Water flow intelligent monitoring system and method

Abstract

The invention discloses a water flow intelligent monitoring system and method, and relates to the technical field of water flow intelligent monitoring, in combination with an actuator health degree evaluation and multi-target prediction control method, the characteristics of a valve and a water pump are finely described by arranging sensors and constructing a digital twinborn model; model predictive control or multi-agent reinforcement learning is adopted to dynamically generate a reliable and efficient control instruction; furthermore, through real-time closed-loop deviation detection and self-adaptive correction, delay, faults or communication abnormity of the actuator can be accurately dealt with; 4, quickly completing self-healing and fault-tolerant control by using a fault prediction and standby scheduling strategy to prevent failure diffusion; and 5, scenes such as rainfall flood, fire fighting and reclaimed water are incorporated into the same framework, and multi-scene coupled comprehensive water resource management is realized, so that water hammer impact and water leakage risks can be effectively reduced, and the safety of the water supply system under complex working conditions or equipment aging is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent water flow monitoring, and in particular to a system and method for intelligent water flow monitoring. Background Art

[0002] In modern urban water systems, water supply networks are not only responsible for ensuring water supply for daily life, but are also often required to provide flexible and adjustable hydraulic support in various scenarios such as rainstorm emergencies, firefighting support, and recycled water reuse.

[0003] To cope with complex and highly dynamic water demand, many cities have installed high-precision sensors at key points in their pipeline networks to collect real-time data on parameters such as pressure, flow, valve opening, and pump motor current. Furthermore, digital twin models are being introduced to simulate the network topology and hydraulic conditions, enabling more timely and secure control and dispatch in the event of leaks, sudden surges, or equipment failures. However, as the network continues to expand, factors such as pipeline aging, local valve corrosion, and increasingly complex system topologies have significantly increased the difficulty of global coordination, making conventional decentralized or rule-based dispatching difficult to meet the high-frequency, sophisticated, and safe requirements. Especially in extreme weather conditions or fire emergencies, water supply networks must coordinate the opening and closing of multiple valves and the switching of pump speeds in a short period of time, while ensuring that water hammer and energy costs are manageable. This poses even greater challenges to the real-time nature of monitoring data and the reliability of dispatching strategies, necessitating an upgrade to a smart pipeline network system with predictive, detection, and self-healing capabilities.

[0004] After searching, a Chinese invention patent with authorization announcement number CN102629106B discloses a water supply control method and system, including: analyzing historical data to establish a water supply prediction model and a function of input layer parameters and water supply output by the output layer; using a genetic algorithm to calculate an optimization plan under restrictive conditions; the above-mentioned water supply control method and system analyzes historical data to establish a water supply prediction model and a water supply function, establishes an objective function to perform combinatorial optimization on water supply scheduling, uses a genetic algorithm to perform solution optimization under restrictive conditions, and simulates the principle of biological evolution to perform optimization with relatively high efficiency, can solve complex combinatorial optimization problems, and verifies the stability and practicality of the optimization method by introducing the restrictive conditions of the water supply model to study, thereby achieving stable supply and the most economical energy-saving optimization control in the water supply link of the water plant.

[0005] In this context, combined with actual application scenarios and existing technologies:

[0006] The core technical problem to be solved by the present invention mainly revolves around how to carry out linkage control and fault-tolerant scheduling of actuators (including valves and water pumps) of large-scale pipeline networks under multi-dimensional real-time monitoring conditions to cope with equipment aging, transient hydraulic shocks and dynamic requirements in multiple scenarios. If only relying on the traditional independent opening and closing of a single valve or manual experience-based scheduling, once the valve is delayed, the pump station fails or the communication is interrupted, it may not be possible to detect, locate and correct its abnormal state in time, resulting in local water pressure imbalance or aggravated water leakage accidents, and even large-scale water supply interruption and water hammer damage to the pipeline. Serious consequences. At the same time, for extended scenarios involving storm water drainage, fire protection and recycled water utilization, if there is a lack of real-time control and fault prediction of cross-system hydraulic distribution and actuator health, problems such as missed scheduling opportunities, sharp increase in energy consumption or inability to switch to backup pipelines in extreme weather or sudden fire situations often occur.

[0007] It can be seen from this that there is an urgent need to build a set of intelligent scheduling and fault-tolerant technical means that can integrate real-time data, digital twin predictions and dynamic health assessments of actuators, which can not only meet the needs of multi-scenario linkage, but also ensure that in the event of equipment aging, failure or sudden high load, it can still quickly restore and maintain highly reliable water supply services.

[0008] To this end, the present invention provides a water flow intelligent monitoring system and method. Summary of the Invention

[0009] (1) Technical problems solved

[0010] To address the shortcomings of existing technologies, the present invention provides an intelligent water flow monitoring system and method. By deploying sensors and constructing digital twin models, the characteristics of valves and pumps are precisely characterized. Model predictive control or multi-agent reinforcement learning is then used to dynamically generate reliable and efficient control instructions. In step three, real-time closed-loop deviation detection and adaptive correction are used to accurately respond to actuator delays, failures, or communication anomalies. In step four, fault prediction and backup scheduling strategies are used to rapidly complete self-healing and fault-tolerant control to prevent the spread of failures. In step five, scenarios such as rainwater, firefighting, and recycled water are incorporated into the same framework to achieve comprehensive water resource management with multi-scenario coupling. This system can effectively reduce the risk of water hammer and water leakage, and significantly improve the safety of water supply systems in complex operating conditions or when equipment is aging. This solves the technical problems described in the background art.

[0011] (2) Technical solution

[0012] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0013] A water flow intelligent monitoring method, comprising:

[0014] Based on multi-source data sets collected by sensors and historical operation logs, a hydraulic digital twin model of the pipeline network is generated through multi-dimensional real-time monitoring and actuator health mapping to dynamically simulate the evolution of hydraulic topology and output actuator health vectors.

[0015] When a coordinated scheduling or water leakage isolation task is detected within the control cycle, a multi-objective model prediction is used to couple the valve opening and closing with the pump speed regulation strategy, calculate the optimal control sequence, and output the action plan and transient response prediction.

[0016] Compare the predicted values ​​of the pipeline network hydraulic digital twin model with the actual sensor information, and use closed-loop deviation detection and adaptive correction to make online adjustments to the valve opening and closing rates and pump station switching timings that have not yet been executed or are being executed, and simultaneously update the actuator health vector;

[0017] When execution layer anomalies are continuously detected or the probability of a fault is judged to be increasing, the remaining availability of the actuator is calculated, and backup valves or adjacent water pumps are quickly called up to simulate the impact of emergency switching on the hydraulics of the pipe network and transmit the fault information back;

[0018] After loading and extending, a hydraulic digital twin model of the integrated pipeline network is constructed. Multi-dimensional water use indicators are measured with multi-objective functions. After cross-system collaborative scheduling is carried out in combination with fault prediction and fault-tolerant strategies, different water supply strategies are solved in layers and resources are reallocated.

[0019] Furthermore, high-precision sensors are deployed at key nodes of the pipe network to collect real-time network management data. Sensor self-test information is collected and integrated into the real-time network management data to obtain the actuator health vector that is updated incrementally over time.

[0020] After integrating historical operation logs, maintenance records, and actuator operation timestamps to form the initial multi-source data set and instantaneous comprehensive monitoring vector, a hydraulic digital twin model of the pipeline network is constructed in the virtual space.

[0021] Furthermore, the state difference and gradient difference are simultaneously considered in the time and space dimensions, and the calibrated hydraulic digital twin model of the pipeline network is obtained by minimizing the error metric function;

[0022] The error metric function is combined with the hydraulic partial differential constraint equation to update the parameters of the pipeline network hydraulic digital twin model in real time, and the actuator health vector is incorporated into the calibration process.

[0023] Furthermore, a multi-objective control model is performed on the valve opening and closing sequence, water pump speed regulation, and pressure distribution within the determined prediction time domain to obtain a multi-objective optimization function;

[0024] Within a certain prediction time domain, distributed reinforcement learning is used to gradually optimize the multi-objective optimization function within the prediction time domain, generate the optimal control sequence for valve opening and closing and water pump speed regulation, and update it over time.

[0025] Furthermore, timing and amplitude constraints are set for valve and pump actions, resulting in a constraint set that couples health and transient hydraulic constraints. If a valve is detected to be stuck or has a switching delay, the corresponding valve's actionable range is explicitly written into the constraint set. Local optimization is performed on each valve and pump, and a secondary solution is obtained based on the multi-objective optimization function.

[0026] A simulation evaluation is performed on multiple candidate control sequences, and the optimal one under the multi-objective optimization function is selected as the actual issued instruction. The predicted state sequence of the corresponding instruction in the digital twin environment is recorded as a control benchmark.

[0027] Furthermore, the deviation ratio of the sensor observation values ​​collected in real time is calculated, and a multi-dimensional deviation measurement function is used to perform multi-dimensional difference evaluation on the predicted values ​​and measured values ​​of key pipeline network nodes within the time window, and abnormal actuator records are obtained;

[0028] If an abnormality is determined to have occurred, a deviation exceeding limit event will be initiated, and the abnormality detection mechanism will be triggered immediately to mark and record the position of the suspicious actuator or pipe section.

[0029] Furthermore, after receiving the deviation exceeding limit event and the corresponding abnormal node information, it interacts with the hydraulic digital twin model of the pipeline network to obtain the latest hydraulic state reassessment results. For the optimal control sequence that has not yet been executed or is being executed, the adaptive correction function outputs a new control instruction sequence through rapid iteration or multi-scenario simulation.

[0030] issuing the updated control instruction sequence and its new predicted state sequence to the field actuators;

[0031] For actions that are already executed halfway, the smooth switching mode is enabled; if an emergency such as a serious water leak occurs, the emergency switching mode is enabled. If abnormalities or large deviations still occur repeatedly after correction, the actuator or pipe section is triggered to conduct in-depth fault analysis and fault-tolerant scheduling.

[0032] Furthermore, based on the executor health vector and abnormal executor records, key health decay patterns are extracted by tracing back the executor's historical operation logs and the intervals between abnormal triggering;

[0033] The actuator simulation correction results provided by the pipeline network hydraulic digital twin model enable dynamic estimation of the probability of failure and map the dynamic estimation results to the actuator's residual availability index;

[0034] The actuator failure probability is obtained and the actuator residual availability is obtained; when the calculated residual availability of a valve or water pump is lower than the corresponding threshold, automatic fault-tolerant scheduling is triggered.

[0035] Furthermore, the hydraulic digital twin model of the pipeline network is called to simulate the hydraulic impact after the backup equipment is switched. If the simulation shows that the risk of pressure fluctuation, local water hammer or energy consumption exceeding the limit is low, a rapid switchover is performed;

[0036] If there is a lack of backup equipment or the health of the backup equipment is low, the optimal control sequence is regenerated with the help of a multi-objective optimization function, so that adjacent actuators share the additional load or the scheduling mode is changed, and a hierarchical strategy is adopted when the pipeline network scale is larger than expected.

[0037] Furthermore, the constructed pipeline network hydraulic digital twin model is expanded to multiple scenarios to form a comprehensive pipeline network hydraulic digital twin model; using the deployed sensors and newly added professional measuring points, the hydraulic parameters of each scenario are uniformly mapped to the same coordinate system, and dynamic calibration of cross-system data is performed through adaptive fusion weights.

[0038] Furthermore, an extended multi-objective optimization function is formed by adding a new objective dimension to the multi-objective optimization function;

[0039] Combining rainwater facilities with recycled water pipelines and fire reserves, a model predictive control mechanism similar to step two is adopted. Local optimization is first performed within the local subsystem, and then secondary coordinated scheduling is performed at the global level to output the final control instructions and distribute them to each subsystem for execution.

[0040] Furthermore, if a key equipment failure is detected in the water supply system or stormwater dispatching system, the integrated pipe network hydraulic digital twin model will be used to determine whether the recycled water pipeline or other backup storage facilities can be used for emergency relay;

[0041] If available, the available hydraulic resources will be allocated to the highest priority scenario first, and the resource allocation strategies of other subsystems will be adjusted synchronously, and the output will be the final cross-system control instructions.

[0042] A water flow intelligent monitoring system, comprising:

[0043] The model building unit generates a hydraulic digital twin model of the pipe network based on multi-source data sets collected by sensors and historical operation logs, using multi-dimensional real-time monitoring and actuator health mapping to dynamically simulate the evolution of hydraulic topology and output actuator health vectors;

[0044] The control strategy output unit, when a coordinated scheduling or water leakage isolation task is detected within the control cycle, uses a multi-objective model prediction to couple the valve opening and closing and pump speed control strategies, calculates the optimal control sequence, and outputs an action plan and transient response prediction;

[0045] The strategy correction unit compares the predicted values ​​of the pipeline network hydraulic digital twin model with the actual sensor information. It uses closed-loop deviation detection and adaptive correction to make online adjustments to the valve opening and closing rates and pump station switching timings that have not yet been executed or are being executed, and simultaneously updates the actuator health vector;

[0046] The compensation control unit, when continuously detecting abnormalities in the execution layer or judging that the probability of a fault is increasing, calculates the remaining availability of the actuator, quickly calls the backup valve or adjacent water pump solution, simulates the impact of emergency switching on the hydraulics of the pipe network, and transmits the fault information back;

[0047] The global scheduling unit builds a hydraulic digital twin model of the integrated pipeline network after loading and extension, measures multi-dimensional water use indicators with multi-objective functions, combines fault prediction and fault-tolerant strategies for cross-system collaborative scheduling, and performs layered solutions and resource reallocation for different water supply strategies.

[0048] (3) Beneficial effects

[0049] The present invention provides a water flow intelligent monitoring system and method, which has the following beneficial effects:

[0050] The introduction of high-precision sensors and a hydraulic digital twin model for the pipeline network, along with the use of health parameters such as the actuator health vector to characterize actuator characteristics, allows for a more refined understanding of the hydraulic and equipment conditions across the entire network, significantly improving the ability to detect potential hazards such as valve wear and motor energy consumption.

[0051] Based on a multi-objective optimization function, the system no longer focuses solely on energy consumption or pressure requirements, but instead integrates network transients, water hammer safety, and equipment health weights to achieve coordinated scheduling among multiple actuators. By adjusting actuator health (such as the probability of a valve sticking) in the optimization weights, the control strategy can accurately adapt to complex operating conditions, avoiding water pressure imbalances and excessive energy consumption that may occur under traditional open-loop control.

[0052] The introduced real-time closed-loop feedback and anomaly detection mechanism compares the deviation between predicted values ​​and actual sensor measurements, triggering adaptive corrections in a short period of time. For unexecuted instructions, the valve opening and closing rates or pump power can be modified online based on new observations, reducing control errors caused by execution layer delays or communication loss, thereby maximizing the robustness of global scheduling.

[0053] Relying on fault prediction logic, the remaining availability of actuators is promptly assessed and fault tolerance plans are triggered. Backup valves or pumps are quickly selected and emergency plans are simulated to prevent sudden failure of critical equipment from causing wider water outages or water hammer. The fault prediction and fault tolerance module can transmit health information back and dynamically adapt to equipment aging or high load conditions in a global manner.

[0054] Integrating multiple scenarios, such as stormwater dispatch, fire water supply, and recycled water reuse, into a single management and control platform creates a cross-system collaborative dispatch network. This not only enhances the city's flexibility in extreme climate and fire emergencies, but also further highlights the combined effects of digital twins and multi-objective optimization. Indicators such as firefighting priority and efficient recycled water reuse are simultaneously compatible and optimized with existing domestic water supply goals, enabling the coordinated dispatch of water resources and risk mitigation within large-scale municipal systems.

[0055] In summary, this solution's deep collaboration in actuator fine monitoring, predictive control, real-time adaptive correction, fault tolerance, and cross-scenario coupling can significantly enhance the security of smart water supply systems and integrated water resources management. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of the water flow intelligent monitoring method of the present invention;

[0057] Figure 2 This is a structural diagram of the water flow intelligent monitoring system of the present invention. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] See also Figure 1 The present invention provides a water flow intelligent monitoring method, comprising:

[0060] Step 1: When the initial pipeline network modeling requirement is detected, the multi-source dataset Sa(t) output by the sensors is parsed and combined with historical operation logs. Through multi-dimensional real-time monitoring and actuator wear characterization, a hydraulic digital twin model Sn of the pipeline network is generated that integrates parameters such as pressure, flow, and valve status. The actuator health vector SH(t) is initialized, and key actuator status information such as valve opening and closing delay thresholds is recorded to calibrate the hydraulic dynamic characteristics.

[0061] The step 1 includes the following:

[0062] Step 101: Multi-dimensional monitoring data collection and high-precision sensor status analysis

[0063] Deploy high-precision sensors at key nodes of the pipeline network (including main pipelines, valves, pump station outlets, etc.) to collect real-time data such as pressure, flow, valve opening, and motor operating current;

[0064] The health and status of each sensor are monitored in real time. Sensor self-test information (such as built-in self-diagnostic signals and calibration coefficients) is collected and integrated into the monitoring data. Historical operation logs, maintenance records, and actuator operation timestamps are integrated to form an initial multi-source dataset Sa(t). This dataset contains several basic sequences:

[0065] Sa(t)={P i (t),Q i (t),A j (t),E j (t)}

[0066] Where: P i (t) represents the pressure sequence measured by the sensor at position i;

[0067] Q i (t) represents the flow sequence measured by the sensor at location i;

[0068] A j (t) represents the actual opening or switching ratio of the jth valve;

[0069] E j (t) represents the operating condition information such as the motor current of the j-th water pump or valve actuator;

[0070] A sensor credibility weighting function based on multinomial power norm is introduced to measure the reliability of the sensor under different operating conditions, and the instantaneous comprehensive monitoring vector SF(t) is obtained:

[0071]

[0072] Where: w i is the credibility weight of the sensor at position i, which ranges from 0 to 1 and is comprehensively evaluated based on the sensor’s historical drift and current calibration status; Φ i (t) is the measurement vector of the sensor in dimensions such as pressure and flow; γ is the fusion power index, which is a real number greater than 1 and is used to enhance the suppression of outliers and reduce the sensitivity to extreme measurement errors;

[0073] Combining sensor self-test parameters (calibration coefficients, offsets) with actuator maintenance logs to form an actuator health vector SH(t) that can be updated incrementally over time. This serves as an important reference for determining whether a sensor or actuator is abnormal.

[0074] When used, each sensor is given a credibility weight ω in this step. iIt will be used to evaluate its reference value in model prediction, and can also improve the accuracy and noise resistance of the original data, provide more reliable basic input for digital twins, and reduce misleading of subsequent control strategies; at the same time, by using multiple power norms and self-test parameter fusion methods, it improves outlier detection and anti-interference capabilities.

[0075] Step 102: Construction of the hydraulic digital twin model of the pipeline network and dynamic parameter calibration

[0076] Based on a multi-source dataset Sa(t) and an instantaneous integrated monitoring vector SF(t), a hydraulic digital twin model Sn of a pipeline network (including network topology, equipment distribution, and real-time hydraulic characteristics) is constructed in virtual space. To construct the hydraulic digital twin model in virtual space, static data such as pipeline nodes, pipe diameters, valves, and pump stations in the GIS / CIM are first mapped into a topology, and static attributes are attached to each node and pipe segment. Multidimensional time-series signals such as pressure, flow, valve opening, and pump speed are then periodically acquired from the SCADA / IoT network. The sensor data are filtered and corrected using weighted-normalized multinomial power norm fusion and non-integer-order error metrics to online calibrate the friction coefficient and local resistance of the pipe segment. Finally, based on this topology and real-time parameters, a one-dimensional hydraulic partial differential equation is used to perform spatiotemporal discrete simulations using explicit finite-difference or preconditioned conjugate gradient numerical methods. The model is then encapsulated as a microservice interface to achieve dynamic mapping, prediction, and visualization of the physical pipeline network.

[0077] Adopting nonlinear constrained optimization method, the model parameters are calibrated online iteratively using pipeline hydraulic equations and actuator characteristic equations, so that the model can dynamically track the actual working conditions.

[0078] To further enhance the transient simulation capability of the model, a set of partial differential equations is defined to describe the coupled relationship between pressure fluctuations and flow distribution in the pipeline. Actuator states such as valve opening and motor speed are incorporated into the set of equations as boundary conditions. The abstract form of the equations can be expressed as follows:

[0079]

[0080] Where: h(x,t) is the state quantity such as water head and flow velocity at the spatial coordinate x and time t;

[0081] u(t) represents the control input (such as opening and speed) of the valve and pump actuator at time t, which is generated by the subsequent step 2; f is a nonlinear operator with hydraulic coupling terms;

[0082] To further improve the precision of the model calibration phase, the following error metric function SCa, which combines integral and gradient types, is introduced. This function simultaneously considers state differences and gradient differences in both time and space. By minimizing the error metric function SCa, a more accurate hydraulic digital twin model Sn of the pipe network is obtained after calibration:

[0083]

[0084] Where: The predicted state quantity of the pipeline network at spatial position x and time t by the hydraulic digital twin model of the pipeline network (for example, head, flow velocity, pressure, etc. can be combined into a vector);

[0085] h(x, t) is the real or fused observation state vector (which may also include water head, pressure, flow rate, etc.) obtained by the sensor or actual monitoring system at the same position x and time t; p is the p-norm, which can be p≥1;

[0086] Ω is the spatial domain of the pipe network (in a multidimensional scenario, it can correspond to the mapping domain of the one-dimensional / two-dimensional / three-dimensional pipe network topology); t0 and t1 are the start and end times of the time integration, which are the fault detection window or daily operation cycle;

[0087] is the gradient operator of the state vector on the spatial coordinate x, which is used to examine the effect of the rate of change of the state quantity in space;

[0088] α is the calibration power exponent, which is a non-integer order or a real number greater than 1, used to enhance sensitivity at extreme values ​​while taking into account robustness; μ is the weight factor of the gradient difference in the total error metric, μ ≥ 0;

[0089] By using the error metric function SCa combined with the hydraulic partial differential constraint equation, the parameters Sn of the pipeline network hydraulic digital twin model (such as pipe roughness coefficient and local flow resistance coefficient) are updated in real time. An optimization solution is performed at fixed time intervals or when new data triggers the calibration threshold, so that the digital twin can accurately reflect the transient characteristics of the pipeline network.

[0090] The actuator health vector SHu(t) is incorporated into the calibration process. If a valve wears out, resulting in a mismatch between the actual and theoretical openings, the valve's equivalent flow resistance or opening and closing delay parameters are automatically corrected during the optimization phase to achieve dynamic correction. The output is a real-time updated hydraulic digital twin model Sn of the pipeline network and calibrated actuator characteristic parameters.

[0091] During use, the model is always synchronized with reality under complex hydraulic transient conditions, avoiding simulation distortion caused by pipeline aging or actuator uncertainty. The multidimensional health of the actuator is coupled with the hydraulic equations, which can not only accurately describe the transient process, but also provide a more reliable reference state for subsequent steps. The actuator health vector SH(t) is explicitly included in the parameter calibration of the partial differential equation, allowing the digital twin to more flexibly respond to changes in pump or valve characteristics. Through an iterative calibration process, the pipeline network hydraulic digital twin model can map the actual pipeline network with high precision, thereby improving the reliability and operational efficiency of the entire smart water supply system.

[0092] Step 2: When the preset control cycle arrives or the water leakage isolation task is triggered, the hydraulic digital twin model Sn of the pipeline network and the actuator health vector SH(t) are read. A multi-objective solution strategy combining multi-agent reinforcement learning and model predictive control is adopted. The water hammer risk constraint and the transient coupling effect of the pipeline network are dynamically calculated to generate the optimal control sequence u for valve opening and closing and water pump speed regulation. * (t);

[0093] The second step includes the following:

[0094] Step 201: Multi-objective optimization model construction and constraint setting

[0095] Based on the hydraulic digital twin model Sn of the pipeline network and the multi-source dataset Sa(t), multi-objective control modeling is performed on valve opening and closing sequences, pump speed regulation, and pressure distribution within a certain prediction time domain (e.g., [t, t+H]).

[0096] Introducing nonlinear coupling constraints, the actuator health, transient hydraulic effects of the pipe network, and energy consumption indicators are incorporated into a unified optimization framework;

[0097] Set the timing and amplitude limits for the valve and pump actions (according to the actuator characteristic parameters, such as switch delay, maximum opening and closing angle speed) to avoid blind actions causing water hammer impact or efficiency reduction, and obtain the constraint set of coupling health and transient hydraulic constraints.

[0098] In which: a multi-objective optimization function Soj is defined that takes into account leakage isolation (or pressure stability), energy consumption minimization, and water hammer safety. The operation process of the pipe network is solved discretely or continuously within the time period [t, t+H]. Its form can be set as:

[0099]

[0100] Where: v(τ) represents dynamic quantities such as the pump output flow rate and the actual valve opening and closing rate, which are used to measure the water hammer effect and energy consumption in real time; p(τ) represents the pressure vector of each key node in the pipeline network at time τ (predicted by the pipeline network hydraulic digital twin model Sn); pref is the target or reference pressure vector; u(τ) represents the action vector of the actuator set (valve opening instruction, water pump speed control instruction) at time τ; Θ(p,v) is the water hammer or transient shock risk function, which can be defined as the sum of the weighted nonlinear penalty of the local pressure gradient and the flow rate mutation: as follows:

[0101]

[0102] Where: p i is the instantaneous pressure at the i-th pipe network node; Represents the spatial gradient of the node along the pipe section (which can be calculated by the difference between two adjacent points);

[0103] v i (t) is the flow rate / flow velocity of the node or pipe segment at time t; v i (t)-v i (t-Δt) measures the sudden change of flow within a time step Δt; λ p ,λ v >0 are weight coefficients for pressure gradient and flow mutation, respectively; μ, ν>1 are non-integer power exponents, used to amplify the penalty for extreme gradients or extreme mutations; N is the total number of sampling nodes;

[0104] By adding this Θ(P,v) to the objective function, a high-weight penalty can be given to drastic pressure changes and flow rate mutations that may cause water hammer or pipeline shock during the optimization process, thereby effectively suppressing the generation of dangerous working conditions.

[0105] α1, α2 are non-integer orders or real numbers greater than 1, which are used to improve the sensitivity to abnormal or extreme values ​​while taking into account the overall robustness; ||·|| r 、||·|| s is the r-norm or s-norm, which is used to measure the magnitude of each state or control vector;

[0106] λ1, λ2, and λ3 are the relative weights of the three objectives in the overall objective function, which can be dynamically adjusted according to the operation strategy;

[0107] If a valve is detected to have a risk of jamming or a switching delay, the valve's actionable range is explicitly written in the constraint, for example: The upper and lower limits depend on the health status of the valve, which can avoid exceeding the equipment capacity during subsequent actions and prevent the control scheme from being theoretically feasible but not practically feasible; combining the pipeline transient equation and the water pump speed regulation model to set the valve opening change rate or the water pump speed change rate The maximum allowable range is set to avoid severe pressure shock or energy consumption surge caused by sudden changes in valve and pump working conditions in a very short time.

[0108] When used, by simultaneously introducing multiple dimensions such as pressure deviation, control energy consumption, and transient impact into the objective function, it is possible to balance safety and economy, consider the actuator health and dynamic timing constraints, ensure that the generated control strategy is truly executable, improve the success rate of subsequent implementation, and reduce the risk of water hammer; coupling the non-integer order norm (α1, α2) with the space-time transient impact function Θ allows the optimization to more comprehensively reflect the impact of extreme working conditions, explicitly incorporating the actuator health status into the multi-objective optimization and constraints, and avoiding potential failures in the strategy generation stage.

[0109] Step 202: Multi-agent control solution and strategy release

[0110] Adopting an advanced model predictive control (MPC) framework, or splitting the control object into multiple agents (valve agent and pump station agent), and using distributed reinforcement learning and other methods to solve the multi-objective optimization function Soj; through iterative solution or simulation-evaluation cycle, gradually optimizing in the prediction time domain, and finally generating the optimal control sequence u for valve opening and closing and pump speed regulation. * (t), and updated over time;

[0111] Taking into account execution latency and communication uncertainty factors, a safe time window is reserved before the strategy is released to ensure that when instructions are issued, hardware delays and network delays will not cause a serious disconnect between the instructions and the prediction model.

[0112] In large-scale pipe networks, each valve and pump can be regarded as an interconnected intelligent agent, which performs local optimization separately. Then, the central coordinator performs a secondary solution based on the multi-objective optimization function Soj, so that the overall control decision score reaches the global optimal or near-optimal state.

[0113] If the reinforcement learning method is used, an adaptive learning rate can be introduced to address hydraulic changes and actuator health changes, which can be used to speed up the update strategy when the network load suddenly increases, and appropriately slow down the update during the stable period to avoid excessive oscillation. Multiple candidate control sequences u are simulated and evaluated, and the optimal one under the multi-objective optimization function Soj is selected as the actual release instruction u. * , and then record the predicted state sequence p of the instruction in the digital twin environment * (t) as a comparison benchmark.

[0114] When in use, with the help of distributed or reinforcement learning algorithms, approximate global optimal solutions can be quickly obtained in large-scale pipeline network scenarios. By incorporating equipment health constraints and transient safety constraints into the solution phase, the risk of water hammer impact can be significantly reduced while meeting energy consumption optimization. Multi-agent reinforcement learning and model predictive control are deeply integrated, and the solution efficiency and network-wide coordination are improved through a combination of distributed and centralized modes. A safety time window is set before the strategy is released, and the delay mechanism is incorporated into the reinforcement learning or MPC process, making the algorithm more reliable in network delays and execution delays.

[0115] However, actual municipal pipe networks may contain uncertainties such as actuator delays, equipment failures, and communication packet loss, leading to deviations between actual operating conditions and predicted results. If these deviations are not detected and corrected promptly, the effectiveness of the control strategy will be reduced, and even new risks such as water hammer and water supply imbalances will arise. Based on this, Step 3 focuses on proposing technical solutions for real-time closed-loop feedback and adaptive correction of anomaly detection. This aims to fully utilize sensor data and the pipe network hydraulic digital twin model for dynamic comparison and secondary scheduling, ensuring that control commands can still be accurately executed in the face of random interference or localized failures.

[0116] Step 3: When the optimal control sequence u * After the signal (t) is issued and the actuator starts to move, the deviation between the predicted value and the observed value of the pipeline network hydraulic digital twin model Sn is compared, and the adaptive correction algorithm is called to adjust the valve closing rate or water pump switching timing that has not been completed. The blocking warning threshold in the actuator health vector SH(t) is also updated synchronously, and the valve position feedback sub-process is triggered to synchronously evaluate the degree of water hammer impact, so as to reduce the execution layer delay and miscontrol risk;

[0117] The step three includes the following:

[0118] Step 301: Real-time monitoring - prediction deviation identification and anomaly detection

[0119] Relying on the predicted state sequence p * (t) and the optimal control sequence u * (t), sensor observation value p collected in real time obs (t) (such as pressure, flow, actual valve opening, pump speed, etc.) deviation ratio, using the multi-dimensional deviation measurement function Dev(t) to perform multi-dimensional difference evaluation on the predicted value and measured value of key pipeline network nodes in the time window [t-Δt, t], and obtain abnormal actuator records;

[0120] When the evaluation result exceeds the predetermined threshold or an abnormal trend appears (such as the actual opening and closing curve of the valve deviates seriously from the prediction), the abnormal detection mechanism is triggered immediately to mark and record the position of the suspicious actuator or pipe section.

[0121] Define a multi-dimensional deviation measurement function Dev(t) to measure the overall difference of key nodes in the pipeline network within the time window [t-Δt,t]:

[0122]

[0123] in: represents the prediction vector of the hydraulic digital twin model of the pipeline network for the head, pressure or flow at the kth monitoring node of the pipeline network; p obs,k (τ) is the corresponding real-time observation vector; ||·|| r represents the r-norm (r≥1), α is a real number greater than 1; ω k is the importance weight of node k, which ranges from 0 to 1 and is used to give higher detection sensitivity in locations such as leak isolation areas and near key valves; Δt is the window length;

[0124] The multi-dimensional deviation measurement function Dev(t) is compared with the preset dynamic threshold Θ S301 (t) contrast;

[0125] If Dev(t)>Θ S301 (t), that is, the deviation is considered abnormal, and the corresponding key node and actuator number are marked, and the dynamic threshold Θ S301 (t) Dynamic adjustment can be made based on the current operation mode of the network (e.g. low load at night vs. high load during the day) and the sensitivity of the control strategy;

[0126] Once an abnormality is determined, a deviation exceeding limit event is initiated to carry out adaptive correction and online adjustment, and the abnormal node and actuator number are recorded.

[0127] When used, use the node importance weight ω k With dynamic threshold Θ S301 (t), differentiated detection can be achieved according to the pipeline network operation strategy, the probability of misjudgment and missed judgment can be reduced, the importance of nodes can be incorporated into a unified detection framework, and higher sensitivity can be given to key areas, which helps to accurately locate local abnormal sources.

[0128] Step 302: Adaptive correction and online instruction rescheduling

[0129] After receiving the deviation exceeding limit event and its corresponding abnormal node information, it interacts with the hydraulic digital twin model Sn of the pipe network to obtain the latest hydraulic state reassessment results; the optimal control sequence u that has not been executed or is being executed is * (t) Perform local or overall rescheduling, including changing the valve closing rate, temporarily activating the backup water pump and other real-time operations; To ensure that the correction does not cause new water hammer waves or energy waste, an adaptive correction function Sap(t) is defined to correct the control instructions online:

[0130] Sap(t)=Φ(u * (t),δp(t),STwin,γ),

[0131] Where: u * (t) is the optimal control instruction; δp(t) represents the main deviation detected, such as the accumulated pressure and flow deviation; STwin is the hydraulic digital twin model of the pipeline network and its current calibration status, which is used to online simulate the impact of the correction scheme on the system hydraulics;

[0132] γ is the adaptive weight coefficient, which is greater than 0 and is used to control the correction amplitude and the required safety margin;

[0133] The adaptive correction function Sap(t) outputs a new control instruction sequence through rapid iteration or multi-scenario simulation Ensure that the target control effect is as close as possible under the safety constraints. After the correction and update are completed, the updated control instruction sequence and its new predicted state sequence Release to field actuators;

[0134] For actions that are already executed halfway, such as closing a valve 50% and then the remaining 50%, the smooth switching mode can be enabled to extend the subsequent valve closing curve for a short period of time to disperse water pressure fluctuations. If an emergency such as a serious water leak occurs, the emergency switching mode can be enabled to quickly close the valve or start the pump, avoiding overly aggressive command changes when the actuator is in poor condition.

[0135] Corrected control instruction sequence It helps to assess whether it is necessary to start a backup valve or water pump to share the load. If abnormalities or large deviations still occur repeatedly after correction, the actuator or pipe section can be triggered for in-depth fault analysis and fault-tolerant scheduling.

[0136] When in use, the adaptive correction mechanism is used to promptly correct instructions that may be out of control or miscontrolled, avoiding secondary risks such as water hammer and unstable supply pressure. Combined with emergency and smooth switching strategies, it can flexibly respond to different fault severities and improve the reliability and real-time performance of the system. By using adaptive correction functions and combining them with real-time simulation of the pipeline network hydraulic digital twin model, local anomalies are quickly iterated, and adjustments are made to balance safety margins and control targets. By integrating emergency switching and smooth switching modes, the execution method can be dynamically selected under the premise of different equipment health levels, greatly reducing pipeline network impacts or actuator damage caused by blind operations. By continuously monitoring the actual actuator status and sensor observations, a comprehensive adaptive correction operation is triggered once an anomaly occurs, and the timing and amplitude of execution instructions such as valves and water pumps are readjusted to ensure the rapid implementation and safety of the strategy.

[0137] However, if the failure risk of a particular actuator continues to accumulate or local anomalies occur frequently, temporary fixes alone will not be enough to fundamentally eliminate potential hazards. Based on this, Step 4 will focus on proposing methods for fault prediction and fault-tolerant scheduling. First, the network hydraulic digital twin model constructed in Step 1 and the actuator health assessment data will be used to predict failure trends for key equipment. Second, when the prediction results indicate high risk, fault-tolerant and self-healing operations will be implemented, such as activating backup valves or dispatching adjacent pumps to share pressure. Furthermore, the network hydraulic digital twin model will be used to simulate the impact of emergency plans on the network's hydraulic safety in advance.

[0138] Step 4: If actuator failure warnings are continuously detected or the probability of failure increases significantly based on the stuck threshold assessment, the actuator health vector SH(t) is obtained and the actuator residual availability Rle(t) is calculated. The backup valve or adjacent water pump relay plan is quickly called and the emergency switching hydraulic impact is simulated in the pipeline network hydraulic digital twin model Sn. The failure information is synchronously fed back to the multi-objective optimization function Soj and the real-time closed-loop deviation detection.

[0139] The step 4 includes the following contents:

[0140] Step 401: Fault prediction and health dynamic assessment

[0141] Based on the actuator health vector SH(t) and abnormal actuator records, its operating characteristics (such as opening and closing time, motor current fluctuation, and jamming frequency) are comprehensively analyzed and a fault prediction model is constructed, where:

[0142] By backtracking the actuator's historical operation logs and the intervals between abnormal triggers, key health decay patterns are extracted. Here, by backtracking the operation logs of each actuator (recording the timestamps of each start / stop or speed regulation instruction) and the abnormal trigger time (the deviation alarm moment identified in step 301), the fault interval between two adjacent abnormalities and the number of operations within this interval are first calculated to obtain the average fault interval and the operation-failure ratio ρ = 1 / N. op ;

[0143] Then, a time decay kernel κ(ΔT) = exp(–β·ΔT) (β>0) is introduced to weight the failure impact at different times. Finally, a discrete decay model h(n) = ∏[1–κ·ρ·κ(ΔT)] or a continuous survival model h(t) = exp[–(t / λ) k ] Fit the decay curve of health with fault accumulation and operation frequency, so as to output the health decay pattern of each actuator in real time, which serves as the key input for subsequent failure probability prediction and fault-tolerant scheduling.

[0144] The execution simulation correction results provided by the pipeline network hydraulic digital twin model Sn are used to dynamically estimate the probability of failure. This dynamic estimation result is mapped to the residual availability index of the actuator to accurately determine whether to activate the backup equipment in the subsequent fault tolerance mechanism. The failure probability function is a combination of the time domain kernel function κ(·) and the multidimensional health mapping function Ψ(·). Its specific form is as follows:

[0145]

[0146] Fdict(t) is the predicted actuator failure probability at time t;

[0147] h(τ) is the local mapping of the actuator health vector SH(t), which contains information such as wear, jam history, number of openings and closings, and motor overload; is the time derivative related to the health vector, reflecting the degradation rate of health over time or the stage-by-stage repair rate;

[0148] κ(t-τ) is a time decay or weight kernel function that describes the contribution of health degradation at time τ to the current time t. κ(Δ) ≥ 0. Common optional forms include exponential decay ε(Δ) = βexp(-ρΔ) or piecewise linear decay. β, ρ> 0 are adjustable parameters. It is a function that maps the health vector and its dynamic change rate to the instant failure rate or degradation rate index. For example, it can be specifically defined as a weighted model of the product of health size and degradation rate. It can be regarded as a generalized disaster rate or hazard degree. exp(·) is the natural exponential function; α0 is the start time of fault prediction, and t is the current prediction time;

[0149] Based on the above failure probability Fdict(t), the actuator residual availability function is defined as: Rle(t) = 1-Fdict(t), which is used to quantify the probability that the device can still work reliably in the short term (such as the next execution cycle);

[0150] When the residual availability function Rle(t) of a valve or pump is calculated to be lower than the corresponding threshold (indicating a high probability of a high-risk failure), this risk information is immediately sent to step 402 to trigger automatic fault-tolerant scheduling;

[0151] Assuming the failure probability is still within an acceptable range, there is no need to immediately implement large-scale fault tolerance. Instead, the results only need to be reported to step 2 or step 3 for dynamic weight adjustment (e.g., reducing the trust in the actuator in the next control optimization cycle).

[0152] Compared to traditional fault prediction methods based on fixed lifespans or simple statistics, this method utilizes a hydraulic digital twin model of the pipeline network and anomaly detection records to assess actuator failure risk at the level of state transition rate, providing greater flexibility. Through the residual availability function Rle(t), failure probability can be quantitatively incorporated into global scheduling and backup strategies, reducing unnecessary maintenance costs and effectively preventing sudden failures. By integrating multidimensional health characteristics and anomaly detection results, a more forward-looking actuator fault prediction method is formed, providing more intuitive control of the remaining healthy resources of each actuator at the scheduling level.

[0153] Step 402: Fault tolerance and self-healing scheduling

[0154] When it is detected that the probability of actuator failure is too high or abnormalities occur frequently, the fault-tolerant logic is activated and the deployed backup valves or pumps that can work in parallel are preferentially retrieved.

[0155] The hydraulic digital twin model Sn of the pipe network is called to simulate the hydraulic impact after the backup equipment is switched. If the simulation shows that the risk of pressure fluctuation, local water hammer or energy consumption exceeding the limit is low, a rapid switch is performed. In the case of a lack of backup equipment or the health of the backup equipment is also worrying, the optimal control sequence u is regenerated with the help of the multi-objective optimization function Soj. * (t) Allow adjacent actuators to share the extra load or change the scheduling mode to ensure the safety and efficiency of the overall pipeline network;

[0156] Define a self-healing scheduling function Sgy(t) to quickly generate alternative plans after a faulty or high-risk executor exits (or degrades):

[0157]

[0158] in: is the instruction sequence that is still being executed or to be executed after adaptive correction; Rle(t) is the remaining availability information of the actuator; Sn is the hydraulic digital twin model of the pipeline network, which is used for hydraulic simulation and backup switching pre-assessment; Soj is a multi-objective optimization function that assists in evaluating energy consumption and water hammer safety; Δ is the emergency dispatch time window;

[0159] The function Ψ(·) is combined with the digital twin simulation through the fault-tolerant control strategy to output the updated optimal control sequence u * (t), which includes refined scheduling instructions such as activating the standby valve B, reducing the opening of the main valve A by 20%, or increasing the speed of the water pump C by 110% to compensate for the flow;

[0160] If the pipe network is large, a layered strategy can be adopted:

[0161] First, fault-tolerant switching is performed in the local pipe section (for example, mutual backup of valves at the community level). If the local demand cannot be met, it is escalated to the whole network level for coordinated rescheduling. Multi-agent reinforcement learning is used to optimize the global fault-tolerant solution in large-scale scenarios. The final control sequence u is generated. ★ (t) The fault information is transmitted back to the hydraulic digital twin model Sn of the pipeline network in step 2, so that the availability of the faulty or high-risk equipment can be downgraded in the next optimization cycle. At the same time, the real-time closed-loop deviation detection in step 3 is also notified to monitor new operating instructions and continue to track abnormal signs.

[0162] During use, when a key actuator faces failure or performance degradation, the close combination of digital twins and fault-tolerant strategies can complete backup switching or rescheduling in a very short time, significantly reducing the risk of accident spread. The layered self-healing method enables large-scale water supply networks to maintain core water supply services after local failures, enhancing the overall resilience of the system. A self-healing scheduling function Sgy(t) based on digital twins and multi-objective optimization functions Soj is proposed to quickly evaluate the comprehensive impact of equipment switching and system rescheduling on water safety and energy consumption. The layered self-healing scheduling concept is introduced to take into account risk control at both local and global scales, making the solution scalable in ultra-large-scale pipe networks. The failure probability and remaining availability can be calculated in real time to form a forward-looking fault prediction result. If it is determined that the risk of actuator failure is high or there is a local fault, the pipe network hydraulic digital twin model is immediately called to perform fault tolerance and self-healing scheduling evaluation to provide the final control sequence u * (t) Provide a feasible and safe switching strategy.

[0163] Step 5: When the pipeline network enters the multi-scenario coupling stage or when demand signals for rainwater, firefighting, and recycled water arrive, an extension module is loaded onto the original pipeline network hydraulic digital twin model Sn to construct a comprehensive pipeline network hydraulic digital twin model Trn. A multi-objective solution function is used to integrate constraints such as domestic water supply, rainwater peak shaving, and firefighting priority, and to dynamically evaluate the efficiency of recycled water utilization. This is coupled with fault prediction and fault-tolerance strategies to implement cross-system resource allocation and risk mitigation.

[0164] The step five includes the following:

[0165] Step 501: Multi-scenario data fusion and coupling extension of the pipeline network hydraulic digital twin model

[0166] Based on the existing pipe network hydraulic digital twin model Sn, data from additional scenarios such as the stormwater pipe network, recycled water pipelines, fire water tanks, and water storage facilities are added to form a comprehensive pipe network hydraulic digital twin model Trn after multi-scenario expansion.

[0167] Leveraging existing sensors and newly added specialized measurement points (such as rain gauges, recycled water outflow metering devices, and fire water intake monitors), the hydraulic parameters (flow rate, available storage capacity, remaining fire water volume, etc.) of each scenario are uniformly mapped to the same coordinate or topological reference system. Dynamic calibration of cross-system data is achieved through adaptive fusion weights.

[0168] In order to take into account complex coupling processes such as rainwater discharge and secondary transportation of recycled water, extended equations or boundary conditions can be added on the basis of the pipeline network hydraulic digital twin model Sn to simulate the evolution of water flow with multiple sources and multiple destinations.

[0169] When in use, cross-system and cross-scenario data (fire water, recycled water, rainwater, etc.) are integrated into the same digital twin platform, so that multi-scenario needs can be considered simultaneously in the scheduling and correction process from step two to step four. Through the composite hydrological and hydraulic model, when faced with extreme climate or sudden firefighting needs, the water volume complementarity and allocation capabilities of multiple scenarios can be calculated in real time, laying a data foundation for the coordinated scheduling of subsequent steps, and can flexibly respond to transient hydraulic changes caused by the interaction of multiple water sources and demands.

[0170] Step 502: Multi-source multi-objective optimization and hierarchical collaborative control

[0171] On the basis of the multi-objective optimization function Soj, new objective dimensions such as rain and flood peak reduction, recycled water reuse efficiency, and fire water supply safety are added to form an expanded multi-objective optimization function Obj, which can be written in general form as follows:

[0172]

[0173] Where p(τ) is the hydraulic state vector after multi-scenario expansion, representing the key states (such as pressure, flow, water level, storage capacity, etc.) of the pipe network under the coupled conditions of multiple scenarios such as domestic, stormwater, recycled water, and fire protection at time τ. It is updated in real time by the integrated pipe network hydraulic digital twin model Trn.

[0174] Γ(p(τ)) is a multi-objective performance indicator:

[0175] Γ(p(τ))=[Γ life (p(τ)),Γ rain (p(τ)),Γ reuse (p(τ)),Γ fire (p(τ))] T

[0176] Γ life Represents comprehensive indicators such as water supply stability or pressure deviation for domestic water; Γ rain Used to measure the peak reduction rate and drainage efficiency of rainwater dispatching; Γ reuse Characterizes the utilization rate or reuse efficiency of recycled water; Γfire It measures the priority level of fire water supply; W is a four-dimensional weight matrix (usually a diagonal matrix), which is used to perform differential weighting on each target dimension before calculating the vector norm. It can be expressed as:

[0177]

[0178] where λ life ,λ rain ,λ reuse ,λ fire , with a value greater than or equal to 0, representing the weight coefficients of domestic water use, rainwater regulation, recycled water utilization, and fire protection demand respectively;

[0179] ||v|| p The meaning is p-norm, which is used to measure the size of vector v and is commonly defined as:

[0180]

[0181] If p = 2 is the Euclidean norm, p≠2 can also be used to enhance sensitivity to outliers or local increases. The value range is: p≧1, which is adjusted according to actual needs and reliability. α is a non-integer or greater than 1 amplification factor, and [t, t+H] is the time interval;

[0182] Hierarchical collaborative control: Combine stormwater facilities (such as flood storage areas and diversion drainage channels) with recycled water pipelines, fire reserves, and other facilities. A model predictive control mechanism similar to step 2 is used, but local optimization is first performed within the local subsystems (such as the stormwater module, recycled water module, and fire pipeline network), and then a secondary collaborative scheduling is performed at the global level.

[0183] Among them: the entire water network is divided into several subsystems according to function:

[0184] Stormwater module (including flood storage areas and diversion drainage channels), recycled water module (from recycled water plant to reuse network), fire pipe network module (fire water tank and fire hydrant network), domestic water supply module (urban main network and terminal water use);

[0185] Each subsystem runs a local MPC controller in its own receding horizon [t,t+H local ], based on its own digital twin model (such as rainwater hydrological model, recycled water pipeline model, etc.) and local objective function (such as flood peak reduction rate, recycled water reuse efficiency or fire protection pressure guarantee), solve:

[0186]

[0187] The local MPC outputs a set of optimized candidate control sequences for each subsystem (rainwater, recycled water, fire protection, and domestic water supply) and the corresponding local prediction status;

[0188] Where: u local (τ) is the control vector at time τ; for example: stormwater module: gate opening and pump station start and stop instructions; recycled water module: reuse pipe network allocation ratio; fire protection module: fire pump supply pressure setting; domestic water supply module: main valve opening and closing and pump speed adjustment;

[0189] Γ local (p local (τ),u local (τ)) is the single-step cost function of the local subsystem, which quantifies the hydraulic performance (such as peak shaving rate, recycling efficiency, pressure deviation), energy consumption or equipment wear of the subsystem at time τ;

[0190] The global coordinator collects candidate sequences and key interface variables (such as interface node flow and shared pipe head) of each subsystem. In the rolling time domain [t, t+H], a global MPC or quadratic optimization is constructed:

[0191] st global interface constraints

[0192] Among them, λ life ,λ rain ,λ reuse and λ fire The weight coefficients of domestic water supply, stormwater peak reduction, recycled water reuse, and fire protection in the global cost function are used to measure the priority of each scenario;

[0193] Γ life , Γ rain , Γ reuse and Γ fire , corresponding to the global cost function components of the four subsystems, respectively measuring: pressure stability and energy consumption of domestic water supply, rain and flood peak reduction efficiency, recycled water utilization rate and fire water supply safety;

[0194] Among them, the global interface constraint ensures the continuity of flow and head of each subsystem at the junction, avoiding local strategy conflicts;

[0195] Thus, the global optimization outputs the final control instruction Distribute to each subsystem for execution, and restart local MPC and global coordination in the next cycle t+Δt to respond to new real-time observations and fault predictions, forming a local-global-local closed-loop hierarchical collaboration.

[0196] During use, the diversified water demand and scheduling objectives are incorporated into a unified optimization framework to avoid interference or conflict between systems. While ensuring priority for domestic water use, it can also take into account rain and flood peak shaving, efficient use of recycled water and fire safety scheduling, and realize the integrated management of urban water resources. Based on the original step 2 multi-objective control, a multi-scenario multi-objective optimization function Obj is introduced, and a hierarchical-collaborative two-level optimization approach is adopted, so that large-scale urban pipeline networks still have real-time and efficient scheduling capabilities when facing multi-source water and multi-destination demands.

[0197] Step 503: Cross-system fault-tolerant switching and emergency linkage

[0198] If a failure of key equipment (such as large-diameter valves and pumping stations) in the water supply system or stormwater dispatching system is detected in step 4, the integrated pipe network hydraulic digital twin model Trn is called to determine whether the recycled water pipeline or other backup storage facilities can be used for emergency relay; a cross-system linkage control function is introduced to automatically generate cross-system emergency switching plans.

[0199] Its input is: the instruction sequence that is still being executed or to be executed after adaptive correction Actuator residual availability function Rle(t); integrated pipe network hydraulic digital twin model Trn; multi-objective performance index Γ(p(τ)); emergency dispatch time window Δ;

[0200] The output is the final cross-system control instruction u * (t);

[0201] When a sudden increase in fire flow or a rapid rise in rain and flood water levels is discovered, for example, the system will automatically link up with adjacent systems (regulating reservoirs, recycled water pipelines) to share part of the flow or water supply load. Based on the fault-tolerant solution in step 4, the system will further connect the multi-objective optimization of recalculation in step 2 that takes into account energy consumption, pressure safety, water hammer risk, and equipment health, and the adaptive correction algorithm in step 3 to conduct online monitoring and dynamic fine-tuning of the fault-tolerant optimized sequence, compressing the actual execution deviation to within the safety threshold to ensure that the scheduling decision remains safe and feasible. In the event of an emergency demand for fire fighting or main water supply, the water supply priority for recycled water reuse or recycled water distribution pipelines can be automatically reduced, and even the relevant valves can be temporarily closed, so that these hydraulic resources can be allocated to key loads such as domestic water supply or fire water, ensuring that the fire pipeline network has sufficient supply pressure and flow.

[0202] When in use, it not only performs self-healing within a single domestic water supply system, but also enables coordinated scheduling across systems. It uses the backup capacity of fire protection, rainwater, and recycled water to alleviate or replace faulty main water supply equipment. In emergencies such as extreme weather or fire, emergency linkage can maximize the utilization of various urban water storage facilities and drainage channels, thereby improving the efficiency of the city's overall disaster prevention, mitigation, and emergency response. It extends cross-system fault-tolerant switching from simple backup valves or parallel water pumps to rainwater and recycled water links, comprehensively improving the resilience of municipal water systems in complex environments. Through cross-system linkage control functions to support multi-scenario parallel scheduling and priority switching, the system has a high degree of flexibility in dynamically responding to emergencies.

[0203] See also Figure 2 The present invention provides a water flow intelligent monitoring system, comprising:

[0204] The model building unit generates a hydraulic digital twin model of the pipe network based on multi-source data sets collected by sensors and historical operation logs, using multi-dimensional real-time monitoring and actuator health mapping to dynamically simulate the evolution of hydraulic topology and output actuator health vectors;

[0205] The control strategy output unit, when a coordinated scheduling or water leakage isolation task is detected within the control cycle, uses a multi-objective model prediction to couple the valve opening and closing and pump speed control strategies, calculates the optimal control sequence, and outputs an action plan and transient response prediction;

[0206] The strategy correction unit compares the predicted values ​​of the pipeline network hydraulic digital twin model with the actual sensor information. It uses closed-loop deviation detection and adaptive correction to make online adjustments to the valve opening and closing rates and pump station switching timings that have not yet been executed or are being executed, and simultaneously updates the actuator health vector;

[0207] The compensation control unit, when continuously detecting abnormalities in the execution layer or judging that the probability of a fault is increasing, calculates the remaining availability of the actuator, quickly calls the backup valve or adjacent water pump solution, simulates the impact of emergency switching on the hydraulics of the pipe network, and transmits the fault information back;

[0208] The global scheduling unit builds a hydraulic digital twin model of the integrated pipeline network after loading and extension, measures multi-dimensional water use indicators with multi-objective functions, combines fault prediction and fault-tolerant strategies for cross-system collaborative scheduling, and performs layered solutions and resource reallocation for different water supply strategies.

[0209] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0210] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0211] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only for some logical functions. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0212] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0213] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A water flow intelligent monitoring method, characterized by: include, Based on multi-source data sets collected by sensors and historical operation logs, a hydraulic digital twin model of the pipeline network is generated through multi-dimensional real-time monitoring and actuator health mapping to dynamically simulate the evolution of hydraulic topology and output actuator health vectors. When a coordinated scheduling or water leakage isolation task is detected within the control cycle, a multi-objective model prediction is used to couple the valve opening and closing with the pump speed regulation strategy, calculate the optimal control sequence, and output the action plan and transient response prediction. Compare the predicted values ​​of the pipeline network hydraulic digital twin model with the actual sensor information, and use closed-loop deviation detection and adaptive correction to make online adjustments to the valve opening and closing rates and pump station switching timings that have not yet been executed or are being executed, and simultaneously update the actuator health vector; When execution layer anomalies are continuously detected or the probability of a fault is judged to be increasing, the remaining availability of the actuator is calculated, and backup valves or adjacent water pumps are quickly called up to simulate the impact of emergency switching on the hydraulics of the pipe network and transmit the fault information back; After loading and extending, a hydraulic digital twin model of the integrated pipeline network is constructed. Multi-dimensional water use indicators are measured with multi-objective functions. After cross-system collaborative scheduling is carried out in combination with fault prediction and fault-tolerant strategies, different water supply strategies are solved in layers and resources are reallocated.

2. A water flow intelligent monitoring method according to claim 1, characterized in that: Deploy high-precision sensors at key nodes of the pipe network to collect real-time network management data, collect sensor self-test information and integrate it into real-time network management data, and obtain actuator health vectors that are updated incrementally over time; The historical operation logs, maintenance records, and actuator operation timestamps are integrated to form the initial multi-source data set and instantaneous comprehensive monitoring vector, and then the hydraulic digital twin model of the pipeline network is constructed.

3. The water flow intelligent monitoring method according to claim 2, characterized in that: The state difference and gradient difference are considered simultaneously in time and space dimensions, and the calibrated hydraulic digital twin model of the pipeline network is obtained by minimizing the error metric function; The error metric function is combined with the hydraulic partial differential constraint equation to update the parameters of the pipeline network hydraulic digital twin model in real time, and the actuator health vector is incorporated into the calibration process.

4. A water flow intelligent monitoring method according to claim 3, characterized in that: Perform multi-objective control modeling on valve opening and closing sequence, pump speed regulation, and pressure distribution within a certain prediction time domain to obtain a multi-objective optimization function. Distributed reinforcement learning is used to gradually optimize the multi-objective optimization function within a certain prediction time domain, generate the optimal control sequence for valve opening and closing and water pump speed regulation, and update it over time.

5. A water flow intelligent monitoring method according to claim 4, characterized in that: Set timing and amplitude limits for valve and pump actions to obtain a constraint set that couples health and transient hydraulic constraints. If a valve is detected to have a stuck risk or switching delay, the corresponding valve's actionable range is explicitly written into the constraint set. Each valve and water pump is locally optimized separately. After a secondary solution based on the multi-objective optimization function, multiple candidate control sequences are simulated and evaluated. The optimal one under the multi-objective optimization function is selected as the actual issued instruction, and the predicted state sequence of the corresponding instruction in the digital twin environment is recorded as a comparison benchmark.

6. A water flow intelligent monitoring method according to claim 5, characterized in that: By comparing the deviation of sensor observations collected in real time, a multi-dimensional deviation measurement function is used to perform multi-dimensional difference evaluation on the predicted and measured values ​​of key pipeline network nodes within the time window, and abnormal actuator records are obtained; If an abnormality is determined to have occurred, a deviation exceeding limit event will be initiated, and the abnormality detection mechanism will be triggered immediately to mark and record the position of the suspicious actuator or pipe section.

7. The water flow intelligent monitoring method according to claim 6, characterized in that: After receiving the deviation exceeding limit event and its corresponding abnormal node information, it interacts with the hydraulic digital twin model of the pipeline network to obtain the latest hydraulic status reassessment results; For the optimal control sequence that has not been executed or is being executed, the adaptive correction function outputs a new control instruction sequence through rapid iteration or multi-scenario simulation, and publishes the updated control instruction sequence and its new predicted state sequence to the on-site actuator; For actions that have been executed halfway, the smooth switching mode is enabled. If an emergency such as a serious water leak occurs, the emergency switching mode is enabled. If abnormalities or large deviations still occur repeatedly after correction, the actuator or pipe section is triggered to conduct in-depth fault analysis and fault-tolerant scheduling.

8. The water flow intelligent monitoring method according to claim 7, characterized in that: Based on the executor health vector and abnormal executor records, key health decay patterns are extracted by tracing back the executor's historical operation logs and the intervals between abnormal triggering. The actuator simulation correction results provided by the pipeline network hydraulic digital twin model enable dynamic estimation of the probability of failure and map the dynamic estimation results to the actuator's residual availability index; The actuator failure probability is obtained and then the actuator residual availability is obtained. When the calculated residual availability of a valve or water pump is lower than the corresponding threshold, automatic fault-tolerant scheduling is triggered.

9. The water flow intelligent monitoring method according to claim 8, characterized in that: The hydraulic digital twin model of the pipe network is used to simulate the hydraulic impact of the backup equipment switchover. If the simulation shows low risk of pressure fluctuations, local water hammer, or excessive energy consumption, a rapid switchover is performed. If there is a lack of backup equipment or the health of the backup equipment is low, the optimal control sequence is regenerated with the help of a multi-objective optimization function to allow adjacent actuators to share the additional load or change the scheduling mode. A hierarchical strategy is adopted when the pipeline network scale is larger than expected.

10. The water flow intelligent monitoring method according to claim 9, characterized in that: The constructed pipeline network hydraulic digital twin model is expanded to multiple scenarios to form a comprehensive pipeline network hydraulic digital twin model; using the deployed sensors and newly added professional measuring points, the hydraulic parameters of each scenario are mapped to the same coordinate system, and dynamic calibration of cross-system data is performed through adaptive fusion weights.

11. The water flow intelligent monitoring method according to claim 10, characterized in that: An extended multi-objective optimization function is formed by adding a new objective dimension to the multi-objective optimization function; Combining rainwater facilities with recycled water pipelines and fire reserves, a model predictive control mechanism is adopted to first perform local optimization within the local subsystem, then conduct secondary coordinated scheduling at the global level, and output the final control instructions to be distributed to each subsystem for execution.

12. A water flow intelligent monitoring method according to claim 11, characterized in that: If a key equipment failure is detected in the water supply system or stormwater dispatching system, the integrated pipe network hydraulic digital twin model will be used to determine whether the recycled water pipeline or other backup storage facilities can be used for emergency relay; If available, the available hydraulic resources will be allocated to the highest priority scenario first, and the resource allocation strategies of other subsystems will be adjusted synchronously, and the output will be the final cross-system control instructions.

13. A water flow intelligent monitoring system, characterized by: include, The model building unit generates a hydraulic digital twin model of the pipe network based on multi-source data sets collected by sensors and historical operation logs, using multi-dimensional real-time monitoring and actuator health mapping to dynamically simulate the evolution of hydraulic topology and output actuator health vectors; The control strategy output unit, when a coordinated scheduling or water leakage isolation task is detected within the control cycle, uses a multi-objective model prediction to couple the valve opening and closing and pump speed control strategies, calculates the optimal control sequence, and outputs an action plan and transient response prediction; The strategy correction unit compares the predicted values ​​of the pipeline network hydraulic digital twin model with the actual sensor information. It uses closed-loop deviation detection and adaptive correction to make online adjustments to the valve opening and closing rates and pump station switching timings that have not yet been executed or are being executed, and simultaneously updates the actuator health vector; The compensation control unit, when continuously detecting abnormalities in the execution layer or judging that the probability of a fault is increasing, calculates the remaining availability of the actuator, quickly calls the backup valve or adjacent water pump solution, simulates the impact of emergency switching on the hydraulics of the pipe network, and transmits the fault information back; The global scheduling unit builds a hydraulic digital twin model of the integrated pipeline network after loading and extension, measures multi-dimensional water use indicators with multi-objective functions, combines fault prediction and fault-tolerant strategies for cross-system collaborative scheduling, and performs layered solutions and resource reallocation for different water supply strategies.

Images