locating the fault in the irrigation network based on the equivalent pressure source vector specifically includes: Find the element with the largest absolute value in the equivalent pressure source vector, determine the node index corresponding to the element as the fault candidate location, and determine the value of the element as the fault intensity. The absolute value of the fault intensity is compared with a preset diagnostic threshold. If the absolute value is greater than the diagnostic threshold, a fault is confirmed to have occurred.
8. The method according to claim 1, characterized in that, Before acquiring the real-time pressure vector, an initialization step is also included: Based on the topology of the irrigation network, an uncalibrated digital twin model is constructed. A nonlinear optimization algorithm is used, with the hydraulic parameters of the pipe section in the uncalibrated digital twin model as the optimization variable, and minimizing the error between the simulated pressure and the measured pressure as the objective function, to perform iterative optimization in order to generate the calibrated digital twin model. Based on the calibrated digital twin model, the transfer admittance matrix is constructed and stored by sequentially applying a unit virtual pressure disturbance to each possible fault node and performing steady-state simulation.
9. The method according to claim 1, characterized in that, The step of locating faults in the irrigation network and executing corresponding controls based on the equivalent pressure source vector includes: Based on the location of the fault, query the pre-stored pipeline topology database for the nearest controllable valve node located upstream of the fault location; A shutdown command is generated for the controllable valve node and sent to the corresponding field controller via an industrial control protocol to isolate the fault location.
10. An intelligent irrigation system based on the Internet of Things, characterized in that, include: The data acquisition and preprocessing module is configured to acquire the real-time pressure vectors of multiple preset pressure sensor nodes within the irrigation network, and to acquire the real-time source status of the source nodes of the irrigation network. The hydraulic digital twin and baseline generation module is configured to generate a theoretical pressure baseline vector aligned with the real-time pressure vector timestamp based on the real-time source state and the pre-stored digital twin model of the irrigation network. The equivalent pressure source inversion and localization module is configured to determine the pressure deviation vector based on the real-time pressure vector and the theoretical pressure baseline vector, and to determine the equivalent pressure source vector characterizing the fault location and intensity based on the pressure deviation vector and the pre-stored transfer admittance matrix using a regularized linear inversion algorithm. The fault diagnosis and control decision module is configured to locate faults in the irrigation network and execute corresponding controls based on the equivalent pressure source vector.