Method for intelligent monitoring and hydraulic balance implementation of capillary network system

Abstract

The application discloses an intelligent monitoring and hydraulic balance implementation method of a capillary tube network system, and belongs to the field of intelligent regulation and control of building heating, ventilation and air conditioning. The method solves the problems of low efficiency of traditional manual debugging and inability to maintain hydraulic balance for a long time. The scheme is as follows: first, intelligent monitoring nodes are arranged at key node positions to collect data, the data are uploaded to a central control system through an edge computing gateway, a built-in hydraulic network calculation model is used to solve system states and output hydraulic imbalance degrees, when the hydraulic imbalance degrees exceed the limit, the system is inversely calculated and encrypted target opening degree instructions are issued to each loop digital twin driven electric balance valve, a closed loop regulation is performed by an embedded controller of the valve, and finally, after installation is completed, a full-automatic initialization debugging is performed, initial opening degree compensation values of the digital twin driven electric balance valves are calculated and written into valve embedded controllers, and presetting is completed. The application is used for realizing automatic construction and continuous and accurate regulation and control of hydraulic balance of the capillary tube network.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control of building heating, ventilation and air conditioning, and specifically relates to a method for intelligent monitoring and hydraulic balance of capillary network system. Background Technology

[0002] In the field of building heating, ventilation, and air conditioning (HVAC), capillary radiant systems are gradually being adopted due to their high comfort and energy-saving potential. However, the system still faces the long-standing technical problem of difficult hydraulic balance adjustment in practical engineering.

[0003] Capillary networks typically consist of multiple parallel loops, characterized by strong hydraulic coupling and significant differences in resistance characteristics between these loops. Traditional hydraulic balancing methods primarily rely on manual on-site commissioning, involving manually adjusting the opening of static balancing valves on each loop to achieve the designed flow distribution. This method has significant limitations: firstly, the commissioning process heavily depends on the experience of technicians and requires repeated measurements and adjustments under system operation, resulting in a lengthy process; secondly, the commissioning result is a static balance achieved under specific operating conditions. When system operating conditions change (e.g., partial load operation, user adjustments) or the physical characteristics of the network change over time (e.g., pipe scaling, component aging), the original balance is easily disrupted, leading to hydraulic imbalances. This manifests as uneven flow distribution between loops, subsequently causing problems such as uneven indoor heating and cooling, and increased system energy consumption.

[0004] To address the aforementioned dynamic misalignment issues, the industry has attempted to introduce electrically controlled regulating valves based on simple feedback control. However, due to the multivariable, strongly coupled, and nonlinear characteristics of capillary network systems, relying solely on single-point feedback control of local loop flow or temperature makes it difficult to achieve coordinated optimization of the overall hydraulic state at the system level. Furthermore, the lack of sufficient coverage for real-time monitoring of key hydraulic and thermal parameters at critical nodes throughout the network prevents the central control system from accurately and promptly acquiring the system's global operating status, thus hindering the application of advanced control strategies. Attempts have also been made to establish theoretical hydraulic models for simulation and guidance, but unavoidable deviations exist between simulation model parameters (such as the actual pipe resistance coefficient) and the actual engineering model, and the model parameters cannot adaptively update with changes in actual system operation, resulting in limited long-term guiding value.

[0005] In summary, the establishment and maintenance of hydraulic balance in capillary network systems in existing technologies mainly face core problems such as over-reliance on manual static commissioning, lack of real-time global system status awareness, and difficulty in long-term dynamic matching between theoretical models and actual systems. These problems affect the stable achievement of energy efficiency and thermal comfort in capillary network systems, and also increase the difficulty and cost of subsequent operation and maintenance. Therefore, how to achieve accurate, automatic, and continuous maintenance of capillary network hydraulic balance is a pressing technical challenge that needs to be addressed in this field. Summary of the Invention

[0006] This invention provides an intelligent monitoring and hydraulic balancing method for capillary network systems, effectively overcoming the technical shortcomings of traditional manual commissioning, such as low efficiency and inability to maintain hydraulic balance sustainably. This method deploys an intelligent monitoring network to perceive the system status in real time, constructs and continuously learns a hydraulic network calculation model for global analysis, combines intelligent valves with digital twin and fault-tolerant functions to perform precise adjustments, and relies on fully automated initialization commissioning to complete the initial matching between the physical system and the digital model. Thus, the hydraulic balancing process is transformed from a static, one-time commissioning relying on experience into a dynamic, continuous process driven by data, with centralized decision-making and automatic execution. This ensures that the system maintains a high-precision hydraulic balance under various operating conditions and during long-term aging, providing intelligent support for the efficient, stable, and comfortable operation of building HVAC systems.

[0007] To achieve these objectives and other advantages of the present invention, a method for intelligent monitoring and hydraulic balancing of a capillary network system is provided, comprising the following steps:

[0008] S1. Based on the hydraulic topology of capillary networks, intelligent monitoring nodes are deployed at preset key node locations to collect time-series data of hydraulic and thermal parameters.

[0009] S2. The time series data is uploaded to the central control system via the edge computing gateway. The capillary network state is solved by the built-in hydraulic network calculation model, and the calculation result set containing the system hydraulic misalignment is output.

[0010] S3. The central control system compares the system hydraulic misalignment in the calculation result set with a preset threshold. When the system hydraulic misalignment exceeds the preset threshold, it generates a target opening instruction set for each loop digital twin-driven electric balancing valve by back-calculating the target opening instruction against the preset valve flow characteristic curve based on the theoretical flow rate of each pipe section in the calculation result set. The instruction is encrypted and sent to the valve embedded controller. The valve embedded controller controls its drive mechanism to read the real-time opening feedback value of the valve absolute encoder and executes the proportional-integral adjustment algorithm with the target opening instruction as the set value until the valve opening of the digital twin-driven electric balancing valve stabilizes within the target range, completing one closed-loop adjustment.

[0011] S4. After the physical installation of the system is completed, a fully automatic initialization and debugging is performed. The central control system sends scanning commands from fully closed to fully open to all digital twin driven electric balancing valves and simultaneously records the pressure and flow data fed back by each intelligent monitoring node. By fitting the correspondence between pressure and flow data, an actual pipeline characteristic curve is generated. The actual pipeline characteristic curve is compared with the pre-stored design model characteristic curve corresponding to the hydraulic network calculation model. The initial opening compensation value of each digital twin driven electric balancing valve is calculated and written into the valve embedded controller to complete the pre-setting.

[0012] Preferably, the preset key node locations in step S1 specifically include the end of the supply manifold and the end of the return manifold of all independent hydraulic loops, as well as the capillary inlet of the key monitoring loop selected based on the weighting factor determined by the hydraulic distance and design heat load of each loop.

[0013] The intelligent monitoring node integrates a pressure sensor, a flow meter, and a temperature sensor.

[0014] Preferably, in step S2, the time-series data is transmitted to an edge computing gateway via a low-power wide-area network. The edge computing gateway cleans, aligns, and compresses the data, packages it into standardized data packets, and uploads them to the central control system. The edge computing gateway incorporates a lightweight hydraulic balance rapid diagnostic algorithm. This algorithm analyzes the spatial gradient distribution of pressure data from each intelligent monitoring node at the same time, identifies and locates suspected blockages or gas-filled abnormal branches in the capillary network in real time, generates corresponding abnormal status markers, and uploads these abnormal status markers along with the standardized data packets to the central control system.

[0015] Preferably, the hydraulic network computation model is an adaptive graph network model that integrates physical mechanisms and data-driven approaches, specifically including:

[0016] S201. Model construction steps: The physical topology of the capillary network is abstracted into a hydraulic calculation diagram, wherein the nodes of the hydraulic calculation diagram correspond to the connection points of each pipe segment and the location of the intelligent monitoring nodes, and the edges of the hydraulic calculation diagram correspond to the capillary mats, branches and manifolds, and each edge is assigned an initial weight coefficient, which represents the comprehensive hydraulic resistance of the corresponding pipe segment under the reference working condition.

[0017] S202. Online solution steps: The pressure and flow data of each node collected in real time are used as the boundary conditions of the hydraulic calculation diagram. By solving the node flow balance equation set and loop pressure loss balance equation set established based on the laws of mass conservation and energy conservation, the actual dynamic weight coefficients of each edge in the hydraulic calculation diagram are derived, and the theoretical flow distribution and hydraulic misalignment of the whole network are calculated accordingly.

[0018] S203, Incremental Learning Step: The hydraulic network calculation model integrates an incremental learning module. After each closed-loop adjustment in step S3, the stabilized system state data is used as a new data sample. The weight coefficients of the corresponding edges in the hydraulic calculation graph are dynamically updated and calibrated using the recursive least squares method or stochastic gradient descent algorithm. This allows the hydraulic network calculation model parameters to continuously track the migration of physical characteristics of the pipeline network caused by scaling, deformation, or changes in valve characteristics.

[0019] Preferably, the valve embedded controller of the digital twin-driven electric balancing valve has a pre-stored data table of valve opening degree-pressure difference-flow relationship obtained through experimental calibration; in step S3, the control process of the valve embedded controller is specifically as follows:

[0020] S301. Real-time acquisition step: Through the differential pressure sensor connected to it, the pressure before and after the valve of the digital twin driven electric balance valve is measured and acquired in real time, and the real-time differential pressure value is calculated.

[0021] S302, Opening command compensation step: Taking the target opening command received from the central control system and the real-time differential pressure value as input, query the valve opening-differential pressure-flow relationship data table, and calculate the compensated opening value required to achieve the target flow under the current differential pressure using a three-dimensional interpolation algorithm;

[0022] S303, Closed-loop control steps: Using the compensated opening value as the set value, read the actual opening value fed back by the valve absolute encoder, execute the proportional-integral-derivative control algorithm, and output the control signal to the drive mechanism of the valve embedded controller to drive the valve core to move to the compensated opening.

[0023] Preferably, a pre-deployment simulation and optimized placement step is performed before step S1, specifically including:

[0024] S001. In Building Information Modeling (BIM), based on the design drawings of capillary networks, a three-dimensional capillary network model containing pipe diameter, length, elevation, and connection relationships is generated using parametric modeling components.

[0025] S002. Import the three-dimensional capillary network model into computational fluid dynamics software, assign material properties to the model, and apply the total flow inlet boundary condition corresponding to the design conditions and the design heat load of each room calculated based on the building thermal performance as the wall heat flux density boundary condition to the boundary of the three-dimensional capillary network model.

[0026] S003. In the computational fluid dynamics software, the k-ε turbulence model and the coupled heat transfer model are selected to perform steady-state and non-isothermal flow simulation calculations on the three-dimensional capillary network model, and the pressure field and flow field distribution data of the three-dimensional capillary network under the whole system operation state are obtained by solving the solution.

[0027] S004. Extract the feature parameter vectors of all hydraulic loops from the simulation results. The feature parameter vectors include at least the total pressure drop of the loop, the average flow rate of the loop, and the design heat load of the room where the loop is located. Use the K-means clustering algorithm to perform cluster analysis on the feature parameter vectors of all hydraulic loops, and select the loop closest to the cluster center from each cluster and mark it as the key monitoring loop.

[0028] Preferably, the pre-deployment simulation and optimized placement step further includes:

[0029] S005. Based on the pressure and flow field distribution data obtained in step S003, extract the pressure and flow reference values ​​of each key monitoring loop under rated operating conditions. Combine the design specifications and historical operating data to set dynamic early warning thresholds and alarm thresholds for pressure and flow for each key monitoring node. Use the initial pressure and flow distribution data of the entire network obtained from the simulation in step S003 as the initial training dataset for the hydraulic network calculation model to complete the pre-training and initialization of the hydraulic network calculation model parameters.

[0030] Preferably, the embedded controller of the digital twin-driven electric balancing valve also integrates a fault self-diagnosis and fault-tolerant control module, which specifically comprises:

[0031] The valve absolutely encoder periodically compares the actual opening value with the drive mechanism commanded opening value. If the deviation continues to exceed the preset tolerance range, it is determined that the valve core is stuck or the transmission mechanism is faulty, and a fault code is generated and uploaded to the central control system.

[0032] When abnormal differential pressure sensor data or communication interruption is detected, the system automatically switches to the opening-flow empirical curve feedforward control mode, maintaining the valve's approximate regulation function based on a simplified relationship fitted from historical normal operation data until the fault is cleared.

[0033] Preferably, the lightweight hydraulic balance fast diagnostic algorithm built into the edge computing gateway further includes:

[0034] Construct a pressure gradient matrix with intelligent monitoring nodes as spatial nodes, and calculate and monitor the gradient vector in the pressure gradient matrix that represents the rate of change of pressure difference between adjacent nodes in real time.

[0035] When the gradient vector magnitude of a certain branch continuously exceeds the dynamic threshold calculated based on the pipeline topology, it is determined that there is a blockage or gas accumulation anomaly in that branch, and the location and severity level of the anomaly are estimated based on the direction and magnitude of the gradient vector.

[0036] A diagnostic report containing the abnormal branch identifier, abnormal type, severity, and suggested troubleshooting measures is generated and uploaded to the central control system along with the abnormal status marker.

[0037] Preferably, during the initialization and debugging process in step S4, a dedicated portable calibrator is used to conduct near-field wireless communication with the on-site intelligent monitoring nodes. The dedicated portable calibrator has a built-in high-precision standard sensor, which can be temporarily connected to the pipeline network measurement point to perform on-site reading comparison and calibration of the fixed-installed intelligent monitoring nodes, and generate calibration coefficients, which are automatically written to the corresponding nodes after being confirmed by the debugging personnel.

[0038] The present invention has at least the following beneficial effects:

[0039] First, this invention transforms the traditional manual, static, and experience-based balancing process into a dynamic process driven by time-series data, decided by the central control system, and automatically executed, by deploying intelligent monitoring nodes, constructing a hydraulic network calculation model, issuing closed-loop control commands, and executing fully automatic initialization and debugging. This not only significantly reduces the manpower costs and time for debugging and subsequent maintenance, but more importantly, it enables the system to continuously operate in a high-precision hydraulic balance state, responding promptly to internal changes and external disturbances, thereby steadily improving system energy efficiency and end-point thermal comfort in the long term. The edge computing gateway cleans, aligns, and compresses the raw time-series data, reducing the communication and computing load on the central control system. Its built-in lightweight hydraulic balance rapid diagnostic algorithm analyzes the spatial distribution anomalies of pressure gradients, enabling it to identify and locate local blockages or air accumulation faults before data upload, greatly shortening the time from fault occurrence to discovery. This facilitates rapid response by maintenance personnel, prevents local problems from escalating into systemic failures, and improves the overall system's operational safety and maintainability.

[0040] Secondly, this invention, by explicitly deploying intelligent monitoring nodes at the ends of the supply and return manifolds of all independent hydraulic loops, as well as at the capillary inlets of key monitoring loops selected based on hydraulic distance and design heat load weighting factors, can capture the most representative hydraulic state information from both the system backbone and key endpoints. This scientific deployment strategy based on hydraulic topology and heat load weighting avoids monitoring blind spots and provides a comprehensive and accurate input data foundation for the hydraulic network calculation model and decision-making of the central control system, which is a prerequisite for achieving accurate full-network hydraulic balance.

[0041] Third, the hydraulic network calculation model of this invention adopts an adaptive graph network model that integrates physical mechanisms and data-driven approaches. Through an incremental learning module, it continuously absorbs the real-state data after the system's closed-loop regulation has stabilized, and dynamically updates its internal weight coefficients (i.e., the comprehensive hydraulic resistance of the pipe section) using recursive least squares or stochastic gradient descent algorithms. This allows it to continuously approximate and reflect the migration of physical characteristics of the pipe network caused by scaling, deformation, or changes in valve characteristics. It effectively overcomes the shortcomings of traditional fixed-parameter models that become inaccurate over time, ensuring that the control commands generated based on this model maintain high accuracy over the long term. This is a key technological support for achieving sustainable intelligent hydraulic balance.

[0042] Fourth, in this invention, the embedded controller of the digital twin-driven electric balancing valve pre-stores a data table of valve opening-pressure difference-flow relationship obtained through experimental calibration. During control, the received target opening command and the real-time measured pressure difference are used as inputs. A three-dimensional interpolation algorithm is used to calculate the compensated opening value required to achieve the target flow rate under the current pressure difference. Then, combined with the actual opening value fed back by the valve absolute encoder, proportional-integral-derivative (PID) closed-loop control is executed. This ensures that the valve control command fully considers the current actual hydraulic conditions, thereby guaranteeing the high stability and accuracy of the valve output flow even under system pressure fluctuations, providing terminal execution guarantee for achieving refined network-wide flow allocation. In addition, by performing pre-deployment simulation and optimized layout steps before construction, namely, performing steady-state non-isothermal flow simulation based on Building Information Modeling (BIM) and Computational Fluid Dynamics (CFD) software, and using the K-means clustering algorithm to analyze the hydraulic loop characteristic parameter vector obtained from the simulation, the most representative key monitoring loop is automatically identified. This method ensures that the most comprehensive system operating characteristics can be covered with minimal monitoring costs before physical construction, making the design of the monitoring network more economical, efficient, and purposeful, thus optimizing the layout of the entire intelligent monitoring system from the source. Simultaneously, simulation data is used to set dynamic early warning and alarm thresholds for key nodes, and the simulation results are used as the initial training dataset for the hydraulic network calculation model, completing the model's pre-training and initialization. This avoids the lengthy and potentially unstable trial-and-error process of the control system learning from "zero knowledge," significantly shortening the system commissioning cycle and improving control stability and reliability in the initial stages of operation.

[0043] Fifth, the digital twin-driven electric balancing valve of this invention integrates a fault self-diagnosis and fault-tolerant control module. By periodically comparing the actual opening value fed back by the valve's absolute encoder with the opening value commanded by the drive mechanism, it can promptly detect valve core jamming or transmission mechanism failure and generate fault codes for reporting. When abnormal differential pressure sensor data or communication interruption is detected, it can automatically switch to a feedforward control mode based on the opening-flow empirical curve fitted from historical normal operating data, maintaining the valve's basic approximate regulation function. This design reduces the dependence on the absolute reliability of individual components, ensuring that the system as a whole can still maintain basic operation in the event of a partial failure, buying time for fault repair, and enhancing the system's availability and resilience.

[0044] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the intelligent monitoring and hydraulic balance implementation method of the capillary network system of the present invention. Detailed Implementation

[0046] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0047] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0048] like Figure 1 As shown, this invention provides an intelligent monitoring and hydraulic balance implementation method for a capillary network system, comprising the following steps:

[0049] S1. Based on the hydraulic topology of capillary networks, intelligent monitoring nodes are deployed at preset key node locations to collect time-series data of hydraulic and thermal parameters.

[0050] S2. The time series data is uploaded to the central control system via the edge computing gateway. The capillary network state is solved by the built-in hydraulic network calculation model, and the calculation result set containing the system hydraulic misalignment is output.

[0051] S3. The central control system compares the system hydraulic misalignment in the calculation result set with a preset threshold. When the system hydraulic misalignment exceeds the preset threshold, it generates a target opening instruction set for each loop digital twin-driven electric balancing valve by back-calculating the target opening instruction against the preset valve flow characteristic curve based on the theoretical flow rate of each pipe section in the calculation result set. The instruction is encrypted and sent to the valve embedded controller. The valve embedded controller controls its drive mechanism to read the real-time opening feedback value of the valve absolute encoder and executes the proportional-integral adjustment algorithm with the target opening instruction as the set value until the valve opening of the digital twin-driven electric balancing valve stabilizes within the target range, completing one closed-loop adjustment.

[0052] S4. After the physical installation of the system is completed, a fully automatic initialization and debugging is performed. The central control system sends scanning commands from fully closed to fully open to all digital twin driven electric balancing valves and simultaneously records the pressure and flow data fed back by each intelligent monitoring node. By fitting the correspondence between pressure and flow data, an actual pipeline characteristic curve is generated. The actual pipeline characteristic curve is compared with the pre-stored design model characteristic curve corresponding to the hydraulic network calculation model. The initial opening compensation value of each digital twin driven electric balancing valve is calculated and written into the valve embedded controller to complete the pre-setting.

[0053] Specifically, the preset key node locations in step S1 include the end of the supply manifold and the end of the return manifold of all independent hydraulic loops, as well as the capillary inlet of the key monitoring loop selected based on the weighting factor determined by the hydraulic distance and design heat load of each loop; the intelligent monitoring node integrates a pressure sensor, a flow meter and a temperature sensor.

[0054] In the above embodiments, the intelligent monitoring and hydraulic balance implementation method for capillary network systems aims to construct a closed-loop control system from precise sensing and intelligent analysis to automatic execution, thereby overcoming the drawbacks of traditional methods that rely on manual and static adjustments. This method achieves continuous and automatic maintenance of hydraulic balance by deeply integrating the physical network with a digital model.

[0055] First, based on the capillary network design drawings and the hydraulic topology formed during actual construction—that is, the spatial connections and routing relationships between the supply manifolds, return manifolds, branch pipes, and capillary mats in the system—intelligent monitoring nodes are strategically deployed during the construction phase. The preset locations of key monitoring nodes are scientifically considered, specifically including: the ends of the supply and return manifolds of each independent hydraulic loop to monitor the hydraulic state of the system's main trunk; and a comprehensive weighting factor calculated based on the hydraulic distance of each loop (the length of fluid flowing from the supply manifold to the end of the loop, calculated by combining local resistance) and the design heat load of the room it serves. Representative loops with higher weighting factors are selected, and monitoring points are placed at their capillary mat inlets. This deployment strategy ensures that the monitoring network simultaneously covers both the system's "main trunk" and key "terminals," thereby obtaining the most representative global and local hydraulic information. The deployed intelligent monitoring node is an integrated measurement terminal, which typically integrates a pressure sensor for measuring water pressure, a flow meter for measuring water flow, and a temperature sensor for measuring the temperature of the supply or return water, in order to collect time-series data of the hydraulic and thermal parameters at that point, providing a high spatiotemporal resolution raw data foundation for subsequent analysis.

[0056] Secondly, the time-series data collected by each intelligent monitoring node is aggregated and transmitted to the edge computing gateway via a field communication network (e.g., RS-485 bus, M-Bus, or low-power wide-area network wireless technologies such as LoRa and NB-IoT). The edge computing gateway, acting as a field-level data processing unit, cleans (removes outliers), aligns (unifies timestamps across nodes), and compresses (reduces data volume while ensuring no loss of critical information) the raw data, packaging it into standardized data packets. These packets are then uploaded to the central control system located in the monitoring center or cloud platform. The core of the central control system is a built-in hydraulic network calculation model, which abstracts the physical pipe network into a computational network and constructs a system of mathematical equations based on the principles of mass and energy conservation. The hydraulic network calculation model receives pressure and flow data of key nodes in the standardized data packet as known boundary conditions. By solving the system of equations, it calculates the theoretical flow distribution of all pipe segments in the network at the current moment and finally calculates a comprehensive evaluation index - system hydraulic imbalance. This index quantitatively reflects the degree to which the actual flow distribution deviates from the ideal design state. Its value is between 0 and 1, and the larger the value, the more serious the imbalance.

[0057] Next, the central control system compares the calculated system hydraulic imbalance with a preset threshold. This preset threshold serves as a "trigger line" for determining whether the system needs to initiate balancing adjustments. Its setting must balance adjustment sensitivity and system stability to avoid frequent malfunctions. As an exemplary, rather than limiting, implementation, this preset threshold can be selected within the range of 0.15 to 0.25, preferably set to 0.2. When the real-time imbalance exceeds this threshold, the system determines it to be unbalanced and initiates an automatic adjustment program. The adjustment is based on the theoretical flow rate of each pipe segment output by the model. The central control system, based on these target flow rate values ​​and combined with the valve flow characteristic curves (which fully describe the mapping relationship between valve opening, pressure difference across the valve, and flow rate) pre-stored in the database for each digital twin-driven electric balancing valve obtained through experimental calibration, performs reverse calculations to determine the target valve opening required to achieve each target flow rate and generates the corresponding target opening instruction set. The instructions are encrypted and sent to the valve embedded controller installed on the valve body. The controller drives its drive mechanism and simultaneously reads the valve spool opening from the valve absolute encoder (a sensor that provides absolute rotational position information of the valve shaft) in real time. This forms a proportional-integral (PI) control closed loop with the target opening as the set value, precisely controlling the valve spool to move to the commanded position, thereby changing the loop resistance and regulating the flow. The target range refers to the allowable deviation range between the actual valve opening and the target opening command. This deviation range can be preset according to the valve accuracy and system requirements, and is typically ±1% to ±5% of the commanded value.

[0058] Finally, before the entire system is physically installed and put into operation for the first time, a fully automatic initialization and commissioning is performed. The central control system sequentially sends scanning commands from fully closed to fully open to all digital twin-driven electric balancing valves. During this process, the valves open gradually according to preset steps, and the system synchronously records the pressure and flow data fed back by each intelligent monitoring node at each opening degree. By fitting these data, an actual pipeline characteristic curve representing the hydraulic characteristics of the actual installed pipeline network is generated. Subsequently, the system compares and analyzes this curve with the design model characteristic curve, which is pre-stored in the hydraulic network calculation model and obtained based on ideal design parameters. By calculating the difference between the two, the system can deduce the inherent errors caused by factors such as construction and installation deviations and differences between the actual pipeline roughness and the design value, and then calculate an initial opening compensation value for each valve. This compensation value is written into the embedded controller of the corresponding valve as the basis for offset correction when it executes adjustment commands later. This step essentially completes the initial calibration and matching of the "digital twin model" and the "physical entity," laying an accurate initial state for subsequent high-precision automatic balancing.

[0059] Compared to the traditional commissioning method that relies entirely on technicians manually measuring and adjusting static balancing valves based on experience, this method, through automated data acquisition, model-based state solving, and command-based closed-loop control, frees manpower from the tedious and time-consuming on-site commissioning work, significantly improving commissioning efficiency. More importantly, the traditional method achieves a "static balance," which is disrupted once operating conditions change or the pipeline ages. The dynamic system constructed in this invention continuously monitors the hydraulic state and automatically triggers rebalancing when the misalignment exceeds a threshold, thus achieving a fundamental shift from "one-time balance" to "continuous balance," enabling the system to maintain optimal hydraulic conditions for extended periods.

[0060] In one specific embodiment, in step S2, the time-series data is transmitted to an edge computing gateway via a low-power wide-area network. The edge computing gateway cleans, aligns, and compresses the data, packages it into a standardized data packet, and uploads it to the central control system. The edge computing gateway incorporates a lightweight hydraulic balance rapid diagnostic algorithm. This algorithm analyzes the spatial gradient distribution of pressure data from each intelligent monitoring node at the same time, identifies and locates suspected blockages or gas-filled abnormal branches in the capillary network in real time, generates corresponding abnormal status markers, and uploads these abnormal status markers along with the standardized data packets to the central control system.

[0061] In the above implementation, in step S2, the time-series data collected by the intelligent monitoring node is first transmitted via a low-power wide-area network (LPWAN). Here, LPWAN refers to a type of wireless network technology specifically designed for IoT devices, supporting long-distance communication and extremely low terminal power consumption, such as LoRa long-range radio or NB-IoT narrowband IoT. This technology is chosen primarily to address the practical problems of capillary network monitoring nodes being typically distributed and battery-powered, thereby achieving stable wireless monitoring for months to years. The data is then sent to an edge computing gateway, a dedicated embedded computing device deployed on the construction site. The core function of the edge computing gateway is to perform real-time preprocessing of the received raw time-series data. This includes cleaning and filtering out obvious erroneous data points caused by transient interference, alignment (synchronizing all data according to a unified time base to ensure time consistency within the same computing cycle), and compression (reducing data volume using algorithms while preserving key characteristics of data change trends). After these processes, the originally scattered raw data is integrated into well-organized and reliable standardized data packets, laying the foundation for subsequent analysis and transmission. This process is typically completed before data is uploaded to the cloud or central server, and its core value lies in shifting a large amount of local computing load from the central system to the edge. As a typical implementation, the interval at which monitoring nodes send data to the edge computing gateway (i.e., the data reporting interval) can be adjusted between 30 seconds and 5 minutes according to the dynamic characteristics of the system, for example, preferably once every 60 seconds, to balance communication real-time performance and node power consumption.

[0062] This paper presents a lightweight hydraulic balance rapid diagnostic algorithm embedded within an edge computing gateway. The algorithm operates based on the physical principles of hydraulics. In a stable, closed-loop pipe network, the pressure distribution between adjacent nodes should exhibit a continuous and gentle spatial gradient. If a branch experiences localized blockage or gas accumulation, it creates abnormal local resistance, causing abrupt changes in the pressure difference between upstream and downstream monitoring nodes, resulting in a significant steep drop or abnormal gradient in the spatial distribution of the pressure field. The algorithm achieves real-time diagnosis by analyzing the spatial gradient distribution of pressure data from each intelligent monitoring node at the same time. Specifically, the hydraulic balance rapid diagnostic algorithm incorporates the topological connections of the capillary network. It periodically acquires the pressure values ​​of all nodes in the same time slice and calculates the pressure difference between every two adjacent topological nodes. By monitoring the instantaneous state and short-term trends of these pressure differences, the algorithm can quickly identify where abnormal pressure gradients that deviate from normal hydraulic patterns occur. Once a suspected anomaly is identified, a corresponding anomaly status flag is immediately generated. This flag includes at least the logical number of the anomalous branch and a preliminary assessment of the anomaly type, such as suspected blockage or suspected gas storage. This flag, along with a processed standardized data packet, is uploaded to the central control system. In this way, the central system receives the data and simultaneously obtains a preliminary diagnostic conclusion based on real-time data from the edge.

[0063] To achieve accurate identification, this lightweight hydraulic balance rapid diagnostic algorithm requires setting reasonable judgment thresholds during operation. The algorithm continuously calculates and tracks the pressure gradient between the nodes at both ends of each branch, i.e., the pressure change per unit pipe length or logical distance. Based on historical data from stable system operation, the algorithm establishes a normal pressure gradient baseline range for each branch. An anomaly judgment is triggered when the real-time calculated pressure gradient value consistently exceeds this baseline range by a certain margin. Here, "consistently" requires defining a time window to avoid false alarms due to transient interference. The algorithm runs independently at fixed diagnostic cycles (e.g., every 10 or 30 seconds), calculating and evaluating the pressure gradient. An anomaly judgment is triggered when an abnormal gradient appears consecutively for more than three consecutive diagnostic cycles. As an example of threshold selection, the anomaly judgment criterion can be set to a real-time pressure gradient value exceeding its historical average by 50% to 150%, preferably exceeding 80%. Furthermore, to distinguish between blockage and gas accumulation, the hydraulic balance rapid diagnostic algorithm can also combine the direction of gradient change and flow readings from adjacent nodes for auxiliary judgment. The entire diagnostic process is completed locally on the edge gateway, with carefully designed computational complexity to ensure real-time operation on the gateway's limited embedded processor resources. The generated diagnostic report, along with the anomaly location and preliminary type, is sent along with the data packet. This significantly reduces the time delay from the occurrence of a physical anomaly to the central control system receiving an alert.

[0064] By introducing an edge computing gateway and its built-in lightweight fast diagnostic algorithm, real-time analysis near the data source enables near-instantaneous capture of local abrupt changes in the hydraulic characteristics of the pipeline network, significantly shortening the response time for anomaly detection and achieving early warning of problems such as blockage and gas accumulation. Secondly, this solution effectively distributes and optimizes the overall system's computational and communication load. The raw sensor data is massive. By cleaning, aligning, and compressing it at the edge, the amount of data that needs to be uploaded is significantly reduced, saving network bandwidth. Simultaneously, by offloading tasks like rapid diagnostics—which have high real-time requirements but relatively fixed computational models—to the edge, the central control system is freed up to handle more complex global optimization calculations, making the entire system architecture more efficient and rational.

[0065] In one specific embodiment, the hydraulic network computation model is an adaptive graph network model that integrates physical mechanisms and data-driven approaches, specifically including:

[0066] S201. Model construction steps: The physical topology of the capillary network is abstracted into a hydraulic calculation diagram, wherein the nodes of the hydraulic calculation diagram correspond to the connection points of each pipe segment and the location of the intelligent monitoring nodes, and the edges of the hydraulic calculation diagram correspond to the capillary mats, branches and manifolds, and each edge is assigned an initial weight coefficient, which represents the comprehensive hydraulic resistance of the corresponding pipe segment under the reference working condition.

[0067] S202. Online solution steps: The pressure and flow data of each node collected in real time are used as the boundary conditions of the hydraulic calculation diagram. By solving the node flow balance equation set and loop pressure loss balance equation set established based on the laws of mass conservation and energy conservation, the actual dynamic weight coefficients of each edge in the hydraulic calculation diagram are derived, and the theoretical flow distribution and hydraulic misalignment of the whole network are calculated accordingly.

[0068] S203, Incremental Learning Step: The hydraulic network calculation model integrates an incremental learning module. After each closed-loop adjustment in step S3, the stabilized system state data is used as a new data sample. The weight coefficients of the corresponding edges in the hydraulic calculation graph are dynamically updated and calibrated using the recursive least squares method or stochastic gradient descent algorithm. This allows the hydraulic network calculation model parameters to continuously track the migration of physical characteristics of the pipeline network caused by scaling, deformation, or changes in valve characteristics.

[0069] In the above implementation, the primary task of constructing the hydraulic network calculation model for step S201 is to transform the complex physical pipe network system into a computable mathematical expression. Specifically, the physical topology of the capillary network, including the connection relationships and spatial orientation of all manifolds, branches, and capillary mats, is abstracted into a dedicated hydraulic calculation diagram. In this hydraulic calculation diagram, the connection points of each pipe segment and the actual locations of the deployed intelligent monitoring nodes are mapped as nodes of the hydraulic calculation diagram, while each specific pipe segment, whether it is a capillary mat, connecting branch, or main manifold, is mapped as an edge of the hydraulic calculation diagram. Assigning an initial weight coefficient to each edge is crucial in this step. This weight coefficient physically represents the comprehensive hydraulic resistance of the corresponding pipe segment under the reference operating conditions, integrating the pipe's friction resistance, local resistance, and additional pressure loss from the connectors. The initial weight can be initially estimated based on pipe design parameters such as pipe diameter, length, and material roughness, and using classical hydraulic formulas such as the Darcy-Visbach formula. This hydraulic computation graph, composed of nodes, edges, and initial weights, serves as a parameterized dynamic framework that encodes the structural constraints and prior knowledge of the physical system, laying the foundation for subsequent data assimilation and online solution.

[0070] For step S202, after the hydraulic network calculation model is constructed, the system enters the online solution step, which is crucial for the model to perform real-time diagnostic functions. In this step, the pressure and flow data of each node, collected and uploaded in real time from the intelligent monitoring nodes, are loaded onto the hydraulic calculation graph as known boundary conditions. The core problem the system needs to solve is two sets of equations based on the laws of mass and energy conservation: Node flow balance equations: requiring the total flow into any node to equal the total flow out of that node. Loop pressure loss balance equations: requiring the sum of pressure losses in each pipe segment of any closed loop to equal the driving head of that loop. Since the actual resistance of the pipe segments (i.e., the dynamic weight coefficients of the edges) is unknown, while the node pressure and some flow are known, this constitutes a typical inversion problem. To solve this problem, this invention integrates the pipe network topology, monitoring data, and fluid mechanics principles to construct a nonlinear optimization model based on a graph structure. This model uses the dynamic resistance coefficients of each pipe segment as optimization variables and the above two types of equations as constraints, and is solved using iterative optimization algorithms (such as the Newton-Raphson method or gradient descent method). The algorithm aims to minimize the residual between the monitored pressure / flow rate values ​​and the model's calculated values, thereby inversely calculating the dynamic weighting coefficients for each pipe segment that best match the current measured data. This numerical inversion process is equivalent to using limited field measurement data to "see through" the real-time hydraulic resistance state of each pipe segment within the entire pipeline network system. Based on the inverted dynamic weighting coefficients of the entire pipeline network, the model can calculate the theoretical flow distribution for all pipe segments in the network. By comparing the theoretical distribution with the design flow rate or ideal parameters, the model ultimately outputs a quantified system hydraulic imbalance, thus accurately assessing the current degree of system imbalance. This solution method can dynamically identify the actual pipeline network state within seconds to minutes, providing real-time and accurate physical parameter basis for subsequent balance calculations and control.

[0071] To ensure the hydraulic network computational model maintains high accuracy over the long term, an incremental learning step was designed for step S203. The incremental learning module integrated into the hydraulic network computational model endows the model with self-learning and continuous calibration capabilities. Its operating mechanism is as follows: each time the central control system completes the closed-loop adjustment in step S3 and brings the system into a new stable operating state, the key data of the system in its stable state (such as stable pressure, flow rates, and valve openings at each monitoring node) are packaged into a new high-confidence data sample. The incremental learning module then starts, using online learning algorithms such as recursive least squares or stochastic gradient descent, with this new data sample as supervision information, to fine-tune and dynamically update the weight coefficients of the corresponding edges in the hydraulic computation graph. For example, the hydraulic network computational model can set a learning rate parameter to control the magnitude of each update; this parameter can be selected between 0.01 and 0.1, preferably 0.05, to balance learning speed and stability. The essence of this process is to allow the parameters of the hydraulic network computational model to continuously track and adapt to the slow migration of physical characteristics caused by factors such as increased inner wall roughness due to scaling, minor deformations due to temperature or mechanical stress, or drift in valve characteristics due to long-term use. Through this continuous "data feeding" and parameter calibration, the hydraulic network computational model can gradually offset the initial modeling errors and dynamically conform to the real trajectory of system aging, thus evolving from a static, idealized design model into a dynamic digital twin that faithfully reflects the current real condition of the specific system.

[0072] Compared to traditional static calculation models or simplified empirical models that use fixed resistance parameters, this model offers a fundamental performance improvement. Its core technical advantage lies in achieving enduring model accuracy over time. Traditional models, once established, have fixed parameters that fail to reflect actual resistance changes after system operation. This leads to a widening discrepancy between theoretical and actual operating conditions over time, rapidly diminishing their guiding significance. In contrast, this hydraulic network calculation model, through an incremental learning mechanism, enables model parameters to evolve with the environment. It continuously absorbs errors caused by slow-changing factors such as system aging and scaling, ensuring highly reliable calculation results and decision-making support throughout the entire system lifecycle.

[0073] In one specific embodiment, the valve embedded controller of the digital twin-driven electric balancing valve has a pre-stored data table of valve opening degree-pressure difference-flow relationship obtained through experimental calibration; in step S3, the control process of the valve embedded controller is specifically as follows:

[0074] S301. Real-time acquisition step: Through the differential pressure sensor connected to it, the pressure before and after the valve of the digital twin driven electric balance valve is measured and acquired in real time, and the real-time differential pressure value is calculated.

[0075] S302, Opening command compensation step: Taking the target opening command received from the central control system and the real-time differential pressure value as input, query the valve opening-differential pressure-flow relationship data table, and calculate the compensated opening value required to achieve the target flow under the current differential pressure using a three-dimensional interpolation algorithm;

[0076] S303, Closed-loop control steps: Using the compensated opening value as the set value, read the actual opening value fed back by the valve absolute encoder, execute the proportional-integral-derivative control algorithm, and output the control signal to the drive mechanism of the valve embedded controller to drive the valve core to move to the compensated opening.

[0077] In the above embodiment, a key valve opening-differential pressure-flow relationship data table is pre-stored in the valve embedded controller of the digital twin-driven electric balancing valve. This data table is the basis for control accuracy and is obtained through experimental calibration. Specifically, under constant temperature laboratory conditions, the digital twin-driven electric balancing valve is installed on a standard test bench, and the stable flow values ​​corresponding to different valve openings and different combinations of pressure differences before and after the valve are systematically measured and recorded, thereby forming a three-dimensional data matrix. During the execution of step S3, control begins with the real-time acquisition step. The valve embedded controller, through a high-precision differential pressure sensor directly connected to it, measures and acquires the pressure before the valve at the valve inlet and the pressure after the valve at the valve outlet in real time, and immediately calculates the difference between the two to obtain the real-time differential pressure value. This real-time differential pressure value is a key dynamic parameter reflecting the current local hydraulic condition of the valve, and its measurement accuracy directly affects the effect of subsequent compensation calculations. The range selection of the differential pressure sensor needs to cover the possible pressure fluctuation range of the system. For example, a range of 0 to 100 kPa can be selected, of which 0 to 60 kPa is the commonly used and optimal operating range.

[0078] After obtaining the real-time differential pressure value, the system enters the valve opening command compensation step. This step is crucial for improving the control accuracy of traditional valves. The valve embedded controller takes the target opening command received from the central control system and the calculated real-time differential pressure value as input, and queries its internally stored valve opening-differential-flow relationship data table. Since the data table contains discrete data points, while the target opening and real-time differential pressure are continuous values, they are usually not exactly equal to a calibration point in the table. Therefore, a three-dimensional interpolation algorithm is needed to calculate the final compensated opening value required to achieve the target flow rate demanded by the central system under the current specific differential pressure. For example, bilinear interpolation or a more precise cubic spline interpolation method can be used. Based on the position of the target opening command and real-time differential pressure in the two-dimensional grid defined in the data table, the flow rate values ​​of surrounding known data points are used for fitting and extrapolation, and the precise opening that can generate the target flow rate is solved in reverse. This process essentially involves dynamically correcting the fixed valve characteristic curve according to the real-time hydraulic conditions on site. The density of the data table determines the interpolation accuracy. As an example of balancing storage space and accuracy, a data table can be built with a 5% opening interval and a 5 kPa pressure difference interval, while a better choice would be a 2% opening interval and a 2 kPa pressure difference interval.

[0079] After calculating the compensated opening value, the valve embedded controller immediately initiates a closed-loop control step to complete the final physical positioning. In this step, the compensated opening value is set as the target value for the control loop. The controller obtains the precise current position of the valve core by reading the actual opening value fed back by the valve absolute encoder mounted on the valve stem. The controller then executes a proportional-integral-derivative (PID) control algorithm. This algorithm continuously calculates the deviation between the compensated opening setpoint and the actual opening value fed back by the absolute encoder, and based on the comprehensive calculation of the three stages (PID, IDD, and DDD), generates and outputs a control signal in real time to the drive mechanism, such as a miniature DC motor. The drive mechanism drives the valve core to move in the direction that reduces the deviation according to this signal. The proportional stage provides a fast response, the integral stage is used to eliminate steady-state error, and the derivative stage is used to suppress overshoot and oscillation. The parameters of the control algorithm can be tuned according to the specific electromechanical characteristics of the valve. For example, the proportional coefficient can be adjusted within the range of 0.5 to 2.0, where 1.0 is a typical and stable preferred value. This closed-loop process continues until the actual opening value stabilizes within the allowable error range of the compensated opening value, such as within ±1%, thereby driving the valve core to move precisely to the optimal position after hydraulic condition compensation.

[0080] The valve-level precision control method defined in this embodiment achieves high-precision and robust regulation of valve flow by integrating real-time differential pressure sensing, three-dimensional interpolation compensation based on pre-calibrated data tables, and proportional-integral-derivative closed-loop control combined with absolute encoder feedback. This method fundamentally overcomes the core defect in traditional electric valve control where system pressure fluctuations cause inaccurate opening-flow relationships, enabling the valve output to closely track the target flow required by the system, significantly improving the accuracy and stability of overall network flow distribution. Simultaneously, localized intelligent compensation calculation and rapid closed-loop response significantly shorten the time for the system to reach hydraulic balance, reducing overshoot and oscillations during regulation, making the regulation process smoother and faster.

[0081] In one specific implementation, a pre-deployment simulation and optimized placement step is performed before step S1, specifically including:

[0082] S001. In Building Information Modeling (BIM), based on the design drawings of capillary networks, a three-dimensional capillary network model containing pipe diameter, length, elevation, and connection relationships is generated using parametric modeling components.

[0083] S002. Import the three-dimensional capillary network model into computational fluid dynamics software, assign material properties to the model, and apply the total flow inlet boundary condition corresponding to the design conditions and the design heat load of each room calculated based on the building thermal performance as the wall heat flux density boundary condition to the boundary of the three-dimensional capillary network model.

[0084] S003. In the computational fluid dynamics software, the k-ε turbulence model and the coupled heat transfer model are selected to perform steady-state and non-isothermal flow simulation calculations on the three-dimensional capillary network model, and the pressure field and flow field distribution data of the three-dimensional capillary network under the whole system operation state are obtained by solving the solution.

[0085] S004. Extract the feature parameter vectors of all hydraulic loops from the simulation results. The feature parameter vectors include at least the total pressure drop of the loop, the average flow rate of the loop, and the design heat load of the room where the loop is located. Use the K-means clustering algorithm to perform cluster analysis on the feature parameter vectors of all hydraulic loops, and select the loop closest to the cluster center from each cluster and mark it as the key monitoring loop.

[0086] The pre-deployment simulation and optimized deployment steps also include:

[0087] S005. Based on the pressure and flow field distribution data obtained in step S003, extract the pressure and flow reference values ​​of each key monitoring loop under rated operating conditions. Combine the design specifications and historical operating data to set dynamic early warning thresholds and alarm thresholds for pressure and flow for each key monitoring node. Use the initial pressure and flow distribution data of the entire network obtained from the simulation in step S003 as the initial training dataset for the hydraulic network calculation model to complete the pre-training and initialization of the hydraulic network calculation model parameters.

[0088] In the above implementation method, before proceeding to actual construction and monitoring node deployment, this method first performs pre-deployment simulation and optimized site layout steps, aiming to scientifically plan the monitoring network in advance through digital means. This process begins in a Building Information Modeling (BIM) environment, based on detailed capillary network design drawings, using parametric modeling components to generate a three-dimensional digital model of the capillary network containing precise pipe diameters, lengths, elevations, and all connection relationships. This model completely reproduces the spatial topology of the future physical system. Subsequently, this three-dimensional model is imported into professional computational fluid dynamics software, assigning realistic material properties such as pipe wall roughness to different pipe segments, and applying hydraulic and thermal conditions at the model boundaries that completely correspond to the design conditions. Specifically, the total flow rate is set at the system's main inlet as the inlet boundary condition, while the design heat load of each room, calculated based on the thermal performance of the building envelope, is converted into the wall heat flux density boundary condition for the corresponding room's ceiling or wall-embedded pipe area. In the software, models such as the k-ε turbulence model and the coupled heat transfer model are selected to perform steady-state, non-isothermal flow simulations on a three-dimensional capillary network model. This simulates the flow and heat transfer processes of water within the network, ultimately obtaining detailed global pressure and flow field distribution data for the system under design conditions. These simulation results provide a high-fidelity data foundation for understanding the hydraulic characteristics of the system.

[0089] Based on detailed pressure and flow field data obtained from computational fluid dynamics simulations, the system enters the core analysis phase of optimizing monitoring point placement. First, characteristic parameter vectors for all independent hydraulic loops are extracted from the simulation results. This vector is a multi-dimensional descriptor, containing at least the total pressure drop of the loop, the average flow rate on the loop, and the design heat load of the room served by the loop. These parameters comprehensively characterize the hydraulic characteristics and functional importance of a loop. Next, the K-means clustering algorithm is used to perform unsupervised clustering analysis on these characteristic parameter vectors of all loops. The purpose of clustering is to automatically group hundreds or even more hydraulic loops into several typical categories based on the similarity of their hydraulic and load characteristics. For example, loops can be clustered into 3 to 7 categories, with 5 categories being a common and effective choice. After the algorithm is executed, from each final cluster, the real loop whose characteristic vector is mathematically closest to the cluster center is selected and identified as the key monitoring loop. This loop is the "most typical representative" of its category, and monitoring its status can reflect the common characteristics of all loops in this category with high efficiency, thus achieving the scientific goal of covering the largest range of system operating conditions with the fewest monitoring points.

[0090] After identifying the critical monitoring loops, the pre-deployment simulation step further extends its value. Based on the pressure and flow field distribution data obtained from the aforementioned simulation, the pressure and flow baseline values ​​under rated operating conditions are extracted for each characteristic location identified as a critical monitoring loop. Combining industry design specifications and historical operating data from similar systems, dynamic early warning and alarm thresholds are set for each critical monitoring node. For example, the pressure early warning threshold can be set to fluctuate within ±15% of the baseline value, while the alarm threshold can be set within ±25%, with ±20% being the preferred value balancing safety and sensitivity. The flow threshold can be set according to similar principles. More importantly, the initial pressure and flow distribution data of the entire network obtained from the simulation calculation is used as the initial training dataset for the hydraulic network calculation model. This high-quality, noise-free simulation data, which is highly consistent with the design, is input into the model to drive its parameter pre-training and initialization. This is equivalent to injecting a set of theoretical "prior knowledge" into the core of intelligent control before the system actually runs, so that it is no longer a blank slate when it is first put into use, but has a cognitive foundation close to the design conditions, creating a superior starting point for a rapid and smooth transition to the adaptive learning stage.

[0091] The pre-deployment simulation and optimized point placement method defined in this embodiment elevates the planning of the monitoring network and the basic training of the control system from a stage of rough estimation relying on experience to a stage of scientific design based on data. It reveals the inherent hydraulic laws of complex pipe network systems in advance through high-fidelity fluid simulation and intelligently selects the most representative key monitoring points from a large number of similar loops using cluster analysis. This allows for the construction of a highly efficient and blind-spot-free sensing network at a cost far lower than full-domain deployment, ensuring the representativeness and effectiveness of subsequent monitoring data from the outset. Simultaneously, this method uses simulation data to provide high-quality prior training and reasonable threshold benchmarks for the intelligent model, enabling the entire intelligent control system to start operating at a high starting point and near-optimal design state when the physical system is first powered on. This significantly shortens the system debugging and learning adaptation period and avoids blind oscillations and low performance in the initial stage.

[0092] In one specific embodiment, the valve embedded controller of the digital twin-driven electric balancing valve further integrates a fault self-diagnosis and fault-tolerant control module, which specifically comprises:

[0093] The valve absolutely encoder periodically compares the actual opening value with the drive mechanism commanded opening value. If the deviation continues to exceed the preset tolerance range, it is determined that the valve core is stuck or the transmission mechanism is faulty, and a fault code is generated and uploaded to the central control system.

[0094] When abnormal differential pressure sensor data or communication interruption is detected, the system automatically switches to the opening-flow empirical curve feedforward control mode, maintaining the valve's approximate regulation function based on a simplified relationship fitted from historical normal operation data until the fault is cleared.

[0095] In the above embodiment, the fault self-diagnosis and fault-tolerant control module integrated within the valve embedded controller of the digital twin-driven electric balancing valve is an intelligent software unit that runs continuously in the background. Its core function is to ensure the reliable operation of the valve body and maintain basic functions under abnormal conditions. The operation of this module is based on the continuous monitoring and analysis of the valve's key status parameters. It periodically collects and compares the actual opening feedback value from the valve's absolute encoder with the command opening value sent by the controller to the drive mechanism. The frequency of this periodic detection can be set according to the system's reliability requirements. For example, a diagnostic cycle can be performed every 10 to 60 seconds, with every 30 seconds being an optimal value that strikes a good balance between diagnostic timeliness and processor load. By continuously comparing these two values, which should be highly consistent, the module can gain insight into the true state of the valve's mechanical actuator.

[0096] When the fault self-diagnosis module detects that the deviation between the actual opening value and the commanded opening value continuously exceeds a preset tolerance range over several consecutive diagnostic cycles, it triggers the fault determination logic. This preset tolerance range is a threshold set to distinguish between normal control errors and genuine mechanical faults. For example, for high-precision control valves, this tolerance range can be set to 2% to 5% of the commanded opening value, where 3% is a typical value that can effectively filter out minor fluctuations while capturing substantial jamming. If the deviation continues to exceed this range, such as three consecutive diagnostic results exceeding the limit, the module determines that there may be mechanical faults inside the valve, such as the valve core being stuck by impurities, damage to the transmission gears, or motor step loss. Once the fault is determined, the module immediately generates a fault information containing a fault type code and a timestamp, and uploads it to the central control system via the communication network, thereby proactively reporting the fault and realizing an automated process from "fault occurrence" to "system awareness".

[0097] In addition to mechanical fault diagnosis, this fault self-diagnosis and fault-tolerant control module also possesses fault-tolerant control capabilities for sensors and communication systems. When the module detects persistently abnormal data from its connected differential pressure sensor—such as readings exceeding the reasonable range, signal values ​​remaining constant or experiencing drastic fluctuations—or detects an interruption in communication with the central control system, it automatically triggers a control mode switch. The valve will no longer rely on unavailable real-time differential pressure data or new commands from the central system, but will automatically switch to a pre-set opening-flow empirical curve feedforward control mode. This empirical curve is a simplified relationship model obtained through statistical analysis and fitting during the valve's historical normal operation; it directly correlates the valve opening with the approximate flow output. In this degraded mode, the valve controller will continue to drive the valve for approximate regulation based on the last received valid target opening command, combined with the locally stored empirical curve. This fault-tolerance mechanism ensures that the valve can maintain a relatively reasonable opening during the period of critical sensor failure or temporary network interruption, avoiding serious deterioration of the hydraulic state of the loop due to complete valve lock-up or uncontrolled full opening / closing, thus providing maintenance personnel with a time window for troubleshooting and repair.

[0098] The fault self-diagnosis and fault-tolerant control module defined in this embodiment injects crucial self-sensing and survivability capabilities into the intelligent balancing valve, transforming the traditional valve's passive waiting for fault alarms or manual inspection into an advanced mode of proactive monitoring, intelligent diagnosis, and autonomous fault tolerance. Through periodic self-checks, it can detect potential mechanical faults such as valve core jamming at an early stage and proactively report them, allowing the maintenance team to calmly plan preventative maintenance and avoid system downtime and greater losses that could result from escalating faults. More importantly, its automatic mode switching function in the event of sensor or communication failure provides a valuable soft-landing mechanism, ensuring that the hydraulic regulation function of the valve and even the entire loop does not completely collapse when a local component fails, maintaining the system's basic operational stability and continuity, and greatly enhancing the overall resilience and availability of complex pipeline systems in the face of local faults.

[0099] In one specific embodiment, the lightweight hydraulic balance fast diagnostic algorithm built into the edge computing gateway further includes:

[0100] Construct a pressure gradient matrix with intelligent monitoring nodes as spatial nodes, and calculate and monitor the gradient vector in the pressure gradient matrix that represents the rate of change of pressure difference between adjacent nodes in real time.

[0101] When the gradient vector magnitude of a certain branch continuously exceeds the dynamic threshold calculated based on the pipeline topology, it is determined that there is a blockage or gas accumulation anomaly in that branch, and the location and severity level of the anomaly are estimated based on the direction and magnitude of the gradient vector.

[0102] A diagnostic report containing the abnormal branch identifier, abnormal type, severity, and suggested troubleshooting measures is generated and uploaded to the central control system along with the abnormal status marker.

[0103] In the above implementation, the lightweight hydraulic balance rapid diagnostic algorithm built into the edge computing gateway is based on the construction and analysis of a mathematical model that can characterize the dynamic spatial pressure distribution of the pipeline network. The algorithm first constructs a pressure gradient matrix with these nodes as spatial nodes, based on the physical locations of the actually deployed intelligent monitoring nodes and their connections within the pipeline network topology. This matrix mathematically describes the pressure relationships between all adjacent monitoring nodes. The algorithm receives pressure data reported by each node in real time and continuously calculates and monitors a series of key gradient vectors in the matrix. Each gradient vector essentially reflects the pressure difference between its corresponding two adjacent nodes at a specific moment and its trend over time. For example, the diagnostic algorithm built into the edge computing gateway can independently of the data reporting process, collecting a pressure snapshot of all network nodes and calculating gradients at a higher frequency (e.g., every 10 to 60 seconds), where 30 seconds is a typical preferred value that balances diagnostic real-time performance with gateway computational load. By continuously tracking these gradient vectors, the algorithm transforms the originally isolated node pressure readings into a dynamic spatial relationship network reflecting the pressure transmission and resistance state within the pipeline network, providing a structured data foundation for anomaly detection.

[0104] After the algorithm completes real-time monitoring of the pressure gradient matrix, it enters the anomaly detection phase. This detection process relies on a dynamic threshold system. This dynamic threshold is not a fixed value but is calculated based on the specific pipeline topology, the calculated length of each pipe segment, and the normal operating pressure differential range of the system. For example, for a branch designed for a normal pressure differential of approximately 20 kPa, its corresponding dynamic threshold for the pressure gradient magnitude might be set between 50% and 150% of the normal average, with 80% considered a preferred value that strikes a good balance between sensitivity and immunity. The specific detection logic is as follows: when the algorithm detects that the magnitude of the gradient vector corresponding to a specific branch—that is, the comprehensive measure of its pressure differential change rate—continuously exceeds the calculated dynamic threshold for several consecutive monitoring cycles, an anomaly detection is triggered. The system will determine that the branch has a physical anomaly, such as blockage or gas accumulation. Furthermore, the algorithm uses the spatial orientation of the gradient vector to locate whether the anomaly is more likely to be located upstream or downstream of the branch, and estimates the severity level of the anomaly based on the magnitude of the magnitude exceeding the threshold, for example, it can be divided into three levels: mild, moderate and severe.

[0105] Once the anomaly detection and preliminary analysis are complete, the algorithm automatically generates a structured diagnostic report. This report not only includes the physical identifier or logical number of the abnormal branch, but also clearly indicates the type of anomaly initially determined, such as suspected solid blockage or suspected gas accumulation. The report includes the severity level of the aforementioned determination and, based on the domain knowledge base and preset rules, provides preliminary suggested troubleshooting measures, such as prioritizing pressure relief checks or venting operations on a specific pipeline section. This diagnostic report is converted into a specific data format and bound together with previously generated anomaly status tags used for simple labeling, and uploaded to the central control system through the data channel of the edge computing gateway. In this way, the central system not only receives an alert about "where an anomaly might occur," but also obtains a comprehensive on-site analysis summary containing preliminary diagnostic conclusions, quantitative severity, and operational recommendations, providing operational and maintenance decisions with in-depth information support far exceeding that of a simple alarm.

[0106] The rapid diagnostic algorithm further defined in this embodiment, by constructing a pressure gradient matrix and implementing intelligent analysis based on dynamic thresholds, elevates the data processing capabilities of the edge computing gateway from simple filtering to a level with basic spatial analysis and diagnostic reasoning, enabling early, accurate, and automated location of pipeline blockages and gas accumulation faults. Compared to traditional methods relying on central system post-processing or manual experience-based judgment, this method completes core analysis near the data source, reducing fault location response time by an order of magnitude and effectively distinguishing fault types and severity. This not only greatly improves the targeting and efficiency of operation and maintenance response and handling, realizing a shift from passive maintenance to proactive early warning, but also significantly enhances the stability and reliability of the entire capillary network system by preventing local hydraulic anomalies from escalating into systemic imbalances.

[0107] In one specific implementation, during the initialization and debugging process in S4, a dedicated portable calibrator is used to conduct near-field wireless communication with the on-site intelligent monitoring node. The dedicated portable calibrator has a built-in high-precision standard sensor, which can be temporarily connected to the pipeline network measurement point to perform on-site reading comparison and calibration of the fixed-installed intelligent monitoring node, and generate calibration coefficients, which are automatically written to the corresponding node after being confirmed by the debugging personnel.

[0108] In the above implementation, a key step in the initialization and commissioning process of step S4 is the introduction of a dedicated portable calibrator to perform on-site accuracy verification and calibration of the fixedly installed intelligent monitoring nodes. This dedicated portable calibrator is a standalone, portable precision measuring device that incorporates high-precision standard pressure and flow sensors. The accuracy of these sensors is typically far higher than that of the industrial-grade sensors in intelligent monitoring nodes designed for long-term stable installation; for example, their accuracy can reach ±0.2% of the measured value, while the accuracy of sensors in fixed nodes is typically around ±1%. The calibrator establishes a temporary, point-to-point data connection with the on-site intelligent monitoring nodes via near-field wireless communication methods, such as Bluetooth Low Energy or Wi-Fi Direct, forming a local calibration network. This communication method is chosen to eliminate the hassle of complex wiring in the on-site commissioning environment, enabling flexible and secure interaction between commissioning personnel and the equipment.

[0109] The specific calibration process is as follows: The calibration personnel temporarily connect a dedicated portable calibrator to the pipeline measurement point where the intelligent monitoring node to be calibrated is located via its equipped quick interface. This is typically achieved through a pre-installed tee or a reserved verification interface, ensuring that the calibrator's high-precision sensor senses the exact same fluid medium state as the fixed node sensor being calibrated, i.e., the same pressure and flow rate. Under stable system conditions, the calibrator and the fixed node synchronously collect hydraulic parameter data over a period of time. For example, they synchronously collect instantaneous pressure and flow rate values ​​over 30 seconds to 2 minutes, with 60 seconds being a preferred duration for obtaining stable statistical results. The calibrator's embedded processor uses the values ​​measured by its own high-precision sensor as the "true value" and compares them in real time with the "measured values" reported by the fixed node sensor. By calculating the difference or proportional relationship between the two at the same timestamp, the system analyzes potential systematic deviations in the fixed node's measurement data, such as fixed zero-point drift or linear gain error.

[0110] Based on the results of real-time comparative analysis, the algorithm within the dedicated portable calibrator automatically generates one or more calibration coefficients. For pressure sensors, these coefficients may include zero-point offset corrections and range slope corrections; for flow meters, they may primarily be flow coefficient corrections. These coefficients are encapsulated in a calibration command packet and sent to the corresponding smart monitoring node via near-field wireless communication. The smart monitoring node's firmware has a dedicated calibration interface that receives and parses the command packet, but it does not take effect immediately. The calibration coefficients are temporarily stored, and the calibration results (such as a comparison of values ​​before and after correction) are displayed on the calibrator's screen or presented via the operator's mobile terminal. After the operator verifies that the calibration results are reasonable, they issue a confirmation command on-site via the calibrator or a connected terminal. This confirmation command triggers the smart monitoring node to formally write the temporarily stored calibration coefficients into a designated area of ​​its non-volatile memory, completing the solidification of the calibration parameters. Subsequently, in all subsequent data acquisition and reporting, the node will automatically apply these calibration coefficients to correct the original sensor signal, thereby ensuring that its measurement output values ​​are traceable to a high-precision standard sensor and guaranteeing its long-term measurement accuracy.

[0111] The on-site calibration mechanism defined in this implementation provides crucial source assurance for the data quality of the entire intelligent monitoring system. By transferring laboratory-level metrological standards to the engineering site, it effectively eliminates initial measurement errors caused by manufacturing discreteness, installation stress, or environmental factors in intelligent monitoring nodes, as well as slow drift that may occur during use. This ensures that all subsequent hydraulic model calculations, balance analyses, and control decisions based on monitoring data are built on an accurate and reliable sensing foundation, avoiding the systemic risk of "distorted data leading to erroneous decisions." Compared to traditional methods that periodically remove sensors for testing or rely on factory calibration without correcting on-site application deviations, this method allows for convenient, rapid, and high-precision online calibration at the system engineering site. This significantly improves the data reliability and authority of the monitoring system throughout its entire lifecycle, reduces the risk of operational misjudgments and subsequent adjustment costs caused by measurement inaccuracies, and strengthens the foundation of the entire intelligent balancing system's effectiveness from the most fundamental level.

[0112] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0113] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details.

Claims

1. A method for intelligent monitoring and hydraulic balance implementation of capillary network systems, characterized in that, Includes the following steps: S1. Based on the hydraulic topology of capillary networks, intelligent monitoring nodes are deployed at preset key node locations to collect time-series data of hydraulic and thermal parameters. S2. The time series data is uploaded to the central control system via the edge computing gateway. The capillary network state is solved by the built-in hydraulic network calculation model, and the calculation result set containing the system hydraulic misalignment is output. S3. The central control system compares the system hydraulic misalignment in the calculation result set with a preset threshold. When the system hydraulic misalignment exceeds the preset threshold, it calculates the target opening instruction set for each loop digital twin-driven electric balancing valve based on the theoretical flow rate of each pipe section in the calculation result set and the preset valve flow characteristic curve. The instruction is encrypted and sent to the valve embedded controller. The valve embedded controller controls its drive mechanism to read the real-time opening feedback value of the valve absolute encoder and executes the proportional-integral control algorithm with the target opening instruction as the set value. S4. After the physical installation of the system is completed, a fully automatic initialization and debugging is performed. The central control system sends a scanning command from fully closed to fully open to all digital twin driven electric balance valves and records the pressure and flow data fed back by each intelligent monitoring node. By fitting the correspondence between pressure and flow data, an actual pipeline characteristic curve is generated. The actual pipeline characteristic curve is compared with the pre-stored design model characteristic curve corresponding to the hydraulic network calculation model. The initial opening compensation value of each digital twin driven electric balance valve is calculated and written into the valve embedded controller to complete the pre-setting. The hydraulic network computation model is an adaptive graph network model that integrates physical mechanisms and data-driven approaches, specifically including: S201. Model construction steps: The physical topology of the capillary network is abstracted into a hydraulic calculation diagram, wherein the nodes of the hydraulic calculation diagram correspond to the connection points of each pipe segment and the location of the intelligent monitoring nodes, and the edges of the hydraulic calculation diagram correspond to the capillary mats, branches and manifolds, and each edge is assigned an initial weight coefficient, which represents the comprehensive hydraulic resistance of the corresponding pipe segment under the reference working condition. S202. Online solution steps: The pressure and flow data of each node collected in real time are used as the boundary conditions of the hydraulic calculation diagram. By solving the node flow balance equation set and loop pressure loss balance equation set established based on the laws of mass conservation and energy conservation, the actual dynamic weight coefficients of each edge in the hydraulic calculation diagram are derived, and the theoretical flow distribution and hydraulic misalignment of the whole network are calculated accordingly. S203, Incremental Learning Step: The hydraulic network calculation model integrates an incremental learning module. After each closed-loop adjustment in step S3, the system state data after the adjustment is stabilized is used as a new data sample. The weight coefficients of the corresponding edges in the hydraulic calculation diagram are dynamically updated and calibrated by recursive least squares method or stochastic gradient descent algorithm, so that the parameters of the hydraulic network calculation model continuously track the migration of physical characteristics of the pipeline network caused by scaling, deformation or changes in valve characteristics. Before step S1, perform pre-deployment simulation and optimized placement steps, specifically including: S001. In Building Information Modeling (BIM), based on the design drawings of capillary networks, a three-dimensional capillary network model containing pipe diameter, length, elevation, and connection relationships is generated using parametric modeling components. S002. Import the three-dimensional capillary network model into computational fluid dynamics software, assign material properties to the model, and apply the total flow inlet boundary condition corresponding to the design conditions and the design heat load of each room calculated based on the building thermal performance as the wall heat flux density boundary condition to the boundary of the three-dimensional capillary network model. S003. In the computational fluid dynamics software, the k-ε turbulence model and the coupled heat transfer model are selected to perform steady-state and non-isothermal flow simulation calculations on the three-dimensional capillary network model, and the pressure field and flow field distribution data of the three-dimensional capillary network under the whole system operation state are obtained by solving the solution. S004. Extract the feature parameter vectors of all hydraulic loops from the simulation results. The feature parameter vectors include at least the total pressure drop of the loop, the average flow rate of the loop, and the design heat load of the room where the loop is located. Use the K-means clustering algorithm to perform cluster analysis on the feature parameter vectors of all hydraulic loops, and select the loop closest to the cluster center from each cluster and mark it as the key monitoring loop.

2. The intelligent monitoring and hydraulic balance implementation method for capillary network systems as described in claim 1, characterized in that, In step S1, the preset key node locations specifically include the end of the water supply manifold and the end of the return water manifold of all independent hydraulic loops, as well as the capillary inlet of the key monitoring loop selected based on the weighting factor determined by the hydraulic distance and design heat load of each loop. The intelligent monitoring node integrates a pressure sensor, a flow meter, and a temperature sensor.

3. The intelligent monitoring and hydraulic balance implementation method for capillary network systems as described in claim 1, characterized in that, In step S2, the time-series data is transmitted to the edge computing gateway via a low-power wide area network. The edge computing gateway cleans, aligns, and compresses the data to form a standardized data packet before uploading it to the central control system. The edge computing gateway has a built-in lightweight hydraulic balance rapid diagnosis algorithm. This algorithm analyzes the spatial gradient distribution of pressure data from each intelligent monitoring node at the same time to identify and locate abnormal branches in the capillary network that are suspected of being blocked or containing gas in real time, and generates a corresponding abnormal status marker. This abnormal status marker is then uploaded to the central control system along with the standardized data packet.

4. The intelligent monitoring and hydraulic balance implementation method for capillary network systems as described in claim 1, characterized in that, The embedded controller of the digital twin-driven electric balancing valve has a pre-stored data table of valve opening degree-pressure difference-flow relationship obtained through experimental calibration; in step S3, the control process of the embedded controller is as follows: S301. Real-time acquisition step: Through the differential pressure sensor connected to it, the pressure before and after the valve of the digital twin driven electric balance valve is measured and acquired in real time, and the real-time differential pressure value is calculated. S302, Opening command compensation step: Taking the target opening command received from the central control system and the real-time differential pressure value as input, query the valve opening-differential pressure-flow relationship data table, and calculate the compensated opening value required to achieve the target flow under the current differential pressure using a three-dimensional interpolation algorithm; S303, Closed-loop control steps: Using the compensated opening value as the set value, read the actual opening value fed back by the valve absolute encoder, execute the proportional-integral-derivative control algorithm, and output the control signal to the drive mechanism of the valve embedded controller to drive the valve core to move to the compensated opening.

5. The intelligent monitoring and hydraulic balance implementation method for a capillary network system as described in claim 1, characterized in that, The pre-deployment simulation and optimized deployment steps also include: S005. Based on the pressure and flow field distribution data obtained in step S003, extract the pressure and flow reference values ​​of each key monitoring loop under rated operating conditions. Combine the design specifications and historical operating data to set dynamic early warning thresholds and alarm thresholds for pressure and flow for each key monitoring node. Use the initial pressure and flow distribution data of the entire network obtained from the simulation in step S003 as the initial training dataset for the hydraulic network calculation model to complete the pre-training and initialization of the hydraulic network calculation model parameters.

6. The intelligent monitoring and hydraulic balance implementation method for a capillary network system as described in claim 4, characterized in that, The embedded controller of the digital twin-driven electric balancing valve also integrates a fault self-diagnosis and fault-tolerant control module, which specifically includes: The valve absolutely encoder periodically compares the actual opening value with the drive mechanism commanded opening value. If the deviation continues to exceed the preset tolerance range, it is determined that the valve core is stuck or the transmission mechanism is faulty, and a fault code is generated and uploaded to the central control system. When abnormal differential pressure sensor data or communication interruption is detected, the system automatically switches to the opening-flow empirical curve feedforward control mode, maintaining the valve's approximate regulation function based on a simplified relationship fitted from historical normal operation data until the fault is cleared.

7. The intelligent monitoring and hydraulic balance implementation method for a capillary network system as described in claim 3, characterized in that, The lightweight hydraulic balance fast diagnostic algorithm built into the edge computing gateway further includes: Construct a pressure gradient matrix with intelligent monitoring nodes as spatial nodes, and calculate and monitor the gradient vector in the pressure gradient matrix that represents the rate of change of pressure difference between adjacent nodes in real time. When the gradient vector magnitude of a certain branch continuously exceeds the dynamic threshold calculated based on the pipeline topology, it is determined that there is a blockage or gas accumulation anomaly in that branch, and the location and severity level of the anomaly are estimated based on the direction and magnitude of the gradient vector. A diagnostic report containing the abnormal branch identifier, abnormal type, severity, and suggested troubleshooting measures is generated and uploaded to the central control system along with the abnormal status marker.

8. The intelligent monitoring and hydraulic balance implementation method for a capillary network system as described in claim 1, characterized in that, During the initialization and debugging process in step S4, a dedicated portable calibrator is used to conduct near-field wireless communication with the on-site intelligent monitoring node. The dedicated portable calibrator has a built-in high-precision standard sensor that can be temporarily connected to pipeline measurement points for on-site reading comparison and calibration of fixed-installation intelligent monitoring nodes, and generates calibration coefficients that are automatically written to the corresponding nodes after confirmation by the commissioning personnel.

Images